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 45e3a9d9fcSOscar Salvador 46e3a9d9fcSOscar Salvador /* 47e3a9d9fcSOscar Salvador * memory_hotplug.memmap_on_memory parameter 48e3a9d9fcSOscar Salvador */ 49e3a9d9fcSOscar Salvador static bool memmap_on_memory __ro_after_init; 50e3a9d9fcSOscar Salvador #ifdef CONFIG_MHP_MEMMAP_ON_MEMORY 51e3a9d9fcSOscar Salvador module_param(memmap_on_memory, bool, 0444); 52e3a9d9fcSOscar Salvador MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug"); 53e3a9d9fcSOscar Salvador #endif 54a08a2ae3SOscar Salvador 559d0ad8caSDaniel Kiper /* 569d0ad8caSDaniel Kiper * online_page_callback contains pointer to current page onlining function. 579d0ad8caSDaniel Kiper * Initially it is generic_online_page(). If it is required it could be 589d0ad8caSDaniel Kiper * changed by calling set_online_page_callback() for callback registration 599d0ad8caSDaniel Kiper * and restore_online_page_callback() for generic callback restore. 609d0ad8caSDaniel Kiper */ 619d0ad8caSDaniel Kiper 629d0ad8caSDaniel Kiper static online_page_callback_t online_page_callback = generic_online_page; 63bfc8c901SVladimir Davydov static DEFINE_MUTEX(online_page_callback_lock); 649d0ad8caSDaniel Kiper 653f906ba2SThomas Gleixner DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); 6620d6c96bSKOSAKI Motohiro 673f906ba2SThomas Gleixner void get_online_mems(void) 683f906ba2SThomas Gleixner { 693f906ba2SThomas Gleixner percpu_down_read(&mem_hotplug_lock); 703f906ba2SThomas Gleixner } 71bfc8c901SVladimir Davydov 723f906ba2SThomas Gleixner void put_online_mems(void) 733f906ba2SThomas Gleixner { 743f906ba2SThomas Gleixner percpu_up_read(&mem_hotplug_lock); 753f906ba2SThomas Gleixner } 76bfc8c901SVladimir Davydov 774932381eSMichal Hocko bool movable_node_enabled = false; 784932381eSMichal Hocko 798604d9e5SVitaly Kuznetsov #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE 801adf8b46SAnshuman Khandual int mhp_default_online_type = MMOP_OFFLINE; 818604d9e5SVitaly Kuznetsov #else 821adf8b46SAnshuman Khandual int mhp_default_online_type = MMOP_ONLINE; 838604d9e5SVitaly Kuznetsov #endif 8431bc3858SVitaly Kuznetsov 8586dd995dSVitaly Kuznetsov static int __init setup_memhp_default_state(char *str) 8686dd995dSVitaly Kuznetsov { 871adf8b46SAnshuman Khandual const int online_type = mhp_online_type_from_str(str); 885f47adf7SDavid Hildenbrand 895f47adf7SDavid Hildenbrand if (online_type >= 0) 901adf8b46SAnshuman Khandual mhp_default_online_type = online_type; 9186dd995dSVitaly Kuznetsov 9286dd995dSVitaly Kuznetsov return 1; 9386dd995dSVitaly Kuznetsov } 9486dd995dSVitaly Kuznetsov __setup("memhp_default_state=", setup_memhp_default_state); 9586dd995dSVitaly Kuznetsov 9630467e0bSDavid Rientjes void mem_hotplug_begin(void) 97bfc8c901SVladimir Davydov { 983f906ba2SThomas Gleixner cpus_read_lock(); 993f906ba2SThomas Gleixner percpu_down_write(&mem_hotplug_lock); 100bfc8c901SVladimir Davydov } 101bfc8c901SVladimir Davydov 10230467e0bSDavid Rientjes void mem_hotplug_done(void) 103bfc8c901SVladimir Davydov { 1043f906ba2SThomas Gleixner percpu_up_write(&mem_hotplug_lock); 1053f906ba2SThomas Gleixner cpus_read_unlock(); 106bfc8c901SVladimir Davydov } 10720d6c96bSKOSAKI Motohiro 108357b4da5SJuergen Gross u64 max_mem_size = U64_MAX; 109357b4da5SJuergen Gross 11045e0b78bSKeith Mannthey /* add this memory to iomem resource */ 1117b7b2721SDavid Hildenbrand static struct resource *register_memory_resource(u64 start, u64 size, 1127b7b2721SDavid Hildenbrand const char *resource_name) 11345e0b78bSKeith Mannthey { 1142794129eSDave Hansen struct resource *res; 1152794129eSDave Hansen unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; 1167b7b2721SDavid Hildenbrand 1177b7b2721SDavid Hildenbrand if (strcmp(resource_name, "System RAM")) 1187cf603d1SDavid Hildenbrand flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED; 119357b4da5SJuergen Gross 120bca3feaaSAnshuman Khandual if (!mhp_range_allowed(start, size, true)) 121bca3feaaSAnshuman Khandual return ERR_PTR(-E2BIG); 122bca3feaaSAnshuman Khandual 123f3cd4c86SBaoquan He /* 124f3cd4c86SBaoquan He * Make sure value parsed from 'mem=' only restricts memory adding 125f3cd4c86SBaoquan He * while booting, so that memory hotplug won't be impacted. Please 126f3cd4c86SBaoquan He * refer to document of 'mem=' in kernel-parameters.txt for more 127f3cd4c86SBaoquan He * details. 128f3cd4c86SBaoquan He */ 129f3cd4c86SBaoquan He if (start + size > max_mem_size && system_state < SYSTEM_RUNNING) 130357b4da5SJuergen Gross return ERR_PTR(-E2BIG); 131357b4da5SJuergen Gross 1322794129eSDave Hansen /* 1332794129eSDave Hansen * Request ownership of the new memory range. This might be 1342794129eSDave Hansen * a child of an existing resource that was present but 1352794129eSDave Hansen * not marked as busy. 1362794129eSDave Hansen */ 1372794129eSDave Hansen res = __request_region(&iomem_resource, start, size, 1382794129eSDave Hansen resource_name, flags); 13945e0b78bSKeith Mannthey 1402794129eSDave Hansen if (!res) { 1412794129eSDave Hansen pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n", 1422794129eSDave Hansen start, start + size); 1436f754ba4SVitaly Kuznetsov return ERR_PTR(-EEXIST); 14445e0b78bSKeith Mannthey } 14545e0b78bSKeith Mannthey return res; 14645e0b78bSKeith Mannthey } 14745e0b78bSKeith Mannthey 14845e0b78bSKeith Mannthey static void release_memory_resource(struct resource *res) 14945e0b78bSKeith Mannthey { 15045e0b78bSKeith Mannthey if (!res) 15145e0b78bSKeith Mannthey return; 15245e0b78bSKeith Mannthey release_resource(res); 15345e0b78bSKeith Mannthey kfree(res); 15445e0b78bSKeith Mannthey } 15545e0b78bSKeith Mannthey 15653947027SKeith Mannthey #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE 15746723bfaSYasuaki Ishimatsu void get_page_bootmem(unsigned long info, struct page *page, 1585f24ce5fSAndrea Arcangeli unsigned long type) 15904753278SYasunori Goto { 160ddffe98dSYasuaki Ishimatsu page->freelist = (void *)type; 16104753278SYasunori Goto SetPagePrivate(page); 16204753278SYasunori Goto set_page_private(page, info); 163fe896d18SJoonsoo Kim page_ref_inc(page); 16404753278SYasunori Goto } 16504753278SYasunori Goto 166170a5a7eSJiang Liu void put_page_bootmem(struct page *page) 16704753278SYasunori Goto { 1685f24ce5fSAndrea Arcangeli unsigned long type; 16904753278SYasunori Goto 170ddffe98dSYasuaki Ishimatsu type = (unsigned long) page->freelist; 1715f24ce5fSAndrea Arcangeli BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || 1725f24ce5fSAndrea Arcangeli type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); 17304753278SYasunori Goto 174fe896d18SJoonsoo Kim if (page_ref_dec_return(page) == 1) { 175ddffe98dSYasuaki Ishimatsu page->freelist = NULL; 17604753278SYasunori Goto ClearPagePrivate(page); 17704753278SYasunori Goto set_page_private(page, 0); 1785f24ce5fSAndrea Arcangeli INIT_LIST_HEAD(&page->lru); 179170a5a7eSJiang Liu free_reserved_page(page); 18004753278SYasunori Goto } 18104753278SYasunori Goto } 18204753278SYasunori Goto 18346723bfaSYasuaki Ishimatsu #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE 18446723bfaSYasuaki Ishimatsu #ifndef CONFIG_SPARSEMEM_VMEMMAP 185d92bc318SAdrian Bunk static void register_page_bootmem_info_section(unsigned long start_pfn) 18604753278SYasunori Goto { 187f1eca35aSDan Williams unsigned long mapsize, section_nr, i; 18804753278SYasunori Goto struct mem_section *ms; 18904753278SYasunori Goto struct page *page, *memmap; 190f1eca35aSDan Williams struct mem_section_usage *usage; 19104753278SYasunori Goto 19204753278SYasunori Goto section_nr = pfn_to_section_nr(start_pfn); 19304753278SYasunori Goto ms = __nr_to_section(section_nr); 19404753278SYasunori Goto 19504753278SYasunori Goto /* Get section's memmap address */ 19604753278SYasunori Goto memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); 19704753278SYasunori Goto 19804753278SYasunori Goto /* 19904753278SYasunori Goto * Get page for the memmap's phys address 20004753278SYasunori Goto * XXX: need more consideration for sparse_vmemmap... 20104753278SYasunori Goto */ 20204753278SYasunori Goto page = virt_to_page(memmap); 20304753278SYasunori Goto mapsize = sizeof(struct page) * PAGES_PER_SECTION; 20404753278SYasunori Goto mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; 20504753278SYasunori Goto 20604753278SYasunori Goto /* remember memmap's page */ 20704753278SYasunori Goto for (i = 0; i < mapsize; i++, page++) 20804753278SYasunori Goto get_page_bootmem(section_nr, page, SECTION_INFO); 20904753278SYasunori Goto 210f1eca35aSDan Williams usage = ms->usage; 211f1eca35aSDan Williams page = virt_to_page(usage); 21204753278SYasunori Goto 213f1eca35aSDan Williams mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; 21404753278SYasunori Goto 21504753278SYasunori Goto for (i = 0; i < mapsize; i++, page++) 216af370fb8SYasunori Goto get_page_bootmem(section_nr, page, MIX_SECTION_INFO); 21704753278SYasunori Goto 21804753278SYasunori Goto } 21946723bfaSYasuaki Ishimatsu #else /* CONFIG_SPARSEMEM_VMEMMAP */ 22046723bfaSYasuaki Ishimatsu static void register_page_bootmem_info_section(unsigned long start_pfn) 22146723bfaSYasuaki Ishimatsu { 222f1eca35aSDan Williams unsigned long mapsize, section_nr, i; 22346723bfaSYasuaki Ishimatsu struct mem_section *ms; 22446723bfaSYasuaki Ishimatsu struct page *page, *memmap; 225f1eca35aSDan Williams struct mem_section_usage *usage; 22646723bfaSYasuaki Ishimatsu 22746723bfaSYasuaki Ishimatsu section_nr = pfn_to_section_nr(start_pfn); 22846723bfaSYasuaki Ishimatsu ms = __nr_to_section(section_nr); 22946723bfaSYasuaki Ishimatsu 23046723bfaSYasuaki Ishimatsu memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); 23146723bfaSYasuaki Ishimatsu 23246723bfaSYasuaki Ishimatsu register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); 23346723bfaSYasuaki Ishimatsu 234f1eca35aSDan Williams usage = ms->usage; 235f1eca35aSDan Williams page = virt_to_page(usage); 23646723bfaSYasuaki Ishimatsu 237f1eca35aSDan Williams mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT; 23846723bfaSYasuaki Ishimatsu 23946723bfaSYasuaki Ishimatsu for (i = 0; i < mapsize; i++, page++) 24046723bfaSYasuaki Ishimatsu get_page_bootmem(section_nr, page, MIX_SECTION_INFO); 24146723bfaSYasuaki Ishimatsu } 24246723bfaSYasuaki Ishimatsu #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ 24304753278SYasunori Goto 2447ded384aSLinus Torvalds void __init register_page_bootmem_info_node(struct pglist_data *pgdat) 24504753278SYasunori Goto { 24604753278SYasunori Goto unsigned long i, pfn, end_pfn, nr_pages; 24704753278SYasunori Goto int node = pgdat->node_id; 24804753278SYasunori Goto struct page *page; 24904753278SYasunori Goto 25004753278SYasunori Goto nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; 25104753278SYasunori Goto page = virt_to_page(pgdat); 25204753278SYasunori Goto 25304753278SYasunori Goto for (i = 0; i < nr_pages; i++, page++) 25404753278SYasunori Goto get_page_bootmem(node, page, NODE_INFO); 25504753278SYasunori Goto 25604753278SYasunori Goto pfn = pgdat->node_start_pfn; 257c1f19495SCody P Schafer end_pfn = pgdat_end_pfn(pgdat); 25804753278SYasunori Goto 2597e9f5eb0STang Chen /* register section info */ 260f14851afSqiuxishi for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { 261f14851afSqiuxishi /* 262f14851afSqiuxishi * Some platforms can assign the same pfn to multiple nodes - on 263f14851afSqiuxishi * node0 as well as nodeN. To avoid registering a pfn against 264f14851afSqiuxishi * multiple nodes we check that this pfn does not already 2657e9f5eb0STang Chen * reside in some other nodes. 266f14851afSqiuxishi */ 267f65e91dfSYang Shi if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) 26804753278SYasunori Goto register_page_bootmem_info_section(pfn); 269f14851afSqiuxishi } 27004753278SYasunori Goto } 27146723bfaSYasuaki Ishimatsu #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ 27204753278SYasunori Goto 2737ea62160SDan Williams static int check_pfn_span(unsigned long pfn, unsigned long nr_pages, 2747ea62160SDan Williams const char *reason) 2757ea62160SDan Williams { 2767ea62160SDan Williams /* 2777ea62160SDan Williams * Disallow all operations smaller than a sub-section and only 2787ea62160SDan Williams * allow operations smaller than a section for 2797ea62160SDan Williams * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range() 2807ea62160SDan Williams * enforces a larger memory_block_size_bytes() granularity for 2817ea62160SDan Williams * memory that will be marked online, so this check should only 2827ea62160SDan Williams * fire for direct arch_{add,remove}_memory() users outside of 2837ea62160SDan Williams * add_memory_resource(). 2847ea62160SDan Williams */ 2857ea62160SDan Williams unsigned long min_align; 2867ea62160SDan Williams 2877ea62160SDan Williams if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 2887ea62160SDan Williams min_align = PAGES_PER_SUBSECTION; 2897ea62160SDan Williams else 2907ea62160SDan Williams min_align = PAGES_PER_SECTION; 2917ea62160SDan Williams if (!IS_ALIGNED(pfn, min_align) 2927ea62160SDan Williams || !IS_ALIGNED(nr_pages, min_align)) { 2937ea62160SDan Williams WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n", 2947ea62160SDan Williams reason, pfn, pfn + nr_pages - 1); 2957ea62160SDan Williams return -EINVAL; 2967ea62160SDan Williams } 2977ea62160SDan Williams return 0; 2987ea62160SDan Williams } 2997ea62160SDan Williams 3004edd7cefSDavid Rientjes /* 3019f605f26SDan Williams * Return page for the valid pfn only if the page is online. All pfn 3029f605f26SDan Williams * walkers which rely on the fully initialized page->flags and others 3039f605f26SDan Williams * should use this rather than pfn_valid && pfn_to_page 3049f605f26SDan Williams */ 3059f605f26SDan Williams struct page *pfn_to_online_page(unsigned long pfn) 3069f605f26SDan Williams { 3079f605f26SDan Williams unsigned long nr = pfn_to_section_nr(pfn); 3081f90a347SDan Williams struct dev_pagemap *pgmap; 3099f9b02e5SDan Williams struct mem_section *ms; 3109f605f26SDan Williams 3119f9b02e5SDan Williams if (nr >= NR_MEM_SECTIONS) 3129f605f26SDan Williams return NULL; 3139f9b02e5SDan Williams 3149f9b02e5SDan Williams ms = __nr_to_section(nr); 3159f9b02e5SDan Williams if (!online_section(ms)) 3169f9b02e5SDan Williams return NULL; 3179f9b02e5SDan Williams 3189f9b02e5SDan Williams /* 3199f9b02e5SDan Williams * Save some code text when online_section() + 3209f9b02e5SDan Williams * pfn_section_valid() are sufficient. 