memory_hotplug.c revision 700c2a46e88265326764197d5b8842490bae5569
1/* 2 * linux/mm/memory_hotplug.c 3 * 4 * Copyright (C) 5 */ 6 7#include <linux/stddef.h> 8#include <linux/mm.h> 9#include <linux/swap.h> 10#include <linux/interrupt.h> 11#include <linux/pagemap.h> 12#include <linux/bootmem.h> 13#include <linux/compiler.h> 14#include <linux/module.h> 15#include <linux/pagevec.h> 16#include <linux/writeback.h> 17#include <linux/slab.h> 18#include <linux/sysctl.h> 19#include <linux/cpu.h> 20#include <linux/memory.h> 21#include <linux/memory_hotplug.h> 22#include <linux/highmem.h> 23#include <linux/vmalloc.h> 24#include <linux/ioport.h> 25#include <linux/delay.h> 26#include <linux/migrate.h> 27#include <linux/page-isolation.h> 28#include <linux/pfn.h> 29#include <linux/suspend.h> 30#include <linux/mm_inline.h> 31#include <linux/firmware-map.h> 32 33#include <asm/tlbflush.h> 34 35#include "internal.h" 36 37DEFINE_MUTEX(mem_hotplug_mutex); 38 39void lock_memory_hotplug(void) 40{ 41 mutex_lock(&mem_hotplug_mutex); 42 43 /* for exclusive hibernation if CONFIG_HIBERNATION=y */ 44 lock_system_sleep(); 45} 46 47void unlock_memory_hotplug(void) 48{ 49 unlock_system_sleep(); 50 mutex_unlock(&mem_hotplug_mutex); 51} 52 53 54/* add this memory to iomem resource */ 55static struct resource *register_memory_resource(u64 start, u64 size) 56{ 57 struct resource *res; 58 res = kzalloc(sizeof(struct resource), GFP_KERNEL); 59 BUG_ON(!res); 60 61 res->name = "System RAM"; 62 res->start = start; 63 res->end = start + size - 1; 64 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; 65 if (request_resource(&iomem_resource, res) < 0) { 66 printk("System RAM resource %llx - %llx cannot be added\n", 67 (unsigned long long)res->start, (unsigned long long)res->end); 68 kfree(res); 69 res = NULL; 70 } 71 return res; 72} 73 74static void release_memory_resource(struct resource *res) 75{ 76 if (!res) 77 return; 78 release_resource(res); 79 kfree(res); 80 return; 81} 82 83#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE 84#ifndef CONFIG_SPARSEMEM_VMEMMAP 85static void get_page_bootmem(unsigned long info, struct page *page, 86 unsigned long type) 87{ 88 page->lru.next = (struct list_head *) type; 89 SetPagePrivate(page); 90 set_page_private(page, info); 91 atomic_inc(&page->_count); 92} 93 94/* reference to __meminit __free_pages_bootmem is valid 95 * so use __ref to tell modpost not to generate a warning */ 96void __ref put_page_bootmem(struct page *page) 97{ 98 unsigned long type; 99 100 type = (unsigned long) page->lru.next; 101 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || 102 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); 103 104 if (atomic_dec_return(&page->_count) == 1) { 105 ClearPagePrivate(page); 106 set_page_private(page, 0); 107 INIT_LIST_HEAD(&page->lru); 108 __free_pages_bootmem(page, 0); 109 } 110 111} 112 113static void register_page_bootmem_info_section(unsigned long start_pfn) 114{ 115 unsigned long *usemap, mapsize, section_nr, i; 116 struct mem_section *ms; 117 struct page *page, *memmap; 118 119 if (!pfn_valid(start_pfn)) 120 return; 121 122 section_nr = pfn_to_section_nr(start_pfn); 123 ms = __nr_to_section(section_nr); 124 125 /* Get section's memmap address */ 126 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); 127 128 /* 129 * Get page for the memmap's phys address 130 * XXX: need more consideration for sparse_vmemmap... 