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/sched/signal.h> 10 #include <linux/swap.h> 11 #include <linux/interrupt.h> 12 #include <linux/pagemap.h> 13 #include <linux/compiler.h> 14 #include <linux/export.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/memremap.h> 22 #include <linux/memory_hotplug.h> 23 #include <linux/highmem.h> 24 #include <linux/vmalloc.h> 25 #include <linux/ioport.h> 26 #include <linux/delay.h> 27 #include <linux/migrate.h> 28 #include <linux/page-isolation.h> 29 #include <linux/pfn.h> 30 #include <linux/suspend.h> 31 #include <linux/mm_inline.h> 32 #include <linux/firmware-map.h> 33 #include <linux/stop_machine.h> 34 #include <linux/hugetlb.h> 35 #include <linux/memblock.h> 36 #include <linux/compaction.h> 37 38 #include <asm/tlbflush.h> 39 40 #include "internal.h" 41 42 /* 43 * online_page_callback contains pointer to current page onlining function. 44 * Initially it is generic_online_page(). If it is required it could be 45 * changed by calling set_online_page_callback() for callback registration 46 * and restore_online_page_callback() for generic callback restore. 47 */ 48 49 static void generic_online_page(struct page *page); 50 51 static online_page_callback_t online_page_callback = generic_online_page; 52 static DEFINE_MUTEX(online_page_callback_lock); 53 54 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); 55 56 void get_online_mems(void) 57 { 58 percpu_down_read(&mem_hotplug_lock); 59 } 60 61 void put_online_mems(void) 62 { 63 percpu_up_read(&mem_hotplug_lock); 64 } 65 66 bool movable_node_enabled = false; 67 68 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE 69 bool memhp_auto_online; 70 #else 71 bool memhp_auto_online = true; 72 #endif 73 EXPORT_SYMBOL_GPL(memhp_auto_online); 74 75 static int __init setup_memhp_default_state(char *str) 76 { 77 if (!strcmp(str, "online")) 78 memhp_auto_online = true; 79 else if (!strcmp(str, "offline")) 80 memhp_auto_online = false; 81 82 return 1; 83 } 84 __setup("memhp_default_state=", setup_memhp_default_state); 85 86 void mem_hotplug_begin(void) 87 { 88 cpus_read_lock(); 89 percpu_down_write(&mem_hotplug_lock); 90 } 91 92 void mem_hotplug_done(void) 93 { 94 percpu_up_write(&mem_hotplug_lock); 95 cpus_read_unlock(); 96 } 97 98 /* add this memory to iomem resource */ 99 static struct resource *register_memory_resource(u64 start, u64 size) 100 { 101 struct resource *res, *conflict; 102 res = kzalloc(sizeof(struct resource), GFP_KERNEL); 103 if (!res) 104 return ERR_PTR(-ENOMEM); 105 106 res->name = "System RAM"; 107 res->start = start; 108 res->end = start + size - 1; 109 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; 110 conflict = request_resource_conflict(&iomem_resource, res); 111 if (conflict) { 112 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) { 113 pr_debug("Device unaddressable memory block " 114 "memory hotplug at %#010llx !\n", 115 (unsigned long long)start); 116 } 117 pr_debug("System RAM resource %pR cannot be added\n", res); 118 kfree(res); 119 return ERR_PTR(-EEXIST); 120 } 121 return res; 122 } 123 124 static void release_memory_resource(struct resource *res) 125 { 126 if (!res) 127 return; 128 release_resource(res); 129 kfree(res); 130 return; 131 } 132 133 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE 134 void get_page_bootmem(unsigned long info, struct page *page, 135 unsigned long type) 136 { 137 page->freelist = (void *)type; 138 SetPagePrivate(page); 139 set_page_private(page, info); 140 page_ref_inc(page); 141 } 142 143 void put_page_bootmem(struct page *page) 144 { 145 unsigned long type; 146 147 type = (unsigned long) page->freelist; 148 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || 149 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); 150 151 if (page_ref_dec_return(page) == 1) { 152 page->freelist = NULL; 153 ClearPagePrivate(page); 154 set_page_private(page, 0); 155 INIT_LIST_HEAD(&page->lru); 156 free_reserved_page(page); 157 } 158 } 159 160 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE 161 #ifndef CONFIG_SPARSEMEM_VMEMMAP 162 static void register_page_bootmem_info_section(unsigned long start_pfn) 163 { 164 unsigned long *usemap, mapsize, section_nr, i; 165 struct mem_section *ms; 166 struct page *page, *memmap; 167 168 section_nr = pfn_to_section_nr(start_pfn); 169 ms = __nr_to_section(section_nr); 170 171 /* Get section's memmap address */ 172 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); 173 174 /* 175 * Get page for the memmap's phys address 176 * XXX: need more consideration for sparse_vmemmap... 177 */ 178 page = virt_to_page(memmap); 179 mapsize = sizeof(struct page) * PAGES_PER_SECTION; 180 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; 181 182 /* remember memmap's page */ 183 for (i = 0; i < mapsize; i++, page++) 184 get_page_bootmem(section_nr, page, SECTION_INFO); 185 186 usemap = ms->pageblock_flags; 187 page = virt_to_page(usemap); 188 189 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; 190 191 for (i = 0; i < mapsize; i++, page++) 192 get_page_bootmem(section_nr, page, MIX_SECTION_INFO); 193 194 } 195 #else /* CONFIG_SPARSEMEM_VMEMMAP */ 196 static void register_page_bootmem_info_section(unsigned long start_pfn) 197 { 198 unsigned long *usemap, mapsize, section_nr, i; 199 struct mem_section *ms; 200 struct page *page, *memmap; 201 202 section_nr = pfn_to_section_nr(start_pfn); 203 ms = __nr_to_section(section_nr); 204 205 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); 206 207 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); 208 209 usemap = ms->pageblock_flags; 210 page = virt_to_page(usemap); 211 212 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; 213 214 for (i = 0; i < mapsize; i++, page++) 215 get_page_bootmem(section_nr, page, MIX_SECTION_INFO); 216 } 217 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ 218 219 void __init register_page_bootmem_info_node(struct pglist_data *pgdat) 220 { 221 unsigned long i, pfn, end_pfn, nr_pages; 222 int node = pgdat->node_id; 223 struct page *page; 224 225 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; 226 page = virt_to_page(pgdat); 227 228 for (i = 0; i < nr_pages; i++, page++) 229 get_page_bootmem(node, page, NODE_INFO); 230 231 pfn = pgdat->node_start_pfn; 232 end_pfn = pgdat_end_pfn(pgdat); 233 234 /* register section info */ 235 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { 236 /* 237 * Some platforms can assign the same pfn to multiple nodes - on 238 * node0 as well as nodeN. To avoid registering a pfn against 239 * multiple nodes we check that this pfn does not already 240 * reside in some other nodes. 