1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 23947be19SDave Hansen /* 33947be19SDave Hansen * linux/mm/memory_hotplug.c 43947be19SDave Hansen * 53947be19SDave Hansen * Copyright (C) 63947be19SDave Hansen */ 73947be19SDave Hansen 83947be19SDave Hansen #include <linux/stddef.h> 93947be19SDave Hansen #include <linux/mm.h> 10174cd4b1SIngo Molnar #include <linux/sched/signal.h> 113947be19SDave Hansen #include <linux/swap.h> 123947be19SDave Hansen #include <linux/interrupt.h> 133947be19SDave Hansen #include <linux/pagemap.h> 143947be19SDave Hansen #include <linux/compiler.h> 15b95f1b31SPaul Gortmaker #include <linux/export.h> 163947be19SDave Hansen #include <linux/pagevec.h> 172d1d43f6SChandra Seetharaman #include <linux/writeback.h> 183947be19SDave Hansen #include <linux/slab.h> 193947be19SDave Hansen #include <linux/sysctl.h> 203947be19SDave Hansen #include <linux/cpu.h> 213947be19SDave Hansen #include <linux/memory.h> 224b94ffdcSDan Williams #include <linux/memremap.h> 233947be19SDave Hansen #include <linux/memory_hotplug.h> 243947be19SDave Hansen #include <linux/highmem.h> 253947be19SDave Hansen #include <linux/vmalloc.h> 260a547039SKAMEZAWA Hiroyuki #include <linux/ioport.h> 270c0e6195SKAMEZAWA Hiroyuki #include <linux/delay.h> 280c0e6195SKAMEZAWA Hiroyuki #include <linux/migrate.h> 290c0e6195SKAMEZAWA Hiroyuki #include <linux/page-isolation.h> 3071088785SBadari Pulavarty #include <linux/pfn.h> 316ad696d2SAndi Kleen #include <linux/suspend.h> 326d9c285aSKOSAKI Motohiro #include <linux/mm_inline.h> 33d96ae530Sakpm@linux-foundation.org #include <linux/firmware-map.h> 3460a5a19eSTang Chen #include <linux/stop_machine.h> 35c8721bbbSNaoya Horiguchi #include <linux/hugetlb.h> 36c5320926STang Chen #include <linux/memblock.h> 37698b1b30SVlastimil Babka #include <linux/compaction.h> 38b15c8726SMichal Hocko #include <linux/rmap.h> 393947be19SDave Hansen 403947be19SDave Hansen #include <asm/tlbflush.h> 413947be19SDave Hansen 421e5ad9a3SAdrian Bunk #include "internal.h" 43e900a918SDan Williams #include "shuffle.h" 441e5ad9a3SAdrian Bunk 459d0ad8caSDaniel Kiper /* 469d0ad8caSDaniel Kiper * online_page_callback contains pointer to current page onlining function. 479d0ad8caSDaniel Kiper * Initially it is generic_online_page(). If it is required it could be 489d0ad8caSDaniel Kiper * changed by calling set_online_page_callback() for callback registration 499d0ad8caSDaniel Kiper * and restore_online_page_callback() for generic callback restore. 509d0ad8caSDaniel Kiper */ 519d0ad8caSDaniel Kiper 52a9cd410aSArun KS static void generic_online_page(struct page *page, unsigned int order); 539d0ad8caSDaniel Kiper 549d0ad8caSDaniel Kiper static online_page_callback_t online_page_callback = generic_online_page; 55bfc8c901SVladimir Davydov static DEFINE_MUTEX(online_page_callback_lock); 569d0ad8caSDaniel Kiper 573f906ba2SThomas Gleixner DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); 5820d6c96bSKOSAKI Motohiro 593f906ba2SThomas Gleixner void get_online_mems(void) 603f906ba2SThomas Gleixner { 613f906ba2SThomas Gleixner percpu_down_read(&mem_hotplug_lock); 623f906ba2SThomas Gleixner } 63bfc8c901SVladimir Davydov 643f906ba2SThomas Gleixner void put_online_mems(void) 653f906ba2SThomas Gleixner { 663f906ba2SThomas Gleixner percpu_up_read(&mem_hotplug_lock); 673f906ba2SThomas Gleixner } 68bfc8c901SVladimir Davydov 694932381eSMichal Hocko bool movable_node_enabled = false; 704932381eSMichal Hocko 718604d9e5SVitaly Kuznetsov #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE 7231bc3858SVitaly Kuznetsov bool memhp_auto_online; 738604d9e5SVitaly Kuznetsov #else 748604d9e5SVitaly Kuznetsov bool memhp_auto_online = true; 758604d9e5SVitaly Kuznetsov #endif 7631bc3858SVitaly Kuznetsov EXPORT_SYMBOL_GPL(memhp_auto_online); 7731bc3858SVitaly Kuznetsov 7886dd995dSVitaly Kuznetsov static int __init setup_memhp_default_state(char *str) 7986dd995dSVitaly Kuznetsov { 8086dd995dSVitaly Kuznetsov if (!strcmp(str, "online")) 8186dd995dSVitaly Kuznetsov memhp_auto_online = true; 8286dd995dSVitaly Kuznetsov else if (!strcmp(str, "offline")) 8386dd995dSVitaly Kuznetsov memhp_auto_online = false; 8486dd995dSVitaly Kuznetsov 8586dd995dSVitaly Kuznetsov return 1; 8686dd995dSVitaly Kuznetsov } 8786dd995dSVitaly Kuznetsov __setup("memhp_default_state=", setup_memhp_default_state); 8886dd995dSVitaly Kuznetsov 8930467e0bSDavid Rientjes void mem_hotplug_begin(void) 90bfc8c901SVladimir Davydov { 913f906ba2SThomas Gleixner cpus_read_lock(); 923f906ba2SThomas Gleixner percpu_down_write(&mem_hotplug_lock); 93bfc8c901SVladimir Davydov } 94bfc8c901SVladimir Davydov 9530467e0bSDavid Rientjes void mem_hotplug_done(void) 96bfc8c901SVladimir Davydov { 973f906ba2SThomas Gleixner percpu_up_write(&mem_hotplug_lock); 983f906ba2SThomas Gleixner cpus_read_unlock(); 99bfc8c901SVladimir Davydov } 10020d6c96bSKOSAKI Motohiro 101357b4da5SJuergen Gross u64 max_mem_size = U64_MAX; 102357b4da5SJuergen Gross 10345e0b78bSKeith Mannthey /* add this memory to iomem resource */ 10445e0b78bSKeith Mannthey static struct resource *register_memory_resource(u64 start, u64 size) 10545e0b78bSKeith Mannthey { 1062794129eSDave Hansen struct resource *res; 1072794129eSDave Hansen unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; 1082794129eSDave Hansen char *resource_name = "System RAM"; 109357b4da5SJuergen Gross 110357b4da5SJuergen Gross if (start + size > max_mem_size) 111357b4da5SJuergen Gross return ERR_PTR(-E2BIG); 112357b4da5SJuergen Gross 1132794129eSDave Hansen /* 1142794129eSDave Hansen * Request ownership of the new memory range. This might be 1152794129eSDave Hansen * a child of an existing resource that was present but 1162794129eSDave Hansen * not marked as busy. 1172794129eSDave Hansen */ 1182794129eSDave Hansen res = __request_region(&iomem_resource, start, size, 1192794129eSDave Hansen resource_name, flags); 12045e0b78bSKeith Mannthey 1212794129eSDave Hansen if (!res) { 1222794129eSDave Hansen pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", 1232794129eSDave Hansen start, start + size); 1246f754ba4SVitaly Kuznetsov return ERR_PTR(-EEXIST); 12545e0b78bSKeith Mannthey } 12645e0b78bSKeith Mannthey return res; 12745e0b78bSKeith Mannthey } 12845e0b78bSKeith Mannthey 12945e0b78bSKeith Mannthey static void release_memory_resource(struct resource *res) 13045e0b78bSKeith Mannthey { 13145e0b78bSKeith Mannthey if (!res) 13245e0b78bSKeith Mannthey return; 13345e0b78bSKeith Mannthey release_resource(res); 13445e0b78bSKeith Mannthey kfree(res); 13545e0b78bSKeith Mannthey } 13645e0b78bSKeith Mannthey 13753947027SKeith Mannthey #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE 13846723bfaSYasuaki Ishimatsu void get_page_bootmem(unsigned long info, struct page *page, 1395f24ce5fSAndrea Arcangeli unsigned long type) 14004753278SYasunori Goto { 141ddffe98dSYasuaki Ishimatsu page->freelist = (void *)type; 14204753278SYasunori Goto SetPagePrivate(page); 14304753278SYasunori Goto set_page_private(page, info); 144fe896d18SJoonsoo Kim page_ref_inc(page); 14504753278SYasunori Goto } 14604753278SYasunori Goto 147170a5a7eSJiang Liu void put_page_bootmem(struct page *page) 14804753278SYasunori Goto { 1495f24ce5fSAndrea Arcangeli unsigned long type; 15004753278SYasunori Goto 151ddffe98dSYasuaki Ishimatsu type = (unsigned long) page->freelist; 1525f24ce5fSAndrea Arcangeli BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || 1535f24ce5fSAndrea Arcangeli type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); 15404753278SYasunori Goto 155fe896d18SJoonsoo Kim if (page_ref_dec_return(page) == 1) { 156ddffe98dSYasuaki Ishimatsu page->freelist = NULL; 15704753278SYasunori Goto ClearPagePrivate(page); 15804753278SYasunori Goto set_page_private(page, 0); 1595f24ce5fSAndrea Arcangeli INIT_LIST_HEAD(&page->lru); 160170a5a7eSJiang Liu free_reserved_page(page); 16104753278SYasunori Goto } 16204753278SYasunori Goto } 16304753278SYasunori Goto 16446723bfaSYasuaki Ishimatsu #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE 16546723bfaSYasuaki Ishimatsu #ifndef CONFIG_SPARSEMEM_VMEMMAP 166d92bc318SAdrian Bunk static void register_page_bootmem_info_section(unsigned long start_pfn) 16704753278SYasunori Goto { 168f1eca35aSDan Williams unsigned long mapsize, section_nr, i; 16904753278SYasunori Goto struct mem_section *ms; 17004753278SYasunori Goto struct page *page, *memmap; 171f1eca35aSDan Williams struct mem_section_usage *usage; 17204753278SYasunori Goto 17304753278SYasunori Goto section_nr = pfn_to_section_nr(start_pfn); 17404753278SYasunori Goto ms = __nr_to_section(section_nr); 17504753278SYasunori Goto 17604753278SYasunori Goto /* Get section's memmap address */ 17704753278SYasunori Goto memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); 17804753278SYasunori Goto 17904753278SYasunori Goto /* 18004753278SYasunori Goto * Get page for the memmap's phys address 18104753278SYasunori Goto * XXX: need more consideration for sparse_vmemmap... 18204753278SYasunori Goto */ 18304753278SYasunori Goto page = virt_to_page(memmap); 18404753278SYasunori Goto mapsize = sizeof(struct page) * PAGES_PER_SECTION; 18504753278SYasunori Goto mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; 18604753278SYasunori Goto 18704753278SYasunori Goto /* remember memmap's page */ 18804753278SYasunori Goto for (i = 0; i < mapsize; i++, page++) 18904753278SYasunori Goto get_page_bootmem(section_nr, page, SECTION_INFO); 19004753278SYasunori Goto 191f1eca35aSDan Williams usage = ms->usage; 192f1eca35aSDan Williams page = virt_to_page(usage); 19304753278SYasunori Goto 194f1eca35aSDan Williams mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; 19504753278SYasunori Goto 19604753278SYasunori Goto for (i = 0; i < mapsize; i++, page++) 197af370fb8SYasunori Goto get_page_bootmem(section_nr, page, MIX_SECTION_INFO); 19804753278SYasunori Goto 19904753278SYasunori Goto } 20046723bfaSYasuaki Ishimatsu #else /* CONFIG_SPARSEMEM_VMEMMAP */ 20146723bfaSYasuaki Ishimatsu static void register_page_bootmem_info_section(unsigned long start_pfn) 20246723bfaSYasuaki Ishimatsu { 203f1eca35aSDan Williams unsigned long mapsize, section_nr, i; 20446723bfaSYasuaki Ishimatsu struct mem_section *ms; 20546723bfaSYasuaki Ishimatsu struct page *page, *memmap; 206f1eca35aSDan Williams struct mem_section_usage *usage; 20746723bfaSYasuaki Ishimatsu 20846723bfaSYasuaki Ishimatsu section_nr = pfn_to_section_nr(start_pfn); 20946723bfaSYasuaki Ishimatsu ms = __nr_to_section(section_nr); 21046723bfaSYasuaki Ishimatsu 21146723bfaSYasuaki Ishimatsu memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); 21246723bfaSYasuaki Ishimatsu 21346723bfaSYasuaki Ishimatsu register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); 21446723bfaSYasuaki Ishimatsu 215f1eca35aSDan Williams usage = ms->usage; 216f1eca35aSDan Williams page = virt_to_page(usage); 21746723bfaSYasuaki Ishimatsu 218f1eca35aSDan Williams mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; 21946723bfaSYasuaki Ishimatsu 22046723bfaSYasuaki Ishimatsu for (i = 0; i < mapsize; i++, page++) 22146723bfaSYasuaki Ishimatsu get_page_bootmem(section_nr, page, MIX_SECTION_INFO); 22246723bfaSYasuaki Ishimatsu } 22346723bfaSYasuaki Ishimatsu #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ 22404753278SYasunori Goto 2257ded384aSLinus Torvalds void __init register_page_bootmem_info_node(struct pglist_data *pgdat) 22604753278SYasunori Goto { 22704753278SYasunori Goto unsigned long i, pfn, end_pfn, nr_pages; 22804753278SYasunori Goto int node = pgdat->node_id; 22904753278SYasunori Goto struct page *page; 23004753278SYasunori Goto 23104753278SYasunori Goto nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; 23204753278SYasunori Goto page = virt_to_page(pgdat); 23304753278SYasunori Goto 23404753278SYasunori Goto for (i = 0; i < nr_pages; i++, page++) 23504753278SYasunori Goto get_page_bootmem(node, page, NODE_INFO); 23604753278SYasunori Goto 23704753278SYasunori Goto pfn = pgdat->node_start_pfn; 238c1f19495SCody P Schafer end_pfn = pgdat_end_pfn(pgdat); 23904753278SYasunori Goto 2407e9f5eb0STang Chen /* register section info */ 241f14851afSqiuxishi for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { 242f14851afSqiuxishi /* 243f14851afSqiuxishi * Some platforms can assign the same pfn to multiple nodes - on 244f14851afSqiuxishi * node0 as well as nodeN. To avoid registering a pfn against 245f14851afSqiuxishi * multiple nodes we check that this pfn does not already 2467e9f5eb0STang Chen * reside in some other nodes. 