131 */ 132 page = virt_to_page(memmap); 133 mapsize = sizeof(struct page) * PAGES_PER_SECTION; 134 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; 135 136 /* remember memmap's page */ 137 for (i = 0; i < mapsize; i++, page++) 138 get_page_bootmem(section_nr, page, SECTION_INFO); 139 140 usemap = __nr_to_section(section_nr)->pageblock_flags; 141 page = virt_to_page(usemap); 142 143 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; 144 145 for (i = 0; i < mapsize; i++, page++) 146 get_page_bootmem(section_nr, page, MIX_SECTION_INFO); 147 148} 149 150void register_page_bootmem_info_node(struct pglist_data *pgdat) 151{ 152 unsigned long i, pfn, end_pfn, nr_pages; 153 int node = pgdat->node_id; 154 struct page *page; 155 struct zone *zone; 156 157 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; 158 page = virt_to_page(pgdat); 159 160 for (i = 0; i < nr_pages; i++, page++) 161 get_page_bootmem(node, page, NODE_INFO); 162 163 zone = &pgdat->node_zones[0]; 164 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { 165 if (zone->wait_table) { 166 nr_pages = zone->wait_table_hash_nr_entries 167 * sizeof(wait_queue_head_t); 168 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; 169 page = virt_to_page(zone->wait_table); 170 171 for (i = 0; i < nr_pages; i++, page++) 172 get_page_bootmem(node, page, NODE_INFO); 173 } 174 } 175 176 pfn = pgdat->node_start_pfn; 177 end_pfn = pfn + pgdat->node_spanned_pages; 178 179 /* register_section info */ 180 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) 181 register_page_bootmem_info_section(pfn); 182 183} 184#endif /* !CONFIG_SPARSEMEM_VMEMMAP */ 185 186static void grow_zone_span(struct zone *zone, unsigned long start_pfn, 187 unsigned long end_pfn) 188{ 189 unsigned long old_zone_end_pfn; 190 191 zone_span_writelock(zone); 192 193 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; 194 if (start_pfn < zone->zone_start_pfn) 195 zone->zone_start_pfn = start_pfn; 196 197 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - 198 zone->zone_start_pfn; 199 200 zone_span_writeunlock(zone); 201} 202 203static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, 204 unsigned long end_pfn) 205{ 206 unsigned long old_pgdat_end_pfn = 207 pgdat->node_start_pfn + pgdat->node_spanned_pages; 208 209 if (start_pfn < pgdat->node_start_pfn) 210 pgdat->node_start_pfn = start_pfn; 211 212 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - 213 pgdat->node_start_pfn; 214} 215 216static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) 217{ 218 struct pglist_data *pgdat = zone->zone_pgdat; 219 int nr_pages = PAGES_PER_SECTION; 220 int nid = pgdat->node_id; 221 int zone_type; 222 unsigned long flags; 223 224 zone_type = zone - pgdat->node_zones; 225 if (!zone->wait_table) { 226 int ret; 227 228 ret = init_currently_empty_zone(zone, phys_start_pfn, 229 nr_pages, MEMMAP_HOTPLUG); 230 if (ret) 231 return ret; 232 } 233 pgdat_resize_lock(zone->zone_pgdat, &flags); 234 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); 235 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, 236 phys_start_pfn + nr_pages); 237 pgdat_resize_unlock(zone->zone_pgdat, &flags); 238 memmap_init_zone(nr_pages, nid, zone_type, 239 phys_start_pfn, MEMMAP_HOTPLUG); 240 return 0; 241} 242 243static int __meminit __add_section(int nid, struct zone *zone, 244 unsigned long phys_start_pfn) 245{ 246 int nr_pages = PAGES_PER_SECTION; 247 int ret; 248 249 if (pfn_valid(phys_start_pfn)) 250 return -EEXIST; 251 252 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); 253 254 if (ret < 0) 255 return ret; 256 257 ret = __add_zone(zone, phys_start_pfn); 258 259 if (ret < 0) 260 return ret; 261 262 return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); 263} 264 265#ifdef CONFIG_SPARSEMEM_VMEMMAP 266static int __remove_section(struct zone *zone, struct mem_section *ms) 267{ 268 /* 269 * XXX: Freeing memmap with vmemmap is not implement yet. 270 * This should be removed later. 