241 */ 242 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) 243 register_page_bootmem_info_section(pfn); 244 } 245 } 246 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ 247 248 static int __meminit __add_section(int nid, unsigned long phys_start_pfn, 249 struct vmem_altmap *altmap, bool want_memblock) 250 { 251 int ret; 252 253 if (pfn_valid(phys_start_pfn)) 254 return -EEXIST; 255 256 ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn, altmap); 257 if (ret < 0) 258 return ret; 259 260 if (!want_memblock) 261 return 0; 262 263 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn)); 264 } 265 266 /* 267 * Reasonably generic function for adding memory. It is 268 * expected that archs that support memory hotplug will 269 * call this function after deciding the zone to which to 270 * add the new pages. 271 */ 272 int __ref __add_pages(int nid, unsigned long phys_start_pfn, 273 unsigned long nr_pages, struct vmem_altmap *altmap, 274 bool want_memblock) 275 { 276 unsigned long i; 277 int err = 0; 278 int start_sec, end_sec; 279 280 /* during initialize mem_map, align hot-added range to section */ 281 start_sec = pfn_to_section_nr(phys_start_pfn); 282 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); 283 284 if (altmap) { 285 /* 286 * Validate altmap is within bounds of the total request 287 */ 288 if (altmap->base_pfn != phys_start_pfn 289 || vmem_altmap_offset(altmap) > nr_pages) { 290 pr_warn_once("memory add fail, invalid altmap\n"); 291 err = -EINVAL; 292 goto out; 293 } 294 altmap->alloc = 0; 295 } 296 297 for (i = start_sec; i <= end_sec; i++) { 298 err = __add_section(nid, section_nr_to_pfn(i), altmap, 299 want_memblock); 300 301 /* 302 * EEXIST is finally dealt with by ioresource collision 303 * check. see add_memory() => register_memory_resource() 304 * Warning will be printed if there is collision. 305 */ 306 if (err && (err != -EEXIST)) 307 break; 308 err = 0; 309 cond_resched(); 310 } 311 vmemmap_populate_print_last(); 312 out: 313 return err; 314 } 315 316 #ifdef CONFIG_MEMORY_HOTREMOVE 317 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ 318 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, 319 unsigned long start_pfn, 320 unsigned long end_pfn) 321 { 322 struct mem_section *ms; 323 324 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { 325 ms = __pfn_to_section(start_pfn); 326 327 if (unlikely(!valid_section(ms))) 328 continue; 329 330 if (unlikely(pfn_to_nid(start_pfn) != nid)) 331 continue; 332 333 if (zone && zone != page_zone(pfn_to_page(start_pfn))) 334 continue; 335 336 return start_pfn; 337 } 338 339 return 0; 340 } 341 342 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ 343 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, 344 unsigned long start_pfn, 345 unsigned long end_pfn) 346 { 347 struct mem_section *ms; 348 unsigned long pfn; 349 350 /* pfn is the end pfn of a memory section. */ 351 pfn = end_pfn - 1; 352 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { 353 ms = __pfn_to_section(pfn); 354 355 if (unlikely(!valid_section(ms))) 356 continue; 357 358 if (unlikely(pfn_to_nid(pfn) != nid)) 359 continue; 360 361 if (zone && zone != page_zone(pfn_to_page(pfn))) 362 continue; 363 364 return pfn; 365 } 366 367 return 0; 368 } 369 370 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, 371 unsigned long end_pfn) 372 { 373 unsigned long zone_start_pfn = zone->zone_start_pfn; 374 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ 375 unsigned long zone_end_pfn = z; 376 unsigned long pfn; 377 struct mem_section *ms; 378 int nid = zone_to_nid(zone); 379 380 zone_span_writelock(zone); 381 if (zone_start_pfn == start_pfn) { 382 /* 383 * If the section is smallest section in the zone, it need 384 * shrink zone->zone_start_pfn and zone->zone_spanned_pages. 385 * In this case, we find second smallest valid mem_section 386 * for shrinking zone. 387 */ 388 pfn = find_smallest_section_pfn(nid, zone, end_pfn, 389 zone_end_pfn); 390 if (pfn) { 391 zone->zone_start_pfn = pfn; 392 zone->spanned_pages = zone_end_pfn - pfn; 393 } 394 } else if (zone_end_pfn == end_pfn) { 395 /* 396 * If the section is biggest section in the zone, it need 397 * shrink zone->spanned_pages. 398 * In this case, we find second biggest valid mem_section for 399 * shrinking zone. 400 */ 401 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, 402 start_pfn); 403 if (pfn) 404 zone->spanned_pages = pfn - zone_start_pfn + 1; 405 } 406 407 /* 408 * The section is not biggest or smallest mem_section in the zone, it 409 * only creates a hole in the zone. So in this case, we need not 410 * change the zone. But perhaps, the zone has only hole data. Thus 411 * it check the zone has only hole or not. 412 */ 413 pfn = zone_start_pfn; 414 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { 415 ms = __pfn_to_section(pfn); 416 417 if (unlikely(!valid_section(ms))) 418 continue; 419 420 if (page_zone(pfn_to_page(pfn)) != zone) 421 continue; 422 423 /* If the section is current section, it continues the loop */ 424 if (start_pfn == pfn) 425 continue; 426 427 /* If we find valid section, we have nothing to do */ 428 zone_span_writeunlock(zone); 429 return; 430 } 431 432 /* The zone has no valid section */ 433 zone->zone_start_pfn = 0; 434 zone->spanned_pages = 0; 435 zone_span_writeunlock(zone); 436 } 437 438 static void shrink_pgdat_span(struct pglist_data *pgdat, 439 unsigned long start_pfn, unsigned long end_pfn) 440 { 441 unsigned long pgdat_start_pfn = pgdat->node_start_pfn; 442 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ 443 unsigned long pgdat_end_pfn = p; 444 unsigned long pfn; 445 struct mem_section *ms; 446 int nid = pgdat->node_id; 447 448 if (pgdat_start_pfn == start_pfn) { 449 /* 450 * If the section is smallest section in the pgdat, it need 451 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. 452 * In this case, we find second smallest valid mem_section 453 * for shrinking zone. 454 */ 455 pfn = find_smallest_section_pfn(nid, NULL, end_pfn, 456 pgdat_end_pfn); 457 if (pfn) { 458 pgdat->node_start_pfn = pfn; 459 pgdat->node_spanned_pages = pgdat_end_pfn - pfn; 460 } 461 } else if (pgdat_end_pfn == end_pfn) { 462 /* 463 * If the section is biggest section in the pgdat, it need 464 * shrink pgdat->node_spanned_pages. 465 * In this case, we find second biggest valid mem_section for 466 * shrinking zone. 467 */ 468 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, 469 start_pfn); 470 if (pfn) 471 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; 472 } 473 474 /* 475 * If the section is not biggest or smallest mem_section in the pgdat, 476 * it only creates a hole in the pgdat. So in this case, we need not 477 * change the pgdat. 