mm/memory_hotplug.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
get_online_mems
put_online_mems
setup_memhp_default_state
mem_hotplug_begin
mem_hotplug_done
register_memory_resource
release_memory_resource
get_page_bootmem
put_page_bootmem
register_page_bootmem_info_section
register_page_bootmem_info_section
register_page_bootmem_info_node
check_pfn_span
__add_pages
find_smallest_section_pfn
find_biggest_section_pfn
shrink_zone_span
update_pgdat_span
remove_pfn_range_from_zone
__remove_section
__remove_pages
set_online_page_callback
restore_online_page_callback
__online_page_set_limits
__online_page_increment_counters
__online_page_free
generic_online_page
online_pages_range
node_states_check_changes_online
node_states_set_node
resize_zone_range
resize_pgdat_range
move_pfn_range_to_zone
default_kernel_zone_for_pfn
default_zone_for_pfn
zone_for_pfn_range
online_pages
reset_node_present_pages
hotadd_new_pgdat
rollback_node_hotadd
__try_online_node
try_online_node
check_hotplug_memory_range
online_memory_block
add_memory_resource
__add_memory
add_memory
pageblock_free
next_active_pageblock
is_pageblock_removable_nolock
is_mem_section_removable
test_pages_in_a_zone
scan_movable_pages
new_node_page
do_migrate_range
offline_isolated_pages_cb
check_pages_isolated_cb
cmdline_parse_movable_node
node_states_check_changes_offline
node_states_clear_node
__offline_pages
offline_pages
check_memblock_offlined_cb
check_cpu_on_node
check_no_memblock_for_node_cb
try_offline_node
__release_memory_resource
try_remove_memory
__remove_memory
remove_memory
   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/mm/memory_hotplug.c
   4  *
   5  *  Copyright (C)
   6  */
   7 
   8 #include <linux/stddef.h>
   9 #include <linux/mm.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/swap.h>
  12 #include <linux/interrupt.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/compiler.h>
  15 #include <linux/export.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/writeback.h>
  18 #include <linux/slab.h>
  19 #include <linux/sysctl.h>
  20 #include <linux/cpu.h>
  21 #include <linux/memory.h>
  22 #include <linux/memremap.h>
  23 #include <linux/memory_hotplug.h>
  24 #include <linux/highmem.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/ioport.h>
  27 #include <linux/delay.h>
  28 #include <linux/migrate.h>
  29 #include <linux/page-isolation.h>
  30 #include <linux/pfn.h>
  31 #include <linux/suspend.h>
  32 #include <linux/mm_inline.h>
  33 #include <linux/firmware-map.h>
  34 #include <linux/stop_machine.h>
  35 #include <linux/hugetlb.h>
  36 #include <linux/memblock.h>
  37 #include <linux/compaction.h>
  38 #include <linux/rmap.h>
  39 
  40 #include <asm/tlbflush.h>
  41 
  42 #include "internal.h"
  43 #include "shuffle.h"
  44 
  45 /*
  46  * online_page_callback contains pointer to current page onlining function.
  47  * Initially it is generic_online_page(). If it is required it could be
  48  * changed by calling set_online_page_callback() for callback registration
  49  * and restore_online_page_callback() for generic callback restore.
  50  */
  51 
  52 static void generic_online_page(struct page *page, unsigned int order);
  53 
  54 static online_page_callback_t online_page_callback = generic_online_page;
  55 static DEFINE_MUTEX(online_page_callback_lock);
  56 
  57 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
  58 
  59 void get_online_mems(void)
  60 {
  61         percpu_down_read(&mem_hotplug_lock);
  62 }
  63 
  64 void put_online_mems(void)
  65 {
  66         percpu_up_read(&mem_hotplug_lock);
  67 }
  68 
  69 bool movable_node_enabled = false;
  70 
  71 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
  72 bool memhp_auto_online;
  73 #else
  74 bool memhp_auto_online = true;
  75 #endif
  76 EXPORT_SYMBOL_GPL(memhp_auto_online);
  77 
  78 static int __init setup_memhp_default_state(char *str)
  79 {
  80         if (!strcmp(str, "online"))
  81                 memhp_auto_online = true;
  82         else if (!strcmp(str, "offline"))
  83                 memhp_auto_online = false;
  84 
  85         return 1;
  86 }
  87 __setup("memhp_default_state=", setup_memhp_default_state);
  88 
  89 void mem_hotplug_begin(void)
  90 {
  91         cpus_read_lock();
  92         percpu_down_write(&mem_hotplug_lock);
  93 }
  94 
  95 void mem_hotplug_done(void)
  96 {
  97         percpu_up_write(&mem_hotplug_lock);
  98         cpus_read_unlock();
  99 }
 100 
 101 u64 max_mem_size = U64_MAX;
 102 
 103 /* add this memory to iomem resource */
 104 static struct resource *register_memory_resource(u64 start, u64 size)
 105 {
 106         struct resource *res;
 107         unsigned long flags =  IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 108         char *resource_name = "System RAM";
 109 
 110         if (start + size > max_mem_size)
 111                 return ERR_PTR(-E2BIG);
 112 
 113         /*
 114          * Request ownership of the new memory range.  This might be
 115          * a child of an existing resource that was present but
 116          * not marked as busy.
 117          */
 118         res = __request_region(&iomem_resource, start, size,
 119                                resource_name, flags);
 120 
 121         if (!res) {
 122                 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
 123                                 start, start + size);
 124                 return ERR_PTR(-EEXIST);
 125         }
 126         return res;
 127 }
 128 
 129 static void release_memory_resource(struct resource *res)
 130 {
 131         if (!res)
 132                 return;
 133         release_resource(res);
 134         kfree(res);
 135 }
 136 
 137 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 138 void get_page_bootmem(unsigned long info,  struct page *page,
 139                       unsigned long type)
 140 {
 141         page->freelist = (void *)type;
 142         SetPagePrivate(page);
 143         set_page_private(page, info);
 144         page_ref_inc(page);
 145 }
 146 
 147 void put_page_bootmem(struct page *page)
 148 {
 149         unsigned long type;
 150 
 151         type = (unsigned long) page->freelist;
 152         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
 153                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
 154 
 155         if (page_ref_dec_return(page) == 1) {
 156                 page->freelist = NULL;
 157                 ClearPagePrivate(page);
 158                 set_page_private(page, 0);
 159                 INIT_LIST_HEAD(&page->lru);
 160                 free_reserved_page(page);
 161         }
 162 }
 163 
 164 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
 165 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 166 static void register_page_bootmem_info_section(unsigned long start_pfn)
 167 {
 168         unsigned long mapsize, section_nr, i;
 169         struct mem_section *ms;
 170         struct page *page, *memmap;
 171         struct mem_section_usage *usage;
 172 
 173         section_nr = pfn_to_section_nr(start_pfn);
 174         ms = __nr_to_section(section_nr);
 175 
 176         /* Get section's memmap address */
 177         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 178 
 179         /*
 180          * Get page for the memmap's phys address
 181          * XXX: need more consideration for sparse_vmemmap...