271 */ 272 return -EBUSY; 273} 274#else 275static int __remove_section(struct zone *zone, struct mem_section *ms) 276{ 277 unsigned long flags; 278 struct pglist_data *pgdat = zone->zone_pgdat; 279 int ret = -EINVAL; 280 281 if (!valid_section(ms)) 282 return ret; 283 284 ret = unregister_memory_section(ms); 285 if (ret) 286 return ret; 287 288 pgdat_resize_lock(pgdat, &flags); 289 sparse_remove_one_section(zone, ms); 290 pgdat_resize_unlock(pgdat, &flags); 291 return 0; 292} 293#endif 294 295/* 296 * Reasonably generic function for adding memory. It is 297 * expected that archs that support memory hotplug will 298 * call this function after deciding the zone to which to 299 * add the new pages. 300 */ 301int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, 302 unsigned long nr_pages) 303{ 304 unsigned long i; 305 int err = 0; 306 int start_sec, end_sec; 307 /* during initialize mem_map, align hot-added range to section */ 308 start_sec = pfn_to_section_nr(phys_start_pfn); 309 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); 310 311 for (i = start_sec; i <= end_sec; i++) { 312 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); 313 314 /* 315 * EEXIST is finally dealt with by ioresource collision 316 * check. see add_memory() => register_memory_resource() 317 * Warning will be printed if there is collision. 318 */ 319 if (err && (err != -EEXIST)) 320 break; 321 err = 0; 322 } 323 324 return err; 325} 326EXPORT_SYMBOL_GPL(__add_pages); 327 328/** 329 * __remove_pages() - remove sections of pages from a zone 330 * @zone: zone from which pages need to be removed 331 * @phys_start_pfn: starting pageframe (must be aligned to start of a section) 332 * @nr_pages: number of pages to remove (must be multiple of section size) 333 * 334 * Generic helper function to remove section mappings and sysfs entries 335 * for the section of the memory we are removing. Caller needs to make 336 * sure that pages are marked reserved and zones are adjust properly by 337 * calling offline_pages(). 338 */ 339int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, 340 unsigned long nr_pages) 341{ 342 unsigned long i, ret = 0; 343 int sections_to_remove; 344 345 /* 346 * We can only remove entire sections 347 */ 348 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); 349 BUG_ON(nr_pages % PAGES_PER_SECTION); 350 351 sections_to_remove = nr_pages / PAGES_PER_SECTION; 352 for (i = 0; i < sections_to_remove; i++) { 353 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; 354 release_mem_region(pfn << PAGE_SHIFT, 355 PAGES_PER_SECTION << PAGE_SHIFT); 356 ret = __remove_section(zone, __pfn_to_section(pfn)); 357 if (ret) 358 break; 359 } 360 return ret; 361} 362EXPORT_SYMBOL_GPL(__remove_pages); 363 364void online_page(struct page *page) 365{ 366 unsigned long pfn = page_to_pfn(page); 367 368 totalram_pages++; 369 if (pfn >= num_physpages) 370 num_physpages = pfn + 1; 371 372#ifdef CONFIG_HIGHMEM 373 if (PageHighMem(page)) 374 totalhigh_pages++; 375#endif 376 377#ifdef CONFIG_FLATMEM 378 max_mapnr = max(pfn, max_mapnr); 379#endif 380 381 ClearPageReserved(page); 382 init_page_count(page); 383 __free_page(page); 384} 385 386static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, 387 void *arg) 388{ 389 unsigned long i; 390 unsigned long onlined_pages = *(unsigned long *)arg; 391 