247f14851afSqiuxishi */ 248f65e91dfSYang Shi if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) 24904753278SYasunori Goto register_page_bootmem_info_section(pfn); 250f14851afSqiuxishi } 25104753278SYasunori Goto } 25246723bfaSYasuaki Ishimatsu #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ 25304753278SYasunori Goto 2547ea62160SDan Williams static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, 2557ea62160SDan Williams const char *reason) 2567ea62160SDan Williams { 2577ea62160SDan Williams /* 2587ea62160SDan Williams * Disallow all operations smaller than a sub-section and only 2597ea62160SDan Williams * allow operations smaller than a section for 2607ea62160SDan Williams * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() 2617ea62160SDan Williams * enforces a larger memory_block_size_bytes() granularity for 2627ea62160SDan Williams * memory that will be marked online, so this check should only 2637ea62160SDan Williams * fire for direct arch_{add,remove}_memory() users outside of 2647ea62160SDan Williams * add_memory_resource(). 2657ea62160SDan Williams */ 2667ea62160SDan Williams unsigned long min_align; 2677ea62160SDan Williams 2687ea62160SDan Williams if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 2697ea62160SDan Williams min_align = PAGES_PER_SUBSECTION; 2707ea62160SDan Williams else 2717ea62160SDan Williams min_align = PAGES_PER_SECTION; 2727ea62160SDan Williams if (!IS_ALIGNED(pfn, min_align) 2737ea62160SDan Williams || !IS_ALIGNED(nr_pages, min_align)) { 2747ea62160SDan Williams WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", 2757ea62160SDan Williams reason, pfn, pfn + nr_pages - 1); 2767ea62160SDan Williams return -EINVAL; 2777ea62160SDan Williams } 2787ea62160SDan Williams return 0; 2797ea62160SDan Williams } 2807ea62160SDan Williams 2814edd7cefSDavid Rientjes /* 2824edd7cefSDavid Rientjes * Reasonably generic function for adding memory. It is 2834edd7cefSDavid Rientjes * expected that archs that support memory hotplug will 2844edd7cefSDavid Rientjes * call this function after deciding the zone to which to 2854edd7cefSDavid Rientjes * add the new pages. 2864edd7cefSDavid Rientjes */ 2877ea62160SDan Williams int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, 2887ea62160SDan Williams struct mhp_restrictions *restrictions) 2894edd7cefSDavid Rientjes { 2909a845030SDan Williams int err; 2919a845030SDan Williams unsigned long nr, start_sec, end_sec; 292940519f0SMichal Hocko struct vmem_altmap *altmap = restrictions->altmap; 2934b94ffdcSDan Williams 2944b94ffdcSDan Williams if (altmap) { 2954b94ffdcSDan Williams /* 2964b94ffdcSDan Williams * Validate altmap is within bounds of the total request 2974b94ffdcSDan Williams */ 2987ea62160SDan Williams if (altmap->base_pfn != pfn 2994b94ffdcSDan Williams || vmem_altmap_offset(altmap) > nr_pages) { 3004b94ffdcSDan Williams pr_warn_once("memory add fail, invalid altmap\n"); 3017ea62160SDan Williams return -EINVAL; 3024b94ffdcSDan Williams } 3034b94ffdcSDan Williams altmap->alloc = 0; 3044b94ffdcSDan Williams } 3054b94ffdcSDan Williams 3067ea62160SDan Williams err = check_pfn_span(pfn, nr_pages, "add"); 3077ea62160SDan Williams if (err) 3087ea62160SDan Williams return err; 3097ea62160SDan Williams 3107ea62160SDan Williams start_sec = pfn_to_section_nr(pfn); 3117ea62160SDan Williams end_sec = pfn_to_section_nr(pfn + nr_pages - 1); 3129a845030SDan Williams for (nr = start_sec; nr <= end_sec; nr++) { 3137ea62160SDan Williams unsigned long pfns; 3147ea62160SDan Williams 3157ea62160SDan Williams pfns = min(nr_pages, PAGES_PER_SECTION 3167ea62160SDan Williams - (pfn & ~PAGE_SECTION_MASK)); 317ba72b4c8SDan Williams err = sparse_add_section(nid, pfn, pfns, altmap); 318ba72b4c8SDan Williams if (err) 319ba72b4c8SDan Williams break; 3207ea62160SDan Williams pfn += pfns; 3217ea62160SDan Williams nr_pages -= pfns; 322f64ac5e6SMichal Hocko cond_resched(); 3234edd7cefSDavid Rientjes } 324c435a390SZhu Guihua vmemmap_populate_print_last(); 3254edd7cefSDavid Rientjes return err; 3264edd7cefSDavid Rientjes } 3274edd7cefSDavid Rientjes 328815121d2SYasuaki Ishimatsu /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ 329d09b0137SYASUAKI ISHIMATSU static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, 330815121d2SYasuaki Ishimatsu unsigned long start_pfn, 331815121d2SYasuaki Ishimatsu unsigned long end_pfn) 332815121d2SYasuaki Ishimatsu { 33349ba3c6bSDan Williams for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { 33449ba3c6bSDan Williams if (unlikely(!pfn_valid(start_pfn))) 335815121d2SYasuaki Ishimatsu continue; 336815121d2SYasuaki Ishimatsu 337815121d2SYasuaki Ishimatsu if (unlikely(pfn_to_nid(start_pfn) != nid)) 338815121d2SYasuaki Ishimatsu continue; 339815121d2SYasuaki Ishimatsu 340815121d2SYasuaki Ishimatsu if (zone && zone != page_zone(pfn_to_page(start_pfn))) 341815121d2SYasuaki Ishimatsu continue; 342815121d2SYasuaki Ishimatsu 343815121d2SYasuaki Ishimatsu return start_pfn; 344815121d2SYasuaki Ishimatsu } 345815121d2SYasuaki Ishimatsu 346815121d2SYasuaki Ishimatsu return 0; 347815121d2SYasuaki Ishimatsu } 348815121d2SYasuaki Ishimatsu 349815121d2SYasuaki Ishimatsu /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ 350d09b0137SYASUAKI ISHIMATSU static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, 351815121d2SYasuaki Ishimatsu unsigned long start_pfn, 352815121d2SYasuaki Ishimatsu unsigned long end_pfn) 353815121d2SYasuaki Ishimatsu { 354815121d2SYasuaki Ishimatsu unsigned long pfn; 355815121d2SYasuaki Ishimatsu 356815121d2SYasuaki Ishimatsu /* pfn is the end pfn of a memory section. */ 357815121d2SYasuaki Ishimatsu pfn = end_pfn - 1; 35849ba3c6bSDan Williams for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { 35949ba3c6bSDan Williams if (unlikely(!pfn_valid(pfn))) 360815121d2SYasuaki Ishimatsu continue; 361815121d2SYasuaki Ishimatsu 362815121d2SYasuaki Ishimatsu if (unlikely(pfn_to_nid(pfn) != nid)) 363815121d2SYasuaki Ishimatsu continue; 364815121d2SYasuaki Ishimatsu 365815121d2SYasuaki Ishimatsu if (zone && zone != page_zone(pfn_to_page(pfn))) 366815121d2SYasuaki Ishimatsu continue; 367815121d2SYasuaki Ishimatsu 368815121d2SYasuaki Ishimatsu return pfn; 369815121d2SYasuaki Ishimatsu } 370815121d2SYasuaki Ishimatsu 371815121d2SYasuaki Ishimatsu return 0; 372815121d2SYasuaki Ishimatsu } 373815121d2SYasuaki Ishimatsu 374815121d2SYasuaki Ishimatsu static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, 375815121d2SYasuaki Ishimatsu unsigned long end_pfn) 376815121d2SYasuaki Ishimatsu { 377815121d2SYasuaki Ishimatsu unsigned long zone_start_pfn = zone->zone_start_pfn; 378c33bc315SXishi Qiu unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ 379c33bc315SXishi Qiu unsigned long zone_end_pfn = z; 380815121d2SYasuaki Ishimatsu unsigned long pfn; 381815121d2SYasuaki Ishimatsu int nid = zone_to_nid(zone); 382815121d2SYasuaki Ishimatsu 383815121d2SYasuaki Ishimatsu zone_span_writelock(zone); 384815121d2SYasuaki Ishimatsu if (zone_start_pfn == start_pfn) { 385815121d2SYasuaki Ishimatsu /* 386815121d2SYasuaki Ishimatsu * If the section is smallest section in the zone, it need 387815121d2SYasuaki Ishimatsu * shrink zone->zone_start_pfn and zone->zone_spanned_pages. 388815121d2SYasuaki Ishimatsu * In this case, we find second smallest valid mem_section 389815121d2SYasuaki Ishimatsu * for shrinking zone. 390815121d2SYasuaki Ishimatsu */ 391815121d2SYasuaki Ishimatsu pfn = find_smallest_section_pfn(nid, zone, end_pfn, 392815121d2SYasuaki Ishimatsu zone_end_pfn); 393815121d2SYasuaki Ishimatsu if (pfn) { 394815121d2SYasuaki Ishimatsu zone->zone_start_pfn = pfn; 395815121d2SYasuaki Ishimatsu zone->spanned_pages = zone_end_pfn - pfn; 396815121d2SYasuaki Ishimatsu } 397815121d2SYasuaki Ishimatsu } else if (zone_end_pfn == end_pfn) { 398815121d2SYasuaki Ishimatsu /* 399815121d2SYasuaki Ishimatsu * If the section is biggest section in the zone, it need 400815121d2SYasuaki Ishimatsu * shrink zone->spanned_pages. 401815121d2SYasuaki Ishimatsu * In this case, we find second biggest valid mem_section for 402815121d2SYasuaki Ishimatsu * shrinking zone. 403815121d2SYasuaki Ishimatsu */ 404815121d2SYasuaki Ishimatsu pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, 405815121d2SYasuaki Ishimatsu start_pfn); 406815121d2SYasuaki Ishimatsu if (pfn) 407815121d2SYasuaki Ishimatsu zone->spanned_pages = pfn - zone_start_pfn + 1; 408815121d2SYasuaki Ishimatsu } 409815121d2SYasuaki Ishimatsu 410815121d2SYasuaki Ishimatsu /* 411815121d2SYasuaki Ishimatsu * The section is not biggest or smallest mem_section in the zone, it 412815121d2SYasuaki Ishimatsu * only creates a hole in the zone. So in this case, we need not 413815121d2SYasuaki Ishimatsu * change the zone. But perhaps, the zone has only hole data. Thus 414815121d2SYasuaki Ishimatsu * it check the zone has only hole or not. 415815121d2SYasuaki Ishimatsu */ 416815121d2SYasuaki Ishimatsu pfn = zone_start_pfn; 41749ba3c6bSDan Williams for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) { 41849ba3c6bSDan Williams if (unlikely(!pfn_valid(pfn))) 419815121d2SYasuaki Ishimatsu continue; 420815121d2SYasuaki Ishimatsu 421815121d2SYasuaki Ishimatsu if (page_zone(pfn_to_page(pfn)) != zone) 422815121d2SYasuaki Ishimatsu continue; 423815121d2SYasuaki Ishimatsu 42449ba3c6bSDan Williams /* Skip range to be removed */ 42549ba3c6bSDan Williams if (pfn >= start_pfn && pfn < end_pfn) 426815121d2SYasuaki Ishimatsu continue; 427815121d2SYasuaki Ishimatsu 428815121d2SYasuaki Ishimatsu /* If we find valid section, we have nothing to do */ 429815121d2SYasuaki Ishimatsu zone_span_writeunlock(zone); 430815121d2SYasuaki Ishimatsu return; 431815121d2SYasuaki Ishimatsu } 432815121d2SYasuaki Ishimatsu 433815121d2SYasuaki Ishimatsu /* The zone has no valid section */ 434815121d2SYasuaki Ishimatsu zone->zone_start_pfn = 0; 435815121d2SYasuaki Ishimatsu zone->spanned_pages = 0; 436815121d2SYasuaki Ishimatsu zone_span_writeunlock(zone); 437815121d2SYasuaki Ishimatsu } 438815121d2SYasuaki Ishimatsu 43900d6c019SDavid Hildenbrand static void update_pgdat_span(struct pglist_data *pgdat) 440815121d2SYasuaki Ishimatsu { 44100d6c019SDavid Hildenbrand unsigned long node_start_pfn = 0, node_end_pfn = 0; 44200d6c019SDavid Hildenbrand struct zone *zone; 443815121d2SYasuaki Ishimatsu 44400d6c019SDavid Hildenbrand for (zone = pgdat->node_zones; 44500d6c019SDavid Hildenbrand zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { 44600d6c019SDavid Hildenbrand unsigned long zone_end_pfn = zone->zone_start_pfn + 44700d6c019SDavid Hildenbrand zone->spanned_pages; 44800d6c019SDavid Hildenbrand 44900d6c019SDavid