 182          */
 183         page = virt_to_page(memmap);
 184         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 185         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 186 
 187         /* remember memmap's page */
 188         for (i = 0; i < mapsize; i++, page++)
 189                 get_page_bootmem(section_nr, page, SECTION_INFO);
 190 
 191         usage = ms->usage;
 192         page = virt_to_page(usage);
 193 
 194         mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
 195 
 196         for (i = 0; i < mapsize; i++, page++)
 197                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 198 
 199 }
 200 #else /* CONFIG_SPARSEMEM_VMEMMAP */
 201 static void register_page_bootmem_info_section(unsigned long start_pfn)
 202 {
 203         unsigned long mapsize, section_nr, i;
 204         struct mem_section *ms;
 205         struct page *page, *memmap;
 206         struct mem_section_usage *usage;
 207 
 208         section_nr = pfn_to_section_nr(start_pfn);
 209         ms = __nr_to_section(section_nr);
 210 
 211         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 212 
 213         register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
 214 
 215         usage = ms->usage;
 216         page = virt_to_page(usage);
 217 
 218         mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
 219 
 220         for (i = 0; i < mapsize; i++, page++)
 221                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 222 }
 223 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 224 
 225 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
 226 {
 227         unsigned long i, pfn, end_pfn, nr_pages;
 228         int node = pgdat->node_id;
 229         struct page *page;
 230 
 231         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 232         page = virt_to_page(pgdat);
 233 
 234         for (i = 0; i < nr_pages; i++, page++)
 235                 get_page_bootmem(node, page, NODE_INFO);
 236 
 237         pfn = pgdat->node_start_pfn;
 238         end_pfn = pgdat_end_pfn(pgdat);
 239 
 240         /* register section info */
 241         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 242                 /*
 243                  * Some platforms can assign the same pfn to multiple nodes - on
 244                  * node0 as well as nodeN.  To avoid registering a pfn against
 245                  * multiple nodes we check that this pfn does not already
 246                  * reside in some other nodes.
 247                  */
 248                 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
 249                         register_page_bootmem_info_section(pfn);
 250         }
 251 }
 252 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
 253 
 254 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
 255                 const char *reason)
 256 {
 257         /*
 258          * Disallow all operations smaller than a sub-section and only
 259          * allow operations smaller than a section for
 260          * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
 261          * enforces a larger memory_block_size_bytes() granularity for
 262          * memory that will be marked online, so this check should only
 263          * fire for direct arch_{add,remove}_memory() users outside of
 264          * add_memory_resource().
 265          */
 266         unsigned long min_align;
 267 
 268         if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
 269                 min_align = PAGES_PER_SUBSECTION;
 270         else
 271                 min_align = PAGES_PER_SECTION;
 272         if (!IS_ALIGNED(pfn, min_align)
 273                         || !IS_ALIGNED(nr_pages, min_align)) {
 274                 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
 275                                 reason, pfn, pfn + nr_pages - 1);
 276                 return -EINVAL;
 277         }
 278         return 0;
 279 }
 280 
 281 /*
 282  * Reasonably generic function for adding memory.  It is
 283  * expected that archs that support memory hotplug will
 284  * call this function after deciding the zone to which to
 285  * add the new pages.
 286  */
 287 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
 288                 struct mhp_restrictions *restrictions)
 289 {
 290         int err;
 291         unsigned long nr, start_sec, end_sec;
 292         struct vmem_altmap *altmap = restrictions->altmap;
 293 
 294         if (altmap) {
 295                 /*
 296                  * Validate altmap is within bounds of the total request
 297                  */
 298                 if (altmap->base_pfn != pfn
 299                                 || vmem_altmap_offset(altmap) > nr_pages) {
 300                         pr_warn_once("memory add fail, invalid altmap\n");
 301                         return -EINVAL;
 302                 }
 303                 altmap->alloc = 0;
 304         }
 305 
 306         err = check_pfn_span(pfn, nr_pages, "add");
 307         if (err)
 308                 return err;
 309 
 310         start_sec = pfn_to_section_nr(pfn);
 311         end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
 312         for (nr = start_sec; nr <= end_sec; nr++) {
 313                 unsigned long pfns;
 314 
 315                 pfns = min(nr_pages, PAGES_PER_SECTION
 316                                 - (pfn & ~PAGE_SECTION_MASK));
 317                 err = sparse_add_section(nid, pfn, pfns, altmap);
 318                 if (err)
 319                         break;
 320                 pfn += pfns;
 321                 nr_pages -= pfns;
 322                 cond_resched();
 323         }
 324         vmemmap_populate_print_last();
 325         return err;
 326 }
 327 
 328 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 329 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
 330                                      unsigned long start_pfn,
 331                                      unsigned long end_pfn)
 332 {
 333         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
 334                 if (unlikely(!pfn_to_online_page(start_pfn)))
 335                         continue;
 336 
 337                 if (unlikely(pfn_to_nid(start_pfn) != nid))
 338                         continue;
 339 
 340                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
 341                         continue;
 342 
 343                 return start_pfn;
 344         }
 345 
 346         return 0;
 347 }
 348 
 349 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 350 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
 351                                     unsigned long start_pfn,
 352                                     unsigned long end_pfn)
 353 {
 354         unsigned long pfn;
 355 
 356         /* pfn is the end pfn of a memory section. */
 357         pfn = end_pfn - 1;
 358         for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
 359                 if (unlikely(!pfn_to_online_page(pfn)))
 360                         continue;
 361 
 362                 if (unlikely(pfn_to_nid(pfn) != nid))
 363                         continue;
 364 
 365                 if (zone && zone != page_zone(pfn_to_page(pfn)))
 366                         continue;
 367 
 368                 return pfn;
 369         }
 370 
 371         return 0;
 372 }
 373 
 374 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 375                              unsigned long end_pfn)
 376 {
 377         unsigned long zone_start_pfn = zone->zone_start_pfn;
 378         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
 379         unsigned long zone_end_pfn = z;
 380         unsigned long pfn;
 381         int nid = zone_to_nid(zone);
 382 
 383         zone_span_writelock(zone);
 384         if (zone_start_pfn == start_pfn) {
 385                 /*
 386                  * If the section is smallest section in the zone, it need
 387                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 388                  * In this case, we find second smallest valid mem_section
 389                  * for shrinking zone.
 390                  */
 391                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 392                                                 zone_end_pfn);
 393                 if (pfn) {
 394                         zone->zone_start_pfn = pfn;
 395                         zone->spanned_pages = zone_end_pfn - pfn;
 396                 }
 397         } else if (zone_end_pfn == end_pfn) {
 398                 /*
 399                  * If the section is biggest section in the zone, it need
 400                  * shrink zone->spanned_pages.
 401                  * In this case, we find second biggest valid mem_section for
 402                  * shrinking zone.
 403                  */
 404                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
 405                                                start_pfn);
 406                 if (pfn)
 407                         zone->spanned_pages = pfn - zone_start_pfn + 1;
 408         }
 409 
 410         /*
 411          * The section is not biggest or smallest mem_section in the zone, it
 412          * only creates a hole in the zone. So in this case, we need not
 413          * change the zone. But perhaps, the zone has only hole data. Thus
 414          * it check the zone has only hole or not.
