blob: c2878d48b8b170ca15fba90e2045e539b5c68f5f [file] [log] [blame]
Dave Hansen3947be12005-10-29 18:16:54 -07001/*
2 * linux/mm/memory_hotplug.c
3 *
4 * Copyright (C)
5 */
6
Dave Hansen3947be12005-10-29 18:16:54 -07007#include <linux/stddef.h>
8#include <linux/mm.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +01009#include <linux/sched/signal.h>
Dave Hansen3947be12005-10-29 18:16:54 -070010#include <linux/swap.h>
11#include <linux/interrupt.h>
12#include <linux/pagemap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070013#include <linux/compiler.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040014#include <linux/export.h>
Dave Hansen3947be12005-10-29 18:16:54 -070015#include <linux/pagevec.h>
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -070016#include <linux/writeback.h>
Dave Hansen3947be12005-10-29 18:16:54 -070017#include <linux/slab.h>
18#include <linux/sysctl.h>
19#include <linux/cpu.h>
20#include <linux/memory.h>
Dan Williams4b94ffd2016-01-15 16:56:22 -080021#include <linux/memremap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070022#include <linux/memory_hotplug.h>
23#include <linux/highmem.h>
24#include <linux/vmalloc.h>
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -070025#include <linux/ioport.h>
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -070026#include <linux/delay.h>
27#include <linux/migrate.h>
28#include <linux/page-isolation.h>
Badari Pulavarty71088782008-10-18 20:25:58 -070029#include <linux/pfn.h>
Andi Kleen6ad696d2009-11-17 14:06:22 -080030#include <linux/suspend.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080031#include <linux/mm_inline.h>
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -080032#include <linux/firmware-map.h>
Tang Chen60a5a192013-02-22 16:33:14 -080033#include <linux/stop_machine.h>
Naoya Horiguchic8721bb2013-09-11 14:22:09 -070034#include <linux/hugetlb.h>
Tang Chenc5320922013-11-12 15:08:10 -080035#include <linux/memblock.h>
Tang Chenf784a3f2014-11-13 15:19:39 -080036#include <linux/bootmem.h>
Vlastimil Babka698b1b32016-03-17 14:18:08 -070037#include <linux/compaction.h>
Michal Hocko2c250712018-12-28 00:38:01 -080038#include <linux/rmap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070039
40#include <asm/tlbflush.h>
41
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030042#include "internal.h"
43
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070044/*
45 * online_page_callback contains pointer to current page onlining function.
46 * Initially it is generic_online_page(). If it is required it could be
47 * changed by calling set_online_page_callback() for callback registration
48 * and restore_online_page_callback() for generic callback restore.
49 */
50
51static void generic_online_page(struct page *page);
52
53static online_page_callback_t online_page_callback = generic_online_page;
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070054static DEFINE_MUTEX(online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070055
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070056DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080057
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070058void get_online_mems(void)
59{
60 percpu_down_read(&mem_hotplug_lock);
61}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070062
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070063void put_online_mems(void)
64{
65 percpu_up_read(&mem_hotplug_lock);
66}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070067
Michal Hocko49323812017-07-06 15:41:05 -070068bool movable_node_enabled = false;
69
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070070#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070071bool memhp_auto_online;
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -070072#else
73bool memhp_auto_online = true;
74#endif
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -070075EXPORT_SYMBOL_GPL(memhp_auto_online);
76
Vitaly Kuznetsov86dd9952016-05-19 17:13:06 -070077static int __init setup_memhp_default_state(char *str)
78{
79 if (!strcmp(str, "online"))
80 memhp_auto_online = true;
81 else if (!strcmp(str, "offline"))
82 memhp_auto_online = false;
83
84 return 1;
85}
86__setup("memhp_default_state=", setup_memhp_default_state);
87
David Rientjes30467e02015-04-14 15:45:11 -070088void mem_hotplug_begin(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070089{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070090 cpus_read_lock();
91 percpu_down_write(&mem_hotplug_lock);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070092}
93
David Rientjes30467e02015-04-14 15:45:11 -070094void mem_hotplug_done(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070095{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -070096 percpu_up_write(&mem_hotplug_lock);
97 cpus_read_unlock();
Vladimir Davydovbfc8c902014-06-04 16:07:18 -070098}
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080099
Keith Mannthey45e0b782006-09-30 23:27:09 -0700100/* add this memory to iomem resource */
101static struct resource *register_memory_resource(u64 start, u64 size)
102{
Jérôme Glisse5042db42017-09-08 16:11:43 -0700103 struct resource *res, *conflict;
Keith Mannthey45e0b782006-09-30 23:27:09 -0700104 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800105 if (!res)
106 return ERR_PTR(-ENOMEM);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700107
108 res->name = "System RAM";
109 res->start = start;
110 res->end = start + size - 1;
Toshi Kani782b8662016-01-26 21:57:24 +0100111 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
Jérôme Glisse5042db42017-09-08 16:11:43 -0700112 conflict = request_resource_conflict(&iomem_resource, res);
113 if (conflict) {
114 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
115 pr_debug("Device unaddressable memory block "
116 "memory hotplug at %#010llx !\n",
117 (unsigned long long)start);
118 }
Toshi Kani4996eed2013-07-03 15:02:39 -0700119 pr_debug("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700120 kfree(res);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800121 return ERR_PTR(-EEXIST);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700122 }
123 return res;
124}
125
126static void release_memory_resource(struct resource *res)
127{
128 if (!res)
129 return;
130 release_resource(res);
131 kfree(res);
132 return;
133}
134
Keith Mannthey53947022006-09-30 23:27:08 -0700135#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800136void get_page_bootmem(unsigned long info, struct page *page,
137 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -0700138{
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800139 page->freelist = (void *)type;
Yasunori Goto04753272008-04-28 02:13:31 -0700140 SetPagePrivate(page);
141 set_page_private(page, info);
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700142 page_ref_inc(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700143}
144
Jiang Liu170a5a72013-07-03 15:03:17 -0700145void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700146{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800147 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700148
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800149 type = (unsigned long) page->freelist;
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800150 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
151 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700152
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700153 if (page_ref_dec_return(page) == 1) {
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800154 page->freelist = NULL;
Yasunori Goto04753272008-04-28 02:13:31 -0700155 ClearPagePrivate(page);
156 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800157 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700158 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700159 }
Yasunori Goto04753272008-04-28 02:13:31 -0700160}
161
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800162#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
163#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700164static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700165{
166 unsigned long *usemap, mapsize, section_nr, i;
167 struct mem_section *ms;
168 struct page *page, *memmap;
169
Yasunori Goto04753272008-04-28 02:13:31 -0700170 section_nr = pfn_to_section_nr(start_pfn);
171 ms = __nr_to_section(section_nr);
172
173 /* Get section's memmap address */
174 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
175
176 /*
177 * Get page for the memmap's phys address
178 * XXX: need more consideration for sparse_vmemmap...
179 */
180 page = virt_to_page(memmap);
181 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
182 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
183
184 /* remember memmap's page */
185 for (i = 0; i < mapsize; i++, page++)
186 get_page_bootmem(section_nr, page, SECTION_INFO);
187
188 usemap = __nr_to_section(section_nr)->pageblock_flags;
189 page = virt_to_page(usemap);
190
191 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
192
193 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700194 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700195
196}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800197#else /* CONFIG_SPARSEMEM_VMEMMAP */
198static void register_page_bootmem_info_section(unsigned long start_pfn)
199{
200 unsigned long *usemap, mapsize, section_nr, i;
201 struct mem_section *ms;
202 struct page *page, *memmap;
203
204 if (!pfn_valid(start_pfn))
205 return;
206
207 section_nr = pfn_to_section_nr(start_pfn);
208 ms = __nr_to_section(section_nr);
209
210 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
211
212 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
213
214 usemap = __nr_to_section(section_nr)->pageblock_flags;
215 page = virt_to_page(usemap);
216
217 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
218
219 for (i = 0; i < mapsize; i++, page++)
220 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
221}
222#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700223
Linus Torvalds7ded3842016-05-27 15:23:32 -0700224void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
Yasunori Goto04753272008-04-28 02:13:31 -0700225{
226 unsigned long i, pfn, end_pfn, nr_pages;
227 int node = pgdat->node_id;
228 struct page *page;
Yasunori Goto04753272008-04-28 02:13:31 -0700229
230 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
231 page = virt_to_page(pgdat);
232
233 for (i = 0; i < nr_pages; i++, page++)
234 get_page_bootmem(node, page, NODE_INFO);
235
Yasunori Goto04753272008-04-28 02:13:31 -0700236 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800237 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700238
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700239 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700240 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
241 /*
242 * Some platforms can assign the same pfn to multiple nodes - on
243 * node0 as well as nodeN. To avoid registering a pfn against
244 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700245 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700246 */
Yang Shif65e91df2016-05-27 14:27:32 -0700247 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
qiuxishif14851a2012-09-17 14:09:24 -0700248 register_page_bootmem_info_section(pfn);
249 }
Yasunori Goto04753272008-04-28 02:13:31 -0700250}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800251#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700252
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700253static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
254 bool want_memblock)
Dave Hansen3947be12005-10-29 18:16:54 -0700255{
Dave Hansen3947be12005-10-29 18:16:54 -0700256 int ret;
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700257 int i;
Dave Hansen3947be12005-10-29 18:16:54 -0700258
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700259 if (pfn_valid(phys_start_pfn))
260 return -EEXIST;
261
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700262 ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn);
Dave Hansen3947be12005-10-29 18:16:54 -0700263 if (ret < 0)
264 return ret;
265
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700266 /*
267 * Make all the pages reserved so that nobody will stumble over half
268 * initialized state.