3219f9b02e5SDan Williams */ 3229f9b02e5SDan Williams if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn)) 3239f9b02e5SDan Williams return NULL; 3249f9b02e5SDan Williams 3259f9b02e5SDan Williams if (!pfn_section_valid(ms, pfn)) 3269f9b02e5SDan Williams return NULL; 3279f9b02e5SDan Williams 3281f90a347SDan Williams if (!online_device_section(ms)) 3291f90a347SDan Williams return pfn_to_page(pfn); 3301f90a347SDan Williams 3311f90a347SDan Williams /* 3321f90a347SDan Williams * Slowpath: when ZONE_DEVICE collides with 3331f90a347SDan Williams * ZONE_{NORMAL,MOVABLE} within the same section some pfns in 3341f90a347SDan Williams * the section may be 'offline' but 'valid'. Only 3351f90a347SDan Williams * get_dev_pagemap() can determine sub-section online status. 3361f90a347SDan Williams */ 3371f90a347SDan Williams pgmap = get_dev_pagemap(pfn, NULL); 3381f90a347SDan Williams put_dev_pagemap(pgmap); 3391f90a347SDan Williams 3401f90a347SDan Williams /* The presence of a pgmap indicates ZONE_DEVICE offline pfn */ 3411f90a347SDan Williams if (pgmap) 3421f90a347SDan Williams return NULL; 3431f90a347SDan Williams 3449f9b02e5SDan Williams return pfn_to_page(pfn); 3459f605f26SDan Williams } 3469f605f26SDan Williams EXPORT_SYMBOL_GPL(pfn_to_online_page); 3479f605f26SDan Williams 3489f605f26SDan Williams /* 3494edd7cefSDavid Rientjes * Reasonably generic function for adding memory. It is 3504edd7cefSDavid Rientjes * expected that archs that support memory hotplug will 3514edd7cefSDavid Rientjes * call this function after deciding the zone to which to 3524edd7cefSDavid Rientjes * add the new pages. 3534edd7cefSDavid Rientjes */ 3547ea62160SDan Williams int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages, 355f5637d3bSLogan Gunthorpe struct mhp_params *params) 3564edd7cefSDavid Rientjes { 3576cdd0b30SDavid Hildenbrand const unsigned long end_pfn = pfn + nr_pages; 3586cdd0b30SDavid Hildenbrand unsigned long cur_nr_pages; 3599a845030SDan Williams int err; 360f5637d3bSLogan Gunthorpe struct vmem_altmap *altmap = params->altmap; 3614b94ffdcSDan Williams 362bfeb022fSLogan Gunthorpe if (WARN_ON_ONCE(!params->pgprot.pgprot)) 363bfeb022fSLogan Gunthorpe return -EINVAL; 364bfeb022fSLogan Gunthorpe 365bca3feaaSAnshuman Khandual VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false)); 366dca4436dSAlastair D'Silva 3674b94ffdcSDan Williams if (altmap) { 3684b94ffdcSDan Williams /* 3694b94ffdcSDan Williams * Validate altmap is within bounds of the total request 3704b94ffdcSDan Williams */ 3717ea62160SDan Williams if (altmap->base_pfn != pfn 3724b94ffdcSDan Williams || vmem_altmap_offset(altmap) > nr_pages) { 3734b94ffdcSDan Williams pr_warn_once("memory add fail, invalid altmap\n"); 3747ea62160SDan Williams return -EINVAL; 3754b94ffdcSDan Williams } 3764b94ffdcSDan Williams altmap->alloc = 0; 3774b94ffdcSDan Williams } 3784b94ffdcSDan Williams 3797ea62160SDan Williams err = check_pfn_span(pfn, nr_pages, "add"); 3807ea62160SDan Williams if (err) 3817ea62160SDan Williams return err; 3827ea62160SDan Williams 3836cdd0b30SDavid Hildenbrand for (; pfn < end_pfn; pfn += cur_nr_pages) { 3846cdd0b30SDavid Hildenbrand /* Select all remaining pages up to the next section boundary */ 3856cdd0b30SDavid Hildenbrand cur_nr_pages = min(end_pfn - pfn, 3866cdd0b30SDavid Hildenbrand SECTION_ALIGN_UP(pfn + 1) - pfn); 3876cdd0b30SDavid Hildenbrand err = sparse_add_section(nid, pfn, cur_nr_pages, altmap); 388ba72b4c8SDan Williams if (err) 389ba72b4c8SDan Williams break; 390f64ac5e6SMichal Hocko cond_resched(); 3914edd7cefSDavid Rientjes } 392c435a390SZhu Guihua vmemmap_populate_print_last(); 3934edd7cefSDavid Rientjes return err; 3944edd7cefSDavid Rientjes } 3954edd7cefSDavid Rientjes 396815121d2SYasuaki Ishimatsu /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ 397d09b0137SYASUAKI ISHIMATSU static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, 398815121d2SYasuaki Ishimatsu unsigned long start_pfn, 399815121d2SYasuaki Ishimatsu unsigned long end_pfn) 400815121d2SYasuaki Ishimatsu { 40149ba3c6bSDan Williams for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) { 4027ce700bfSDavid Hildenbrand if (unlikely(!pfn_to_online_page(start_pfn))) 403815121d2SYasuaki Ishimatsu continue; 404815121d2SYasuaki Ishimatsu 405815121d2SYasuaki Ishimatsu if (unlikely(pfn_to_nid(start_pfn) != nid)) 406815121d2SYasuaki Ishimatsu continue; 407815121d2SYasuaki Ishimatsu 4089b05158fSDavid Hildenbrand if (zone != page_zone(pfn_to_page(start_pfn))) 409815121d2SYasuaki Ishimatsu continue; 410815121d2SYasuaki Ishimatsu 411815121d2SYasuaki Ishimatsu return start_pfn; 412815121d2SYasuaki Ishimatsu } 413815121d2SYasuaki Ishimatsu 414815121d2SYasuaki Ishimatsu return 0; 415815121d2SYasuaki Ishimatsu } 416815121d2SYasuaki Ishimatsu 417815121d2SYasuaki Ishimatsu /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ 418d09b0137SYASUAKI ISHIMATSU static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, 419815121d2SYasuaki Ishimatsu unsigned long start_pfn, 420815121d2SYasuaki Ishimatsu unsigned long end_pfn) 421815121d2SYasuaki Ishimatsu { 422815121d2SYasuaki Ishimatsu unsigned long pfn; 423815121d2SYasuaki Ishimatsu 424815121d2SYasuaki Ishimatsu /* pfn is the end pfn of a memory section. */ 425815121d2SYasuaki Ishimatsu pfn = end_pfn - 1; 42649ba3c6bSDan Williams for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) { 4277ce700bfSDavid Hildenbrand if (unlikely(!pfn_to_online_page(pfn))) 428815121d2SYasuaki Ishimatsu continue; 429815121d2SYasuaki Ishimatsu 430815121d2SYasuaki Ishimatsu if (unlikely(pfn_to_nid(pfn) != nid)) 431815121d2SYasuaki Ishimatsu continue; 432815121d2SYasuaki Ishimatsu 4339b05158fSDavid Hildenbrand if (zone != page_zone(pfn_to_page(pfn))) 434815121d2SYasuaki Ishimatsu continue; 435815121d2SYasuaki Ishimatsu 436815121d2SYasuaki Ishimatsu return pfn; 437815121d2SYasuaki Ishimatsu } 438815121d2SYasuaki Ishimatsu 439815121d2SYasuaki Ishimatsu return 0; 440815121d2SYasuaki Ishimatsu } 441815121d2SYasuaki Ishimatsu 442815121d2SYasuaki Ishimatsu static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, 443815121d2SYasuaki Ishimatsu unsigned long end_pfn) 444815121d2SYasuaki Ishimatsu { 445815121d2SYasuaki Ishimatsu unsigned long pfn; 446815121d2SYasuaki Ishimatsu int nid = zone_to_nid(zone); 447815121d2SYasuaki Ishimatsu 448815121d2SYasuaki Ishimatsu zone_span_writelock(zone); 4495d12071cSDavid Hildenbrand if (zone->zone_start_pfn == start_pfn) { 450815121d2SYasuaki Ishimatsu /* 451815121d2SYasuaki Ishimatsu * If the section is smallest section in the zone, it need 452815121d2SYasuaki Ishimatsu * shrink zone->zone_start_pfn and zone->zone_spanned_pages. 453815121d2SYasuaki Ishimatsu * In this case, we find second smallest valid mem_section 454815121d2SYasuaki Ishimatsu * for shrinking zone. 455815121d2SYasuaki Ishimatsu */ 456815121d2SYasuaki Ishimatsu pfn = find_smallest_section_pfn(nid, zone, end_pfn, 4575d12071cSDavid Hildenbrand zone_end_pfn(zone)); 458815121d2SYasuaki Ishimatsu if (pfn) { 4595d12071cSDavid Hildenbrand zone->spanned_pages = zone_end_pfn(zone) - pfn; 460815121d2SYasuaki Ishimatsu zone->zone_start_pfn = pfn; 461950b68d9SDavid Hildenbrand } else { 462950b68d9SDavid Hildenbrand zone->zone_start_pfn = 0; 463950b68d9SDavid Hildenbrand zone->spanned_pages = 0; 464815121d2SYasuaki Ishimatsu } 4655d12071cSDavid Hildenbrand } else if (zone_end_pfn(zone) == end_pfn) { 466815121d2SYasuaki Ishimatsu /* 467815121d2SYasuaki Ishimatsu * If the section is biggest section in the zone, it need 468815121d2SYasuaki Ishimatsu * shrink zone->spanned_pages. 469815121d2SYasuaki Ishimatsu * In this case, we find second biggest valid mem_section for 470815121d2SYasuaki Ishimatsu * shrinking zone. 471815121d2SYasuaki Ishimatsu */ 4725d12071cSDavid Hildenbrand pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn, 473815121d2SYasuaki Ishimatsu start_pfn); 474815121d2SYasuaki Ishimatsu if (pfn) 4755d12071cSDavid Hildenbrand zone->spanned_pages = pfn - zone->zone_start_pfn + 1; 476950b68d9SDavid Hildenbrand else { 477815121d2SYasuaki Ishimatsu zone->zone_start_pfn = 0; 478815121d2SYasuaki Ishimatsu zone->spanned_pages = 0; 479950b68d9SDavid Hildenbrand } 480950b68d9SDavid Hildenbrand } 481815121d2SYasuaki Ishimatsu zone_span_writeunlock(zone); 482815121d2SYasuaki Ishimatsu } 483815121d2SYasuaki Ishimatsu 48400d6c019SDavid Hildenbrand static void update_pgdat_span(struct pglist_data *pgdat) 485815121d2SYasuaki Ishimatsu { 48600d6c019SDavid Hildenbrand unsigned long node_start_pfn = 0, node_end_pfn = 0; 48700d6c019SDavid Hildenbrand struct zone *zone; 488815121d2SYasuaki Ishimatsu 48900d6c019SDavid Hildenbrand for (zone = pgdat->node_zones; 49000d6c019SDavid Hildenbrand zone < pgdat->node_zones + MAX_NR_ZONES; zone++) { 4916c922cf7SMiaohe Lin unsigned long end_pfn = zone_end_pfn(zone); 49200d6c019SDavid Hildenbrand 49300d6c019SDavid Hildenbrand /* No need to lock the zones, they can't change. */ 494656d5711SDavid Hildenbrand if (!zone->spanned_pages) 495656d5711SDavid Hildenbrand continue; 496656d5711SDavid Hildenbrand if (!node_end_pfn) { 497656d5711SDavid Hildenbrand node_start_pfn = zone->zone_start_pfn; 4986c922cf7SMiaohe Lin node_end_pfn = end_pfn; 499656d5711SDavid Hildenbrand continue; 500656d5711SDavid Hildenbrand } 501656d5711SDavid Hildenbrand 5026c922cf7SMiaohe Lin if (end_pfn > node_end_pfn) 5036c922cf7SMiaohe Lin node_end_pfn = end_pfn; 50400d6c019SDavid Hildenbrand if (zone->zone_start_pfn < node_start_pfn) 50500d6c019SDavid Hildenbrand node_start_pfn = zone->zone_start_pfn; 506815121d2SYasuaki Ishimatsu } 507815121d2SYasuaki Ishimatsu 50800d6c019SDavid Hildenbrand pgdat->node_start_pfn = node_start_pfn; 50900d6c019SDavid Hildenbrand pgdat->node_spanned_pages = node_end_pfn - node_start_pfn; 510815121d2SYasuaki Ishimatsu } 511815121d2SYasuaki Ishimatsu 512feee6b29SDavid Hildenbrand void __ref remove_pfn_range_from_zone(struct zone *zone, 513feee6b29SDavid Hildenbrand unsigned long start_pfn, 5147ea62160SDan Williams unsigned long nr_pages) 515815121d2SYasuaki Ishimatsu { 516b7e3debdSBen Widawsky const unsigned long end_pfn = start_pfn + nr_pages; 517815121d2SYasuaki Ishimatsu struct pglist_data *pgdat = zone->zone_pgdat; 518b7e3debdSBen Widawsky unsigned long pfn, cur_nr_pages, flags; 519815121d2SYasuaki Ishimatsu 520d33695b1SDavid Hildenbrand /* Poison struct pages because they are now uninitialized again. */ 521b7e3debdSBen Widawsky for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) { 522b7e3debdSBen Widawsky cond_resched(); 523b7e3debdSBen Widawsky 524b7e3debdSBen Widawsky /* Select all remaining pages up to the next section boundary */ 525b7e3debdSBen Widawsky cur_nr_pages = 526b7e3debdSBen Widawsky min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn); 527b7e3debdSBen Widawsky page_init_poison(pfn_to_page(pfn), 528b7e3debdSBen Widawsky sizeof(struct page) * cur_nr_pages); 529b7e3debdSBen Widawsky } 530d33695b1SDavid Hildenbrand 5317ce700bfSDavid Hildenbrand #ifdef CONFIG_ZONE_DEVICE 5327ce700bfSDavid Hildenbrand /* 5337ce700bfSDavid Hildenbrand * Zone shrinking code cannot properly deal with ZONE_DEVICE. So 5347ce700bfSDavid Hildenbrand * we will not try to shrink the zones - which is okay as 5357ce700bfSDavid Hildenbrand * set_zone_contiguous() cannot deal with ZONE_DEVICE either way. 5367ce700bfSDavid Hildenbrand */ 5377ce700bfSDavid Hildenbrand if (zone_idx(zone) == ZONE_DEVICE) 5387ce700bfSDavid Hildenbrand return; 5397ce700bfSDavid Hildenbrand #endif 5407ce700bfSDavid Hildenbrand 541feee6b29SDavid Hildenbrand clear_zone_contiguous(zone); 542feee6b29SDavid Hildenbrand 543815121d2SYasuaki Ishimatsu pgdat_resize_lock(zone->zone_pgdat, &flags); 544815121d2SYasuaki Ishimatsu shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); 54500d6c019SDavid Hildenbrand update_pgdat_span(pgdat); 546815121d2SYasuaki Ishimatsu pgdat_resize_unlock(zone->zone_pgdat, &flags); 547feee6b29SDavid Hildenbrand 548feee6b29SDavid Hildenbrand set_zone_contiguous(zone); 549815121d2SYasuaki Ishimatsu } 550815121d2SYasuaki Ishimatsu 551feee6b29SDavid Hildenbrand static void __remove_section(unsigned long pfn, unsigned long nr_pages, 552feee6b29SDavid Hildenbrand unsigned long map_offset, 5539d1d887dSDavid Hildenbrand struct vmem_altmap *altmap) 554ea01ea93SBadari Pulavarty { 55510404901Schenqiwu struct mem_section *ms = __pfn_to_section(pfn); 556ea01ea93SBadari Pulavarty 5579d1d887dSDavid Hildenbrand if (WARN_ON_ONCE(!valid_section(ms))) 5589d1d887dSDavid Hildenbrand return; 559ea01ea93SBadari Pulavarty 560ba72b4c8SDan Williams sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap); 561ea01ea93SBadari Pulavarty } 562ea01ea93SBadari Pulavarty 563ea01ea93SBadari Pulavarty /** 564feee6b29SDavid Hildenbrand * __remove_pages() - remove sections of pages 5657ea62160SDan Williams * @pfn: starting pageframe (must be aligned to start of a section) 566ea01ea93SBadari Pulavarty * @nr_pages: number of pages to remove (must be multiple of section size) 567e8b098fcSMike Rapoport * @altmap: alternative device page map or %NULL if default memmap is used 568ea01ea93SBadari Pulavarty * 569ea01ea93SBadari Pulavarty * Generic helper function to remove section mappings and sysfs entries 570ea01ea93SBadari Pulavarty * for the section of the memory we are removing. Caller needs to make 571ea01ea93SBadari Pulavarty * sure that pages are marked reserved and zones are adjust properly by 572ea01ea93SBadari Pulavarty * calling offline_pages(). 