Thomas Gleixner | 09c434b | 2019-05-19 13:08:20 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 2 | /* |
| 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 Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 27 | #include <linux/sched.h> |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 28 | #include <linux/cpumask.h> |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 29 | #include <linux/list.h> |
| 30 | #include <linux/mm.h> |
| 31 | #include <linux/module.h> |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 32 | #include <linux/page-flags.h> |
| 33 | #include <linux/migrate.h> |
| 34 | #include <linux/node.h> |
| 35 | #include <linux/compaction.h> |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 36 | #include <linux/percpu.h> |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 37 | #include <linux/preempt.h> |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 38 | #include <linux/workqueue.h> |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 39 | #include <linux/slab.h> |
| 40 | #include <linux/spinlock.h> |
| 41 | #include <linux/zpool.h> |
Qian Cai | af4798a | 2020-05-27 22:20:40 -0700 | [diff] [blame] | 42 | #include <linux/kmemleak.h> |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 43 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 44 | /* |
| 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 Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 48 | * 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 Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 53 | */ |
| 54 | #define NCHUNKS_ORDER 6 |
| 55 | |
| 56 | #define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER) |
| 57 | #define CHUNK_SIZE (1 << CHUNK_SHIFT) |
Vitaly Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 58 | #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 Lin | e3c0db4 | 2021-06-30 18:50:24 -0700 | [diff] [blame] | 61 | #define NCHUNKS (TOTAL_CHUNKS - ZHDR_CHUNKS) |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 62 | |
zhong jiang | f201ebd | 2017-02-22 15:46:51 -0800 | [diff] [blame] | 63 | #define BUDDY_MASK (0x3) |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 64 | #define BUDDY_SHIFT 2 |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 65 | #define SLOTS_ALIGN (0x40) |
| 66 | |
| 67 | /***************** |
| 68 | * Structures |
| 69 | *****************/ |
| 70 | struct z3fold_pool; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 71 | |
| 72 | enum buddy { |
| 73 | HEADLESS = 0, |
| 74 | FIRST, |
| 75 | MIDDLE, |
| 76 | LAST, |
| 77 | BUDDIES_MAX = LAST |
| 78 | }; |
| 79 | |
| 80 | struct 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 Wool | fc54886 | 2020-12-14 19:12:30 -0800 | [diff] [blame] | 86 | unsigned long pool; /* back link */ |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 87 | rwlock_t lock; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 88 | }; |
| 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 Wool | bb9a374 | 2019-07-16 16:25:48 -0700 | [diff] [blame] | 100 | * @pool: pointer to the containing pool |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 101 | * @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 Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 106 | * @mapped_count: the number of objects currently mapped |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 107 | */ |
| 108 | struct 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 Wool | bb9a374 | 2019-07-16 16:25:48 -0700 | [diff] [blame] | 114 | struct z3fold_pool *pool; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 115 | 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 Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 121 | unsigned short mapped_count:2; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 122 | unsigned short foreign_handles:2; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 123 | }; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 124 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 125 | /** |
| 126 | * struct z3fold_pool - stores metadata for each z3fold pool |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 127 | * @name: pool name |
| 128 | * @lock: protects pool unbuddied/lru lists |
| 129 | * @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 Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 133 | * @lru: list tracking the z3fold pages in LRU order by most recently |
| 134 | * added buddy. |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 135 | * @stale: list of pages marked for freeing |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 136 | * @pages_nr: number of z3fold pages in the pool. |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 137 | * @c_handle: cache for z3fold_buddy_slots allocation |
Mel Gorman | 3052217 | 2021-06-30 18:53:44 -0700 | [diff] [blame] | 138 | * @zpool: zpool driver |
| 139 | * @zpool_ops: zpool operations structure with an evict callback |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 140 | * @compact_wq: workqueue for page layout background optimization |
| 141 | * @release_wq: workqueue for safe page release |
| 142 | * @work: work_struct for safe page release |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 143 | * |
| 144 | * This structure is allocated at pool creation time and maintains metadata |
| 145 | * pertaining to a particular z3fold pool. |
| 146 | */ |
| 147 | struct z3fold_pool { |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 148 | const char *name; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 149 | spinlock_t lock; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 150 | spinlock_t stale_lock; |
| 151 | struct list_head *unbuddied; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 152 | struct list_head lru; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 153 | struct list_head stale; |
Vitaly Wool | 12d59ae | 2017-02-24 14:57:15 -0800 | [diff] [blame] | 154 | atomic64_t pages_nr; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 155 | struct kmem_cache *c_handle; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 156 | struct zpool *zpool; |
| 157 | const struct zpool_ops *zpool_ops; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 158 | struct workqueue_struct *compact_wq; |
| 159 | struct workqueue_struct *release_wq; |
| 160 | struct work_struct work; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 161 | }; |
| 162 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 163 | /* |
| 164 | * Internal z3fold page flags |
| 165 | */ |
| 166 | enum z3fold_page_flags { |
Vitaly Wool | 5a27aa8 | 2017-02-24 14:57:26 -0800 | [diff] [blame] | 167 | PAGE_HEADLESS = 0, |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 168 | MIDDLE_CHUNK_MAPPED, |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 169 | NEEDS_COMPACTING, |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 170 | PAGE_STALE, |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 171 | PAGE_CLAIMED, /* by either reclaim or free */ |
Miaohe Lin | 943fb61 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 172 | PAGE_MIGRATED, /* page is migrated and soon to be released */ |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 173 | }; |
| 174 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 175 | /* |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 176 | * handle flags, go under HANDLE_FLAG_MASK |
| 177 | */ |
| 178 | enum z3fold_handle_flags { |
| 179 | HANDLES_NOFREE = 0, |
| 180 | }; |
| 181 | |
| 182 | /* |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 183 | * Forward declarations |
| 184 | */ |
| 185 | static struct z3fold_header *__z3fold_alloc(struct z3fold_pool *, size_t, bool); |
| 186 | static void compact_page_work(struct work_struct *w); |
| 187 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 188 | /***************** |
| 189 | * Helpers |
| 190 | *****************/ |
| 191 | |
| 192 | /* Converts an allocation size in bytes to size in z3fold chunks */ |
| 193 | static int size_to_chunks(size_t size) |
| 194 | { |
| 195 | return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT; |
| 196 | } |
| 197 | |
| 198 | #define for_each_unbuddied_list(_iter, _begin) \ |
| 199 | for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++) |
| 200 | |
Vitaly Wool | bb9f6f6 | 2019-05-31 22:30:39 -0700 | [diff] [blame] | 201 | static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool, |
| 202 | gfp_t gfp) |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 203 | { |
Miaohe Lin | f4bad64 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 204 | struct z3fold_buddy_slots *slots = kmem_cache_zalloc(pool->c_handle, |
| 205 | gfp); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 206 | |
| 207 | if (slots) { |
Qian Cai | af4798a | 2020-05-27 22:20:40 -0700 | [diff] [blame] | 208 | /* It will be freed separately in free_handle(). */ |
| 209 | kmemleak_not_leak(slots); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 210 | slots->pool = (unsigned long)pool; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 211 | rwlock_init(&slots->lock); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 212 | } |
| 213 | |
| 214 | return slots; |
| 215 | } |
| 216 | |
| 217 | static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s) |
| 218 | { |
| 219 | return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK); |
| 220 | } |
| 221 | |
| 222 | static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle) |
| 223 | { |
| 224 | return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1)); |
| 225 | } |
| 226 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 227 | /* Lock a z3fold page */ |
| 228 | static inline void z3fold_page_lock(struct z3fold_header *zhdr) |
| 229 | { |
| 230 | spin_lock(&zhdr->page_lock); |
| 231 | } |
| 232 | |
| 233 | /* Try to lock a z3fold page */ |
| 234 | static inline int z3fold_page_trylock(struct z3fold_header *zhdr) |
| 235 | { |
| 236 | return spin_trylock(&zhdr->page_lock); |
| 237 | } |
| 238 | |
| 239 | /* Unlock a z3fold page */ |
| 240 | static inline void z3fold_page_unlock(struct z3fold_header *zhdr) |
| 241 | { |
| 242 | spin_unlock(&zhdr->page_lock); |
| 243 | } |
| 244 | |
Miaohe Lin | 767cc6c | 2021-06-30 18:50:33 -0700 | [diff] [blame] | 245 | /* return locked z3fold page if it's not headless */ |