269 * FIXME: We also have to associate it with a node because pfn_to_node
270 * relies on having page with the proper node.
271 */
272 for (i = 0; i < PAGES_PER_SECTION; i++) {
273 unsigned long pfn = phys_start_pfn + i;
274 struct page *page;
275 if (!pfn_valid(pfn))
276 continue;
Yasunori Goto718127c2006-06-23 02:03:10 -0700277
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700278 page = pfn_to_page(pfn);
279 set_page_node(page, nid);
280 SetPageReserved(page);
281 }
Yasunori Goto718127c2006-06-23 02:03:10 -0700282
Michal Hocko1b862ae2017-07-06 15:37:45 -0700283 if (!want_memblock)
284 return 0;
285
Gary Hadec04fc582009-01-06 14:39:14 -0800286 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700287}
288
David Rientjes4edd7ce2013-04-29 15:08:22 -0700289/*
290 * Reasonably generic function for adding memory. It is
291 * expected that archs that support memory hotplug will
292 * call this function after deciding the zone to which to
293 * add the new pages.
294 */
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700295int __ref __add_pages(int nid, unsigned long phys_start_pfn,
Michal Hocko1b862ae2017-07-06 15:37:45 -0700296 unsigned long nr_pages, bool want_memblock)
David Rientjes4edd7ce2013-04-29 15:08:22 -0700297{
298 unsigned long i;
299 int err = 0;
300 int start_sec, end_sec;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800301 struct vmem_altmap *altmap;
302
David Rientjes4edd7ce2013-04-29 15:08:22 -0700303 /* during initialize mem_map, align hot-added range to section */
304 start_sec = pfn_to_section_nr(phys_start_pfn);
305 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
306
Dan Williams4b94ffd2016-01-15 16:56:22 -0800307 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
308 if (altmap) {
309 /*
310 * Validate altmap is within bounds of the total request
311 */
312 if (altmap->base_pfn != phys_start_pfn
313 || vmem_altmap_offset(altmap) > nr_pages) {
314 pr_warn_once("memory add fail, invalid altmap\n");
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700315 err = -EINVAL;
316 goto out;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800317 }
318 altmap->alloc = 0;
319 }
320
David Rientjes4edd7ce2013-04-29 15:08:22 -0700321 for (i = start_sec; i <= end_sec; i++) {
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700322 err = __add_section(nid, section_nr_to_pfn(i), want_memblock);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700323
324 /*
325 * EEXIST is finally dealt with by ioresource collision
326 * check. see add_memory() => register_memory_resource()
327 * Warning will be printed if there is collision.
328 */
329 if (err && (err != -EEXIST))
330 break;
331 err = 0;
Michal Hockof64ac5e2017-10-03 16:16:16 -0700332 cond_resched();
David Rientjes4edd7ce2013-04-29 15:08:22 -0700333 }
Zhu Guihuac435a392015-06-24 16:58:42 -0700334 vmemmap_populate_print_last();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700335out:
David Rientjes4edd7ce2013-04-29 15:08:22 -0700336 return err;
337}
338EXPORT_SYMBOL_GPL(__add_pages);
339
340#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800341/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700342static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800343 unsigned long start_pfn,
344 unsigned long end_pfn)
345{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800346 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
David Hildenbrandb5ca5ac2019-11-21 17:53:56 -0800347 if (unlikely(!pfn_to_online_page(start_pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800348 continue;
349
350 if (unlikely(pfn_to_nid(start_pfn) != nid))
351 continue;
352
353 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
354 continue;
355
356 return start_pfn;
357 }
358
359 return 0;
360}
361
362/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700363static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800364 unsigned long start_pfn,
365 unsigned long end_pfn)
366{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800367 unsigned long pfn;
368
369 /* pfn is the end pfn of a memory section. */
370 pfn = end_pfn - 1;
371 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
David Hildenbrandb5ca5ac2019-11-21 17:53:56 -0800372 if (unlikely(!pfn_to_online_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800373 continue;
374
375 if (unlikely(pfn_to_nid(pfn) != nid))
376 continue;
377
378 if (zone && zone != page_zone(pfn_to_page(pfn)))
379 continue;
380
381 return pfn;
382 }
383
384 return 0;
385}
386
387static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
388 unsigned long end_pfn)
389{
Xishi Qiuc33bc312013-09-11 14:21:44 -0700390 unsigned long zone_start_pfn = zone->zone_start_pfn;
391 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
392 unsigned long zone_end_pfn = z;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800393 unsigned long pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800394 int nid = zone_to_nid(zone);
395
396 zone_span_writelock(zone);
397 if (zone_start_pfn == start_pfn) {
398 /*
399 * If the section is smallest section in the zone, it need
400 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
401 * In this case, we find second smallest valid mem_section
402 * for shrinking zone.
403 */
404 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
405 zone_end_pfn);
406 if (pfn) {
407 zone->zone_start_pfn = pfn;
408 zone->spanned_pages = zone_end_pfn - pfn;
409 }
410 } else if (zone_end_pfn == end_pfn) {
411 /*
412 * If the section is biggest section in the zone, it need
413 * shrink zone->spanned_pages.
414 * In this case, we find second biggest valid mem_section for
415 * shrinking zone.
416 */
417 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
418 start_pfn);
419 if (pfn)
420 zone->spanned_pages = pfn - zone_start_pfn + 1;
421 }
422
423 /*
424 * The section is not biggest or smallest mem_section in the zone, it
425 * only creates a hole in the zone. So in this case, we need not
426 * change the zone. But perhaps, the zone has only hole data. Thus
427 * it check the zone has only hole or not.