573ea01ea93SBadari Pulavarty */ 574feee6b29SDavid Hildenbrand void __remove_pages(unsigned long pfn, unsigned long nr_pages, 575feee6b29SDavid Hildenbrand struct vmem_altmap *altmap) 576ea01ea93SBadari Pulavarty { 57752fb87c8SDavid Hildenbrand const unsigned long end_pfn = pfn + nr_pages; 57852fb87c8SDavid Hildenbrand unsigned long cur_nr_pages; 5794b94ffdcSDan Williams unsigned long map_offset = 0; 5804b94ffdcSDan Williams 5814b94ffdcSDan Williams map_offset = vmem_altmap_offset(altmap); 582ea01ea93SBadari Pulavarty 5837ea62160SDan Williams if (check_pfn_span(pfn, nr_pages, "remove")) 5847ea62160SDan Williams return; 585ea01ea93SBadari Pulavarty 58652fb87c8SDavid Hildenbrand for (; pfn < end_pfn; pfn += cur_nr_pages) { 587dd33ad7bSMichal Hocko cond_resched(); 58852fb87c8SDavid Hildenbrand /* Select all remaining pages up to the next section boundary */ 589a11b9419SDavid Hildenbrand cur_nr_pages = min(end_pfn - pfn, 590a11b9419SDavid Hildenbrand SECTION_ALIGN_UP(pfn + 1) - pfn); 59152fb87c8SDavid Hildenbrand __remove_section(pfn, cur_nr_pages, map_offset, altmap); 5924b94ffdcSDan Williams map_offset = 0; 593ea01ea93SBadari Pulavarty } 594ea01ea93SBadari Pulavarty } 595ea01ea93SBadari Pulavarty 5969d0ad8caSDaniel Kiper int set_online_page_callback(online_page_callback_t callback) 5979d0ad8caSDaniel Kiper { 5989d0ad8caSDaniel Kiper int rc = -EINVAL; 5999d0ad8caSDaniel Kiper 600bfc8c901SVladimir Davydov get_online_mems(); 601bfc8c901SVladimir Davydov mutex_lock(&online_page_callback_lock); 6029d0ad8caSDaniel Kiper 6039d0ad8caSDaniel Kiper if (online_page_callback == generic_online_page) { 6049d0ad8caSDaniel Kiper online_page_callback = callback; 6059d0ad8caSDaniel Kiper rc = 0; 6069d0ad8caSDaniel Kiper } 6079d0ad8caSDaniel Kiper 608bfc8c901SVladimir Davydov mutex_unlock(&online_page_callback_lock); 609bfc8c901SVladimir Davydov put_online_mems(); 6109d0ad8caSDaniel Kiper 6119d0ad8caSDaniel Kiper return rc; 6129d0ad8caSDaniel Kiper } 6139d0ad8caSDaniel Kiper EXPORT_SYMBOL_GPL(set_online_page_callback); 6149d0ad8caSDaniel Kiper 6159d0ad8caSDaniel Kiper int restore_online_page_callback(online_page_callback_t callback) 6169d0ad8caSDaniel Kiper { 6179d0ad8caSDaniel Kiper int rc = -EINVAL; 6189d0ad8caSDaniel Kiper 619bfc8c901SVladimir Davydov get_online_mems(); 620bfc8c901SVladimir Davydov mutex_lock(&online_page_callback_lock); 6219d0ad8caSDaniel Kiper 6229d0ad8caSDaniel Kiper if (online_page_callback == callback) { 6239d0ad8caSDaniel Kiper online_page_callback = generic_online_page; 6249d0ad8caSDaniel Kiper rc = 0; 6259d0ad8caSDaniel Kiper } 6269d0ad8caSDaniel Kiper 627bfc8c901SVladimir Davydov mutex_unlock(&online_page_callback_lock); 628bfc8c901SVladimir Davydov put_online_mems(); 6299d0ad8caSDaniel Kiper 6309d0ad8caSDaniel Kiper return rc; 6319d0ad8caSDaniel Kiper } 6329d0ad8caSDaniel Kiper EXPORT_SYMBOL_GPL(restore_online_page_callback); 6339d0ad8caSDaniel Kiper 63418db1491SDavid Hildenbrand void generic_online_page(struct page *page, unsigned int order) 6359d0ad8caSDaniel Kiper { 636c87cbc1fSVlastimil Babka /* 637c87cbc1fSVlastimil Babka * Freeing the page with debug_pagealloc enabled will try to unmap it, 638c87cbc1fSVlastimil Babka * so we should map it first. This is better than introducing a special 639c87cbc1fSVlastimil Babka * case in page freeing fast path. 640c87cbc1fSVlastimil Babka */ 64177bc7fd6SMike Rapoport debug_pagealloc_map_pages(page, 1 << order); 642a9cd410aSArun KS __free_pages_core(page, order); 643a9cd410aSArun KS totalram_pages_add(1UL << order); 644a9cd410aSArun KS #ifdef CONFIG_HIGHMEM 645a9cd410aSArun KS if (PageHighMem(page)) 646a9cd410aSArun KS totalhigh_pages_add(1UL << order); 647a9cd410aSArun KS #endif 648a9cd410aSArun KS } 64918db1491SDavid Hildenbrand EXPORT_SYMBOL_GPL(generic_online_page); 650a9cd410aSArun KS 651aac65321SDavid Hildenbrand static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages) 6523947be19SDave Hansen { 653b2c2ab20SDavid Hildenbrand const unsigned long end_pfn = start_pfn + nr_pages; 654b2c2ab20SDavid Hildenbrand unsigned long pfn; 6552d070eabSMichal Hocko 656b2c2ab20SDavid Hildenbrand /* 657aac65321SDavid Hildenbrand * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might 658aac65321SDavid Hildenbrand * decide to not expose all pages to the buddy (e.g., expose them 659aac65321SDavid Hildenbrand * later). We account all pages as being online and belonging to this 660aac65321SDavid Hildenbrand * zone ("present"). 661a08a2ae3SOscar Salvador * When using memmap_on_memory, the range might not be aligned to 662a08a2ae3SOscar Salvador * MAX_ORDER_NR_PAGES - 1, but pageblock aligned. __ffs() will detect 663a08a2ae3SOscar Salvador * this and the first chunk to online will be pageblock_nr_pages. 664b2c2ab20SDavid Hildenbrand */ 665a08a2ae3SOscar Salvador for (pfn = start_pfn; pfn < end_pfn;) { 666a08a2ae3SOscar Salvador int order = min(MAX_ORDER - 1UL, __ffs(pfn)); 667a08a2ae3SOscar Salvador 668a08a2ae3SOscar Salvador (*online_page_callback)(pfn_to_page(pfn), order); 669a08a2ae3SOscar Salvador pfn += (1UL << order); 670a08a2ae3SOscar Salvador } 6712d070eabSMichal Hocko 672b2c2ab20SDavid Hildenbrand /* mark all involved sections as online */ 673b2c2ab20SDavid Hildenbrand online_mem_sections(start_pfn, end_pfn); 67475884fb1SKAMEZAWA Hiroyuki } 67575884fb1SKAMEZAWA Hiroyuki 676d9713679SLai Jiangshan /* check which state of node_states will be changed when online memory */ 677d9713679SLai Jiangshan static void node_states_check_changes_online(unsigned long nr_pages, 678d9713679SLai Jiangshan struct zone *zone, struct memory_notify *arg) 679d9713679SLai Jiangshan { 680d9713679SLai Jiangshan int nid = zone_to_nid(zone); 681d9713679SLai Jiangshan 68298fa15f3SAnshuman Khandual arg->status_change_nid = NUMA_NO_NODE; 68398fa15f3SAnshuman Khandual arg->status_change_nid_normal = NUMA_NO_NODE; 68498fa15f3SAnshuman Khandual arg->status_change_nid_high = NUMA_NO_NODE; 6856715ddf9SLai Jiangshan 6866715ddf9SLai Jiangshan if (!node_state(nid, N_MEMORY)) 687d9713679SLai Jiangshan arg->status_change_nid = nid; 6888efe33f4SOscar Salvador if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY)) 6898efe33f4SOscar Salvador arg->status_change_nid_normal = nid; 6908efe33f4SOscar Salvador #ifdef CONFIG_HIGHMEM 691d3ba3ae1SBaoquan He if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY)) 6928efe33f4SOscar Salvador arg->status_change_nid_high = nid; 6938efe33f4SOscar Salvador #endif 694d9713679SLai Jiangshan } 695d9713679SLai Jiangshan 696d9713679SLai Jiangshan static void node_states_set_node(int node, struct memory_notify *arg) 697d9713679SLai Jiangshan { 698d9713679SLai Jiangshan if (arg->status_change_nid_normal >= 0) 699d9713679SLai Jiangshan node_set_state(node, N_NORMAL_MEMORY); 700d9713679SLai Jiangshan 7016715ddf9SLai Jiangshan if (arg->status_change_nid_high >= 0) 702d9713679SLai Jiangshan node_set_state(node, N_HIGH_MEMORY); 7036715ddf9SLai Jiangshan 70483d83612SOscar Salvador if (arg->status_change_nid >= 0) 7056715ddf9SLai Jiangshan node_set_state(node, N_MEMORY); 706d9713679SLai Jiangshan } 707d9713679SLai Jiangshan 708f1dd2cd1SMichal Hocko static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, 709f1dd2cd1SMichal Hocko unsigned long nr_pages) 710f1dd2cd1SMichal Hocko { 711f1dd2cd1SMichal Hocko unsigned long old_end_pfn = zone_end_pfn(zone); 712f1dd2cd1SMichal Hocko 713f1dd2cd1SMichal Hocko if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) 714f1dd2cd1SMichal Hocko zone->zone_start_pfn = start_pfn; 715f1dd2cd1SMichal Hocko 716f1dd2cd1SMichal Hocko zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; 717f1dd2cd1SMichal Hocko } 718f1dd2cd1SMichal Hocko 719f1dd2cd1SMichal Hocko static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, 720f1dd2cd1SMichal Hocko unsigned long nr_pages) 721f1dd2cd1SMichal Hocko { 722f1dd2cd1SMichal Hocko unsigned long old_end_pfn = pgdat_end_pfn(pgdat); 723f1dd2cd1SMichal Hocko 724f1dd2cd1SMichal Hocko if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) 725f1dd2cd1SMichal Hocko pgdat->node_start_pfn = start_pfn; 726f1dd2cd1SMichal Hocko 727f1dd2cd1SMichal Hocko pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; 728f1dd2cd1SMichal Hocko 7293fccb74cSDavid Hildenbrand } 7301f90a347SDan Williams 7311f90a347SDan Williams static void section_taint_zone_device(unsigned long pfn) 7321f90a347SDan Williams { 7331f90a347SDan Williams struct mem_section *ms = __pfn_to_section(pfn); 7341f90a347SDan Williams 7351f90a347SDan Williams ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE; 7361f90a347SDan Williams } 7371f90a347SDan Williams 7383fccb74cSDavid Hildenbrand /* 7393fccb74cSDavid Hildenbrand * Associate the pfn range with the given zone, initializing the memmaps 7403fccb74cSDavid Hildenbrand * and resizing the pgdat/zone data to span the added pages. After this 7413fccb74cSDavid Hildenbrand * call, all affected pages are PG_reserved. 742d882c006SDavid Hildenbrand * 743d882c006SDavid Hildenbrand * All aligned pageblocks are initialized to the specified migratetype 744d882c006SDavid Hildenbrand * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related 745d882c006SDavid Hildenbrand * zone stats (e.g., nr_isolate_pageblock) are touched. 7463fccb74cSDavid Hildenbrand */ 747a99583e7SChristoph Hellwig void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn, 748d882c006SDavid Hildenbrand unsigned long nr_pages, 749d882c006SDavid Hildenbrand struct vmem_altmap *altmap, int migratetype) 750f1dd2cd1SMichal Hocko { 751f1dd2cd1SMichal Hocko struct pglist_data *pgdat = zone->zone_pgdat; 752f1dd2cd1SMichal Hocko int nid = pgdat->node_id; 753f1dd2cd1SMichal Hocko unsigned long flags; 754f1dd2cd1SMichal Hocko 755f1dd2cd1SMichal Hocko clear_zone_contiguous(zone); 756f1dd2cd1SMichal Hocko 757f1dd2cd1SMichal Hocko /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ 758f1dd2cd1SMichal Hocko pgdat_resize_lock(pgdat, &flags); 759f1dd2cd1SMichal Hocko zone_span_writelock(zone); 760fa004ab7SWei Yang if (zone_is_empty(zone)) 761fa004ab7SWei Yang init_currently_empty_zone(zone, start_pfn, nr_pages); 762f1dd2cd1SMichal Hocko resize_zone_range(zone, start_pfn, nr_pages); 763f1dd2cd1SMichal Hocko zone_span_writeunlock(zone); 764f1dd2cd1SMichal Hocko resize_pgdat_range(pgdat, start_pfn, nr_pages); 765f1dd2cd1SMichal Hocko pgdat_resize_unlock(pgdat, &flags); 766f1dd2cd1SMichal Hocko 767f1dd2cd1SMichal Hocko /* 7681f90a347SDan Williams * Subsection population requires care in pfn_to_online_page(). 7691f90a347SDan Williams * Set the taint to enable the slow path detection of 7701f90a347SDan Williams * ZONE_DEVICE pages in an otherwise ZONE_{NORMAL,MOVABLE} 7711f90a347SDan Williams * section. 7721f90a347SDan Williams */ 7731f90a347SDan Williams if (zone_is_zone_device(zone)) { 7741f90a347SDan Williams if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION)) 7751f90a347SDan Williams section_taint_zone_device(start_pfn); 7761f90a347SDan Williams if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)) 7771f90a347SDan Williams section_taint_zone_device(start_pfn + nr_pages); 7781f90a347SDan Williams } 7791f90a347SDan Williams 7801f90a347SDan Williams /* 781f1dd2cd1SMichal Hocko * TODO now we have a visible range of pages which are not associated 782f1dd2cd1SMichal Hocko * with their zone properly. Not nice but set_pfnblock_flags_mask 783f1dd2cd1SMichal Hocko * expects the zone spans the pfn range. All the pages in the range 784f1dd2cd1SMichal Hocko * are reserved so nobody should be touching them so we should be safe 785f1dd2cd1SMichal Hocko */ 786ab28cb6eSBaoquan He memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0, 787d882c006SDavid Hildenbrand MEMINIT_HOTPLUG, altmap, migratetype); 788f1dd2cd1SMichal Hocko 789f1dd2cd1SMichal Hocko set_zone_contiguous(zone); 790f1dd2cd1SMichal Hocko } 791f1dd2cd1SMichal Hocko 792f1dd2cd1SMichal Hocko /* 793c246a213SMichal Hocko * Returns a default kernel memory zone for the given pfn range. 794c246a213SMichal Hocko * If no kernel zone covers this pfn range it will automatically go 795c246a213SMichal Hocko * to the ZONE_NORMAL. 796c246a213SMichal Hocko */ 797c6f03e29SMichal Hocko static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, 798c246a213SMichal Hocko unsigned long nr_pages) 799c246a213SMichal Hocko { 800c246a213SMichal Hocko struct pglist_data *pgdat = NODE_DATA(nid); 801c246a213SMichal Hocko int zid; 802c246a213SMichal Hocko 803c246a213SMichal Hocko for (zid = 0; zid <= ZONE_NORMAL; zid++) { 804c246a213SMichal Hocko struct zone *zone = &pgdat->node_zones[zid]; 805c246a213SMichal Hocko 806c246a213SMichal Hocko if (zone_intersects(zone, start_pfn, nr_pages)) 807c246a213SMichal Hocko return zone; 808c246a213SMichal Hocko } 809c246a213SMichal Hocko 810c246a213SMichal Hocko return &pgdat->node_zones[ZONE_NORMAL]; 811c246a213SMichal Hocko } 812c246a213SMichal Hocko 813c6f03e29SMichal Hocko static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, 814c6f03e29SMichal Hocko unsigned long nr_pages) 815e5e68930SMichal Hocko { 816c6f03e29SMichal Hocko struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, 817c6f03e29SMichal Hocko nr_pages); 818c6f03e29SMichal Hocko struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; 819c6f03e29SMichal Hocko bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); 820c6f03e29SMichal Hocko bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); 821e5e68930SMichal Hocko 822e5e68930SMichal Hocko /* 823c6f03e29SMichal Hocko * We inherit the existing zone in a simple case where zones do not 824c6f03e29SMichal Hocko * overlap in the given range 825e5e68930SMichal Hocko */ 826c6f03e29SMichal Hocko if (in_kernel ^ in_movable) 827c6f03e29SMichal Hocko return (in_kernel) ? kernel_zone : movable_zone; 828e5e68930SMichal Hocko 829c6f03e29SMichal Hocko /* 830c6f03e29SMichal Hocko * If the range doesn't belong to any zone or two zones overlap in the 831c6f03e29SMichal Hocko * given range then we use movable zone only if movable_node is 832c6f03e29SMichal Hocko * enabled because we always online to a kernel zone by default. 833c6f03e29SMichal Hocko */ 834c6f03e29SMichal Hocko return movable_node_enabled ? movable_zone : kernel_zone; 8359f123ab5SMichal Hocko } 8369f123ab5SMichal Hocko 837e5e68930SMichal Hocko struct zone *zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, 838e5e68930SMichal Hocko unsigned long nr_pages) 839f1dd2cd1SMichal Hocko { 840c6f03e29SMichal Hocko if (online_type == MMOP_ONLINE_KERNEL) 841c6f03e29SMichal Hocko return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); 842f1dd2cd1SMichal Hocko 843c6f03e29SMichal Hocko if (online_type == MMOP_ONLINE_MOVABLE) 844c6f03e29SMichal Hocko return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; 845f1dd2cd1SMichal Hocko 846c6f03e29SMichal Hocko return default_zone_for_pfn(nid, start_pfn, nr_pages); 847e5e68930SMichal Hocko } 848e5e68930SMichal Hocko 849a08a2ae3SOscar Salvador /* 850a08a2ae3SOscar Salvador * This function should only be called by memory_block_{online,offline}, 851a08a2ae3SOscar Salvador * and {online,offline}_pages. 