478 * But perhaps, the pgdat has only hole data. Thus it check the pgdat 479 * has only hole or not. 480 */ 481 pfn = pgdat_start_pfn; 482 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { 483 ms = __pfn_to_section(pfn); 484 485 if (unlikely(!valid_section(ms))) 486 continue; 487 488 if (pfn_to_nid(pfn) != nid) 489 continue; 490 491 /* If the section is current section, it continues the loop */ 492 if (start_pfn == pfn) 493 continue; 494 495 /* If we find valid section, we have nothing to do */ 496 return; 497 } 498 499 /* The pgdat has no valid section */ 500 pgdat->node_start_pfn = 0; 501 pgdat->node_spanned_pages = 0; 502 } 503 504 static void __remove_zone(struct zone *zone, unsigned long start_pfn) 505 { 506 struct pglist_data *pgdat = zone->zone_pgdat; 507 int nr_pages = PAGES_PER_SECTION; 508 unsigned long flags; 509 510 pgdat_resize_lock(zone->zone_pgdat, &flags); 511 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); 512 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); 513 pgdat_resize_unlock(zone->zone_pgdat, &flags); 514 } 515 516 static int __remove_section(struct zone *zone, struct mem_section *ms, 517 unsigned long map_offset, struct vmem_altmap *altmap) 518 { 519 unsigned long start_pfn; 520 int scn_nr; 521 int ret = -EINVAL; 522 523 if (!valid_section(ms)) 524 return ret; 525 526 ret = unregister_memory_section(ms); 527 if (ret) 528 return ret; 529 530 scn_nr = __section_nr(ms); 531 start_pfn = section_nr_to_pfn((unsigned long)scn_nr); 532 __remove_zone(zone, start_pfn); 533 534 sparse_remove_one_section(zone, ms, map_offset, altmap); 535 return 0; 536 } 537 538 /** 539 * __remove_pages() - remove sections of pages from a zone 540 * @zone: zone from which pages need to be removed 541 * @phys_start_pfn: starting pageframe (must be aligned to start of a section) 542 * @nr_pages: number of pages to remove (must be multiple of section size) 543 * @altmap: alternative device page map or %NULL if default memmap is used 544 * 545 * Generic helper function to remove section mappings and sysfs entries 546 * for the section of the memory we are removing. Caller needs to make 547 * sure that pages are marked reserved and zones are adjust properly by 548 * calling offline_pages(). 549 */ 550 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, 551 unsigned long nr_pages, struct vmem_altmap *altmap) 552 { 553 unsigned long i; 554 unsigned long map_offset = 0; 555 int sections_to_remove, ret = 0; 556 557 /* In the ZONE_DEVICE case device driver owns the memory region */ 558 if (is_dev_zone(zone)) { 559 if (altmap) 560 map_offset = vmem_altmap_offset(altmap); 561 } else { 562 resource_size_t start, size; 563 564 start = phys_start_pfn << PAGE_SHIFT; 565 size = nr_pages * PAGE_SIZE; 566 567 ret = release_mem_region_adjustable(&iomem_resource, start, 568 size); 569 if (ret) { 570 resource_size_t endres = start + size - 1; 571 572 pr_warn("Unable to release resource <%pa-%pa> (%d)\n", 573 &start, &endres, ret); 574 } 575 } 576 577 clear_zone_contiguous(zone); 578 579 /* 580 * We can only remove entire sections 581 */ 582 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); 583 BUG_ON(nr_pages % PAGES_PER_SECTION); 584 585 sections_to_remove = nr_pages / PAGES_PER_SECTION; 586 for (i = 0; i < sections_to_remove; i++) { 587 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; 588 589 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset, 590 altmap); 591 map_offset = 0; 592 if (ret) 593 break; 594 } 595 596 set_zone_contiguous(zone); 597 598 return ret; 599 } 600 #endif /* CONFIG_MEMORY_HOTREMOVE */ 601 602 int set_online_page_callback(online_page_callback_t callback) 603 { 604 int rc = -EINVAL; 605 606 get_online_mems(); 607 mutex_lock(&online_page_callback_lock); 608 609 if (online_page_callback == generic_online_page) { 610 online_page_callback = callback; 611 rc = 0; 612 } 613 614 mutex_unlock(&online_page_callback_lock); 615 put_online_mems(); 616 617 return rc; 618 } 619 EXPORT_SYMBOL_GPL(set_online_page_callback); 620 621 int restore_online_page_callback(online_page_callback_t callback) 622 { 623 int rc = -EINVAL; 624 625 get_online_mems(); 626 mutex_lock(&online_page_callback_lock); 627 628 if (online_page_callback == callback) { 629 online_page_callback = generic_online_page; 630 rc = 0; 631 } 632 633 mutex_unlock(&online_page_callback_lock); 634 put_online_mems(); 635 636 return rc; 637 } 638 EXPORT_SYMBOL_GPL(restore_online_page_callback); 639 640 void __online_page_set_limits(struct page *page) 641 { 642 } 643 EXPORT_SYMBOL_GPL(__online_page_set_limits); 644 645 void __online_page_increment_counters(struct page *page) 646 { 647 adjust_managed_page_count(page, 1); 648 } 649 EXPORT_SYMBOL_GPL(__online_page_increment_counters); 650 651 void __online_page_free(struct page *page) 652 { 653 __free_reserved_page(page); 654 } 655 EXPORT_SYMBOL_GPL(__online_page_free); 656 657 static void generic_online_page(struct page *page) 658 { 659 __online_page_set_limits(page); 660 __online_page_increment_counters(page); 661 __online_page_free(page); 662 } 663 664 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, 665 void *arg) 666 { 667 unsigned long i; 668 unsigned long onlined_pages = *(unsigned long *)arg; 669 struct page *page; 670 671 if (PageReserved(pfn_to_page(start_pfn))) 672 for (i = 0; i < nr_pages; i++) { 673 page = pfn_to_page(start_pfn + i); 674 (*online_page_callback)(page); 675 onlined_pages++; 676 } 677 678 online_mem_sections(start_pfn, start_pfn + nr_pages); 679 680 *(unsigned long *)arg = onlined_pages; 681 return 0; 682 } 683 684 /* check which state of node_states will be changed when online memory */ 685 static void node_states_check_changes_online(unsigned long nr_pages, 686 struct zone *zone, struct memory_notify *arg) 687 { 688 int nid = zone_to_nid(zone); 689 690 arg->status_change_nid = -1; 691 arg->status_change_nid_normal = -1; 692 arg->status_change_nid_high = -1; 693 694 if (!node_state(nid, N_MEMORY)) 695 arg->status_change_nid = nid; 696 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) 697 arg->status_change_nid_normal = nid; 698 #ifdef CONFIG_HIGHMEM 699 if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY)) 700 arg->status_change_nid_high = nid; 701 #endif 702 } 703 704 static void node_states_set_node(int node, struct memory_notify *arg) 705 { 706 if (arg->status_change_nid_normal >= 0) 707 node_set_state(node, N_NORMAL_MEMORY); 708 709 if (arg->status_change_nid_high >= 0) 710 