| 1 | /* |
| 2 | * linux/kernel/power/snapshot.c |
| 3 | * |
| 4 | * This file provides system snapshot/restore functionality for swsusp. |
| 5 | * |
| 6 | * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz> |
| 7 | * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl> |
| 8 | * |
| 9 | * This file is released under the GPLv2. |
| 10 | * |
| 11 | */ |
| 12 | |
| 13 | #include <linux/version.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <linux/mm.h> |
| 16 | #include <linux/suspend.h> |
| 17 | #include <linux/delay.h> |
| 18 | #include <linux/bitops.h> |
| 19 | #include <linux/spinlock.h> |
| 20 | #include <linux/kernel.h> |
| 21 | #include <linux/pm.h> |
| 22 | #include <linux/device.h> |
| 23 | #include <linux/init.h> |
| 24 | #include <linux/bootmem.h> |
| 25 | #include <linux/syscalls.h> |
| 26 | #include <linux/console.h> |
| 27 | #include <linux/highmem.h> |
| 28 | #include <linux/list.h> |
| 29 | #include <linux/slab.h> |
| 30 | #include <linux/compiler.h> |
| 31 | #include <linux/ktime.h> |
| 32 | |
| 33 | #include <asm/uaccess.h> |
| 34 | #include <asm/mmu_context.h> |
| 35 | #include <asm/pgtable.h> |
| 36 | #include <asm/tlbflush.h> |
| 37 | #include <asm/io.h> |
| 38 | |
| 39 | #include "power.h" |
| 40 | |
| 41 | #ifdef CONFIG_DEBUG_RODATA |
| 42 | static bool hibernate_restore_protection; |
| 43 | static bool hibernate_restore_protection_active; |
| 44 | |
| 45 | void enable_restore_image_protection(void) |
| 46 | { |
| 47 | hibernate_restore_protection = true; |
| 48 | } |
| 49 | |
| 50 | static inline void hibernate_restore_protection_begin(void) |
| 51 | { |
| 52 | hibernate_restore_protection_active = hibernate_restore_protection; |
| 53 | } |
| 54 | |
| 55 | static inline void hibernate_restore_protection_end(void) |
| 56 | { |
| 57 | hibernate_restore_protection_active = false; |
| 58 | } |
| 59 | |
| 60 | static inline void hibernate_restore_protect_page(void *page_address) |
| 61 | { |
| 62 | if (hibernate_restore_protection_active) |
| 63 | set_memory_ro((unsigned long)page_address, 1); |
| 64 | } |
| 65 | |
| 66 | static inline void hibernate_restore_unprotect_page(void *page_address) |
| 67 | { |
| 68 | if (hibernate_restore_protection_active) |
| 69 | set_memory_rw((unsigned long)page_address, 1); |
| 70 | } |
| 71 | #else |
| 72 | static inline void hibernate_restore_protection_begin(void) {} |
| 73 | static inline void hibernate_restore_protection_end(void) {} |
| 74 | static inline void hibernate_restore_protect_page(void *page_address) {} |
| 75 | static inline void hibernate_restore_unprotect_page(void *page_address) {} |
| 76 | #endif /* CONFIG_DEBUG_RODATA */ |
| 77 | |
| 78 | static int swsusp_page_is_free(struct page *); |
| 79 | static void swsusp_set_page_forbidden(struct page *); |
| 80 | static void swsusp_unset_page_forbidden(struct page *); |
| 81 | |
| 82 | /* |
| 83 | * Number of bytes to reserve for memory allocations made by device drivers |
| 84 | * from their ->freeze() and ->freeze_noirq() callbacks so that they don't |
| 85 | * cause image creation to fail (tunable via /sys/power/reserved_size). |
| 86 | */ |
| 87 | unsigned long reserved_size; |
| 88 | |
| 89 | void __init hibernate_reserved_size_init(void) |
| 90 | { |
| 91 | reserved_size = SPARE_PAGES * PAGE_SIZE; |
| 92 | } |
| 93 | |
| 94 | /* |
| 95 | * Preferred image size in bytes (tunable via /sys/power/image_size). |
| 96 | * When it is set to N, swsusp will do its best to ensure the image |
| 97 | * size will not exceed N bytes, but if that is impossible, it will |
| 98 | * try to create the smallest image possible. |
| 99 | */ |
| 100 | unsigned long image_size; |
| 101 | |
| 102 | void __init hibernate_image_size_init(void) |
| 103 | { |
| 104 | image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE; |
| 105 | } |
| 106 | |
| 107 | /* |
| 108 | * List of PBEs needed for restoring the pages that were allocated before |
| 109 | * the suspend and included in the suspend image, but have also been |
| 110 | * allocated by the "resume" kernel, so their contents cannot be written |
| 111 | * directly to their "original" page frames. |
| 112 | */ |
| 113 | struct pbe *restore_pblist; |
| 114 | |
| 115 | /* struct linked_page is used to build chains of pages */ |
| 116 | |
| 117 | #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *)) |
| 118 | |
| 119 | struct linked_page { |
| 120 | struct linked_page *next; |
| 121 | char data[LINKED_PAGE_DATA_SIZE]; |
| 122 | } __packed; |
| 123 | |
| 124 | /* |
| 125 | * List of "safe" pages (ie. pages that were not used by the image kernel |
| 126 | * before hibernation) that may be used as temporary storage for image kernel |
| 127 | * memory contents. |
| 128 | */ |
| 129 | static struct linked_page *safe_pages_list; |
| 130 | |
| 131 | /* Pointer to an auxiliary buffer (1 page) */ |
| 132 | static void *buffer; |
| 133 | |
| 134 | #define PG_ANY 0 |
| 135 | #define PG_SAFE 1 |
| 136 | #define PG_UNSAFE_CLEAR 1 |
| 137 | #define PG_UNSAFE_KEEP 0 |
| 138 | |
| 139 | static unsigned int allocated_unsafe_pages; |
| 140 | |
| 141 | /** |
| 142 | * get_image_page - Allocate a page for a hibernation image. |
| 143 | * @gfp_mask: GFP mask for the allocation. |
| 144 | * @safe_needed: Get pages that were not used before hibernation (restore only) |
| 145 | * |
| 146 | * During image restoration, for storing the PBE list and the image data, we can |
| 147 | * only use memory pages that do not conflict with the pages used before |
| 148 | * hibernation. The "unsafe" pages have PageNosaveFree set and we count them |
| 149 | * using allocated_unsafe_pages. |
| 150 | * |
| 151 | * Each allocated image page is marked as PageNosave and PageNosaveFree so that |
| 152 | * swsusp_free() can release it. |
| 153 | */ |
| 154 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) |
| 155 | { |
| 156 | void *res; |
| 157 | |
| 158 | res = (void *)get_zeroed_page(gfp_mask); |
| 159 | if (safe_needed) |
| 160 | while (res && swsusp_page_is_free(virt_to_page(res))) { |
| 161 | /* The page is unsafe, mark it for swsusp_free() */ |
| 162 | swsusp_set_page_forbidden(virt_to_page(res)); |
| 163 | allocated_unsafe_pages++; |
| 164 | res = (void *)get_zeroed_page(gfp_mask); |
| 165 | } |
| 166 | if (res) { |
| 167 | swsusp_set_page_forbidden(virt_to_page(res)); |
| 168 | swsusp_set_page_free(virt_to_page(res)); |
| 169 | } |
| 170 | return res; |
| 171 | } |
| 172 | |
| 173 | static void *__get_safe_page(gfp_t gfp_mask) |
| 174 | { |
| 175 | if (safe_pages_list) { |
| 176 | void *ret = safe_pages_list; |
| 177 | |
| 178 | safe_pages_list = safe_pages_list->next; |
| 179 | memset(ret, 0, PAGE_SIZE); |
| 180 | return ret; |
| 181 | } |
| 182 | return get_image_page(gfp_mask, PG_SAFE); |
| 183 | } |
| 184 | |
| 185 | unsigned long get_safe_page(gfp_t gfp_mask) |
| 186 | { |
| 187 | return (unsigned long)__get_safe_page(gfp_mask); |
| 188 | } |
| 189 | |
| 190 | static struct page *alloc_image_page(gfp_t gfp_mask) |
| 191 | { |
| 192 | struct page *page; |
| 193 | |
| 194 | page = alloc_page(gfp_mask); |
| 195 | if (page) { |
| 196 | swsusp_set_page_forbidden(page); |
| 197 | swsusp_set_page_free(page); |
| 198 | } |
| 199 | return page; |
| 200 | } |
| 201 | |
| 202 | static void recycle_safe_page(void *page_address) |
| 203 | { |
| 204 | struct linked_page *lp = page_address; |
| 205 | |
| 206 | lp->next = safe_pages_list; |
| 207 | safe_pages_list = lp; |
| 208 | } |
| 209 | |
| 210 | /** |
| 211 | * free_image_page - Free a page allocated for hibernation image. |
| 212 | * @addr: Address of the page to free. |
| 213 | * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page. |
| 214 | * |
| 215 | * The page to free should have been allocated by get_image_page() (page flags |
| 216 | * set by it are affected). |
| 217 | */ |
| 218 | static inline void free_image_page(void *addr, int clear_nosave_free) |
| 219 | { |
| 220 | struct page *page; |
| 221 | |
| 222 | BUG_ON(!virt_addr_valid(addr)); |
| 223 | |
| 224 | page = virt_to_page(addr); |
| 225 | |
| 226 | swsusp_unset_page_forbidden(page); |
| 227 | if (clear_nosave_free) |
| 228 | swsusp_unset_page_free(page); |
| 229 | |
| 230 | __free_page(page); |
| 231 | } |
| 232 | |
| 233 | static inline void free_list_of_pages(struct linked_page *list, |
| 234 | int clear_page_nosave) |
| 235 | { |
| 236 | while (list) { |
| 237 | struct linked_page *lp = list->next; |
| 238 | |
| 239 | free_image_page(list, clear_page_nosave); |
| 240 | list = lp; |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | /* |
| 245 | * struct chain_allocator is used for allocating small objects out of |
| 246 | * a linked list of pages called 'the chain'. |
| 247 | * |
| 248 | * The chain grows each time when there is no room for a new object in |
| 249 | * the current page. The allocated objects cannot be freed individually. |
| 250 | * It is only possible to free them all at once, by freeing the entire |
| 251 | * chain. |
| 252 | * |
| 253 | * NOTE: The chain allocator may be inefficient if the allocated objects |
| 254 | * are not much smaller than PAGE_SIZE. |
| 255 | */ |
| 256 | struct chain_allocator { |
| 257 | struct linked_page *chain; /* the chain */ |
| 258 | unsigned int used_space; /* total size of objects allocated out |
| 259 | of the current page */ |
| 260 | gfp_t gfp_mask; /* mask for allocating pages */ |
| 261 | int safe_needed; /* if set, only "safe" pages are allocated */ |
| 262 | }; |
| 263 | |
| 264 | static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask, |
| 265 | int safe_needed) |
| 266 | { |
| 267 | ca->chain = NULL; |
| 268 | ca->used_space = LINKED_PAGE_DATA_SIZE; |
| 269 | ca->gfp_mask = gfp_mask; |
| 270 | ca->safe_needed = safe_needed; |
| 271 | } |
| 272 | |
| 273 | static void *chain_alloc(struct chain_allocator *ca, unsigned int size) |
| 274 | { |
| 275 | void *ret; |
| 276 | |
| 277 | if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) { |
| 278 | struct linked_page *lp; |
| 279 | |
| 280 | lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) : |
| 281 | get_image_page(ca->gfp_mask, PG_ANY); |
| 282 | if (!lp) |
| 283 | return NULL; |
| 284 | |
| 285 | lp->next = ca->chain; |
| 286 | ca->chain = lp; |
| 287 | ca->used_space = 0; |
| 288 | } |
| 289 | ret = ca->chain->data + ca->used_space; |
| 290 | ca->used_space += size; |
| 291 | return ret; |
| 292 | } |
| 293 | |
| 294 | /** |
| 295 | * Data types related to memory bitmaps. |
| 296 | * |
| 297 | * Memory bitmap is a structure consiting of many linked lists of |
| 298 | * objects. The main list's elements are of type struct zone_bitmap |
| 299 | * and each of them corresonds to one zone. For each zone bitmap |
| 300 | * object there is a list of objects of type struct bm_block that |
| 301 | * represent each blocks of bitmap in which information is stored. |
| 302 | * |
| 303 | * struct memory_bitmap contains a pointer to the main list of zone |
| 304 | * bitmap objects, a struct bm_position used for browsing the bitmap, |
| 305 | * and a pointer to the list of pages used for allocating all of the |
| 306 | * zone bitmap objects and bitmap block objects. |
| 307 | * |
| 308 | * NOTE: It has to be possible to lay out the bitmap in memory |
| 309 | * using only allocations of order 0. Additionally, the bitmap is |
| 310 | * designed to work with arbitrary number of zones (this is over the |
| 311 | * top for now, but let's avoid making unnecessary assumptions ;-). |
| 312 | * |
| 313 | * struct zone_bitmap contains a pointer to a list of bitmap block |
| 314 | * objects and a pointer to the bitmap block object that has been |
| 315 | * most recently used for setting bits. Additionally, it contains the |
| 316 | * PFNs that correspond to the start and end of the represented zone. |
| 317 | * |
| 318 | * struct bm_block contains a pointer to the memory page in which |
| 319 | * information is stored (in the form of a block of bitmap) |
| 320 | * It also contains the pfns that correspond to the start and end of |
| 321 | * the represented memory area. |
| 322 | * |
| 323 | * The memory bitmap is organized as a radix tree to guarantee fast random |