| 246 | static inline struct z3fold_header *get_z3fold_header(unsigned long handle) |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 247 | { |
| 248 | struct z3fold_buddy_slots *slots; |
| 249 | struct z3fold_header *zhdr; |
| 250 | int locked = 0; |
| 251 | |
| 252 | if (!(handle & (1 << PAGE_HEADLESS))) { |
| 253 | slots = handle_to_slots(handle); |
| 254 | do { |
| 255 | unsigned long addr; |
| 256 | |
| 257 | read_lock(&slots->lock); |
| 258 | addr = *(unsigned long *)handle; |
| 259 | zhdr = (struct z3fold_header *)(addr & PAGE_MASK); |
Miaohe Lin | 767cc6c | 2021-06-30 18:50:33 -0700 | [diff] [blame] | 260 | locked = z3fold_page_trylock(zhdr); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 261 | read_unlock(&slots->lock); |
Miaohe Lin | 943fb61 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 262 | if (locked) { |
| 263 | struct page *page = virt_to_page(zhdr); |
| 264 | |
| 265 | if (!test_bit(PAGE_MIGRATED, &page->private)) |
| 266 | break; |
| 267 | z3fold_page_unlock(zhdr); |
| 268 | } |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 269 | cpu_relax(); |
Miaohe Lin | 767cc6c | 2021-06-30 18:50:33 -0700 | [diff] [blame] | 270 | } while (true); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 271 | } else { |
| 272 | zhdr = (struct z3fold_header *)(handle & PAGE_MASK); |
| 273 | } |
| 274 | |
| 275 | return zhdr; |
| 276 | } |
| 277 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 278 | static inline void put_z3fold_header(struct z3fold_header *zhdr) |
| 279 | { |
| 280 | struct page *page = virt_to_page(zhdr); |
| 281 | |
| 282 | if (!test_bit(PAGE_HEADLESS, &page->private)) |
| 283 | z3fold_page_unlock(zhdr); |
| 284 | } |
| 285 | |
Vitaly Wool | fc54886 | 2020-12-14 19:12:30 -0800 | [diff] [blame] | 286 | static inline void free_handle(unsigned long handle, struct z3fold_header *zhdr) |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 287 | { |
| 288 | struct z3fold_buddy_slots *slots; |
| 289 | int i; |
| 290 | bool is_free; |
| 291 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 292 | if (WARN_ON(*(unsigned long *)handle == 0)) |
| 293 | return; |
| 294 | |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 295 | slots = handle_to_slots(handle); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 296 | write_lock(&slots->lock); |
| 297 | *(unsigned long *)handle = 0; |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 298 | |
| 299 | if (test_bit(HANDLES_NOFREE, &slots->pool)) { |
| 300 | write_unlock(&slots->lock); |
| 301 | return; /* simple case, nothing else to do */ |
| 302 | } |
| 303 | |
Vitaly Wool | fc54886 | 2020-12-14 19:12:30 -0800 | [diff] [blame] | 304 | if (zhdr->slots != slots) |
| 305 | zhdr->foreign_handles--; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 306 | |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 307 | is_free = true; |
| 308 | for (i = 0; i <= BUDDY_MASK; i++) { |
| 309 | if (slots->slot[i]) { |
| 310 | is_free = false; |
| 311 | break; |
| 312 | } |
| 313 | } |
Uladzislau Rezki | d8f117a | 2020-05-22 22:23:12 -0700 | [diff] [blame] | 314 | write_unlock(&slots->lock); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 315 | |
| 316 | if (is_free) { |
| 317 | struct z3fold_pool *pool = slots_to_pool(slots); |
| 318 | |
Vitaly Wool | fc54886 | 2020-12-14 19:12:30 -0800 | [diff] [blame] | 319 | if (zhdr->slots == slots) |
| 320 | zhdr->slots = NULL; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 321 | kmem_cache_free(pool->c_handle, slots); |
| 322 | } |
| 323 | } |
| 324 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 325 | /* Initializes the z3fold header of a newly allocated z3fold page */ |
Vitaly Wool | 6339841 | 2019-09-23 15:36:51 -0700 | [diff] [blame] | 326 | static struct z3fold_header *init_z3fold_page(struct page *page, bool headless, |
Vitaly Wool | bb9f6f6 | 2019-05-31 22:30:39 -0700 | [diff] [blame] | 327 | struct z3fold_pool *pool, gfp_t gfp) |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 328 | { |
| 329 | struct z3fold_header *zhdr = page_address(page); |
Vitaly Wool | 6339841 | 2019-09-23 15:36:51 -0700 | [diff] [blame] | 330 | struct z3fold_buddy_slots *slots; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 331 | |
| 332 | INIT_LIST_HEAD(&page->lru); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 333 | clear_bit(PAGE_HEADLESS, &page->private); |
| 334 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 335 | clear_bit(NEEDS_COMPACTING, &page->private); |
| 336 | clear_bit(PAGE_STALE, &page->private); |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 337 | clear_bit(PAGE_CLAIMED, &page->private); |
Miaohe Lin | 943fb61 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 338 | clear_bit(PAGE_MIGRATED, &page->private); |
Vitaly Wool | 6339841 | 2019-09-23 15:36:51 -0700 | [diff] [blame] | 339 | if (headless) |
| 340 | return zhdr; |
| 341 | |
| 342 | slots = alloc_slots(pool, gfp); |
| 343 | if (!slots) |
| 344 | return NULL; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 345 | |
Miaohe Lin | c457cd9 | 2021-02-24 12:09:22 -0800 | [diff] [blame] | 346 | memset(zhdr, 0, sizeof(*zhdr)); |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 347 | spin_lock_init(&zhdr->page_lock); |
Vitaly Wool | 5a27aa8 | 2017-02-24 14:57:26 -0800 | [diff] [blame] | 348 | kref_init(&zhdr->refcount); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 349 | zhdr->cpu = -1; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 350 | zhdr->slots = slots; |
Vitaly Wool | bb9a374 | 2019-07-16 16:25:48 -0700 | [diff] [blame] | 351 | zhdr->pool = pool; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 352 | INIT_LIST_HEAD(&zhdr->buddy); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 353 | INIT_WORK(&zhdr->work, compact_page_work); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 354 | return zhdr; |
| 355 | } |
| 356 | |
| 357 | /* Resets the struct page fields and frees the page */ |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 358 | static void free_z3fold_page(struct page *page, bool headless) |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 359 | { |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 360 | if (!headless) { |
| 361 | lock_page(page); |
| 362 | __ClearPageMovable(page); |
| 363 | unlock_page(page); |
| 364 | } |
Vitaly Wool | 5a27aa8 | 2017-02-24 14:57:26 -0800 | [diff] [blame] | 365 | __free_page(page); |
| 366 | } |
| 367 | |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 368 | /* Helper function to build the index */ |
| 369 | static inline int __idx(struct z3fold_header *zhdr, enum buddy bud) |
| 370 | { |
| 371 | return (bud + zhdr->first_num) & BUDDY_MASK; |
| 372 | } |
| 373 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 374 | /* |
| 375 | * Encodes the handle of a particular buddy within a z3fold page |
| 376 | * Pool lock should be held as this function accesses first_num |
| 377 | */ |
Vitaly Wool | 3f9d2b5 | 2019-09-23 15:33:02 -0700 | [diff] [blame] | 378 | static unsigned long __encode_handle(struct z3fold_header *zhdr, |
| 379 | struct z3fold_buddy_slots *slots, |
| 380 | enum buddy bud) |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 381 | { |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 382 | unsigned long h = (unsigned long)zhdr; |
| 383 | int idx = 0; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 384 | |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 385 | /* |
| 386 | * For a headless page, its handle is its pointer with the extra |
| 387 | * PAGE_HEADLESS bit set |
| 388 | */ |
| 389 | if (bud == HEADLESS) |
| 390 | return h | (1 << PAGE_HEADLESS); |
| 391 | |
| 392 | /* otherwise, return pointer to encoded handle */ |
| 393 | idx = __idx(zhdr, bud); |
| 394 | h += idx; |
| 395 | if (bud == LAST) |
| 396 | h |= (zhdr->last_chunks << BUDDY_SHIFT); |
| 397 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 398 | write_lock(&slots->lock); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 399 | slots->slot[idx] = h; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 400 | write_unlock(&slots->lock); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 401 | return (unsigned long)&slots->slot[idx]; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 402 | } |
| 403 | |
Vitaly Wool | 3f9d2b5 | 2019-09-23 15:33:02 -0700 | [diff] [blame] | 404 | static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud) |
| 405 | { |
| 406 | return __encode_handle(zhdr, zhdr->slots, bud); |
| 407 | } |
| 408 | |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 409 | /* only for LAST bud, returns zero otherwise */ |
| 410 | static unsigned short handle_to_chunks(unsigned long handle) |
| 411 | { |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 412 | struct z3fold_buddy_slots *slots = handle_to_slots(handle); |
| 413 | unsigned long addr; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 414 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 415 | read_lock(&slots->lock); |
| 416 | addr = *(unsigned long *)handle; |
| 417 | read_unlock(&slots->lock); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 418 | return (addr & ~PAGE_MASK) >> BUDDY_SHIFT; |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 419 | } |
| 420 | |
zhong jiang | f201ebd | 2017-02-22 15:46:51 -0800 | [diff] [blame] | 421 | /* |
| 422 | * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle |
| 423 | * but that doesn't matter. because the masking will result in the |
| 424 | * correct buddy number. |
| 425 | */ |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 426 | static enum buddy handle_to_buddy(unsigned long handle) |
| 427 | { |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 428 | struct z3fold_header *zhdr; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 429 | struct z3fold_buddy_slots *slots = handle_to_slots(handle); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 430 | unsigned long addr; |
| 431 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 432 | read_lock(&slots->lock); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 433 | WARN_ON(handle & (1 << PAGE_HEADLESS)); |
| 434 | addr = *(unsigned long *)handle; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 435 | read_unlock(&slots->lock); |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 436 | zhdr = (struct z3fold_header *)(addr & PAGE_MASK); |