node_set_state(node, N_HIGH_MEMORY); 711 712 if (arg->status_change_nid >= 0) 713 node_set_state(node, N_MEMORY); 714 } 715 716 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, 717 unsigned long nr_pages) 718 { 719 unsigned long old_end_pfn = zone_end_pfn(zone); 720 721 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) 722 zone->zone_start_pfn = start_pfn; 723 724 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; 725 } 726 727 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, 728 unsigned long nr_pages) 729 { 730 unsigned long old_end_pfn = pgdat_end_pfn(pgdat); 731 732 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) 733 pgdat->node_start_pfn = start_pfn; 734 735 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; 736 } 737 738 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, 739 unsigned long nr_pages, struct vmem_altmap *altmap) 740 { 741 struct pglist_data *pgdat = zone->zone_pgdat; 742 int nid = pgdat->node_id; 743 unsigned long flags; 744 745 if (zone_is_empty(zone)) 746 init_currently_empty_zone(zone, start_pfn, nr_pages); 747 748 clear_zone_contiguous(zone); 749 750 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ 751 pgdat_resize_lock(pgdat, &flags); 752 zone_span_writelock(zone); 753 resize_zone_range(zone, start_pfn, nr_pages); 754 zone_span_writeunlock(zone); 755 resize_pgdat_range(pgdat, start_pfn, nr_pages); 756 pgdat_resize_unlock(pgdat, &flags); 757 758 /* 759 * TODO now we have a visible range of pages which are not associated 760 * with their zone properly. Not nice but set_pfnblock_flags_mask 761 * expects the zone spans the pfn range. All the pages in the range 762 * are reserved so nobody should be touching them so we should be safe 763 */ 764 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, 765 MEMMAP_HOTPLUG, altmap); 766 767 set_zone_contiguous(zone); 768 } 769 770 /* 771 * Returns a default kernel memory zone for the given pfn range. 772 * If no kernel zone covers this pfn range it will automatically go 773 * to the ZONE_NORMAL. 774 */ 775 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, 776 unsigned long nr_pages) 777 { 778 struct pglist_data *pgdat = NODE_DATA(nid); 779 int zid; 780 781 for (zid = 0; zid <= ZONE_NORMAL; zid++) { 782 struct zone *zone = &pgdat->node_zones[zid]; 783 784 if (zone_intersects(zone, start_pfn, nr_pages)) 785 return zone; 786 } 787 788 return &pgdat->node_zones[ZONE_NORMAL]; 789 } 790 791 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, 792 unsigned long nr_pages) 793 { 794 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, 795 nr_pages); 796 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; 797 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); 798 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); 799 800 /* 801 * We inherit the existing zone in a simple case where zones do not 802 * overlap in the given range 803 */ 804 if (in_kernel ^ in_movable) 805 return (in_kernel) ? kernel_zone : movable_zone; 806 807 /* 808 * If the range doesn't belong to any zone or two zones overlap in the 809 * given range then we use movable zone only if movable_node is 810 * enabled because we always online to a kernel zone by default. 811 */ 812 return movable_node_enabled ? movable_zone : kernel_zone; 813 } 814 815 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, 816 unsigned long nr_pages) 817 { 818 if (online_type == MMOP_ONLINE_KERNEL) 819 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); 820 821 if (online_type == MMOP_ONLINE_MOVABLE) 822 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; 823 824 return default_zone_for_pfn(nid, start_pfn, nr_pages); 825 } 826 827 /* 828 * Associates the given pfn range with the given node and the zone appropriate 829 * for the given online type. 830 */ 831 static struct zone * __meminit move_pfn_range(int online_type, int nid, 832 unsigned long start_pfn, unsigned long nr_pages) 833 { 834 struct zone *zone; 835 836 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); 837 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL); 838 return zone; 839 } 840 841 /* Must be protected by mem_hotplug_begin() or a device_lock */ 842 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) 843 { 844 unsigned long flags; 845 unsigned long onlined_pages = 0; 846 struct zone *zone; 847 int need_zonelists_rebuild = 0; 848 int nid; 849 int ret; 850 struct memory_notify arg; 851 struct memory_block *mem; 852 853 /* 854 * We can't use pfn_to_nid() because nid might be stored in struct page 855 * which is not yet initialized. Instead, we find nid from memory block. 856 */ 857 mem = find_memory_block(__pfn_to_section(pfn)); 858 nid = mem->nid; 859 860 /* associate pfn range with the zone */ 861 zone = move_pfn_range(online_type, nid, pfn, nr_pages); 862 863 arg.start_pfn = pfn; 864 arg.nr_pages = nr_pages; 865 node_states_check_changes_online(nr_pages, zone, &arg); 866 867 ret = memory_notify(MEM_GOING_ONLINE, &arg); 868 ret = notifier_to_errno(ret); 869 if (ret) 870 goto failed_addition; 871 872 /* 873 * If this zone is not populated, then it is not in zonelist. 874 * This means the page allocator ignores this zone. 875 * So, zonelist must be updated after online. 876 */ 877 if (!populated_zone(zone)) { 878 need_zonelists_rebuild = 1; 879 setup_zone_pageset(zone); 880 } 881 882 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, 883 online_pages_range); 884 if (ret) { 885 if (need_zonelists_rebuild) 886 zone_pcp_reset(zone); 887 goto failed_addition; 888 } 889 890 zone->present_pages += onlined_pages; 891 892 pgdat_resize_lock(zone->zone_pgdat, &flags); 893 zone->zone_pgdat->node_present_pages += onlined_pages; 894 pgdat_resize_unlock(zone->zone_pgdat, &flags); 895 896 if (onlined_pages) { 897 node_states_set_node(nid, &arg); 898 if (need_zonelists_rebuild) 899 build_all_zonelists(NULL); 900 else 901 zone_pcp_update(zone); 902 } 903 904 init_per_zone_wmark_min(); 905 906 if (onlined_pages) { 907 kswapd_run(nid); 908 kcompactd_run(nid); 909 } 910 911 vm_total_pages = nr_free_pagecache_pages(); 912 913 writeback_set_ratelimit(); 914 915 if (onlined_pages) 916 memory_notify(MEM_ONLINE, &arg); 917 return 0; 918 919 failed_addition: 920 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", 921 (unsigned long long) pfn << PAGE_SHIFT, 922 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); 923 memory_notify(MEM_CANCEL_ONLINE, &arg); 924 return ret; 925 } 926 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ 927 928 static void reset_node_present_pages(pg_data_t *pgdat) 929 { 930 struct zone *z; 931 932 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 933 z->present_pages = 0; 934 935 pgdat->node_present_pages = 0; 936 } 937 938 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ 939 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) 940 { 941 struct pglist_data *pgdat; 942 unsigned long start_pfn = PFN_DOWN(start); 943 944 pgdat = NODE_DATA(nid); 945 if (!pgdat) { 946 pgdat = arch_alloc_nodedata(nid); 947 if (!pgdat) 948 return NULL; 949 950 arch_refresh_nodedata(nid, pgdat); 951 } else { 952 /* 953 * Reset the nr_zones, order and classzone_idx before reuse. 