| 324 | * access to the bits. There is one radix tree for each zone (as returned |
| 325 | * from create_mem_extents). |
| 326 | * |
| 327 | * One radix tree is represented by one struct mem_zone_bm_rtree. There are |
| 328 | * two linked lists for the nodes of the tree, one for the inner nodes and |
| 329 | * one for the leave nodes. The linked leave nodes are used for fast linear |
| 330 | * access of the memory bitmap. |
| 331 | * |
| 332 | * The struct rtree_node represents one node of the radix tree. |
| 333 | */ |
| 334 | |
| 335 | #define BM_END_OF_MAP (~0UL) |
| 336 | |
| 337 | #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE) |
| 338 | #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3) |
| 339 | #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1) |
| 340 | |
| 341 | /* |
| 342 | * struct rtree_node is a wrapper struct to link the nodes |
| 343 | * of the rtree together for easy linear iteration over |
| 344 | * bits and easy freeing |
| 345 | */ |
| 346 | struct rtree_node { |
| 347 | struct list_head list; |
| 348 | unsigned long *data; |
| 349 | }; |
| 350 | |
| 351 | /* |
| 352 | * struct mem_zone_bm_rtree represents a bitmap used for one |
| 353 | * populated memory zone. |
| 354 | */ |
| 355 | struct mem_zone_bm_rtree { |
| 356 | struct list_head list; /* Link Zones together */ |
| 357 | struct list_head nodes; /* Radix Tree inner nodes */ |
| 358 | struct list_head leaves; /* Radix Tree leaves */ |
| 359 | unsigned long start_pfn; /* Zone start page frame */ |
| 360 | unsigned long end_pfn; /* Zone end page frame + 1 */ |
| 361 | struct rtree_node *rtree; /* Radix Tree Root */ |
| 362 | int levels; /* Number of Radix Tree Levels */ |
| 363 | unsigned int blocks; /* Number of Bitmap Blocks */ |
| 364 | }; |
| 365 | |
| 366 | /* strcut bm_position is used for browsing memory bitmaps */ |
| 367 | |
| 368 | struct bm_position { |
| 369 | struct mem_zone_bm_rtree *zone; |
| 370 | struct rtree_node *node; |
| 371 | unsigned long node_pfn; |
| 372 | int node_bit; |
| 373 | }; |
| 374 | |
| 375 | struct memory_bitmap { |
| 376 | struct list_head zones; |
| 377 | struct linked_page *p_list; /* list of pages used to store zone |
| 378 | bitmap objects and bitmap block |
| 379 | objects */ |
| 380 | struct bm_position cur; /* most recently used bit position */ |
| 381 | }; |
| 382 | |
| 383 | /* Functions that operate on memory bitmaps */ |
| 384 | |
| 385 | #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long)) |
| 386 | #if BITS_PER_LONG == 32 |
| 387 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2) |
| 388 | #else |
| 389 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3) |
| 390 | #endif |
| 391 | #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1) |
| 392 | |
| 393 | /** |
| 394 | * alloc_rtree_node - Allocate a new node and add it to the radix tree. |
| 395 | * |
| 396 | * This function is used to allocate inner nodes as well as the |
| 397 | * leave nodes of the radix tree. It also adds the node to the |
| 398 | * corresponding linked list passed in by the *list parameter. |
| 399 | */ |
| 400 | static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed, |
| 401 | struct chain_allocator *ca, |
| 402 | struct list_head *list) |
| 403 | { |
| 404 | struct rtree_node *node; |
| 405 | |
| 406 | node = chain_alloc(ca, sizeof(struct rtree_node)); |
| 407 | if (!node) |
| 408 | return NULL; |
| 409 | |
| 410 | node->data = get_image_page(gfp_mask, safe_needed); |
| 411 | if (!node->data) |
| 412 | return NULL; |
| 413 | |
| 414 | list_add_tail(&node->list, list); |
| 415 | |
| 416 | return node; |
| 417 | } |
| 418 | |
| 419 | /** |
| 420 | * add_rtree_block - Add a new leave node to the radix tree. |
| 421 | * |
| 422 | * The leave nodes need to be allocated in order to keep the leaves |
| 423 | * linked list in order. This is guaranteed by the zone->blocks |
| 424 | * counter. |
| 425 | */ |
| 426 | static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask, |
| 427 | int safe_needed, struct chain_allocator *ca) |
| 428 | { |
| 429 | struct rtree_node *node, *block, **dst; |
| 430 | unsigned int levels_needed, block_nr; |
| 431 | int i; |
| 432 | |
| 433 | block_nr = zone->blocks; |
| 434 | levels_needed = 0; |
| 435 | |
| 436 | /* How many levels do we need for this block nr? */ |
| 437 | while (block_nr) { |
| 438 | levels_needed += 1; |
| 439 | block_nr >>= BM_RTREE_LEVEL_SHIFT; |
| 440 | } |
| 441 | |
| 442 | /* Make sure the rtree has enough levels */ |
| 443 | for (i = zone->levels; i < levels_needed; i++) { |
| 444 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, |
| 445 | &zone->nodes); |
| 446 | if (!node) |
| 447 | return -ENOMEM; |
| 448 | |
| 449 | node->data[0] = (unsigned long)zone->rtree; |
| 450 | zone->rtree = node; |
| 451 | zone->levels += 1; |
| 452 | } |
| 453 | |
| 454 | /* Allocate new block */ |
| 455 | block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves); |
| 456 | if (!block) |
| 457 | return -ENOMEM; |
| 458 | |
| 459 | /* Now walk the rtree to insert the block */ |
| 460 | node = zone->rtree; |
| 461 | dst = &zone->rtree; |
| 462 | block_nr = zone->blocks; |
| 463 | for (i = zone->levels; i > 0; i--) { |
| 464 | int index; |
| 465 | |
| 466 | if (!node) { |
| 467 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, |
| 468 | &zone->nodes); |
| 469 | if (!node) |
| 470 | return -ENOMEM; |
| 471 | *dst = node; |
| 472 | } |
| 473 | |
| 474 | index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT); |
| 475 | index &= BM_RTREE_LEVEL_MASK; |
| 476 | dst = (struct rtree_node **)&((*dst)->data[index]); |
| 477 | node = *dst; |
| 478 | } |
| 479 | |
| 480 | zone->blocks += 1; |
| 481 | *dst = block; |
| 482 | |
| 483 | return 0; |
| 484 | } |
| 485 | |
| 486 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, |
| 487 | int clear_nosave_free); |
| 488 | |
| 489 | /** |
| 490 | * create_zone_bm_rtree - Create a radix tree for one zone. |
| 491 | * |
| 492 | * Allocated the mem_zone_bm_rtree structure and initializes it. |
| 493 | * This function also allocated and builds the radix tree for the |
| 494 | * zone. |
| 495 | */ |
| 496 | static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask, |
| 497 | int safe_needed, |
| 498 | struct chain_allocator *ca, |
| 499 | unsigned long start, |
| 500 | unsigned long end) |
| 501 | { |
| 502 | struct mem_zone_bm_rtree *zone; |
| 503 | unsigned int i, nr_blocks; |
| 504 | unsigned long pages; |
| 505 | |
| 506 | pages = end - start; |
| 507 | zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree)); |
| 508 | if (!zone) |
| 509 | return NULL; |
| 510 | |
| 511 | INIT_LIST_HEAD(&zone->nodes); |
| 512 | INIT_LIST_HEAD(&zone->leaves); |
| 513 | zone->start_pfn = start; |
| 514 | zone->end_pfn = end; |
| 515 | nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); |
| 516 | |
| 517 | for (i = 0; i < nr_blocks; i++) { |
| 518 | if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) { |
| 519 | free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR); |
| 520 | return NULL; |
| 521 | } |
| 522 | } |
| 523 | |
| 524 | return zone; |
| 525 | } |
| 526 | |
| 527 | /** |
| 528 | * free_zone_bm_rtree - Free the memory of the radix tree. |
| 529 | * |
| 530 | * Free all node pages of the radix tree. The mem_zone_bm_rtree |
| 531 | * structure itself is not freed here nor are the rtree_node |
| 532 | * structs. |
| 533 | */ |
| 534 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, |
| 535 | int clear_nosave_free) |
| 536 | { |
| 537 | struct rtree_node *node; |
| 538 | |
| 539 | list_for_each_entry(node, &zone->nodes, list) |
| 540 | free_image_page(node->data, clear_nosave_free); |
| 541 | |
| 542 | list_for_each_entry(node, &zone->leaves, list) |
| 543 | free_image_page(node->data, clear_nosave_free); |
| 544 | } |
| 545 | |
| 546 | static void memory_bm_position_reset(struct memory_bitmap *bm) |
| 547 | { |
| 548 | bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree, |
| 549 | list); |
| 550 | bm->cur.node = list_entry(bm->cur.zone->leaves.next, |
| 551 | struct rtree_node, list); |
| 552 | bm->cur.node_pfn = 0; |
| 553 | bm->cur.node_bit = 0; |
| 554 | } |
| 555 | |
| 556 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); |
| 557 | |
| 558 | struct mem_extent { |
| 559 | struct list_head hook; |
| 560 | unsigned long start; |
| 561 | unsigned long end; |
| 562 | }; |
| 563 | |
| 564 | /** |
| 565 | * free_mem_extents - Free a list of memory extents. |
| 566 | * @list: List of extents to free. |
| 567 | */ |
| 568 | static void free_mem_extents(struct list_head *list) |
| 569 | { |
| 570 | struct mem_extent *ext, *aux; |
| 571 | |
| 572 | list_for_each_entry_safe(ext, aux, list, hook) { |
| 573 | list_del(&ext->hook); |
| 574 | kfree(ext); |
| 575 | } |
| 576 | } |
| 577 | |
| 578 | /** |
| 579 | * create_mem_extents - Create a list of memory extents. |
| 580 | * @list: List to put the extents into. |
| 581 | * @gfp_mask: Mask to use for memory allocations. |
| 582 | * |
| 583 | * The extents represent contiguous ranges of PFNs. |
| 584 | */ |
| 585 | static int create_mem_extents(struct list_head *list, gfp_t gfp_mask) |
| 586 | { |
| 587 | struct zone *zone; |
| 588 | |
| 589 | INIT_LIST_HEAD(list); |
| 590 | |
| 591 | for_each_populated_zone(zone) { |
| 592 | unsigned long zone_start, zone_end; |
| 593 | struct mem_extent *ext, *cur, *aux; |
| 594 | |
| 595 | zone_start = zone->zone_start_pfn; |
| 596 | zone_end = zone_end_pfn(zone); |
| 597 | |
| 598 | list_for_each_entry(ext, list, hook) |
| 599 | if (zone_start <= ext->end) |
| 600 | break; |
| 601 | |
| 602 | if (&ext->hook == list || zone_end < ext->start) { |
| 603 | /* New extent is necessary */ |
| 604 | struct mem_extent *new_ext; |
| 605 | |
| 606 | new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask); |
| 607 | if (!new_ext) { |
| 608 | free_mem_extents(list); |
| 609 | return -ENOMEM; |
| 610 | } |
| 611 | new_ext->start = zone_start; |
| 612 | new_ext->end = zone_end; |
| 613 | list_add_tail(&new_ext->hook, &ext->hook); |
| 614 | continue; |
| 615 | } |
| 616 | |
| 617 | /* Merge this zone's range of PFNs with the existing one */ |
| 618 | if (zone_start < ext->start) |
| 619 | ext->start = zone_start; |
| 620 | if (zone_end > ext->end) |
| 621 | ext->end = zone_end; |
| 622 | |
| 623 | /* More merging may be possible */ |
| 624 | cur = ext; |
| 625 | list_for_each_entry_safe_continue(cur, aux, list, hook) { |
| 626 | if (zone_end < cur->start) |
| 627 | break; |
| 628 | if (zone_end < cur->end) |
| 629 | ext->end = cur->end; |
| 630 | list_del(&cur->hook); |
| 631 | kfree(cur); |
| 632 | } |
| 633 | } |
| 634 | |
| 635 | return 0; |
| 636 | } |
| 637 | |
| 638 | /** |
| 639 | * memory_bm_create - Allocate memory for a memory bitmap. |
| 640 | */ |
| 641 | static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, |
| 642 | int safe_needed) |
| 643 | { |
| 644 | struct chain_allocator ca; |
| 645 | struct list_head mem_extents; |
| 646 | struct mem_extent *ext; |
| 647 | int error; |
| 648 | |
| 649 | chain_init(&ca, gfp_mask, safe_needed); |
| 650 | INIT_LIST_HEAD(&bm->zones); |
| 651 | |
| 652 | error = create_mem_extents(&mem_extents, gfp_mask); |
| 653 | if (error) |
| 654 | return error; |
| 655 | |
| 656 | list_for_each_entry(ext, &mem_extents, hook) { |
| 657 | struct mem_zone_bm_rtree *zone; |
| 658 | |
| 659 | zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca, |
| 660 | ext->start, ext->end); |
| 661 | if (!zone) { |
| 662 | error = -ENOMEM; |
| 663 | goto Error; |
| 664 | } |
| 665 | list_add_tail(&zone->list, &bm->zones); |
| 666 | } |
| 667 | |
| 668 | bm->p_list = ca.chain; |
| 669 | memory_bm_position_reset(bm); |
| 670 | Exit: |
| 671 | free_mem_extents(&mem_extents); |
| 672 | return error; |
| 673 | |
| 674 | Error: |
| 675 | bm->p_list = ca.chain; |
| 676 | memory_bm_free(bm, PG_UNSAFE_CLEAR); |
| 677 | goto Exit; |
| 678 | } |
| 679 | |
| 680 | /** |
| 681 | * memory_bm_free - Free memory occupied by the memory bitmap. |
| 682 | * @bm: Memory bitmap. |
| 683 | */ |
| 684 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) |
| 685 | { |