| 437 | return (addr - zhdr->first_num) & BUDDY_MASK; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 438 | } |
| 439 | |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 440 | static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr) |
| 441 | { |
Vitaly Wool | bb9a374 | 2019-07-16 16:25:48 -0700 | [diff] [blame] | 442 | return zhdr->pool; |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 443 | } |
| 444 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 445 | static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked) |
| 446 | { |
| 447 | struct page *page = virt_to_page(zhdr); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 448 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 449 | |
| 450 | WARN_ON(!list_empty(&zhdr->buddy)); |
| 451 | set_bit(PAGE_STALE, &page->private); |
Vitaly Wool | 3552935 | 2017-10-03 16:15:06 -0700 | [diff] [blame] | 452 | clear_bit(NEEDS_COMPACTING, &page->private); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 453 | spin_lock(&pool->lock); |
| 454 | if (!list_empty(&page->lru)) |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 455 | list_del_init(&page->lru); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 456 | spin_unlock(&pool->lock); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 457 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 458 | if (locked) |
| 459 | z3fold_page_unlock(zhdr); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 460 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 461 | spin_lock(&pool->stale_lock); |
| 462 | list_add(&zhdr->buddy, &pool->stale); |
| 463 | queue_work(pool->release_wq, &pool->work); |
| 464 | spin_unlock(&pool->stale_lock); |
Miaohe Lin | 5e36c25 | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 465 | |
| 466 | atomic64_dec(&pool->pages_nr); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 467 | } |
| 468 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 469 | static void release_z3fold_page_locked(struct kref *ref) |
| 470 | { |
| 471 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, |
| 472 | refcount); |
| 473 | WARN_ON(z3fold_page_trylock(zhdr)); |
| 474 | __release_z3fold_page(zhdr, true); |
| 475 | } |
| 476 | |
| 477 | static void release_z3fold_page_locked_list(struct kref *ref) |
| 478 | { |
| 479 | struct z3fold_header *zhdr = container_of(ref, struct z3fold_header, |
| 480 | refcount); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 481 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 482 | |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 483 | spin_lock(&pool->lock); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 484 | list_del_init(&zhdr->buddy); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 485 | spin_unlock(&pool->lock); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 486 | |
| 487 | WARN_ON(z3fold_page_trylock(zhdr)); |
| 488 | __release_z3fold_page(zhdr, true); |
| 489 | } |
| 490 | |
| 491 | static void free_pages_work(struct work_struct *w) |
| 492 | { |
| 493 | struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work); |
| 494 | |
| 495 | spin_lock(&pool->stale_lock); |
| 496 | while (!list_empty(&pool->stale)) { |
| 497 | struct z3fold_header *zhdr = list_first_entry(&pool->stale, |
| 498 | struct z3fold_header, buddy); |
| 499 | struct page *page = virt_to_page(zhdr); |
| 500 | |
| 501 | list_del(&zhdr->buddy); |
| 502 | if (WARN_ON(!test_bit(PAGE_STALE, &page->private))) |
| 503 | continue; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 504 | spin_unlock(&pool->stale_lock); |
| 505 | cancel_work_sync(&zhdr->work); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 506 | free_z3fold_page(page, false); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 507 | cond_resched(); |
| 508 | spin_lock(&pool->stale_lock); |
| 509 | } |
| 510 | spin_unlock(&pool->stale_lock); |
| 511 | } |
| 512 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 513 | /* |
| 514 | * Returns the number of free chunks in a z3fold page. |
| 515 | * NB: can't be used with HEADLESS pages. |
| 516 | */ |
| 517 | static int num_free_chunks(struct z3fold_header *zhdr) |
| 518 | { |
| 519 | int nfree; |
| 520 | /* |
| 521 | * If there is a middle object, pick up the bigger free space |
| 522 | * either before or after it. Otherwise just subtract the number |
| 523 | * of chunks occupied by the first and the last objects. |
| 524 | */ |
| 525 | if (zhdr->middle_chunks != 0) { |
| 526 | int nfree_before = zhdr->first_chunks ? |
Vitaly Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 527 | 0 : zhdr->start_middle - ZHDR_CHUNKS; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 528 | int nfree_after = zhdr->last_chunks ? |
Vitaly Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 529 | 0 : TOTAL_CHUNKS - |
| 530 | (zhdr->start_middle + zhdr->middle_chunks); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 531 | nfree = max(nfree_before, nfree_after); |
| 532 | } else |
| 533 | nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks; |
| 534 | return nfree; |
| 535 | } |
| 536 | |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 537 | /* Add to the appropriate unbuddied list */ |
| 538 | static inline void add_to_unbuddied(struct z3fold_pool *pool, |
| 539 | struct z3fold_header *zhdr) |
| 540 | { |
| 541 | if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 || |
| 542 | zhdr->middle_chunks == 0) { |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 543 | struct list_head *unbuddied; |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 544 | int freechunks = num_free_chunks(zhdr); |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 545 | |
| 546 | migrate_disable(); |
| 547 | unbuddied = this_cpu_ptr(pool->unbuddied); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 548 | spin_lock(&pool->lock); |
| 549 | list_add(&zhdr->buddy, &unbuddied[freechunks]); |
| 550 | spin_unlock(&pool->lock); |
| 551 | zhdr->cpu = smp_processor_id(); |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 552 | migrate_enable(); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 553 | } |
| 554 | } |
| 555 | |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 556 | static inline enum buddy get_free_buddy(struct z3fold_header *zhdr, int chunks) |
| 557 | { |
| 558 | enum buddy bud = HEADLESS; |
| 559 | |
| 560 | if (zhdr->middle_chunks) { |
| 561 | if (!zhdr->first_chunks && |
| 562 | chunks <= zhdr->start_middle - ZHDR_CHUNKS) |
| 563 | bud = FIRST; |
| 564 | else if (!zhdr->last_chunks) |
| 565 | bud = LAST; |
| 566 | } else { |
| 567 | if (!zhdr->first_chunks) |
| 568 | bud = FIRST; |
| 569 | else if (!zhdr->last_chunks) |
| 570 | bud = LAST; |
| 571 | else |
| 572 | bud = MIDDLE; |
| 573 | } |
| 574 | |
| 575 | return bud; |
| 576 | } |
| 577 | |
Vitaly Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 578 | static inline void *mchunk_memmove(struct z3fold_header *zhdr, |
| 579 | unsigned short dst_chunk) |
| 580 | { |
| 581 | void *beg = zhdr; |
| 582 | return memmove(beg + (dst_chunk << CHUNK_SHIFT), |
| 583 | beg + (zhdr->start_middle << CHUNK_SHIFT), |
| 584 | zhdr->middle_chunks << CHUNK_SHIFT); |
| 585 | } |
| 586 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 587 | static inline bool buddy_single(struct z3fold_header *zhdr) |
| 588 | { |
| 589 | return !((zhdr->first_chunks && zhdr->middle_chunks) || |
| 590 | (zhdr->first_chunks && zhdr->last_chunks) || |
| 591 | (zhdr->middle_chunks && zhdr->last_chunks)); |
| 592 | } |
| 593 | |
| 594 | static struct z3fold_header *compact_single_buddy(struct z3fold_header *zhdr) |
| 595 | { |
| 596 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
| 597 | void *p = zhdr; |
| 598 | unsigned long old_handle = 0; |
| 599 | size_t sz = 0; |
| 600 | struct z3fold_header *new_zhdr = NULL; |
| 601 | int first_idx = __idx(zhdr, FIRST); |
| 602 | int middle_idx = __idx(zhdr, MIDDLE); |
| 603 | int last_idx = __idx(zhdr, LAST); |
| 604 | unsigned short *moved_chunks = NULL; |
| 605 | |
| 606 | /* |
| 607 | * No need to protect slots here -- all the slots are "local" and |
| 608 | * the page lock is already taken |
| 609 | */ |
| 610 | if (zhdr->first_chunks && zhdr->slots->slot[first_idx]) { |
| 611 | p += ZHDR_SIZE_ALIGNED; |
| 612 | sz = zhdr->first_chunks << CHUNK_SHIFT; |
| 613 | old_handle = (unsigned long)&zhdr->slots->slot[first_idx]; |
| 614 | moved_chunks = &zhdr->first_chunks; |
| 615 | } else if (zhdr->middle_chunks && zhdr->slots->slot[middle_idx]) { |
| 616 | p += zhdr->start_middle << CHUNK_SHIFT; |
| 617 | sz = zhdr->middle_chunks << CHUNK_SHIFT; |
| 618 | old_handle = (unsigned long)&zhdr->slots->slot[middle_idx]; |
| 619 | moved_chunks = &zhdr->middle_chunks; |
| 620 | } else if (zhdr->last_chunks && zhdr->slots->slot[last_idx]) { |
| 621 | p += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT); |
| 622 | sz = zhdr->last_chunks << CHUNK_SHIFT; |
| 623 | old_handle = (unsigned long)&zhdr->slots->slot[last_idx]; |
| 624 | moved_chunks = &zhdr->last_chunks; |
| 625 | } |
| 626 | |
| 627 | if (sz > 0) { |
| 628 | enum buddy new_bud = HEADLESS; |
| 629 | short chunks = size_to_chunks(sz); |
| 630 | void *q; |
| 631 | |
| 632 | new_zhdr = __z3fold_alloc(pool, sz, false); |
| 633 | if (!new_zhdr) |
| 634 | return NULL; |
| 635 | |
| 636 | if (WARN_ON(new_zhdr == zhdr)) |
| 637 | goto out_fail; |
| 638 | |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 639 | new_bud = get_free_buddy(new_zhdr, chunks); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 640 | q = new_zhdr; |
| 641 | switch (new_bud) { |
| 642 | case FIRST: |
| 643 | new_zhdr->first_chunks = chunks; |
| 644 | q += ZHDR_SIZE_ALIGNED; |
| 645 | break; |
| 646 | case MIDDLE: |
| 647 | new_zhdr->middle_chunks = chunks; |
| 648 | new_zhdr->start_middle = |
| 649 | new_zhdr->first_chunks + ZHDR_CHUNKS; |
| 650 | q += new_zhdr->start_middle << CHUNK_SHIFT; |
| 651 | break; |
| 652 | case LAST: |
| 653 | new_zhdr->last_chunks = chunks; |
| 654 | q += PAGE_SIZE - (new_zhdr->last_chunks << CHUNK_SHIFT); |
| 655 | break; |
| 656 | default: |
| 657 | goto out_fail; |
| 658 | } |
| 659 | new_zhdr->foreign_handles++; |
| 660 | memcpy(q, p, sz); |
| 661 | write_lock(&zhdr->slots->lock); |