428 */
429 pfn = zone_start_pfn;
430 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
David Hildenbrandb5ca5ac2019-11-21 17:53:56 -0800431 if (unlikely(!pfn_to_online_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800432 continue;
433
434 if (page_zone(pfn_to_page(pfn)) != zone)
435 continue;
436
437 /* If the section is current section, it continues the loop */
438 if (start_pfn == pfn)
439 continue;
440
441 /* If we find valid section, we have nothing to do */
442 zone_span_writeunlock(zone);
443 return;
444 }
445
446 /* The zone has no valid section */
447 zone->zone_start_pfn = 0;
448 zone->spanned_pages = 0;
449 zone_span_writeunlock(zone);
450}
451
David Hildenbrandb55841e2019-10-18 20:19:33 -0700452static void update_pgdat_span(struct pglist_data *pgdat)
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800453{
David Hildenbrandb55841e2019-10-18 20:19:33 -0700454 unsigned long node_start_pfn = 0, node_end_pfn = 0;
455 struct zone *zone;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800456
David Hildenbrandb55841e2019-10-18 20:19:33 -0700457 for (zone = pgdat->node_zones;
458 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
459 unsigned long zone_end_pfn = zone->zone_start_pfn +
460 zone->spanned_pages;
461
462 /* No need to lock the zones, they can't change. */
David Hildenbrand630ca7d2019-11-05 21:17:10 -0800463 if (!zone->spanned_pages)
464 continue;
465 if (!node_end_pfn) {
466 node_start_pfn = zone->zone_start_pfn;
467 node_end_pfn = zone_end_pfn;
468 continue;
469 }
470
David Hildenbrandb55841e2019-10-18 20:19:33 -0700471 if (zone_end_pfn > node_end_pfn)
472 node_end_pfn = zone_end_pfn;
473 if (zone->zone_start_pfn < node_start_pfn)
474 node_start_pfn = zone->zone_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800475 }
476
David Hildenbrandb55841e2019-10-18 20:19:33 -0700477 pgdat->node_start_pfn = node_start_pfn;
478 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800479}
480
481static void __remove_zone(struct zone *zone, unsigned long start_pfn)
482{
483 struct pglist_data *pgdat = zone->zone_pgdat;
484 int nr_pages = PAGES_PER_SECTION;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800485 unsigned long flags;
486
David Hildenbrandb5ca5ac2019-11-21 17:53:56 -0800487#ifdef CONFIG_ZONE_DEVICE
488 /*
489 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
490 * we will not try to shrink the zones - which is okay as
491 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
492 */
493 if (zone_idx(zone) == ZONE_DEVICE)
494 return;
495#endif
496
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800497 pgdat_resize_lock(zone->zone_pgdat, &flags);
498 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
David Hildenbrandb55841e2019-10-18 20:19:33 -0700499 update_pgdat_span(pgdat);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800500 pgdat_resize_unlock(zone->zone_pgdat, &flags);
501}
502
Dan Williams4b94ffd2016-01-15 16:56:22 -0800503static int __remove_section(struct zone *zone, struct mem_section *ms,
504 unsigned long map_offset)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700505{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800506 unsigned long start_pfn;
507 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700508 int ret = -EINVAL;
509
510 if (!valid_section(ms))
511 return ret;
512
513 ret = unregister_memory_section(ms);
514 if (ret)
515 return ret;
516
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800517 scn_nr = __section_nr(ms);
YASUAKI ISHIMATSU1dd2bfc2017-10-03 16:16:29 -0700518 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800519 __remove_zone(zone, start_pfn);
520
Dan Williams4b94ffd2016-01-15 16:56:22 -0800521 sparse_remove_one_section(zone, ms, map_offset);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700522 return 0;
523}
524
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700525/**
526 * __remove_pages() - remove sections of pages from a zone
527 * @zone: zone from which pages need to be removed
528 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
529 * @nr_pages: number of pages to remove (must be multiple of section size)
530 *
531 * Generic helper function to remove section mappings and sysfs entries
532 * for the section of the memory we are removing. Caller needs to make
533 * sure that pages are marked reserved and zones are adjust properly by
534 * calling offline_pages().
535 */
536int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
537 unsigned long nr_pages)
538{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700539 unsigned long i;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800540 unsigned long map_offset = 0;
541 int sections_to_remove, ret = 0;
542
543 /* In the ZONE_DEVICE case device driver owns the memory region */
544 if (is_dev_zone(zone)) {
545 struct page *page = pfn_to_page(phys_start_pfn);
546 struct vmem_altmap *altmap;
547
548 altmap = to_vmem_altmap((unsigned long) page);
549 if (altmap)
550 map_offset = vmem_altmap_offset(altmap);
551 } else {
552 resource_size_t start, size;
553
554 start = phys_start_pfn << PAGE_SHIFT;
555 size = nr_pages * PAGE_SIZE;
556
557 ret = release_mem_region_adjustable(&iomem_resource, start,
558 size);
559 if (ret) {
560 resource_size_t endres = start + size - 1;
561
562 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
563 &start, &endres, ret);
564 }
565 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700566
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700567 clear_zone_contiguous(zone);
568
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700569 /*
570 * We can only remove entire sections
571 */
572 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
573 BUG_ON(nr_pages % PAGES_PER_SECTION);
574
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700575 sections_to_remove = nr_pages / PAGES_PER_SECTION;
576 for (i = 0; i < sections_to_remove; i++) {
577 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800578
579 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
580 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700581 if (ret)
582 break;
583 }
Joonsoo Kim7cf91a92016-03-15 14:57:51 -0700584
585 set_zone_contiguous(zone);
586
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700587 return ret;
588}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700589#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700590
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700591int set_online_page_callback(online_page_callback_t callback)
592{
593 int rc = -EINVAL;
594
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700595 get_online_mems();
596 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700597
598 if (online_page_callback == generic_online_page) {
599 online_page_callback = callback;
600 rc = 0;
601 }
602
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700603 mutex_unlock(&online_page_callback_lock);
604 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700605
606 return rc;
607}
608EXPORT_SYMBOL_GPL(set_online_page_callback);
609
610int restore_online_page_callback(online_page_callback_t callback)
611{
612 int rc = -EINVAL;
613
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700614 get_online_mems();
615 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700616
617 if (online_page_callback == callback) {
618 online_page_callback = generic_online_page;
619 rc = 0;
620 }
621
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700622 mutex_unlock(&online_page_callback_lock);
623 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700624
625 return rc;
626}
627EXPORT_SYMBOL_GPL(restore_online_page_callback);
628
629void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700630{
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700631}
632EXPORT_SYMBOL_GPL(__online_page_set_limits);
633
634void __online_page_increment_counters(struct page *page)
635{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700636 adjust_managed_page_count(page, 1);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700637}
638EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700639
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700640void __online_page_free(struct page *page)
641{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700642 __free_reserved_page(page);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700643}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700644EXPORT_SYMBOL_GPL(__online_page_free);
645
646static void generic_online_page(struct page *page)
647{
648 __online_page_set_limits(page);
649 __online_page_increment_counters(page);
650 __online_page_free(page);
651}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700652
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700653static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
654 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700655{
656 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700657 unsigned long onlined_pages = *(unsigned long *)arg;
658 struct page *page;
Michal Hocko2d070ea2017-07-06 15:37:56 -0700659
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700660 if (PageReserved(pfn_to_page(start_pfn)))
661 for (i = 0; i < nr_pages; i++) {
662 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700663 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700664 onlined_pages++;
665 }
Michal Hocko2d070ea2017-07-06 15:37:56 -0700666
667 online_mem_sections(start_pfn, start_pfn + nr_pages);
668
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700669 *(unsigned long *)arg = onlined_pages;
670 return 0;
671}
672
Lai Jiangshand9713672012-12-11 16:01:03 -0800673/* check which state of node_states will be changed when online memory */
674static void node_states_check_changes_online(unsigned long nr_pages,
675 struct zone *zone, struct memory_notify *arg)
676{
677 int nid = zone_to_nid(zone);
678 enum zone_type zone_last = ZONE_NORMAL;
679
680 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800681 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
682 * contains nodes which have zones of 0...ZONE_NORMAL,
683 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -0800684 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800685 * If we don't have HIGHMEM nor movable node,
686 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
687 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -0800688 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800689 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -0800690 zone_last = ZONE_MOVABLE;
691
692 /*
693 * if the memory to be online is in a zone of 0...zone_last, and
694 * the zones of 0...zone_last don't have memory before online, we will
695 * need to set the node to node_states[N_NORMAL_MEMORY] after
696 * the memory is online.
697 */
698 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
699 arg->status_change_nid_normal = nid;
700 else
701 arg->status_change_nid_normal = -1;
702
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800703#ifdef CONFIG_HIGHMEM
704 /*
705 * If we have movable node, node_states[N_HIGH_MEMORY]
706 * contains nodes which have zones of 0...ZONE_HIGHMEM,
707 * set zone_last to ZONE_HIGHMEM.
708 *
709 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
710 * contains nodes which have zones of 0...ZONE_MOVABLE,
711 * set zone_last to ZONE_MOVABLE.
712 */
713 zone_last = ZONE_HIGHMEM;
714 if (N_MEMORY == N_HIGH_MEMORY)
715 zone_last = ZONE_MOVABLE;
716
717 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
718 arg->status_change_nid_high = nid;
719 else
720 arg->status_change_nid_high = -1;
721#else
722 arg->status_change_nid_high = arg->status_change_nid_normal;
723#endif
724
Lai Jiangshand9713672012-12-11 16:01:03 -0800725 /*
726 * if the node don't have memory befor online, we will need to
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800727 * set the node to node_states[N_MEMORY] after the memory
Lai Jiangshand9713672012-12-11 16:01:03 -0800728 * is online.