852a08a2ae3SOscar Salvador */ 853a08a2ae3SOscar Salvador void adjust_present_page_count(struct zone *zone, long nr_pages) 854f9901144SDavid Hildenbrand { 855f9901144SDavid Hildenbrand unsigned long flags; 856f9901144SDavid Hildenbrand 857f9901144SDavid Hildenbrand zone->present_pages += nr_pages; 858f9901144SDavid Hildenbrand pgdat_resize_lock(zone->zone_pgdat, &flags); 859f9901144SDavid Hildenbrand zone->zone_pgdat->node_present_pages += nr_pages; 860f9901144SDavid Hildenbrand pgdat_resize_unlock(zone->zone_pgdat, &flags); 861f9901144SDavid Hildenbrand } 862f9901144SDavid Hildenbrand 863a08a2ae3SOscar Salvador int mhp_init_memmap_on_memory(unsigned long pfn, unsigned long nr_pages, 864a08a2ae3SOscar Salvador struct zone *zone) 865a08a2ae3SOscar Salvador { 866a08a2ae3SOscar Salvador unsigned long end_pfn = pfn + nr_pages; 867a08a2ae3SOscar Salvador int ret; 868a08a2ae3SOscar Salvador 869a08a2ae3SOscar Salvador ret = kasan_add_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages)); 870a08a2ae3SOscar Salvador if (ret) 871a08a2ae3SOscar Salvador return ret; 872a08a2ae3SOscar Salvador 873a08a2ae3SOscar Salvador move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_UNMOVABLE); 874a08a2ae3SOscar Salvador 875a08a2ae3SOscar Salvador /* 876a08a2ae3SOscar Salvador * It might be that the vmemmap_pages fully span sections. If that is 877a08a2ae3SOscar Salvador * the case, mark those sections online here as otherwise they will be 878a08a2ae3SOscar Salvador * left offline. 879a08a2ae3SOscar Salvador */ 880a08a2ae3SOscar Salvador if (nr_pages >= PAGES_PER_SECTION) 881a08a2ae3SOscar Salvador online_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION)); 882a08a2ae3SOscar Salvador 883a08a2ae3SOscar Salvador return ret; 884a08a2ae3SOscar Salvador } 885a08a2ae3SOscar Salvador 886a08a2ae3SOscar Salvador void mhp_deinit_memmap_on_memory(unsigned long pfn, unsigned long nr_pages) 887a08a2ae3SOscar Salvador { 888a08a2ae3SOscar Salvador unsigned long end_pfn = pfn + nr_pages; 889a08a2ae3SOscar Salvador 890a08a2ae3SOscar Salvador /* 891a08a2ae3SOscar Salvador * It might be that the vmemmap_pages fully span sections. If that is 892a08a2ae3SOscar Salvador * the case, mark those sections offline here as otherwise they will be 893a08a2ae3SOscar Salvador * left online. 894a08a2ae3SOscar Salvador */ 895a08a2ae3SOscar Salvador if (nr_pages >= PAGES_PER_SECTION) 896a08a2ae3SOscar Salvador offline_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION)); 897a08a2ae3SOscar Salvador 898a08a2ae3SOscar Salvador /* 899a08a2ae3SOscar Salvador * The pages associated with this vmemmap have been offlined, so 900a08a2ae3SOscar Salvador * we can reset its state here. 901a08a2ae3SOscar Salvador */ 902a08a2ae3SOscar Salvador remove_pfn_range_from_zone(page_zone(pfn_to_page(pfn)), pfn, nr_pages); 903a08a2ae3SOscar Salvador kasan_remove_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages)); 904a08a2ae3SOscar Salvador } 905a08a2ae3SOscar Salvador 906a08a2ae3SOscar Salvador int __ref online_pages(unsigned long pfn, unsigned long nr_pages, struct zone *zone) 90775884fb1SKAMEZAWA Hiroyuki { 908aa47228aSCody P Schafer unsigned long flags; 9096811378eSYasunori Goto int need_zonelists_rebuild = 0; 910a08a2ae3SOscar Salvador const int nid = zone_to_nid(zone); 9117b78d335SYasunori Goto int ret; 9127b78d335SYasunori Goto struct memory_notify arg; 9133947be19SDave Hansen 914dd8e2f23SOscar Salvador /* 915dd8e2f23SOscar Salvador * {on,off}lining is constrained to full memory sections (or more 916dd8e2f23SOscar Salvador * precisly to memory blocks from the user space POV). 917dd8e2f23SOscar Salvador * memmap_on_memory is an exception because it reserves initial part 918dd8e2f23SOscar Salvador * of the physical memory space for vmemmaps. That space is pageblock 919dd8e2f23SOscar Salvador * aligned. 920dd8e2f23SOscar Salvador */ 9214986fac1SDavid Hildenbrand if (WARN_ON_ONCE(!nr_pages || 922dd8e2f23SOscar Salvador !IS_ALIGNED(pfn, pageblock_nr_pages) || 923dd8e2f23SOscar Salvador !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION))) 9244986fac1SDavid Hildenbrand return -EINVAL; 9254986fac1SDavid Hildenbrand 926381eab4aSDavid Hildenbrand mem_hotplug_begin(); 927381eab4aSDavid Hildenbrand 928f1dd2cd1SMichal Hocko /* associate pfn range with the zone */ 929b30c5927SDavid Hildenbrand move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE); 930511c2abaSLai Jiangshan 9317b78d335SYasunori Goto arg.start_pfn = pfn; 9327b78d335SYasunori Goto arg.nr_pages = nr_pages; 933d9713679SLai Jiangshan node_states_check_changes_online(nr_pages, zone, &arg); 9347b78d335SYasunori Goto 9357b78d335SYasunori Goto ret = memory_notify(MEM_GOING_ONLINE, &arg); 9367b78d335SYasunori Goto ret = notifier_to_errno(ret); 937e33e33b4SChen Yucong if (ret) 938e33e33b4SChen Yucong goto failed_addition; 939e33e33b4SChen Yucong 9403947be19SDave Hansen /* 941b30c5927SDavid Hildenbrand * Fixup the number of isolated pageblocks before marking the sections 942b30c5927SDavid Hildenbrand * onlining, such that undo_isolate_page_range() works correctly. 943b30c5927SDavid Hildenbrand */ 944b30c5927SDavid Hildenbrand spin_lock_irqsave(&zone->lock, flags); 945b30c5927SDavid Hildenbrand zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages; 946b30c5927SDavid Hildenbrand spin_unlock_irqrestore(&zone->lock, flags); 947b30c5927SDavid Hildenbrand 948b30c5927SDavid Hildenbrand /* 9496811378eSYasunori Goto * If this zone is not populated, then it is not in zonelist. 9506811378eSYasunori Goto * This means the page allocator ignores this zone. 9516811378eSYasunori Goto * So, zonelist must be updated after online. 9526811378eSYasunori Goto */ 9536dcd73d7SWen Congyang if (!populated_zone(zone)) { 9546811378eSYasunori Goto need_zonelists_rebuild = 1; 95572675e13SMichal Hocko setup_zone_pageset(zone); 9566dcd73d7SWen Congyang } 9576811378eSYasunori Goto 958aac65321SDavid Hildenbrand online_pages_range(pfn, nr_pages); 959f9901144SDavid Hildenbrand adjust_present_page_count(zone, nr_pages); 960aa47228aSCody P Schafer 961b30c5927SDavid Hildenbrand node_states_set_node(nid, &arg); 962b30c5927SDavid Hildenbrand if (need_zonelists_rebuild) 963b30c5927SDavid Hildenbrand build_all_zonelists(NULL); 964b30c5927SDavid Hildenbrand 965b30c5927SDavid Hildenbrand /* Basic onlining is complete, allow allocation of onlined pages. */ 966b30c5927SDavid Hildenbrand undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE); 967b30c5927SDavid Hildenbrand 96893146d98SDavid Hildenbrand /* 969b86c5fc4SDavid Hildenbrand * Freshly onlined pages aren't shuffled (e.g., all pages are placed to 970b86c5fc4SDavid Hildenbrand * the tail of the freelist when undoing isolation). Shuffle the whole 971b86c5fc4SDavid Hildenbrand * zone to make sure the just onlined pages are properly distributed 972b86c5fc4SDavid Hildenbrand * across the whole freelist - to create an initial shuffle. 97393146d98SDavid Hildenbrand */ 974e900a918SDan Williams shuffle_zone(zone); 975e900a918SDan Williams 976*b92ca18eSMel Gorman /* reinitialise watermarks and update pcp limits */ 9771b79acc9SKOSAKI Motohiro init_per_zone_wmark_min(); 9781b79acc9SKOSAKI Motohiro 979e888ca35SVlastimil Babka kswapd_run(nid); 980698b1b30SVlastimil Babka kcompactd_run(nid); 98161b13993SDave Hansen 9822d1d43f6SChandra Seetharaman writeback_set_ratelimit(); 9837b78d335SYasunori Goto 9847b78d335SYasunori Goto memory_notify(MEM_ONLINE, &arg); 985381eab4aSDavid Hildenbrand mem_hotplug_done(); 98630467e0bSDavid Rientjes return 0; 987e33e33b4SChen Yucong 988e33e33b4SChen Yucong failed_addition: 989e33e33b4SChen Yucong pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", 990e33e33b4SChen Yucong (unsigned long long) pfn << PAGE_SHIFT, 991e33e33b4SChen Yucong (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); 992e33e33b4SChen Yucong memory_notify(MEM_CANCEL_ONLINE, &arg); 993feee6b29SDavid Hildenbrand remove_pfn_range_from_zone(zone, pfn, nr_pages); 994381eab4aSDavid Hildenbrand mem_hotplug_done(); 995e33e33b4SChen Yucong return ret; 9963947be19SDave Hansen } 99753947027SKeith Mannthey #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ 998bc02af93SYasunori Goto 9990bd85420STang Chen static void reset_node_present_pages(pg_data_t *pgdat) 10000bd85420STang Chen { 10010bd85420STang Chen struct zone *z; 10020bd85420STang Chen 10030bd85420STang Chen for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 10040bd85420STang Chen z->present_pages = 0; 10050bd85420STang Chen 10060bd85420STang Chen pgdat->node_present_pages = 0; 10070bd85420STang Chen } 10080bd85420STang Chen 1009e1319331SHidetoshi Seto /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ 1010c68ab18cSDavid Hildenbrand static pg_data_t __ref *hotadd_new_pgdat(int nid) 10119af3c2deSYasunori Goto { 10129af3c2deSYasunori Goto struct pglist_data *pgdat; 10139af3c2deSYasunori Goto 1014a1e565aaSTang Chen pgdat = NODE_DATA(nid); 1015a1e565aaSTang Chen if (!pgdat) { 10169af3c2deSYasunori Goto pgdat = arch_alloc_nodedata(nid); 10179af3c2deSYasunori Goto if (!pgdat) 10189af3c2deSYasunori Goto return NULL; 10199af3c2deSYasunori Goto 102033fce011SWei Yang pgdat->per_cpu_nodestats = 102133fce011SWei Yang alloc_percpu(struct per_cpu_nodestat); 10229af3c2deSYasunori Goto arch_refresh_nodedata(nid, pgdat); 1023b0dc3a34SGu Zheng } else { 102433fce011SWei Yang int cpu; 1025e716f2ebSMel Gorman /* 102697a225e6SJoonsoo Kim * Reset the nr_zones, order and highest_zoneidx before reuse. 102797a225e6SJoonsoo Kim * Note that kswapd will init kswapd_highest_zoneidx properly 1028e716f2ebSMel Gorman * when it starts in the near future. 1029e716f2ebSMel Gorman */ 1030b0dc3a34SGu Zheng pgdat->nr_zones = 0; 103138087d9bSMel Gorman pgdat->kswapd_order = 0; 103297a225e6SJoonsoo Kim pgdat->kswapd_highest_zoneidx = 0; 103333fce011SWei Yang for_each_online_cpu(cpu) { 103433fce011SWei Yang struct per_cpu_nodestat *p; 103533fce011SWei Yang 103633fce011SWei Yang p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); 103733fce011SWei Yang memset(p, 0, sizeof(*p)); 103833fce011SWei Yang } 1039a1e565aaSTang Chen } 10409af3c2deSYasunori Goto 10419af3c2deSYasunori Goto /* we can use NODE_DATA(nid) from here */ 104203e85f9dSOscar Salvador pgdat->node_id = nid; 1043c68ab18cSDavid Hildenbrand pgdat->node_start_pfn = 0; 104403e85f9dSOscar Salvador 10459af3c2deSYasunori Goto /* init node's zones as empty zones, we don't have any present pages.*/ 104603e85f9dSOscar Salvador free_area_init_core_hotplug(nid); 10479af3c2deSYasunori Goto 1048959ecc48SKAMEZAWA Hiroyuki /* 1049959ecc48SKAMEZAWA Hiroyuki * The node we allocated has no zone fallback lists. For avoiding 1050959ecc48SKAMEZAWA Hiroyuki * to access not-initialized zonelist, build here. 1051959ecc48SKAMEZAWA Hiroyuki */ 105272675e13SMichal Hocko build_all_zonelists(pgdat); 1053959ecc48SKAMEZAWA Hiroyuki 1054f784a3f1STang Chen /* 10550bd85420STang Chen * When memory is hot-added, all the memory is in offline state. So 10560bd85420STang Chen * clear all zones' present_pages because they will be updated in 10570bd85420STang Chen * online_pages() and offline_pages(). 10580bd85420STang Chen */ 105903e85f9dSOscar Salvador reset_node_managed_pages(pgdat); 10600bd85420STang Chen reset_node_present_pages(pgdat); 10610bd85420STang Chen 10629af3c2deSYasunori Goto return pgdat; 10639af3c2deSYasunori Goto } 10649af3c2deSYasunori Goto 1065b9ff0360SOscar Salvador static void rollback_node_hotadd(int nid) 10669af3c2deSYasunori Goto { 1067b9ff0360SOscar Salvador pg_data_t *pgdat = NODE_DATA(nid); 1068b9ff0360SOscar Salvador 10699af3c2deSYasunori Goto arch_refresh_nodedata(nid, NULL); 10705830169fSReza Arbab free_percpu(pgdat->per_cpu_nodestats); 10719af3c2deSYasunori Goto arch_free_nodedata(pgdat); 10729af3c2deSYasunori Goto } 10739af3c2deSYasunori Goto 10740a547039SKAMEZAWA Hiroyuki 107501b0f197SToshi Kani /** 107601b0f197SToshi Kani * try_online_node - online a node if offlined 1077e8b098fcSMike Rapoport * @nid: the node ID 1078b9ff0360SOscar Salvador * @set_node_online: Whether we want to online the node 1079cf23422bSminskey guo * called by cpu_up() to online a node without onlined memory. 1080b9ff0360SOscar Salvador * 1081b9ff0360SOscar Salvador * Returns: 1082b9ff0360SOscar Salvador * 1 -> a new node has been allocated 1083b9ff0360SOscar Salvador * 0 -> the node is already online 1084b9ff0360SOscar Salvador * -ENOMEM -> the node could not be allocated 1085cf23422bSminskey guo */ 1086c68ab18cSDavid Hildenbrand static int __try_online_node(int nid, bool set_node_online) 1087cf23422bSminskey guo { 1088cf23422bSminskey guo pg_data_t *pgdat; 1089b9ff0360SOscar Salvador int ret = 1; 1090cf23422bSminskey guo 109101b0f197SToshi Kani if (node_online(nid)) 109201b0f197SToshi Kani return 0; 109301b0f197SToshi Kani 1094c68ab18cSDavid Hildenbrand pgdat = hotadd_new_pgdat(nid); 10957553e8f2SDavid Rientjes if (!pgdat) { 109601b0f197SToshi Kani pr_err("Cannot online node %d due to NULL pgdat\n", nid); 1097cf23422bSminskey guo ret = -ENOMEM; 1098cf23422bSminskey guo goto out; 1099cf23422bSminskey guo } 1100b9ff0360SOscar Salvador 1101b9ff0360SOscar Salvador if (set_node_online) { 1102cf23422bSminskey guo node_set_online(nid); 1103cf23422bSminskey guo ret = register_one_node(nid); 1104cf23422bSminskey guo BUG_ON(ret); 1105b9ff0360SOscar Salvador } 1106cf23422bSminskey guo out: 1107b9ff0360SOscar Salvador return ret; 1108b9ff0360SOscar Salvador } 1109b9ff0360SOscar Salvador 1110b9ff0360SOscar Salvador /* 1111b9ff0360SOscar Salvador * Users of this function always want to online/register the node 1112b9ff0360SOscar Salvador */ 1113b9ff0360SOscar Salvador int try_online_node(int nid) 1114b9ff0360SOscar Salvador { 1115b9ff0360SOscar Salvador int ret; 1116b9ff0360SOscar Salvador 1117b9ff0360SOscar Salvador mem_hotplug_begin(); 1118c68ab18cSDavid Hildenbrand ret = __try_online_node(nid, true); 1119bfc8c901SVladimir Davydov mem_hotplug_done(); 1120cf23422bSminskey guo return ret; 1121cf23422bSminskey guo } 1122cf23422bSminskey guo 112327356f54SToshi Kani static int check_hotplug_memory_range(u64 start, u64 size) 112427356f54SToshi Kani { 1125ba325585SPavel Tatashin /* memory range must be block size aligned */ 1126cec3ebd0SDavid Hildenbrand if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) || 1127cec3ebd0SDavid Hildenbrand !IS_ALIGNED(size, memory_block_size_bytes())) { 1128ba325585SPavel Tatashin pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx", 1129cec3ebd0SDavid Hildenbrand memory_block_size_bytes(), start, size); 113027356f54SToshi Kani return -EINVAL; 113127356f54SToshi Kani } 113227356f54SToshi Kani 113327356f54SToshi Kani return 0; 113427356f54SToshi Kani } 113527356f54SToshi Kani 113631bc3858SVitaly Kuznetsov static int online_memory_block(struct memory_block *mem, void *arg) 113731bc3858SVitaly Kuznetsov { 11381adf8b46SAnshuman Khandual mem->online_type = mhp_default_online_type; 1139dc18d706SNathan Fontenot return device_online(&mem->dev); 114031bc3858SVitaly Kuznetsov } 114131bc3858SVitaly Kuznetsov 1142a08a2ae3SOscar Salvador bool mhp_supports_memmap_on_memory(unsigned long size) 1143a08a2ae3SOscar Salvador { 1144a08a2ae3SOscar Salvador unsigned long nr_vmemmap_pages = size / PAGE_SIZE; 1145a08a2ae3SOscar Salvador unsigned long vmemmap_size = nr_vmemmap_pages * sizeof(struct page); 1146a08a2ae3SOscar Salvador unsigned long remaining_size = size - vmemmap_size; 1147a08a2ae3SOscar Salvador 1148a08a2ae3SOscar Salvador /* 1149a08a2ae3SOscar Salvador * Besides having arch support and the feature enabled at runtime, we 1150a08a2ae3SOscar Salvador * need a few more assumptions to hold true: 1151a08a2ae3SOscar Salvador * 1152a08a2ae3SOscar Salvador * a) We span a single memory block: memory onlining/offlinin;g happens 1153a08a2ae3SOscar Salvador * in memory block granularity. We don't want the vmemmap of online 1154a08a2ae3SOscar Salvador * memory blocks to reside on offline memory blocks. In the future, 1155a08a2ae3SOscar Salvador * we might want to support variable-sized memory blocks to make the 1156a08a2ae3SOscar Salvador * feature more versatile. 