| 686 | struct mem_zone_bm_rtree *zone; |
| 687 | |
| 688 | list_for_each_entry(zone, &bm->zones, list) |
| 689 | free_zone_bm_rtree(zone, clear_nosave_free); |
| 690 | |
| 691 | free_list_of_pages(bm->p_list, clear_nosave_free); |
| 692 | |
| 693 | INIT_LIST_HEAD(&bm->zones); |
| 694 | } |
| 695 | |
| 696 | /** |
| 697 | * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap. |
| 698 | * |
| 699 | * Find the bit in memory bitmap @bm that corresponds to the given PFN. |
| 700 | * The cur.zone, cur.block and cur.node_pfn members of @bm are updated. |
| 701 | * |
| 702 | * Walk the radix tree to find the page containing the bit that represents @pfn |
| 703 | * and return the position of the bit in @addr and @bit_nr. |
| 704 | */ |
| 705 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
| 706 | void **addr, unsigned int *bit_nr) |
| 707 | { |
| 708 | struct mem_zone_bm_rtree *curr, *zone; |
| 709 | struct rtree_node *node; |
| 710 | int i, block_nr; |
| 711 | |
| 712 | zone = bm->cur.zone; |
| 713 | |
| 714 | if (pfn >= zone->start_pfn && pfn < zone->end_pfn) |
| 715 | goto zone_found; |
| 716 | |
| 717 | zone = NULL; |
| 718 | |
| 719 | /* Find the right zone */ |
| 720 | list_for_each_entry(curr, &bm->zones, list) { |
| 721 | if (pfn >= curr->start_pfn && pfn < curr->end_pfn) { |
| 722 | zone = curr; |
| 723 | break; |
| 724 | } |
| 725 | } |
| 726 | |
| 727 | if (!zone) |
| 728 | return -EFAULT; |
| 729 | |
| 730 | zone_found: |
| 731 | /* |
| 732 | * We have found the zone. Now walk the radix tree to find the leaf node |
| 733 | * for our PFN. |
| 734 | */ |
| 735 | node = bm->cur.node; |
| 736 | if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn) |
| 737 | goto node_found; |
| 738 | |
| 739 | node = zone->rtree; |
| 740 | block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT; |
| 741 | |
| 742 | for (i = zone->levels; i > 0; i--) { |
| 743 | int index; |
| 744 | |
| 745 | index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT); |
| 746 | index &= BM_RTREE_LEVEL_MASK; |
| 747 | BUG_ON(node->data[index] == 0); |
| 748 | node = (struct rtree_node *)node->data[index]; |
| 749 | } |
| 750 | |
| 751 | node_found: |
| 752 | /* Update last position */ |
| 753 | bm->cur.zone = zone; |
| 754 | bm->cur.node = node; |
| 755 | bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK; |
| 756 | |
| 757 | /* Set return values */ |
| 758 | *addr = node->data; |
| 759 | *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK; |
| 760 | |
| 761 | return 0; |
| 762 | } |
| 763 | |
| 764 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) |
| 765 | { |
| 766 | void *addr; |
| 767 | unsigned int bit; |
| 768 | int error; |
| 769 | |
| 770 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
| 771 | BUG_ON(error); |
| 772 | set_bit(bit, addr); |
| 773 | } |
| 774 | |
| 775 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) |
| 776 | { |
| 777 | void *addr; |
| 778 | unsigned int bit; |
| 779 | int error; |
| 780 | |
| 781 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
| 782 | if (!error) |
| 783 | set_bit(bit, addr); |
| 784 | |
| 785 | return error; |
| 786 | } |
| 787 | |
| 788 | static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn) |
| 789 | { |
| 790 | void *addr; |
| 791 | unsigned int bit; |
| 792 | int error; |
| 793 | |
| 794 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
| 795 | BUG_ON(error); |
| 796 | clear_bit(bit, addr); |
| 797 | } |
| 798 | |
| 799 | static void memory_bm_clear_current(struct memory_bitmap *bm) |
| 800 | { |
| 801 | int bit; |
| 802 | |
| 803 | bit = max(bm->cur.node_bit - 1, 0); |
| 804 | clear_bit(bit, bm->cur.node->data); |
| 805 | } |
| 806 | |
| 807 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) |
| 808 | { |
| 809 | void *addr; |
| 810 | unsigned int bit; |
| 811 | int error; |
| 812 | |
| 813 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
| 814 | BUG_ON(error); |
| 815 | return test_bit(bit, addr); |
| 816 | } |
| 817 | |
| 818 | static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn) |
| 819 | { |
| 820 | void *addr; |
| 821 | unsigned int bit; |
| 822 | |
| 823 | return !memory_bm_find_bit(bm, pfn, &addr, &bit); |
| 824 | } |
| 825 | |
| 826 | /* |
| 827 | * rtree_next_node - Jump to the next leaf node. |
| 828 | * |
| 829 | * Set the position to the beginning of the next node in the |
| 830 | * memory bitmap. This is either the next node in the current |
| 831 | * zone's radix tree or the first node in the radix tree of the |
| 832 | * next zone. |
| 833 | * |
| 834 | * Return true if there is a next node, false otherwise. |
| 835 | */ |
| 836 | static bool rtree_next_node(struct memory_bitmap *bm) |
| 837 | { |
| 838 | if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) { |
| 839 | bm->cur.node = list_entry(bm->cur.node->list.next, |
| 840 | struct rtree_node, list); |
| 841 | bm->cur.node_pfn += BM_BITS_PER_BLOCK; |
| 842 | bm->cur.node_bit = 0; |
| 843 | touch_softlockup_watchdog(); |
| 844 | return true; |
| 845 | } |
| 846 | |
| 847 | /* No more nodes, goto next zone */ |
| 848 | if (!list_is_last(&bm->cur.zone->list, &bm->zones)) { |
| 849 | bm->cur.zone = list_entry(bm->cur.zone->list.next, |
| 850 | struct mem_zone_bm_rtree, list); |
| 851 | bm->cur.node = list_entry(bm->cur.zone->leaves.next, |
| 852 | struct rtree_node, list); |
| 853 | bm->cur.node_pfn = 0; |
| 854 | bm->cur.node_bit = 0; |
| 855 | return true; |
| 856 | } |
| 857 | |
| 858 | /* No more zones */ |
| 859 | return false; |
| 860 | } |
| 861 | |
| 862 | /** |
| 863 | * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap. |
| 864 | * @bm: Memory bitmap. |
| 865 | * |
| 866 | * Starting from the last returned position this function searches for the next |
| 867 | * set bit in @bm and returns the PFN represented by it. If no more bits are |
| 868 | * set, BM_END_OF_MAP is returned. |
| 869 | * |
| 870 | * It is required to run memory_bm_position_reset() before the first call to |
| 871 | * this function for the given memory bitmap. |
| 872 | */ |
| 873 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) |
| 874 | { |
| 875 | unsigned long bits, pfn, pages; |
| 876 | int bit; |
| 877 | |
| 878 | do { |
| 879 | pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn; |
| 880 | bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK); |
| 881 | bit = find_next_bit(bm->cur.node->data, bits, |
| 882 | bm->cur.node_bit); |
| 883 | if (bit < bits) { |
| 884 | pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit; |
| 885 | bm->cur.node_bit = bit + 1; |
| 886 | return pfn; |
| 887 | } |
| 888 | } while (rtree_next_node(bm)); |
| 889 | |
| 890 | return BM_END_OF_MAP; |
| 891 | } |
| 892 | |
| 893 | /* |
| 894 | * This structure represents a range of page frames the contents of which |
| 895 | * should not be saved during hibernation. |
| 896 | */ |
| 897 | struct nosave_region { |
| 898 | struct list_head list; |
| 899 | unsigned long start_pfn; |
| 900 | unsigned long end_pfn; |
| 901 | }; |
| 902 | |
| 903 | static LIST_HEAD(nosave_regions); |
| 904 | |
| 905 | static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone) |
| 906 | { |
| 907 | struct rtree_node *node; |
| 908 | |
| 909 | list_for_each_entry(node, &zone->nodes, list) |
| 910 | recycle_safe_page(node->data); |
| 911 | |
| 912 | list_for_each_entry(node, &zone->leaves, list) |
| 913 | recycle_safe_page(node->data); |
| 914 | } |
| 915 | |
| 916 | static void memory_bm_recycle(struct memory_bitmap *bm) |
| 917 | { |
| 918 | struct mem_zone_bm_rtree *zone; |
| 919 | struct linked_page *p_list; |
| 920 | |
| 921 | list_for_each_entry(zone, &bm->zones, list) |
| 922 | recycle_zone_bm_rtree(zone); |
| 923 | |
| 924 | p_list = bm->p_list; |
| 925 | while (p_list) { |
| 926 | struct linked_page *lp = p_list; |
| 927 | |
| 928 | p_list = lp->next; |
| 929 | recycle_safe_page(lp); |
| 930 | } |
| 931 | } |
| 932 | |
| 933 | /** |
| 934 | * register_nosave_region - Register a region of unsaveable memory. |
| 935 | * |
| 936 | * Register a range of page frames the contents of which should not be saved |
| 937 | * during hibernation (to be used in the early initialization code). |
| 938 | */ |
| 939 | void __init __register_nosave_region(unsigned long start_pfn, |
| 940 | unsigned long end_pfn, int use_kmalloc) |
| 941 | { |
| 942 | struct nosave_region *region; |
| 943 | |
| 944 | if (start_pfn >= end_pfn) |
| 945 | return; |
| 946 | |
| 947 | if (!list_empty(&nosave_regions)) { |
| 948 | /* Try to extend the previous region (they should be sorted) */ |
| 949 | region = list_entry(nosave_regions.prev, |
| 950 | struct nosave_region, list); |
| 951 | if (region->end_pfn == start_pfn) { |
| 952 | region->end_pfn = end_pfn; |
| 953 | goto Report; |
| 954 | } |
| 955 | } |
| 956 | if (use_kmalloc) { |
| 957 | /* During init, this shouldn't fail */ |
| 958 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); |
| 959 | BUG_ON(!region); |
| 960 | } else { |
| 961 | /* This allocation cannot fail */ |
| 962 | region = memblock_virt_alloc(sizeof(struct nosave_region), 0); |
| 963 | } |
| 964 | region->start_pfn = start_pfn; |
| 965 | region->end_pfn = end_pfn; |
| 966 | list_add_tail(®ion->list, &nosave_regions); |
| 967 | Report: |
| 968 | printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n", |
| 969 | (unsigned long long) start_pfn << PAGE_SHIFT, |
| 970 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); |
| 971 | } |
| 972 | |
| 973 | /* |
| 974 | * Set bits in this map correspond to the page frames the contents of which |
| 975 | * should not be saved during the suspend. |
| 976 | */ |
| 977 | static struct memory_bitmap *forbidden_pages_map; |
| 978 | |
| 979 | /* Set bits in this map correspond to free page frames. */ |
| 980 | static struct memory_bitmap *free_pages_map; |
| 981 | |
| 982 | /* |
| 983 | * Each page frame allocated for creating the image is marked by setting the |
| 984 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously |
| 985 | */ |
| 986 | |
| 987 | void swsusp_set_page_free(struct page *page) |
| 988 | { |
| 989 | if (free_pages_map) |
| 990 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); |
| 991 | } |
| 992 | |
| 993 | static int swsusp_page_is_free(struct page *page) |
| 994 | { |
| 995 | return free_pages_map ? |
| 996 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; |
| 997 | } |
| 998 | |
| 999 | void swsusp_unset_page_free(struct page *page) |
| 1000 | { |
| 1001 | if (free_pages_map) |
| 1002 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); |
| 1003 | } |
| 1004 | |
| 1005 | static void swsusp_set_page_forbidden(struct page *page) |
| 1006 | { |
| 1007 | if (forbidden_pages_map) |
| 1008 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); |
| 1009 | } |
| 1010 | |
| 1011 | int swsusp_page_is_forbidden(struct page *page) |
| 1012 | { |
| 1013 | return forbidden_pages_map ? |
| 1014 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; |
| 1015 | } |
| 1016 | |
| 1017 | static void swsusp_unset_page_forbidden(struct page *page) |
| 1018 | { |
| 1019 | if (forbidden_pages_map) |
| 1020 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); |
| 1021 | } |
| 1022 | |
| 1023 | /** |
| 1024 | * mark_nosave_pages - Mark pages that should not be saved. |
| 1025 | * @bm: Memory bitmap. |
| 1026 | * |
| 1027 | * Set the bits in @bm that correspond to the page frames the contents of which |
| 1028 | * should not be saved. |
| 1029 | */ |
| 1030 | static void mark_nosave_pages(struct memory_bitmap *bm) |
| 1031 | { |
| 1032 | struct nosave_region *region; |
| 1033 | |
| 1034 | if (list_empty(&nosave_regions)) |
| 1035 | return; |
| 1036 | |
| 1037 | list_for_each_entry(region, &nosave_regions, list) { |
| 1038 | unsigned long pfn; |
| 1039 | |
| 1040 | pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n", |
| 1041 | (unsigned long long) region->start_pfn << PAGE_SHIFT, |
| 1042 | ((unsigned long long) region->end_pfn << PAGE_SHIFT) |
| 1043 | - 1); |
| 1044 | |
| 1045 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) |
| 1046 | if (pfn_valid(pfn)) { |
| 1047 | /* |
| 1048 | * It is safe to ignore the result of |