| 662 | *(unsigned long *)old_handle = (unsigned long)new_zhdr + |
| 663 | __idx(new_zhdr, new_bud); |
| 664 | if (new_bud == LAST) |
| 665 | *(unsigned long *)old_handle |= |
| 666 | (new_zhdr->last_chunks << BUDDY_SHIFT); |
| 667 | write_unlock(&zhdr->slots->lock); |
| 668 | add_to_unbuddied(pool, new_zhdr); |
| 669 | z3fold_page_unlock(new_zhdr); |
| 670 | |
| 671 | *moved_chunks = 0; |
| 672 | } |
| 673 | |
| 674 | return new_zhdr; |
| 675 | |
| 676 | out_fail: |
Miaohe Lin | 5e36c25 | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 677 | if (new_zhdr && !kref_put(&new_zhdr->refcount, release_z3fold_page_locked)) { |
| 678 | add_to_unbuddied(pool, new_zhdr); |
| 679 | z3fold_page_unlock(new_zhdr); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 680 | } |
| 681 | return NULL; |
| 682 | |
| 683 | } |
| 684 | |
Vitaly Wool | 1b096e5 | 2017-02-24 14:57:20 -0800 | [diff] [blame] | 685 | #define BIG_CHUNK_GAP 3 |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 686 | /* Has to be called with lock held */ |
| 687 | static int z3fold_compact_page(struct z3fold_header *zhdr) |
| 688 | { |
| 689 | struct page *page = virt_to_page(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 690 | |
Vitaly Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 691 | if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private)) |
| 692 | return 0; /* can't move middle chunk, it's used */ |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 693 | |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 694 | if (unlikely(PageIsolated(page))) |
| 695 | return 0; |
| 696 | |
Vitaly Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 697 | if (zhdr->middle_chunks == 0) |
| 698 | return 0; /* nothing to compact */ |
| 699 | |
| 700 | if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) { |
| 701 | /* move to the beginning */ |
| 702 | mchunk_memmove(zhdr, ZHDR_CHUNKS); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 703 | zhdr->first_chunks = zhdr->middle_chunks; |
| 704 | zhdr->middle_chunks = 0; |
| 705 | zhdr->start_middle = 0; |
| 706 | zhdr->first_num++; |
Vitaly Wool | 1b096e5 | 2017-02-24 14:57:20 -0800 | [diff] [blame] | 707 | return 1; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 708 | } |
Vitaly Wool | 1b096e5 | 2017-02-24 14:57:20 -0800 | [diff] [blame] | 709 | |
| 710 | /* |
| 711 | * moving data is expensive, so let's only do that if |
| 712 | * there's substantial gain (at least BIG_CHUNK_GAP chunks) |
| 713 | */ |
| 714 | if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 && |
| 715 | zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >= |
| 716 | BIG_CHUNK_GAP) { |
| 717 | mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS); |
| 718 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; |
| 719 | return 1; |
| 720 | } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 && |
| 721 | TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle |
| 722 | + zhdr->middle_chunks) >= |
| 723 | BIG_CHUNK_GAP) { |
| 724 | unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks - |
| 725 | zhdr->middle_chunks; |
| 726 | mchunk_memmove(zhdr, new_start); |
| 727 | zhdr->start_middle = new_start; |
| 728 | return 1; |
| 729 | } |
| 730 | |
| 731 | return 0; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 732 | } |
| 733 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 734 | static void do_compact_page(struct z3fold_header *zhdr, bool locked) |
| 735 | { |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 736 | struct z3fold_pool *pool = zhdr_to_pool(zhdr); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 737 | struct page *page; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 738 | |
| 739 | page = virt_to_page(zhdr); |
| 740 | if (locked) |
| 741 | WARN_ON(z3fold_page_trylock(zhdr)); |
| 742 | else |
| 743 | z3fold_page_lock(zhdr); |
Vitaly Wool | 5d03a66 | 2017-11-17 15:26:16 -0800 | [diff] [blame] | 744 | if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) { |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 745 | z3fold_page_unlock(zhdr); |
| 746 | return; |
| 747 | } |
| 748 | spin_lock(&pool->lock); |
| 749 | list_del_init(&zhdr->buddy); |
| 750 | spin_unlock(&pool->lock); |
| 751 | |
Miaohe Lin | 5e36c25 | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 752 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) |
Vitaly Wool | 5d03a66 | 2017-11-17 15:26:16 -0800 | [diff] [blame] | 753 | return; |
Vitaly Wool | 5d03a66 | 2017-11-17 15:26:16 -0800 | [diff] [blame] | 754 | |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 755 | if (test_bit(PAGE_STALE, &page->private) || |
| 756 | test_and_set_bit(PAGE_CLAIMED, &page->private)) { |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 757 | z3fold_page_unlock(zhdr); |
| 758 | return; |
| 759 | } |
| 760 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 761 | if (!zhdr->foreign_handles && buddy_single(zhdr) && |
| 762 | zhdr->mapped_count == 0 && compact_single_buddy(zhdr)) { |
Miaohe Lin | 5e36c25 | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 763 | if (!kref_put(&zhdr->refcount, release_z3fold_page_locked)) { |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 764 | clear_bit(PAGE_CLAIMED, &page->private); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 765 | z3fold_page_unlock(zhdr); |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 766 | } |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 767 | return; |
| 768 | } |
| 769 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 770 | z3fold_compact_page(zhdr); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 771 | add_to_unbuddied(pool, zhdr); |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 772 | clear_bit(PAGE_CLAIMED, &page->private); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 773 | z3fold_page_unlock(zhdr); |
| 774 | } |
| 775 | |
| 776 | static void compact_page_work(struct work_struct *w) |
| 777 | { |
| 778 | struct z3fold_header *zhdr = container_of(w, struct z3fold_header, |
| 779 | work); |
| 780 | |
| 781 | do_compact_page(zhdr, false); |
| 782 | } |
| 783 | |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 784 | /* returns _locked_ z3fold page header or NULL */ |
| 785 | static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool, |
| 786 | size_t size, bool can_sleep) |
| 787 | { |
| 788 | struct z3fold_header *zhdr = NULL; |
| 789 | struct page *page; |
| 790 | struct list_head *unbuddied; |
| 791 | int chunks = size_to_chunks(size), i; |
| 792 | |
| 793 | lookup: |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 794 | migrate_disable(); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 795 | /* First, try to find an unbuddied z3fold page. */ |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 796 | unbuddied = this_cpu_ptr(pool->unbuddied); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 797 | for_each_unbuddied_list(i, chunks) { |
| 798 | struct list_head *l = &unbuddied[i]; |
| 799 | |
| 800 | zhdr = list_first_entry_or_null(READ_ONCE(l), |
| 801 | struct z3fold_header, buddy); |
| 802 | |
| 803 | if (!zhdr) |
| 804 | continue; |
| 805 | |
| 806 | /* Re-check under lock. */ |
| 807 | spin_lock(&pool->lock); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 808 | if (unlikely(zhdr != list_first_entry(READ_ONCE(l), |
| 809 | struct z3fold_header, buddy)) || |
| 810 | !z3fold_page_trylock(zhdr)) { |
| 811 | spin_unlock(&pool->lock); |
| 812 | zhdr = NULL; |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 813 | migrate_enable(); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 814 | if (can_sleep) |
| 815 | cond_resched(); |
| 816 | goto lookup; |
| 817 | } |
| 818 | list_del_init(&zhdr->buddy); |
| 819 | zhdr->cpu = -1; |
| 820 | spin_unlock(&pool->lock); |
| 821 | |
| 822 | page = virt_to_page(zhdr); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 823 | if (test_bit(NEEDS_COMPACTING, &page->private) || |
| 824 | test_bit(PAGE_CLAIMED, &page->private)) { |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 825 | z3fold_page_unlock(zhdr); |
| 826 | zhdr = NULL; |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 827 | migrate_enable(); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 828 | if (can_sleep) |
| 829 | cond_resched(); |
| 830 | goto lookup; |
| 831 | } |
| 832 | |
| 833 | /* |
| 834 | * this page could not be removed from its unbuddied |
| 835 | * list while pool lock was held, and then we've taken |
| 836 | * page lock so kref_put could not be called before |
| 837 | * we got here, so it's safe to just call kref_get() |
| 838 | */ |
| 839 | kref_get(&zhdr->refcount); |
| 840 | break; |
| 841 | } |
Vitaly Wool | 135f97f | 2020-12-14 19:12:36 -0800 | [diff] [blame] | 842 | migrate_enable(); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 843 | |
Vitaly Wool | 351618b | 2019-05-13 17:22:46 -0700 | [diff] [blame] | 844 | if (!zhdr) { |
| 845 | int cpu; |
| 846 | |
| 847 | /* look for _exact_ match on other cpus' lists */ |
| 848 | for_each_online_cpu(cpu) { |
| 849 | struct list_head *l; |
| 850 | |
| 851 | unbuddied = per_cpu_ptr(pool->unbuddied, cpu); |
| 852 | spin_lock(&pool->lock); |
| 853 | l = &unbuddied[chunks]; |
| 854 | |
| 855 | zhdr = list_first_entry_or_null(READ_ONCE(l), |
| 856 | struct z3fold_header, buddy); |
| 857 | |
| 858 | if (!zhdr || !z3fold_page_trylock(zhdr)) { |
| 859 | spin_unlock(&pool->lock); |
| 860 | zhdr = NULL; |
| 861 | continue; |
| 862 | } |
| 863 | list_del_init(&zhdr->buddy); |
| 864 | zhdr->cpu = -1; |
| 865 | spin_unlock(&pool->lock); |
| 866 | |
| 867 | page = virt_to_page(zhdr); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 868 | if (test_bit(NEEDS_COMPACTING, &page->private) || |
| 869 | test_bit(PAGE_CLAIMED, &page->private)) { |
Vitaly Wool | 351618b | 2019-05-13 17:22:46 -0700 | [diff] [blame] | 870 | z3fold_page_unlock(zhdr); |
| 871 | zhdr = NULL; |
| 872 | if (can_sleep) |
| 873 | cond_resched(); |
| 874 | continue; |
| 875 | } |
| 876 | kref_get(&zhdr->refcount); |
| 877 | break; |
| 878 | } |
| 879 | } |
| 880 | |
Miaohe Lin | 7c61c35 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 881 | if (zhdr && !zhdr->slots) { |
Miaohe Lin | 4c6bdb3 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 882 | zhdr->slots = alloc_slots(pool, GFP_ATOMIC); |