1157a08a2ae3SOscar Salvador * 1158a08a2ae3SOscar Salvador * b) The vmemmap pages span complete PMDs: We don't want vmemmap code 1159a08a2ae3SOscar Salvador * to populate memory from the altmap for unrelated parts (i.e., 1160a08a2ae3SOscar Salvador * other memory blocks) 1161a08a2ae3SOscar Salvador * 1162a08a2ae3SOscar Salvador * c) The vmemmap pages (and thereby the pages that will be exposed to 1163a08a2ae3SOscar Salvador * the buddy) have to cover full pageblocks: memory onlining/offlining 1164a08a2ae3SOscar Salvador * code requires applicable ranges to be page-aligned, for example, to 1165a08a2ae3SOscar Salvador * set the migratetypes properly. 1166a08a2ae3SOscar Salvador * 1167a08a2ae3SOscar Salvador * TODO: Although we have a check here to make sure that vmemmap pages 1168a08a2ae3SOscar Salvador * fully populate a PMD, it is not the right place to check for 1169a08a2ae3SOscar Salvador * this. A much better solution involves improving vmemmap code 1170a08a2ae3SOscar Salvador * to fallback to base pages when trying to populate vmemmap using 1171a08a2ae3SOscar Salvador * altmap as an alternative source of memory, and we do not exactly 1172a08a2ae3SOscar Salvador * populate a single PMD. 1173a08a2ae3SOscar Salvador */ 1174a08a2ae3SOscar Salvador return memmap_on_memory && 1175a08a2ae3SOscar Salvador IS_ENABLED(CONFIG_MHP_MEMMAP_ON_MEMORY) && 1176a08a2ae3SOscar Salvador size == memory_block_size_bytes() && 1177a08a2ae3SOscar Salvador IS_ALIGNED(vmemmap_size, PMD_SIZE) && 1178a08a2ae3SOscar Salvador IS_ALIGNED(remaining_size, (pageblock_nr_pages << PAGE_SHIFT)); 1179a08a2ae3SOscar Salvador } 1180a08a2ae3SOscar Salvador 11818df1d0e4SDavid Hildenbrand /* 11828df1d0e4SDavid Hildenbrand * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 11838df1d0e4SDavid Hildenbrand * and online/offline operations (triggered e.g. by sysfs). 11848df1d0e4SDavid Hildenbrand * 11858df1d0e4SDavid Hildenbrand * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG 11868df1d0e4SDavid Hildenbrand */ 1187b6117199SDavid Hildenbrand int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags) 1188bc02af93SYasunori Goto { 1189d15dfd31SCatalin Marinas struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) }; 1190a08a2ae3SOscar Salvador struct vmem_altmap mhp_altmap = {}; 119162cedb9fSDavid Vrabel u64 start, size; 1192b9ff0360SOscar Salvador bool new_node = false; 1193bc02af93SYasunori Goto int ret; 1194bc02af93SYasunori Goto 119562cedb9fSDavid Vrabel start = res->start; 119662cedb9fSDavid Vrabel size = resource_size(res); 119762cedb9fSDavid Vrabel 119827356f54SToshi Kani ret = check_hotplug_memory_range(start, size); 119927356f54SToshi Kani if (ret) 120027356f54SToshi Kani return ret; 120127356f54SToshi Kani 1202fa6d9ec7SVishal Verma if (!node_possible(nid)) { 1203fa6d9ec7SVishal Verma WARN(1, "node %d was absent from the node_possible_map\n", nid); 1204fa6d9ec7SVishal Verma return -EINVAL; 1205fa6d9ec7SVishal Verma } 1206fa6d9ec7SVishal Verma 1207bfc8c901SVladimir Davydov mem_hotplug_begin(); 1208ac13c462SNathan Zimmer 120952219aeaSDavid Hildenbrand if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 12107f36e3e5STang Chen memblock_add_node(start, size, nid); 12117f36e3e5STang Chen 1212c68ab18cSDavid Hildenbrand ret = __try_online_node(nid, false); 1213b9ff0360SOscar Salvador if (ret < 0) 121441b9e2d7SWen Congyang goto error; 1215b9ff0360SOscar Salvador new_node = ret; 12169af3c2deSYasunori Goto 1217a08a2ae3SOscar Salvador /* 1218a08a2ae3SOscar Salvador * Self hosted memmap array 1219a08a2ae3SOscar Salvador */ 1220a08a2ae3SOscar Salvador if (mhp_flags & MHP_MEMMAP_ON_MEMORY) { 1221a08a2ae3SOscar Salvador if (!mhp_supports_memmap_on_memory(size)) { 1222a08a2ae3SOscar Salvador ret = -EINVAL; 1223a08a2ae3SOscar Salvador goto error; 1224a08a2ae3SOscar Salvador } 1225a08a2ae3SOscar Salvador mhp_altmap.free = PHYS_PFN(size); 1226a08a2ae3SOscar Salvador mhp_altmap.base_pfn = PHYS_PFN(start); 1227a08a2ae3SOscar Salvador params.altmap = &mhp_altmap; 1228a08a2ae3SOscar Salvador } 1229a08a2ae3SOscar Salvador 1230bc02af93SYasunori Goto /* call arch's memory hotadd */ 1231f5637d3bSLogan Gunthorpe ret = arch_add_memory(nid, start, size, ¶ms); 12329af3c2deSYasunori Goto if (ret < 0) 12339af3c2deSYasunori Goto goto error; 12349af3c2deSYasunori Goto 1235db051a0dSDavid Hildenbrand /* create memory block devices after memory was added */ 1236a08a2ae3SOscar Salvador ret = create_memory_block_devices(start, size, mhp_altmap.alloc); 1237db051a0dSDavid Hildenbrand if (ret) { 1238db051a0dSDavid Hildenbrand arch_remove_memory(nid, start, size, NULL); 1239db051a0dSDavid Hildenbrand goto error; 1240db051a0dSDavid Hildenbrand } 1241db051a0dSDavid Hildenbrand 1242a1e565aaSTang Chen if (new_node) { 1243d5b6f6a3SOscar Salvador /* If sysfs file of new node can't be created, cpu on the node 12440fc44159SYasunori Goto * can't be hot-added. There is no rollback way now. 12450fc44159SYasunori Goto * So, check by BUG_ON() to catch it reluctantly.. 1246d5b6f6a3SOscar Salvador * We online node here. We can't roll back from here. 12470fc44159SYasunori Goto */ 1248d5b6f6a3SOscar Salvador node_set_online(nid); 1249d5b6f6a3SOscar Salvador ret = __register_one_node(nid); 12500fc44159SYasunori Goto BUG_ON(ret); 12510fc44159SYasunori Goto } 12520fc44159SYasunori Goto 1253d5b6f6a3SOscar Salvador /* link memory sections under this node.*/ 125490c7eaebSLaurent Dufour link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1), 1255f85086f9SLaurent Dufour MEMINIT_HOTPLUG); 1256d5b6f6a3SOscar Salvador 1257d96ae530Sakpm@linux-foundation.org /* create new memmap entry */ 12587b7b2721SDavid Hildenbrand if (!strcmp(res->name, "System RAM")) 1259d96ae530Sakpm@linux-foundation.org firmware_map_add_hotplug(start, start + size, "System RAM"); 1260d96ae530Sakpm@linux-foundation.org 1261381eab4aSDavid Hildenbrand /* device_online() will take the lock when calling online_pages() */ 1262381eab4aSDavid Hildenbrand mem_hotplug_done(); 1263381eab4aSDavid Hildenbrand 12649ca6551eSDavid Hildenbrand /* 12659ca6551eSDavid Hildenbrand * In case we're allowed to merge the resource, flag it and trigger 12669ca6551eSDavid Hildenbrand * merging now that adding succeeded. 12679ca6551eSDavid Hildenbrand */ 126826011267SDavid Hildenbrand if (mhp_flags & MHP_MERGE_RESOURCE) 12699ca6551eSDavid Hildenbrand merge_system_ram_resource(res); 12709ca6551eSDavid Hildenbrand 127131bc3858SVitaly Kuznetsov /* online pages if requested */ 12721adf8b46SAnshuman Khandual if (mhp_default_online_type != MMOP_OFFLINE) 1273fbcf73ceSDavid Hildenbrand walk_memory_blocks(start, size, NULL, online_memory_block); 127431bc3858SVitaly Kuznetsov 1275381eab4aSDavid Hildenbrand return ret; 12769af3c2deSYasunori Goto error: 12779af3c2deSYasunori Goto /* rollback pgdat allocation and others */ 1278b9ff0360SOscar Salvador if (new_node) 1279b9ff0360SOscar Salvador rollback_node_hotadd(nid); 128052219aeaSDavid Hildenbrand if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) 12817f36e3e5STang Chen memblock_remove(start, size); 1282bfc8c901SVladimir Davydov mem_hotplug_done(); 1283bc02af93SYasunori Goto return ret; 1284bc02af93SYasunori Goto } 128562cedb9fSDavid Vrabel 12868df1d0e4SDavid Hildenbrand /* requires device_hotplug_lock, see add_memory_resource() */ 1287b6117199SDavid Hildenbrand int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) 128862cedb9fSDavid Vrabel { 128962cedb9fSDavid Vrabel struct resource *res; 129062cedb9fSDavid Vrabel int ret; 129162cedb9fSDavid Vrabel 12927b7b2721SDavid Hildenbrand res = register_memory_resource(start, size, "System RAM"); 12936f754ba4SVitaly Kuznetsov if (IS_ERR(res)) 12946f754ba4SVitaly Kuznetsov return PTR_ERR(res); 129562cedb9fSDavid Vrabel 1296b6117199SDavid Hildenbrand ret = add_memory_resource(nid, res, mhp_flags); 129762cedb9fSDavid Vrabel if (ret < 0) 129862cedb9fSDavid Vrabel release_memory_resource(res); 129962cedb9fSDavid Vrabel return ret; 130062cedb9fSDavid Vrabel } 13018df1d0e4SDavid Hildenbrand 1302b6117199SDavid Hildenbrand int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags) 13038df1d0e4SDavid Hildenbrand { 13048df1d0e4SDavid Hildenbrand int rc; 13058df1d0e4SDavid Hildenbrand 13068df1d0e4SDavid Hildenbrand lock_device_hotplug(); 1307b6117199SDavid Hildenbrand rc = __add_memory(nid, start, size, mhp_flags); 13088df1d0e4SDavid Hildenbrand unlock_device_hotplug(); 13098df1d0e4SDavid Hildenbrand 13108df1d0e4SDavid Hildenbrand return rc; 13118df1d0e4SDavid Hildenbrand } 1312bc02af93SYasunori Goto EXPORT_SYMBOL_GPL(add_memory); 13130c0e6195SKAMEZAWA Hiroyuki 13147b7b2721SDavid Hildenbrand /* 13157b7b2721SDavid Hildenbrand * Add special, driver-managed memory to the system as system RAM. Such 13167b7b2721SDavid Hildenbrand * memory is not exposed via the raw firmware-provided memmap as system 13177b7b2721SDavid Hildenbrand * RAM, instead, it is detected and added by a driver - during cold boot, 13187b7b2721SDavid Hildenbrand * after a reboot, and after kexec. 13197b7b2721SDavid Hildenbrand * 13207b7b2721SDavid Hildenbrand * Reasons why this memory should not be used for the initial memmap of a 13217b7b2721SDavid Hildenbrand * kexec kernel or for placing kexec images: 13227b7b2721SDavid Hildenbrand * - The booting kernel is in charge of determining how this memory will be 13237b7b2721SDavid Hildenbrand * used (e.g., use persistent memory as system RAM) 13247b7b2721SDavid Hildenbrand * - Coordination with a hypervisor is required before this memory 13257b7b2721SDavid Hildenbrand * can be used (e.g., inaccessible parts). 13267b7b2721SDavid Hildenbrand * 13277b7b2721SDavid Hildenbrand * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided 13287b7b2721SDavid Hildenbrand * memory map") are created. Also, the created memory resource is flagged 13297cf603d1SDavid Hildenbrand * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case 13307b7b2721SDavid Hildenbrand * this memory as well (esp., not place kexec images onto it). 13317b7b2721SDavid Hildenbrand * 13327b7b2721SDavid Hildenbrand * The resource_name (visible via /proc/iomem) has to have the format 13337b7b2721SDavid Hildenbrand * "System RAM ($DRIVER)". 13347b7b2721SDavid Hildenbrand */ 13357b7b2721SDavid Hildenbrand int add_memory_driver_managed(int nid, u64 start, u64 size, 1336b6117199SDavid Hildenbrand const char *resource_name, mhp_t mhp_flags) 13377b7b2721SDavid Hildenbrand { 13387b7b2721SDavid Hildenbrand struct resource *res; 13397b7b2721SDavid Hildenbrand int rc; 13407b7b2721SDavid Hildenbrand 13417b7b2721SDavid Hildenbrand if (!resource_name || 13427b7b2721SDavid Hildenbrand strstr(resource_name, "System RAM (") != resource_name || 13437b7b2721SDavid Hildenbrand resource_name[strlen(resource_name) - 1] != ')') 13447b7b2721SDavid Hildenbrand return -EINVAL; 13457b7b2721SDavid Hildenbrand 13467b7b2721SDavid Hildenbrand lock_device_hotplug(); 13477b7b2721SDavid Hildenbrand 13487b7b2721SDavid Hildenbrand res = register_memory_resource(start, size, resource_name); 13497b7b2721SDavid Hildenbrand if (IS_ERR(res)) { 13507b7b2721SDavid Hildenbrand rc = PTR_ERR(res); 13517b7b2721SDavid Hildenbrand goto out_unlock; 13527b7b2721SDavid Hildenbrand } 13537b7b2721SDavid Hildenbrand 1354b6117199SDavid Hildenbrand rc = add_memory_resource(nid, res, mhp_flags); 13557b7b2721SDavid Hildenbrand if (rc < 0) 13567b7b2721SDavid Hildenbrand release_memory_resource(res); 13577b7b2721SDavid Hildenbrand 13587b7b2721SDavid Hildenbrand out_unlock: 13597b7b2721SDavid Hildenbrand unlock_device_hotplug(); 13607b7b2721SDavid Hildenbrand return rc; 13617b7b2721SDavid Hildenbrand } 13627b7b2721SDavid Hildenbrand EXPORT_SYMBOL_GPL(add_memory_driver_managed); 13637b7b2721SDavid Hildenbrand 1364bca3feaaSAnshuman Khandual /* 1365bca3feaaSAnshuman Khandual * Platforms should define arch_get_mappable_range() that provides 1366bca3feaaSAnshuman Khandual * maximum possible addressable physical memory range for which the 1367bca3feaaSAnshuman Khandual * linear mapping could be created. The platform returned address 1368bca3feaaSAnshuman Khandual * range must adhere to these following semantics. 1369bca3feaaSAnshuman Khandual * 1370bca3feaaSAnshuman Khandual * - range.start <= range.end 1371bca3feaaSAnshuman Khandual * - Range includes both end points [range.start..range.end] 1372bca3feaaSAnshuman Khandual * 1373bca3feaaSAnshuman Khandual * There is also a fallback definition provided here, allowing the 1374bca3feaaSAnshuman Khandual * entire possible physical address range in case any platform does 1375bca3feaaSAnshuman Khandual * not define arch_get_mappable_range(). 