| 1049 | * mem_bm_set_bit_check() here, since we won't |
| 1050 | * touch the PFNs for which the error is |
| 1051 | * returned anyway. |
| 1052 | */ |
| 1053 | mem_bm_set_bit_check(bm, pfn); |
| 1054 | } |
| 1055 | } |
| 1056 | } |
| 1057 | |
| 1058 | /** |
| 1059 | * create_basic_memory_bitmaps - Create bitmaps to hold basic page information. |
| 1060 | * |
| 1061 | * Create bitmaps needed for marking page frames that should not be saved and |
| 1062 | * free page frames. The forbidden_pages_map and free_pages_map pointers are |
| 1063 | * only modified if everything goes well, because we don't want the bits to be |
| 1064 | * touched before both bitmaps are set up. |
| 1065 | */ |
| 1066 | int create_basic_memory_bitmaps(void) |
| 1067 | { |
| 1068 | struct memory_bitmap *bm1, *bm2; |
| 1069 | int error = 0; |
| 1070 | |
| 1071 | if (forbidden_pages_map && free_pages_map) |
| 1072 | return 0; |
| 1073 | else |
| 1074 | BUG_ON(forbidden_pages_map || free_pages_map); |
| 1075 | |
| 1076 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
| 1077 | if (!bm1) |
| 1078 | return -ENOMEM; |
| 1079 | |
| 1080 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
| 1081 | if (error) |
| 1082 | goto Free_first_object; |
| 1083 | |
| 1084 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
| 1085 | if (!bm2) |
| 1086 | goto Free_first_bitmap; |
| 1087 | |
| 1088 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
| 1089 | if (error) |
| 1090 | goto Free_second_object; |
| 1091 | |
| 1092 | forbidden_pages_map = bm1; |
| 1093 | free_pages_map = bm2; |
| 1094 | mark_nosave_pages(forbidden_pages_map); |
| 1095 | |
| 1096 | pr_debug("PM: Basic memory bitmaps created\n"); |
| 1097 | |
| 1098 | return 0; |
| 1099 | |
| 1100 | Free_second_object: |
| 1101 | kfree(bm2); |
| 1102 | Free_first_bitmap: |
| 1103 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); |
| 1104 | Free_first_object: |
| 1105 | kfree(bm1); |
| 1106 | return -ENOMEM; |
| 1107 | } |
| 1108 | |
| 1109 | /** |
| 1110 | * free_basic_memory_bitmaps - Free memory bitmaps holding basic information. |
| 1111 | * |
| 1112 | * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The |
| 1113 | * auxiliary pointers are necessary so that the bitmaps themselves are not |
| 1114 | * referred to while they are being freed. |
| 1115 | */ |
| 1116 | void free_basic_memory_bitmaps(void) |
| 1117 | { |
| 1118 | struct memory_bitmap *bm1, *bm2; |
| 1119 | |
| 1120 | if (WARN_ON(!(forbidden_pages_map && free_pages_map))) |
| 1121 | return; |
| 1122 | |
| 1123 | bm1 = forbidden_pages_map; |
| 1124 | bm2 = free_pages_map; |
| 1125 | forbidden_pages_map = NULL; |
| 1126 | free_pages_map = NULL; |
| 1127 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); |
| 1128 | kfree(bm1); |
| 1129 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); |
| 1130 | kfree(bm2); |
| 1131 | |
| 1132 | pr_debug("PM: Basic memory bitmaps freed\n"); |
| 1133 | } |
| 1134 | |
| 1135 | /** |
| 1136 | * snapshot_additional_pages - Estimate the number of extra pages needed. |
| 1137 | * @zone: Memory zone to carry out the computation for. |
| 1138 | * |
| 1139 | * Estimate the number of additional pages needed for setting up a hibernation |
| 1140 | * image data structures for @zone (usually, the returned value is greater than |
| 1141 | * the exact number). |
| 1142 | */ |
| 1143 | unsigned int snapshot_additional_pages(struct zone *zone) |
| 1144 | { |
| 1145 | unsigned int rtree, nodes; |
| 1146 | |
| 1147 | rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); |
| 1148 | rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node), |
| 1149 | LINKED_PAGE_DATA_SIZE); |
| 1150 | while (nodes > 1) { |
| 1151 | nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL); |
| 1152 | rtree += nodes; |
| 1153 | } |
| 1154 | |
| 1155 | return 2 * rtree; |
| 1156 | } |
| 1157 | |
| 1158 | #ifdef CONFIG_HIGHMEM |
| 1159 | /** |
| 1160 | * count_free_highmem_pages - Compute the total number of free highmem pages. |
| 1161 | * |
| 1162 | * The returned number is system-wide. |
| 1163 | */ |
| 1164 | static unsigned int count_free_highmem_pages(void) |
| 1165 | { |
| 1166 | struct zone *zone; |
| 1167 | unsigned int cnt = 0; |
| 1168 | |
| 1169 | for_each_populated_zone(zone) |
| 1170 | if (is_highmem(zone)) |
| 1171 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
| 1172 | |
| 1173 | return cnt; |
| 1174 | } |
| 1175 | |
| 1176 | /** |
| 1177 | * saveable_highmem_page - Check if a highmem page is saveable. |
| 1178 | * |
| 1179 | * Determine whether a highmem page should be included in a hibernation image. |
| 1180 | * |
| 1181 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, |
| 1182 | * and it isn't part of a free chunk of pages. |
| 1183 | */ |
| 1184 | static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) |
| 1185 | { |
| 1186 | struct page *page; |
| 1187 | |
| 1188 | if (!pfn_valid(pfn)) |
| 1189 | return NULL; |
| 1190 | |
| 1191 | page = pfn_to_page(pfn); |
| 1192 | if (page_zone(page) != zone) |
| 1193 | return NULL; |
| 1194 | |
| 1195 | BUG_ON(!PageHighMem(page)); |
| 1196 | |
| 1197 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
| 1198 | PageReserved(page)) |
| 1199 | return NULL; |
| 1200 | |
| 1201 | if (page_is_guard(page)) |
| 1202 | return NULL; |
| 1203 | |
| 1204 | return page; |
| 1205 | } |
| 1206 | |
| 1207 | /** |
| 1208 | * count_highmem_pages - Compute the total number of saveable highmem pages. |
| 1209 | */ |
| 1210 | static unsigned int count_highmem_pages(void) |
| 1211 | { |
| 1212 | struct zone *zone; |
| 1213 | unsigned int n = 0; |
| 1214 | |
| 1215 | for_each_populated_zone(zone) { |
| 1216 | unsigned long pfn, max_zone_pfn; |
| 1217 | |
| 1218 | if (!is_highmem(zone)) |
| 1219 | continue; |
| 1220 | |
| 1221 | mark_free_pages(zone); |
| 1222 | max_zone_pfn = zone_end_pfn(zone); |
| 1223 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
| 1224 | if (saveable_highmem_page(zone, pfn)) |
| 1225 | n++; |
| 1226 | } |
| 1227 | return n; |
| 1228 | } |
| 1229 | #else |
| 1230 | static inline void *saveable_highmem_page(struct zone *z, unsigned long p) |
| 1231 | { |
| 1232 | return NULL; |
| 1233 | } |
| 1234 | #endif /* CONFIG_HIGHMEM */ |
| 1235 | |
| 1236 | /** |
| 1237 | * saveable_page - Check if the given page is saveable. |
| 1238 | * |
| 1239 | * Determine whether a non-highmem page should be included in a hibernation |
| 1240 | * image. |
| 1241 | * |
| 1242 | * We should save the page if it isn't Nosave, and is not in the range |
| 1243 | * of pages statically defined as 'unsaveable', and it isn't part of |
| 1244 | * a free chunk of pages. |
| 1245 | */ |
| 1246 | static struct page *saveable_page(struct zone *zone, unsigned long pfn) |
| 1247 | { |
| 1248 | struct page *page; |
| 1249 | |
| 1250 | if (!pfn_valid(pfn)) |
| 1251 | return NULL; |
| 1252 | |
| 1253 | page = pfn_to_page(pfn); |
| 1254 | if (page_zone(page) != zone) |
| 1255 | return NULL; |
| 1256 | |
| 1257 | BUG_ON(PageHighMem(page)); |
| 1258 | |
| 1259 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
| 1260 | return NULL; |
| 1261 | |
| 1262 | if (PageReserved(page) |
| 1263 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) |
| 1264 | return NULL; |
| 1265 | |
| 1266 | if (page_is_guard(page)) |
| 1267 | return NULL; |
| 1268 | |
| 1269 | return page; |
| 1270 | } |
| 1271 | |
| 1272 | /** |
| 1273 | * count_data_pages - Compute the total number of saveable non-highmem pages. |
| 1274 | */ |
| 1275 | static unsigned int count_data_pages(void) |
| 1276 | { |
| 1277 | struct zone *zone; |
| 1278 | unsigned long pfn, max_zone_pfn; |
| 1279 | unsigned int n = 0; |
| 1280 | |
| 1281 | for_each_populated_zone(zone) { |
| 1282 | if (is_highmem(zone)) |
| 1283 | continue; |
| 1284 | |
| 1285 | mark_free_pages(zone); |
| 1286 | max_zone_pfn = zone_end_pfn(zone); |
| 1287 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
| 1288 | if (saveable_page(zone, pfn)) |
| 1289 | n++; |
| 1290 | } |
| 1291 | return n; |
| 1292 | } |
| 1293 | |
| 1294 | /* |
| 1295 | * This is needed, because copy_page and memcpy are not usable for copying |
| 1296 | * task structs. |
| 1297 | */ |
| 1298 | static inline void do_copy_page(long *dst, long *src) |
| 1299 | { |
| 1300 | int n; |
| 1301 | |
| 1302 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
| 1303 | *dst++ = *src++; |
| 1304 | } |
| 1305 | |
| 1306 | /** |
| 1307 | * safe_copy_page - Copy a page in a safe way. |
| 1308 | * |
| 1309 | * Check if the page we are going to copy is marked as present in the kernel |
| 1310 | * page tables (this always is the case if CONFIG_DEBUG_PAGEALLOC is not set |
| 1311 | * and in that case kernel_page_present() always returns 'true'). |
| 1312 | */ |
| 1313 | static void safe_copy_page(void *dst, struct page *s_page) |
| 1314 | { |
| 1315 | if (kernel_page_present(s_page)) { |
| 1316 | do_copy_page(dst, page_address(s_page)); |
| 1317 | } else { |
| 1318 | kernel_map_pages(s_page, 1, 1); |
| 1319 | do_copy_page(dst, page_address(s_page)); |
| 1320 | kernel_map_pages(s_page, 1, 0); |
| 1321 | } |
| 1322 | } |
| 1323 | |
| 1324 | #ifdef CONFIG_HIGHMEM |
| 1325 | static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn) |
| 1326 | { |
| 1327 | return is_highmem(zone) ? |
| 1328 | saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn); |
| 1329 | } |
| 1330 | |
| 1331 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
| 1332 | { |
| 1333 | struct page *s_page, *d_page; |
| 1334 | void *src, *dst; |
| 1335 | |
| 1336 | s_page = pfn_to_page(src_pfn); |
| 1337 | d_page = pfn_to_page(dst_pfn); |
| 1338 | if (PageHighMem(s_page)) { |
| 1339 | src = kmap_atomic(s_page); |
| 1340 | dst = kmap_atomic(d_page); |
| 1341 | do_copy_page(dst, src); |
| 1342 | kunmap_atomic(dst); |
| 1343 | kunmap_atomic(src); |
| 1344 | } else { |
| 1345 | if (PageHighMem(d_page)) { |
| 1346 | /* |
| 1347 | * The page pointed to by src may contain some kernel |
| 1348 | * data modified by kmap_atomic() |
| 1349 | */ |
| 1350 | safe_copy_page(buffer, s_page); |
| 1351 | dst = kmap_atomic(d_page); |
| 1352 | copy_page(dst, buffer); |
| 1353 | kunmap_atomic(dst); |
| 1354 | } else { |
| 1355 | safe_copy_page(page_address(d_page), s_page); |
| 1356 | } |
| 1357 | } |
| 1358 | } |
| 1359 | #else |
| 1360 | #define page_is_saveable(zone, pfn) saveable_page(zone, pfn) |
| 1361 | |
| 1362 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
| 1363 | { |
| 1364 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), |
| 1365 | pfn_to_page(src_pfn)); |
| 1366 | } |
| 1367 | #endif /* CONFIG_HIGHMEM */ |
| 1368 | |
| 1369 | static void copy_data_pages(struct memory_bitmap *copy_bm, |
| 1370 | struct memory_bitmap *orig_bm) |
| 1371 | { |
| 1372 | struct zone *zone; |
| 1373 | unsigned long pfn; |
| 1374 | |
| 1375 | for_each_populated_zone(zone) { |
| 1376 | unsigned long max_zone_pfn; |
| 1377 | |
| 1378 | mark_free_pages(zone); |
| 1379 | max_zone_pfn = zone_end_pfn(zone); |
| 1380 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
| 1381 | if (page_is_saveable(zone, pfn)) |
| 1382 | memory_bm_set_bit(orig_bm, pfn); |
| 1383 | } |
| 1384 | memory_bm_position_reset(orig_bm); |
| 1385 | memory_bm_position_reset(copy_bm); |
| 1386 | for(;;) { |
| 1387 | pfn = memory_bm_next_pfn(orig_bm); |
| 1388 | if (unlikely(pfn == BM_END_OF_MAP)) |
| 1389 | break; |
| 1390 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); |
| 1391 | } |
| 1392 | } |
| 1393 | |
| 1394 | /* Total number of image pages */ |
| 1395 | static unsigned int nr_copy_pages; |
| 1396 | /* Number of pages needed for saving the original pfns of the image pages */ |
| 1397 | static unsigned int nr_meta_pages; |
| 1398 | /* |
| 1399 | * Numbers of normal and highmem page frames allocated for hibernation image |
| 1400 | * before suspending devices. |
| 1401 | */ |
| 1402 | unsigned int alloc_normal, alloc_highmem; |
| 1403 | /* |
| 1404 | * Memory bitmap used for marking saveable pages (during hibernation) or |
| 1405 | * hibernation image pages (during restore) |
| 1406 | */ |
| 1407 | static struct memory_bitmap orig_bm; |
| 1408 | /* |