Hildenbrand /* No need to lock the zones, they can't change. */ 450*656d5711SDavid Hildenbrand if (!zone->spanned_pages) 451*656d5711SDavid Hildenbrand continue; 452*656d5711SDavid Hildenbrand if (!node_end_pfn) { 453*656d5711SDavid Hildenbrand node_start_pfn = zone->zone_start_pfn; 454*656d5711SDavid Hildenbrand node_end_pfn = zone_end_pfn; 455*656d5711SDavid Hildenbrand continue; 456*656d5711SDavid Hildenbrand } 457*656d5711SDavid Hildenbrand 45800d6c019SDavid Hildenbrand if (zone_end_pfn > node_end_pfn) 45900d6c019SDavid Hildenbrand node_end_pfn = zone_end_pfn; 46000d6c019SDavid Hildenbrand if (zone->zone_start_pfn < node_start_pfn) 46100d6c019SDavid Hildenbrand node_start_pfn = zone->zone_start_pfn; 462815121d2SYasuaki Ishimatsu } 463815121d2SYasuaki Ishimatsu 46400d6c019SDavid Hildenbrand pgdat->node_start_pfn = node_start_pfn; 46500d6c019SDavid Hildenbrand pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; 466815121d2SYasuaki Ishimatsu } 467815121d2SYasuaki Ishimatsu 4687ea62160SDan Williams static void __remove_zone(struct zone *zone, unsigned long start_pfn, 4697ea62160SDan Williams unsigned long nr_pages) 470815121d2SYasuaki Ishimatsu { 471815121d2SYasuaki Ishimatsu struct pglist_data *pgdat = zone->zone_pgdat; 472815121d2SYasuaki Ishimatsu unsigned long flags; 473815121d2SYasuaki Ishimatsu 474815121d2SYasuaki Ishimatsu pgdat_resize_lock(zone->zone_pgdat, &flags); 475815121d2SYasuaki Ishimatsu shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); 47600d6c019SDavid Hildenbrand update_pgdat_span(pgdat); 477815121d2SYasuaki Ishimatsu pgdat_resize_unlock(zone->zone_pgdat, &flags); 478815121d2SYasuaki Ishimatsu } 479815121d2SYasuaki Ishimatsu 4807ea62160SDan Williams static void __remove_section(struct zone *zone, unsigned long pfn, 4817ea62160SDan Williams unsigned long nr_pages, unsigned long map_offset, 4829d1d887dSDavid Hildenbrand struct vmem_altmap *altmap) 483ea01ea93SBadari Pulavarty { 4847ea62160SDan Williams struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn)); 485ea01ea93SBadari Pulavarty 4869d1d887dSDavid Hildenbrand if (WARN_ON_ONCE(!valid_section(ms))) 4879d1d887dSDavid Hildenbrand return; 488ea01ea93SBadari Pulavarty 4897ea62160SDan Williams __remove_zone(zone, pfn, nr_pages); 490ba72b4c8SDan Williams sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); 491ea01ea93SBadari Pulavarty } 492ea01ea93SBadari Pulavarty 493ea01ea93SBadari Pulavarty /** 494ea01ea93SBadari Pulavarty * __remove_pages() - remove sections of pages from a zone 495ea01ea93SBadari Pulavarty * @zone: zone from which pages need to be removed 4967ea62160SDan Williams * @pfn: starting pageframe (must be aligned to start of a section) 497ea01ea93SBadari Pulavarty * @nr_pages: number of pages to remove (must be multiple of section size) 498e8b098fcSMike Rapoport * @altmap: alternative device page map or %NULL if default memmap is used 499ea01ea93SBadari Pulavarty * 500ea01ea93SBadari Pulavarty * Generic helper function to remove section mappings and sysfs entries 501ea01ea93SBadari Pulavarty * for the section of the memory we are removing. Caller needs to make 502ea01ea93SBadari Pulavarty * sure that pages are marked reserved and zones are adjust properly by 503ea01ea93SBadari Pulavarty * calling offline_pages(). 504ea01ea93SBadari Pulavarty */ 5057ea62160SDan Williams void __remove_pages(struct zone *zone, unsigned long pfn, 506da024512SChristoph Hellwig unsigned long nr_pages, struct vmem_altmap *altmap) 507ea01ea93SBadari Pulavarty { 5084b94ffdcSDan Williams unsigned long map_offset = 0; 5099a845030SDan Williams unsigned long nr, start_sec, end_sec; 5104b94ffdcSDan Williams 5114b94ffdcSDan Williams map_offset = vmem_altmap_offset(altmap); 512ea01ea93SBadari Pulavarty 5137cf91a98SJoonsoo Kim clear_zone_contiguous(zone); 5147cf91a98SJoonsoo Kim 5157ea62160SDan Williams if (check_pfn_span(pfn, nr_pages, "remove")) 5167ea62160SDan Williams return; 517ea01ea93SBadari Pulavarty 5187ea62160SDan Williams start_sec = pfn_to_section_nr(pfn); 5197ea62160SDan Williams end_sec = pfn_to_section_nr(pfn + nr_pages - 1); 5209a845030SDan Williams for (nr = start_sec; nr <= end_sec; nr++) { 5217ea62160SDan Williams unsigned long pfns; 5224b94ffdcSDan Williams 523dd33ad7bSMichal Hocko cond_resched(); 5247ea62160SDan Williams pfns = min(nr_pages, PAGES_PER_SECTION 5257ea62160SDan Williams - (pfn & ~PAGE_SECTION_MASK)); 5267ea62160SDan Williams __remove_section(zone, pfn, pfns, map_offset, altmap); 5277ea62160SDan Williams pfn += pfns; 5287ea62160SDan Williams nr_pages -= pfns; 5294b94ffdcSDan Williams map_offset = 0; 530ea01ea93SBadari Pulavarty } 5317cf91a98SJoonsoo Kim 5327cf91a98SJoonsoo Kim set_zone_contiguous(zone); 533ea01ea93SBadari Pulavarty } 534ea01ea93SBadari Pulavarty 5359d0ad8caSDaniel Kiper int set_online_page_callback(online_page_callback_t callback) 5369d0ad8caSDaniel Kiper { 5379d0ad8caSDaniel Kiper int rc = -EINVAL; 5389d0ad8caSDaniel Kiper 539bfc8c901SVladimir Davydov get_online_mems(); 540bfc8c901SVladimir Davydov mutex_lock(&online_page_callback_lock); 5419d0ad8caSDaniel Kiper 5429d0ad8caSDaniel Kiper if (online_page_callback == generic_online_page) { 5439d0ad8caSDaniel Kiper online_page_callback = callback; 5449d0ad8caSDaniel Kiper rc = 0; 5459d0ad8caSDaniel Kiper } 5469d0ad8caSDaniel Kiper 547bfc8c901SVladimir Davydov mutex_unlock(&online_page_callback_lock); 548bfc8c901SVladimir Davydov put_online_mems(); 5499d0ad8caSDaniel Kiper 5509d0ad8caSDaniel Kiper return rc; 5519d0ad8caSDaniel Kiper } 5529d0ad8caSDaniel Kiper EXPORT_SYMBOL_GPL(set_online_page_callback); 5539d0ad8caSDaniel Kiper 5549d0ad8caSDaniel Kiper int restore_online_page_callback(online_page_callback_t callback) 5559d0ad8caSDaniel Kiper { 5569d0ad8caSDaniel Kiper int rc = -EINVAL; 5579d0ad8caSDaniel Kiper 558bfc8c901SVladimir Davydov get_online_mems(); 559bfc8c901SVladimir Davydov mutex_lock(&online_page_callback_lock); 5609d0ad8caSDaniel Kiper 5619d0ad8caSDaniel Kiper if (online_page_callback == callback) { 5629d0ad8caSDaniel Kiper online_page_callback = generic_online_page; 5639d0ad8caSDaniel Kiper rc = 0; 5649d0ad8caSDaniel Kiper } 5659d0ad8caSDaniel Kiper 566bfc8c901SVladimir Davydov mutex_unlock(&online_page_callback_lock); 567bfc8c901SVladimir Davydov put_online_mems(); 5689d0ad8caSDaniel Kiper 5699d0ad8caSDaniel Kiper return rc; 5709d0ad8caSDaniel Kiper } 5719d0ad8caSDaniel Kiper EXPORT_SYMBOL_GPL(restore_online_page_callback); 5729d0ad8caSDaniel Kiper 5739d0ad8caSDaniel Kiper void __online_page_set_limits(struct page *page) 574180c06efSJeremy Fitzhardinge { 5759d0ad8caSDaniel Kiper } 5769d0ad8caSDaniel Kiper EXPORT_SYMBOL_GPL(__online_page_set_limits); 5779d0ad8caSDaniel Kiper 5789d0ad8caSDaniel Kiper void __online_page_increment_counters(struct page *page) 5799d0ad8caSDaniel Kiper { 5803dcc0571SJiang Liu adjust_managed_page_count(page, 1); 5819d0ad8caSDaniel Kiper } 5829d0ad8caSDaniel Kiper EXPORT_SYMBOL_GPL(__online_page_increment_counters); 583180c06efSJeremy Fitzhardinge 5849d0ad8caSDaniel Kiper void __online_page_free(struct page *page) 5859d0ad8caSDaniel Kiper { 5863dcc0571SJiang Liu __free_reserved_page(page); 587180c06efSJeremy Fitzhardinge } 5889d0ad8caSDaniel Kiper EXPORT_SYMBOL_GPL(__online_page_free); 5899d0ad8caSDaniel Kiper 590a9cd410aSArun KS static void generic_online_page(struct page *page, unsigned int order) 5919d0ad8caSDaniel Kiper { 592cd02cf1aSQian Cai kernel_map_pages(page, 1 << order, 1); 593a9cd410aSArun KS __free_pages_core(page, order); 594a9cd410aSArun KS totalram_pages_add(1UL << order); 595a9cd410aSArun KS #ifdef CONFIG_HIGHMEM 596a9cd410aSArun KS if (PageHighMem(page)) 597a9cd410aSArun KS totalhigh_pages_add(1UL << order); 598a9cd410aSArun KS #endif 599a9cd410aSArun KS } 600a9cd410aSArun KS 60175884fb1SKAMEZAWA Hiroyuki static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, 60275884fb1SKAMEZAWA Hiroyuki void *arg) 6033947be19SDave Hansen { 604b2c2ab20SDavid Hildenbrand const unsigned long end_pfn = start_pfn + nr_pages; 605b2c2ab20SDavid Hildenbrand unsigned long pfn; 606b2c2ab20SDavid Hildenbrand int order; 6072d070eabSMichal Hocko 608b2c2ab20SDavid Hildenbrand /* 609b2c2ab20SDavid Hildenbrand * Online the pages. The callback might decide to keep some pages 610b2c2ab20SDavid Hildenbrand * PG_reserved (to add them to the buddy later), but we still account 611b2c2ab20SDavid Hildenbrand * them as being online/belonging to this zone ("present"). 612b2c2ab20SDavid Hildenbrand */ 613b2c2ab20SDavid Hildenbrand for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) { 614b2c2ab20SDavid Hildenbrand order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn))); 615bd02cc01SDavid Hildenbrand /* __free_pages_core() wants pfns to be aligned to the order */ 616bd02cc01SDavid Hildenbrand if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order))) 617bd02cc01SDavid Hildenbrand order = 0; 618b2c2ab20SDavid Hildenbrand (*online_page_callback)(pfn_to_page(pfn), order); 619b2c2ab20SDavid Hildenbrand } 6202d070eabSMichal Hocko 621b2c2ab20SDavid Hildenbrand /* mark all involved sections as online */ 622b2c2ab20SDavid Hildenbrand online_mem_sections(start_pfn, end_pfn); 623b2c2ab20SDavid Hildenbrand 624b2c2ab20SDavid Hildenbrand *(unsigned long *)arg += nr_pages; 62575884fb1SKAMEZAWA Hiroyuki return 0; 62675884fb1SKAMEZAWA Hiroyuki } 62775884fb1SKAMEZAWA Hiroyuki 628d9713679SLai Jiangshan /* check which state of node_states will be changed when online memory */ 629d9713679SLai Jiangshan static void node_states_check_changes_online(unsigned long nr_pages, 630d9713679SLai Jiangshan struct zone *zone, struct memory_notify *arg) 631d9713679SLai Jiangshan { 632d9713679SLai Jiangshan int nid = zone_to_nid(zone); 633d9713679SLai Jiangshan 63498fa15f3SAnshuman Khandual arg->status_change_nid = NUMA_NO_NODE; 63598fa15f3SAnshuman Khandual arg->status_change_nid_normal = NUMA_NO_NODE; 63698fa15f3SAnshuman Khandual arg->status_change_nid_high = NUMA_NO_NODE; 6376715ddf9SLai Jiangshan 6386715ddf9SLai Jiangshan if (!node_state(nid, N_MEMORY)) 639d9713679SLai Jiangshan arg->status_change_nid = nid; 6408efe33f4SOscar Salvador if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) 6418efe33f4SOscar Salvador arg->status_change_nid_normal = nid; 6428efe33f4SOscar Salvador #ifdef CONFIG_HIGHMEM 643d3ba3ae1SBaoquan He if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) 6448efe33f4SOscar Salvador arg->status_change_nid_high = nid; 6458efe33f4SOscar Salvador #endif 646d9713679SLai Jiangshan } 647d9713679SLai Jiangshan 648d9713679SLai Jiangshan static void node_states_set_node(int node, struct memory_notify *arg) 649d9713679SLai Jiangshan { 650d9713679SLai Jiangshan if (arg->status_change_nid_normal >= 0) 651d9713679SLai Jiangshan node_set_state(node, N_NORMAL_MEMORY); 652d9713679SLai Jiangshan 6536715ddf9SLai Jiangshan if (arg->status_change_nid_high >= 0) 654d9713679SLai Jiangshan node_set_state(node, N_HIGH_MEMORY); 6556715ddf9SLai Jiangshan 65683d83612SOscar Salvador if (arg->status_change_nid >= 0) 6576715ddf9SLai Jiangshan node_set_state(node, N_MEMORY); 658d9713679SLai Jiangshan } 659d9713679SLai Jiangshan 660f1dd2cd1SMichal Hocko static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, 661f1dd2cd1SMichal Hocko unsigned long nr_pages) 662f1dd2cd1SMichal Hocko { 663f1dd2cd1SMichal Hocko unsigned long old_end_pfn = zone_end_pfn(zone); 664f1dd2cd1SMichal Hocko 665f1dd2cd1SMichal Hocko if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) 666f1dd2cd1SMichal Hocko zone->zone_start_pfn = start_pfn; 667f1dd2cd1SMichal Hocko 668f1dd2cd1SMichal Hocko zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; 669f1dd2cd1SMichal Hocko } 670f1dd2cd1SMichal Hocko 671f1dd2cd1SMichal Hocko static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, 672f1dd2cd1SMichal Hocko unsigned long nr_pages) 673f1dd2cd1SMichal Hocko { 674f1dd2cd1SMichal Hocko unsigned long old_end_pfn = pgdat_end_pfn(pgdat); 675f1dd2cd1SMichal Hocko 676f1dd2cd1SMichal Hocko if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) 677f1dd2cd1SMichal Hocko pgdat->node_start_pfn = start_pfn; 678f1dd2cd1SMichal Hocko 679f1dd2cd1SMichal Hocko pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; 680f1dd2cd1SMichal Hocko 6813fccb74cSDavid Hildenbrand } 6823fccb74cSDavid Hildenbrand /* 6833fccb74cSDavid Hildenbrand * Associate the pfn range with the given zone, initializing the memmaps 6843fccb74cSDavid Hildenbrand * and resizing the pgdat/zone data to span the added pages. After this 6853fccb74cSDavid Hildenbrand * call, all affected pages are PG_reserved. 