 415          */
 416         pfn = zone_start_pfn;
 417         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) {
 418                 if (unlikely(!pfn_to_online_page(pfn)))
 419                         continue;
 420 
 421                 if (page_zone(pfn_to_page(pfn)) != zone)
 422                         continue;
 423 
 424                 /* Skip range to be removed */
 425                 if (pfn >= start_pfn && pfn < end_pfn)
 426                         continue;
 427 
 428                 /* If we find valid section, we have nothing to do */
 429                 zone_span_writeunlock(zone);
 430                 return;
 431         }
 432 
 433         /* The zone has no valid section */
 434         zone->zone_start_pfn = 0;
 435         zone->spanned_pages = 0;
 436         zone_span_writeunlock(zone);
 437 }
 438 
 439 static void update_pgdat_span(struct pglist_data *pgdat)
 440 {
 441         unsigned long node_start_pfn = 0, node_end_pfn = 0;
 442         struct zone *zone;
 443 
 444         for (zone = pgdat->node_zones;
 445              zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
 446                 unsigned long zone_end_pfn = zone->zone_start_pfn +
 447                                              zone->spanned_pages;
 448 
 449                 /* No need to lock the zones, they can't change. */
 450                 if (!zone->spanned_pages)
 451                         continue;
 452                 if (!node_end_pfn) {
 453                         node_start_pfn = zone->zone_start_pfn;
 454                         node_end_pfn = zone_end_pfn;
 455                         continue;
 456                 }
 457 
 458                 if (zone_end_pfn > node_end_pfn)
 459                         node_end_pfn = zone_end_pfn;
 460                 if (zone->zone_start_pfn < node_start_pfn)
 461                         node_start_pfn = zone->zone_start_pfn;
 462         }
 463 
 464         pgdat->node_start_pfn = node_start_pfn;
 465         pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
 466 }
 467 
 468 void __ref remove_pfn_range_from_zone(struct zone *zone,
 469                                       unsigned long start_pfn,
 470                                       unsigned long nr_pages)
 471 {
 472         struct pglist_data *pgdat = zone->zone_pgdat;
 473         unsigned long flags;
 474 
 475 #ifdef CONFIG_ZONE_DEVICE
 476         /*
 477          * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
 478          * we will not try to shrink the zones - which is okay as
 479          * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
 480          */
 481         if (zone_idx(zone) == ZONE_DEVICE)
 482                 return;
 483 #endif
 484 
 485         clear_zone_contiguous(zone);
 486 
 487         pgdat_resize_lock(zone->zone_pgdat, &flags);
 488         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 489         update_pgdat_span(pgdat);
 490         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 491 
 492         set_zone_contiguous(zone);
 493 }
 494 
 495 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
 496                              unsigned long map_offset,
 497                              struct vmem_altmap *altmap)
 498 {
 499         struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn));
 500 
 501         if (WARN_ON_ONCE(!valid_section(ms)))
 502                 return;
 503 
 504         sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
 505 }
 506 
 507 /**
 508  * __remove_pages() - remove sections of pages
 509  * @pfn: starting pageframe (must be aligned to start of a section)
 510  * @nr_pages: number of pages to remove (must be multiple of section size)
 511  * @altmap: alternative device page map or %NULL if default memmap is used
 512  *
 513  * Generic helper function to remove section mappings and sysfs entries
 514  * for the section of the memory we are removing. Caller needs to make
 515  * sure that pages are marked reserved and zones are adjust properly by
 516  * calling offline_pages().
 517  */
 518 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
 519                     struct vmem_altmap *altmap)
 520 {
 521         unsigned long map_offset = 0;
 522         unsigned long nr, start_sec, end_sec;
 523 
 524         map_offset = vmem_altmap_offset(altmap);
 525 
 526         if (check_pfn_span(pfn, nr_pages, "remove"))
 527                 return;
 528 
 529         start_sec = pfn_to_section_nr(pfn);
 530         end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
 531         for (nr = start_sec; nr <= end_sec; nr++) {
 532                 unsigned long pfns;
 533 
 534                 cond_resched();
 535                 pfns = min(nr_pages, PAGES_PER_SECTION
 536                                 - (pfn & ~PAGE_SECTION_MASK));
 537                 __remove_section(pfn, pfns, map_offset, altmap);
 538                 pfn += pfns;
 539                 nr_pages -= pfns;
 540                 map_offset = 0;
 541         }
 542 }
 543 
 544 int set_online_page_callback(online_page_callback_t callback)
 545 {
 546         int rc = -EINVAL;
 547 
 548         get_online_mems();
 549         mutex_lock(&online_page_callback_lock);
 550 
 551         if (online_page_callback == generic_online_page) {
 552                 online_page_callback = callback;
 553                 rc = 0;
 554         }
 555 
 556         mutex_unlock(&online_page_callback_lock);
 557         put_online_mems();
 558 
 559         return rc;
 560 }
 561 EXPORT_SYMBOL_GPL(set_online_page_callback);
 562 
 563 int restore_online_page_callback(online_page_callback_t callback)
 564 {
 565         int rc = -EINVAL;
 566 
 567         get_online_mems();
 568         mutex_lock(&online_page_callback_lock);
 569 
 570         if (online_page_callback == callback) {
 571                 online_page_callback = generic_online_page;
 572                 rc = 0;
 573         }
 574 
 575         mutex_unlock(&online_page_callback_lock);
 576         put_online_mems();
 577 
 578         return rc;
 579 }
 580 EXPORT_SYMBOL_GPL(restore_online_page_callback);
 581 
 582 void __online_page_set_limits(struct page *page)
 583 {
 584 }
 585 EXPORT_SYMBOL_GPL(__online_page_set_limits);
 586 
 587 void __online_page_increment_counters(struct page *page)
 588 {
 589         adjust_managed_page_count(page, 1);
 590 }
 591 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 592 
 593 void __online_page_free(struct page *page)
 594 {
 595         __free_reserved_page(page);
 596 }
 597 EXPORT_SYMBOL_GPL(__online_page_free);
 598 
 599 static void generic_online_page(struct page *page, unsigned int order)
 600 {
 601         /*
 602          * Freeing the page with debug_pagealloc enabled will try to unmap it,
 603          * so we should map it first. This is better than introducing a special
 604          * case in page freeing fast path.
 605          */
 606         if (debug_pagealloc_enabled_static())
 607                 kernel_map_pages(page, 1 << order, 1);
 608         __free_pages_core(page, order);
 609         totalram_pages_add(1UL << order);
 610 #ifdef CONFIG_HIGHMEM
 611         if (PageHighMem(page))
 612                 totalhigh_pages_add(1UL << order);
 613 #endif
 614 }
 615 
 616 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 617                         void *arg)
 618 {
 619         const unsigned long end_pfn = start_pfn + nr_pages;
 620         unsigned long pfn;
 621         int order;
 622 
 623         /*
 624          * Online the pages. The callback might decide to keep some pages
 625          * PG_reserved (to add them to the buddy later), but we still account
 626          * them as being online/belonging to this zone ("present").