struct page *page; 392 if (PageReserved(pfn_to_page(start_pfn))) 393 for (i = 0; i < nr_pages; i++) { 394 page = pfn_to_page(start_pfn + i); 395 online_page(page); 396 onlined_pages++; 397 } 398 *(unsigned long *)arg = onlined_pages; 399 return 0; 400} 401 402 403int __ref online_pages(unsigned long pfn, unsigned long nr_pages) 404{ 405 unsigned long onlined_pages = 0; 406 struct zone *zone; 407 int need_zonelists_rebuild = 0; 408 int nid; 409 int ret; 410 struct memory_notify arg; 411 412 lock_memory_hotplug(); 413 arg.start_pfn = pfn; 414 arg.nr_pages = nr_pages; 415 arg.status_change_nid = -1; 416 417 nid = page_to_nid(pfn_to_page(pfn)); 418 if (node_present_pages(nid) == 0) 419 arg.status_change_nid = nid; 420 421 ret = memory_notify(MEM_GOING_ONLINE, &arg); 422 ret = notifier_to_errno(ret); 423 if (ret) { 424 memory_notify(MEM_CANCEL_ONLINE, &arg); 425 unlock_memory_hotplug(); 426 return ret; 427 } 428 /* 429 * This doesn't need a lock to do pfn_to_page(). 430 * The section can't be removed here because of the 431 * memory_block->state_mutex. 432 */ 433 zone = page_zone(pfn_to_page(pfn)); 434 /* 435 * If this zone is not populated, then it is not in zonelist. 436 * This means the page allocator ignores this zone. 437 * So, zonelist must be updated after online. 438 */ 439 mutex_lock(&zonelists_mutex); 440 if (!populated_zone(zone)) 441 need_zonelists_rebuild = 1; 442 443 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, 444 online_pages_range); 445 if (ret) { 446 mutex_unlock(&zonelists_mutex); 447 printk(KERN_DEBUG "online_pages %lx at %lx failed\n", 448 nr_pages, pfn); 449 memory_notify(MEM_CANCEL_ONLINE, &arg); 450 unlock_memory_hotplug(); 451 return ret; 452 } 453 454 zone->present_pages += onlined_pages; 455 zone->zone_pgdat->node_present_pages += onlined_pages; 456 if (need_zonelists_rebuild) 457 build_all_zonelists(zone); 458 else 459 zone_pcp_update(zone); 460 461 mutex_unlock(&zonelists_mutex); 462 463 init_per_zone_wmark_min(); 464 465 if (onlined_pages) { 466 kswapd_run(zone_to_nid(zone)); 467 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY); 468 } 469 470 vm_total_pages = nr_free_pagecache_pages(); 471 472 writeback_set_ratelimit(); 473 474 if (onlined_pages) 475 memory_notify(MEM_ONLINE, &arg); 476 unlock_memory_hotplug(); 477 478 return 0; 479} 480#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ 481 482/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ 483static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) 484{ 485 struct pglist_data *pgdat; 486 unsigned long zones_size[MAX_NR_ZONES] = {0}; 487 unsigned long zholes_size[MAX_NR_ZONES] = {0}; 488 unsigned long start_pfn = start >> PAGE_SHIFT; 489 490 pgdat = arch_alloc_nodedata(nid); 491 if (!pgdat) 492 return NULL; 493 494 arch_refresh_nodedata(nid, pgdat); 495 496 /* we can use NODE_DATA(nid) from here */ 497 498 /* init node's zones as empty zones, we don't have any present pages.*/ 499 free_area_init_node(nid, zones_size, start_pfn, zholes_size); 500 501 return pgdat; 502} 503 504static void rollback_node_hotadd(int nid, pg_data_t *pgdat) 505{ 506 arch_refresh_nodedata(nid, NULL); 507 arch_free_nodedata(pgdat); 508 return; 509} 510 511 512/* 513 * called by cpu_up() to online a node without onlined memory. 