6863fccb74cSDavid Hildenbrand */ 687a99583e7SChristoph Hellwig void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, 688a99583e7SChristoph Hellwig unsigned long nr_pages, struct vmem_altmap *altmap) 689f1dd2cd1SMichal Hocko { 690f1dd2cd1SMichal Hocko struct pglist_data *pgdat = zone->zone_pgdat; 691f1dd2cd1SMichal Hocko int nid = pgdat->node_id; 692f1dd2cd1SMichal Hocko unsigned long flags; 693f1dd2cd1SMichal Hocko 694f1dd2cd1SMichal Hocko clear_zone_contiguous(zone); 695f1dd2cd1SMichal Hocko 696f1dd2cd1SMichal Hocko /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ 697f1dd2cd1SMichal Hocko pgdat_resize_lock(pgdat, &flags); 698f1dd2cd1SMichal Hocko zone_span_writelock(zone); 699fa004ab7SWei Yang if (zone_is_empty(zone)) 700fa004ab7SWei Yang init_currently_empty_zone(zone, start_pfn, nr_pages); 701f1dd2cd1SMichal Hocko resize_zone_range(zone, start_pfn, nr_pages); 702f1dd2cd1SMichal Hocko zone_span_writeunlock(zone); 703f1dd2cd1SMichal Hocko resize_pgdat_range(pgdat, start_pfn, nr_pages); 704f1dd2cd1SMichal Hocko pgdat_resize_unlock(pgdat, &flags); 705f1dd2cd1SMichal Hocko 706f1dd2cd1SMichal Hocko /* 707f1dd2cd1SMichal Hocko * TODO now we have a visible range of pages which are not associated 708f1dd2cd1SMichal Hocko * with their zone properly. Not nice but set_pfnblock_flags_mask 709f1dd2cd1SMichal Hocko * expects the zone spans the pfn range. All the pages in the range 710f1dd2cd1SMichal Hocko * are reserved so nobody should be touching them so we should be safe 711f1dd2cd1SMichal Hocko */ 712a99583e7SChristoph Hellwig memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, 713a99583e7SChristoph Hellwig MEMMAP_HOTPLUG, altmap); 714f1dd2cd1SMichal Hocko 715f1dd2cd1SMichal Hocko set_zone_contiguous(zone); 716f1dd2cd1SMichal Hocko } 717f1dd2cd1SMichal Hocko 718f1dd2cd1SMichal Hocko /* 719c246a213SMichal Hocko * Returns a default kernel memory zone for the given pfn range. 720c246a213SMichal Hocko * If no kernel zone covers this pfn range it will automatically go 721c246a213SMichal Hocko * to the ZONE_NORMAL. 722c246a213SMichal Hocko */ 723c6f03e29SMichal Hocko static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, 724c246a213SMichal Hocko unsigned long nr_pages) 725c246a213SMichal Hocko { 726c246a213SMichal Hocko struct pglist_data *pgdat = NODE_DATA(nid); 727c246a213SMichal Hocko int zid; 728c246a213SMichal Hocko 729c246a213SMichal Hocko for (zid = 0; zid <= ZONE_NORMAL; zid++) { 730c246a213SMichal Hocko struct zone *zone = &pgdat->node_zones[zid]; 731c246a213SMichal Hocko 732c246a213SMichal Hocko if (zone_intersects(zone, start_pfn, nr_pages)) 733c246a213SMichal Hocko return zone; 734c246a213SMichal Hocko } 735c246a213SMichal Hocko 736c246a213SMichal Hocko return &pgdat->node_zones[ZONE_NORMAL]; 737c246a213SMichal Hocko } 738c246a213SMichal Hocko 739c6f03e29SMichal Hocko static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, 740c6f03e29SMichal Hocko unsigned long nr_pages) 741e5e68930SMichal Hocko { 742c6f03e29SMichal Hocko struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, 743c6f03e29SMichal Hocko nr_pages); 744c6f03e29SMichal Hocko struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; 745c6f03e29SMichal Hocko bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); 746c6f03e29SMichal Hocko bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); 747e5e68930SMichal Hocko 748e5e68930SMichal Hocko /* 749c6f03e29SMichal Hocko * We inherit the existing zone in a simple case where zones do not 750c6f03e29SMichal Hocko * overlap in the given range 751e5e68930SMichal Hocko */ 752c6f03e29SMichal Hocko if (in_kernel ^ in_movable) 753c6f03e29SMichal Hocko return (in_kernel) ? kernel_zone : movable_zone; 754e5e68930SMichal Hocko 755c6f03e29SMichal Hocko /* 756c6f03e29SMichal Hocko * If the range doesn't belong to any zone or two zones overlap in the 757c6f03e29SMichal Hocko * given range then we use movable zone only if movable_node is 758c6f03e29SMichal Hocko * enabled because we always online to a kernel zone by default. 759c6f03e29SMichal Hocko */ 760c6f03e29SMichal Hocko return movable_node_enabled ? movable_zone : kernel_zone; 7619f123ab5SMichal Hocko } 7629f123ab5SMichal Hocko 763e5e68930SMichal Hocko struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, 764e5e68930SMichal Hocko unsigned long nr_pages) 765f1dd2cd1SMichal Hocko { 766c6f03e29SMichal Hocko if (online_type == MMOP_ONLINE_KERNEL) 767c6f03e29SMichal Hocko return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); 768f1dd2cd1SMichal Hocko 769c6f03e29SMichal Hocko if (online_type == MMOP_ONLINE_MOVABLE) 770c6f03e29SMichal Hocko return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; 771f1dd2cd1SMichal Hocko 772c6f03e29SMichal Hocko return default_zone_for_pfn(nid, start_pfn, nr_pages); 773e5e68930SMichal Hocko } 774e5e68930SMichal Hocko 775511c2abaSLai Jiangshan int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) 77675884fb1SKAMEZAWA Hiroyuki { 777aa47228aSCody P Schafer unsigned long flags; 7783947be19SDave Hansen unsigned long onlined_pages = 0; 7793947be19SDave Hansen struct zone *zone; 7806811378eSYasunori Goto int need_zonelists_rebuild = 0; 7817b78d335SYasunori Goto int nid; 7827b78d335SYasunori Goto int ret; 7837b78d335SYasunori Goto struct memory_notify arg; 784d0dc12e8SPavel Tatashin struct memory_block *mem; 7853947be19SDave Hansen 786381eab4aSDavid Hildenbrand mem_hotplug_begin(); 787381eab4aSDavid Hildenbrand 788d0dc12e8SPavel Tatashin /* 789d0dc12e8SPavel Tatashin * We can't use pfn_to_nid() because nid might be stored in struct page 790d0dc12e8SPavel Tatashin * which is not yet initialized. Instead, we find nid from memory block. 791d0dc12e8SPavel Tatashin */ 792d0dc12e8SPavel Tatashin mem = find_memory_block(__pfn_to_section(pfn)); 793d0dc12e8SPavel Tatashin nid = mem->nid; 79489c02e69SDavid Hildenbrand put_device(&mem->dev); 795d0dc12e8SPavel Tatashin 796f1dd2cd1SMichal Hocko /* associate pfn range with the zone */ 7973fccb74cSDavid Hildenbrand zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages); 7983fccb74cSDavid Hildenbrand move_pfn_range_to_zone(zone, pfn, nr_pages, NULL); 799511c2abaSLai Jiangshan 8007b78d335SYasunori Goto arg.start_pfn = pfn; 8017b78d335SYasunori Goto arg.nr_pages = nr_pages; 802d9713679SLai Jiangshan node_states_check_changes_online(nr_pages, zone, &arg); 8037b78d335SYasunori Goto 8047b78d335SYasunori Goto ret = memory_notify(MEM_GOING_ONLINE, &arg); 8057b78d335SYasunori Goto ret = notifier_to_errno(ret); 806e33e33b4SChen Yucong if (ret) 807e33e33b4SChen Yucong goto failed_addition; 808e33e33b4SChen Yucong 8093947be19SDave Hansen /* 8106811378eSYasunori Goto * If this zone is not populated, then it is not in zonelist. 8116811378eSYasunori Goto * This means the page allocator ignores this zone. 8126811378eSYasunori Goto * So, zonelist must be updated after online. 8136811378eSYasunori Goto */ 8146dcd73d7SWen Congyang if (!populated_zone(zone)) { 8156811378eSYasunori Goto need_zonelists_rebuild = 1; 81672675e13SMichal Hocko setup_zone_pageset(zone); 8176dcd73d7SWen Congyang } 8186811378eSYasunori Goto 819908eedc6SKAMEZAWA Hiroyuki ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, 82075884fb1SKAMEZAWA Hiroyuki online_pages_range); 821fd8a4221SGeoff Levand if (ret) { 822ca9a46f8SDavid Hildenbrand /* not a single memory resource was applicable */ 8236dcd73d7SWen Congyang if (need_zonelists_rebuild) 8246dcd73d7SWen Congyang zone_pcp_reset(zone); 825e33e33b4SChen Yucong goto failed_addition; 826fd8a4221SGeoff Levand } 827fd8a4221SGeoff Levand 8283947be19SDave Hansen zone->present_pages += onlined_pages; 829aa47228aSCody P Schafer 830aa47228aSCody P Schafer pgdat_resize_lock(zone->zone_pgdat, &flags); 831f2937be5SYasunori Goto zone->zone_pgdat->node_present_pages += onlined_pages; 832aa47228aSCody P Schafer pgdat_resize_unlock(zone->zone_pgdat, &flags); 833aa47228aSCody P Schafer 834e900a918SDan Williams shuffle_zone(zone); 835e900a918SDan Williams 836e888ca35SVlastimil Babka node_states_set_node(nid, &arg); 8371f522509SHaicheng Li if (need_zonelists_rebuild) 83872675e13SMichal Hocko build_all_zonelists(NULL); 8391f522509SHaicheng Li else 840112067f0SShaohua Li zone_pcp_update(zone); 8411f522509SHaicheng Li 8421b79acc9SKOSAKI Motohiro init_per_zone_wmark_min(); 8431b79acc9SKOSAKI Motohiro 844e888ca35SVlastimil Babka kswapd_run(nid); 845698b1b30SVlastimil Babka kcompactd_run(nid); 84661b13993SDave Hansen 8475a4d4361SKAMEZAWA Hiroyuki vm_total_pages = nr_free_pagecache_pages(); 8482f7f24ecSKent Liu 8492d1d43f6SChandra Seetharaman writeback_set_ratelimit(); 8507b78d335SYasunori Goto 8517b78d335SYasunori Goto memory_notify(MEM_ONLINE, &arg); 852381eab4aSDavid Hildenbrand mem_hotplug_done(); 85330467e0bSDavid Rientjes return 0; 854e33e33b4SChen Yucong 855e33e33b4SChen Yucong failed_addition: 856e33e33b4SChen Yucong pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", 857e33e33b4SChen Yucong (unsigned long long) pfn << PAGE_SHIFT, 858e33e33b4SChen Yucong (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); 859e33e33b4SChen Yucong memory_notify(MEM_CANCEL_ONLINE, &arg); 860381eab4aSDavid Hildenbrand mem_hotplug_done(); 861e33e33b4SChen Yucong return ret; 8623947be19SDave Hansen } 86353947027SKeith Mannthey #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ 864bc02af93SYasunori Goto 8650bd85420STang Chen static void reset_node_present_pages(pg_data_t *pgdat) 8660bd85420STang Chen { 8670bd85420STang Chen struct zone *z; 8680bd85420STang Chen 8690bd85420STang Chen for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 8700bd85420STang Chen z->present_pages = 0; 8710bd85420STang Chen 8720bd85420STang Chen pgdat->node_present_pages = 0; 8730bd85420STang Chen } 8740bd85420STang Chen 875e1319331SHidetoshi Seto /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ 876e1319331SHidetoshi Seto static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) 8779af3c2deSYasunori Goto { 8789af3c2deSYasunori Goto struct pglist_data *pgdat; 879c8e861a5SFabian Frederick unsigned long start_pfn = PFN_DOWN(start); 8809af3c2deSYasunori Goto 881a1e565aaSTang Chen pgdat = NODE_DATA(nid); 882a1e565aaSTang Chen if (!pgdat) { 8839af3c2deSYasunori Goto pgdat = arch_alloc_nodedata(nid); 8849af3c2deSYasunori Goto if (!pgdat) 8859af3c2deSYasunori Goto return NULL; 8869af3c2deSYasunori Goto 88733fce011SWei Yang pgdat->per_cpu_nodestats = 88833fce011SWei Yang alloc_percpu(struct per_cpu_nodestat); 8899af3c2deSYasunori Goto arch_refresh_nodedata(nid, pgdat); 890b0dc3a34SGu Zheng } else { 89133fce011SWei Yang int cpu; 892e716f2ebSMel Gorman /* 893e716f2ebSMel Gorman * Reset the nr_zones, order and classzone_idx before reuse. 