729 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800730 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800731 arg->status_change_nid = nid;
732 else
733 arg->status_change_nid = -1;
734}
735
736static void node_states_set_node(int node, struct memory_notify *arg)
737{
738 if (arg->status_change_nid_normal >= 0)
739 node_set_state(node, N_NORMAL_MEMORY);
740
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800741 if (arg->status_change_nid_high >= 0)
742 node_set_state(node, N_HIGH_MEMORY);
743
744 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800745}
746
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700747static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
748 unsigned long nr_pages)
749{
750 unsigned long old_end_pfn = zone_end_pfn(zone);
751
752 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
753 zone->zone_start_pfn = start_pfn;
754
755 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
756}
757
758static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
759 unsigned long nr_pages)
760{
761 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
762
763 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
764 pgdat->node_start_pfn = start_pfn;
765
766 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
767}
768
Michal Hockocdf72f22017-07-06 15:38:25 -0700769void __ref move_pfn_range_to_zone(struct zone *zone,
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700770 unsigned long start_pfn, unsigned long nr_pages)
771{
772 struct pglist_data *pgdat = zone->zone_pgdat;
773 int nid = pgdat->node_id;
774 unsigned long flags;
775
776 if (zone_is_empty(zone))
777 init_currently_empty_zone(zone, start_pfn, nr_pages);
778
779 clear_zone_contiguous(zone);
780
781 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
782 pgdat_resize_lock(pgdat, &flags);
783 zone_span_writelock(zone);
784 resize_zone_range(zone, start_pfn, nr_pages);
785 zone_span_writeunlock(zone);
786 resize_pgdat_range(pgdat, start_pfn, nr_pages);
787 pgdat_resize_unlock(pgdat, &flags);
788
789 /*
790 * TODO now we have a visible range of pages which are not associated
791 * with their zone properly. Not nice but set_pfnblock_flags_mask
792 * expects the zone spans the pfn range. All the pages in the range
793 * are reserved so nobody should be touching them so we should be safe
794 */
795 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG);
796
797 set_zone_contiguous(zone);
798}
799
800/*
Michal Hockoc246a212017-07-06 15:38:18 -0700801 * Returns a default kernel memory zone for the given pfn range.
802 * If no kernel zone covers this pfn range it will automatically go
803 * to the ZONE_NORMAL.
804 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700805static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
Michal Hockoc246a212017-07-06 15:38:18 -0700806 unsigned long nr_pages)
807{
808 struct pglist_data *pgdat = NODE_DATA(nid);
809 int zid;
810
811 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
812 struct zone *zone = &pgdat->node_zones[zid];
813
814 if (zone_intersects(zone, start_pfn, nr_pages))
815 return zone;
816 }
817
818 return &pgdat->node_zones[ZONE_NORMAL];
819}
820
Michal Hockoc6f03e22017-09-06 16:19:40 -0700821static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
822 unsigned long nr_pages)
Michal Hockoe5e68932017-09-06 16:19:37 -0700823{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700824 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
825 nr_pages);
826 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
827 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
828 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700829
830 /*
Michal Hockoc6f03e22017-09-06 16:19:40 -0700831 * We inherit the existing zone in a simple case where zones do not
832 * overlap in the given range
Michal Hockoe5e68932017-09-06 16:19:37 -0700833 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700834 if (in_kernel ^ in_movable)
835 return (in_kernel) ? kernel_zone : movable_zone;
Michal Hockoe5e68932017-09-06 16:19:37 -0700836
Michal Hockoc6f03e22017-09-06 16:19:40 -0700837 /*
838 * If the range doesn't belong to any zone or two zones overlap in the
839 * given range then we use movable zone only if movable_node is
840 * enabled because we always online to a kernel zone by default.
841 */
842 return movable_node_enabled ? movable_zone : kernel_zone;
Michal Hocko9f123ab2017-07-10 15:48:37 -0700843}
844
David Hildenbrand699bbc82021-09-07 19:54:59 -0700845struct zone *zone_for_pfn_range(int online_type, int nid,
846 unsigned long start_pfn, unsigned long nr_pages)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700847{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700848 if (online_type == MMOP_ONLINE_KERNEL)
849 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700850
Michal Hockoc6f03e22017-09-06 16:19:40 -0700851 if (online_type == MMOP_ONLINE_MOVABLE)
852 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
Reza Arbabdf429ac2016-07-26 15:22:23 -0700853
Michal Hockoc6f03e22017-09-06 16:19:40 -0700854 return default_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700855}
856
857/*
858 * Associates the given pfn range with the given node and the zone appropriate
859 * for the given online type.
860 */
861static struct zone * __meminit move_pfn_range(int online_type, int nid,
862 unsigned long start_pfn, unsigned long nr_pages)
863{
864 struct zone *zone;
865
866 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700867 move_pfn_range_to_zone(zone, start_pfn, nr_pages);
868 return zone;
Reza Arbabdf429ac2016-07-26 15:22:23 -0700869}
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700870
Michal Hockob93e0f32017-09-06 16:20:37 -0700871/* Must be protected by mem_hotplug_begin() or a device_lock */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800872int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700873{
Cody P Schaferaa472282013-07-03 15:02:10 -0700874 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700875 unsigned long onlined_pages = 0;
876 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700877 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700878 int nid;
879 int ret;
880 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700881
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700882 nid = pfn_to_nid(pfn);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700883 /* associate pfn range with the zone */
884 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
885
Yasunori Goto7b78d332007-10-21 16:41:36 -0700886 arg.start_pfn = pfn;
887 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -0800888 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700889
Yasunori Goto7b78d332007-10-21 16:41:36 -0700890 ret = memory_notify(MEM_GOING_ONLINE, &arg);
891 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -0700892 if (ret)
893 goto failed_addition;
894
Dave Hansen3947be12005-10-29 18:16:54 -0700895 /*
Yasunori Goto68113782006-06-23 02:03:11 -0700896 * If this zone is not populated, then it is not in zonelist.
897 * This means the page allocator ignores this zone.
898 * So, zonelist must be updated after online.
899 */
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800900 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -0700901 need_zonelists_rebuild = 1;
Michal Hocko72675e12017-09-06 16:20:24 -0700902 setup_zone_pageset(zone);
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800903 }
Yasunori Goto68113782006-06-23 02:03:11 -0700904
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700905 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700906 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700907 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800908 if (need_zonelists_rebuild)
909 zone_pcp_reset(zone);
Chen Yuconge33e33b2016-03-17 14:19:35 -0700910 goto failed_addition;
Geoff Levandfd8a4222008-05-14 16:05:50 -0700911 }
912
Dave Hansen3947be12005-10-29 18:16:54 -0700913 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700914
915 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -0800916 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700917 pgdat_resize_unlock(zone->zone_pgdat, &flags);
918
Jiang Liu08dff7b2012-07-31 16:43:30 -0700919 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -0700920 node_states_set_node(nid, &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -0700921 if (need_zonelists_rebuild)
Michal Hocko72675e12017-09-06 16:20:24 -0700922 build_all_zonelists(NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -0700923 else
924 zone_pcp_update(zone);
925 }
Dave Hansen3947be12005-10-29 18:16:54 -0700926
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700927 init_per_zone_wmark_min();
928
Vlastimil Babka698b1b32016-03-17 14:18:08 -0700929 if (onlined_pages) {
Vlastimil Babkae888ca32016-03-17 14:18:12 -0700930 kswapd_run(nid);
Vlastimil Babka698b1b32016-03-17 14:18:08 -0700931 kcompactd_run(nid);
932 }
Dave Hansen61b13992005-10-29 18:16:56 -0700933
Haicheng Li1f522502010-05-24 14:32:51 -0700934 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700935
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700936 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700937
938 if (onlined_pages)
939 memory_notify(MEM_ONLINE, &arg);
David Rientjes30467e02015-04-14 15:45:11 -0700940 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -0700941
942failed_addition:
943 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
944 (unsigned long long) pfn << PAGE_SHIFT,
945 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
946 memory_notify(MEM_CANCEL_ONLINE, &arg);
947 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700948}
Keith Mannthey53947022006-09-30 23:27:08 -0700949#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700950
Tang Chen0bd85422014-11-13 15:19:41 -0800951static void reset_node_present_pages(pg_data_t *pgdat)
952{
953 struct zone *z;
954
955 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
956 z->present_pages = 0;
957
958 pgdat->node_present_pages = 0;
959}
960
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800961/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
962static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700963{
964 struct pglist_data *pgdat;
965 unsigned long zones_size[MAX_NR_ZONES] = {0};
966 unsigned long zholes_size[MAX_NR_ZONES] = {0};
Fabian Frederickc8e861a2014-06-04 16:07:51 -0700967 unsigned long start_pfn = PFN_DOWN(start);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700968
Tang Chena1e565a2013-02-22 16:33:18 -0800969 pgdat = NODE_DATA(nid);
970 if (!pgdat) {
971 pgdat = arch_alloc_nodedata(nid);
972 if (!pgdat)
973 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700974
Tang Chena1e565a2013-02-22 16:33:18 -0800975 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700976 } else {
Mel Gormane716f2e2017-05-03 14:53:45 -0700977 /*
978 * Reset the nr_zones, order and classzone_idx before reuse.