 627          */
 628         for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
 629                 order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
 630                 /* __free_pages_core() wants pfns to be aligned to the order */
 631                 if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order)))
 632                         order = 0;
 633                 (*online_page_callback)(pfn_to_page(pfn), order);
 634         }
 635 
 636         /* mark all involved sections as online */
 637         online_mem_sections(start_pfn, end_pfn);
 638 
 639         *(unsigned long *)arg += nr_pages;
 640         return 0;
 641 }
 642 
 643 /* check which state of node_states will be changed when online memory */
 644 static void node_states_check_changes_online(unsigned long nr_pages,
 645         struct zone *zone, struct memory_notify *arg)
 646 {
 647         int nid = zone_to_nid(zone);
 648 
 649         arg->status_change_nid = NUMA_NO_NODE;
 650         arg->status_change_nid_normal = NUMA_NO_NODE;
 651         arg->status_change_nid_high = NUMA_NO_NODE;
 652 
 653         if (!node_state(nid, N_MEMORY))
 654                 arg->status_change_nid = nid;
 655         if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
 656                 arg->status_change_nid_normal = nid;
 657 #ifdef CONFIG_HIGHMEM
 658         if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
 659                 arg->status_change_nid_high = nid;
 660 #endif
 661 }
 662 
 663 static void node_states_set_node(int node, struct memory_notify *arg)
 664 {
 665         if (arg->status_change_nid_normal >= 0)
 666                 node_set_state(node, N_NORMAL_MEMORY);
 667 
 668         if (arg->status_change_nid_high >= 0)
 669                 node_set_state(node, N_HIGH_MEMORY);
 670 
 671         if (arg->status_change_nid >= 0)
 672                 node_set_state(node, N_MEMORY);
 673 }
 674 
 675 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
 676                 unsigned long nr_pages)
 677 {
 678         unsigned long old_end_pfn = zone_end_pfn(zone);
 679 
 680         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
 681                 zone->zone_start_pfn = start_pfn;
 682 
 683         zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
 684 }
 685 
 686 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
 687                                      unsigned long nr_pages)
 688 {
 689         unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
 690 
 691         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
 692                 pgdat->node_start_pfn = start_pfn;
 693 
 694         pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
 695 
 696 }
 697 /*
 698  * Associate the pfn range with the given zone, initializing the memmaps
 699  * and resizing the pgdat/zone data to span the added pages. After this
 700  * call, all affected pages are PG_reserved.
 701  */
 702 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
 703                 unsigned long nr_pages, struct vmem_altmap *altmap)
 704 {
 705         struct pglist_data *pgdat = zone->zone_pgdat;
 706         int nid = pgdat->node_id;
 707         unsigned long flags;
 708 
 709         clear_zone_contiguous(zone);
 710 
 711         /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
 712         pgdat_resize_lock(pgdat, &flags);
 713         zone_span_writelock(zone);
 714         if (zone_is_empty(zone))
 715                 init_currently_empty_zone(zone, start_pfn, nr_pages);
 716         resize_zone_range(zone, start_pfn, nr_pages);
 717         zone_span_writeunlock(zone);
 718         resize_pgdat_range(pgdat, start_pfn, nr_pages);
 719         pgdat_resize_unlock(pgdat, &flags);
 720 
 721         /*
 722          * TODO now we have a visible range of pages which are not associated
 723          * with their zone properly. Not nice but set_pfnblock_flags_mask
 724          * expects the zone spans the pfn range. All the pages in the range
 725          * are reserved so nobody should be touching them so we should be safe
 726          */
 727         memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
 728                         MEMMAP_HOTPLUG, altmap);
 729 
 730         set_zone_contiguous(zone);
 731 }
 732 
 733 /*
 734  * Returns a default kernel memory zone for the given pfn range.
 735  * If no kernel zone covers this pfn range it will automatically go
 736  * to the ZONE_NORMAL.
 737  */
 738 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
 739                 unsigned long nr_pages)
 740 {
 741         struct pglist_data *pgdat = NODE_DATA(nid);
 742         int zid;
 743 
 744         for (zid = 0; zid <= ZONE_NORMAL; zid++) {
 745                 struct zone *zone = &pgdat->node_zones[zid];
 746 
 747                 if (zone_intersects(zone, start_pfn, nr_pages))
 748                         return zone;
 749         }
 750 
 751         return &pgdat->node_zones[ZONE_NORMAL];
 752 }
 753 
 754 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
 755                 unsigned long nr_pages)
 756 {
 757         struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
 758                         nr_pages);
 759         struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
 760         bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
 761         bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
 762 
 763         /*
 764          * We inherit the existing zone in a simple case where zones do not
 765          * overlap in the given range
 766          */
 767         if (in_kernel ^ in_movable)
 768                 return (in_kernel) ? kernel_zone : movable_zone;
 769 
 770         /*
 771          * If the range doesn't belong to any zone or two zones overlap in the
 772          * given range then we use movable zone only if movable_node is
 773          * enabled because we always online to a kernel zone by default.
 774          */
 775         return movable_node_enabled ? movable_zone : kernel_zone;
 776 }
 777 
 778 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
 779                 unsigned long nr_pages)
 780 {
 781         if (online_type == MMOP_ONLINE_KERNEL)
 782                 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
 783 
 784         if (online_type == MMOP_ONLINE_MOVABLE)
 785                 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
 786 
 787         return default_zone_for_pfn(nid, start_pfn, nr_pages);
 788 }
 789 
 790 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
 791 {
 792         unsigned long flags;
 793         unsigned long onlined_pages = 0;
 794         struct zone *zone;
 795         int need_zonelists_rebuild = 0;
 796         int nid;
 797         int ret;
 798         struct memory_notify arg;
 799         struct memory_block *mem;
 800 
 801         mem_hotplug_begin();
 802 
 803         /*
 804          * We can't use pfn_to_nid() because nid might be stored in struct page
 805          * which is not yet initialized. Instead, we find nid from memory block.
 806          */
 807         mem = find_memory_block(__pfn_to_section(pfn));
 808         nid = mem->nid;
 809         put_device(&mem->dev);
 810 
 811         /* associate pfn range with the zone */
 812         zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
 813         move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
 814 
 815         arg.start_pfn = pfn;
 816         arg.nr_pages = nr_pages;
 817         node_states_check_changes_online(nr_pages, zone, &arg);
 818 
 819         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 820         ret = notifier_to_errno(ret);
 821         if (ret)
 822                 goto failed_addition;
 823 
 824         /*
 825          * If this zone is not populated, then it is not in zonelist.
 826          * This means the page allocator ignores this zone.
 827          * So, zonelist must be updated after online.
 828          */
 829         if (!populated_zone(zone)) {
 830                 need_zonelists_rebuild = 1;
 831                 setup_zone_pageset(zone);
 832         }
 833 
 834         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
 835                 online_pages_range);
 836         if (ret) {
 837                 /* not a single memory resource was applicable */
 838                 if (need_zonelists_rebuild)
 839                         zone_pcp_reset(zone);
 840                 goto failed_addition;
 841         }
 842 
 843         zone->present_pages += onlined_pages;
 844 
 845         pgdat_resize_lock(zone->zone_pgdat, &flags);
 846         zone->zone_pgdat->node_present_pages += onlined_pages;
 847         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 848 
 849         shuffle_zone(zone);
 850 
 851         node_states_set_node(nid, &arg);
 852         if (need_zonelists_rebuild)
 853                 build_all_zonelists(NULL);
 854         else
 855                 zone_pcp_update(zone);
 856 
 857         init_per_zone_wmark_min();
 858 
 859         kswapd_run(nid);
 860         kcompactd_run(nid);
 861 
 862         vm_total_pages = nr_free_pagecache_pages();
 863 
 864         writeback_set_ratelimit();
 865 
 866         memory_notify(MEM_ONLINE, &arg);
 867         mem_hotplug_done();
 868         return 0;
 869 
 870 failed_addition:
 871         pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
 872                  (unsigned long long) pfn << PAGE_SHIFT,
 873                  (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
 874         memory_notify(MEM_CANCEL_ONLINE, &arg);
 875         remove_pfn_range_from_zone(zone, pfn, nr_pages);
 876         mem_hotplug_done();
 877         return ret;
 878 }
 879 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 880 
 881 static void reset_node_present_pages(pg_data_t *pgdat)
 882 {
 883         struct zone *z;
 884 
 885         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 886                 z->present_pages = 0;
 887 
 888         pgdat->node_present_pages = 0;
 889 }
 890 
 891 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 892 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 893 {
 894         struct pglist_data *pgdat;
 895         unsigned long start_pfn = PFN_DOWN(start);
 896 
 897         pgdat = NODE_DATA(nid);
 898         if (!pgdat) {
 899                 pgdat = arch_alloc_nodedata(nid);
 900                 if (!pgdat)
 901                         return NULL;
 902 
 903                 pgdat->per_cpu_nodestats =
 904                         alloc_percpu(struct per_cpu_nodestat);
 905                 arch_refresh_nodedata(nid, pgdat);
 906         } else {
 907                 int cpu;
 908                 /*
 909                  * Reset the nr_zones, order and classzone_idx before reuse.