| 1409 | * Memory bitmap used during hibernation for marking allocated page frames that |
| 1410 | * will contain copies of saveable pages. During restore it is initially used |
| 1411 | * for marking hibernation image pages, but then the set bits from it are |
| 1412 | * duplicated in @orig_bm and it is released. On highmem systems it is next |
| 1413 | * used for marking "safe" highmem pages, but it has to be reinitialized for |
| 1414 | * this purpose. |
| 1415 | */ |
| 1416 | static struct memory_bitmap copy_bm; |
| 1417 | |
| 1418 | /** |
| 1419 | * swsusp_free - Free pages allocated for hibernation image. |
| 1420 | * |
| 1421 | * Image pages are alocated before snapshot creation, so they need to be |
| 1422 | * released after resume. |
| 1423 | */ |
| 1424 | void swsusp_free(void) |
| 1425 | { |
| 1426 | unsigned long fb_pfn, fr_pfn; |
| 1427 | |
| 1428 | if (!forbidden_pages_map || !free_pages_map) |
| 1429 | goto out; |
| 1430 | |
| 1431 | memory_bm_position_reset(forbidden_pages_map); |
| 1432 | memory_bm_position_reset(free_pages_map); |
| 1433 | |
| 1434 | loop: |
| 1435 | fr_pfn = memory_bm_next_pfn(free_pages_map); |
| 1436 | fb_pfn = memory_bm_next_pfn(forbidden_pages_map); |
| 1437 | |
| 1438 | /* |
| 1439 | * Find the next bit set in both bitmaps. This is guaranteed to |
| 1440 | * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP. |
| 1441 | */ |
| 1442 | do { |
| 1443 | if (fb_pfn < fr_pfn) |
| 1444 | fb_pfn = memory_bm_next_pfn(forbidden_pages_map); |
| 1445 | if (fr_pfn < fb_pfn) |
| 1446 | fr_pfn = memory_bm_next_pfn(free_pages_map); |
| 1447 | } while (fb_pfn != fr_pfn); |
| 1448 | |
| 1449 | if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) { |
| 1450 | struct page *page = pfn_to_page(fr_pfn); |
| 1451 | |
| 1452 | memory_bm_clear_current(forbidden_pages_map); |
| 1453 | memory_bm_clear_current(free_pages_map); |
| 1454 | hibernate_restore_unprotect_page(page_address(page)); |
| 1455 | __free_page(page); |
| 1456 | goto loop; |
| 1457 | } |
| 1458 | |
| 1459 | out: |
| 1460 | nr_copy_pages = 0; |
| 1461 | nr_meta_pages = 0; |
| 1462 | restore_pblist = NULL; |
| 1463 | buffer = NULL; |
| 1464 | alloc_normal = 0; |
| 1465 | alloc_highmem = 0; |
| 1466 | hibernate_restore_protection_end(); |
| 1467 | } |
| 1468 | |
| 1469 | /* Helper functions used for the shrinking of memory. */ |
| 1470 | |
| 1471 | #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN) |
| 1472 | |
| 1473 | /** |
| 1474 | * preallocate_image_pages - Allocate a number of pages for hibernation image. |
| 1475 | * @nr_pages: Number of page frames to allocate. |
| 1476 | * @mask: GFP flags to use for the allocation. |
| 1477 | * |
| 1478 | * Return value: Number of page frames actually allocated |
| 1479 | */ |
| 1480 | static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask) |
| 1481 | { |
| 1482 | unsigned long nr_alloc = 0; |
| 1483 | |
| 1484 | while (nr_pages > 0) { |
| 1485 | struct page *page; |
| 1486 | |
| 1487 | page = alloc_image_page(mask); |
| 1488 | if (!page) |
| 1489 | break; |
| 1490 | memory_bm_set_bit(©_bm, page_to_pfn(page)); |
| 1491 | if (PageHighMem(page)) |
| 1492 | alloc_highmem++; |
| 1493 | else |
| 1494 | alloc_normal++; |
| 1495 | nr_pages--; |
| 1496 | nr_alloc++; |
| 1497 | } |
| 1498 | |
| 1499 | return nr_alloc; |
| 1500 | } |
| 1501 | |
| 1502 | static unsigned long preallocate_image_memory(unsigned long nr_pages, |
| 1503 | unsigned long avail_normal) |
| 1504 | { |
| 1505 | unsigned long alloc; |
| 1506 | |
| 1507 | if (avail_normal <= alloc_normal) |
| 1508 | return 0; |
| 1509 | |
| 1510 | alloc = avail_normal - alloc_normal; |
| 1511 | if (nr_pages < alloc) |
| 1512 | alloc = nr_pages; |
| 1513 | |
| 1514 | return preallocate_image_pages(alloc, GFP_IMAGE); |
| 1515 | } |
| 1516 | |
| 1517 | #ifdef CONFIG_HIGHMEM |
| 1518 | static unsigned long preallocate_image_highmem(unsigned long nr_pages) |
| 1519 | { |
| 1520 | return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM); |
| 1521 | } |
| 1522 | |
| 1523 | /** |
| 1524 | * __fraction - Compute (an approximation of) x * (multiplier / base). |
| 1525 | */ |
| 1526 | static unsigned long __fraction(u64 x, u64 multiplier, u64 base) |
| 1527 | { |
| 1528 | x *= multiplier; |
| 1529 | do_div(x, base); |
| 1530 | return (unsigned long)x; |
| 1531 | } |
| 1532 | |
| 1533 | static unsigned long preallocate_highmem_fraction(unsigned long nr_pages, |
| 1534 | unsigned long highmem, |
| 1535 | unsigned long total) |
| 1536 | { |
| 1537 | unsigned long alloc = __fraction(nr_pages, highmem, total); |
| 1538 | |
| 1539 | return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM); |
| 1540 | } |
| 1541 | #else /* CONFIG_HIGHMEM */ |
| 1542 | static inline unsigned long preallocate_image_highmem(unsigned long nr_pages) |
| 1543 | { |
| 1544 | return 0; |
| 1545 | } |
| 1546 | |
| 1547 | static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages, |
| 1548 | unsigned long highmem, |
| 1549 | unsigned long total) |
| 1550 | { |
| 1551 | return 0; |
| 1552 | } |
| 1553 | #endif /* CONFIG_HIGHMEM */ |
| 1554 | |
| 1555 | /** |
| 1556 | * free_unnecessary_pages - Release preallocated pages not needed for the image. |
| 1557 | */ |
| 1558 | static unsigned long free_unnecessary_pages(void) |
| 1559 | { |
| 1560 | unsigned long save, to_free_normal, to_free_highmem, free; |
| 1561 | |
| 1562 | save = count_data_pages(); |
| 1563 | if (alloc_normal >= save) { |
| 1564 | to_free_normal = alloc_normal - save; |
| 1565 | save = 0; |
| 1566 | } else { |
| 1567 | to_free_normal = 0; |
| 1568 | save -= alloc_normal; |
| 1569 | } |
| 1570 | save += count_highmem_pages(); |
| 1571 | if (alloc_highmem >= save) { |
| 1572 | to_free_highmem = alloc_highmem - save; |
| 1573 | } else { |
| 1574 | to_free_highmem = 0; |
| 1575 | save -= alloc_highmem; |
| 1576 | if (to_free_normal > save) |
| 1577 | to_free_normal -= save; |
| 1578 | else |
| 1579 | to_free_normal = 0; |
| 1580 | } |
| 1581 | free = to_free_normal + to_free_highmem; |
| 1582 | |
| 1583 | memory_bm_position_reset(©_bm); |
| 1584 | |
| 1585 | while (to_free_normal > 0 || to_free_highmem > 0) { |
| 1586 | unsigned long pfn = memory_bm_next_pfn(©_bm); |
| 1587 | struct page *page = pfn_to_page(pfn); |
| 1588 | |
| 1589 | if (PageHighMem(page)) { |
| 1590 | if (!to_free_highmem) |
| 1591 | continue; |
| 1592 | to_free_highmem--; |
| 1593 | alloc_highmem--; |
| 1594 | } else { |
| 1595 | if (!to_free_normal) |
| 1596 | continue; |
| 1597 | to_free_normal--; |
| 1598 | alloc_normal--; |
| 1599 | } |
| 1600 | memory_bm_clear_bit(©_bm, pfn); |
| 1601 | swsusp_unset_page_forbidden(page); |
| 1602 | swsusp_unset_page_free(page); |
| 1603 | __free_page(page); |
| 1604 | } |
| 1605 | |
| 1606 | return free; |
| 1607 | } |
| 1608 | |
| 1609 | /** |
| 1610 | * minimum_image_size - Estimate the minimum acceptable size of an image. |
| 1611 | * @saveable: Number of saveable pages in the system. |
| 1612 | * |
| 1613 | * We want to avoid attempting to free too much memory too hard, so estimate the |
| 1614 | * minimum acceptable size of a hibernation image to use as the lower limit for |
| 1615 | * preallocating memory. |
| 1616 | * |
| 1617 | * We assume that the minimum image size should be proportional to |
| 1618 | * |
| 1619 | * [number of saveable pages] - [number of pages that can be freed in theory] |
| 1620 | * |
| 1621 | * where the second term is the sum of (1) reclaimable slab pages, (2) active |
| 1622 | * and (3) inactive anonymous pages, (4) active and (5) inactive file pages, |
| 1623 | * minus mapped file pages. |
| 1624 | */ |
| 1625 | static unsigned long minimum_image_size(unsigned long saveable) |
| 1626 | { |
| 1627 | unsigned long size; |
| 1628 | |
| 1629 | size = global_page_state(NR_SLAB_RECLAIMABLE) |
| 1630 | + global_node_page_state(NR_ACTIVE_ANON) |
| 1631 | + global_node_page_state(NR_INACTIVE_ANON) |
| 1632 | + global_node_page_state(NR_ACTIVE_FILE) |
| 1633 | + global_node_page_state(NR_INACTIVE_FILE) |
| 1634 | - global_node_page_state(NR_FILE_MAPPED); |
| 1635 | |
| 1636 | return saveable <= size ? 0 : saveable - size; |
| 1637 | } |
| 1638 | |
| 1639 | /** |
| 1640 | * hibernate_preallocate_memory - Preallocate memory for hibernation image. |
| 1641 | * |
| 1642 | * To create a hibernation image it is necessary to make a copy of every page |
| 1643 | * frame in use. We also need a number of page frames to be free during |
| 1644 | * hibernation for allocations made while saving the image and for device |
| 1645 | * drivers, in case they need to allocate memory from their hibernation |
| 1646 | * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough |
| 1647 | * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through |
| 1648 | * /sys/power/reserved_size, respectively). To make this happen, we compute the |
| 1649 | * total number of available page frames and allocate at least |
| 1650 | * |
| 1651 | * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 |
| 1652 | * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE) |
| 1653 | * |
| 1654 | * of them, which corresponds to the maximum size of a hibernation image. |
| 1655 | * |
| 1656 | * If image_size is set below the number following from the above formula, |
| 1657 | * the preallocation of memory is continued until the total number of saveable |
| 1658 | * pages in the system is below the requested image size or the minimum |
| 1659 | * acceptable image size returned by minimum_image_size(), whichever is greater. |
| 1660 | */ |
| 1661 | int hibernate_preallocate_memory(void) |
| 1662 | { |
| 1663 | struct zone *zone; |
| 1664 | unsigned long saveable, size, max_size, count, highmem, pages = 0; |
| 1665 | unsigned long alloc, save_highmem, pages_highmem, avail_normal; |
| 1666 | ktime_t start, stop; |
| 1667 | int error; |
| 1668 | |
| 1669 | printk(KERN_INFO "PM: Preallocating image memory... "); |
| 1670 | start = ktime_get(); |
| 1671 | |
| 1672 | error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY); |
| 1673 | if (error) |
| 1674 | goto err_out; |
| 1675 | |
| 1676 | error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY); |
| 1677 | if (error) |
| 1678 | goto err_out; |
| 1679 | |
| 1680 | alloc_normal = 0; |
| 1681 | alloc_highmem = 0; |
| 1682 | |
| 1683 | /* Count the number of saveable data pages. */ |
| 1684 | save_highmem = count_highmem_pages(); |
| 1685 | saveable = count_data_pages(); |
| 1686 | |
| 1687 | /* |
| 1688 | * Compute the total number of page frames we can use (count) and the |
| 1689 | * number of pages needed for image metadata (size). |
| 1690 | */ |
| 1691 | count = saveable; |
| 1692 | saveable += save_highmem; |
| 1693 | highmem = save_highmem; |
| 1694 | size = 0; |
| 1695 | for_each_populated_zone(zone) { |
| 1696 | size += snapshot_additional_pages(zone); |
| 1697 | if (is_highmem(zone)) |
| 1698 | highmem += zone_page_state(zone, NR_FREE_PAGES); |
| 1699 | else |
| 1700 | count += zone_page_state(zone, NR_FREE_PAGES); |
| 1701 | } |
| 1702 | avail_normal = count; |
| 1703 | count += highmem; |
| 1704 | count -= totalreserve_pages; |
| 1705 | |
| 1706 | /* Add number of pages required for page keys (s390 only). */ |
| 1707 | size += page_key_additional_pages(saveable); |
| 1708 | |
| 1709 | /* Compute the maximum number of saveable pages to leave in memory. */ |
| 1710 | max_size = (count - (size + PAGES_FOR_IO)) / 2 |
| 1711 | - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE); |
| 1712 | /* Compute the desired number of image pages specified by image_size. */ |
| 1713 | size = DIV_ROUND_UP(image_size, PAGE_SIZE); |
| 1714 | if (size > max_size) |
| 1715 | size = max_size; |
| 1716 | /* |
| 1717 | * If the desired number of image pages is at least as large as the |
| 1718 | * current number of saveable pages in memory, allocate page frames for |
| 1719 | * the image and we're done. |
| 1720 | */ |
| 1721 | if (size >= saveable) { |
| 1722 | pages = preallocate_image_highmem(save_highmem); |
| 1723 | pages += preallocate_image_memory(saveable - pages, avail_normal); |
| 1724 | goto out; |
| 1725 | } |
| 1726 | |
| 1727 | /* Estimate the minimum size of the image. */ |