514 */ 515int mem_online_node(int nid) 516{ 517 pg_data_t *pgdat; 518 int ret; 519 520 lock_memory_hotplug(); 521 pgdat = hotadd_new_pgdat(nid, 0); 522 if (pgdat) { 523 ret = -ENOMEM; 524 goto out; 525 } 526 node_set_online(nid); 527 ret = register_one_node(nid); 528 BUG_ON(ret); 529 530out: 531 unlock_memory_hotplug(); 532 return ret; 533} 534 535/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ 536int __ref add_memory(int nid, u64 start, u64 size) 537{ 538 pg_data_t *pgdat = NULL; 539 int new_pgdat = 0; 540 struct resource *res; 541 int ret; 542 543 lock_memory_hotplug(); 544 545 res = register_memory_resource(start, size); 546 ret = -EEXIST; 547 if (!res) 548 goto out; 549 550 if (!node_online(nid)) { 551 pgdat = hotadd_new_pgdat(nid, start); 552 ret = -ENOMEM; 553 if (!pgdat) 554 goto out; 555 new_pgdat = 1; 556 } 557 558 /* call arch's memory hotadd */ 559 ret = arch_add_memory(nid, start, size); 560 561 if (ret < 0) 562 goto error; 563 564 /* we online node here. we can't roll back from here. */ 565 node_set_online(nid); 566 567 if (new_pgdat) { 568 ret = register_one_node(nid); 569 /* 570 * If sysfs file of new node can't create, cpu on the node 571 * can't be hot-added. There is no rollback way now. 572 * So, check by BUG_ON() to catch it reluctantly.. 573 */ 574 BUG_ON(ret); 575 } 576 577 /* create new memmap entry */ 578 firmware_map_add_hotplug(start, start + size, "System RAM"); 579 580 goto out; 581 582error: 583 /* rollback pgdat allocation and others */ 584 if (new_pgdat) 585 rollback_node_hotadd(nid, pgdat); 586 if (res) 587 release_memory_resource(res); 588 589out: 590 unlock_memory_hotplug(); 591 return ret; 592} 593EXPORT_SYMBOL_GPL(add_memory); 594 595#ifdef CONFIG_MEMORY_HOTREMOVE 596/* 597 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy 598 * set and the size of the free page is given by page_order(). Using this, 599 * the function determines if the pageblock contains only free pages. 600 * Due to buddy contraints, a free page at least the size of a pageblock will 601 * be located at the start of the pageblock 602 */ 603static inline int pageblock_free(struct page *page) 604{ 605 return PageBuddy(page) && page_order(page) >= pageblock_order; 606} 607 608/* Return the start of the next active pageblock after a given page */ 609static struct page *next_active_pageblock(struct page *page) 610{ 611 /* Ensure the starting page is pageblock-aligned */ 612 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); 613 614 /* If the entire pageblock is free, move to the end of free page */ 615 if (pageblock_free(page)) { 616 int order; 617 /* be careful. we don't have locks, page_order can be changed.*/ 618 order = page_order(page); 619 if ((order < MAX_ORDER) && (order >= pageblock_order)) 620 return page + (1 << order); 621 } 622 623 return page + pageblock_nr_pages; 624} 625 626/* Checks if this range of memory is likely to be hot-removable. */ 627int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) 628{ 629 struct page *page = pfn_to_page(start_pfn); 630 struct page *end_page = page + nr_pages; 631 632 /* Check the starting page of each pageblock within the range */ 633 for (; page < end_page; page = next_active_pageblock(page)) { 634 if (!is_pageblock_removable_nolock(page)) 635 return 0; 636 cond_resched(); 637 } 638 639 /* All pageblocks in the memory block are likely to be hot-removable */ 640 return 1; 641} 642 643/* 644 * Confirm all pages in a range [start, end) is belongs to the same zone. 645 */ 646static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) 647{ 648 unsigned long pfn; 649 struct zone *zone = NULL; 650 struct page *page; 651 int i; 652 for (pfn = start_pfn; 653 pfn < end_pfn; 654 pfn += MAX_ORDER_NR_PAGES) { 655 i = 0; 656 /* This is just a CONFIG_HOLES_IN_ZONE check.