 910                  * Note that kswapd will init kswapd_classzone_idx properly
 911                  * when it starts in the near future.
 912                  */
 913                 pgdat->nr_zones = 0;
 914                 pgdat->kswapd_order = 0;
 915                 pgdat->kswapd_classzone_idx = 0;
 916                 for_each_online_cpu(cpu) {
 917                         struct per_cpu_nodestat *p;
 918 
 919                         p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
 920                         memset(p, 0, sizeof(*p));
 921                 }
 922         }
 923 
 924         /* we can use NODE_DATA(nid) from here */
 925 
 926         pgdat->node_id = nid;
 927         pgdat->node_start_pfn = start_pfn;
 928 
 929         /* init node's zones as empty zones, we don't have any present pages.*/
 930         free_area_init_core_hotplug(nid);
 931 
 932         /*
 933          * The node we allocated has no zone fallback lists. For avoiding
 934          * to access not-initialized zonelist, build here.
 935          */
 936         build_all_zonelists(pgdat);
 937 
 938         /*
 939          * When memory is hot-added, all the memory is in offline state. So
 940          * clear all zones' present_pages because they will be updated in
 941          * online_pages() and offline_pages().
 942          */
 943         reset_node_managed_pages(pgdat);
 944         reset_node_present_pages(pgdat);
 945 
 946         return pgdat;
 947 }
 948 
 949 static void rollback_node_hotadd(int nid)
 950 {
 951         pg_data_t *pgdat = NODE_DATA(nid);
 952 
 953         arch_refresh_nodedata(nid, NULL);
 954         free_percpu(pgdat->per_cpu_nodestats);
 955         arch_free_nodedata(pgdat);
 956 }
 957 
 958 
 959 /**
 960  * try_online_node - online a node if offlined
 961  * @nid: the node ID
 962  * @start: start addr of the node
 963  * @set_node_online: Whether we want to online the node
 964  * called by cpu_up() to online a node without onlined memory.
 965  *
 966  * Returns:
 967  * 1 -> a new node has been allocated
 968  * 0 -> the node is already online
 969  * -ENOMEM -> the node could not be allocated
 970  */
 971 static int __try_online_node(int nid, u64 start, bool set_node_online)
 972 {
 973         pg_data_t *pgdat;
 974         int ret = 1;
 975 
 976         if (node_online(nid))
 977                 return 0;
 978 
 979         pgdat = hotadd_new_pgdat(nid, start);
 980         if (!pgdat) {
 981                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
 982                 ret = -ENOMEM;
 983                 goto out;
 984         }
 985 
 986         if (set_node_online) {
 987                 node_set_online(nid);
 988                 ret = register_one_node(nid);
 989                 BUG_ON(ret);
 990         }
 991 out:
 992         return ret;
 993 }
 994 
 995 /*
 996  * Users of this function always want to online/register the node
 997  */
 998 int try_online_node(int nid)
 999 {
1000         int ret;
1001 
1002         mem_hotplug_begin();
1003         ret =  __try_online_node(nid, 0, true);
1004         mem_hotplug_done();
1005         return ret;
1006 }
1007 
1008 static int check_hotplug_memory_range(u64 start, u64 size)
1009 {
1010         /* memory range must be block size aligned */
1011         if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1012             !IS_ALIGNED(size, memory_block_size_bytes())) {
1013                 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1014                        memory_block_size_bytes(), start, size);
1015                 return -EINVAL;
1016         }
1017 
1018         return 0;
1019 }
1020 
1021 static int online_memory_block(struct memory_block *mem, void *arg)
1022 {
1023         return device_online(&mem->dev);
1024 }
1025 
1026 /*
1027  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1028  * and online/offline operations (triggered e.g. by sysfs).
1029  *
1030  * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1031  */
1032 int __ref add_memory_resource(int nid, struct resource *res)
1033 {
1034         struct mhp_restrictions restrictions = {};
1035         u64 start, size;
1036         bool new_node = false;
1037         int ret;
1038 
1039         start = res->start;
1040         size = resource_size(res);
1041 
1042         ret = check_hotplug_memory_range(start, size);
1043         if (ret)
1044                 return ret;
1045 
1046         mem_hotplug_begin();
1047 
1048         /*
1049          * Add new range to memblock so that when hotadd_new_pgdat() is called
1050          * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1051          * this new range and calculate total pages correctly.  The range will
1052          * be removed at hot-remove time.
1053          */
1054         memblock_add_node(start, size, nid);
1055 
1056         ret = __try_online_node(nid, start, false);
1057         if (ret < 0)
1058                 goto error;
1059         new_node = ret;
1060 
1061         /* call arch's memory hotadd */
1062         ret = arch_add_memory(nid, start, size, &restrictions);
1063         if (ret < 0)
1064                 goto error;
1065 
1066         /* create memory block devices after memory was added */
1067         ret = create_memory_block_devices(start, size);
1068         if (ret) {
1069                 arch_remove_memory(nid, start, size, NULL);
1070                 goto error;
1071         }
1072 
1073         if (new_node) {
1074                 /* If sysfs file of new node can't be created, cpu on the node
1075                  * can't be hot-added. There is no rollback way now.
1076                  * So, check by BUG_ON() to catch it reluctantly..
1077                  * We online node here. We can't roll back from here.
1078                  */
1079                 node_set_online(nid);
1080                 ret = __register_one_node(nid);
1081                 BUG_ON(ret);
1082         }
1083 
1084         /* link memory sections under this node.*/
1085         ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1086         BUG_ON(ret);
1087 
1088         /* create new memmap entry */
1089         firmware_map_add_hotplug(start, start + size, "System RAM");
1090 
1091         /* device_online() will take the lock when calling online_pages() */
1092         mem_hotplug_done();
1093 
1094         /* online pages if requested */
1095         if (memhp_auto_online)
1096                 walk_memory_blocks(start, size, NULL, online_memory_block);
1097 
1098         return ret;
1099 error:
1100         /* rollback pgdat allocation and others */
1101         if (new_node)
1102                 rollback_node_hotadd(nid);
1103         memblock_remove(start, size);
1104         mem_hotplug_done();
1105         return ret;
1106 }
1107 
1108 /* requires device_hotplug_lock, see add_memory_resource() */
1109 int __ref __add_memory(int nid, u64 start, u64 size)
1110 {
1111         struct resource *res;
1112         int ret;
1113 
1114         res = register_memory_resource(start, size);
1115         if (IS_ERR(res))
1116                 return PTR_ERR(res);
1117 
1118         ret = add_memory_resource(nid, res);
1119         if (ret < 0)
1120                 release_memory_resource(res);
1121         return ret;
1122 }
1123 
1124 int add_memory(int nid, u64 start, u64 size)
1125 {
1126         int rc;
1127 
1128         lock_device_hotplug();
1129         rc = __add_memory(nid, start, size);
1130         unlock_device_hotplug();
1131 
1132         return rc;
1133 }
1134 EXPORT_SYMBOL_GPL(add_memory);
1135 
1136 #ifdef CONFIG_MEMORY_HOTREMOVE
1137 /*
1138  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1139  * set and the size of the free page is given by page_order(). Using this,
1140  * the function determines if the pageblock contains only free pages.