954 * Note that kswapd will init kswapd_classzone_idx properly 955 * when it starts in the near future. 956 */ 957 pgdat->nr_zones = 0; 958 pgdat->kswapd_order = 0; 959 pgdat->kswapd_classzone_idx = 0; 960 } 961 962 /* we can use NODE_DATA(nid) from here */ 963 964 pgdat->node_id = nid; 965 pgdat->node_start_pfn = start_pfn; 966 967 /* init node's zones as empty zones, we don't have any present pages.*/ 968 free_area_init_core_hotplug(nid); 969 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); 970 971 /* 972 * The node we allocated has no zone fallback lists. For avoiding 973 * to access not-initialized zonelist, build here. 974 */ 975 build_all_zonelists(pgdat); 976 977 /* 978 * When memory is hot-added, all the memory is in offline state. So 979 * clear all zones' present_pages because they will be updated in 980 * online_pages() and offline_pages(). 981 */ 982 reset_node_managed_pages(pgdat); 983 reset_node_present_pages(pgdat); 984 985 return pgdat; 986 } 987 988 static void rollback_node_hotadd(int nid) 989 { 990 pg_data_t *pgdat = NODE_DATA(nid); 991 992 arch_refresh_nodedata(nid, NULL); 993 free_percpu(pgdat->per_cpu_nodestats); 994 arch_free_nodedata(pgdat); 995 return; 996 } 997 998 999 /** 1000 * try_online_node - online a node if offlined 1001 * @nid: the node ID 1002 * @start: start addr of the node 1003 * @set_node_online: Whether we want to online the node 1004 * called by cpu_up() to online a node without onlined memory. 1005 * 1006 * Returns: 1007 * 1 -> a new node has been allocated 1008 * 0 -> the node is already online 1009 * -ENOMEM -> the node could not be allocated 1010 */ 1011 static int __try_online_node(int nid, u64 start, bool set_node_online) 1012 { 1013 pg_data_t *pgdat; 1014 int ret = 1; 1015 1016 if (node_online(nid)) 1017 return 0; 1018 1019 pgdat = hotadd_new_pgdat(nid, start); 1020 if (!pgdat) { 1021 pr_err("Cannot online node %d due to NULL pgdat\n", nid); 1022 ret = -ENOMEM; 1023 goto out; 1024 } 1025 1026 if (set_node_online) { 1027 node_set_online(nid); 1028 ret = register_one_node(nid); 1029 BUG_ON(ret); 1030 } 1031 out: 1032 return ret; 1033 } 1034 1035 /* 1036 * Users of this function always want to online/register the node 1037 */ 1038 int try_online_node(int nid) 1039 { 1040 int ret; 1041 1042 mem_hotplug_begin(); 1043 ret = __try_online_node(nid, 0, true); 1044 mem_hotplug_done(); 1045 return ret; 1046 } 1047 1048 static int check_hotplug_memory_range(u64 start, u64 size) 1049 { 1050 unsigned long block_sz = memory_block_size_bytes(); 1051 u64 block_nr_pages = block_sz >> PAGE_SHIFT; 1052 u64 nr_pages = size >> PAGE_SHIFT; 1053 u64 start_pfn = PFN_DOWN(start); 1054 1055 /* memory range must be block size aligned */ 1056 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) || 1057 !IS_ALIGNED(nr_pages, block_nr_pages)) { 1058 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", 1059 block_sz, start, size); 1060 return -EINVAL; 1061 } 1062 1063 return 0; 1064 } 1065 1066 static int online_memory_block(struct memory_block *mem, void *arg) 1067 { 1068 return device_online(&mem->dev); 1069 } 1070 1071 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ 1072 int __ref add_memory_resource(int nid, struct resource *res, bool online) 1073 { 1074 u64 start, size; 1075 bool new_node = false; 1076 int ret; 1077 1078 start = res->start; 1079 size = resource_size(res); 1080 1081 ret = check_hotplug_memory_range(start, size); 1082 if (ret) 1083 return ret; 1084 1085 mem_hotplug_begin(); 1086 1087 /* 1088 * Add new range to memblock so that when hotadd_new_pgdat() is called 1089 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find 1090 * this new range and calculate total pages correctly. The range will 1091 * be removed at hot-remove time. 1092 */ 1093 memblock_add_node(start, size, nid); 1094 1095 ret = __try_online_node(nid, start, false); 1096 if (ret < 0) 1097 goto error; 1098 new_node = ret; 1099 1100 /* call arch's memory hotadd */ 1101 ret = arch_add_memory(nid, start, size, NULL, true); 1102 if (ret < 0) 1103 goto error; 1104 1105 if (new_node) { 1106 /* If sysfs file of new node can't be created, cpu on the node 1107 * can't be hot-added. There is no rollback way now. 1108 * So, check by BUG_ON() to catch it reluctantly.. 1109 * We online node here. We can't roll back from here. 1110 */ 1111 node_set_online(nid); 1112 ret = __register_one_node(nid); 1113 BUG_ON(ret); 1114 } 1115 1116 /* link memory sections under this node.*/ 1117 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); 1118 BUG_ON(ret); 1119 1120 /* create new memmap entry */ 1121 firmware_map_add_hotplug(start, start + size, "System RAM"); 1122 1123 /* online pages if requested */ 1124 if (online) 1125 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), 1126 NULL, online_memory_block); 1127 1128 goto out; 1129 1130 error: 1131 /* rollback pgdat allocation and others */ 1132 if (new_node) 1133 rollback_node_hotadd(nid); 1134 memblock_remove(start, size); 1135 1136 out: 1137 mem_hotplug_done(); 1138 return ret; 1139 } 1140 EXPORT_SYMBOL_GPL(add_memory_resource); 1141 1142 int __ref add_memory(int nid, u64 start, u64 size) 1143 { 1144 struct resource *res; 1145 int ret; 1146 1147 res = register_memory_resource(start, size); 1148 if (IS_ERR(res)) 1149 return PTR_ERR(res); 1150 1151 ret = add_memory_resource(nid, res, memhp_auto_online); 1152 if (ret < 0) 1153 release_memory_resource(res); 1154 return ret; 1155 } 1156 EXPORT_SYMBOL_GPL(add_memory); 1157 1158 #ifdef CONFIG_MEMORY_HOTREMOVE 1159 /* 1160 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy 1161 * set and the size of the free page is given by page_order(). Using this, 1162 * the function determines if the pageblock contains only free pages. 