1376bca3feaaSAnshuman Khandual */ 1377bca3feaaSAnshuman Khandual struct range __weak arch_get_mappable_range(void) 1378bca3feaaSAnshuman Khandual { 1379bca3feaaSAnshuman Khandual struct range mhp_range = { 1380bca3feaaSAnshuman Khandual .start = 0UL, 1381bca3feaaSAnshuman Khandual .end = -1ULL, 1382bca3feaaSAnshuman Khandual }; 1383bca3feaaSAnshuman Khandual return mhp_range; 1384bca3feaaSAnshuman Khandual } 1385bca3feaaSAnshuman Khandual 1386bca3feaaSAnshuman Khandual struct range mhp_get_pluggable_range(bool need_mapping) 1387bca3feaaSAnshuman Khandual { 1388bca3feaaSAnshuman Khandual const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1; 1389bca3feaaSAnshuman Khandual struct range mhp_range; 1390bca3feaaSAnshuman Khandual 1391bca3feaaSAnshuman Khandual if (need_mapping) { 1392bca3feaaSAnshuman Khandual mhp_range = arch_get_mappable_range(); 1393bca3feaaSAnshuman Khandual if (mhp_range.start > max_phys) { 1394bca3feaaSAnshuman Khandual mhp_range.start = 0; 1395bca3feaaSAnshuman Khandual mhp_range.end = 0; 1396bca3feaaSAnshuman Khandual } 1397bca3feaaSAnshuman Khandual mhp_range.end = min_t(u64, mhp_range.end, max_phys); 1398bca3feaaSAnshuman Khandual } else { 1399bca3feaaSAnshuman Khandual mhp_range.start = 0; 1400bca3feaaSAnshuman Khandual mhp_range.end = max_phys; 1401bca3feaaSAnshuman Khandual } 1402bca3feaaSAnshuman Khandual return mhp_range; 1403bca3feaaSAnshuman Khandual } 1404bca3feaaSAnshuman Khandual EXPORT_SYMBOL_GPL(mhp_get_pluggable_range); 1405bca3feaaSAnshuman Khandual 1406bca3feaaSAnshuman Khandual bool mhp_range_allowed(u64 start, u64 size, bool need_mapping) 1407bca3feaaSAnshuman Khandual { 1408bca3feaaSAnshuman Khandual struct range mhp_range = mhp_get_pluggable_range(need_mapping); 1409bca3feaaSAnshuman Khandual u64 end = start + size; 1410bca3feaaSAnshuman Khandual 1411bca3feaaSAnshuman Khandual if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end) 1412bca3feaaSAnshuman Khandual return true; 1413bca3feaaSAnshuman Khandual 1414bca3feaaSAnshuman Khandual pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n", 1415bca3feaaSAnshuman Khandual start, end, mhp_range.start, mhp_range.end); 1416bca3feaaSAnshuman Khandual return false; 1417bca3feaaSAnshuman Khandual } 1418bca3feaaSAnshuman Khandual 14190c0e6195SKAMEZAWA Hiroyuki #ifdef CONFIG_MEMORY_HOTREMOVE 14200c0e6195SKAMEZAWA Hiroyuki /* 142192917998SDavid Hildenbrand * Confirm all pages in a range [start, end) belong to the same zone (skipping 142292917998SDavid Hildenbrand * memory holes). When true, return the zone. 14230c0e6195SKAMEZAWA Hiroyuki */ 142492917998SDavid Hildenbrand struct zone *test_pages_in_a_zone(unsigned long start_pfn, 142592917998SDavid Hildenbrand unsigned long end_pfn) 14260c0e6195SKAMEZAWA Hiroyuki { 14275f0f2887SAndrew Banman unsigned long pfn, sec_end_pfn; 14280c0e6195SKAMEZAWA Hiroyuki struct zone *zone = NULL; 14290c0e6195SKAMEZAWA Hiroyuki struct page *page; 14300c0e6195SKAMEZAWA Hiroyuki int i; 1431deb88a2aSToshi Kani for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); 14320c0e6195SKAMEZAWA Hiroyuki pfn < end_pfn; 1433deb88a2aSToshi Kani pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { 14345f0f2887SAndrew Banman /* Make sure the memory section is present first */ 14355f0f2887SAndrew Banman if (!present_section_nr(pfn_to_section_nr(pfn))) 14365f0f2887SAndrew Banman continue; 14375f0f2887SAndrew Banman for (; pfn < sec_end_pfn && pfn < end_pfn; 14380c0e6195SKAMEZAWA Hiroyuki pfn += MAX_ORDER_NR_PAGES) { 14390c0e6195SKAMEZAWA Hiroyuki i = 0; 14400c0e6195SKAMEZAWA Hiroyuki /* This is just a CONFIG_HOLES_IN_ZONE check.*/ 14415f0f2887SAndrew Banman while ((i < MAX_ORDER_NR_PAGES) && 14425f0f2887SAndrew Banman !pfn_valid_within(pfn + i)) 14430c0e6195SKAMEZAWA Hiroyuki i++; 1444d6d8c8a4Szhong jiang if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) 14450c0e6195SKAMEZAWA Hiroyuki continue; 144624feb47cSMikhail Zaslonko /* Check if we got outside of the zone */ 144724feb47cSMikhail Zaslonko if (zone && !zone_spans_pfn(zone, pfn + i)) 144892917998SDavid Hildenbrand return NULL; 14490c0e6195SKAMEZAWA Hiroyuki page = pfn_to_page(pfn + i); 14500c0e6195SKAMEZAWA Hiroyuki if (zone && page_zone(page) != zone) 145192917998SDavid Hildenbrand return NULL; 14520c0e6195SKAMEZAWA Hiroyuki zone = page_zone(page); 14530c0e6195SKAMEZAWA Hiroyuki } 14545f0f2887SAndrew Banman } 1455deb88a2aSToshi Kani 145692917998SDavid Hildenbrand return zone; 1457a96dfddbSToshi Kani } 14580c0e6195SKAMEZAWA Hiroyuki 14590c0e6195SKAMEZAWA Hiroyuki /* 14600efadf48SYisheng Xie * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, 1461aa218795SDavid Hildenbrand * non-lru movable pages and hugepages). Will skip over most unmovable 1462aa218795SDavid Hildenbrand * pages (esp., pages that can be skipped when offlining), but bail out on 1463aa218795SDavid Hildenbrand * definitely unmovable pages. 1464aa218795SDavid Hildenbrand * 1465aa218795SDavid Hildenbrand * Returns: 1466aa218795SDavid Hildenbrand * 0 in case a movable page is found and movable_pfn was updated. 1467aa218795SDavid Hildenbrand * -ENOENT in case no movable page was found. 1468aa218795SDavid Hildenbrand * -EBUSY in case a definitely unmovable page was found. 14690c0e6195SKAMEZAWA Hiroyuki */ 1470aa218795SDavid Hildenbrand static int scan_movable_pages(unsigned long start, unsigned long end, 1471aa218795SDavid Hildenbrand unsigned long *movable_pfn) 14720c0e6195SKAMEZAWA Hiroyuki { 14730c0e6195SKAMEZAWA Hiroyuki unsigned long pfn; 1474eeb0efd0SOscar Salvador 14750c0e6195SKAMEZAWA Hiroyuki for (pfn = start; pfn < end; pfn++) { 1476eeb0efd0SOscar Salvador struct page *page, *head; 1477eeb0efd0SOscar Salvador unsigned long skip; 1478eeb0efd0SOscar Salvador 1479eeb0efd0SOscar Salvador if (!pfn_valid(pfn)) 1480eeb0efd0SOscar Salvador continue; 14810c0e6195SKAMEZAWA Hiroyuki page = pfn_to_page(pfn); 14820c0e6195SKAMEZAWA Hiroyuki if (PageLRU(page)) 1483aa218795SDavid Hildenbrand goto found; 14840efadf48SYisheng Xie if (__PageMovable(page)) 1485aa218795SDavid Hildenbrand goto found; 1486aa218795SDavid Hildenbrand 1487aa218795SDavid Hildenbrand /* 1488aa218795SDavid Hildenbrand * PageOffline() pages that are not marked __PageMovable() and 1489aa218795SDavid Hildenbrand * have a reference count > 0 (after MEM_GOING_OFFLINE) are 1490aa218795SDavid Hildenbrand * definitely unmovable. If their reference count would be 0, 1491aa218795SDavid Hildenbrand * they could at least be skipped when offlining memory. 1492aa218795SDavid Hildenbrand */ 1493aa218795SDavid Hildenbrand if (PageOffline(page) && page_count(page)) 1494aa218795SDavid Hildenbrand return -EBUSY; 1495eeb0efd0SOscar Salvador 1496eeb0efd0SOscar Salvador if (!PageHuge(page)) 1497eeb0efd0SOscar Salvador continue; 1498eeb0efd0SOscar Salvador head = compound_head(page); 14998f251a3dSMike Kravetz /* 15008f251a3dSMike Kravetz * This test is racy as we hold no reference or lock. The 15018f251a3dSMike Kravetz * hugetlb page could have been free'ed and head is no longer 15028f251a3dSMike Kravetz * a hugetlb page before the following check. In such unlikely 15038f251a3dSMike Kravetz * cases false positives and negatives are possible. Calling 15048f251a3dSMike Kravetz * code must deal with these scenarios. 15058f251a3dSMike Kravetz */ 15068f251a3dSMike Kravetz if (HPageMigratable(head)) 1507aa218795SDavid Hildenbrand goto found; 1508d8c6546bSMatthew Wilcox (Oracle) skip = compound_nr(head) - (page - head); 1509eeb0efd0SOscar Salvador pfn += skip - 1; 15100c0e6195SKAMEZAWA Hiroyuki } 1511aa218795SDavid Hildenbrand return -ENOENT; 1512aa218795SDavid Hildenbrand found: 1513aa218795SDavid Hildenbrand *movable_pfn = pfn; 15140c0e6195SKAMEZAWA Hiroyuki return 0; 15150c0e6195SKAMEZAWA Hiroyuki } 15160c0e6195SKAMEZAWA Hiroyuki 15170c0e6195SKAMEZAWA Hiroyuki static int 15180c0e6195SKAMEZAWA Hiroyuki do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) 15190c0e6195SKAMEZAWA Hiroyuki { 15200c0e6195SKAMEZAWA Hiroyuki unsigned long pfn; 15216c357848SMatthew Wilcox (Oracle) struct page *page, *head; 15220c0e6195SKAMEZAWA Hiroyuki int ret = 0; 15230c0e6195SKAMEZAWA Hiroyuki LIST_HEAD(source); 15240c0e6195SKAMEZAWA Hiroyuki 1525a85009c3SMichal Hocko for (pfn = start_pfn; pfn < end_pfn; pfn++) { 15260c0e6195SKAMEZAWA Hiroyuki if (!pfn_valid(pfn)) 15270c0e6195SKAMEZAWA Hiroyuki continue; 15280c0e6195SKAMEZAWA Hiroyuki page = pfn_to_page(pfn); 15296c357848SMatthew Wilcox (Oracle) head = compound_head(page); 1530c8721bbbSNaoya Horiguchi 1531c8721bbbSNaoya Horiguchi if (PageHuge(page)) { 1532d8c6546bSMatthew Wilcox (Oracle) pfn = page_to_pfn(head) + compound_nr(head) - 1; 1533daf3538aSOscar Salvador isolate_huge_page(head, &source); 1534c8721bbbSNaoya Horiguchi continue; 153594723aafSMichal Hocko } else if (PageTransHuge(page)) 15366c357848SMatthew Wilcox (Oracle) pfn = page_to_pfn(head) + thp_nr_pages(page) - 1; 1537c8721bbbSNaoya Horiguchi 1538b15c8726SMichal Hocko /* 1539b15c8726SMichal Hocko * HWPoison pages have elevated reference counts so the migration would 1540b15c8726SMichal Hocko * fail on them. It also doesn't make any sense to migrate them in the 1541b15c8726SMichal Hocko * first place. Still try to unmap such a page in case it is still mapped 1542b15c8726SMichal Hocko * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep 1543b15c8726SMichal Hocko * the unmap as the catch all safety net). 1544b15c8726SMichal Hocko */ 1545b15c8726SMichal Hocko if (PageHWPoison(page)) { 1546b15c8726SMichal Hocko if (WARN_ON(PageLRU(page))) 1547b15c8726SMichal Hocko isolate_lru_page(page); 1548b15c8726SMichal Hocko if (page_mapped(page)) 1549013339dfSShakeel Butt try_to_unmap(page, TTU_IGNORE_MLOCK); 1550b15c8726SMichal Hocko continue; 1551b15c8726SMichal Hocko } 1552b15c8726SMichal Hocko 1553700c2a46SKonstantin Khlebnikov if (!get_page_unless_zero(page)) 15540c0e6195SKAMEZAWA Hiroyuki continue; 15550c0e6195SKAMEZAWA Hiroyuki /* 15560efadf48SYisheng Xie * We can skip free pages. And we can deal with pages on 15570efadf48SYisheng Xie * LRU and non-lru movable pages. 15580c0e6195SKAMEZAWA Hiroyuki */ 15590efadf48SYisheng Xie if (PageLRU(page)) 156062695a84SNick Piggin ret = isolate_lru_page(page); 15610efadf48SYisheng Xie else 15620efadf48SYisheng Xie ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); 15630c0e6195SKAMEZAWA Hiroyuki if (!ret) { /* Success */ 156462695a84SNick Piggin list_add_tail(&page->lru, &source); 15650efadf48SYisheng Xie if (!__PageMovable(page)) 1566599d0c95SMel Gorman inc_node_page_state(page, NR_ISOLATED_ANON + 15679de4f22aSHuang Ying page_is_file_lru(page)); 15686d9c285aSKOSAKI Motohiro 15690c0e6195SKAMEZAWA Hiroyuki } else { 15702932c8b0SMichal Hocko pr_warn("failed to isolate pfn %lx\n", pfn); 15710efadf48SYisheng Xie dump_page(page, "isolation failed"); 15721723058eSOscar Salvador } 1573700c2a46SKonstantin Khlebnikov put_page(page); 15740c0e6195SKAMEZAWA Hiroyuki } 1575f3ab2636SBob Liu if (!list_empty(&source)) { 1576203e6e5cSJoonsoo Kim nodemask_t nmask = node_states[N_MEMORY]; 1577203e6e5cSJoonsoo Kim struct migration_target_control mtc = { 1578203e6e5cSJoonsoo Kim .nmask = &nmask, 1579203e6e5cSJoonsoo Kim .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, 1580203e6e5cSJoonsoo Kim }; 1581203e6e5cSJoonsoo Kim 1582203e6e5cSJoonsoo Kim /* 1583203e6e5cSJoonsoo Kim * We have checked that migration range is on a single zone so 1584203e6e5cSJoonsoo Kim * we can use the nid of the first page to all the others. 1585203e6e5cSJoonsoo Kim */ 1586203e6e5cSJoonsoo Kim mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru)); 1587203e6e5cSJoonsoo Kim 1588203e6e5cSJoonsoo Kim /* 1589203e6e5cSJoonsoo Kim * try to allocate from a different node but reuse this node 1590203e6e5cSJoonsoo Kim * if there are no other online nodes to be used (e.g. we are 1591203e6e5cSJoonsoo Kim * offlining a part of the only existing node) 1592203e6e5cSJoonsoo Kim */ 1593203e6e5cSJoonsoo Kim node_clear(mtc.nid, nmask); 1594203e6e5cSJoonsoo Kim if (nodes_empty(nmask)) 1595203e6e5cSJoonsoo Kim node_set(mtc.nid, nmask); 1596203e6e5cSJoonsoo Kim ret = migrate_pages(&source, alloc_migration_target, NULL, 1597203e6e5cSJoonsoo Kim (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG); 15982932c8b0SMichal Hocko if (ret) { 15992932c8b0SMichal Hocko list_for_each_entry(page, &source, lru) { 16002932c8b0SMichal Hocko pr_warn("migrating pfn %lx failed ret:%d ", 16012932c8b0SMichal Hocko page_to_pfn(page), ret); 16022932c8b0SMichal Hocko dump_page(page, "migration failure"); 16032932c8b0SMichal Hocko } 1604c8721bbbSNaoya Horiguchi putback_movable_pages(&source); 1605f3ab2636SBob Liu } 16062932c8b0SMichal Hocko } 16071723058eSOscar Salvador 16080c0e6195SKAMEZAWA Hiroyuki return ret; 16090c0e6195SKAMEZAWA Hiroyuki } 16100c0e6195SKAMEZAWA Hiroyuki 1611c5320926STang Chen static int __init cmdline_parse_movable_node(char *p) 1612c5320926STang Chen { 161355ac590cSTang Chen movable_node_enabled = true; 1614c5320926STang Chen return 0; 1615c5320926STang Chen } 1616c5320926STang Chen early_param("movable_node", cmdline_parse_movable_node); 1617c5320926STang Chen 1618d9713679SLai Jiangshan /* check which state of node_states will be changed when offline memory */ 1619d9713679SLai Jiangshan static void node_states_check_changes_offline(unsigned long nr_pages, 1620d9713679SLai Jiangshan struct zone *zone, struct memory_notify *arg) 1621d9713679SLai Jiangshan { 1622d9713679SLai Jiangshan struct pglist_data *pgdat = zone->zone_pgdat; 1623d9713679SLai Jiangshan unsigned long present_pages = 0; 162486b27beaSOscar Salvador enum zone_type zt; 1625d9713679SLai Jiangshan 162698fa15f3SAnshuman Khandual arg->status_change_nid = NUMA_NO_NODE; 162798fa15f3SAnshuman Khandual arg->status_change_nid_normal = NUMA_NO_NODE; 162898fa15f3SAnshuman Khandual arg->status_change_nid_high = NUMA_NO_NODE; 162986b27beaSOscar Salvador 163086b27beaSOscar Salvador /* 163186b27beaSOscar Salvador * Check whether node_states[N_NORMAL_MEMORY] will be changed. 163286b27beaSOscar Salvador * If the memory to be offline is within the range 163386b27beaSOscar Salvador * [0..ZONE_NORMAL], and it is the last present memory there, 163486b27beaSOscar Salvador * the zones in that range will become empty after the offlining, 163586b27beaSOscar Salvador * thus we can determine that we need to clear the node from 163686b27beaSOscar Salvador * node_states[N_NORMAL_MEMORY]. 163786b27beaSOscar Salvador */ 163886b27beaSOscar Salvador for (zt = 0; zt <= ZONE_NORMAL; zt++) 163986b27beaSOscar Salvador present_pages += pgdat->node_zones[zt].present_pages; 164086b27beaSOscar Salvador if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages) 164186b27beaSOscar Salvador arg->status_change_nid_normal = zone_to_nid(zone); 1642d9713679SLai Jiangshan 16436715ddf9SLai Jiangshan #ifdef CONFIG_HIGHMEM 16446715ddf9SLai Jiangshan /* 164586b27beaSOscar Salvador * node_states[N_HIGH_MEMORY] contains nodes which 164686b27beaSOscar Salvador * have normal memory or high memory. 