894e716f2ebSMel Gorman * Note that kswapd will init kswapd_classzone_idx properly 895e716f2ebSMel Gorman * when it starts in the near future. 896e716f2ebSMel Gorman */ 897b0dc3a34SGu Zheng pgdat->nr_zones = 0; 89838087d9bSMel Gorman pgdat->kswapd_order = 0; 89938087d9bSMel Gorman pgdat->kswapd_classzone_idx = 0; 90033fce011SWei Yang for_each_online_cpu(cpu) { 90133fce011SWei Yang struct per_cpu_nodestat *p; 90233fce011SWei Yang 90333fce011SWei Yang p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); 90433fce011SWei Yang memset(p, 0, sizeof(*p)); 90533fce011SWei Yang } 906a1e565aaSTang Chen } 9079af3c2deSYasunori Goto 9089af3c2deSYasunori Goto /* we can use NODE_DATA(nid) from here */ 9099af3c2deSYasunori Goto 91003e85f9dSOscar Salvador pgdat->node_id = nid; 91103e85f9dSOscar Salvador pgdat->node_start_pfn = start_pfn; 91203e85f9dSOscar Salvador 9139af3c2deSYasunori Goto /* init node's zones as empty zones, we don't have any present pages.*/ 91403e85f9dSOscar Salvador free_area_init_core_hotplug(nid); 9159af3c2deSYasunori Goto 916959ecc48SKAMEZAWA Hiroyuki /* 917959ecc48SKAMEZAWA Hiroyuki * The node we allocated has no zone fallback lists. For avoiding 918959ecc48SKAMEZAWA Hiroyuki * to access not-initialized zonelist, build here. 919959ecc48SKAMEZAWA Hiroyuki */ 92072675e13SMichal Hocko build_all_zonelists(pgdat); 921959ecc48SKAMEZAWA Hiroyuki 922f784a3f1STang Chen /* 9230bd85420STang Chen * When memory is hot-added, all the memory is in offline state. So 9240bd85420STang Chen * clear all zones' present_pages because they will be updated in 9250bd85420STang Chen * online_pages() and offline_pages(). 9260bd85420STang Chen */ 92703e85f9dSOscar Salvador reset_node_managed_pages(pgdat); 9280bd85420STang Chen reset_node_present_pages(pgdat); 9290bd85420STang Chen 9309af3c2deSYasunori Goto return pgdat; 9319af3c2deSYasunori Goto } 9329af3c2deSYasunori Goto 933b9ff0360SOscar Salvador static void rollback_node_hotadd(int nid) 9349af3c2deSYasunori Goto { 935b9ff0360SOscar Salvador pg_data_t *pgdat = NODE_DATA(nid); 936b9ff0360SOscar Salvador 9379af3c2deSYasunori Goto arch_refresh_nodedata(nid, NULL); 9385830169fSReza Arbab free_percpu(pgdat->per_cpu_nodestats); 9399af3c2deSYasunori Goto arch_free_nodedata(pgdat); 9409af3c2deSYasunori Goto } 9419af3c2deSYasunori Goto 9420a547039SKAMEZAWA Hiroyuki 94301b0f197SToshi Kani /** 94401b0f197SToshi Kani * try_online_node - online a node if offlined 945e8b098fcSMike Rapoport * @nid: the node ID 946b9ff0360SOscar Salvador * @start: start addr of the node 947b9ff0360SOscar Salvador * @set_node_online: Whether we want to online the node 948cf23422bSminskey guo * called by cpu_up() to online a node without onlined memory. 949b9ff0360SOscar Salvador * 950b9ff0360SOscar Salvador * Returns: 951b9ff0360SOscar Salvador * 1 -> a new node has been allocated 952b9ff0360SOscar Salvador * 0 -> the node is already online 953b9ff0360SOscar Salvador * -ENOMEM -> the node could not be allocated 954cf23422bSminskey guo */ 955b9ff0360SOscar Salvador static int __try_online_node(int nid, u64 start, bool set_node_online) 956cf23422bSminskey guo { 957cf23422bSminskey guo pg_data_t *pgdat; 958b9ff0360SOscar Salvador int ret = 1; 959cf23422bSminskey guo 96001b0f197SToshi Kani if (node_online(nid)) 96101b0f197SToshi Kani return 0; 96201b0f197SToshi Kani 963b9ff0360SOscar Salvador pgdat = hotadd_new_pgdat(nid, start); 9647553e8f2SDavid Rientjes if (!pgdat) { 96501b0f197SToshi Kani pr_err("Cannot online node %d due to NULL pgdat\n", nid); 966cf23422bSminskey guo ret = -ENOMEM; 967cf23422bSminskey guo goto out; 968cf23422bSminskey guo } 969b9ff0360SOscar Salvador 970b9ff0360SOscar Salvador if (set_node_online) { 971cf23422bSminskey guo node_set_online(nid); 972cf23422bSminskey guo ret = register_one_node(nid); 973cf23422bSminskey guo BUG_ON(ret); 974b9ff0360SOscar Salvador } 975cf23422bSminskey guo out: 976b9ff0360SOscar Salvador return ret; 977b9ff0360SOscar Salvador } 978b9ff0360SOscar Salvador 979b9ff0360SOscar Salvador /* 980b9ff0360SOscar Salvador * Users of this function always want to online/register the node 981b9ff0360SOscar Salvador */ 982b9ff0360SOscar Salvador int try_online_node(int nid) 983b9ff0360SOscar Salvador { 984b9ff0360SOscar Salvador int ret; 985b9ff0360SOscar Salvador 986b9ff0360SOscar Salvador mem_hotplug_begin(); 987b9ff0360SOscar Salvador ret = __try_online_node(nid, 0, true); 988bfc8c901SVladimir Davydov mem_hotplug_done(); 989cf23422bSminskey guo return ret; 990cf23422bSminskey guo } 991cf23422bSminskey guo 99227356f54SToshi Kani static int check_hotplug_memory_range(u64 start, u64 size) 99327356f54SToshi Kani { 994ba325585SPavel Tatashin /* memory range must be block size aligned */ 995cec3ebd0SDavid Hildenbrand if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || 996cec3ebd0SDavid Hildenbrand !IS_ALIGNED(size, memory_block_size_bytes())) { 997ba325585SPavel Tatashin pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", 998cec3ebd0SDavid Hildenbrand memory_block_size_bytes(), start, size); 99927356f54SToshi Kani return -EINVAL; 100027356f54SToshi Kani } 100127356f54SToshi Kani 100227356f54SToshi Kani return 0; 100327356f54SToshi Kani } 100427356f54SToshi Kani 100531bc3858SVitaly Kuznetsov static int online_memory_block(struct memory_block *mem, void *arg) 100631bc3858SVitaly Kuznetsov { 1007dc18d706SNathan Fontenot return device_online(&mem->dev); 100831bc3858SVitaly Kuznetsov } 100931bc3858SVitaly Kuznetsov 10108df1d0e4SDavid Hildenbrand /* 10118df1d0e4SDavid Hildenbrand * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 10128df1d0e4SDavid Hildenbrand * and online/offline operations (triggered e.g. by sysfs). 10138df1d0e4SDavid Hildenbrand * 10148df1d0e4SDavid Hildenbrand * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG 10158df1d0e4SDavid Hildenbrand */ 1016f29d8e9cSDavid Hildenbrand int __ref add_memory_resource(int nid, struct resource *res) 1017bc02af93SYasunori Goto { 101805f800a0SDavid Hildenbrand struct mhp_restrictions restrictions = {}; 101962cedb9fSDavid Vrabel u64 start, size; 1020b9ff0360SOscar Salvador bool new_node = false; 1021bc02af93SYasunori Goto int ret; 1022bc02af93SYasunori Goto 102362cedb9fSDavid Vrabel start = res->start; 102462cedb9fSDavid Vrabel size = resource_size(res); 102562cedb9fSDavid Vrabel 102627356f54SToshi Kani ret = check_hotplug_memory_range(start, size); 102727356f54SToshi Kani if (ret) 102827356f54SToshi Kani return ret; 102927356f54SToshi Kani 1030bfc8c901SVladimir Davydov mem_hotplug_begin(); 1031ac13c462SNathan Zimmer 10327f36e3e5STang Chen /* 10337f36e3e5STang Chen * Add new range to memblock so that when hotadd_new_pgdat() is called 10347f36e3e5STang Chen * to allocate new pgdat, get_pfn_range_for_nid() will be able to find 10357f36e3e5STang Chen * this new range and calculate total pages correctly. The range will 10367f36e3e5STang Chen * be removed at hot-remove time. 10377f36e3e5STang Chen */ 10387f36e3e5STang Chen memblock_add_node(start, size, nid); 10397f36e3e5STang Chen 1040b9ff0360SOscar Salvador ret = __try_online_node(nid, start, false); 1041b9ff0360SOscar Salvador if (ret < 0) 104241b9e2d7SWen Congyang goto error; 1043b9ff0360SOscar Salvador new_node = ret; 10449af3c2deSYasunori Goto 1045bc02af93SYasunori Goto /* call arch's memory hotadd */ 1046940519f0SMichal Hocko ret = arch_add_memory(nid, start, size, &restrictions); 10479af3c2deSYasunori Goto if (ret < 0) 10489af3c2deSYasunori Goto goto error; 10499af3c2deSYasunori Goto 1050db051a0dSDavid Hildenbrand /* create memory block devices after memory was added */ 1051db051a0dSDavid Hildenbrand ret = create_memory_block_devices(start, size); 1052db051a0dSDavid Hildenbrand if (ret) { 1053db051a0dSDavid Hildenbrand arch_remove_memory(nid, start, size, NULL); 1054db051a0dSDavid Hildenbrand goto error; 1055db051a0dSDavid Hildenbrand } 1056db051a0dSDavid Hildenbrand 1057a1e565aaSTang Chen if (new_node) { 1058d5b6f6a3SOscar Salvador /* If sysfs file of new node can't be created, cpu on the node 10590fc44159SYasunori Goto * can't be hot-added. There is no rollback way now. 10600fc44159SYasunori Goto * So, check by BUG_ON() to catch it reluctantly.. 1061d5b6f6a3SOscar Salvador * We online node here. We can't roll back from here. 10620fc44159SYasunori Goto */ 1063d5b6f6a3SOscar Salvador node_set_online(nid); 1064d5b6f6a3SOscar Salvador ret = __register_one_node(nid); 10650fc44159SYasunori Goto BUG_ON(ret); 10660fc44159SYasunori Goto } 10670fc44159SYasunori Goto 1068d5b6f6a3SOscar Salvador /* link memory sections under this node.*/ 10694fbce633SOscar Salvador ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1)); 1070d5b6f6a3SOscar Salvador BUG_ON(ret); 1071d5b6f6a3SOscar Salvador 1072d96ae530Sakpm@linux-foundation.org /* create new memmap entry */ 1073d96ae530Sakpm@linux-foundation.org firmware_map_add_hotplug(start, start + size, "System RAM"); 1074d96ae530Sakpm@linux-foundation.org 1075381eab4aSDavid Hildenbrand /* device_online() will take the lock when calling online_pages() */ 1076381eab4aSDavid Hildenbrand mem_hotplug_done(); 1077381eab4aSDavid Hildenbrand 107831bc3858SVitaly Kuznetsov /* online pages if requested */ 1079f29d8e9cSDavid Hildenbrand if (memhp_auto_online) 1080fbcf73ceSDavid Hildenbrand walk_memory_blocks(start, size, NULL, online_memory_block); 108131bc3858SVitaly Kuznetsov 1082381eab4aSDavid Hildenbrand return ret; 10839af3c2deSYasunori Goto error: 10849af3c2deSYasunori Goto /* rollback pgdat allocation and others */ 1085b9ff0360SOscar Salvador if (new_node) 1086b9ff0360SOscar Salvador rollback_node_hotadd(nid); 10877f36e3e5STang Chen memblock_remove(start, size); 1088bfc8c901SVladimir Davydov mem_hotplug_done(); 1089bc02af93SYasunori Goto return ret; 1090bc02af93SYasunori Goto } 109162cedb9fSDavid Vrabel 10928df1d0e4SDavid Hildenbrand /* requires device_hotplug_lock, see add_memory_resource() */ 10938df1d0e4SDavid Hildenbrand int __ref __add_memory(int nid, u64 start, u64 size) 109462cedb9fSDavid Vrabel { 109562cedb9fSDavid Vrabel struct resource *res; 109662cedb9fSDavid Vrabel int ret; 109762cedb9fSDavid Vrabel 109862cedb9fSDavid Vrabel res = register_memory_resource(start, size); 10996f754ba4SVitaly Kuznetsov if (IS_ERR(res)) 11006f754ba4SVitaly Kuznetsov return PTR_ERR(res); 110162cedb9fSDavid Vrabel 1102f29d8e9cSDavid Hildenbrand ret = add_memory_resource(nid, res); 110362cedb9fSDavid Vrabel if (ret < 0) 110462cedb9fSDavid Vrabel release_memory_resource(res); 110562cedb9fSDavid Vrabel return ret; 110662cedb9fSDavid Vrabel } 11078df1d0e4SDavid Hildenbrand 11088df1d0e4SDavid Hildenbrand int add_memory(int nid, u64 start, u64 size) 11098df1d0e4SDavid Hildenbrand { 11108df1d0e4SDavid Hildenbrand int rc; 11118df1d0e4SDavid Hildenbrand 11128df1d0e4SDavid Hildenbrand lock_device_hotplug(); 11138df1d0e4SDavid Hildenbrand rc = __add_memory(nid, start, size); 11148df1d0e4SDavid Hildenbrand unlock_device_hotplug(); 11158df1d0e4SDavid Hildenbrand 11168df1d0e4SDavid Hildenbrand return rc; 11178df1d0e4SDavid Hildenbrand } 1118bc02af93SYasunori Goto EXPORT_SYMBOL_GPL(add_memory); 11190c0e6195SKAMEZAWA Hiroyuki 11200c0e6195SKAMEZAWA Hiroyuki #ifdef CONFIG_MEMORY_HOTREMOVE 11210c0e6195SKAMEZAWA Hiroyuki /* 11225c755e9fSBadari Pulavarty * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy 11235c755e9fSBadari Pulavarty * set and the size of the free page is given by page_order(). Using this, 11245c755e9fSBadari Pulavarty * the function determines if the pageblock contains only free pages. 