979 * Note that kswapd will init kswapd_classzone_idx properly
980 * when it starts in the near future.
981 */
Gu Zhengb0dc3a32015-03-25 15:55:20 -0700982 pgdat->nr_zones = 0;
Mel Gorman38087d92016-07-28 15:45:49 -0700983 pgdat->kswapd_order = 0;
984 pgdat->kswapd_classzone_idx = 0;
Tang Chena1e565a2013-02-22 16:33:18 -0800985 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700986
987 /* we can use NODE_DATA(nid) from here */
988
989 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -0700990 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Reza Arbab58301692016-08-11 15:33:12 -0700991 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700992
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700993 /*
994 * The node we allocated has no zone fallback lists. For avoiding
995 * to access not-initialized zonelist, build here.
996 */
Michal Hocko72675e12017-09-06 16:20:24 -0700997 build_all_zonelists(pgdat);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700998
Tang Chenf784a3f2014-11-13 15:19:39 -0800999 /*
1000 * zone->managed_pages is set to an approximate value in
1001 * free_area_init_core(), which will cause
1002 * /sys/device/system/node/nodeX/meminfo has wrong data.
1003 * So reset it to 0 before any memory is onlined.
1004 */
1005 reset_node_managed_pages(pgdat);
1006
Tang Chen0bd85422014-11-13 15:19:41 -08001007 /*
1008 * When memory is hot-added, all the memory is in offline state. So
1009 * clear all zones' present_pages because they will be updated in
1010 * online_pages() and offline_pages().
1011 */
1012 reset_node_present_pages(pgdat);
1013
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001014 return pgdat;
1015}
1016
1017static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1018{
1019 arch_refresh_nodedata(nid, NULL);
Reza Arbab58301692016-08-11 15:33:12 -07001020 free_percpu(pgdat->per_cpu_nodestats);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001021 arch_free_nodedata(pgdat);
1022 return;
1023}
1024
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001025
Toshi Kani01b0f192013-11-12 15:07:25 -08001026/**
1027 * try_online_node - online a node if offlined
1028 *
minskey guocf234222010-05-24 14:32:41 -07001029 * called by cpu_up() to online a node without onlined memory.
1030 */
Toshi Kani01b0f192013-11-12 15:07:25 -08001031int try_online_node(int nid)
minskey guocf234222010-05-24 14:32:41 -07001032{
1033 pg_data_t *pgdat;
1034 int ret;
1035
Toshi Kani01b0f192013-11-12 15:07:25 -08001036 if (node_online(nid))
1037 return 0;
1038
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001039 mem_hotplug_begin();
minskey guocf234222010-05-24 14:32:41 -07001040 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001041 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001042 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001043 ret = -ENOMEM;
1044 goto out;
1045 }
1046 node_set_online(nid);
1047 ret = register_one_node(nid);
1048 BUG_ON(ret);
minskey guocf234222010-05-24 14:32:41 -07001049out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001050 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001051 return ret;
1052}
1053
Toshi Kani27356f52013-09-11 14:21:49 -07001054static int check_hotplug_memory_range(u64 start, u64 size)
1055{
Fabian Frederickc8e861a2014-06-04 16:07:51 -07001056 u64 start_pfn = PFN_DOWN(start);
Toshi Kani27356f52013-09-11 14:21:49 -07001057 u64 nr_pages = size >> PAGE_SHIFT;
1058
1059 /* Memory range must be aligned with section */
1060 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1061 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1062 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1063 (unsigned long long)start,
1064 (unsigned long long)size);
1065 return -EINVAL;
1066 }
1067
1068 return 0;
1069}
1070
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001071static int online_memory_block(struct memory_block *mem, void *arg)
1072{
Nathan Fontenotdc18d702017-02-24 15:00:02 -08001073 return device_online(&mem->dev);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001074}
1075
David Hildenbrand5cb83882018-10-30 15:10:24 -07001076/*
1077 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1078 * and online/offline operations (triggered e.g. by sysfs).
1079 *
1080 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1081 */
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001082int __ref add_memory_resource(int nid, struct resource *res, bool online)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001083{
David Vrabel62cedb92015-06-25 16:35:49 +01001084 u64 start, size;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001085 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001086 bool new_pgdat;
1087 bool new_node;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001088 int ret;
1089
David Vrabel62cedb92015-06-25 16:35:49 +01001090 start = res->start;
1091 size = resource_size(res);
1092
Toshi Kani27356f52013-09-11 14:21:49 -07001093 ret = check_hotplug_memory_range(start, size);
1094 if (ret)
1095 return ret;
1096
Tang Chena1e565a2013-02-22 16:33:18 -08001097 { /* Stupid hack to suppress address-never-null warning */
1098 void *p = NODE_DATA(nid);
1099 new_pgdat = !p;
1100 }
Nathan Zimmerac13c462014-01-23 15:53:26 -08001101
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001102 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001103
Tang Chen7f36e3e2015-09-04 15:42:32 -07001104 /*
1105 * Add new range to memblock so that when hotadd_new_pgdat() is called
1106 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1107 * this new range and calculate total pages correctly. The range will
1108 * be removed at hot-remove time.
1109 */
1110 memblock_add_node(start, size, nid);
1111
Tang Chena1e565a2013-02-22 16:33:18 -08001112 new_node = !node_online(nid);
1113 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001114 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001115 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001116 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001117 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001118 }
1119
Yasunori Gotobc02af92006-06-27 02:53:30 -07001120 /* call arch's memory hotadd */
Michal Hocko3d79a722017-07-06 15:38:21 -07001121 ret = arch_add_memory(nid, start, size, true);
Yasunori Gotobc02af92006-06-27 02:53:30 -07001122
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001123 if (ret < 0)
1124 goto error;
1125
Yasunori Goto0fc44152006-06-27 02:53:38 -07001126 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001127 node_set_online(nid);
1128
Tang Chena1e565a2013-02-22 16:33:18 -08001129 if (new_node) {
Michal Hocko9037a992017-07-06 15:37:49 -07001130 unsigned long start_pfn = start >> PAGE_SHIFT;
1131 unsigned long nr_pages = size >> PAGE_SHIFT;
1132
1133 ret = __register_one_node(nid);
1134 if (ret)
1135 goto register_fail;
1136
1137 /*
1138 * link memory sections under this node. This is already
1139 * done when creatig memory section in register_new_memory
1140 * but that depends to have the node registered so offline
1141 * nodes have to go through register_node.
1142 * TODO clean up this mess.
1143 */
1144 ret = link_mem_sections(nid, start_pfn, nr_pages);
1145register_fail:
Yasunori Goto0fc44152006-06-27 02:53:38 -07001146 /*
1147 * If sysfs file of new node can't create, cpu on the node
1148 * can't be hot-added. There is no rollback way now.
1149 * So, check by BUG_ON() to catch it reluctantly..
1150 */
1151 BUG_ON(ret);
1152 }
1153
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001154 /* create new memmap entry */
1155 firmware_map_add_hotplug(start, start + size, "System RAM");
1156
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001157 /* online pages if requested */
1158 if (online)
1159 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1160 NULL, online_memory_block);
1161
Andi Kleen6ad696d2009-11-17 14:06:22 -08001162 goto out;
1163
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001164error:
1165 /* rollback pgdat allocation and others */
Gustavo A. R. Silvadbac61a2017-07-10 15:47:23 -07001166 if (new_pgdat && pgdat)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001167 rollback_node_hotadd(nid, pgdat);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001168 memblock_remove(start, size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001169
Andi Kleen6ad696d2009-11-17 14:06:22 -08001170out:
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001171 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001172 return ret;
1173}
David Vrabel62cedb92015-06-25 16:35:49 +01001174
David Hildenbrand5cb83882018-10-30 15:10:24 -07001175/* requires device_hotplug_lock, see add_memory_resource() */
1176int __ref __add_memory(int nid, u64 start, u64 size)
David Vrabel62cedb92015-06-25 16:35:49 +01001177{
1178 struct resource *res;
1179 int ret;
1180
1181 res = register_memory_resource(start, size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001182 if (IS_ERR(res))
1183 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001184
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001185 ret = add_memory_resource(nid, res, memhp_auto_online);
David Vrabel62cedb92015-06-25 16:35:49 +01001186 if (ret < 0)
1187 release_memory_resource(res);
1188 return ret;
1189}
David Hildenbrand5cb83882018-10-30 15:10:24 -07001190
1191int add_memory(int nid, u64 start, u64 size)
1192{
1193 int rc;
1194
1195 lock_device_hotplug();
1196 rc = __add_memory(nid, start, size);
1197 unlock_device_hotplug();
1198
1199 return rc;
1200}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001201EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001202
1203#ifdef CONFIG_MEMORY_HOTREMOVE
1204/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001205 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1206 * set and the size of the free page is given by page_order(). Using this,
1207 * the function determines if the pageblock contains only free pages.