1141  * Due to buddy contraints, a free page at least the size of a pageblock will
1142  * be located at the start of the pageblock
1143  */
1144 static inline int pageblock_free(struct page *page)
1145 {
1146         return PageBuddy(page) && page_order(page) >= pageblock_order;
1147 }
1148 
1149 /* Return the pfn of the start of the next active pageblock after a given pfn */
1150 static unsigned long next_active_pageblock(unsigned long pfn)
1151 {
1152         struct page *page = pfn_to_page(pfn);
1153 
1154         /* Ensure the starting page is pageblock-aligned */
1155         BUG_ON(pfn & (pageblock_nr_pages - 1));
1156 
1157         /* If the entire pageblock is free, move to the end of free page */
1158         if (pageblock_free(page)) {
1159                 int order;
1160                 /* be careful. we don't have locks, page_order can be changed.*/
1161                 order = page_order(page);
1162                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1163                         return pfn + (1 << order);
1164         }
1165 
1166         return pfn + pageblock_nr_pages;
1167 }
1168 
1169 static bool is_pageblock_removable_nolock(unsigned long pfn)
1170 {
1171         struct page *page = pfn_to_page(pfn);
1172         struct zone *zone;
1173 
1174         /*
1175          * We have to be careful here because we are iterating over memory
1176          * sections which are not zone aware so we might end up outside of
1177          * the zone but still within the section.
1178          * We have to take care about the node as well. If the node is offline
1179          * its NODE_DATA will be NULL - see page_zone.
1180          */
1181         if (!node_online(page_to_nid(page)))
1182                 return false;
1183 
1184         zone = page_zone(page);
1185         pfn = page_to_pfn(page);
1186         if (!zone_spans_pfn(zone, pfn))
1187                 return false;
1188 
1189         return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1190 }
1191 
1192 /* Checks if this range of memory is likely to be hot-removable. */
1193 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1194 {
1195         unsigned long end_pfn, pfn;
1196 
1197         end_pfn = min(start_pfn + nr_pages,
1198                         zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1199 
1200         /* Check the starting page of each pageblock within the range */
1201         for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1202                 if (!is_pageblock_removable_nolock(pfn))
1203                         return false;
1204                 cond_resched();
1205         }
1206 
1207         /* All pageblocks in the memory block are likely to be hot-removable */
1208         return true;
1209 }
1210 
1211 /*
1212  * Confirm all pages in a range [start, end) belong to the same zone.
1213  * When true, return its valid [start, end).
1214  */
1215 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1216                          unsigned long *valid_start, unsigned long *valid_end)
1217 {
1218         unsigned long pfn, sec_end_pfn;
1219         unsigned long start, end;
1220         struct zone *zone = NULL;
1221         struct page *page;
1222         int i;
1223         for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1224              pfn < end_pfn;
1225              pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1226                 /* Make sure the memory section is present first */
1227                 if (!present_section_nr(pfn_to_section_nr(pfn)))
1228                         continue;
1229                 for (; pfn < sec_end_pfn && pfn < end_pfn;
1230                      pfn += MAX_ORDER_NR_PAGES) {
1231                         i = 0;
1232                         /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1233                         while ((i < MAX_ORDER_NR_PAGES) &&
1234                                 !pfn_valid_within(pfn + i))
1235                                 i++;
1236                         if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1237                                 continue;
1238                         /* Check if we got outside of the zone */
1239                         if (zone && !zone_spans_pfn(zone, pfn + i))
1240                                 return 0;
1241                         page = pfn_to_page(pfn + i);
1242                         if (zone && page_zone(page) != zone)
1243                                 return 0;
1244                         if (!zone)
1245                                 start = pfn + i;
1246                         zone = page_zone(page);
1247                         end = pfn + MAX_ORDER_NR_PAGES;
1248                 }
1249         }
1250 
1251         if (zone) {
1252                 *valid_start = start;
1253                 *valid_end = min(end, end_pfn);
1254                 return 1;
1255         } else {
1256                 return 0;
1257         }
1258 }
1259 
1260 /*
1261  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1262  * non-lru movable pages and hugepages). We scan pfn because it's much
1263  * easier than scanning over linked list. This function returns the pfn
1264  * of the first found movable page if it's found, otherwise 0.
1265  */
1266 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1267 {
1268         unsigned long pfn;
1269 
1270         for (pfn = start; pfn < end; pfn++) {
1271                 struct page *page, *head;
1272                 unsigned long skip;
1273 
1274                 if (!pfn_valid(pfn))
1275                         continue;
1276                 page = pfn_to_page(pfn);
1277                 if (PageLRU(page))
1278                         return pfn;
1279                 if (__PageMovable(page))
1280                         return pfn;
1281 
1282                 if (!PageHuge(page))
1283                         continue;
1284                 head = compound_head(page);
1285                 if (page_huge_active(head))
1286                         return pfn;
1287                 skip = compound_nr(head) - (page - head);
1288                 pfn += skip - 1;
1289         }
1290         return 0;
1291 }
1292 
1293 static struct page *new_node_page(struct page *page, unsigned long private)
1294 {
1295         int nid = page_to_nid(page);
1296         nodemask_t nmask = node_states[N_MEMORY];
1297 
1298         /*
1299          * try to allocate from a different node but reuse this node if there
1300          * are no other online nodes to be used (e.g. we are offlining a part
1301          * of the only existing node)
1302          */
1303         node_clear(nid, nmask);
1304         if (nodes_empty(nmask))
1305                 node_set(nid, nmask);
1306 
1307         return new_page_nodemask(page, nid, &nmask);
1308 }
1309 
1310 static int
1311 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1312 {
1313         unsigned long pfn;
1314         struct page *page;
1315         int ret = 0;
1316         LIST_HEAD(source);
1317 
1318         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1319                 if (!pfn_valid(pfn))
1320                         continue;
1321                 page = pfn_to_page(pfn);
1322 
1323                 if (PageHuge(page)) {
1324                         struct page *head = compound_head(page);
1325                         pfn = page_to_pfn(head) + compound_nr(head) - 1;
1326                         isolate_huge_page(head, &source);
1327                         continue;
1328                 } else if (PageTransHuge(page))
1329                         pfn = page_to_pfn(compound_head(page))
1330                                 + hpage_nr_pages(page) - 1;
1331 
1332                 /*
1333                  * HWPoison pages have elevated reference counts so the migration would
1334                  * fail on them. It also doesn't make any sense to migrate them in the
1335                  * first place. Still try to unmap such a page in case it is still mapped
1336                  * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1337                  * the unmap as the catch all safety net).
1338                  */
1339                 if (PageHWPoison(page)) {
1340                         if (WARN_ON(PageLRU(page)))
1341                                 isolate_lru_page(page);
1342                         if (page_mapped(page))
1343                                 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1344                         continue;
1345                 }
1346 
1347                 if (!get_page_unless_zero(page))
1348                         continue;
1349                 /*
1350                  * We can skip free pages. And we can deal with pages on
1351                  * LRU and non-lru movable pages.