| 1728 | pages = minimum_image_size(saveable); |
| 1729 | /* |
| 1730 | * To avoid excessive pressure on the normal zone, leave room in it to |
| 1731 | * accommodate an image of the minimum size (unless it's already too |
| 1732 | * small, in which case don't preallocate pages from it at all). |
| 1733 | */ |
| 1734 | if (avail_normal > pages) |
| 1735 | avail_normal -= pages; |
| 1736 | else |
| 1737 | avail_normal = 0; |
| 1738 | if (size < pages) |
| 1739 | size = min_t(unsigned long, pages, max_size); |
| 1740 | |
| 1741 | /* |
| 1742 | * Let the memory management subsystem know that we're going to need a |
| 1743 | * large number of page frames to allocate and make it free some memory. |
| 1744 | * NOTE: If this is not done, performance will be hurt badly in some |
| 1745 | * test cases. |
| 1746 | */ |
| 1747 | shrink_all_memory(saveable - size); |
| 1748 | |
| 1749 | /* |
| 1750 | * The number of saveable pages in memory was too high, so apply some |
| 1751 | * pressure to decrease it. First, make room for the largest possible |
| 1752 | * image and fail if that doesn't work. Next, try to decrease the size |
| 1753 | * of the image as much as indicated by 'size' using allocations from |
| 1754 | * highmem and non-highmem zones separately. |
| 1755 | */ |
| 1756 | pages_highmem = preallocate_image_highmem(highmem / 2); |
| 1757 | alloc = count - max_size; |
| 1758 | if (alloc > pages_highmem) |
| 1759 | alloc -= pages_highmem; |
| 1760 | else |
| 1761 | alloc = 0; |
| 1762 | pages = preallocate_image_memory(alloc, avail_normal); |
| 1763 | if (pages < alloc) { |
| 1764 | /* We have exhausted non-highmem pages, try highmem. */ |
| 1765 | alloc -= pages; |
| 1766 | pages += pages_highmem; |
| 1767 | pages_highmem = preallocate_image_highmem(alloc); |
| 1768 | if (pages_highmem < alloc) |
| 1769 | goto err_out; |
| 1770 | pages += pages_highmem; |
| 1771 | /* |
| 1772 | * size is the desired number of saveable pages to leave in |
| 1773 | * memory, so try to preallocate (all memory - size) pages. |
| 1774 | */ |
| 1775 | alloc = (count - pages) - size; |
| 1776 | pages += preallocate_image_highmem(alloc); |
| 1777 | } else { |
| 1778 | /* |
| 1779 | * There are approximately max_size saveable pages at this point |
| 1780 | * and we want to reduce this number down to size. |
| 1781 | */ |
| 1782 | alloc = max_size - size; |
| 1783 | size = preallocate_highmem_fraction(alloc, highmem, count); |
| 1784 | pages_highmem += size; |
| 1785 | alloc -= size; |
| 1786 | size = preallocate_image_memory(alloc, avail_normal); |
| 1787 | pages_highmem += preallocate_image_highmem(alloc - size); |
| 1788 | pages += pages_highmem + size; |
| 1789 | } |
| 1790 | |
| 1791 | /* |
| 1792 | * We only need as many page frames for the image as there are saveable |
| 1793 | * pages in memory, but we have allocated more. Release the excessive |
| 1794 | * ones now. |
| 1795 | */ |
| 1796 | pages -= free_unnecessary_pages(); |
| 1797 | |
| 1798 | out: |
| 1799 | stop = ktime_get(); |
| 1800 | printk(KERN_CONT "done (allocated %lu pages)\n", pages); |
| 1801 | swsusp_show_speed(start, stop, pages, "Allocated"); |
| 1802 | |
| 1803 | return 0; |
| 1804 | |
| 1805 | err_out: |
| 1806 | printk(KERN_CONT "\n"); |
| 1807 | swsusp_free(); |
| 1808 | return -ENOMEM; |
| 1809 | } |
| 1810 | |
| 1811 | #ifdef CONFIG_HIGHMEM |
| 1812 | /** |
| 1813 | * count_pages_for_highmem - Count non-highmem pages needed for copying highmem. |
| 1814 | * |
| 1815 | * Compute the number of non-highmem pages that will be necessary for creating |
| 1816 | * copies of highmem pages. |
| 1817 | */ |
| 1818 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) |
| 1819 | { |
| 1820 | unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem; |
| 1821 | |
| 1822 | if (free_highmem >= nr_highmem) |
| 1823 | nr_highmem = 0; |
| 1824 | else |
| 1825 | nr_highmem -= free_highmem; |
| 1826 | |
| 1827 | return nr_highmem; |
| 1828 | } |
| 1829 | #else |
| 1830 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; } |
| 1831 | #endif /* CONFIG_HIGHMEM */ |
| 1832 | |
| 1833 | /** |
| 1834 | * enough_free_mem - Check if there is enough free memory for the image. |
| 1835 | */ |
| 1836 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
| 1837 | { |
| 1838 | struct zone *zone; |
| 1839 | unsigned int free = alloc_normal; |
| 1840 | |
| 1841 | for_each_populated_zone(zone) |
| 1842 | if (!is_highmem(zone)) |
| 1843 | free += zone_page_state(zone, NR_FREE_PAGES); |
| 1844 | |
| 1845 | nr_pages += count_pages_for_highmem(nr_highmem); |
| 1846 | pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n", |
| 1847 | nr_pages, PAGES_FOR_IO, free); |
| 1848 | |
| 1849 | return free > nr_pages + PAGES_FOR_IO; |
| 1850 | } |
| 1851 | |
| 1852 | #ifdef CONFIG_HIGHMEM |
| 1853 | /** |
| 1854 | * get_highmem_buffer - Allocate a buffer for highmem pages. |
| 1855 | * |
| 1856 | * If there are some highmem pages in the hibernation image, we may need a |
| 1857 | * buffer to copy them and/or load their data. |
| 1858 | */ |
| 1859 | static inline int get_highmem_buffer(int safe_needed) |
| 1860 | { |
| 1861 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); |
| 1862 | return buffer ? 0 : -ENOMEM; |
| 1863 | } |
| 1864 | |
| 1865 | /** |
| 1866 | * alloc_highmem_image_pages - Allocate some highmem pages for the image. |
| 1867 | * |
| 1868 | * Try to allocate as many pages as needed, but if the number of free highmem |
| 1869 | * pages is less than that, allocate them all. |
| 1870 | */ |
| 1871 | static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm, |
| 1872 | unsigned int nr_highmem) |
| 1873 | { |
| 1874 | unsigned int to_alloc = count_free_highmem_pages(); |
| 1875 | |
| 1876 | if (to_alloc > nr_highmem) |
| 1877 | to_alloc = nr_highmem; |
| 1878 | |
| 1879 | nr_highmem -= to_alloc; |
| 1880 | while (to_alloc-- > 0) { |
| 1881 | struct page *page; |
| 1882 | |
| 1883 | page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM); |
| 1884 | memory_bm_set_bit(bm, page_to_pfn(page)); |
| 1885 | } |
| 1886 | return nr_highmem; |
| 1887 | } |
| 1888 | #else |
| 1889 | static inline int get_highmem_buffer(int safe_needed) { return 0; } |
| 1890 | |
| 1891 | static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm, |
| 1892 | unsigned int n) { return 0; } |
| 1893 | #endif /* CONFIG_HIGHMEM */ |
| 1894 | |
| 1895 | /** |
| 1896 | * swsusp_alloc - Allocate memory for hibernation image. |
| 1897 | * |
| 1898 | * We first try to allocate as many highmem pages as there are |
| 1899 | * saveable highmem pages in the system. If that fails, we allocate |
| 1900 | * non-highmem pages for the copies of the remaining highmem ones. |
| 1901 | * |
| 1902 | * In this approach it is likely that the copies of highmem pages will |
| 1903 | * also be located in the high memory, because of the way in which |
| 1904 | * copy_data_pages() works. |
| 1905 | */ |
| 1906 | static int swsusp_alloc(struct memory_bitmap *orig_bm, |
| 1907 | struct memory_bitmap *copy_bm, |
| 1908 | unsigned int nr_pages, unsigned int nr_highmem) |
| 1909 | { |
| 1910 | if (nr_highmem > 0) { |
| 1911 | if (get_highmem_buffer(PG_ANY)) |
| 1912 | goto err_out; |
| 1913 | if (nr_highmem > alloc_highmem) { |
| 1914 | nr_highmem -= alloc_highmem; |
| 1915 | nr_pages += alloc_highmem_pages(copy_bm, nr_highmem); |
| 1916 | } |
| 1917 | } |
| 1918 | if (nr_pages > alloc_normal) { |
| 1919 | nr_pages -= alloc_normal; |
| 1920 | while (nr_pages-- > 0) { |
| 1921 | struct page *page; |
| 1922 | |
| 1923 | page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); |
| 1924 | if (!page) |
| 1925 | goto err_out; |
| 1926 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); |
| 1927 | } |
| 1928 | } |
| 1929 | |
| 1930 | return 0; |
| 1931 | |
| 1932 | err_out: |
| 1933 | swsusp_free(); |
| 1934 | return -ENOMEM; |
| 1935 | } |
| 1936 | |
| 1937 | asmlinkage __visible int swsusp_save(void) |
| 1938 | { |
| 1939 | unsigned int nr_pages, nr_highmem; |
| 1940 | |
| 1941 | printk(KERN_INFO "PM: Creating hibernation image:\n"); |
| 1942 | |
| 1943 | drain_local_pages(NULL); |
| 1944 | nr_pages = count_data_pages(); |
| 1945 | nr_highmem = count_highmem_pages(); |
| 1946 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); |
| 1947 | |
| 1948 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
| 1949 | printk(KERN_ERR "PM: Not enough free memory\n"); |
| 1950 | return -ENOMEM; |
| 1951 | } |
| 1952 | |
| 1953 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
| 1954 | printk(KERN_ERR "PM: Memory allocation failed\n"); |
| 1955 | return -ENOMEM; |
| 1956 | } |
| 1957 | |
| 1958 | /* |
| 1959 | * During allocating of suspend pagedir, new cold pages may appear. |
| 1960 | * Kill them. |
| 1961 | */ |
| 1962 | drain_local_pages(NULL); |
| 1963 | copy_data_pages(©_bm, &orig_bm); |
| 1964 | |
| 1965 | /* |
| 1966 | * End of critical section. From now on, we can write to memory, |
| 1967 | * but we should not touch disk. This specially means we must _not_ |
| 1968 | * touch swap space! Except we must write out our image of course. |
| 1969 | */ |
| 1970 | |
| 1971 | nr_pages += nr_highmem; |
| 1972 | nr_copy_pages = nr_pages; |
| 1973 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
| 1974 | |
| 1975 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", |
| 1976 | nr_pages); |
| 1977 | |
| 1978 | return 0; |
| 1979 | } |
| 1980 | |
| 1981 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER |
| 1982 | static int init_header_complete(struct swsusp_info *info) |
| 1983 | { |
| 1984 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
| 1985 | info->version_code = LINUX_VERSION_CODE; |
| 1986 | return 0; |
| 1987 | } |
| 1988 | |
| 1989 | static char *check_image_kernel(struct swsusp_info *info) |
| 1990 | { |
| 1991 | if (info->version_code != LINUX_VERSION_CODE) |
| 1992 | return "kernel version"; |
| 1993 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) |
| 1994 | return "system type"; |
| 1995 | if (strcmp(info->uts.release,init_utsname()->release)) |
| 1996 | return "kernel release"; |
| 1997 | if (strcmp(info->uts.version,init_utsname()->version)) |
| 1998 | return "version"; |
| 1999 | if (strcmp(info->uts.machine,init_utsname()->machine)) |
| 2000 | return "machine"; |
| 2001 | return NULL; |
| 2002 | } |
| 2003 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ |
| 2004 | |
| 2005 | unsigned long snapshot_get_image_size(void) |
| 2006 | { |
| 2007 | return nr_copy_pages + nr_meta_pages + 1; |
| 2008 | } |
| 2009 | |
| 2010 | static int init_header(struct swsusp_info *info) |
| 2011 | { |
| 2012 | memset(info, 0, sizeof(struct swsusp_info)); |
| 2013 | info->num_physpages = get_num_physpages(); |
| 2014 | info->image_pages = nr_copy_pages; |
| 2015 | info->pages = snapshot_get_image_size(); |
| 2016 | info->size = info->pages; |
| 2017 | info->size <<= PAGE_SHIFT; |
| 2018 | return init_header_complete(info); |
| 2019 | } |
| 2020 | |
| 2021 | /** |
| 2022 | * pack_pfns - Prepare PFNs for saving. |
| 2023 | * @bm: Memory bitmap. |
| 2024 | * @buf: Memory buffer to store the PFNs in. |
| 2025 | * |
| 2026 | * PFNs corresponding to set bits in @bm are stored in the area of memory |
| 2027 | * pointed to by @buf (1 page at a time). |
| 2028 | */ |
| 2029 | static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
| 2030 | { |
| 2031 | int j; |
| 2032 | |
| 2033 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
| 2034 | buf[j] = memory_bm_next_pfn(bm); |
| 2035 | if (unlikely(buf[j] == BM_END_OF_MAP)) |
| 2036 | break; |
| 2037 | /* Save page key for data page (s390 only). */ |
| 2038 | page_key_read(buf + j); |
| 2039 | } |
| 2040 | } |
| 2041 | |
| 2042 | /** |
| 2043 | * snapshot_read_next - Get the address to read the next image page from. |
| 2044 | * @handle: Snapshot handle to be used for the reading. |
| 2045 | * |
| 2046 | * On the first call, @handle should point to a zeroed snapshot_handle |
| 2047 | * structure. The structure gets populated then and a pointer to it should be |
| 2048 | * passed to this function every next time. |
| 2049 | * |
| 2050 | * On success, the function returns a positive number. Then, the caller |