1163 * Due to buddy contraints, a free page at least the size of a pageblock will 1164 * be located at the start of the pageblock 1165 */ 1166 static inline int pageblock_free(struct page *page) 1167 { 1168 return PageBuddy(page) && page_order(page) >= pageblock_order; 1169 } 1170 1171 /* Return the start of the next active pageblock after a given page */ 1172 static struct page *next_active_pageblock(struct page *page) 1173 { 1174 /* Ensure the starting page is pageblock-aligned */ 1175 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); 1176 1177 /* If the entire pageblock is free, move to the end of free page */ 1178 if (pageblock_free(page)) { 1179 int order; 1180 /* be careful. we don't have locks, page_order can be changed.*/ 1181 order = page_order(page); 1182 if ((order < MAX_ORDER) && (order >= pageblock_order)) 1183 return page + (1 << order); 1184 } 1185 1186 return page + pageblock_nr_pages; 1187 } 1188 1189 static bool is_pageblock_removable_nolock(struct page *page) 1190 { 1191 struct zone *zone; 1192 unsigned long pfn; 1193 1194 /* 1195 * We have to be careful here because we are iterating over memory 1196 * sections which are not zone aware so we might end up outside of 1197 * the zone but still within the section. 1198 * We have to take care about the node as well. If the node is offline 1199 * its NODE_DATA will be NULL - see page_zone. 1200 */ 1201 if (!node_online(page_to_nid(page))) 1202 return false; 1203 1204 zone = page_zone(page); 1205 pfn = page_to_pfn(page); 1206 if (!zone_spans_pfn(zone, pfn)) 1207 return false; 1208 1209 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true); 1210 } 1211 1212 /* Checks if this range of memory is likely to be hot-removable. */ 1213 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) 1214 { 1215 struct page *page = pfn_to_page(start_pfn); 1216 struct page *end_page = page + nr_pages; 1217 1218 /* Check the starting page of each pageblock within the range */ 1219 for (; page < end_page; page = next_active_pageblock(page)) { 1220 if (!is_pageblock_removable_nolock(page)) 1221 return false; 1222 cond_resched(); 1223 } 1224 1225 /* All pageblocks in the memory block are likely to be hot-removable */ 1226 return true; 1227 } 1228 1229 /* 1230 * Confirm all pages in a range [start, end) belong to the same zone. 1231 * When true, return its valid [start, end). 1232 */ 1233 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, 1234 unsigned long *valid_start, unsigned long *valid_end) 1235 { 1236 unsigned long pfn, sec_end_pfn; 1237 unsigned long start, end; 1238 struct zone *zone = NULL; 1239 struct page *page; 1240 int i; 1241 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); 1242 pfn < end_pfn; 1243 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { 1244 /* Make sure the memory section is present first */ 1245 if (!present_section_nr(pfn_to_section_nr(pfn))) 1246 continue; 1247 for (; pfn < sec_end_pfn && pfn < end_pfn; 1248 pfn += MAX_ORDER_NR_PAGES) { 1249 i = 0; 1250 /* This is just a CONFIG_HOLES_IN_ZONE check.*/ 1251 while ((i < MAX_ORDER_NR_PAGES) && 1252 !pfn_valid_within(pfn + i)) 1253 i++; 1254 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) 1255 continue; 1256 page = pfn_to_page(pfn + i); 1257 if (zone && page_zone(page) != zone) 1258 return 0; 1259 if (!zone) 1260 start = pfn + i; 1261 zone = page_zone(page); 1262 end = pfn + MAX_ORDER_NR_PAGES; 1263 } 1264 } 1265 1266 if (zone) { 1267 *valid_start = start; 1268 *valid_end = min(end, end_pfn); 1269 return 1; 1270 } else { 1271 return 0; 1272 } 1273 } 1274 1275 /* 1276 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, 1277 * non-lru movable pages and hugepages). We scan pfn because it's much 1278 * easier than scanning over linked list. This function returns the pfn 1279 * of the first found movable page if it's found, otherwise 0. 1280 */ 1281 static unsigned long scan_movable_pages(unsigned long start, unsigned long end) 1282 { 1283 unsigned long pfn; 1284 struct page *page; 1285 for (pfn = start; pfn < end; pfn++) { 1286 if (pfn_valid(pfn)) { 1287 page = pfn_to_page(pfn); 1288 if (PageLRU(page)) 1289 return pfn; 1290 if (__PageMovable(page)) 1291 return pfn; 1292 if (PageHuge(page)) { 1293 if (hugepage_migration_supported(page_hstate(page)) && 1294 page_huge_active(page)) 1295 return pfn; 1296 else 1297 pfn = round_up(pfn + 1, 1298 1 << compound_order(page)) - 1; 1299 } 1300 } 1301 } 1302 return 0; 1303 } 1304 1305 static struct page *new_node_page(struct page *page, unsigned long private) 1306 { 1307 int nid = page_to_nid(page); 1308 nodemask_t nmask = node_states[N_MEMORY]; 1309 1310 /* 1311 * try to allocate from a different node but reuse this node if there 1312 * are no other online nodes to be used (e.g. we are offlining a part 1313 * of the only existing node) 1314 */ 1315 node_clear(nid, nmask); 1316 if (nodes_empty(nmask)) 1317 node_set(nid, nmask); 1318 1319 return new_page_nodemask(page, nid, &nmask); 1320 } 1321 1322 #define NR_OFFLINE_AT_ONCE_PAGES (256) 1323 static int 1324 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) 1325 { 1326 unsigned long pfn; 1327 struct page *page; 1328 int move_pages = NR_OFFLINE_AT_ONCE_PAGES; 1329 int not_managed = 0; 1330 int ret = 0; 1331 LIST_HEAD(source); 1332 1333 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { 1334 if (!pfn_valid(pfn)) 1335 continue; 1336 page = pfn_to_page(pfn); 1337 1338 if (PageHuge(page)) { 1339 struct page *head = compound_head(page); 1340 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; 1341 if (compound_order(head) > PFN_SECTION_SHIFT) { 1342 ret = -EBUSY; 1343 break; 1344 } 1345 if (isolate_huge_page(page, &source)) 1346 move_pages -= 1 << compound_order(head); 1347 continue; 1348 } else if (PageTransHuge(page)) 1349 pfn = page_to_pfn(compound_head(page)) 1350 + hpage_nr_pages(page) - 1; 1351 1352 if (!get_page_unless_zero(page)) 1353 continue; 1354 /* 1355 * We can skip free pages. And we can deal with pages on 1356 * LRU and non-lru movable pages. 1357 */ 1358 if (PageLRU(page)) 1359 ret = isolate_lru_page(page); 1360 else 1361 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); 1362 if (!ret) { /* Success */ 1363 put_page(page); 1364 list_add_tail(&page->lru, &source); 1365 move_pages--; 1366 if (!__PageMovable(page)) 1367 inc_node_page_state(page, NR_ISOLATED_ANON + 1368 page_is_file_cache(page)); 1369 1370 } else { 1371 #ifdef CONFIG_DEBUG_VM 1372 pr_alert("failed to isolate pfn %lx\n", pfn); 1373 dump_page(page, "isolation failed"); 1374 #endif 1375 put_page(page); 1376 /* Because we don't have big zone->lock. we should 1377 check this again here. */ 1378 if (page_count(page)) { 1379 not_managed++; 1380 ret = -EBUSY; 1381 break; 1382 } 1383 } 1384 } 1385 if (!list_empty(&source)) { 1386 if (not_managed) { 1387 putback_movable_pages(&source); 1388 goto out; 1389 } 1390 1391 /* Allocate a new page from the nearest neighbor node */ 1392 ret = migrate_pages(&source, new_node_page, NULL, 0, 1393 MIGRATE_SYNC, MR_MEMORY_HOTPLUG); 1394 if (ret) 1395 putback_movable_pages(&source); 1396 } 1397 out: 1398 return ret; 1399 } 1400 1401 /* 1402 * remove from free_area[] and mark all as Reserved. 