Miaohe Lin | 7c61c35 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 883 | if (!zhdr->slots) |
| 884 | goto out_fail; |
| 885 | } |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 886 | return zhdr; |
Miaohe Lin | 7c61c35 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 887 | |
| 888 | out_fail: |
| 889 | if (!kref_put(&zhdr->refcount, release_z3fold_page_locked)) { |
| 890 | add_to_unbuddied(pool, zhdr); |
| 891 | z3fold_page_unlock(zhdr); |
| 892 | } |
| 893 | return NULL; |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 894 | } |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 895 | |
| 896 | /* |
| 897 | * API Functions |
| 898 | */ |
| 899 | |
| 900 | /** |
| 901 | * z3fold_create_pool() - create a new z3fold pool |
| 902 | * @name: pool name |
| 903 | * @gfp: gfp flags when allocating the z3fold pool structure |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 904 | * |
| 905 | * Return: pointer to the new z3fold pool or NULL if the metadata allocation |
| 906 | * failed. |
| 907 | */ |
Johannes Weiner | 6a05aa3 | 2022-11-28 11:16:11 -0800 | [diff] [blame] | 908 | static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp) |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 909 | { |
| 910 | struct z3fold_pool *pool = NULL; |
| 911 | int i, cpu; |
| 912 | |
| 913 | pool = kzalloc(sizeof(struct z3fold_pool), gfp); |
| 914 | if (!pool) |
| 915 | goto out; |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 916 | pool->c_handle = kmem_cache_create("z3fold_handle", |
| 917 | sizeof(struct z3fold_buddy_slots), |
| 918 | SLOTS_ALIGN, 0, NULL); |
| 919 | if (!pool->c_handle) |
| 920 | goto out_c; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 921 | spin_lock_init(&pool->lock); |
| 922 | spin_lock_init(&pool->stale_lock); |
Miaohe Lin | e891f60 | 2021-06-30 18:50:30 -0700 | [diff] [blame] | 923 | pool->unbuddied = __alloc_percpu(sizeof(struct list_head) * NCHUNKS, |
| 924 | __alignof__(struct list_head)); |
Xidong Wang | 1ec6995 | 2018-04-10 16:29:34 -0700 | [diff] [blame] | 925 | if (!pool->unbuddied) |
| 926 | goto out_pool; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 927 | for_each_possible_cpu(cpu) { |
| 928 | struct list_head *unbuddied = |
| 929 | per_cpu_ptr(pool->unbuddied, cpu); |
| 930 | for_each_unbuddied_list(i, 0) |
| 931 | INIT_LIST_HEAD(&unbuddied[i]); |
| 932 | } |
| 933 | INIT_LIST_HEAD(&pool->lru); |
| 934 | 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 Wang | 1ec6995 | 2018-04-10 16:29:34 -0700 | [diff] [blame] | 939 | goto out_unbuddied; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 940 | 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 Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 944 | return pool; |
| 945 | |
| 946 | out_wq: |
| 947 | destroy_workqueue(pool->compact_wq); |
Xidong Wang | 1ec6995 | 2018-04-10 16:29:34 -0700 | [diff] [blame] | 948 | out_unbuddied: |
| 949 | free_percpu(pool->unbuddied); |
| 950 | out_pool: |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 951 | kmem_cache_destroy(pool->c_handle); |
| 952 | out_c: |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 953 | kfree(pool); |
Xidong Wang | 1ec6995 | 2018-04-10 16:29:34 -0700 | [diff] [blame] | 954 | out: |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 955 | 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 | */ |
| 964 | static void z3fold_destroy_pool(struct z3fold_pool *pool) |
| 965 | { |
Vitaly Wool | 7c2b8ba | 2019-05-13 17:22:49 -0700 | [diff] [blame] | 966 | kmem_cache_destroy(pool->c_handle); |
Henry Burns | 6051d3b | 2019-08-13 15:37:21 -0700 | [diff] [blame] | 967 | |
| 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 Burns | b997052 | 2019-08-13 15:37:25 -0700 | [diff] [blame] | 972 | * |
| 973 | * There are still outstanding pages until both workqueues are drained, |
| 974 | * so we cannot unregister migration until then. |
Henry Burns | 6051d3b | 2019-08-13 15:37:21 -0700 | [diff] [blame] | 975 | */ |
| 976 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 977 | destroy_workqueue(pool->compact_wq); |
Henry Burns | 6051d3b | 2019-08-13 15:37:21 -0700 | [diff] [blame] | 978 | destroy_workqueue(pool->release_wq); |
Miaohe Lin | dac0d1c | 2021-06-30 18:50:36 -0700 | [diff] [blame] | 979 | free_percpu(pool->unbuddied); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 980 | kfree(pool); |
| 981 | } |
| 982 | |
Matthew Wilcox (Oracle) | 68f2736 | 2022-06-07 15:38:48 -0400 | [diff] [blame] | 983 | static const struct movable_operations z3fold_mops; |
| 984 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 985 | /** |
| 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 Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 997 | * 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 | */ |
| 1001 | static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp, |
| 1002 | unsigned long *handle) |
| 1003 | { |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 1004 | int chunks = size_to_chunks(size); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1005 | struct z3fold_header *zhdr = NULL; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1006 | struct page *page = NULL; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1007 | enum buddy bud; |
Matthew Wilcox | 8a97ea54 | 2018-04-10 16:29:37 -0700 | [diff] [blame] | 1008 | bool can_sleep = gfpflags_allow_blocking(gfp); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1009 | |
Miaohe Lin | f4bad64 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1010 | if (!size || (gfp & __GFP_HIGHMEM)) |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1011 | 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 Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 1019 | retry: |
| 1020 | zhdr = __z3fold_alloc(pool, size, can_sleep); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1021 | if (zhdr) { |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1022 | bud = get_free_buddy(zhdr, chunks); |
| 1023 | if (bud == HEADLESS) { |
Miaohe Lin | 5e36c25 | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 1024 | if (!kref_put(&zhdr->refcount, |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1025 | release_z3fold_page_locked)) |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1026 | z3fold_page_unlock(zhdr); |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1027 | pr_err("No free chunks in unbuddied\n"); |
| 1028 | WARN_ON(1); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 1029 | goto retry; |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1030 | } |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 1031 | page = virt_to_page(zhdr); |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1032 | goto found; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1033 | } |
| 1034 | bud = FIRST; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1035 | } |
| 1036 | |
Miaohe Lin | df6f0f1 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1037 | page = alloc_page(gfp); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1038 | if (!page) |
| 1039 | return -ENOMEM; |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1040 | |
Vitaly Wool | 6339841 | 2019-09-23 15:36:51 -0700 | [diff] [blame] | 1041 | zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp); |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 1042 | if (!zhdr) { |
| 1043 | __free_page(page); |
| 1044 | return -ENOMEM; |
| 1045 | } |
| 1046 | atomic64_inc(&pool->pages_nr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1047 | |
| 1048 | if (bud == HEADLESS) { |
| 1049 | set_bit(PAGE_HEADLESS, &page->private); |
| 1050 | goto headless; |
| 1051 | } |
Henry Burns | 810481a | 2019-07-11 20:52:14 -0700 | [diff] [blame] | 1052 | if (can_sleep) { |
| 1053 | lock_page(page); |
Matthew Wilcox (Oracle) | 68f2736 | 2022-06-07 15:38:48 -0400 | [diff] [blame] | 1054 | __SetPageMovable(page, &z3fold_mops); |
Henry Burns | 810481a | 2019-07-11 20:52:14 -0700 | [diff] [blame] | 1055 | unlock_page(page); |
| 1056 | } else { |
Miaohe Lin | 2c0f351 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1057 | WARN_ON(!trylock_page(page)); |
Matthew Wilcox (Oracle) | 68f2736 | 2022-06-07 15:38:48 -0400 | [diff] [blame] | 1058 | __SetPageMovable(page, &z3fold_mops); |
Miaohe Lin | 2c0f351 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1059 | unlock_page(page); |
Henry Burns | 810481a | 2019-07-11 20:52:14 -0700 | [diff] [blame] | 1060 | } |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1061 | z3fold_page_lock(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1062 | |
| 1063 | found: |
| 1064 | if (bud == FIRST) |
| 1065 | zhdr->first_chunks = chunks; |
| 1066 | else if (bud == LAST) |
| 1067 | zhdr->last_chunks = chunks; |
| 1068 | else { |
| 1069 | zhdr->middle_chunks = chunks; |
Vitaly Wool | ede9321 | 2017-02-24 14:57:17 -0800 | [diff] [blame] | 1070 | zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1071 | } |
Vitaly Wool | 9050cce | 2019-05-13 17:22:43 -0700 | [diff] [blame] | 1072 | add_to_unbuddied(pool, zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1073 | |
| 1074 | headless: |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1075 | spin_lock(&pool->lock); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1076 | /* Add/move z3fold page to beginning of LRU */ |
| 1077 | if (!list_empty(&page->lru)) |
| 1078 | list_del(&page->lru); |
| 1079 | |
| 1080 | list_add(&page->lru, &pool->lru); |
| 1081 | |
| 1082 | *handle = encode_handle(zhdr, bud); |
| 1083 | spin_unlock(&pool->lock); |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1084 | if (bud != HEADLESS) |
| 1085 | z3fold_page_unlock(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1086 | |
| 1087 | return 0; |
| 1088 | } |
| 1089 | |
| 1090 | /** |
| 1091 | * z3fold_free() - frees the allocation associated with the given handle |
| 1092 | * @pool: pool in which the allocation resided |
| 1093 | * @handle: handle associated with the allocation returned by z3fold_alloc() |
| 1094 | * |
| 1095 | * In the case that the z3fold page in which the allocation resides is under |
Miaohe Lin | ed0e5dc | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 1096 | * reclaim, as indicated by the PAGE_CLAIMED flag being set, this function |