11255c755e9fSBadari Pulavarty * Due to buddy contraints, a free page at least the size of a pageblock will 11265c755e9fSBadari Pulavarty * be located at the start of the pageblock 11275c755e9fSBadari Pulavarty */ 11285c755e9fSBadari Pulavarty static inline int pageblock_free(struct page *page) 11295c755e9fSBadari Pulavarty { 11305c755e9fSBadari Pulavarty return PageBuddy(page) && page_order(page) >= pageblock_order; 11315c755e9fSBadari Pulavarty } 11325c755e9fSBadari Pulavarty 1133891cb2a7SMichal Hocko /* Return the pfn of the start of the next active pageblock after a given pfn */ 1134891cb2a7SMichal Hocko static unsigned long next_active_pageblock(unsigned long pfn) 11355c755e9fSBadari Pulavarty { 1136891cb2a7SMichal Hocko struct page *page = pfn_to_page(pfn); 1137891cb2a7SMichal Hocko 11385c755e9fSBadari Pulavarty /* Ensure the starting page is pageblock-aligned */ 1139891cb2a7SMichal Hocko BUG_ON(pfn & (pageblock_nr_pages - 1)); 11405c755e9fSBadari Pulavarty 11415c755e9fSBadari Pulavarty /* If the entire pageblock is free, move to the end of free page */ 11420dcc48c1SKAMEZAWA Hiroyuki if (pageblock_free(page)) { 11430dcc48c1SKAMEZAWA Hiroyuki int order; 11440dcc48c1SKAMEZAWA Hiroyuki /* be careful. we don't have locks, page_order can be changed.*/ 11450dcc48c1SKAMEZAWA Hiroyuki order = page_order(page); 11460dcc48c1SKAMEZAWA Hiroyuki if ((order < MAX_ORDER) && (order >= pageblock_order)) 1147891cb2a7SMichal Hocko return pfn + (1 << order); 11480dcc48c1SKAMEZAWA Hiroyuki } 11495c755e9fSBadari Pulavarty 1150891cb2a7SMichal Hocko return pfn + pageblock_nr_pages; 11515c755e9fSBadari Pulavarty } 11525c755e9fSBadari Pulavarty 1153891cb2a7SMichal Hocko static bool is_pageblock_removable_nolock(unsigned long pfn) 1154fb52bbaeSMathieu Malaterre { 1155891cb2a7SMichal Hocko struct page *page = pfn_to_page(pfn); 1156fb52bbaeSMathieu Malaterre struct zone *zone; 1157fb52bbaeSMathieu Malaterre 1158fb52bbaeSMathieu Malaterre /* 1159fb52bbaeSMathieu Malaterre * We have to be careful here because we are iterating over memory 1160fb52bbaeSMathieu Malaterre * sections which are not zone aware so we might end up outside of 1161fb52bbaeSMathieu Malaterre * the zone but still within the section. 1162fb52bbaeSMathieu Malaterre * We have to take care about the node as well. If the node is offline 1163fb52bbaeSMathieu Malaterre * its NODE_DATA will be NULL - see page_zone. 1164fb52bbaeSMathieu Malaterre */ 1165fb52bbaeSMathieu Malaterre if (!node_online(page_to_nid(page))) 1166fb52bbaeSMathieu Malaterre return false; 1167fb52bbaeSMathieu Malaterre 1168fb52bbaeSMathieu Malaterre zone = page_zone(page); 1169fb52bbaeSMathieu Malaterre pfn = page_to_pfn(page); 1170fb52bbaeSMathieu Malaterre if (!zone_spans_pfn(zone, pfn)) 1171fb52bbaeSMathieu Malaterre return false; 1172fb52bbaeSMathieu Malaterre 1173d381c547SMichal Hocko return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON); 1174fb52bbaeSMathieu Malaterre } 1175fb52bbaeSMathieu Malaterre 11765c755e9fSBadari Pulavarty /* Checks if this range of memory is likely to be hot-removable. */ 1177c98940f6SYaowei Bai bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) 11785c755e9fSBadari Pulavarty { 1179891cb2a7SMichal Hocko unsigned long end_pfn, pfn; 1180891cb2a7SMichal Hocko 1181891cb2a7SMichal Hocko end_pfn = min(start_pfn + nr_pages, 1182891cb2a7SMichal Hocko zone_end_pfn(page_zone(pfn_to_page(start_pfn)))); 11835c755e9fSBadari Pulavarty 11845c755e9fSBadari Pulavarty /* Check the starting page of each pageblock within the range */ 1185891cb2a7SMichal Hocko for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) { 1186891cb2a7SMichal Hocko if (!is_pageblock_removable_nolock(pfn)) 1187c98940f6SYaowei Bai return false; 118849ac8255SKAMEZAWA Hiroyuki cond_resched(); 11895c755e9fSBadari Pulavarty } 11905c755e9fSBadari Pulavarty 11915c755e9fSBadari Pulavarty /* All pageblocks in the memory block are likely to be hot-removable */ 1192c98940f6SYaowei Bai return true; 11935c755e9fSBadari Pulavarty } 11945c755e9fSBadari Pulavarty 11955c755e9fSBadari Pulavarty /* 1196deb88a2aSToshi Kani * Confirm all pages in a range [start, end) belong to the same zone. 1197a96dfddbSToshi Kani * When true, return its valid [start, end). 11980c0e6195SKAMEZAWA Hiroyuki */ 1199a96dfddbSToshi Kani int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, 1200a96dfddbSToshi Kani unsigned long *valid_start, unsigned long *valid_end) 12010c0e6195SKAMEZAWA Hiroyuki { 12025f0f2887SAndrew Banman unsigned long pfn, sec_end_pfn; 1203a96dfddbSToshi Kani unsigned long start, end; 12040c0e6195SKAMEZAWA Hiroyuki struct zone *zone = NULL; 12050c0e6195SKAMEZAWA Hiroyuki struct page *page; 12060c0e6195SKAMEZAWA Hiroyuki int i; 1207deb88a2aSToshi Kani for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); 12080c0e6195SKAMEZAWA Hiroyuki pfn < end_pfn; 1209deb88a2aSToshi Kani pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { 12105f0f2887SAndrew Banman /* Make sure the memory section is present first */ 12115f0f2887SAndrew Banman if (!present_section_nr(pfn_to_section_nr(pfn))) 12125f0f2887SAndrew Banman continue; 12135f0f2887SAndrew Banman for (; pfn < sec_end_pfn && pfn < end_pfn; 12140c0e6195SKAMEZAWA Hiroyuki pfn += MAX_ORDER_NR_PAGES) { 12150c0e6195SKAMEZAWA Hiroyuki i = 0; 12160c0e6195SKAMEZAWA Hiroyuki /* This is just a CONFIG_HOLES_IN_ZONE check.*/ 12175f0f2887SAndrew Banman while ((i < MAX_ORDER_NR_PAGES) && 12185f0f2887SAndrew Banman !pfn_valid_within(pfn + i)) 12190c0e6195SKAMEZAWA Hiroyuki i++; 1220d6d8c8a4Szhong jiang if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) 12210c0e6195SKAMEZAWA Hiroyuki continue; 122224feb47cSMikhail Zaslonko /* Check if we got outside of the zone */ 122324feb47cSMikhail Zaslonko if (zone && !zone_spans_pfn(zone, pfn + i)) 122424feb47cSMikhail Zaslonko return 0; 12250c0e6195SKAMEZAWA Hiroyuki page = pfn_to_page(pfn + i); 12260c0e6195SKAMEZAWA Hiroyuki if (zone && page_zone(page) != zone) 12270c0e6195SKAMEZAWA Hiroyuki return 0; 1228a96dfddbSToshi Kani if (!zone) 1229a96dfddbSToshi Kani start = pfn + i; 12300c0e6195SKAMEZAWA Hiroyuki zone = page_zone(page); 1231a96dfddbSToshi Kani end = pfn + MAX_ORDER_NR_PAGES; 12320c0e6195SKAMEZAWA Hiroyuki } 12335f0f2887SAndrew Banman } 1234deb88a2aSToshi Kani 1235a96dfddbSToshi Kani if (zone) { 1236a96dfddbSToshi Kani *valid_start = start; 1237d6d8c8a4Szhong jiang *valid_end = min(end, end_pfn); 12380c0e6195SKAMEZAWA Hiroyuki return 1; 1239a96dfddbSToshi Kani } else { 1240deb88a2aSToshi Kani return 0; 12410c0e6195SKAMEZAWA Hiroyuki } 1242a96dfddbSToshi Kani } 12430c0e6195SKAMEZAWA Hiroyuki 12440c0e6195SKAMEZAWA Hiroyuki /* 12450efadf48SYisheng Xie * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, 12460efadf48SYisheng Xie * non-lru movable pages and hugepages). We scan pfn because it's much 12470efadf48SYisheng Xie * easier than scanning over linked list. This function returns the pfn 12480efadf48SYisheng Xie * of the first found movable page if it's found, otherwise 0. 12490c0e6195SKAMEZAWA Hiroyuki */ 1250c8721bbbSNaoya Horiguchi static unsigned long scan_movable_pages(unsigned long start, unsigned long end) 12510c0e6195SKAMEZAWA Hiroyuki { 12520c0e6195SKAMEZAWA Hiroyuki unsigned long pfn; 1253eeb0efd0SOscar Salvador 12540c0e6195SKAMEZAWA Hiroyuki for (pfn = start; pfn < end; pfn++) { 1255eeb0efd0SOscar Salvador struct page *page, *head; 1256eeb0efd0SOscar Salvador unsigned long skip; 1257eeb0efd0SOscar Salvador 1258eeb0efd0SOscar Salvador if (!pfn_valid(pfn)) 1259eeb0efd0SOscar Salvador continue; 12600c0e6195SKAMEZAWA Hiroyuki page = pfn_to_page(pfn); 12610c0e6195SKAMEZAWA Hiroyuki if (PageLRU(page)) 12620c0e6195SKAMEZAWA Hiroyuki return pfn; 12630efadf48SYisheng Xie if (__PageMovable(page)) 12640efadf48SYisheng Xie return pfn; 1265eeb0efd0SOscar Salvador 1266eeb0efd0SOscar Salvador if (!PageHuge(page)) 1267eeb0efd0SOscar Salvador continue; 1268eeb0efd0SOscar Salvador head = compound_head(page); 126939186cbeSOscar Salvador if (page_huge_active(head)) 1270c8721bbbSNaoya Horiguchi return pfn; 1271d8c6546bSMatthew Wilcox (Oracle) skip = compound_nr(head) - (page - head); 1272eeb0efd0SOscar Salvador pfn += skip - 1; 12730c0e6195SKAMEZAWA Hiroyuki } 12740c0e6195SKAMEZAWA Hiroyuki return 0; 12750c0e6195SKAMEZAWA Hiroyuki } 12760c0e6195SKAMEZAWA Hiroyuki 1277666feb21SMichal Hocko static struct page *new_node_page(struct page *page, unsigned long private) 1278394e31d2SXishi Qiu { 1279394e31d2SXishi Qiu int nid = page_to_nid(page); 1280231e97e2SLi Zhong nodemask_t nmask = node_states[N_MEMORY]; 12817f252f27SMichal Hocko 12827f252f27SMichal Hocko /* 12837f252f27SMichal Hocko * try to allocate from a different node but reuse this node if there 12847f252f27SMichal Hocko * are no other online nodes to be used (e.g. we are offlining a part 12857f252f27SMichal Hocko * of the only existing node) 12867f252f27SMichal Hocko */ 12877f252f27SMichal Hocko node_clear(nid, nmask); 12887f252f27SMichal Hocko if (nodes_empty(nmask)) 12897f252f27SMichal Hocko node_set(nid, nmask); 1290394e31d2SXishi Qiu 12918b913238SMichal Hocko return new_page_nodemask(page, nid, &nmask); 1292394e31d2SXishi Qiu } 1293394e31d2SXishi Qiu 12940c0e6195SKAMEZAWA Hiroyuki static int 12950c0e6195SKAMEZAWA Hiroyuki do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) 12960c0e6195SKAMEZAWA Hiroyuki { 12970c0e6195SKAMEZAWA Hiroyuki unsigned long pfn; 12980c0e6195SKAMEZAWA Hiroyuki struct page *page; 12990c0e6195SKAMEZAWA Hiroyuki int ret = 0; 13000c0e6195SKAMEZAWA Hiroyuki LIST_HEAD(source); 13010c0e6195SKAMEZAWA Hiroyuki 1302a85009c3SMichal Hocko for (pfn = start_pfn; pfn < end_pfn; pfn++) { 13030c0e6195SKAMEZAWA Hiroyuki if (!pfn_valid(pfn)) 13040c0e6195SKAMEZAWA Hiroyuki continue; 13050c0e6195SKAMEZAWA Hiroyuki page = pfn_to_page(pfn); 1306c8721bbbSNaoya Horiguchi 1307c8721bbbSNaoya Horiguchi if (PageHuge(page)) { 1308c8721bbbSNaoya Horiguchi struct page *head = compound_head(page); 1309d8c6546bSMatthew Wilcox (Oracle) pfn = page_to_pfn(head) + compound_nr(head) - 1; 1310daf3538aSOscar Salvador isolate_huge_page(head, &source); 1311c8721bbbSNaoya Horiguchi continue; 131294723aafSMichal Hocko } else if (PageTransHuge(page)) 13138135d892SNaoya Horiguchi pfn = page_to_pfn(compound_head(page)) 13148135d892SNaoya Horiguchi + hpage_nr_pages(page) - 1; 1315c8721bbbSNaoya Horiguchi 1316b15c8726SMichal Hocko /* 1317b15c8726SMichal Hocko * HWPoison pages have elevated reference counts so the migration would 1318b15c8726SMichal Hocko * fail on them. It also doesn't make any sense to migrate them in the 1319b15c8726SMichal Hocko * first place. Still try to unmap such a page in case it is still mapped 1320b15c8726SMichal Hocko * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep 1321b15c8726SMichal Hocko * the unmap as the catch all safety net). 1322b15c8726SMichal Hocko */ 1323b15c8726SMichal Hocko if (PageHWPoison(page)) { 1324b15c8726SMichal Hocko if (WARN_ON(PageLRU(page))) 1325b15c8726SMichal Hocko isolate_lru_page(page); 1326b15c8726SMichal Hocko if (page_mapped(page)) 1327b15c8726SMichal Hocko try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); 1328b15c8726SMichal Hocko continue; 1329b15c8726SMichal Hocko } 1330b15c8726SMichal Hocko 1331700c2a46SKonstantin Khlebnikov if (!get_page_unless_zero(page)) 13320c0e6195SKAMEZAWA Hiroyuki continue; 13330c0e6195SKAMEZAWA Hiroyuki /* 13340efadf48SYisheng Xie * We can skip free pages. And we can deal with pages on 13350efadf48SYisheng Xie * LRU and non-lru movable pages. 