1208 * Due to buddy contraints, a free page at least the size of a pageblock will
1209 * be located at the start of the pageblock
1210 */
1211static inline int pageblock_free(struct page *page)
1212{
1213 return PageBuddy(page) && page_order(page) >= pageblock_order;
1214}
1215
1216/* Return the start of the next active pageblock after a given page */
1217static struct page *next_active_pageblock(struct page *page)
1218{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001219 /* Ensure the starting page is pageblock-aligned */
1220 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1221
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001222 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001223 if (pageblock_free(page)) {
1224 int order;
1225 /* be careful. we don't have locks, page_order can be changed.*/
1226 order = page_order(page);
1227 if ((order < MAX_ORDER) && (order >= pageblock_order))
1228 return page + (1 << order);
1229 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001230
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001231 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001232}
1233
1234/* Checks if this range of memory is likely to be hot-removable. */
Yaowei Baic98940f2016-05-19 17:11:26 -07001235bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001236{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001237 struct page *page = pfn_to_page(start_pfn);
Michal Hocko659052d2019-02-01 14:20:34 -08001238 unsigned long end_pfn = min(start_pfn + nr_pages, zone_end_pfn(page_zone(page)));
1239 struct page *end_page = pfn_to_page(end_pfn);
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001240
1241 /* Check the starting page of each pageblock within the range */
1242 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001243 if (!is_pageblock_removable_nolock(page))
Yaowei Baic98940f2016-05-19 17:11:26 -07001244 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001245 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001246 }
1247
1248 /* All pageblocks in the memory block are likely to be hot-removable */
Yaowei Baic98940f2016-05-19 17:11:26 -07001249 return true;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001250}
1251
1252/*
Toshi Kanideb88a22017-02-03 13:13:20 -08001253 * Confirm all pages in a range [start, end) belong to the same zone.
Toshi Kania96dfdd2017-02-03 13:13:23 -08001254 * When true, return its valid [start, end).
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001255 */
Toshi Kania96dfdd2017-02-03 13:13:23 -08001256int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1257 unsigned long *valid_start, unsigned long *valid_end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001258{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001259 unsigned long pfn, sec_end_pfn;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001260 unsigned long start, end;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001261 struct zone *zone = NULL;
1262 struct page *page;
1263 int i;
Toshi Kanideb88a22017-02-03 13:13:20 -08001264 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001265 pfn < end_pfn;
Toshi Kanideb88a22017-02-03 13:13:20 -08001266 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
Andrew Banman5f0f2882015-12-29 14:54:25 -08001267 /* Make sure the memory section is present first */
1268 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001269 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001270 for (; pfn < sec_end_pfn && pfn < end_pfn;
1271 pfn += MAX_ORDER_NR_PAGES) {
1272 i = 0;
1273 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1274 while ((i < MAX_ORDER_NR_PAGES) &&
1275 !pfn_valid_within(pfn + i))
1276 i++;
zhong jiangd6d8c8a2017-02-24 14:59:30 -08001277 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
Andrew Banman5f0f2882015-12-29 14:54:25 -08001278 continue;
Mikhail Zaslonkod83382c2019-02-01 14:20:38 -08001279 /* Check if we got outside of the zone */
1280 if (zone && !zone_spans_pfn(zone, pfn + i))
1281 return 0;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001282 page = pfn_to_page(pfn + i);
1283 if (zone && page_zone(page) != zone)
1284 return 0;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001285 if (!zone)
1286 start = pfn + i;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001287 zone = page_zone(page);
Toshi Kania96dfdd2017-02-03 13:13:23 -08001288 end = pfn + MAX_ORDER_NR_PAGES;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001289 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001290 }
Toshi Kanideb88a22017-02-03 13:13:20 -08001291
Toshi Kania96dfdd2017-02-03 13:13:23 -08001292 if (zone) {
1293 *valid_start = start;
zhong jiangd6d8c8a2017-02-24 14:59:30 -08001294 *valid_end = min(end, end_pfn);
Toshi Kanideb88a22017-02-03 13:13:20 -08001295 return 1;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001296 } else {
Toshi Kanideb88a22017-02-03 13:13:20 -08001297 return 0;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001298 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001299}
1300
1301/*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001302 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1303 * non-lru movable pages and hugepages). We scan pfn because it's much
1304 * easier than scanning over linked list. This function returns the pfn
1305 * of the first found movable page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001306 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001307static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001308{
1309 unsigned long pfn;
1310 struct page *page;
1311 for (pfn = start; pfn < end; pfn++) {
1312 if (pfn_valid(pfn)) {
1313 page = pfn_to_page(pfn);
1314 if (PageLRU(page))
1315 return pfn;
Yisheng Xie0efadf42017-02-24 14:57:39 -08001316 if (__PageMovable(page))
1317 return pfn;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001318 if (PageHuge(page)) {
Naoya Horiguchi7e1f0492015-04-15 16:14:41 -07001319 if (page_huge_active(page))
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001320 return pfn;
1321 else
1322 pfn = round_up(pfn + 1,
1323 1 << compound_order(page)) - 1;
1324 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001325 }
1326 }
1327 return 0;
1328}
1329
Xishi Qiu394e31d2016-07-28 15:48:53 -07001330static struct page *new_node_page(struct page *page, unsigned long private,
1331 int **result)
1332{
Xishi Qiu394e31d2016-07-28 15:48:53 -07001333 int nid = page_to_nid(page);
Li Zhong231e97e2016-09-28 15:22:38 -07001334 nodemask_t nmask = node_states[N_MEMORY];
Michal Hocko7f252f22017-07-10 15:48:41 -07001335
1336 /*
1337 * try to allocate from a different node but reuse this node if there
1338 * are no other online nodes to be used (e.g. we are offlining a part
1339 * of the only existing node)
1340 */
1341 node_clear(nid, nmask);
1342 if (nodes_empty(nmask))
1343 node_set(nid, nmask);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001344
Michal Hocko8b913232017-07-10 15:48:47 -07001345 return new_page_nodemask(page, nid, &nmask);
Xishi Qiu394e31d2016-07-28 15:48:53 -07001346}
1347
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001348#define NR_OFFLINE_AT_ONCE_PAGES (256)
1349static int
1350do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1351{
1352 unsigned long pfn;
1353 struct page *page;
1354 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1355 int not_managed = 0;
1356 int ret = 0;
1357 LIST_HEAD(source);
1358
1359 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1360 if (!pfn_valid(pfn))
1361 continue;
1362 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001363
1364 if (PageHuge(page)) {
1365 struct page *head = compound_head(page);
1366 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1367 if (compound_order(head) > PFN_SECTION_SHIFT) {
1368 ret = -EBUSY;
1369 break;
1370 }
1371 if (isolate_huge_page(page, &source))
1372 move_pages -= 1 << compound_order(head);
1373 continue;
Naoya Horiguchi8135d892017-09-08 16:11:15 -07001374 } else if (thp_migration_supported() && PageTransHuge(page))
1375 pfn = page_to_pfn(compound_head(page))
1376 + hpage_nr_pages(page) - 1;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001377
Michal Hocko2c250712018-12-28 00:38:01 -08001378 /*
1379 * HWPoison pages have elevated reference counts so the migration would
1380 * fail on them. It also doesn't make any sense to migrate them in the
1381 * first place. Still try to unmap such a page in case it is still mapped
1382 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1383 * the unmap as the catch all safety net).