1403 */ 1404 static int 1405 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, 1406 void *data) 1407 { 1408 __offline_isolated_pages(start, start + nr_pages); 1409 return 0; 1410 } 1411 1412 static void 1413 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) 1414 { 1415 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, 1416 offline_isolated_pages_cb); 1417 } 1418 1419 /* 1420 * Check all pages in range, recoreded as memory resource, are isolated. 1421 */ 1422 static int 1423 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, 1424 void *data) 1425 { 1426 int ret; 1427 long offlined = *(long *)data; 1428 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); 1429 offlined = nr_pages; 1430 if (!ret) 1431 *(long *)data += offlined; 1432 return ret; 1433 } 1434 1435 static long 1436 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) 1437 { 1438 long offlined = 0; 1439 int ret; 1440 1441 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, 1442 check_pages_isolated_cb); 1443 if (ret < 0) 1444 offlined = (long)ret; 1445 return offlined; 1446 } 1447 1448 static int __init cmdline_parse_movable_node(char *p) 1449 { 1450 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1451 movable_node_enabled = true; 1452 #else 1453 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); 1454 #endif 1455 return 0; 1456 } 1457 early_param("movable_node", cmdline_parse_movable_node); 1458 1459 /* check which state of node_states will be changed when offline memory */ 1460 static void node_states_check_changes_offline(unsigned long nr_pages, 1461 struct zone *zone, struct memory_notify *arg) 1462 { 1463 struct pglist_data *pgdat = zone->zone_pgdat; 1464 unsigned long present_pages = 0; 1465 enum zone_type zt; 1466 1467 arg->status_change_nid = -1; 1468 arg->status_change_nid_normal = -1; 1469 arg->status_change_nid_high = -1; 1470 1471 /* 1472 * Check whether node_states[N_NORMAL_MEMORY] will be changed. 1473 * If the memory to be offline is within the range 1474 * [0..ZONE_NORMAL], and it is the last present memory there, 1475 * the zones in that range will become empty after the offlining, 1476 * thus we can determine that we need to clear the node from 1477 * node_states[N_NORMAL_MEMORY]. 1478 */ 1479 for (zt = 0; zt <= ZONE_NORMAL; zt++) 1480 present_pages += pgdat->node_zones[zt].present_pages; 1481 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) 1482 arg->status_change_nid_normal = zone_to_nid(zone); 1483 1484 #ifdef CONFIG_HIGHMEM 1485 /* 1486 * node_states[N_HIGH_MEMORY] contains nodes which 1487 * have normal memory or high memory. 1488 * Here we add the present_pages belonging to ZONE_HIGHMEM. 1489 * If the zone is within the range of [0..ZONE_HIGHMEM), and 1490 * we determine that the zones in that range become empty, 1491 * we need to clear the node for N_HIGH_MEMORY. 1492 */ 1493 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; 1494 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) 1495 arg->status_change_nid_high = zone_to_nid(zone); 1496 #endif 1497 1498 /* 1499 * We have accounted the pages from [0..ZONE_NORMAL), and 1500 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM 1501 * as well. 1502 * Here we count the possible pages from ZONE_MOVABLE. 1503 * If after having accounted all the pages, we see that the nr_pages 1504 * to be offlined is over or equal to the accounted pages, 1505 * we know that the node will become empty, and so, we can clear 1506 * it for N_MEMORY as well. 1507 */ 1508 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; 1509 1510 if (nr_pages >= present_pages) 1511 arg->status_change_nid = zone_to_nid(zone); 1512 } 1513 1514 static void node_states_clear_node(int node, struct memory_notify *arg) 1515 { 1516 if (arg->status_change_nid_normal >= 0) 1517 node_clear_state(node, N_NORMAL_MEMORY); 1518 1519 if (arg->status_change_nid_high >= 0) 1520 node_clear_state(node, N_HIGH_MEMORY); 1521 1522 if (arg->status_change_nid >= 0) 1523 node_clear_state(node, N_MEMORY); 1524 } 1525 1526 static int __ref __offline_pages(unsigned long start_pfn, 1527 unsigned long end_pfn) 1528 { 1529 unsigned long pfn, nr_pages; 1530 long offlined_pages; 1531 int ret, node; 1532 unsigned long flags; 1533 unsigned long valid_start, valid_end; 1534 struct zone *zone; 1535 struct memory_notify arg; 1536 1537 /* at least, alignment against pageblock is necessary */ 1538 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) 1539 return -EINVAL; 1540 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) 1541 return -EINVAL; 1542 /* This makes hotplug much easier...and readable. 1543 we assume this for now. .*/ 1544 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) 1545 return -EINVAL; 1546 1547 zone = page_zone(pfn_to_page(valid_start)); 1548 node = zone_to_nid(zone); 1549 nr_pages = end_pfn - start_pfn; 1550 1551 /* set above range as isolated */ 1552 ret = start_isolate_page_range(start_pfn, end_pfn, 1553 MIGRATE_MOVABLE, true); 1554 if (ret) 1555 return ret; 1556 1557 arg.start_pfn = start_pfn; 1558 arg.nr_pages = nr_pages; 1559 node_states_check_changes_offline(nr_pages, zone, &arg); 1560 1561 ret = memory_notify(MEM_GOING_OFFLINE, &arg); 1562 ret = notifier_to_errno(ret); 1563 if (ret) 1564 goto failed_removal; 1565 1566 pfn = start_pfn; 1567 repeat: 1568 /* start memory hot removal */ 1569 ret = -EINTR; 1570 if (signal_pending(current)) 1571 goto failed_removal; 1572 1573 cond_resched(); 1574 lru_add_drain_all(); 1575 drain_all_pages(zone); 1576 1577 pfn = scan_movable_pages(start_pfn, end_pfn); 1578 if (pfn) { /* We have movable pages */ 1579 ret = do_migrate_range(pfn, end_pfn); 1580 goto repeat; 1581 } 1582 1583 /* 1584 * dissolve free hugepages in the memory block before doing offlining 1585 * actually in order to make hugetlbfs's object counting consistent. 1586 */ 1587 ret = dissolve_free_huge_pages(start_pfn, end_pfn); 1588 if (ret) 1589 goto failed_removal; 1590 /* check again */ 1591 offlined_pages = check_pages_isolated(start_pfn, end_pfn); 1592 if (offlined_pages < 0) 1593 goto repeat; 1594 pr_info("Offlined Pages %ld\n", offlined_pages); 1595 /* Ok, all of our target is isolated. 