13360c0e6195SKAMEZAWA Hiroyuki */ 13370efadf48SYisheng Xie if (PageLRU(page)) 133862695a84SNick Piggin ret = isolate_lru_page(page); 13390efadf48SYisheng Xie else 13400efadf48SYisheng Xie ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); 13410c0e6195SKAMEZAWA Hiroyuki if (!ret) { /* Success */ 134262695a84SNick Piggin list_add_tail(&page->lru, &source); 13430efadf48SYisheng Xie if (!__PageMovable(page)) 1344599d0c95SMel Gorman inc_node_page_state(page, NR_ISOLATED_ANON + 13456d9c285aSKOSAKI Motohiro page_is_file_cache(page)); 13466d9c285aSKOSAKI Motohiro 13470c0e6195SKAMEZAWA Hiroyuki } else { 13482932c8b0SMichal Hocko pr_warn("failed to isolate pfn %lx\n", pfn); 13490efadf48SYisheng Xie dump_page(page, "isolation failed"); 13501723058eSOscar Salvador } 1351700c2a46SKonstantin Khlebnikov put_page(page); 13520c0e6195SKAMEZAWA Hiroyuki } 1353f3ab2636SBob Liu if (!list_empty(&source)) { 1354394e31d2SXishi Qiu /* Allocate a new page from the nearest neighbor node */ 1355394e31d2SXishi Qiu ret = migrate_pages(&source, new_node_page, NULL, 0, 13569c620e2bSHugh Dickins MIGRATE_SYNC, MR_MEMORY_HOTPLUG); 13572932c8b0SMichal Hocko if (ret) { 13582932c8b0SMichal Hocko list_for_each_entry(page, &source, lru) { 13592932c8b0SMichal Hocko pr_warn("migrating pfn %lx failed ret:%d ", 13602932c8b0SMichal Hocko page_to_pfn(page), ret); 13612932c8b0SMichal Hocko dump_page(page, "migration failure"); 13622932c8b0SMichal Hocko } 1363c8721bbbSNaoya Horiguchi putback_movable_pages(&source); 1364f3ab2636SBob Liu } 13652932c8b0SMichal Hocko } 13661723058eSOscar Salvador 13670c0e6195SKAMEZAWA Hiroyuki return ret; 13680c0e6195SKAMEZAWA Hiroyuki } 13690c0e6195SKAMEZAWA Hiroyuki 13700c0e6195SKAMEZAWA Hiroyuki /* 13710c0e6195SKAMEZAWA Hiroyuki * remove from free_area[] and mark all as Reserved. 13720c0e6195SKAMEZAWA Hiroyuki */ 13730c0e6195SKAMEZAWA Hiroyuki static int 13740c0e6195SKAMEZAWA Hiroyuki offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, 13750c0e6195SKAMEZAWA Hiroyuki void *data) 13760c0e6195SKAMEZAWA Hiroyuki { 13775557c766SMichal Hocko unsigned long *offlined_pages = (unsigned long *)data; 13780c0e6195SKAMEZAWA Hiroyuki 13795557c766SMichal Hocko *offlined_pages += __offline_isolated_pages(start, start + nr_pages); 13805557c766SMichal Hocko return 0; 13810c0e6195SKAMEZAWA Hiroyuki } 13820c0e6195SKAMEZAWA Hiroyuki 13830c0e6195SKAMEZAWA Hiroyuki /* 13840c0e6195SKAMEZAWA Hiroyuki * Check all pages in range, recoreded as memory resource, are isolated. 13850c0e6195SKAMEZAWA Hiroyuki */ 13860c0e6195SKAMEZAWA Hiroyuki static int 13870c0e6195SKAMEZAWA Hiroyuki check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, 13880c0e6195SKAMEZAWA Hiroyuki void *data) 13890c0e6195SKAMEZAWA Hiroyuki { 13905557c766SMichal Hocko return test_pages_isolated(start_pfn, start_pfn + nr_pages, true); 13910c0e6195SKAMEZAWA Hiroyuki } 13920c0e6195SKAMEZAWA Hiroyuki 1393c5320926STang Chen static int __init cmdline_parse_movable_node(char *p) 1394c5320926STang Chen { 13954932381eSMichal Hocko #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 139655ac590cSTang Chen movable_node_enabled = true; 13974932381eSMichal Hocko #else 13984932381eSMichal Hocko pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); 13994932381eSMichal Hocko #endif 1400c5320926STang Chen return 0; 1401c5320926STang Chen } 1402c5320926STang Chen early_param("movable_node", cmdline_parse_movable_node); 1403c5320926STang Chen 1404d9713679SLai Jiangshan /* check which state of node_states will be changed when offline memory */ 1405d9713679SLai Jiangshan static void node_states_check_changes_offline(unsigned long nr_pages, 1406d9713679SLai Jiangshan struct zone *zone, struct memory_notify *arg) 1407d9713679SLai Jiangshan { 1408d9713679SLai Jiangshan struct pglist_data *pgdat = zone->zone_pgdat; 1409d9713679SLai Jiangshan unsigned long present_pages = 0; 141086b27beaSOscar Salvador enum zone_type zt; 1411d9713679SLai Jiangshan 141298fa15f3SAnshuman Khandual arg->status_change_nid = NUMA_NO_NODE; 141398fa15f3SAnshuman Khandual arg->status_change_nid_normal = NUMA_NO_NODE; 141498fa15f3SAnshuman Khandual arg->status_change_nid_high = NUMA_NO_NODE; 141586b27beaSOscar Salvador 141686b27beaSOscar Salvador /* 141786b27beaSOscar Salvador * Check whether node_states[N_NORMAL_MEMORY] will be changed. 141886b27beaSOscar Salvador * If the memory to be offline is within the range 141986b27beaSOscar Salvador * [0..ZONE_NORMAL], and it is the last present memory there, 142086b27beaSOscar Salvador * the zones in that range will become empty after the offlining, 142186b27beaSOscar Salvador * thus we can determine that we need to clear the node from 142286b27beaSOscar Salvador * node_states[N_NORMAL_MEMORY]. 142386b27beaSOscar Salvador */ 142486b27beaSOscar Salvador for (zt = 0; zt <= ZONE_NORMAL; zt++) 142586b27beaSOscar Salvador present_pages += pgdat->node_zones[zt].present_pages; 142686b27beaSOscar Salvador if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) 142786b27beaSOscar Salvador arg->status_change_nid_normal = zone_to_nid(zone); 1428d9713679SLai Jiangshan 14296715ddf9SLai Jiangshan #ifdef CONFIG_HIGHMEM 14306715ddf9SLai Jiangshan /* 143186b27beaSOscar Salvador * node_states[N_HIGH_MEMORY] contains nodes which 143286b27beaSOscar Salvador * have normal memory or high memory. 143386b27beaSOscar Salvador * Here we add the present_pages belonging to ZONE_HIGHMEM. 143486b27beaSOscar Salvador * If the zone is within the range of [0..ZONE_HIGHMEM), and 143586b27beaSOscar Salvador * we determine that the zones in that range become empty, 143686b27beaSOscar Salvador * we need to clear the node for N_HIGH_MEMORY. 14376715ddf9SLai Jiangshan */ 143886b27beaSOscar Salvador present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; 143986b27beaSOscar Salvador if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) 14406715ddf9SLai Jiangshan arg->status_change_nid_high = zone_to_nid(zone); 14416715ddf9SLai Jiangshan #endif 14426715ddf9SLai Jiangshan 1443d9713679SLai Jiangshan /* 144486b27beaSOscar Salvador * We have accounted the pages from [0..ZONE_NORMAL), and 144586b27beaSOscar Salvador * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM 144686b27beaSOscar Salvador * as well. 144786b27beaSOscar Salvador * Here we count the possible pages from ZONE_MOVABLE. 144886b27beaSOscar Salvador * If after having accounted all the pages, we see that the nr_pages 144986b27beaSOscar Salvador * to be offlined is over or equal to the accounted pages, 145086b27beaSOscar Salvador * we know that the node will become empty, and so, we can clear 145186b27beaSOscar Salvador * it for N_MEMORY as well. 1452d9713679SLai Jiangshan */ 145386b27beaSOscar Salvador present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; 1454d9713679SLai Jiangshan 1455d9713679SLai Jiangshan if (nr_pages >= present_pages) 1456d9713679SLai Jiangshan arg->status_change_nid = zone_to_nid(zone); 1457d9713679SLai Jiangshan } 1458d9713679SLai Jiangshan 1459d9713679SLai Jiangshan static void node_states_clear_node(int node, struct memory_notify *arg) 1460d9713679SLai Jiangshan { 1461d9713679SLai Jiangshan if (arg->status_change_nid_normal >= 0) 1462d9713679SLai Jiangshan node_clear_state(node, N_NORMAL_MEMORY); 1463d9713679SLai Jiangshan 1464cf01f6f5SOscar Salvador if (arg->status_change_nid_high >= 0) 1465d9713679SLai Jiangshan node_clear_state(node, N_HIGH_MEMORY); 14666715ddf9SLai Jiangshan 1467cf01f6f5SOscar Salvador if (arg->status_change_nid >= 0) 14686715ddf9SLai Jiangshan node_clear_state(node, N_MEMORY); 1469d9713679SLai Jiangshan } 1470d9713679SLai Jiangshan 1471a16cee10SWen Congyang static int __ref __offline_pages(unsigned long start_pfn, 1472ecde0f3eSMichal Hocko unsigned long end_pfn) 14730c0e6195SKAMEZAWA Hiroyuki { 1474ecde0f3eSMichal Hocko unsigned long pfn, nr_pages; 14755557c766SMichal Hocko unsigned long offlined_pages = 0; 14769b7ea46aSQian Cai int ret, node, nr_isolate_pageblock; 1477d702909fSCody P Schafer unsigned long flags; 1478a96dfddbSToshi Kani unsigned long valid_start, valid_end; 14790c0e6195SKAMEZAWA Hiroyuki struct zone *zone; 14807b78d335SYasunori Goto struct memory_notify arg; 148179605093SMichal Hocko char *reason; 14820c0e6195SKAMEZAWA Hiroyuki 1483381eab4aSDavid Hildenbrand mem_hotplug_begin(); 1484381eab4aSDavid Hildenbrand 14850c0e6195SKAMEZAWA Hiroyuki /* This makes hotplug much easier...and readable. 14860c0e6195SKAMEZAWA Hiroyuki we assume this for now. .*/ 1487381eab4aSDavid Hildenbrand if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, 1488381eab4aSDavid Hildenbrand &valid_end)) { 148979605093SMichal Hocko ret = -EINVAL; 149079605093SMichal Hocko reason = "multizone range"; 149179605093SMichal Hocko goto failed_removal; 1492381eab4aSDavid Hildenbrand } 14937b78d335SYasunori Goto 1494a96dfddbSToshi Kani zone = page_zone(pfn_to_page(valid_start)); 14957b78d335SYasunori Goto node = zone_to_nid(zone); 14967b78d335SYasunori Goto nr_pages = end_pfn - start_pfn; 14977b78d335SYasunori Goto 14980c0e6195SKAMEZAWA Hiroyuki /* set above range as isolated */ 1499b023f468SWen Congyang ret = start_isolate_page_range(start_pfn, end_pfn, 1500d381c547SMichal Hocko MIGRATE_MOVABLE, 1501d381c547SMichal Hocko SKIP_HWPOISON | REPORT_FAILURE); 15029b7ea46aSQian Cai if (ret < 0) { 150379605093SMichal Hocko reason = "failure to isolate range"; 150479605093SMichal Hocko goto failed_removal; 1505381eab4aSDavid Hildenbrand } 15069b7ea46aSQian Cai nr_isolate_pageblock = ret; 15077b78d335SYasunori Goto 15087b78d335SYasunori Goto arg.start_pfn = start_pfn; 15097b78d335SYasunori Goto arg.nr_pages = nr_pages; 1510d9713679SLai Jiangshan node_states_check_changes_offline(nr_pages, zone, &arg); 15117b78d335SYasunori Goto 15127b78d335SYasunori Goto ret = memory_notify(MEM_GOING_OFFLINE, &arg); 15137b78d335SYasunori Goto ret = notifier_to_errno(ret); 151479605093SMichal Hocko if (ret) { 151579605093SMichal Hocko reason = "notifier failure"; 151679605093SMichal Hocko goto failed_removal_isolated; 151779605093SMichal Hocko } 15187b78d335SYasunori Goto 1519bb8965bdSMichal Hocko do { 1520bb8965bdSMichal Hocko for (pfn = start_pfn; pfn;) { 152179605093SMichal Hocko if (signal_pending(current)) { 1522bb8965bdSMichal Hocko ret = -EINTR; 152379605093SMichal Hocko reason = "signal backoff"; 152479605093SMichal Hocko goto failed_removal_isolated; 152579605093SMichal Hocko } 152672b39cfcSMichal Hocko 15270c0e6195SKAMEZAWA Hiroyuki cond_resched(); 15289852a721SMichal Hocko lru_add_drain_all(); 15290c0e6195SKAMEZAWA Hiroyuki 1530bb8965bdSMichal Hocko pfn = scan_movable_pages(pfn, end_pfn); 1531bb8965bdSMichal Hocko if (pfn) { 1532bb8965bdSMichal Hocko /* 1533bb8965bdSMichal Hocko * TODO: fatal migration failures should bail 1534bb8965bdSMichal Hocko * out 1535bb8965bdSMichal Hocko */ 1536bb8965bdSMichal Hocko do_migrate_range(pfn, end_pfn); 1537bb8965bdSMichal Hocko } 15380c0e6195SKAMEZAWA Hiroyuki } 153972b39cfcSMichal Hocko 1540c8721bbbSNaoya Horiguchi /* 1541bb8965bdSMichal Hocko * Dissolve free hugepages in the memory block before doing 1542bb8965bdSMichal Hocko * offlining actually in order to make hugetlbfs's object 1543bb8965bdSMichal Hocko * counting consistent. 1544c8721bbbSNaoya Horiguchi */ 1545082d5b6bSGerald Schaefer ret = dissolve_free_huge_pages(start_pfn, end_pfn); 154679605093SMichal Hocko if (ret) { 154779605093SMichal Hocko reason = "failure to dissolve huge pages"; 154879605093SMichal Hocko goto failed_removal_isolated; 154979605093SMichal Hocko } 15500c0e6195SKAMEZAWA Hiroyuki /* check again */ 15515557c766SMichal Hocko ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, 15525557c766SMichal Hocko NULL, check_pages_isolated_cb); 15535557c766SMichal Hocko } while (ret); 1554bb8965bdSMichal Hocko 1555b3834be5SAdam Buchbinder /* Ok, all of our target is isolated. 