| 1097 | * only sets the first|middle|last_chunks to 0. The page is actually freed |
| 1098 | * once all buddies are evicted (see z3fold_reclaim_page() below). |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1099 | */ |
| 1100 | static void z3fold_free(struct z3fold_pool *pool, unsigned long handle) |
| 1101 | { |
| 1102 | struct z3fold_header *zhdr; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1103 | struct page *page; |
| 1104 | enum buddy bud; |
Vitaly Wool | 5b6807d | 2019-10-06 17:58:22 -0700 | [diff] [blame] | 1105 | bool page_claimed; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1106 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1107 | zhdr = get_z3fold_header(handle); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1108 | page = virt_to_page(zhdr); |
Vitaly Wool | 5b6807d | 2019-10-06 17:58:22 -0700 | [diff] [blame] | 1109 | page_claimed = test_and_set_bit(PAGE_CLAIMED, &page->private); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1110 | |
| 1111 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1112 | /* if a headless page is under reclaim, just leave. |
| 1113 | * NB: we use test_and_set_bit for a reason: if the bit |
| 1114 | * has not been set before, we release this page |
| 1115 | * immediately so we don't care about its value any more. |
| 1116 | */ |
Vitaly Wool | 5b6807d | 2019-10-06 17:58:22 -0700 | [diff] [blame] | 1117 | if (!page_claimed) { |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1118 | spin_lock(&pool->lock); |
| 1119 | list_del(&page->lru); |
| 1120 | spin_unlock(&pool->lock); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1121 | put_z3fold_header(zhdr); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1122 | free_z3fold_page(page, true); |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1123 | atomic64_dec(&pool->pages_nr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1124 | } |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1125 | return; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1126 | } |
| 1127 | |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1128 | /* Non-headless case */ |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1129 | bud = handle_to_buddy(handle); |
| 1130 | |
| 1131 | switch (bud) { |
| 1132 | case FIRST: |
| 1133 | zhdr->first_chunks = 0; |
| 1134 | break; |
| 1135 | case MIDDLE: |
| 1136 | zhdr->middle_chunks = 0; |
| 1137 | break; |
| 1138 | case LAST: |
| 1139 | zhdr->last_chunks = 0; |
| 1140 | break; |
| 1141 | default: |
| 1142 | pr_err("%s: unknown bud %d\n", __func__, bud); |
| 1143 | WARN_ON(1); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1144 | put_z3fold_header(zhdr); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1145 | return; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1146 | } |
| 1147 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1148 | if (!page_claimed) |
Vitaly Wool | fc54886 | 2020-12-14 19:12:30 -0800 | [diff] [blame] | 1149 | free_handle(handle, zhdr); |
Miaohe Lin | 5e36c25 | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 1150 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1151 | return; |
Vitaly Wool | 5b6807d | 2019-10-06 17:58:22 -0700 | [diff] [blame] | 1152 | if (page_claimed) { |
| 1153 | /* the page has not been claimed by us */ |
Miaohe Lin | ed0e5dc | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 1154 | put_z3fold_header(zhdr); |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1155 | return; |
| 1156 | } |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1157 | if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) { |
Vitaly Wool | 5b6807d | 2019-10-06 17:58:22 -0700 | [diff] [blame] | 1158 | clear_bit(PAGE_CLAIMED, &page->private); |
Miaohe Lin | 4a1c383 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1159 | put_z3fold_header(zhdr); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1160 | return; |
| 1161 | } |
| 1162 | if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) { |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1163 | zhdr->cpu = -1; |
Vitaly Wool | 5d03a66 | 2017-11-17 15:26:16 -0800 | [diff] [blame] | 1164 | kref_get(&zhdr->refcount); |
Vitaly Wool | 5b6807d | 2019-10-06 17:58:22 -0700 | [diff] [blame] | 1165 | clear_bit(PAGE_CLAIMED, &page->private); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1166 | do_compact_page(zhdr, true); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1167 | return; |
| 1168 | } |
Vitaly Wool | 5d03a66 | 2017-11-17 15:26:16 -0800 | [diff] [blame] | 1169 | kref_get(&zhdr->refcount); |
Vitaly Wool | 5b6807d | 2019-10-06 17:58:22 -0700 | [diff] [blame] | 1170 | clear_bit(PAGE_CLAIMED, &page->private); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1171 | queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work); |
| 1172 | put_z3fold_header(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1173 | } |
| 1174 | |
| 1175 | /** |
| 1176 | * z3fold_reclaim_page() - evicts allocations from a pool page and frees it |
| 1177 | * @pool: pool from which a page will attempt to be evicted |
Mike Rapoport | f144c39 | 2018-02-06 15:42:16 -0800 | [diff] [blame] | 1178 | * @retries: number of pages on the LRU list for which eviction will |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1179 | * be attempted before failing |
| 1180 | * |
| 1181 | * z3fold reclaim is different from normal system reclaim in that it is done |
| 1182 | * from the bottom, up. This is because only the bottom layer, z3fold, has |
| 1183 | * information on how the allocations are organized within each z3fold page. |
| 1184 | * This has the potential to create interesting locking situations between |
| 1185 | * z3fold and the user, however. |
| 1186 | * |
| 1187 | * To avoid these, this is how z3fold_reclaim_page() should be called: |
Mike Rapoport | f144c39 | 2018-02-06 15:42:16 -0800 | [diff] [blame] | 1188 | * |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1189 | * The user detects a page should be reclaimed and calls z3fold_reclaim_page(). |
| 1190 | * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and |
| 1191 | * call the user-defined eviction handler with the pool and handle as |
| 1192 | * arguments. |
| 1193 | * |
| 1194 | * If the handle can not be evicted, the eviction handler should return |
| 1195 | * non-zero. z3fold_reclaim_page() will add the z3fold page back to the |
| 1196 | * appropriate list and try the next z3fold page on the LRU up to |
| 1197 | * a user defined number of retries. |
| 1198 | * |
| 1199 | * If the handle is successfully evicted, the eviction handler should |
| 1200 | * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free() |
| 1201 | * contains logic to delay freeing the page if the page is under reclaim, |
| 1202 | * as indicated by the setting of the PG_reclaim flag on the underlying page. |
| 1203 | * |
| 1204 | * If all buddies in the z3fold page are successfully evicted, then the |
| 1205 | * z3fold page can be freed. |
| 1206 | * |
| 1207 | * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are |
| 1208 | * no pages to evict or an eviction handler is not registered, -EAGAIN if |
| 1209 | * the retry limit was hit. |
| 1210 | */ |
| 1211 | static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries) |
| 1212 | { |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1213 | int i, ret = -1; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1214 | struct z3fold_header *zhdr = NULL; |
| 1215 | struct page *page = NULL; |
| 1216 | struct list_head *pos; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1217 | unsigned long first_handle = 0, middle_handle = 0, last_handle = 0; |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1218 | struct z3fold_buddy_slots slots __attribute__((aligned(SLOTS_ALIGN))); |
| 1219 | |
| 1220 | rwlock_init(&slots.lock); |
| 1221 | slots.pool = (unsigned long)pool | (1 << HANDLES_NOFREE); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1222 | |
| 1223 | spin_lock(&pool->lock); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1224 | for (i = 0; i < retries; i++) { |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1225 | if (list_empty(&pool->lru)) { |
| 1226 | spin_unlock(&pool->lock); |
| 1227 | return -EINVAL; |
| 1228 | } |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1229 | list_for_each_prev(pos, &pool->lru) { |
| 1230 | page = list_entry(pos, struct page, lru); |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1231 | |
Vitaly Wool | 3f9d2b5 | 2019-09-23 15:33:02 -0700 | [diff] [blame] | 1232 | zhdr = page_address(page); |
Thomas Hebb | 6d67957 | 2021-03-24 21:37:29 -0700 | [diff] [blame] | 1233 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
| 1234 | /* |
| 1235 | * For non-headless pages, we wait to do this |
| 1236 | * until we have the page lock to avoid racing |
| 1237 | * with __z3fold_alloc(). Headless pages don't |
| 1238 | * have a lock (and __z3fold_alloc() will never |
| 1239 | * see them), but we still need to test and set |
| 1240 | * PAGE_CLAIMED to avoid racing with |
| 1241 | * z3fold_free(), so just do it now before |
| 1242 | * leaving the loop. |
| 1243 | */ |
| 1244 | if (test_and_set_bit(PAGE_CLAIMED, &page->private)) |
| 1245 | continue; |
| 1246 | |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1247 | break; |
Thomas Hebb | 6d67957 | 2021-03-24 21:37:29 -0700 | [diff] [blame] | 1248 | } |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1249 | |
| 1250 | if (!z3fold_page_trylock(zhdr)) { |
| 1251 | zhdr = NULL; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1252 | continue; /* can't evict at this point */ |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1253 | } |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1254 | |
| 1255 | /* test_and_set_bit is of course atomic, but we still |
| 1256 | * need to do it under page lock, otherwise checking |
| 1257 | * that bit in __z3fold_alloc wouldn't make sense |
| 1258 | */ |
| 1259 | if (zhdr->foreign_handles || |
| 1260 | test_and_set_bit(PAGE_CLAIMED, &page->private)) { |
Miaohe Lin | 0409422 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1261 | z3fold_page_unlock(zhdr); |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1262 | zhdr = NULL; |
| 1263 | continue; /* can't evict such page */ |