1384 */
1385 if (PageHWPoison(page)) {
1386 if (WARN_ON(PageLRU(page)))
1387 isolate_lru_page(page);
1388 if (page_mapped(page))
1389 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1390 continue;
1391 }
1392
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001393 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001394 continue;
1395 /*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001396 * We can skip free pages. And we can deal with pages on
1397 * LRU and non-lru movable pages.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001398 */
Yisheng Xie0efadf42017-02-24 14:57:39 -08001399 if (PageLRU(page))
1400 ret = isolate_lru_page(page);
1401 else
1402 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001403 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001404 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001405 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001406 move_pages--;
Yisheng Xie0efadf42017-02-24 14:57:39 -08001407 if (!__PageMovable(page))
1408 inc_node_page_state(page, NR_ISOLATED_ANON +
1409 page_is_file_cache(page));
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001410
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001411 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001412#ifdef CONFIG_DEBUG_VM
Yisheng Xie0efadf42017-02-24 14:57:39 -08001413 pr_alert("failed to isolate pfn %lx\n", pfn);
1414 dump_page(page, "isolation failed");
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001415#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001416 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001417 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001418 check this again here. */
1419 if (page_count(page)) {
1420 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001421 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001422 break;
1423 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001424 }
1425 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001426 if (!list_empty(&source)) {
1427 if (not_managed) {
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001428 putback_movable_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001429 goto out;
1430 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001431
Xishi Qiu394e31d2016-07-28 15:48:53 -07001432 /* Allocate a new page from the nearest neighbor node */
1433 ret = migrate_pages(&source, new_node_page, NULL, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001434 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001435 if (ret)
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001436 putback_movable_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001437 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001438out:
1439 return ret;
1440}
1441
1442/*
1443 * remove from free_area[] and mark all as Reserved.
1444 */
1445static int
1446offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1447 void *data)
1448{
1449 __offline_isolated_pages(start, start + nr_pages);
1450 return 0;
1451}
1452
1453static void
1454offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1455{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001456 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001457 offline_isolated_pages_cb);
1458}
1459
1460/*
1461 * Check all pages in range, recoreded as memory resource, are isolated.
1462 */
1463static int
1464check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1465 void *data)
1466{
1467 int ret;
1468 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001469 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001470 offlined = nr_pages;
1471 if (!ret)
1472 *(long *)data += offlined;
1473 return ret;
1474}
1475
1476static long
1477check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1478{
1479 long offlined = 0;
1480 int ret;
1481
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001482 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001483 check_pages_isolated_cb);
1484 if (ret < 0)
1485 offlined = (long)ret;
1486 return offlined;
1487}
1488
Tang Chenc5320922013-11-12 15:08:10 -08001489static int __init cmdline_parse_movable_node(char *p)
1490{
Michal Hocko49323812017-07-06 15:41:05 -07001491#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Tang Chen55ac5902014-01-21 15:49:35 -08001492 movable_node_enabled = true;
Michal Hocko49323812017-07-06 15:41:05 -07001493#else
1494 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1495#endif
Tang Chenc5320922013-11-12 15:08:10 -08001496 return 0;
1497}
1498early_param("movable_node", cmdline_parse_movable_node);
1499
Lai Jiangshand9713672012-12-11 16:01:03 -08001500/* check which state of node_states will be changed when offline memory */
1501static void node_states_check_changes_offline(unsigned long nr_pages,
1502 struct zone *zone, struct memory_notify *arg)
1503{
1504 struct pglist_data *pgdat = zone->zone_pgdat;
1505 unsigned long present_pages = 0;
1506 enum zone_type zt, zone_last = ZONE_NORMAL;
1507
1508 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001509 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1510 * contains nodes which have zones of 0...ZONE_NORMAL,
1511 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001512 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001513 * If we don't have HIGHMEM nor movable node,
1514 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1515 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001516 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001517 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001518 zone_last = ZONE_MOVABLE;
1519
1520 /*
1521 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1522 * If the memory to be offline is in a zone of 0...zone_last,
1523 * and it is the last present memory, 0...zone_last will
1524 * become empty after offline , thus we can determind we will
1525 * need to clear the node from node_states[N_NORMAL_MEMORY].
1526 */
1527 for (zt = 0; zt <= zone_last; zt++)
1528 present_pages += pgdat->node_zones[zt].present_pages;
1529 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1530 arg->status_change_nid_normal = zone_to_nid(zone);
1531 else
1532 arg->status_change_nid_normal = -1;
1533
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001534#ifdef CONFIG_HIGHMEM
1535 /*
1536 * If we have movable node, node_states[N_HIGH_MEMORY]
1537 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1538 * set zone_last to ZONE_HIGHMEM.
1539 *
1540 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1541 * contains nodes which have zones of 0...ZONE_MOVABLE,
1542 * set zone_last to ZONE_MOVABLE.
1543 */
1544 zone_last = ZONE_HIGHMEM;
1545 if (N_MEMORY == N_HIGH_MEMORY)
1546 zone_last = ZONE_MOVABLE;
1547
1548 for (; zt <= zone_last; zt++)
1549 present_pages += pgdat->node_zones[zt].present_pages;
1550 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1551 arg->status_change_nid_high = zone_to_nid(zone);
1552 else
1553 arg->status_change_nid_high = -1;
1554#else
1555 arg->status_change_nid_high = arg->status_change_nid_normal;
1556#endif
1557
Lai Jiangshand9713672012-12-11 16:01:03 -08001558 /*
1559 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1560 */
1561 zone_last = ZONE_MOVABLE;
1562
1563 /*
1564 * check whether node_states[N_HIGH_MEMORY] will be changed
1565 * If we try to offline the last present @nr_pages from the node,
1566 * we can determind we will need to clear the node from
1567 * node_states[N_HIGH_MEMORY].
1568 */
1569 for (; zt <= zone_last; zt++)
1570 present_pages += pgdat->node_zones[zt].present_pages;
1571 if (nr_pages >= present_pages)
1572 arg->status_change_nid = zone_to_nid(zone);
1573 else
1574 arg->status_change_nid = -1;
1575}
1576
1577static void node_states_clear_node(int node, struct memory_notify *arg)
1578{
1579 if (arg->status_change_nid_normal >= 0)
1580 node_clear_state(node, N_NORMAL_MEMORY);
1581
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001582 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1583 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001584 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001585
1586 if ((N_MEMORY != N_HIGH_MEMORY) &&
1587 (arg->status_change_nid >= 0))
1588 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001589}
1590
Wen Congyanga16cee12012-10-08 16:33:58 -07001591static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001592 unsigned long end_pfn, unsigned long timeout)
1593{
1594 unsigned long pfn, nr_pages, expire;
1595 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001596 int ret, drain, retry_max, node;
Cody P Schaferd7029092013-07-03 15:02:11 -07001597 unsigned long flags;
Toshi Kania96dfdd2017-02-03 13:13:23 -08001598 unsigned long valid_start, valid_end;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001599 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001600 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001601
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001602 /* at least, alignment against pageblock is necessary */
1603 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1604 return -EINVAL;
1605 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1606 return -EINVAL;
1607 /* This makes hotplug much easier...and readable.
1608 we assume this for now. .*/
Toshi Kania96dfdd2017-02-03 13:13:23 -08001609 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001610 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001611
Toshi Kania96dfdd2017-02-03 13:13:23 -08001612 zone = page_zone(pfn_to_page(valid_start));
Yasunori Goto7b78d332007-10-21 16:41:36 -07001613 node = zone_to_nid(zone);
1614 nr_pages = end_pfn - start_pfn;
1615
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001616 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001617 ret = start_isolate_page_range(start_pfn, end_pfn,
1618 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001619 if (ret)
David Rientjes30467e02015-04-14 15:45:11 -07001620 return ret;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001621
1622 arg.start_pfn = start_pfn;
1623 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001624 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001625
1626 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1627 ret = notifier_to_errno(ret);
1628 if (ret)
1629 goto failed_removal;
1630
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001631 pfn = start_pfn;
1632 expire = jiffies + timeout;
1633 drain = 0;
1634 retry_max = 5;
1635repeat:
1636 /* start memory hot removal */
1637 ret = -EAGAIN;
1638 if (time_after(jiffies, expire))
1639 goto failed_removal;
1640 ret = -EINTR;
1641 if (signal_pending(current))
1642 goto failed_removal;
1643 ret = 0;
1644 if (drain) {
Thomas Gleixner3f906ba2017-07-10 15:50:09 -07001645 lru_add_drain_all_cpuslocked();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001646 cond_resched();
Vlastimil Babkac0554322014-12-10 15:43:10 -08001647 drain_all_pages(zone);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001648 }
1649
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001650 pfn = scan_movable_pages(start_pfn, end_pfn);
1651 if (pfn) { /* We have movable pages */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001652 ret = do_migrate_range(pfn, end_pfn);
1653 if (!ret) {
1654 drain = 1;
1655 goto repeat;
1656 } else {
1657 if (ret < 0)
1658 if (--retry_max == 0)
1659 goto failed_removal;
1660 yield();
1661 drain = 1;
1662 goto repeat;
1663 }
1664 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001665 /* drain all zone's lru pagevec, this is asynchronous... */
Thomas Gleixner3f906ba2017-07-10 15:50:09 -07001666 lru_add_drain_all_cpuslocked();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001667 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001668 /* drain pcp pages, this is synchronous. */
Vlastimil Babkac0554322014-12-10 15:43:10 -08001669 drain_all_pages(zone);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001670 /*
1671 * dissolve free hugepages in the memory block before doing offlining
1672 * actually in order to make hugetlbfs's object counting consistent.