1352                  */
1353                 if (PageLRU(page))
1354                         ret = isolate_lru_page(page);
1355                 else
1356                         ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1357                 if (!ret) { /* Success */
1358                         list_add_tail(&page->lru, &source);
1359                         if (!__PageMovable(page))
1360                                 inc_node_page_state(page, NR_ISOLATED_ANON +
1361                                                     page_is_file_cache(page));
1362 
1363                 } else {
1364                         pr_warn("failed to isolate pfn %lx\n", pfn);
1365                         dump_page(page, "isolation failed");
1366                 }
1367                 put_page(page);
1368         }
1369         if (!list_empty(&source)) {
1370                 /* Allocate a new page from the nearest neighbor node */
1371                 ret = migrate_pages(&source, new_node_page, NULL, 0,
1372                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1373                 if (ret) {
1374                         list_for_each_entry(page, &source, lru) {
1375                                 pr_warn("migrating pfn %lx failed ret:%d ",
1376                                        page_to_pfn(page), ret);
1377                                 dump_page(page, "migration failure");
1378                         }
1379                         putback_movable_pages(&source);
1380                 }
1381         }
1382 
1383         return ret;
1384 }
1385 
1386 /*
1387  * remove from free_area[] and mark all as Reserved.
1388  */
1389 static int
1390 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1391                         void *data)
1392 {
1393         unsigned long *offlined_pages = (unsigned long *)data;
1394 
1395         *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1396         return 0;
1397 }
1398 
1399 /*
1400  * Check all pages in range, recoreded as memory resource, are isolated.
1401  */
1402 static int
1403 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1404                         void *data)
1405 {
1406         return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1407 }
1408 
1409 static int __init cmdline_parse_movable_node(char *p)
1410 {
1411 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1412         movable_node_enabled = true;
1413 #else
1414         pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1415 #endif
1416         return 0;
1417 }
1418 early_param("movable_node", cmdline_parse_movable_node);
1419 
1420 /* check which state of node_states will be changed when offline memory */
1421 static void node_states_check_changes_offline(unsigned long nr_pages,
1422                 struct zone *zone, struct memory_notify *arg)
1423 {
1424         struct pglist_data *pgdat = zone->zone_pgdat;
1425         unsigned long present_pages = 0;
1426         enum zone_type zt;
1427 
1428         arg->status_change_nid = NUMA_NO_NODE;
1429         arg->status_change_nid_normal = NUMA_NO_NODE;
1430         arg->status_change_nid_high = NUMA_NO_NODE;
1431 
1432         /*
1433          * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1434          * If the memory to be offline is within the range
1435          * [0..ZONE_NORMAL], and it is the last present memory there,
1436          * the zones in that range will become empty after the offlining,
1437          * thus we can determine that we need to clear the node from
1438          * node_states[N_NORMAL_MEMORY].
1439          */
1440         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1441                 present_pages += pgdat->node_zones[zt].present_pages;
1442         if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1443                 arg->status_change_nid_normal = zone_to_nid(zone);
1444 
1445 #ifdef CONFIG_HIGHMEM
1446         /*
1447          * node_states[N_HIGH_MEMORY] contains nodes which
1448          * have normal memory or high memory.
1449          * Here we add the present_pages belonging to ZONE_HIGHMEM.
1450          * If the zone is within the range of [0..ZONE_HIGHMEM), and
1451          * we determine that the zones in that range become empty,
1452          * we need to clear the node for N_HIGH_MEMORY.
1453          */
1454         present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1455         if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1456                 arg->status_change_nid_high = zone_to_nid(zone);
1457 #endif
1458 
1459         /*
1460          * We have accounted the pages from [0..ZONE_NORMAL), and
1461          * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1462          * as well.
1463          * Here we count the possible pages from ZONE_MOVABLE.
1464          * If after having accounted all the pages, we see that the nr_pages
1465          * to be offlined is over or equal to the accounted pages,
1466          * we know that the node will become empty, and so, we can clear
1467          * it for N_MEMORY as well.
1468          */
1469         present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1470 
1471         if (nr_pages >= present_pages)
1472                 arg->status_change_nid = zone_to_nid(zone);
1473 }
1474 
1475 static void node_states_clear_node(int node, struct memory_notify *arg)
1476 {
1477         if (arg->status_change_nid_normal >= 0)
1478                 node_clear_state(node, N_NORMAL_MEMORY);
1479 
1480         if (arg->status_change_nid_high >= 0)
1481                 node_clear_state(node, N_HIGH_MEMORY);
1482 
1483         if (arg->status_change_nid >= 0)
1484                 node_clear_state(node, N_MEMORY);
1485 }
1486 
1487 static int __ref __offline_pages(unsigned long start_pfn,
1488                   unsigned long end_pfn)
1489 {
1490         unsigned long pfn, nr_pages;
1491         unsigned long offlined_pages = 0;
1492         int ret, node, nr_isolate_pageblock;
1493         unsigned long flags;
1494         unsigned long valid_start, valid_end;
1495         struct zone *zone;
1496         struct memory_notify arg;
1497         char *reason;
1498 
1499         mem_hotplug_begin();
1500 
1501         /* This makes hotplug much easier...and readable.
1502            we assume this for now. .*/
1503         if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1504                                   &valid_end)) {
1505                 ret = -EINVAL;
1506                 reason = "multizone range";
1507                 goto failed_removal;
1508         }
1509 
1510         zone = page_zone(pfn_to_page(valid_start));
1511         node = zone_to_nid(zone);
1512         nr_pages = end_pfn - start_pfn;
1513 
1514         /* set above range as isolated */
1515         ret = start_isolate_page_range(start_pfn, end_pfn,
1516                                        MIGRATE_MOVABLE,
1517                                        SKIP_HWPOISON | REPORT_FAILURE);
1518         if (ret < 0) {
1519                 reason = "failure to isolate range";
1520                 goto failed_removal;
1521         }
1522         nr_isolate_pageblock = ret;
1523 
1524         arg.start_pfn = start_pfn;
1525         arg.nr_pages = nr_pages;
1526         node_states_check_changes_offline(nr_pages, zone, &arg);
1527 
1528         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1529         ret = notifier_to_errno(ret);
1530         if (ret) {
1531                 reason = "notifier failure";
1532                 goto failed_removal_isolated;
1533         }
1534 
1535         do {
1536                 for (pfn = start_pfn; pfn;) {
1537                         if (signal_pending(current)) {
1538                                 ret = -EINTR;
1539                                 reason = "signal backoff";
1540                                 goto failed_removal_isolated;
1541                         }
1542 
1543                         cond_resched();
1544                         lru_add_drain_all();
1545 
1546                         pfn = scan_movable_pages(pfn, end_pfn);
1547                         if (pfn) {
1548                                 /*
1549                                  * TODO: fatal migration failures should bail
1550                                  * out
1551                                  */
1552                                 do_migrate_range(pfn, end_pfn);
1553                         }
1554                 }
1555 
1556                 /*
1557                  * Dissolve free hugepages in the memory block before doing
1558                  * offlining actually in order to make hugetlbfs's object
1559                  * counting consistent.
1560                  */
1561                 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1562                 if (ret) {
1563                         reason = "failure to dissolve huge pages";
1564                         goto failed_removal_isolated;
1565                 }
1566                 /* check again */
1567                 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1568                                             NULL, check_pages_isolated_cb);
1569         } while (ret);
1570 
1571         /* Ok, all of our target is isolated.