*/ 657 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) 658 i++; 659 if (i == MAX_ORDER_NR_PAGES) 660 continue; 661 page = pfn_to_page(pfn + i); 662 if (zone && page_zone(page) != zone) 663 return 0; 664 zone = page_zone(page); 665 } 666 return 1; 667} 668 669/* 670 * Scanning pfn is much easier than scanning lru list. 671 * Scan pfn from start to end and Find LRU page. 672 */ 673static unsigned long scan_lru_pages(unsigned long start, unsigned long end) 674{ 675 unsigned long pfn; 676 struct page *page; 677 for (pfn = start; pfn < end; pfn++) { 678 if (pfn_valid(pfn)) { 679 page = pfn_to_page(pfn); 680 if (PageLRU(page)) 681 return pfn; 682 } 683 } 684 return 0; 685} 686 687static struct page * 688hotremove_migrate_alloc(struct page *page, unsigned long private, int **x) 689{ 690 /* This should be improooooved!! */ 691 return alloc_page(GFP_HIGHUSER_MOVABLE); 692} 693 694#define NR_OFFLINE_AT_ONCE_PAGES (256) 695static int 696do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) 697{ 698 unsigned long pfn; 699 struct page *page; 700 int move_pages = NR_OFFLINE_AT_ONCE_PAGES; 701 int not_managed = 0; 702 int ret = 0; 703 LIST_HEAD(source); 704 705 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { 706 if (!pfn_valid(pfn)) 707 continue; 708 page = pfn_to_page(pfn); 709 if (!get_page_unless_zero(page)) 710 continue; 711 /* 712 * We can skip free pages. And we can only deal with pages on 713 * LRU. 714 */ 715 ret = isolate_lru_page(page); 716 if (!ret) { /* Success */ 717 put_page(page); 718 list_add_tail(&page->lru, &source); 719 move_pages--; 720 inc_zone_page_state(page, NR_ISOLATED_ANON + 721 page_is_file_cache(page)); 722 723 } else { 724#ifdef CONFIG_DEBUG_VM 725 printk(KERN_ALERT "removing pfn %lx from LRU failed\n", 726 pfn); 727 dump_page(page); 728#endif 729 put_page(page); 730 /* Because we don't have big zone->lock. we should 731 check this again here. */ 732 if (page_count(page)) { 733 not_managed++; 734 ret = -EBUSY; 735 break; 736 } 737 } 738 } 739 if (!list_empty(&source)) { 740 if (not_managed) { 741 putback_lru_pages(&source); 742 goto out; 743 } 744 /* this function returns # of failed pages */ 745 ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 746 true, true); 747 if (ret) 748 putback_lru_pages(&source); 749 } 750out: 751 return ret; 752} 753 754/* 755 * remove from free_area[] and mark all as Reserved. 756 */ 757static int 758offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, 759 void *data) 760{ 761 __offline_isolated_pages(start, start + nr_pages); 762 return 0; 763} 764 765static void 766offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) 767{ 768 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, 769 offline_isolated_pages_cb); 770} 771 772/* 773 * Check all pages in range, recoreded as memory resource, are isolated. 774 */ 775static int 776check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, 777 void *data) 778{ 779 int ret; 780 long offlined = *(long *)data; 781 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages); 782 offlined = nr_pages; 783 if (!ret) 784 *(long *)data += offlined; 785 return ret; 786} 787 788static long 789check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) 790{ 791 long offlined = 0; 792 int ret; 793 794 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, 795 check_pages_isolated_cb); 796 if (ret < 0) 797 offlined = (long)ret; 798 return offlined; 799} 800 801static int __ref offline_pages(unsigned long start_pfn, 802 unsigned long end_pfn, unsigned long timeout) 803{ 804 unsigned long pfn, nr_pages, expire; 805 long offlined_pages; 806 int ret, drain, retry_max, node; 807 struct zone *zone; 808 struct memory_notify arg; 809 810 BUG_ON(start_pfn >= end_pfn); 811 /* at least, alignment against pageblock is necessary */ 812 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) 813 return -EINVAL; 814 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) 815 return -EINVAL; 816 /* This makes hotplug much easier...and readable. 