| 1264 | } |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1265 | list_del_init(&zhdr->buddy); |
| 1266 | zhdr->cpu = -1; |
Miaohe Lin | 0409422 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1267 | /* See comment in __z3fold_alloc. */ |
| 1268 | kref_get(&zhdr->refcount); |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1269 | break; |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1270 | } |
| 1271 | |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1272 | if (!zhdr) |
| 1273 | break; |
| 1274 | |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1275 | list_del_init(&page->lru); |
| 1276 | spin_unlock(&pool->lock); |
| 1277 | |
| 1278 | if (!test_bit(PAGE_HEADLESS, &page->private)) { |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1279 | /* |
Vitaly Wool | 3f9d2b5 | 2019-09-23 15:33:02 -0700 | [diff] [blame] | 1280 | * We need encode the handles before unlocking, and |
| 1281 | * use our local slots structure because z3fold_free |
| 1282 | * can zero out zhdr->slots and we can't do much |
| 1283 | * about that |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1284 | */ |
| 1285 | first_handle = 0; |
| 1286 | last_handle = 0; |
| 1287 | middle_handle = 0; |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1288 | memset(slots.slot, 0, sizeof(slots.slot)); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1289 | if (zhdr->first_chunks) |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1290 | first_handle = __encode_handle(zhdr, &slots, |
| 1291 | FIRST); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1292 | if (zhdr->middle_chunks) |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1293 | middle_handle = __encode_handle(zhdr, &slots, |
| 1294 | MIDDLE); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1295 | if (zhdr->last_chunks) |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1296 | last_handle = __encode_handle(zhdr, &slots, |
| 1297 | LAST); |
Vitaly Wool | d30561c | 2017-09-06 16:24:47 -0700 | [diff] [blame] | 1298 | /* |
| 1299 | * it's safe to unlock here because we hold a |
| 1300 | * reference to this page |
| 1301 | */ |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1302 | z3fold_page_unlock(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1303 | } else { |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1304 | first_handle = encode_handle(zhdr, HEADLESS); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1305 | last_handle = middle_handle = 0; |
| 1306 | } |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1307 | /* Issue the eviction callback(s) */ |
| 1308 | if (middle_handle) { |
Johannes Weiner | 6a05aa3 | 2022-11-28 11:16:11 -0800 | [diff] [blame] | 1309 | ret = pool->zpool_ops->evict(pool->zpool, middle_handle); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1310 | if (ret) |
| 1311 | goto next; |
| 1312 | } |
| 1313 | if (first_handle) { |
Johannes Weiner | 6a05aa3 | 2022-11-28 11:16:11 -0800 | [diff] [blame] | 1314 | ret = pool->zpool_ops->evict(pool->zpool, first_handle); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1315 | if (ret) |
| 1316 | goto next; |
| 1317 | } |
| 1318 | if (last_handle) { |
Johannes Weiner | 6a05aa3 | 2022-11-28 11:16:11 -0800 | [diff] [blame] | 1319 | ret = pool->zpool_ops->evict(pool->zpool, last_handle); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1320 | if (ret) |
| 1321 | goto next; |
| 1322 | } |
| 1323 | next: |
Vitaly Wool | 5a27aa8 | 2017-02-24 14:57:26 -0800 | [diff] [blame] | 1324 | if (test_bit(PAGE_HEADLESS, &page->private)) { |
| 1325 | if (ret == 0) { |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1326 | free_z3fold_page(page, true); |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1327 | atomic64_dec(&pool->pages_nr); |
Vitaly Wool | 5a27aa8 | 2017-02-24 14:57:26 -0800 | [diff] [blame] | 1328 | return 0; |
Vitaly Wool | 5a27aa8 | 2017-02-24 14:57:26 -0800 | [diff] [blame] | 1329 | } |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1330 | spin_lock(&pool->lock); |
| 1331 | list_add(&page->lru, &pool->lru); |
Vitaly Wool | d5567c9 | 2017-10-03 16:14:47 -0700 | [diff] [blame] | 1332 | spin_unlock(&pool->lock); |
Vitaly Wool | 3f9d2b5 | 2019-09-23 15:33:02 -0700 | [diff] [blame] | 1333 | clear_bit(PAGE_CLAIMED, &page->private); |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1334 | } else { |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1335 | struct z3fold_buddy_slots *slots = zhdr->slots; |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1336 | z3fold_page_lock(zhdr); |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1337 | if (kref_put(&zhdr->refcount, |
| 1338 | release_z3fold_page_locked)) { |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1339 | kmem_cache_free(pool->c_handle, slots); |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1340 | return 0; |
| 1341 | } |
| 1342 | /* |
| 1343 | * if we are here, the page is still not completely |
| 1344 | * free. Take the global pool lock then to be able |
| 1345 | * to add it back to the lru list |
| 1346 | */ |
| 1347 | spin_lock(&pool->lock); |
| 1348 | list_add(&page->lru, &pool->lru); |
| 1349 | spin_unlock(&pool->lock); |
Miaohe Lin | 6cf9a34 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1350 | if (list_empty(&zhdr->buddy)) |
| 1351 | add_to_unbuddied(pool, zhdr); |
Vitaly Wool | 3f9d2b5 | 2019-09-23 15:33:02 -0700 | [diff] [blame] | 1352 | clear_bit(PAGE_CLAIMED, &page->private); |
Miaohe Lin | 4a1c383 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1353 | z3fold_page_unlock(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1354 | } |
| 1355 | |
Vitaly Wool | 6098d7e | 2018-05-11 16:01:46 -0700 | [diff] [blame] | 1356 | /* We started off locked to we need to lock the pool back */ |
| 1357 | spin_lock(&pool->lock); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1358 | } |
| 1359 | spin_unlock(&pool->lock); |
| 1360 | return -EAGAIN; |
| 1361 | } |
| 1362 | |
| 1363 | /** |
| 1364 | * z3fold_map() - maps the allocation associated with the given handle |
| 1365 | * @pool: pool in which the allocation resides |
| 1366 | * @handle: handle associated with the allocation to be mapped |
| 1367 | * |
| 1368 | * Extracts the buddy number from handle and constructs the pointer to the |
| 1369 | * correct starting chunk within the page. |
| 1370 | * |
| 1371 | * Returns: a pointer to the mapped allocation |
| 1372 | */ |
| 1373 | static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle) |
| 1374 | { |
| 1375 | struct z3fold_header *zhdr; |
| 1376 | struct page *page; |
| 1377 | void *addr; |
| 1378 | enum buddy buddy; |
| 1379 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1380 | zhdr = get_z3fold_header(handle); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1381 | addr = zhdr; |
| 1382 | page = virt_to_page(zhdr); |
| 1383 | |
| 1384 | if (test_bit(PAGE_HEADLESS, &page->private)) |
| 1385 | goto out; |
| 1386 | |
| 1387 | buddy = handle_to_buddy(handle); |
| 1388 | switch (buddy) { |
| 1389 | case FIRST: |
| 1390 | addr += ZHDR_SIZE_ALIGNED; |
| 1391 | break; |
| 1392 | case MIDDLE: |
| 1393 | addr += zhdr->start_middle << CHUNK_SHIFT; |
| 1394 | set_bit(MIDDLE_CHUNK_MAPPED, &page->private); |
| 1395 | break; |
| 1396 | case LAST: |
Vitaly Wool | ca0246b | 2018-11-16 15:07:56 -0800 | [diff] [blame] | 1397 | addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1398 | break; |
| 1399 | default: |
| 1400 | pr_err("unknown buddy id %d\n", buddy); |
| 1401 | WARN_ON(1); |
| 1402 | addr = NULL; |
| 1403 | break; |
| 1404 | } |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1405 | |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1406 | if (addr) |
| 1407 | zhdr->mapped_count++; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1408 | out: |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1409 | put_z3fold_header(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1410 | return addr; |
| 1411 | } |
| 1412 | |
| 1413 | /** |
| 1414 | * z3fold_unmap() - unmaps the allocation associated with the given handle |
| 1415 | * @pool: pool in which the allocation resides |
| 1416 | * @handle: handle associated with the allocation to be unmapped |
| 1417 | */ |
| 1418 | static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle) |
| 1419 | { |
| 1420 | struct z3fold_header *zhdr; |
| 1421 | struct page *page; |
| 1422 | enum buddy buddy; |
| 1423 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1424 | zhdr = get_z3fold_header(handle); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1425 | page = virt_to_page(zhdr); |
| 1426 | |
Vitaly Wool | 2f1e5e4 | 2017-02-24 14:57:23 -0800 | [diff] [blame] | 1427 | if (test_bit(PAGE_HEADLESS, &page->private)) |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1428 | return; |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1429 | |
| 1430 | buddy = handle_to_buddy(handle); |
| 1431 | if (buddy == MIDDLE) |
| 1432 | clear_bit(MIDDLE_CHUNK_MAPPED, &page->private); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1433 | zhdr->mapped_count--; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1434 | put_z3fold_header(zhdr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1435 | } |
| 1436 | |
| 1437 | /** |
| 1438 | * z3fold_get_pool_size() - gets the z3fold pool size in pages |
| 1439 | * @pool: pool whose size is being queried |
| 1440 | * |
Vitaly Wool | 12d59ae | 2017-02-24 14:57:15 -0800 | [diff] [blame] | 1441 | * Returns: size in pages of the given pool. |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1442 | */ |
| 1443 | static u64 z3fold_get_pool_size(struct z3fold_pool *pool) |
| 1444 | { |
Vitaly Wool | 12d59ae | 2017-02-24 14:57:15 -0800 | [diff] [blame] | 1445 | return atomic64_read(&pool->pages_nr); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1446 | } |
| 1447 | |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1448 | static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode) |
| 1449 | { |
| 1450 | struct z3fold_header *zhdr; |
| 1451 | struct z3fold_pool *pool; |
| 1452 | |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1453 | VM_BUG_ON_PAGE(PageIsolated(page), page); |
| 1454 | |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1455 | if (test_bit(PAGE_HEADLESS, &page->private)) |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1456 | return false; |