1596 We cannot do rollback at this point. */ 1597 offline_isolated_pages(start_pfn, end_pfn); 1598 /* reset pagetype flags and makes migrate type to be MOVABLE */ 1599 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); 1600 /* removal success */ 1601 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); 1602 zone->present_pages -= offlined_pages; 1603 1604 pgdat_resize_lock(zone->zone_pgdat, &flags); 1605 zone->zone_pgdat->node_present_pages -= offlined_pages; 1606 pgdat_resize_unlock(zone->zone_pgdat, &flags); 1607 1608 init_per_zone_wmark_min(); 1609 1610 if (!populated_zone(zone)) { 1611 zone_pcp_reset(zone); 1612 build_all_zonelists(NULL); 1613 } else 1614 zone_pcp_update(zone); 1615 1616 node_states_clear_node(node, &arg); 1617 if (arg.status_change_nid >= 0) { 1618 kswapd_stop(node); 1619 kcompactd_stop(node); 1620 } 1621 1622 vm_total_pages = nr_free_pagecache_pages(); 1623 writeback_set_ratelimit(); 1624 1625 memory_notify(MEM_OFFLINE, &arg); 1626 return 0; 1627 1628 failed_removal: 1629 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", 1630 (unsigned long long) start_pfn << PAGE_SHIFT, 1631 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); 1632 memory_notify(MEM_CANCEL_OFFLINE, &arg); 1633 /* pushback to free area */ 1634 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); 1635 return ret; 1636 } 1637 1638 /* Must be protected by mem_hotplug_begin() or a device_lock */ 1639 int offline_pages(unsigned long start_pfn, unsigned long nr_pages) 1640 { 1641 return __offline_pages(start_pfn, start_pfn + nr_pages); 1642 } 1643 #endif /* CONFIG_MEMORY_HOTREMOVE */ 1644 1645 /** 1646 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) 1647 * @start_pfn: start pfn of the memory range 1648 * @end_pfn: end pfn of the memory range 1649 * @arg: argument passed to func 1650 * @func: callback for each memory section walked 1651 * 1652 * This function walks through all present mem sections in range 1653 * [start_pfn, end_pfn) and call func on each mem section. 1654 * 1655 * Returns the return value of func. 1656 */ 1657 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, 1658 void *arg, int (*func)(struct memory_block *, void *)) 1659 { 1660 struct memory_block *mem = NULL; 1661 struct mem_section *section; 1662 unsigned long pfn, section_nr; 1663 int ret; 1664 1665 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { 1666 section_nr = pfn_to_section_nr(pfn); 1667 if (!present_section_nr(section_nr)) 1668 continue; 1669 1670 section = __nr_to_section(section_nr); 1671 /* same memblock? */ 1672 if (mem) 1673 if ((section_nr >= mem->start_section_nr) && 1674 (section_nr <= mem->end_section_nr)) 1675 continue; 1676 1677 mem = find_memory_block_hinted(section, mem); 1678 if (!mem) 1679 continue; 1680 1681 ret = func(mem, arg); 1682 if (ret) { 1683 kobject_put(&mem->dev.kobj); 1684 return ret; 1685 } 1686 } 1687 1688 if (mem) 1689 kobject_put(&mem->dev.kobj); 1690 1691 return 0; 1692 } 1693 1694 #ifdef CONFIG_MEMORY_HOTREMOVE 1695 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) 1696 { 1697 int ret = !is_memblock_offlined(mem); 1698 1699 if (unlikely(ret)) { 1700 phys_addr_t beginpa, endpa; 1701 1702 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); 1703 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; 1704 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", 1705 &beginpa, &endpa); 1706 } 1707 1708 return ret; 1709 } 1710 1711 static int check_cpu_on_node(pg_data_t *pgdat) 1712 { 1713 int cpu; 1714 1715 for_each_present_cpu(cpu) { 1716 if (cpu_to_node(cpu) == pgdat->node_id) 1717 /* 1718 * the cpu on this node isn't removed, and we can't 1719 * offline this node. 1720 */ 1721 return -EBUSY; 1722 } 1723 1724 return 0; 1725 } 1726 1727 static void unmap_cpu_on_node(pg_data_t *pgdat) 1728 { 1729 #ifdef CONFIG_ACPI_NUMA 1730 int cpu; 1731 1732 for_each_possible_cpu(cpu) 1733 if (cpu_to_node(cpu) == pgdat->node_id) 1734 numa_clear_node(cpu); 1735 #endif 1736 } 1737 1738 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) 1739 { 1740 int ret; 1741 1742 ret = check_cpu_on_node(pgdat); 1743 if (ret) 1744 return ret; 1745 1746 /* 1747 * the node will be offlined when we come here, so we can clear 1748 * the cpu_to_node() now. 1749 */ 1750 1751 unmap_cpu_on_node(pgdat); 1752 return 0; 1753 } 1754 1755 /** 1756 * try_offline_node 1757 * @nid: the node ID 1758 * 1759 * Offline a node if all memory sections and cpus of the node are removed. 1760 * 1761 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 1762 * and online/offline operations before this call. 1763 */ 1764 void try_offline_node(int nid) 1765 { 1766 pg_data_t *pgdat = NODE_DATA(nid); 1767 unsigned long start_pfn = pgdat->node_start_pfn; 1768 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; 1769 unsigned long pfn; 1770 1771 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { 1772 unsigned long section_nr = pfn_to_section_nr(pfn); 1773 1774 if (!present_section_nr(section_nr)) 1775 continue; 1776 1777 if (pfn_to_nid(pfn) != nid) 1778 continue; 1779 1780 /* 1781 * some memory sections of this node are not removed, and we 1782 * can't offline node now. 1783 */ 1784 return; 1785 } 1786 1787 if (check_and_unmap_cpu_on_node(pgdat)) 1788 return; 1789 1790 /* 1791 * all memory/cpu of this node are removed, we can offline this 1792 * node now. 1793 */ 1794 node_set_offline(nid); 1795 unregister_one_node(nid); 1796 } 1797 EXPORT_SYMBOL(try_offline_node); 1798 1799 /** 1800 * remove_memory 1801 * @nid: the node ID 1802 * @start: physical address of the region to remove 1803 * @size: size of the region to remove 1804 * 1805 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 1806 * and online/offline operations before this call, as required by 1807 * try_offline_node(). 1808 */ 1809 void __ref __remove_memory(int nid, u64 start, u64 size) 1810 { 1811 int ret; 1812 1813 BUG_ON(check_hotplug_memory_range(start, size)); 1814 1815 mem_hotplug_begin(); 1816 1817 /* 1818 * All memory blocks must be offlined before removing memory. Check 1819 * whether all memory blocks in question are offline and trigger a BUG() 1820 * if this is not the case. 1821 */ 1822 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, 1823 check_memblock_offlined_cb); 1824 if (ret) 1825 BUG(); 1826 1827 /* remove memmap entry */ 1828 firmware_map_remove(start, start + size, "System RAM"); 1829 memblock_free(start, size); 1830 memblock_remove(start, size); 1831 1832 arch_remove_memory(start, size, NULL); 1833 1834 try_offline_node(nid); 1835 1836 mem_hotplug_done(); 1837 } 1838 1839 void remove_memory(int nid, u64 start, u64 size) 1840 { 1841 lock_device_hotplug(); 1842 __remove_memory(nid, start, size); 1843 unlock_device_hotplug(); 1844 } 1845 EXPORT_SYMBOL_GPL(remove_memory); 1846 #endif /* CONFIG_MEMORY_HOTREMOVE */ 1847