15560c0e6195SKAMEZAWA Hiroyuki We cannot do rollback at this point. */ 15575557c766SMichal Hocko walk_system_ram_range(start_pfn, end_pfn - start_pfn, 15585557c766SMichal Hocko &offlined_pages, offline_isolated_pages_cb); 15595557c766SMichal Hocko pr_info("Offlined Pages %ld\n", offlined_pages); 15609b7ea46aSQian Cai /* 15619b7ea46aSQian Cai * Onlining will reset pagetype flags and makes migrate type 15629b7ea46aSQian Cai * MOVABLE, so just need to decrease the number of isolated 15639b7ea46aSQian Cai * pageblocks zone counter here. 15649b7ea46aSQian Cai */ 15659b7ea46aSQian Cai spin_lock_irqsave(&zone->lock, flags); 15669b7ea46aSQian Cai zone->nr_isolate_pageblock -= nr_isolate_pageblock; 15679b7ea46aSQian Cai spin_unlock_irqrestore(&zone->lock, flags); 15689b7ea46aSQian Cai 15690c0e6195SKAMEZAWA Hiroyuki /* removal success */ 15703dcc0571SJiang Liu adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); 15710c0e6195SKAMEZAWA Hiroyuki zone->present_pages -= offlined_pages; 1572d702909fSCody P Schafer 1573d702909fSCody P Schafer pgdat_resize_lock(zone->zone_pgdat, &flags); 15740c0e6195SKAMEZAWA Hiroyuki zone->zone_pgdat->node_present_pages -= offlined_pages; 1575d702909fSCody P Schafer pgdat_resize_unlock(zone->zone_pgdat, &flags); 15767b78d335SYasunori Goto 15771b79acc9SKOSAKI Motohiro init_per_zone_wmark_min(); 15781b79acc9SKOSAKI Motohiro 15791e8537baSXishi Qiu if (!populated_zone(zone)) { 1580340175b7SJiang Liu zone_pcp_reset(zone); 158172675e13SMichal Hocko build_all_zonelists(NULL); 15821e8537baSXishi Qiu } else 15831e8537baSXishi Qiu zone_pcp_update(zone); 1584340175b7SJiang Liu 1585d9713679SLai Jiangshan node_states_clear_node(node, &arg); 1586698b1b30SVlastimil Babka if (arg.status_change_nid >= 0) { 15878fe23e05SDavid Rientjes kswapd_stop(node); 1588698b1b30SVlastimil Babka kcompactd_stop(node); 1589698b1b30SVlastimil Babka } 1590bce7394aSMinchan Kim 15910c0e6195SKAMEZAWA Hiroyuki vm_total_pages = nr_free_pagecache_pages(); 15920c0e6195SKAMEZAWA Hiroyuki writeback_set_ratelimit(); 15937b78d335SYasunori Goto 15947b78d335SYasunori Goto memory_notify(MEM_OFFLINE, &arg); 1595381eab4aSDavid Hildenbrand mem_hotplug_done(); 15960c0e6195SKAMEZAWA Hiroyuki return 0; 15970c0e6195SKAMEZAWA Hiroyuki 159879605093SMichal Hocko failed_removal_isolated: 159979605093SMichal Hocko undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); 1600c4efe484SQian Cai memory_notify(MEM_CANCEL_OFFLINE, &arg); 16010c0e6195SKAMEZAWA Hiroyuki failed_removal: 160279605093SMichal Hocko pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", 1603a62e2f4fSBjorn Helgaas (unsigned long long) start_pfn << PAGE_SHIFT, 160479605093SMichal Hocko ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, 160579605093SMichal Hocko reason); 16060c0e6195SKAMEZAWA Hiroyuki /* pushback to free area */ 1607381eab4aSDavid Hildenbrand mem_hotplug_done(); 16080c0e6195SKAMEZAWA Hiroyuki return ret; 16090c0e6195SKAMEZAWA Hiroyuki } 161071088785SBadari Pulavarty 1611a16cee10SWen Congyang int offline_pages(unsigned long start_pfn, unsigned long nr_pages) 1612a16cee10SWen Congyang { 1613ecde0f3eSMichal Hocko return __offline_pages(start_pfn, start_pfn + nr_pages); 1614a16cee10SWen Congyang } 1615a16cee10SWen Congyang 1616d6de9d53SXishi Qiu static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) 1617bbc76be6SWen Congyang { 1618bbc76be6SWen Congyang int ret = !is_memblock_offlined(mem); 1619bbc76be6SWen Congyang 1620349daa0fSRandy Dunlap if (unlikely(ret)) { 1621349daa0fSRandy Dunlap phys_addr_t beginpa, endpa; 1622349daa0fSRandy Dunlap 1623349daa0fSRandy Dunlap beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); 1624b6c88d3bSDavid Hildenbrand endpa = beginpa + memory_block_size_bytes() - 1; 1625756a025fSJoe Perches pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", 1626349daa0fSRandy Dunlap &beginpa, &endpa); 1627bbc76be6SWen Congyang 1628eca499abSPavel Tatashin return -EBUSY; 1629eca499abSPavel Tatashin } 1630eca499abSPavel Tatashin return 0; 1631bbc76be6SWen Congyang } 1632bbc76be6SWen Congyang 16330f1cfe9dSToshi Kani static int check_cpu_on_node(pg_data_t *pgdat) 163460a5a19eSTang Chen { 163560a5a19eSTang Chen int cpu; 163660a5a19eSTang Chen 163760a5a19eSTang Chen for_each_present_cpu(cpu) { 163860a5a19eSTang Chen if (cpu_to_node(cpu) == pgdat->node_id) 163960a5a19eSTang Chen /* 164060a5a19eSTang Chen * the cpu on this node isn't removed, and we can't 164160a5a19eSTang Chen * offline this node. 164260a5a19eSTang Chen */ 164360a5a19eSTang Chen return -EBUSY; 164460a5a19eSTang Chen } 164560a5a19eSTang Chen 164660a5a19eSTang Chen return 0; 164760a5a19eSTang Chen } 164860a5a19eSTang Chen 16490f1cfe9dSToshi Kani /** 16500f1cfe9dSToshi Kani * try_offline_node 1651e8b098fcSMike Rapoport * @nid: the node ID 16520f1cfe9dSToshi Kani * 16530f1cfe9dSToshi Kani * Offline a node if all memory sections and cpus of the node are removed. 16540f1cfe9dSToshi Kani * 16550f1cfe9dSToshi Kani * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 16560f1cfe9dSToshi Kani * and online/offline operations before this call. 16570f1cfe9dSToshi Kani */ 165890b30cdcSWen Congyang void try_offline_node(int nid) 165960a5a19eSTang Chen { 1660d822b86aSWen Congyang pg_data_t *pgdat = NODE_DATA(nid); 1661d822b86aSWen Congyang unsigned long start_pfn = pgdat->node_start_pfn; 1662d822b86aSWen Congyang unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; 166360a5a19eSTang Chen unsigned long pfn; 166460a5a19eSTang Chen 166560a5a19eSTang Chen for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { 166660a5a19eSTang Chen unsigned long section_nr = pfn_to_section_nr(pfn); 166760a5a19eSTang Chen 166860a5a19eSTang Chen if (!present_section_nr(section_nr)) 166960a5a19eSTang Chen continue; 167060a5a19eSTang Chen 167160a5a19eSTang Chen if (pfn_to_nid(pfn) != nid) 167260a5a19eSTang Chen continue; 167360a5a19eSTang Chen 167460a5a19eSTang Chen /* 167560a5a19eSTang Chen * some memory sections of this node are not removed, and we 167660a5a19eSTang Chen * can't offline node now. 167760a5a19eSTang Chen */ 167860a5a19eSTang Chen return; 167960a5a19eSTang Chen } 168060a5a19eSTang Chen 168146a3679bSMichal Hocko if (check_cpu_on_node(pgdat)) 168260a5a19eSTang Chen return; 168360a5a19eSTang Chen 168460a5a19eSTang Chen /* 168560a5a19eSTang Chen * all memory/cpu of this node are removed, we can offline this 168660a5a19eSTang Chen * node now. 168760a5a19eSTang Chen */ 168860a5a19eSTang Chen node_set_offline(nid); 168960a5a19eSTang Chen unregister_one_node(nid); 169060a5a19eSTang Chen } 169190b30cdcSWen Congyang EXPORT_SYMBOL(try_offline_node); 169260a5a19eSTang Chen 1693d9eb1417SDavid Hildenbrand static void __release_memory_resource(resource_size_t start, 1694d9eb1417SDavid Hildenbrand resource_size_t size) 1695d9eb1417SDavid Hildenbrand { 1696d9eb1417SDavid Hildenbrand int ret; 1697d9eb1417SDavid Hildenbrand 1698d9eb1417SDavid Hildenbrand /* 1699d9eb1417SDavid Hildenbrand * When removing memory in the same granularity as it was added, 1700d9eb1417SDavid Hildenbrand * this function never fails. It might only fail if resources 1701d9eb1417SDavid Hildenbrand * have to be adjusted or split. We'll ignore the error, as 1702d9eb1417SDavid Hildenbrand * removing of memory cannot fail. 1703d9eb1417SDavid Hildenbrand */ 1704d9eb1417SDavid Hildenbrand ret = release_mem_region_adjustable(&iomem_resource, start, size); 1705d9eb1417SDavid Hildenbrand if (ret) { 1706d9eb1417SDavid Hildenbrand resource_size_t endres = start + size - 1; 1707d9eb1417SDavid Hildenbrand 1708d9eb1417SDavid Hildenbrand pr_warn("Unable to release resource <%pa-%pa> (%d)\n", 1709d9eb1417SDavid Hildenbrand &start, &endres, ret); 1710d9eb1417SDavid Hildenbrand } 1711d9eb1417SDavid Hildenbrand } 1712d9eb1417SDavid Hildenbrand 1713eca499abSPavel Tatashin static int __ref try_remove_memory(int nid, u64 start, u64 size) 1714bbc76be6SWen Congyang { 1715eca499abSPavel Tatashin int rc = 0; 1716993c1aadSWen Congyang 171727356f54SToshi Kani BUG_ON(check_hotplug_memory_range(start, size)); 171827356f54SToshi Kani 1719bfc8c901SVladimir Davydov mem_hotplug_begin(); 17206677e3eaSYasuaki Ishimatsu 17216677e3eaSYasuaki Ishimatsu /* 1722242831ebSRafael J. Wysocki * All memory blocks must be offlined before removing memory. Check 1723eca499abSPavel Tatashin * whether all memory blocks in question are offline and return error 1724242831ebSRafael J. Wysocki * if this is not the case. 17256677e3eaSYasuaki Ishimatsu */ 1726fbcf73ceSDavid Hildenbrand rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); 1727eca499abSPavel Tatashin if (rc) 1728eca499abSPavel Tatashin goto done; 17296677e3eaSYasuaki Ishimatsu 173046c66c4bSYasuaki Ishimatsu /* remove memmap entry */ 173146c66c4bSYasuaki Ishimatsu firmware_map_remove(start, start + size, "System RAM"); 1732f9126ab9SXishi Qiu memblock_free(start, size); 1733f9126ab9SXishi Qiu memblock_remove(start, size); 173446c66c4bSYasuaki Ishimatsu 17354c4b7f9bSDavid Hildenbrand /* remove memory block devices before removing memory */ 17364c4b7f9bSDavid Hildenbrand remove_memory_block_devices(start, size); 17374c4b7f9bSDavid Hildenbrand 17382c2a5af6SOscar Salvador arch_remove_memory(nid, start, size, NULL); 1739d9eb1417SDavid Hildenbrand __release_memory_resource(start, size); 174024d335caSWen Congyang 174160a5a19eSTang Chen try_offline_node(nid); 174260a5a19eSTang Chen 1743eca499abSPavel Tatashin done: 1744bfc8c901SVladimir Davydov mem_hotplug_done(); 1745eca499abSPavel Tatashin return rc; 174671088785SBadari Pulavarty } 1747d15e5926SDavid Hildenbrand 1748eca499abSPavel Tatashin /** 1749eca499abSPavel Tatashin * remove_memory 1750eca499abSPavel Tatashin * @nid: the node ID 1751eca499abSPavel Tatashin * @start: physical address of the region to remove 1752eca499abSPavel Tatashin * @size: size of the region to remove 1753eca499abSPavel Tatashin * 1754eca499abSPavel Tatashin * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 1755eca499abSPavel Tatashin * and online/offline operations before this call, as required by 1756eca499abSPavel Tatashin * try_offline_node(). 1757eca499abSPavel Tatashin */ 1758eca499abSPavel Tatashin void __remove_memory(int nid, u64 start, u64 size) 1759d15e5926SDavid Hildenbrand { 1760eca499abSPavel Tatashin 1761eca499abSPavel Tatashin /* 176229a90db9SSouptick Joarder * trigger BUG() if some memory is not offlined prior to calling this 1763eca499abSPavel Tatashin * function 1764eca499abSPavel Tatashin */ 1765eca499abSPavel Tatashin if (try_remove_memory(nid, start, size)) 1766eca499abSPavel Tatashin BUG(); 1767eca499abSPavel Tatashin } 1768eca499abSPavel Tatashin 1769eca499abSPavel Tatashin /* 1770eca499abSPavel Tatashin * Remove memory if every memory block is offline, otherwise return -EBUSY is 1771eca499abSPavel Tatashin * some memory is not offline 1772eca499abSPavel Tatashin */ 1773eca499abSPavel Tatashin int remove_memory(int nid, u64 start, u64 size) 1774eca499abSPavel Tatashin { 1775eca499abSPavel Tatashin int rc; 1776eca499abSPavel Tatashin 1777d15e5926SDavid Hildenbrand lock_device_hotplug(); 1778eca499abSPavel Tatashin rc = try_remove_memory(nid, start, size); 1779d15e5926SDavid Hildenbrand unlock_device_hotplug(); 1780eca499abSPavel Tatashin 1781eca499abSPavel Tatashin return rc; 1782d15e5926SDavid Hildenbrand } 178371088785SBadari Pulavarty EXPORT_SYMBOL_GPL(remove_memory); 1784aba6efc4SRafael J. Wysocki #endif /* CONFIG_MEMORY_HOTREMOVE */ 1785