| 1457 | |
| 1458 | zhdr = page_address(page); |
| 1459 | z3fold_page_lock(zhdr); |
| 1460 | if (test_bit(NEEDS_COMPACTING, &page->private) || |
| 1461 | test_bit(PAGE_STALE, &page->private)) |
| 1462 | goto out; |
| 1463 | |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1464 | if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) |
| 1465 | goto out; |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1466 | |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1467 | if (test_and_set_bit(PAGE_CLAIMED, &page->private)) |
| 1468 | goto out; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1469 | pool = zhdr_to_pool(zhdr); |
| 1470 | spin_lock(&pool->lock); |
| 1471 | if (!list_empty(&zhdr->buddy)) |
| 1472 | list_del_init(&zhdr->buddy); |
| 1473 | if (!list_empty(&page->lru)) |
| 1474 | list_del_init(&page->lru); |
| 1475 | spin_unlock(&pool->lock); |
| 1476 | |
| 1477 | kref_get(&zhdr->refcount); |
| 1478 | z3fold_page_unlock(zhdr); |
| 1479 | return true; |
| 1480 | |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1481 | out: |
| 1482 | z3fold_page_unlock(zhdr); |
| 1483 | return false; |
| 1484 | } |
| 1485 | |
Matthew Wilcox (Oracle) | 68f2736 | 2022-06-07 15:38:48 -0400 | [diff] [blame] | 1486 | static int z3fold_page_migrate(struct page *newpage, struct page *page, |
| 1487 | enum migrate_mode mode) |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1488 | { |
| 1489 | struct z3fold_header *zhdr, *new_zhdr; |
| 1490 | struct z3fold_pool *pool; |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1491 | |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1492 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1493 | VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED, &page->private), page); |
Henry Burns | 810481a | 2019-07-11 20:52:14 -0700 | [diff] [blame] | 1494 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1495 | |
| 1496 | zhdr = page_address(page); |
| 1497 | pool = zhdr_to_pool(zhdr); |
| 1498 | |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1499 | if (!z3fold_page_trylock(zhdr)) |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1500 | return -EAGAIN; |
Vitaly Wool | 4a3ac93 | 2019-11-30 17:56:11 -0800 | [diff] [blame] | 1501 | if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) { |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1502 | clear_bit(PAGE_CLAIMED, &page->private); |
Miaohe Lin | 4a1c383 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1503 | z3fold_page_unlock(zhdr); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1504 | return -EBUSY; |
| 1505 | } |
Henry Burns | c92d2f3 | 2019-07-16 16:26:21 -0700 | [diff] [blame] | 1506 | if (work_pending(&zhdr->work)) { |
| 1507 | z3fold_page_unlock(zhdr); |
| 1508 | return -EAGAIN; |
| 1509 | } |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1510 | new_zhdr = page_address(newpage); |
| 1511 | memcpy(new_zhdr, zhdr, PAGE_SIZE); |
| 1512 | newpage->private = page->private; |
Miaohe Lin | 943fb61 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1513 | set_bit(PAGE_MIGRATED, &page->private); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1514 | z3fold_page_unlock(zhdr); |
| 1515 | spin_lock_init(&new_zhdr->page_lock); |
Henry Burns | c92d2f3 | 2019-07-16 16:26:21 -0700 | [diff] [blame] | 1516 | INIT_WORK(&new_zhdr->work, compact_page_work); |
| 1517 | /* |
| 1518 | * z3fold_page_isolate() ensures that new_zhdr->buddy is empty, |
| 1519 | * so we only have to reinitialize it. |
| 1520 | */ |
| 1521 | INIT_LIST_HEAD(&new_zhdr->buddy); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1522 | __ClearPageMovable(page); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1523 | |
| 1524 | get_page(newpage); |
| 1525 | z3fold_page_lock(new_zhdr); |
| 1526 | if (new_zhdr->first_chunks) |
| 1527 | encode_handle(new_zhdr, FIRST); |
| 1528 | if (new_zhdr->last_chunks) |
| 1529 | encode_handle(new_zhdr, LAST); |
| 1530 | if (new_zhdr->middle_chunks) |
| 1531 | encode_handle(new_zhdr, MIDDLE); |
| 1532 | set_bit(NEEDS_COMPACTING, &newpage->private); |
| 1533 | new_zhdr->cpu = smp_processor_id(); |
| 1534 | spin_lock(&pool->lock); |
| 1535 | list_add(&newpage->lru, &pool->lru); |
| 1536 | spin_unlock(&pool->lock); |
Matthew Wilcox (Oracle) | 68f2736 | 2022-06-07 15:38:48 -0400 | [diff] [blame] | 1537 | __SetPageMovable(newpage, &z3fold_mops); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1538 | z3fold_page_unlock(new_zhdr); |
| 1539 | |
| 1540 | queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work); |
| 1541 | |
Miaohe Lin | 943fb61 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1542 | /* PAGE_CLAIMED and PAGE_MIGRATED are cleared now. */ |
| 1543 | page->private = 0; |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1544 | put_page(page); |
| 1545 | return 0; |
| 1546 | } |
| 1547 | |
| 1548 | static void z3fold_page_putback(struct page *page) |
| 1549 | { |
| 1550 | struct z3fold_header *zhdr; |
| 1551 | struct z3fold_pool *pool; |
| 1552 | |
| 1553 | zhdr = page_address(page); |
| 1554 | pool = zhdr_to_pool(zhdr); |
| 1555 | |
| 1556 | z3fold_page_lock(zhdr); |
| 1557 | if (!list_empty(&zhdr->buddy)) |
| 1558 | list_del_init(&zhdr->buddy); |
| 1559 | INIT_LIST_HEAD(&page->lru); |
Miaohe Lin | 5e36c25 | 2022-04-28 23:16:05 -0700 | [diff] [blame] | 1560 | if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1561 | return; |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1562 | spin_lock(&pool->lock); |
| 1563 | list_add(&page->lru, &pool->lru); |
| 1564 | spin_unlock(&pool->lock); |
Miaohe Lin | 6cf9a34 | 2022-04-29 14:40:43 +0800 | [diff] [blame] | 1565 | if (list_empty(&zhdr->buddy)) |
| 1566 | add_to_unbuddied(pool, zhdr); |
Vitaly Wool | dcf5aed | 2020-12-14 19:12:33 -0800 | [diff] [blame] | 1567 | clear_bit(PAGE_CLAIMED, &page->private); |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1568 | z3fold_page_unlock(zhdr); |
| 1569 | } |
| 1570 | |
Matthew Wilcox (Oracle) | 68f2736 | 2022-06-07 15:38:48 -0400 | [diff] [blame] | 1571 | static const struct movable_operations z3fold_mops = { |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1572 | .isolate_page = z3fold_page_isolate, |
Matthew Wilcox (Oracle) | 68f2736 | 2022-06-07 15:38:48 -0400 | [diff] [blame] | 1573 | .migrate_page = z3fold_page_migrate, |
Vitaly Wool | 1f86298 | 2019-05-13 17:22:52 -0700 | [diff] [blame] | 1574 | .putback_page = z3fold_page_putback, |
| 1575 | }; |
| 1576 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1577 | /***************** |
| 1578 | * zpool |
| 1579 | ****************/ |
| 1580 | |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1581 | static void *z3fold_zpool_create(const char *name, gfp_t gfp, |
| 1582 | const struct zpool_ops *zpool_ops, |
| 1583 | struct zpool *zpool) |
| 1584 | { |
| 1585 | struct z3fold_pool *pool; |
| 1586 | |
Johannes Weiner | 6a05aa3 | 2022-11-28 11:16:11 -0800 | [diff] [blame] | 1587 | pool = z3fold_create_pool(name, gfp); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1588 | if (pool) { |
| 1589 | pool->zpool = zpool; |
| 1590 | pool->zpool_ops = zpool_ops; |
| 1591 | } |
| 1592 | return pool; |
| 1593 | } |
| 1594 | |
| 1595 | static void z3fold_zpool_destroy(void *pool) |
| 1596 | { |
| 1597 | z3fold_destroy_pool(pool); |
| 1598 | } |
| 1599 | |
| 1600 | static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp, |
| 1601 | unsigned long *handle) |
| 1602 | { |
| 1603 | return z3fold_alloc(pool, size, gfp, handle); |
| 1604 | } |
| 1605 | static void z3fold_zpool_free(void *pool, unsigned long handle) |
| 1606 | { |
| 1607 | z3fold_free(pool, handle); |
| 1608 | } |
| 1609 | |
| 1610 | static int z3fold_zpool_shrink(void *pool, unsigned int pages, |
| 1611 | unsigned int *reclaimed) |
| 1612 | { |
| 1613 | unsigned int total = 0; |
| 1614 | int ret = -EINVAL; |
| 1615 | |
| 1616 | while (total < pages) { |
| 1617 | ret = z3fold_reclaim_page(pool, 8); |
| 1618 | if (ret < 0) |
| 1619 | break; |
| 1620 | total++; |
| 1621 | } |
| 1622 | |
| 1623 | if (reclaimed) |
| 1624 | *reclaimed = total; |
| 1625 | |
| 1626 | return ret; |
| 1627 | } |
| 1628 | |
| 1629 | static void *z3fold_zpool_map(void *pool, unsigned long handle, |
| 1630 | enum zpool_mapmode mm) |
| 1631 | { |
| 1632 | return z3fold_map(pool, handle); |
| 1633 | } |
| 1634 | static void z3fold_zpool_unmap(void *pool, unsigned long handle) |
| 1635 | { |
| 1636 | z3fold_unmap(pool, handle); |
| 1637 | } |
| 1638 | |
| 1639 | static u64 z3fold_zpool_total_size(void *pool) |
| 1640 | { |
| 1641 | return z3fold_get_pool_size(pool) * PAGE_SIZE; |
| 1642 | } |
| 1643 | |
| 1644 | static struct zpool_driver z3fold_zpool_driver = { |
| 1645 | .type = "z3fold", |
Tian Tao | e818e82 | 2021-02-25 17:18:22 -0800 | [diff] [blame] | 1646 | .sleep_mapped = true, |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1647 | .owner = THIS_MODULE, |
| 1648 | .create = z3fold_zpool_create, |
| 1649 | .destroy = z3fold_zpool_destroy, |
| 1650 | .malloc = z3fold_zpool_malloc, |
| 1651 | .free = z3fold_zpool_free, |
| 1652 | .shrink = z3fold_zpool_shrink, |
| 1653 | .map = z3fold_zpool_map, |
| 1654 | .unmap = z3fold_zpool_unmap, |
| 1655 | .total_size = z3fold_zpool_total_size, |
| 1656 | }; |
| 1657 | |
| 1658 | MODULE_ALIAS("zpool-z3fold"); |
| 1659 | |
| 1660 | static int __init init_z3fold(void) |
| 1661 | { |
Miaohe Lin | 014284a | 2021-06-30 18:50:27 -0700 | [diff] [blame] | 1662 | /* |
| 1663 | * Make sure the z3fold header is not larger than the page size and |
| 1664 | * there has remaining spaces for its buddy. |
| 1665 | */ |
| 1666 | BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE - CHUNK_SIZE); |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1667 | zpool_register_driver(&z3fold_zpool_driver); |
| 1668 | |
| 1669 | return 0; |
| 1670 | } |
| 1671 | |
| 1672 | static void __exit exit_z3fold(void) |
| 1673 | { |
Vitaly Wool | 9a001fc | 2016-05-20 16:58:30 -0700 | [diff] [blame] | 1674 | zpool_unregister_driver(&z3fold_zpool_driver); |
| 1675 | } |
| 1676 | |
| 1677 | module_init(init_z3fold); |
| 1678 | module_exit(exit_z3fold); |
| 1679 | |
| 1680 | MODULE_LICENSE("GPL"); |
| 1681 | MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>"); |
| 1682 | MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages"); |