164786b27beaSOscar Salvador * Here we add the present_pages belonging to ZONE_HIGHMEM. 164886b27beaSOscar Salvador * If the zone is within the range of [0..ZONE_HIGHMEM), and 164986b27beaSOscar Salvador * we determine that the zones in that range become empty, 165086b27beaSOscar Salvador * we need to clear the node for N_HIGH_MEMORY. 16516715ddf9SLai Jiangshan */ 165286b27beaSOscar Salvador present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages; 165386b27beaSOscar Salvador if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages) 16546715ddf9SLai Jiangshan arg->status_change_nid_high = zone_to_nid(zone); 16556715ddf9SLai Jiangshan #endif 16566715ddf9SLai Jiangshan 1657d9713679SLai Jiangshan /* 165886b27beaSOscar Salvador * We have accounted the pages from [0..ZONE_NORMAL), and 165986b27beaSOscar Salvador * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM 166086b27beaSOscar Salvador * as well. 166186b27beaSOscar Salvador * Here we count the possible pages from ZONE_MOVABLE. 166286b27beaSOscar Salvador * If after having accounted all the pages, we see that the nr_pages 166386b27beaSOscar Salvador * to be offlined is over or equal to the accounted pages, 166486b27beaSOscar Salvador * we know that the node will become empty, and so, we can clear 166586b27beaSOscar Salvador * it for N_MEMORY as well. 1666d9713679SLai Jiangshan */ 166786b27beaSOscar Salvador present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages; 1668d9713679SLai Jiangshan 1669d9713679SLai Jiangshan if (nr_pages >= present_pages) 1670d9713679SLai Jiangshan arg->status_change_nid = zone_to_nid(zone); 1671d9713679SLai Jiangshan } 1672d9713679SLai Jiangshan 1673d9713679SLai Jiangshan static void node_states_clear_node(int node, struct memory_notify *arg) 1674d9713679SLai Jiangshan { 1675d9713679SLai Jiangshan if (arg->status_change_nid_normal >= 0) 1676d9713679SLai Jiangshan node_clear_state(node, N_NORMAL_MEMORY); 1677d9713679SLai Jiangshan 1678cf01f6f5SOscar Salvador if (arg->status_change_nid_high >= 0) 1679d9713679SLai Jiangshan node_clear_state(node, N_HIGH_MEMORY); 16806715ddf9SLai Jiangshan 1681cf01f6f5SOscar Salvador if (arg->status_change_nid >= 0) 16826715ddf9SLai Jiangshan node_clear_state(node, N_MEMORY); 1683d9713679SLai Jiangshan } 1684d9713679SLai Jiangshan 1685c5e79ef5SDavid Hildenbrand static int count_system_ram_pages_cb(unsigned long start_pfn, 1686c5e79ef5SDavid Hildenbrand unsigned long nr_pages, void *data) 1687c5e79ef5SDavid Hildenbrand { 1688c5e79ef5SDavid Hildenbrand unsigned long *nr_system_ram_pages = data; 1689c5e79ef5SDavid Hildenbrand 1690c5e79ef5SDavid Hildenbrand *nr_system_ram_pages += nr_pages; 1691c5e79ef5SDavid Hildenbrand return 0; 1692c5e79ef5SDavid Hildenbrand } 1693c5e79ef5SDavid Hildenbrand 169473a11c96SDavid Hildenbrand int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages) 16950c0e6195SKAMEZAWA Hiroyuki { 169673a11c96SDavid Hildenbrand const unsigned long end_pfn = start_pfn + nr_pages; 16970a1a9a00SDavid Hildenbrand unsigned long pfn, system_ram_pages = 0; 1698d702909fSCody P Schafer unsigned long flags; 16990c0e6195SKAMEZAWA Hiroyuki struct zone *zone; 17007b78d335SYasunori Goto struct memory_notify arg; 1701ea15153cSDavid Hildenbrand int ret, node; 170279605093SMichal Hocko char *reason; 17030c0e6195SKAMEZAWA Hiroyuki 1704dd8e2f23SOscar Salvador /* 1705dd8e2f23SOscar Salvador * {on,off}lining is constrained to full memory sections (or more 1706dd8e2f23SOscar Salvador * precisly to memory blocks from the user space POV). 1707dd8e2f23SOscar Salvador * memmap_on_memory is an exception because it reserves initial part 1708dd8e2f23SOscar Salvador * of the physical memory space for vmemmaps. That space is pageblock 1709dd8e2f23SOscar Salvador * aligned. 1710dd8e2f23SOscar Salvador */ 17114986fac1SDavid Hildenbrand if (WARN_ON_ONCE(!nr_pages || 1712dd8e2f23SOscar Salvador !IS_ALIGNED(start_pfn, pageblock_nr_pages) || 1713dd8e2f23SOscar Salvador !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))) 17144986fac1SDavid Hildenbrand return -EINVAL; 17154986fac1SDavid Hildenbrand 1716381eab4aSDavid Hildenbrand mem_hotplug_begin(); 1717381eab4aSDavid Hildenbrand 1718c5e79ef5SDavid Hildenbrand /* 1719c5e79ef5SDavid Hildenbrand * Don't allow to offline memory blocks that contain holes. 1720c5e79ef5SDavid Hildenbrand * Consequently, memory blocks with holes can never get onlined 1721c5e79ef5SDavid Hildenbrand * via the hotplug path - online_pages() - as hotplugged memory has 1722c5e79ef5SDavid Hildenbrand * no holes. This way, we e.g., don't have to worry about marking 1723c5e79ef5SDavid Hildenbrand * memory holes PG_reserved, don't need pfn_valid() checks, and can 1724c5e79ef5SDavid Hildenbrand * avoid using walk_system_ram_range() later. 1725c5e79ef5SDavid Hildenbrand */ 172673a11c96SDavid Hildenbrand walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages, 1727c5e79ef5SDavid Hildenbrand count_system_ram_pages_cb); 172873a11c96SDavid Hildenbrand if (system_ram_pages != nr_pages) { 1729c5e79ef5SDavid Hildenbrand ret = -EINVAL; 1730c5e79ef5SDavid Hildenbrand reason = "memory holes"; 1731c5e79ef5SDavid Hildenbrand goto failed_removal; 1732c5e79ef5SDavid Hildenbrand } 1733c5e79ef5SDavid Hildenbrand 17340c0e6195SKAMEZAWA Hiroyuki /* This makes hotplug much easier...and readable. 17350c0e6195SKAMEZAWA Hiroyuki we assume this for now. .*/ 173692917998SDavid Hildenbrand zone = test_pages_in_a_zone(start_pfn, end_pfn); 173792917998SDavid Hildenbrand if (!zone) { 173879605093SMichal Hocko ret = -EINVAL; 173979605093SMichal Hocko reason = "multizone range"; 174079605093SMichal Hocko goto failed_removal; 1741381eab4aSDavid Hildenbrand } 17427b78d335SYasunori Goto node = zone_to_nid(zone); 17437b78d335SYasunori Goto 1744ec6e8c7eSVlastimil Babka /* 1745ec6e8c7eSVlastimil Babka * Disable pcplists so that page isolation cannot race with freeing 1746ec6e8c7eSVlastimil Babka * in a way that pages from isolated pageblock are left on pcplists. 1747ec6e8c7eSVlastimil Babka */ 1748ec6e8c7eSVlastimil Babka zone_pcp_disable(zone); 1749d479960eSMinchan Kim lru_cache_disable(); 1750ec6e8c7eSVlastimil Babka 17510c0e6195SKAMEZAWA Hiroyuki /* set above range as isolated */ 1752b023f468SWen Congyang ret = start_isolate_page_range(start_pfn, end_pfn, 1753d381c547SMichal Hocko MIGRATE_MOVABLE, 1754756d25beSDavid Hildenbrand MEMORY_OFFLINE | REPORT_FAILURE); 17553fa0c7c7SDavid Hildenbrand if (ret) { 175679605093SMichal Hocko reason = "failure to isolate range"; 1757ec6e8c7eSVlastimil Babka goto failed_removal_pcplists_disabled; 1758381eab4aSDavid Hildenbrand } 17597b78d335SYasunori Goto 17607b78d335SYasunori Goto arg.start_pfn = start_pfn; 17617b78d335SYasunori Goto arg.nr_pages = nr_pages; 1762d9713679SLai Jiangshan node_states_check_changes_offline(nr_pages, zone, &arg); 17637b78d335SYasunori Goto 17647b78d335SYasunori Goto ret = memory_notify(MEM_GOING_OFFLINE, &arg); 17657b78d335SYasunori Goto ret = notifier_to_errno(ret); 176679605093SMichal Hocko if (ret) { 176779605093SMichal Hocko reason = "notifier failure"; 176879605093SMichal Hocko goto failed_removal_isolated; 176979605093SMichal Hocko } 17707b78d335SYasunori Goto 1771bb8965bdSMichal Hocko do { 1772aa218795SDavid Hildenbrand pfn = start_pfn; 1773aa218795SDavid Hildenbrand do { 177479605093SMichal Hocko if (signal_pending(current)) { 1775bb8965bdSMichal Hocko ret = -EINTR; 177679605093SMichal Hocko reason = "signal backoff"; 177779605093SMichal Hocko goto failed_removal_isolated; 177879605093SMichal Hocko } 177972b39cfcSMichal Hocko 17800c0e6195SKAMEZAWA Hiroyuki cond_resched(); 17810c0e6195SKAMEZAWA Hiroyuki 1782aa218795SDavid Hildenbrand ret = scan_movable_pages(pfn, end_pfn, &pfn); 1783aa218795SDavid Hildenbrand if (!ret) { 1784bb8965bdSMichal Hocko /* 1785bb8965bdSMichal Hocko * TODO: fatal migration failures should bail 1786bb8965bdSMichal Hocko * out 1787bb8965bdSMichal Hocko */ 1788bb8965bdSMichal Hocko do_migrate_range(pfn, end_pfn); 1789bb8965bdSMichal Hocko } 1790aa218795SDavid Hildenbrand } while (!ret); 1791aa218795SDavid Hildenbrand 1792aa218795SDavid Hildenbrand if (ret != -ENOENT) { 1793aa218795SDavid Hildenbrand reason = "unmovable page"; 1794aa218795SDavid Hildenbrand goto failed_removal_isolated; 17950c0e6195SKAMEZAWA Hiroyuki } 179672b39cfcSMichal Hocko 1797c8721bbbSNaoya Horiguchi /* 1798bb8965bdSMichal Hocko * Dissolve free hugepages in the memory block before doing 1799bb8965bdSMichal Hocko * offlining actually in order to make hugetlbfs's object 1800bb8965bdSMichal Hocko * counting consistent. 1801c8721bbbSNaoya Horiguchi */ 1802082d5b6bSGerald Schaefer ret = dissolve_free_huge_pages(start_pfn, end_pfn); 180379605093SMichal Hocko if (ret) { 180479605093SMichal Hocko reason = "failure to dissolve huge pages"; 180579605093SMichal Hocko goto failed_removal_isolated; 180679605093SMichal Hocko } 18070a1a9a00SDavid Hildenbrand 18080a1a9a00SDavid Hildenbrand ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE); 1809ec6e8c7eSVlastimil Babka 18105557c766SMichal Hocko } while (ret); 1811bb8965bdSMichal Hocko 18120a1a9a00SDavid Hildenbrand /* Mark all sections offline and remove free pages from the buddy. */ 18130a1a9a00SDavid Hildenbrand __offline_isolated_pages(start_pfn, end_pfn); 18147c33023aSLaurent Dufour pr_debug("Offlined Pages %ld\n", nr_pages); 18150a1a9a00SDavid Hildenbrand 18169b7ea46aSQian Cai /* 1817b30c5927SDavid Hildenbrand * The memory sections are marked offline, and the pageblock flags 1818b30c5927SDavid Hildenbrand * effectively stale; nobody should be touching them. Fixup the number 1819b30c5927SDavid Hildenbrand * of isolated pageblocks, memory onlining will properly revert this. 18209b7ea46aSQian Cai */ 18219b7ea46aSQian Cai spin_lock_irqsave(&zone->lock, flags); 1822ea15153cSDavid Hildenbrand zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages; 18239b7ea46aSQian Cai spin_unlock_irqrestore(&zone->lock, flags); 18249b7ea46aSQian Cai 1825d479960eSMinchan Kim lru_cache_enable(); 1826ec6e8c7eSVlastimil Babka zone_pcp_enable(zone); 1827ec6e8c7eSVlastimil Babka 18280c0e6195SKAMEZAWA Hiroyuki /* removal success */ 18290a1a9a00SDavid Hildenbrand adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages); 1830f9901144SDavid Hildenbrand adjust_present_page_count(zone, -nr_pages); 18317b78d335SYasunori Goto 1832*b92ca18eSMel Gorman /* reinitialise watermarks and update pcp limits */ 18331b79acc9SKOSAKI Motohiro init_per_zone_wmark_min(); 18341b79acc9SKOSAKI Motohiro 18351e8537baSXishi Qiu if (!populated_zone(zone)) { 1836340175b7SJiang Liu zone_pcp_reset(zone); 183772675e13SMichal Hocko build_all_zonelists(NULL); 1838*b92ca18eSMel Gorman } 1839340175b7SJiang Liu 1840d9713679SLai Jiangshan node_states_clear_node(node, &arg); 1841698b1b30SVlastimil Babka if (arg.status_change_nid >= 0) { 18428fe23e05SDavid Rientjes kswapd_stop(node); 1843698b1b30SVlastimil Babka kcompactd_stop(node); 1844698b1b30SVlastimil Babka } 1845bce7394aSMinchan Kim 18460c0e6195SKAMEZAWA Hiroyuki writeback_set_ratelimit(); 18477b78d335SYasunori Goto 18487b78d335SYasunori Goto memory_notify(MEM_OFFLINE, &arg); 1849feee6b29SDavid Hildenbrand remove_pfn_range_from_zone(zone, start_pfn, nr_pages); 1850381eab4aSDavid Hildenbrand mem_hotplug_done(); 18510c0e6195SKAMEZAWA Hiroyuki return 0; 18520c0e6195SKAMEZAWA Hiroyuki 185379605093SMichal Hocko failed_removal_isolated: 185479605093SMichal Hocko undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); 1855c4efe484SQian Cai memory_notify(MEM_CANCEL_OFFLINE, &arg); 1856ec6e8c7eSVlastimil Babka failed_removal_pcplists_disabled: 1857ec6e8c7eSVlastimil Babka zone_pcp_enable(zone); 18580c0e6195SKAMEZAWA Hiroyuki failed_removal: 185979605093SMichal Hocko pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n", 1860a62e2f4fSBjorn Helgaas (unsigned long long) start_pfn << PAGE_SHIFT, 186179605093SMichal Hocko ((unsigned long long) end_pfn << PAGE_SHIFT) - 1, 186279605093SMichal Hocko reason); 18630c0e6195SKAMEZAWA Hiroyuki /* pushback to free area */ 1864381eab4aSDavid Hildenbrand mem_hotplug_done(); 18650c0e6195SKAMEZAWA Hiroyuki return ret; 18660c0e6195SKAMEZAWA Hiroyuki } 186771088785SBadari Pulavarty 1868d6de9d53SXishi Qiu static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) 1869bbc76be6SWen Congyang { 1870bbc76be6SWen Congyang int ret = !is_memblock_offlined(mem); 1871bbc76be6SWen Congyang 1872349daa0fSRandy Dunlap if (unlikely(ret)) { 1873349daa0fSRandy Dunlap phys_addr_t beginpa, endpa; 1874349daa0fSRandy Dunlap 1875349daa0fSRandy Dunlap beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); 1876b6c88d3bSDavid Hildenbrand endpa = beginpa + memory_block_size_bytes() - 1; 1877756a025fSJoe Perches pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", 1878349daa0fSRandy Dunlap &beginpa, &endpa); 1879bbc76be6SWen Congyang 1880eca499abSPavel Tatashin return -EBUSY; 1881eca499abSPavel Tatashin } 1882eca499abSPavel Tatashin return 0; 1883bbc76be6SWen Congyang } 1884bbc76be6SWen Congyang 1885a08a2ae3SOscar Salvador static int get_nr_vmemmap_pages_cb(struct memory_block *mem, void *arg) 1886a08a2ae3SOscar Salvador { 1887a08a2ae3SOscar Salvador /* 1888a08a2ae3SOscar Salvador * If not set, continue with the next block. 1889a08a2ae3SOscar Salvador */ 1890a08a2ae3SOscar Salvador return mem->nr_vmemmap_pages; 1891a08a2ae3SOscar Salvador } 1892a08a2ae3SOscar Salvador 18930f1cfe9dSToshi Kani static int check_cpu_on_node(pg_data_t *pgdat) 189460a5a19eSTang Chen { 189560a5a19eSTang Chen int cpu; 189660a5a19eSTang Chen 189760a5a19eSTang Chen for_each_present_cpu(cpu) { 189860a5a19eSTang Chen if (cpu_to_node(cpu) == pgdat->node_id) 189960a5a19eSTang Chen /* 190060a5a19eSTang Chen * the cpu on this node isn't removed, and we can't 190160a5a19eSTang Chen * offline this node. 190260a5a19eSTang Chen */ 190360a5a19eSTang Chen return -EBUSY; 190460a5a19eSTang Chen } 190560a5a19eSTang Chen 190660a5a19eSTang Chen return 0; 190760a5a19eSTang Chen } 190860a5a19eSTang Chen 19092c91f8fcSDavid Hildenbrand static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg) 19102c91f8fcSDavid Hildenbrand { 19112c91f8fcSDavid Hildenbrand int nid = *(int *)arg; 19122c91f8fcSDavid Hildenbrand 19132c91f8fcSDavid Hildenbrand /* 19142c91f8fcSDavid Hildenbrand * If a memory block belongs to multiple nodes, the stored nid is not 19152c91f8fcSDavid Hildenbrand * reliable. However, such blocks are always online (e.g., cannot get 19162c91f8fcSDavid Hildenbrand * offlined) and, therefore, are still spanned by the node. 19172c91f8fcSDavid Hildenbrand */ 19182c91f8fcSDavid Hildenbrand return mem->nid == nid ? -EEXIST : 0; 19192c91f8fcSDavid Hildenbrand } 19202c91f8fcSDavid Hildenbrand 19210f1cfe9dSToshi Kani /** 19220f1cfe9dSToshi Kani * try_offline_node 1923e8b098fcSMike Rapoport * @nid: the node ID 19240f1cfe9dSToshi Kani * 19250f1cfe9dSToshi Kani * Offline a node if all memory sections and cpus of the node are removed. 