| 2051 | * is allowed to read up to the returned number of bytes from the memory |
| 2052 | * location computed by the data_of() macro. |
| 2053 | * |
| 2054 | * The function returns 0 to indicate the end of the data stream condition, |
| 2055 | * and negative numbers are returned on errors. If that happens, the structure |
| 2056 | * pointed to by @handle is not updated and should not be used any more. |
| 2057 | */ |
| 2058 | int snapshot_read_next(struct snapshot_handle *handle) |
| 2059 | { |
| 2060 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
| 2061 | return 0; |
| 2062 | |
| 2063 | if (!buffer) { |
| 2064 | /* This makes the buffer be freed by swsusp_free() */ |
| 2065 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
| 2066 | if (!buffer) |
| 2067 | return -ENOMEM; |
| 2068 | } |
| 2069 | if (!handle->cur) { |
| 2070 | int error; |
| 2071 | |
| 2072 | error = init_header((struct swsusp_info *)buffer); |
| 2073 | if (error) |
| 2074 | return error; |
| 2075 | handle->buffer = buffer; |
| 2076 | memory_bm_position_reset(&orig_bm); |
| 2077 | memory_bm_position_reset(©_bm); |
| 2078 | } else if (handle->cur <= nr_meta_pages) { |
| 2079 | clear_page(buffer); |
| 2080 | pack_pfns(buffer, &orig_bm); |
| 2081 | } else { |
| 2082 | struct page *page; |
| 2083 | |
| 2084 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
| 2085 | if (PageHighMem(page)) { |
| 2086 | /* |
| 2087 | * Highmem pages are copied to the buffer, |
| 2088 | * because we can't return with a kmapped |
| 2089 | * highmem page (we may not be called again). |
| 2090 | */ |
| 2091 | void *kaddr; |
| 2092 | |
| 2093 | kaddr = kmap_atomic(page); |
| 2094 | copy_page(buffer, kaddr); |
| 2095 | kunmap_atomic(kaddr); |
| 2096 | handle->buffer = buffer; |
| 2097 | } else { |
| 2098 | handle->buffer = page_address(page); |
| 2099 | } |
| 2100 | } |
| 2101 | handle->cur++; |
| 2102 | return PAGE_SIZE; |
| 2103 | } |
| 2104 | |
| 2105 | static void duplicate_memory_bitmap(struct memory_bitmap *dst, |
| 2106 | struct memory_bitmap *src) |
| 2107 | { |
| 2108 | unsigned long pfn; |
| 2109 | |
| 2110 | memory_bm_position_reset(src); |
| 2111 | pfn = memory_bm_next_pfn(src); |
| 2112 | while (pfn != BM_END_OF_MAP) { |
| 2113 | memory_bm_set_bit(dst, pfn); |
| 2114 | pfn = memory_bm_next_pfn(src); |
| 2115 | } |
| 2116 | } |
| 2117 | |
| 2118 | /** |
| 2119 | * mark_unsafe_pages - Mark pages that were used before hibernation. |
| 2120 | * |
| 2121 | * Mark the pages that cannot be used for storing the image during restoration, |
| 2122 | * because they conflict with the pages that had been used before hibernation. |
| 2123 | */ |
| 2124 | static void mark_unsafe_pages(struct memory_bitmap *bm) |
| 2125 | { |
| 2126 | unsigned long pfn; |
| 2127 | |
| 2128 | /* Clear the "free"/"unsafe" bit for all PFNs */ |
| 2129 | memory_bm_position_reset(free_pages_map); |
| 2130 | pfn = memory_bm_next_pfn(free_pages_map); |
| 2131 | while (pfn != BM_END_OF_MAP) { |
| 2132 | memory_bm_clear_current(free_pages_map); |
| 2133 | pfn = memory_bm_next_pfn(free_pages_map); |
| 2134 | } |
| 2135 | |
| 2136 | /* Mark pages that correspond to the "original" PFNs as "unsafe" */ |
| 2137 | duplicate_memory_bitmap(free_pages_map, bm); |
| 2138 | |
| 2139 | allocated_unsafe_pages = 0; |
| 2140 | } |
| 2141 | |
| 2142 | static int check_header(struct swsusp_info *info) |
| 2143 | { |
| 2144 | char *reason; |
| 2145 | |
| 2146 | reason = check_image_kernel(info); |
| 2147 | if (!reason && info->num_physpages != get_num_physpages()) |
| 2148 | reason = "memory size"; |
| 2149 | if (reason) { |
| 2150 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); |
| 2151 | return -EPERM; |
| 2152 | } |
| 2153 | return 0; |
| 2154 | } |
| 2155 | |
| 2156 | /** |
| 2157 | * load header - Check the image header and copy the data from it. |
| 2158 | */ |
| 2159 | static int load_header(struct swsusp_info *info) |
| 2160 | { |
| 2161 | int error; |
| 2162 | |
| 2163 | restore_pblist = NULL; |
| 2164 | error = check_header(info); |
| 2165 | if (!error) { |
| 2166 | nr_copy_pages = info->image_pages; |
| 2167 | nr_meta_pages = info->pages - info->image_pages - 1; |
| 2168 | } |
| 2169 | return error; |
| 2170 | } |
| 2171 | |
| 2172 | /** |
| 2173 | * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap. |
| 2174 | * @bm: Memory bitmap. |
| 2175 | * @buf: Area of memory containing the PFNs. |
| 2176 | * |
| 2177 | * For each element of the array pointed to by @buf (1 page at a time), set the |
| 2178 | * corresponding bit in @bm. |
| 2179 | */ |
| 2180 | static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) |
| 2181 | { |
| 2182 | int j; |
| 2183 | |
| 2184 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
| 2185 | if (unlikely(buf[j] == BM_END_OF_MAP)) |
| 2186 | break; |
| 2187 | |
| 2188 | /* Extract and buffer page key for data page (s390 only). */ |
| 2189 | page_key_memorize(buf + j); |
| 2190 | |
| 2191 | if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j])) |
| 2192 | memory_bm_set_bit(bm, buf[j]); |
| 2193 | else |
| 2194 | return -EFAULT; |
| 2195 | } |
| 2196 | |
| 2197 | return 0; |
| 2198 | } |
| 2199 | |
| 2200 | #ifdef CONFIG_HIGHMEM |
| 2201 | /* |
| 2202 | * struct highmem_pbe is used for creating the list of highmem pages that |
| 2203 | * should be restored atomically during the resume from disk, because the page |
| 2204 | * frames they have occupied before the suspend are in use. |
| 2205 | */ |
| 2206 | struct highmem_pbe { |
| 2207 | struct page *copy_page; /* data is here now */ |
| 2208 | struct page *orig_page; /* data was here before the suspend */ |
| 2209 | struct highmem_pbe *next; |
| 2210 | }; |
| 2211 | |
| 2212 | /* |
| 2213 | * List of highmem PBEs needed for restoring the highmem pages that were |
| 2214 | * allocated before the suspend and included in the suspend image, but have |
| 2215 | * also been allocated by the "resume" kernel, so their contents cannot be |
| 2216 | * written directly to their "original" page frames. |
| 2217 | */ |
| 2218 | static struct highmem_pbe *highmem_pblist; |
| 2219 | |
| 2220 | /** |
| 2221 | * count_highmem_image_pages - Compute the number of highmem pages in the image. |
| 2222 | * @bm: Memory bitmap. |
| 2223 | * |
| 2224 | * The bits in @bm that correspond to image pages are assumed to be set. |
| 2225 | */ |
| 2226 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) |
| 2227 | { |
| 2228 | unsigned long pfn; |
| 2229 | unsigned int cnt = 0; |
| 2230 | |
| 2231 | memory_bm_position_reset(bm); |
| 2232 | pfn = memory_bm_next_pfn(bm); |
| 2233 | while (pfn != BM_END_OF_MAP) { |
| 2234 | if (PageHighMem(pfn_to_page(pfn))) |
| 2235 | cnt++; |
| 2236 | |
| 2237 | pfn = memory_bm_next_pfn(bm); |
| 2238 | } |
| 2239 | return cnt; |
| 2240 | } |
| 2241 | |
| 2242 | static unsigned int safe_highmem_pages; |
| 2243 | |
| 2244 | static struct memory_bitmap *safe_highmem_bm; |
| 2245 | |
| 2246 | /** |
| 2247 | * prepare_highmem_image - Allocate memory for loading highmem data from image. |
| 2248 | * @bm: Pointer to an uninitialized memory bitmap structure. |
| 2249 | * @nr_highmem_p: Pointer to the number of highmem image pages. |
| 2250 | * |
| 2251 | * Try to allocate as many highmem pages as there are highmem image pages |
| 2252 | * (@nr_highmem_p points to the variable containing the number of highmem image |
| 2253 | * pages). The pages that are "safe" (ie. will not be overwritten when the |
| 2254 | * hibernation image is restored entirely) have the corresponding bits set in |
| 2255 | * @bm (it must be unitialized). |
| 2256 | * |
| 2257 | * NOTE: This function should not be called if there are no highmem image pages. |
| 2258 | */ |
| 2259 | static int prepare_highmem_image(struct memory_bitmap *bm, |
| 2260 | unsigned int *nr_highmem_p) |
| 2261 | { |
| 2262 | unsigned int to_alloc; |
| 2263 | |
| 2264 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) |
| 2265 | return -ENOMEM; |
| 2266 | |
| 2267 | if (get_highmem_buffer(PG_SAFE)) |
| 2268 | return -ENOMEM; |
| 2269 | |
| 2270 | to_alloc = count_free_highmem_pages(); |
| 2271 | if (to_alloc > *nr_highmem_p) |
| 2272 | to_alloc = *nr_highmem_p; |
| 2273 | else |
| 2274 | *nr_highmem_p = to_alloc; |
| 2275 | |
| 2276 | safe_highmem_pages = 0; |
| 2277 | while (to_alloc-- > 0) { |
| 2278 | struct page *page; |
| 2279 | |
| 2280 | page = alloc_page(__GFP_HIGHMEM); |
| 2281 | if (!swsusp_page_is_free(page)) { |
| 2282 | /* The page is "safe", set its bit the bitmap */ |
| 2283 | memory_bm_set_bit(bm, page_to_pfn(page)); |
| 2284 | safe_highmem_pages++; |
| 2285 | } |
| 2286 | /* Mark the page as allocated */ |
| 2287 | swsusp_set_page_forbidden(page); |
| 2288 | swsusp_set_page_free(page); |
| 2289 | } |
| 2290 | memory_bm_position_reset(bm); |
| 2291 | safe_highmem_bm = bm; |
| 2292 | return 0; |
| 2293 | } |
| 2294 | |
| 2295 | static struct page *last_highmem_page; |
| 2296 | |
| 2297 | /** |
| 2298 | * get_highmem_page_buffer - Prepare a buffer to store a highmem image page. |
| 2299 | * |
| 2300 | * For a given highmem image page get a buffer that suspend_write_next() should |
| 2301 | * return to its caller to write to. |
| 2302 | * |
| 2303 | * If the page is to be saved to its "original" page frame or a copy of |
| 2304 | * the page is to be made in the highmem, @buffer is returned. Otherwise, |
| 2305 | * the copy of the page is to be made in normal memory, so the address of |
| 2306 | * the copy is returned. |
| 2307 | * |
| 2308 | * If @buffer is returned, the caller of suspend_write_next() will write |
| 2309 | * the page's contents to @buffer, so they will have to be copied to the |
| 2310 | * right location on the next call to suspend_write_next() and it is done |
| 2311 | * with the help of copy_last_highmem_page(). For this purpose, if |
| 2312 | * @buffer is returned, @last_highmem_page is set to the page to which |
| 2313 | * the data will have to be copied from @buffer. |
| 2314 | */ |
| 2315 | static void *get_highmem_page_buffer(struct page *page, |
| 2316 | struct chain_allocator *ca) |
| 2317 | { |
| 2318 | struct highmem_pbe *pbe; |
| 2319 | void *kaddr; |
| 2320 | |
| 2321 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
| 2322 | /* |
| 2323 | * We have allocated the "original" page frame and we can |
| 2324 | * use it directly to store the loaded page. |
| 2325 | */ |
| 2326 | last_highmem_page = page; |
| 2327 | return buffer; |
| 2328 | } |
| 2329 | /* |
| 2330 | * The "original" page frame has not been allocated and we have to |
| 2331 | * use a "safe" page frame to store the loaded page. |
| 2332 | */ |
| 2333 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); |
| 2334 | if (!pbe) { |
| 2335 | swsusp_free(); |
| 2336 | return ERR_PTR(-ENOMEM); |
| 2337 | } |
| 2338 | pbe->orig_page = page; |
| 2339 | if (safe_highmem_pages > 0) { |
| 2340 | struct page *tmp; |
| 2341 | |
| 2342 | /* Copy of the page will be stored in high memory */ |
| 2343 | kaddr = buffer; |
| 2344 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); |
| 2345 | safe_highmem_pages--; |
| 2346 | last_highmem_page = tmp; |
| 2347 | pbe->copy_page = tmp; |
| 2348 | } else { |
| 2349 | /* Copy of the page will be stored in normal memory */ |
| 2350 | kaddr = safe_pages_list; |
| 2351 | safe_pages_list = safe_pages_list->next; |
| 2352 | pbe->copy_page = virt_to_page(kaddr); |
| 2353 | } |
| 2354 | pbe->next = highmem_pblist; |
| 2355 | highmem_pblist = pbe; |
| 2356 | return kaddr; |
| 2357 | } |
| 2358 | |
| 2359 | /** |
| 2360 | * copy_last_highmem_page - Copy most the most recent highmem image page. |
| 2361 | * |
| 2362 | * Copy the contents of a highmem image from @buffer, where the caller of |
| 2363 | * snapshot_write_next() has stored them, to the right location represented by |
| 2364 | * @last_highmem_page . |
| 2365 | */ |
| 2366 | static void copy_last_highmem_page(void) |
| 2367 | { |
| 2368 | if (last_highmem_page) { |
| 2369 | void *dst; |
| 2370 | |
| 2371 | dst = kmap_atomic(last_highmem_page); |
| 2372 | copy_page(dst, buffer); |
| 2373 | kunmap_atomic(dst); |
| 2374 | last_highmem_page = NULL; |