817 we assume this for now. .*/ 818 if (!test_pages_in_a_zone(start_pfn, end_pfn)) 819 return -EINVAL; 820 821 lock_memory_hotplug(); 822 823 zone = page_zone(pfn_to_page(start_pfn)); 824 node = zone_to_nid(zone); 825 nr_pages = end_pfn - start_pfn; 826 827 /* set above range as isolated */ 828 ret = start_isolate_page_range(start_pfn, end_pfn); 829 if (ret) 830 goto out; 831 832 arg.start_pfn = start_pfn; 833 arg.nr_pages = nr_pages; 834 arg.status_change_nid = -1; 835 if (nr_pages >= node_present_pages(node)) 836 arg.status_change_nid = node; 837 838 ret = memory_notify(MEM_GOING_OFFLINE, &arg); 839 ret = notifier_to_errno(ret); 840 if (ret) 841 goto failed_removal; 842 843 pfn = start_pfn; 844 expire = jiffies + timeout; 845 drain = 0; 846 retry_max = 5; 847repeat: 848 /* start memory hot removal */ 849 ret = -EAGAIN; 850 if (time_after(jiffies, expire)) 851 goto failed_removal; 852 ret = -EINTR; 853 if (signal_pending(current)) 854 goto failed_removal; 855 ret = 0; 856 if (drain) { 857 lru_add_drain_all(); 858 cond_resched(); 859 drain_all_pages(); 860 } 861 862 pfn = scan_lru_pages(start_pfn, end_pfn); 863 if (pfn) { /* We have page on LRU */ 864 ret = do_migrate_range(pfn, end_pfn); 865 if (!ret) { 866 drain = 1; 867 goto repeat; 868 } else { 869 if (ret < 0) 870 if (--retry_max == 0) 871 goto failed_removal; 872 yield(); 873 drain = 1; 874 goto repeat; 875 } 876 } 877 /* drain all zone's lru pagevec, this is asyncronous... */ 878 lru_add_drain_all(); 879 yield(); 880 /* drain pcp pages , this is synchrouns. */ 881 drain_all_pages(); 882 /* check again */ 883 offlined_pages = check_pages_isolated(start_pfn, end_pfn); 884 if (offlined_pages < 0) { 885 ret = -EBUSY; 886 goto failed_removal; 887 } 888 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); 889 /* Ok, all of our target is islaoted. 890 We cannot do rollback at this point. */ 891 offline_isolated_pages(start_pfn, end_pfn); 892 /* reset pagetype flags and makes migrate type to be MOVABLE */ 893 undo_isolate_page_range(start_pfn, end_pfn); 894 /* removal success */ 895 zone->present_pages -= offlined_pages; 896 zone->zone_pgdat->node_present_pages -= offlined_pages; 897 totalram_pages -= offlined_pages; 898 899 init_per_zone_wmark_min(); 900 901 if (!node_present_pages(node)) { 902 node_clear_state(node, N_HIGH_MEMORY); 903 kswapd_stop(node); 904 } 905 906 vm_total_pages = nr_free_pagecache_pages(); 907 writeback_set_ratelimit(); 908 909 memory_notify(MEM_OFFLINE, &arg); 910 unlock_memory_hotplug(); 911 return 0; 912 913failed_removal: 914 printk(KERN_INFO "memory offlining %lx to %lx failed\n", 915 start_pfn, end_pfn); 916 memory_notify(MEM_CANCEL_OFFLINE, &arg); 917 /* pushback to free area */ 918 undo_isolate_page_range(start_pfn, end_pfn); 919 920out: 921 unlock_memory_hotplug(); 922 return ret; 923} 924 925int remove_memory(u64 start, u64 size) 926{ 927 unsigned long start_pfn, end_pfn; 928 929 start_pfn = PFN_DOWN(start); 930 end_pfn = start_pfn + PFN_DOWN(size); 931 return offline_pages(start_pfn, end_pfn, 120 * HZ); 932} 933#else 934int remove_memory(u64 start, u64 size) 935{ 936 return -EINVAL; 937} 938#endif /* CONFIG_MEMORY_HOTREMOVE */ 939EXPORT_SYMBOL_GPL(remove_memory); 940