1673 */
Gerald Schaefer082d5b62016-10-07 17:01:10 -07001674 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1675 if (ret)
1676 goto failed_removal;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001677 /* check again */
1678 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1679 if (offlined_pages < 0) {
1680 ret = -EBUSY;
1681 goto failed_removal;
1682 }
Chen Yuconge33e33b2016-03-17 14:19:35 -07001683 pr_info("Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001684 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001685 We cannot do rollback at this point. */
1686 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001687 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001688 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001689 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001690 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001691 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001692
1693 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001694 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001695 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001696
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001697 init_per_zone_wmark_min();
1698
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001699 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001700 zone_pcp_reset(zone);
Michal Hocko72675e12017-09-06 16:20:24 -07001701 build_all_zonelists(NULL);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001702 } else
1703 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001704
Lai Jiangshand9713672012-12-11 16:01:03 -08001705 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001706 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001707 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001708 kcompactd_stop(node);
1709 }
Minchan Kimbce73942009-06-16 15:32:50 -07001710
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001711 vm_total_pages = nr_free_pagecache_pages();
1712 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001713
1714 memory_notify(MEM_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001715 return 0;
1716
1717failed_removal:
Chen Yuconge33e33b2016-03-17 14:19:35 -07001718 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1719 (unsigned long long) start_pfn << PAGE_SHIFT,
1720 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001721 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001722 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001723 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001724 return ret;
1725}
Badari Pulavarty71088782008-10-18 20:25:58 -07001726
Michal Hockob93e0f32017-09-06 16:20:37 -07001727/* Must be protected by mem_hotplug_begin() or a device_lock */
Wen Congyanga16cee12012-10-08 16:33:58 -07001728int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1729{
1730 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1731}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001732#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07001733
Wen Congyangbbc76be2013-02-22 16:32:54 -08001734/**
1735 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1736 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001737 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001738 * @arg: argument passed to func
1739 * @func: callback for each memory section walked
1740 *
1741 * This function walks through all present mem sections in range
1742 * [start_pfn, end_pfn) and call func on each mem section.
1743 *
1744 * Returns the return value of func.
1745 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001746int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08001747 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001748{
Wen Congyange90bdb72012-10-08 16:34:01 -07001749 struct memory_block *mem = NULL;
1750 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001751 unsigned long pfn, section_nr;
1752 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001753
Wen Congyange90bdb72012-10-08 16:34:01 -07001754 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1755 section_nr = pfn_to_section_nr(pfn);
1756 if (!present_section_nr(section_nr))
1757 continue;
1758
1759 section = __nr_to_section(section_nr);
1760 /* same memblock? */
1761 if (mem)
1762 if ((section_nr >= mem->start_section_nr) &&
1763 (section_nr <= mem->end_section_nr))
1764 continue;
1765
1766 mem = find_memory_block_hinted(section, mem);
1767 if (!mem)
1768 continue;
1769
Wen Congyangbbc76be2013-02-22 16:32:54 -08001770 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001771 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001772 kobject_put(&mem->dev.kobj);
1773 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001774 }
1775 }
1776
1777 if (mem)
1778 kobject_put(&mem->dev.kobj);
1779
Wen Congyangbbc76be2013-02-22 16:32:54 -08001780 return 0;
1781}
1782
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001783#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08001784static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001785{
1786 int ret = !is_memblock_offlined(mem);
1787
Randy Dunlap349daa02013-04-29 15:08:49 -07001788 if (unlikely(ret)) {
1789 phys_addr_t beginpa, endpa;
1790
1791 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1792 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Joe Perches756a0252016-03-17 14:19:47 -07001793 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07001794 &beginpa, &endpa);
1795 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001796
1797 return ret;
1798}
1799
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001800static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001801{
Tang Chen60a5a192013-02-22 16:33:14 -08001802 int cpu;
1803
1804 for_each_present_cpu(cpu) {
1805 if (cpu_to_node(cpu) == pgdat->node_id)
1806 /*
1807 * the cpu on this node isn't removed, and we can't
1808 * offline this node.
1809 */
1810 return -EBUSY;
1811 }
1812
1813 return 0;
1814}
1815
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001816static void unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001817{
1818#ifdef CONFIG_ACPI_NUMA
Wen Congyange13fe862013-02-22 16:33:31 -08001819 int cpu;
1820
1821 for_each_possible_cpu(cpu)
1822 if (cpu_to_node(cpu) == pgdat->node_id)
1823 numa_clear_node(cpu);
1824#endif
1825}
1826
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001827static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001828{
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001829 int ret;
Wen Congyange13fe862013-02-22 16:33:31 -08001830
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001831 ret = check_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001832 if (ret)
1833 return ret;
1834
1835 /*
1836 * the node will be offlined when we come here, so we can clear
1837 * the cpu_to_node() now.
1838 */
1839
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001840 unmap_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001841 return 0;
1842}
1843
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001844/**
1845 * try_offline_node
1846 *
1847 * Offline a node if all memory sections and cpus of the node are removed.
1848 *
1849 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1850 * and online/offline operations before this call.
1851 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001852void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001853{
Wen Congyangd822b862013-02-22 16:33:16 -08001854 pg_data_t *pgdat = NODE_DATA(nid);
1855 unsigned long start_pfn = pgdat->node_start_pfn;
1856 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08001857 unsigned long pfn;
1858
1859 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1860 unsigned long section_nr = pfn_to_section_nr(pfn);
1861
1862 if (!present_section_nr(section_nr))
1863 continue;
1864
1865 if (pfn_to_nid(pfn) != nid)
1866 continue;
1867
1868 /*
1869 * some memory sections of this node are not removed, and we
1870 * can't offline node now.
1871 */
1872 return;
1873 }
1874
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001875 if (check_and_unmap_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08001876 return;
1877
1878 /*
1879 * all memory/cpu of this node are removed, we can offline this
1880 * node now.
1881 */
1882 node_set_offline(nid);
1883 unregister_one_node(nid);
1884}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001885EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001886
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001887/**
1888 * remove_memory
1889 *
1890 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1891 * and online/offline operations before this call, as required by
1892 * try_offline_node().
1893 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001894void __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001895{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001896 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08001897
Toshi Kani27356f52013-09-11 14:21:49 -07001898 BUG_ON(check_hotplug_memory_range(start, size));
1899
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001900 mem_hotplug_begin();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001901
1902 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001903 * All memory blocks must be offlined before removing memory. Check
1904 * whether all memory blocks in question are offline and trigger a BUG()
1905 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001906 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001907 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08001908 check_memblock_offlined_cb);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001909 if (ret)
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001910 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001911
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001912 /* remove memmap entry */
1913 firmware_map_remove(start, start + size, "System RAM");
Xishi Qiuf9126ab2015-08-14 15:35:16 -07001914 memblock_free(start, size);
1915 memblock_remove(start, size);
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001916
Wen Congyang24d335c2013-02-22 16:32:58 -08001917 arch_remove_memory(start, size);
1918
Tang Chen60a5a192013-02-22 16:33:14 -08001919 try_offline_node(nid);
1920
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001921 mem_hotplug_done();
Badari Pulavarty71088782008-10-18 20:25:58 -07001922}
Badari Pulavarty71088782008-10-18 20:25:58 -07001923EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02001924#endif /* CONFIG_MEMORY_HOTREMOVE */