19260f1cfe9dSToshi Kani * 19270f1cfe9dSToshi Kani * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 19280f1cfe9dSToshi Kani * and online/offline operations before this call. 19290f1cfe9dSToshi Kani */ 193090b30cdcSWen Congyang void try_offline_node(int nid) 193160a5a19eSTang Chen { 1932d822b86aSWen Congyang pg_data_t *pgdat = NODE_DATA(nid); 19332c91f8fcSDavid Hildenbrand int rc; 193460a5a19eSTang Chen 193560a5a19eSTang Chen /* 19362c91f8fcSDavid Hildenbrand * If the node still spans pages (especially ZONE_DEVICE), don't 19372c91f8fcSDavid Hildenbrand * offline it. A node spans memory after move_pfn_range_to_zone(), 19382c91f8fcSDavid Hildenbrand * e.g., after the memory block was onlined. 193960a5a19eSTang Chen */ 19402c91f8fcSDavid Hildenbrand if (pgdat->node_spanned_pages) 194160a5a19eSTang Chen return; 19422c91f8fcSDavid Hildenbrand 19432c91f8fcSDavid Hildenbrand /* 19442c91f8fcSDavid Hildenbrand * Especially offline memory blocks might not be spanned by the 19452c91f8fcSDavid Hildenbrand * node. They will get spanned by the node once they get onlined. 19462c91f8fcSDavid Hildenbrand * However, they link to the node in sysfs and can get onlined later. 19472c91f8fcSDavid Hildenbrand */ 19482c91f8fcSDavid Hildenbrand rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb); 19492c91f8fcSDavid Hildenbrand if (rc) 19502c91f8fcSDavid Hildenbrand return; 195160a5a19eSTang Chen 195246a3679bSMichal Hocko if (check_cpu_on_node(pgdat)) 195360a5a19eSTang Chen return; 195460a5a19eSTang Chen 195560a5a19eSTang Chen /* 195660a5a19eSTang Chen * all memory/cpu of this node are removed, we can offline this 195760a5a19eSTang Chen * node now. 195860a5a19eSTang Chen */ 195960a5a19eSTang Chen node_set_offline(nid); 196060a5a19eSTang Chen unregister_one_node(nid); 196160a5a19eSTang Chen } 196290b30cdcSWen Congyang EXPORT_SYMBOL(try_offline_node); 196360a5a19eSTang Chen 1964eca499abSPavel Tatashin static int __ref try_remove_memory(int nid, u64 start, u64 size) 1965bbc76be6SWen Congyang { 1966eca499abSPavel Tatashin int rc = 0; 1967a08a2ae3SOscar Salvador struct vmem_altmap mhp_altmap = {}; 1968a08a2ae3SOscar Salvador struct vmem_altmap *altmap = NULL; 1969a08a2ae3SOscar Salvador unsigned long nr_vmemmap_pages; 1970993c1aadSWen Congyang 197127356f54SToshi Kani BUG_ON(check_hotplug_memory_range(start, size)); 197227356f54SToshi Kani 19736677e3eaSYasuaki Ishimatsu /* 1974242831ebSRafael J. Wysocki * All memory blocks must be offlined before removing memory. Check 1975eca499abSPavel Tatashin * whether all memory blocks in question are offline and return error 1976242831ebSRafael J. Wysocki * if this is not the case. 19776677e3eaSYasuaki Ishimatsu */ 1978fbcf73ceSDavid Hildenbrand rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb); 1979eca499abSPavel Tatashin if (rc) 1980b4223a51SJia He return rc; 19816677e3eaSYasuaki Ishimatsu 1982a08a2ae3SOscar Salvador /* 1983a08a2ae3SOscar Salvador * We only support removing memory added with MHP_MEMMAP_ON_MEMORY in 1984a08a2ae3SOscar Salvador * the same granularity it was added - a single memory block. 1985a08a2ae3SOscar Salvador */ 1986a08a2ae3SOscar Salvador if (memmap_on_memory) { 1987a08a2ae3SOscar Salvador nr_vmemmap_pages = walk_memory_blocks(start, size, NULL, 1988a08a2ae3SOscar Salvador get_nr_vmemmap_pages_cb); 1989a08a2ae3SOscar Salvador if (nr_vmemmap_pages) { 1990a08a2ae3SOscar Salvador if (size != memory_block_size_bytes()) { 1991a08a2ae3SOscar Salvador pr_warn("Refuse to remove %#llx - %#llx," 1992a08a2ae3SOscar Salvador "wrong granularity\n", 1993a08a2ae3SOscar Salvador start, start + size); 1994a08a2ae3SOscar Salvador return -EINVAL; 1995a08a2ae3SOscar Salvador } 1996a08a2ae3SOscar Salvador 1997a08a2ae3SOscar Salvador /* 1998a08a2ae3SOscar Salvador * Let remove_pmd_table->free_hugepage_table do the 1999a08a2ae3SOscar Salvador * right thing if we used vmem_altmap when hot-adding 2000a08a2ae3SOscar Salvador * the range. 2001a08a2ae3SOscar Salvador */ 2002a08a2ae3SOscar Salvador mhp_altmap.alloc = nr_vmemmap_pages; 2003a08a2ae3SOscar Salvador altmap = &mhp_altmap; 2004a08a2ae3SOscar Salvador } 2005a08a2ae3SOscar Salvador } 2006a08a2ae3SOscar Salvador 200746c66c4bSYasuaki Ishimatsu /* remove memmap entry */ 200846c66c4bSYasuaki Ishimatsu firmware_map_remove(start, start + size, "System RAM"); 200946c66c4bSYasuaki Ishimatsu 2010f1037ec0SDan Williams /* 2011f1037ec0SDan Williams * Memory block device removal under the device_hotplug_lock is 2012f1037ec0SDan Williams * a barrier against racing online attempts. 2013f1037ec0SDan Williams */ 20144c4b7f9bSDavid Hildenbrand remove_memory_block_devices(start, size); 20154c4b7f9bSDavid Hildenbrand 2016f1037ec0SDan Williams mem_hotplug_begin(); 2017f1037ec0SDan Williams 2018a08a2ae3SOscar Salvador arch_remove_memory(nid, start, size, altmap); 201952219aeaSDavid Hildenbrand 202052219aeaSDavid Hildenbrand if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) { 202132d1fe8fSAnshuman Khandual memblock_free(start, size); 202232d1fe8fSAnshuman Khandual memblock_remove(start, size); 202352219aeaSDavid Hildenbrand } 202452219aeaSDavid Hildenbrand 2025cb8e3c8bSDavid Hildenbrand release_mem_region_adjustable(start, size); 202624d335caSWen Congyang 202760a5a19eSTang Chen try_offline_node(nid); 202860a5a19eSTang Chen 2029bfc8c901SVladimir Davydov mem_hotplug_done(); 2030b4223a51SJia He return 0; 203171088785SBadari Pulavarty } 2032d15e5926SDavid Hildenbrand 2033eca499abSPavel Tatashin /** 2034eca499abSPavel Tatashin * remove_memory 2035eca499abSPavel Tatashin * @nid: the node ID 2036eca499abSPavel Tatashin * @start: physical address of the region to remove 2037eca499abSPavel Tatashin * @size: size of the region to remove 2038eca499abSPavel Tatashin * 2039eca499abSPavel Tatashin * NOTE: The caller must call lock_device_hotplug() to serialize hotplug 2040eca499abSPavel Tatashin * and online/offline operations before this call, as required by 2041eca499abSPavel Tatashin * try_offline_node(). 2042eca499abSPavel Tatashin */ 2043eca499abSPavel Tatashin void __remove_memory(int nid, u64 start, u64 size) 2044d15e5926SDavid Hildenbrand { 2045eca499abSPavel Tatashin 2046eca499abSPavel Tatashin /* 204729a90db9SSouptick Joarder * trigger BUG() if some memory is not offlined prior to calling this 2048eca499abSPavel Tatashin * function 2049eca499abSPavel Tatashin */ 2050eca499abSPavel Tatashin if (try_remove_memory(nid, start, size)) 2051eca499abSPavel Tatashin BUG(); 2052eca499abSPavel Tatashin } 2053eca499abSPavel Tatashin 2054eca499abSPavel Tatashin /* 2055eca499abSPavel Tatashin * Remove memory if every memory block is offline, otherwise return -EBUSY is 2056eca499abSPavel Tatashin * some memory is not offline 2057eca499abSPavel Tatashin */ 2058eca499abSPavel Tatashin int remove_memory(int nid, u64 start, u64 size) 2059eca499abSPavel Tatashin { 2060eca499abSPavel Tatashin int rc; 2061eca499abSPavel Tatashin 2062d15e5926SDavid Hildenbrand lock_device_hotplug(); 2063eca499abSPavel Tatashin rc = try_remove_memory(nid, start, size); 2064d15e5926SDavid Hildenbrand unlock_device_hotplug(); 2065eca499abSPavel Tatashin 2066eca499abSPavel Tatashin return rc; 2067d15e5926SDavid Hildenbrand } 206871088785SBadari Pulavarty EXPORT_SYMBOL_GPL(remove_memory); 206908b3acd7SDavid Hildenbrand 20708dc4bb58SDavid Hildenbrand static int try_offline_memory_block(struct memory_block *mem, void *arg) 20718dc4bb58SDavid Hildenbrand { 20728dc4bb58SDavid Hildenbrand uint8_t online_type = MMOP_ONLINE_KERNEL; 20738dc4bb58SDavid Hildenbrand uint8_t **online_types = arg; 20748dc4bb58SDavid Hildenbrand struct page *page; 20758dc4bb58SDavid Hildenbrand int rc; 20768dc4bb58SDavid Hildenbrand 207708b3acd7SDavid Hildenbrand /* 20788dc4bb58SDavid Hildenbrand * Sense the online_type via the zone of the memory block. Offlining 20798dc4bb58SDavid Hildenbrand * with multiple zones within one memory block will be rejected 20808dc4bb58SDavid Hildenbrand * by offlining code ... so we don't care about that. 20818dc4bb58SDavid Hildenbrand */ 20828dc4bb58SDavid Hildenbrand page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr)); 20838dc4bb58SDavid Hildenbrand if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE) 20848dc4bb58SDavid Hildenbrand online_type = MMOP_ONLINE_MOVABLE; 20858dc4bb58SDavid Hildenbrand 20868dc4bb58SDavid Hildenbrand rc = device_offline(&mem->dev); 20878dc4bb58SDavid Hildenbrand /* 20888dc4bb58SDavid Hildenbrand * Default is MMOP_OFFLINE - change it only if offlining succeeded, 20898dc4bb58SDavid Hildenbrand * so try_reonline_memory_block() can do the right thing. 20908dc4bb58SDavid Hildenbrand */ 20918dc4bb58SDavid Hildenbrand if (!rc) 20928dc4bb58SDavid Hildenbrand **online_types = online_type; 20938dc4bb58SDavid Hildenbrand 20948dc4bb58SDavid Hildenbrand (*online_types)++; 20958dc4bb58SDavid Hildenbrand /* Ignore if already offline. */ 20968dc4bb58SDavid Hildenbrand return rc < 0 ? rc : 0; 20978dc4bb58SDavid Hildenbrand } 20988dc4bb58SDavid Hildenbrand 20998dc4bb58SDavid Hildenbrand static int try_reonline_memory_block(struct memory_block *mem, void *arg) 21008dc4bb58SDavid Hildenbrand { 21018dc4bb58SDavid Hildenbrand uint8_t **online_types = arg; 21028dc4bb58SDavid Hildenbrand int rc; 21038dc4bb58SDavid Hildenbrand 21048dc4bb58SDavid Hildenbrand if (**online_types != MMOP_OFFLINE) { 21058dc4bb58SDavid Hildenbrand mem->online_type = **online_types; 21068dc4bb58SDavid Hildenbrand rc = device_online(&mem->dev); 21078dc4bb58SDavid Hildenbrand if (rc < 0) 21088dc4bb58SDavid Hildenbrand pr_warn("%s: Failed to re-online memory: %d", 21098dc4bb58SDavid Hildenbrand __func__, rc); 21108dc4bb58SDavid Hildenbrand } 21118dc4bb58SDavid Hildenbrand 21128dc4bb58SDavid Hildenbrand /* Continue processing all remaining memory blocks. */ 21138dc4bb58SDavid Hildenbrand (*online_types)++; 21148dc4bb58SDavid Hildenbrand return 0; 21158dc4bb58SDavid Hildenbrand } 21168dc4bb58SDavid Hildenbrand 21178dc4bb58SDavid Hildenbrand /* 21188dc4bb58SDavid Hildenbrand * Try to offline and remove memory. Might take a long time to finish in case 21198dc4bb58SDavid Hildenbrand * memory is still in use. Primarily useful for memory devices that logically 21208dc4bb58SDavid Hildenbrand * unplugged all memory (so it's no longer in use) and want to offline + remove 21218dc4bb58SDavid Hildenbrand * that memory. 212208b3acd7SDavid Hildenbrand */ 212308b3acd7SDavid Hildenbrand int offline_and_remove_memory(int nid, u64 start, u64 size) 212408b3acd7SDavid Hildenbrand { 21258dc4bb58SDavid Hildenbrand const unsigned long mb_count = size / memory_block_size_bytes(); 21268dc4bb58SDavid Hildenbrand uint8_t *online_types, *tmp; 21278dc4bb58SDavid Hildenbrand int rc; 212808b3acd7SDavid Hildenbrand 212908b3acd7SDavid Hildenbrand if (!IS_ALIGNED(start, memory_block_size_bytes()) || 21308dc4bb58SDavid Hildenbrand !IS_ALIGNED(size, memory_block_size_bytes()) || !size) 21318dc4bb58SDavid Hildenbrand return -EINVAL; 213208b3acd7SDavid Hildenbrand 213308b3acd7SDavid Hildenbrand /* 21348dc4bb58SDavid Hildenbrand * We'll remember the old online type of each memory block, so we can 21358dc4bb58SDavid Hildenbrand * try to revert whatever we did when offlining one memory block fails 21368dc4bb58SDavid Hildenbrand * after offlining some others succeeded. 21378dc4bb58SDavid Hildenbrand */ 21388dc4bb58SDavid Hildenbrand online_types = kmalloc_array(mb_count, sizeof(*online_types), 21398dc4bb58SDavid Hildenbrand GFP_KERNEL); 21408dc4bb58SDavid Hildenbrand if (!online_types) 21418dc4bb58SDavid Hildenbrand return -ENOMEM; 21428dc4bb58SDavid Hildenbrand /* 21438dc4bb58SDavid Hildenbrand * Initialize all states to MMOP_OFFLINE, so when we abort processing in 21448dc4bb58SDavid Hildenbrand * try_offline_memory_block(), we'll skip all unprocessed blocks in 21458dc4bb58SDavid Hildenbrand * try_reonline_memory_block(). 21468dc4bb58SDavid Hildenbrand */ 21478dc4bb58SDavid Hildenbrand memset(online_types, MMOP_OFFLINE, mb_count); 21488dc4bb58SDavid Hildenbrand 21498dc4bb58SDavid Hildenbrand lock_device_hotplug(); 21508dc4bb58SDavid Hildenbrand 21518dc4bb58SDavid Hildenbrand tmp = online_types; 21528dc4bb58SDavid Hildenbrand rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block); 21538dc4bb58SDavid Hildenbrand 21548dc4bb58SDavid Hildenbrand /* 21558dc4bb58SDavid Hildenbrand * In case we succeeded to offline all memory, remove it. 215608b3acd7SDavid Hildenbrand * This cannot fail as it cannot get onlined in the meantime. 215708b3acd7SDavid Hildenbrand */ 215808b3acd7SDavid Hildenbrand if (!rc) { 215908b3acd7SDavid Hildenbrand rc = try_remove_memory(nid, start, size); 21608dc4bb58SDavid Hildenbrand if (rc) 21618dc4bb58SDavid Hildenbrand pr_err("%s: Failed to remove memory: %d", __func__, rc); 21628dc4bb58SDavid Hildenbrand } 21638dc4bb58SDavid Hildenbrand 21648dc4bb58SDavid Hildenbrand /* 21658dc4bb58SDavid Hildenbrand * Rollback what we did. While memory onlining might theoretically fail 21668dc4bb58SDavid Hildenbrand * (nacked by a notifier), it barely ever happens. 21678dc4bb58SDavid Hildenbrand */ 21688dc4bb58SDavid Hildenbrand if (rc) { 21698dc4bb58SDavid Hildenbrand tmp = online_types; 21708dc4bb58SDavid Hildenbrand walk_memory_blocks(start, size, &tmp, 21718dc4bb58SDavid Hildenbrand try_reonline_memory_block); 217208b3acd7SDavid Hildenbrand } 217308b3acd7SDavid Hildenbrand unlock_device_hotplug(); 217408b3acd7SDavid Hildenbrand 21758dc4bb58SDavid Hildenbrand kfree(online_types); 217608b3acd7SDavid Hildenbrand return rc; 217708b3acd7SDavid Hildenbrand } 217808b3acd7SDavid Hildenbrand EXPORT_SYMBOL_GPL(offline_and_remove_memory); 2179aba6efc4SRafael J. Wysocki #endif /* CONFIG_MEMORY_HOTREMOVE */ 2180