| 2375 | } |
| 2376 | } |
| 2377 | |
| 2378 | static inline int last_highmem_page_copied(void) |
| 2379 | { |
| 2380 | return !last_highmem_page; |
| 2381 | } |
| 2382 | |
| 2383 | static inline void free_highmem_data(void) |
| 2384 | { |
| 2385 | if (safe_highmem_bm) |
| 2386 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); |
| 2387 | |
| 2388 | if (buffer) |
| 2389 | free_image_page(buffer, PG_UNSAFE_CLEAR); |
| 2390 | } |
| 2391 | #else |
| 2392 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } |
| 2393 | |
| 2394 | static inline int prepare_highmem_image(struct memory_bitmap *bm, |
| 2395 | unsigned int *nr_highmem_p) { return 0; } |
| 2396 | |
| 2397 | static inline void *get_highmem_page_buffer(struct page *page, |
| 2398 | struct chain_allocator *ca) |
| 2399 | { |
| 2400 | return ERR_PTR(-EINVAL); |
| 2401 | } |
| 2402 | |
| 2403 | static inline void copy_last_highmem_page(void) {} |
| 2404 | static inline int last_highmem_page_copied(void) { return 1; } |
| 2405 | static inline void free_highmem_data(void) {} |
| 2406 | #endif /* CONFIG_HIGHMEM */ |
| 2407 | |
| 2408 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
| 2409 | |
| 2410 | /** |
| 2411 | * prepare_image - Make room for loading hibernation image. |
| 2412 | * @new_bm: Unitialized memory bitmap structure. |
| 2413 | * @bm: Memory bitmap with unsafe pages marked. |
| 2414 | * |
| 2415 | * Use @bm to mark the pages that will be overwritten in the process of |
| 2416 | * restoring the system memory state from the suspend image ("unsafe" pages) |
| 2417 | * and allocate memory for the image. |
| 2418 | * |
| 2419 | * The idea is to allocate a new memory bitmap first and then allocate |
| 2420 | * as many pages as needed for image data, but without specifying what those |
| 2421 | * pages will be used for just yet. Instead, we mark them all as allocated and |
| 2422 | * create a lists of "safe" pages to be used later. On systems with high |
| 2423 | * memory a list of "safe" highmem pages is created too. |
| 2424 | */ |
| 2425 | static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) |
| 2426 | { |
| 2427 | unsigned int nr_pages, nr_highmem; |
| 2428 | struct linked_page *lp; |
| 2429 | int error; |
| 2430 | |
| 2431 | /* If there is no highmem, the buffer will not be necessary */ |
| 2432 | free_image_page(buffer, PG_UNSAFE_CLEAR); |
| 2433 | buffer = NULL; |
| 2434 | |
| 2435 | nr_highmem = count_highmem_image_pages(bm); |
| 2436 | mark_unsafe_pages(bm); |
| 2437 | |
| 2438 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); |
| 2439 | if (error) |
| 2440 | goto Free; |
| 2441 | |
| 2442 | duplicate_memory_bitmap(new_bm, bm); |
| 2443 | memory_bm_free(bm, PG_UNSAFE_KEEP); |
| 2444 | if (nr_highmem > 0) { |
| 2445 | error = prepare_highmem_image(bm, &nr_highmem); |
| 2446 | if (error) |
| 2447 | goto Free; |
| 2448 | } |
| 2449 | /* |
| 2450 | * Reserve some safe pages for potential later use. |
| 2451 | * |
| 2452 | * NOTE: This way we make sure there will be enough safe pages for the |
| 2453 | * chain_alloc() in get_buffer(). It is a bit wasteful, but |
| 2454 | * nr_copy_pages cannot be greater than 50% of the memory anyway. |
| 2455 | * |
| 2456 | * nr_copy_pages cannot be less than allocated_unsafe_pages too. |
| 2457 | */ |
| 2458 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
| 2459 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
| 2460 | while (nr_pages > 0) { |
| 2461 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
| 2462 | if (!lp) { |
| 2463 | error = -ENOMEM; |
| 2464 | goto Free; |
| 2465 | } |
| 2466 | lp->next = safe_pages_list; |
| 2467 | safe_pages_list = lp; |
| 2468 | nr_pages--; |
| 2469 | } |
| 2470 | /* Preallocate memory for the image */ |
| 2471 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
| 2472 | while (nr_pages > 0) { |
| 2473 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); |
| 2474 | if (!lp) { |
| 2475 | error = -ENOMEM; |
| 2476 | goto Free; |
| 2477 | } |
| 2478 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
| 2479 | /* The page is "safe", add it to the list */ |
| 2480 | lp->next = safe_pages_list; |
| 2481 | safe_pages_list = lp; |
| 2482 | } |
| 2483 | /* Mark the page as allocated */ |
| 2484 | swsusp_set_page_forbidden(virt_to_page(lp)); |
| 2485 | swsusp_set_page_free(virt_to_page(lp)); |
| 2486 | nr_pages--; |
| 2487 | } |
| 2488 | return 0; |
| 2489 | |
| 2490 | Free: |
| 2491 | swsusp_free(); |
| 2492 | return error; |
| 2493 | } |
| 2494 | |
| 2495 | /** |
| 2496 | * get_buffer - Get the address to store the next image data page. |
| 2497 | * |
| 2498 | * Get the address that snapshot_write_next() should return to its caller to |
| 2499 | * write to. |
| 2500 | */ |
| 2501 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) |
| 2502 | { |
| 2503 | struct pbe *pbe; |
| 2504 | struct page *page; |
| 2505 | unsigned long pfn = memory_bm_next_pfn(bm); |
| 2506 | |
| 2507 | if (pfn == BM_END_OF_MAP) |
| 2508 | return ERR_PTR(-EFAULT); |
| 2509 | |
| 2510 | page = pfn_to_page(pfn); |
| 2511 | if (PageHighMem(page)) |
| 2512 | return get_highmem_page_buffer(page, ca); |
| 2513 | |
| 2514 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
| 2515 | /* |
| 2516 | * We have allocated the "original" page frame and we can |
| 2517 | * use it directly to store the loaded page. |
| 2518 | */ |
| 2519 | return page_address(page); |
| 2520 | |
| 2521 | /* |
| 2522 | * The "original" page frame has not been allocated and we have to |
| 2523 | * use a "safe" page frame to store the loaded page. |
| 2524 | */ |
| 2525 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
| 2526 | if (!pbe) { |
| 2527 | swsusp_free(); |
| 2528 | return ERR_PTR(-ENOMEM); |
| 2529 | } |
| 2530 | pbe->orig_address = page_address(page); |
| 2531 | pbe->address = safe_pages_list; |
| 2532 | safe_pages_list = safe_pages_list->next; |
| 2533 | pbe->next = restore_pblist; |
| 2534 | restore_pblist = pbe; |
| 2535 | return pbe->address; |
| 2536 | } |
| 2537 | |
| 2538 | /** |
| 2539 | * snapshot_write_next - Get the address to store the next image page. |
| 2540 | * @handle: Snapshot handle structure to guide the writing. |
| 2541 | * |
| 2542 | * On the first call, @handle should point to a zeroed snapshot_handle |
| 2543 | * structure. The structure gets populated then and a pointer to it should be |
| 2544 | * passed to this function every next time. |
| 2545 | * |
| 2546 | * On success, the function returns a positive number. Then, the caller |
| 2547 | * is allowed to write up to the returned number of bytes to the memory |
| 2548 | * location computed by the data_of() macro. |
| 2549 | * |
| 2550 | * The function returns 0 to indicate the "end of file" condition. Negative |
| 2551 | * numbers are returned on errors, in which cases the structure pointed to by |
| 2552 | * @handle is not updated and should not be used any more. |
| 2553 | */ |
| 2554 | int snapshot_write_next(struct snapshot_handle *handle) |
| 2555 | { |
| 2556 | static struct chain_allocator ca; |
| 2557 | int error = 0; |
| 2558 | |
| 2559 | /* Check if we have already loaded the entire image */ |
| 2560 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) |
| 2561 | return 0; |
| 2562 | |
| 2563 | handle->sync_read = 1; |
| 2564 | |
| 2565 | if (!handle->cur) { |
| 2566 | if (!buffer) |
| 2567 | /* This makes the buffer be freed by swsusp_free() */ |
| 2568 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
| 2569 | |
| 2570 | if (!buffer) |
| 2571 | return -ENOMEM; |
| 2572 | |
| 2573 | handle->buffer = buffer; |
| 2574 | } else if (handle->cur == 1) { |
| 2575 | error = load_header(buffer); |
| 2576 | if (error) |
| 2577 | return error; |
| 2578 | |
| 2579 | safe_pages_list = NULL; |
| 2580 | |
| 2581 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); |
| 2582 | if (error) |
| 2583 | return error; |
| 2584 | |
| 2585 | /* Allocate buffer for page keys. */ |
| 2586 | error = page_key_alloc(nr_copy_pages); |
| 2587 | if (error) |
| 2588 | return error; |
| 2589 | |
| 2590 | hibernate_restore_protection_begin(); |
| 2591 | } else if (handle->cur <= nr_meta_pages + 1) { |
| 2592 | error = unpack_orig_pfns(buffer, ©_bm); |
| 2593 | if (error) |
| 2594 | return error; |
| 2595 | |
| 2596 | if (handle->cur == nr_meta_pages + 1) { |
| 2597 | error = prepare_image(&orig_bm, ©_bm); |
| 2598 | if (error) |
| 2599 | return error; |
| 2600 | |
| 2601 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); |
| 2602 | memory_bm_position_reset(&orig_bm); |
| 2603 | restore_pblist = NULL; |
| 2604 | handle->buffer = get_buffer(&orig_bm, &ca); |
| 2605 | handle->sync_read = 0; |
| 2606 | if (IS_ERR(handle->buffer)) |
| 2607 | return PTR_ERR(handle->buffer); |
| 2608 | } |
| 2609 | } else { |
| 2610 | copy_last_highmem_page(); |
| 2611 | /* Restore page key for data page (s390 only). */ |
| 2612 | page_key_write(handle->buffer); |
| 2613 | hibernate_restore_protect_page(handle->buffer); |
| 2614 | handle->buffer = get_buffer(&orig_bm, &ca); |
| 2615 | if (IS_ERR(handle->buffer)) |
| 2616 | return PTR_ERR(handle->buffer); |
| 2617 | if (handle->buffer != buffer) |
| 2618 | handle->sync_read = 0; |
| 2619 | } |
| 2620 | handle->cur++; |
| 2621 | return PAGE_SIZE; |
| 2622 | } |
| 2623 | |
| 2624 | /** |
| 2625 | * snapshot_write_finalize - Complete the loading of a hibernation image. |
| 2626 | * |
| 2627 | * Must be called after the last call to snapshot_write_next() in case the last |
| 2628 | * page in the image happens to be a highmem page and its contents should be |
| 2629 | * stored in highmem. Additionally, it recycles bitmap memory that's not |
| 2630 | * necessary any more. |
| 2631 | */ |
| 2632 | void snapshot_write_finalize(struct snapshot_handle *handle) |
| 2633 | { |
| 2634 | copy_last_highmem_page(); |
| 2635 | /* Restore page key for data page (s390 only). */ |
| 2636 | page_key_write(handle->buffer); |
| 2637 | page_key_free(); |
| 2638 | hibernate_restore_protect_page(handle->buffer); |
| 2639 | /* Do that only if we have loaded the image entirely */ |
| 2640 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) { |
| 2641 | memory_bm_recycle(&orig_bm); |
| 2642 | free_highmem_data(); |
| 2643 | } |
| 2644 | } |
| 2645 | |
| 2646 | int snapshot_image_loaded(struct snapshot_handle *handle) |
| 2647 | { |
| 2648 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
| 2649 | handle->cur <= nr_meta_pages + nr_copy_pages); |
| 2650 | } |
| 2651 | |
| 2652 | #ifdef CONFIG_HIGHMEM |
| 2653 | /* Assumes that @buf is ready and points to a "safe" page */ |
| 2654 | static inline void swap_two_pages_data(struct page *p1, struct page *p2, |
| 2655 | void *buf) |
| 2656 | { |
| 2657 | void *kaddr1, *kaddr2; |
| 2658 | |
| 2659 | kaddr1 = kmap_atomic(p1); |
| 2660 | kaddr2 = kmap_atomic(p2); |
| 2661 | copy_page(buf, kaddr1); |
| 2662 | copy_page(kaddr1, kaddr2); |
| 2663 | copy_page(kaddr2, buf); |
| 2664 | kunmap_atomic(kaddr2); |
| 2665 | kunmap_atomic(kaddr1); |
| 2666 | } |
| 2667 | |
| 2668 | /** |
| 2669 | * restore_highmem - Put highmem image pages into their original locations. |
| 2670 | * |
| 2671 | * For each highmem page that was in use before hibernation and is included in |
| 2672 | * the image, and also has been allocated by the "restore" kernel, swap its |
| 2673 | * current contents with the previous (ie. "before hibernation") ones. |
| 2674 | * |
| 2675 | * If the restore eventually fails, we can call this function once again and |
| 2676 | * restore the highmem state as seen by the restore kernel. |
| 2677 | */ |
| 2678 | int restore_highmem(void) |
| 2679 | { |
| 2680 | struct highmem_pbe *pbe = highmem_pblist; |
| 2681 | void *buf; |
| 2682 | |
| 2683 | if (!pbe) |
| 2684 | return 0; |
| 2685 | |
| 2686 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); |
| 2687 | if (!buf) |
| 2688 | return -ENOMEM; |
| 2689 | |
| 2690 | while (pbe) { |
| 2691 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); |
| 2692 | pbe = pbe->next; |
| 2693 | } |
| 2694 | free_image_page(buf, PG_UNSAFE_CLEAR); |
| 2695 | return 0; |
| 2696 | } |
| 2697 | #endif /* CONFIG_HIGHMEM */ |