1572            We cannot do rollback at this point. */
1573         walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1574                               &offlined_pages, offline_isolated_pages_cb);
1575         pr_info("Offlined Pages %ld\n", offlined_pages);
1576         /*
1577          * Onlining will reset pagetype flags and makes migrate type
1578          * MOVABLE, so just need to decrease the number of isolated
1579          * pageblocks zone counter here.
1580          */
1581         spin_lock_irqsave(&zone->lock, flags);
1582         zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1583         spin_unlock_irqrestore(&zone->lock, flags);
1584 
1585         /* removal success */
1586         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1587         zone->present_pages -= offlined_pages;
1588 
1589         pgdat_resize_lock(zone->zone_pgdat, &flags);
1590         zone->zone_pgdat->node_present_pages -= offlined_pages;
1591         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1592 
1593         init_per_zone_wmark_min();
1594 
1595         if (!populated_zone(zone)) {
1596                 zone_pcp_reset(zone);
1597                 build_all_zonelists(NULL);
1598         } else
1599                 zone_pcp_update(zone);
1600 
1601         node_states_clear_node(node, &arg);
1602         if (arg.status_change_nid >= 0) {
1603                 kswapd_stop(node);
1604                 kcompactd_stop(node);
1605         }
1606 
1607         vm_total_pages = nr_free_pagecache_pages();
1608         writeback_set_ratelimit();
1609 
1610         memory_notify(MEM_OFFLINE, &arg);
1611         remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1612         mem_hotplug_done();
1613         return 0;
1614 
1615 failed_removal_isolated:
1616         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1617         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1618 failed_removal:
1619         pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1620                  (unsigned long long) start_pfn << PAGE_SHIFT,
1621                  ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1622                  reason);
1623         /* pushback to free area */
1624         mem_hotplug_done();
1625         return ret;
1626 }
1627 
1628 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1629 {
1630         return __offline_pages(start_pfn, start_pfn + nr_pages);
1631 }
1632 
1633 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1634 {
1635         int ret = !is_memblock_offlined(mem);
1636 
1637         if (unlikely(ret)) {
1638                 phys_addr_t beginpa, endpa;
1639 
1640                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1641                 endpa = beginpa + memory_block_size_bytes() - 1;
1642                 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1643                         &beginpa, &endpa);
1644 
1645                 return -EBUSY;
1646         }
1647         return 0;
1648 }
1649 
1650 static int check_cpu_on_node(pg_data_t *pgdat)
1651 {
1652         int cpu;
1653 
1654         for_each_present_cpu(cpu) {
1655                 if (cpu_to_node(cpu) == pgdat->node_id)
1656                         /*
1657                          * the cpu on this node isn't removed, and we can't
1658                          * offline this node.
1659                          */
1660                         return -EBUSY;
1661         }
1662 
1663         return 0;
1664 }
1665 
1666 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1667 {
1668         int nid = *(int *)arg;
1669 
1670         /*
1671          * If a memory block belongs to multiple nodes, the stored nid is not
1672          * reliable. However, such blocks are always online (e.g., cannot get
1673          * offlined) and, therefore, are still spanned by the node.
1674          */
1675         return mem->nid == nid ? -EEXIST : 0;
1676 }
1677 
1678 /**
1679  * try_offline_node
1680  * @nid: the node ID
1681  *
1682  * Offline a node if all memory sections and cpus of the node are removed.
1683  *
1684  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1685  * and online/offline operations before this call.
1686  */
1687 void try_offline_node(int nid)
1688 {
1689         pg_data_t *pgdat = NODE_DATA(nid);
1690         int rc;
1691 
1692         /*
1693          * If the node still spans pages (especially ZONE_DEVICE), don't
1694          * offline it. A node spans memory after move_pfn_range_to_zone(),
1695          * e.g., after the memory block was onlined.
1696          */
1697         if (pgdat->node_spanned_pages)
1698                 return;
1699 
1700         /*
1701          * Especially offline memory blocks might not be spanned by the
1702          * node. They will get spanned by the node once they get onlined.
1703          * However, they link to the node in sysfs and can get onlined later.
1704          */
1705         rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1706         if (rc)
1707                 return;
1708 
1709         if (check_cpu_on_node(pgdat))
1710                 return;
1711 
1712         /*
1713          * all memory/cpu of this node are removed, we can offline this
1714          * node now.
1715          */
1716         node_set_offline(nid);
1717         unregister_one_node(nid);
1718 }
1719 EXPORT_SYMBOL(try_offline_node);
1720 
1721 static void __release_memory_resource(resource_size_t start,
1722                                       resource_size_t size)
1723 {
1724         int ret;
1725 
1726         /*
1727          * When removing memory in the same granularity as it was added,
1728          * this function never fails. It might only fail if resources
1729          * have to be adjusted or split. We'll ignore the error, as
1730          * removing of memory cannot fail.
1731          */
1732         ret = release_mem_region_adjustable(&iomem_resource, start, size);
1733         if (ret) {
1734                 resource_size_t endres = start + size - 1;
1735 
1736                 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1737                         &start, &endres, ret);
1738         }
1739 }
1740 
1741 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1742 {
1743         int rc = 0;
1744 
1745         BUG_ON(check_hotplug_memory_range(start, size));
1746 
1747         /*
1748          * All memory blocks must be offlined before removing memory.  Check
1749          * whether all memory blocks in question are offline and return error
1750          * if this is not the case.
1751          */
1752         rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1753         if (rc)
1754                 goto done;
1755 
1756         /* remove memmap entry */
1757         firmware_map_remove(start, start + size, "System RAM");
1758         memblock_free(start, size);
1759         memblock_remove(start, size);
1760 
1761         /*
1762          * Memory block device removal under the device_hotplug_lock is
1763          * a barrier against racing online attempts.
1764          */
1765         remove_memory_block_devices(start, size);
1766 
1767         mem_hotplug_begin();
1768 
1769         arch_remove_memory(nid, start, size, NULL);
1770         __release_memory_resource(start, size);
1771 
1772         try_offline_node(nid);
1773 
1774 done:
1775         mem_hotplug_done();
1776         return rc;
1777 }
1778 
1779 /**
1780  * remove_memory
1781  * @nid: the node ID
1782  * @start: physical address of the region to remove
1783  * @size: size of the region to remove
1784  *
1785  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1786  * and online/offline operations before this call, as required by
1787  * try_offline_node().
1788  */
1789 void __remove_memory(int nid, u64 start, u64 size)
1790 {
1791 
1792         /*
1793          * trigger BUG() if some memory is not offlined prior to calling this
1794          * function
1795          */
1796         if (try_remove_memory(nid, start, size))
1797                 BUG();
1798 }
1799 
1800 /*
1801  * Remove memory if every memory block is offline, otherwise return -EBUSY is
1802  * some memory is not offline
1803  */
1804 int remove_memory(int nid, u64 start, u64 size)
1805 {
1806         int rc;
1807 
1808         lock_device_hotplug();
1809         rc  = try_remove_memory(nid, start, size);
1810         unlock_device_hotplug();
1811 
1812         return rc;
1813 }
1814 EXPORT_SYMBOL_GPL(remove_memory);
1815 #endif /* CONFIG_MEMORY_HOTREMOVE */
/* [<][>][^][v][top][bottom][index][help] */
root/mm/memory_hotplug.c

DEFINITIONS
