blob: 5118028468eb8fd04311b2de2644cabd3f33fbda [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>
9#include <linux/swap.h>
10#include <linux/interrupt.h>
11#include <linux/pagemap.h>
12#include <linux/bootmem.h>
13#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>
21#include <linux/memory_hotplug.h>
22#include <linux/highmem.h>
23#include <linux/vmalloc.h>
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -070024#include <linux/ioport.h>
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -070025#include <linux/delay.h>
26#include <linux/migrate.h>
27#include <linux/page-isolation.h>
Badari Pulavarty71088782008-10-18 20:25:58 -070028#include <linux/pfn.h>
Andi Kleen6ad696d2009-11-17 14:06:22 -080029#include <linux/suspend.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080030#include <linux/mm_inline.h>
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -080031#include <linux/firmware-map.h>
Tang Chen60a5a192013-02-22 16:33:14 -080032#include <linux/stop_machine.h>
Naoya Horiguchic8721bb2013-09-11 14:22:09 -070033#include <linux/hugetlb.h>
Dave Hansen3947be12005-10-29 18:16:54 -070034
35#include <asm/tlbflush.h>
36
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030037#include "internal.h"
38
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070039/*
40 * online_page_callback contains pointer to current page onlining function.
41 * Initially it is generic_online_page(). If it is required it could be
42 * changed by calling set_online_page_callback() for callback registration
43 * and restore_online_page_callback() for generic callback restore.
44 */
45
46static void generic_online_page(struct page *page);
47
48static online_page_callback_t online_page_callback = generic_online_page;
49
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080050DEFINE_MUTEX(mem_hotplug_mutex);
51
52void lock_memory_hotplug(void)
53{
54 mutex_lock(&mem_hotplug_mutex);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080055}
56
57void unlock_memory_hotplug(void)
58{
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080059 mutex_unlock(&mem_hotplug_mutex);
60}
61
62
Keith Mannthey45e0b782006-09-30 23:27:09 -070063/* add this memory to iomem resource */
64static struct resource *register_memory_resource(u64 start, u64 size)
65{
66 struct resource *res;
67 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
68 BUG_ON(!res);
69
70 res->name = "System RAM";
71 res->start = start;
72 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080073 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070074 if (request_resource(&iomem_resource, res) < 0) {
Toshi Kani4996eed2013-07-03 15:02:39 -070075 pr_debug("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -070076 kfree(res);
77 res = NULL;
78 }
79 return res;
80}
81
82static void release_memory_resource(struct resource *res)
83{
84 if (!res)
85 return;
86 release_resource(res);
87 kfree(res);
88 return;
89}
90
Keith Mannthey53947022006-09-30 23:27:08 -070091#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -080092void get_page_bootmem(unsigned long info, struct page *page,
93 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -070094{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -080095 page->lru.next = (struct list_head *) type;
Yasunori Goto04753272008-04-28 02:13:31 -070096 SetPagePrivate(page);
97 set_page_private(page, info);
98 atomic_inc(&page->_count);
99}
100
Jiang Liu170a5a72013-07-03 15:03:17 -0700101void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700102{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800103 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700104
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800105 type = (unsigned long) page->lru.next;
106 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
107 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700108
109 if (atomic_dec_return(&page->_count) == 1) {
110 ClearPagePrivate(page);
111 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800112 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700113 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700114 }
Yasunori Goto04753272008-04-28 02:13:31 -0700115}
116
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800117#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
118#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700119static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700120{
121 unsigned long *usemap, mapsize, section_nr, i;
122 struct mem_section *ms;
123 struct page *page, *memmap;
124
Yasunori Goto04753272008-04-28 02:13:31 -0700125 section_nr = pfn_to_section_nr(start_pfn);
126 ms = __nr_to_section(section_nr);
127
128 /* Get section's memmap address */
129 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
130
131 /*
132 * Get page for the memmap's phys address
133 * XXX: need more consideration for sparse_vmemmap...
134 */
135 page = virt_to_page(memmap);
136 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
137 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
138
139 /* remember memmap's page */
140 for (i = 0; i < mapsize; i++, page++)
141 get_page_bootmem(section_nr, page, SECTION_INFO);
142
143 usemap = __nr_to_section(section_nr)->pageblock_flags;
144 page = virt_to_page(usemap);
145
146 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
147
148 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700149 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700150
151}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800152#else /* CONFIG_SPARSEMEM_VMEMMAP */
153static void register_page_bootmem_info_section(unsigned long start_pfn)
154{
155 unsigned long *usemap, mapsize, section_nr, i;
156 struct mem_section *ms;
157 struct page *page, *memmap;
158
159 if (!pfn_valid(start_pfn))
160 return;
161
162 section_nr = pfn_to_section_nr(start_pfn);
163 ms = __nr_to_section(section_nr);
164
165 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
166
167 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
168
169 usemap = __nr_to_section(section_nr)->pageblock_flags;
170 page = virt_to_page(usemap);
171
172 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
173
174 for (i = 0; i < mapsize; i++, page++)
175 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
176}
177#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700178
179void register_page_bootmem_info_node(struct pglist_data *pgdat)
180{
181 unsigned long i, pfn, end_pfn, nr_pages;
182 int node = pgdat->node_id;
183 struct page *page;
184 struct zone *zone;
185
186 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
187 page = virt_to_page(pgdat);
188
189 for (i = 0; i < nr_pages; i++, page++)
190 get_page_bootmem(node, page, NODE_INFO);
191
192 zone = &pgdat->node_zones[0];
193 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
Xishi Qiu139c2d72013-09-11 14:21:46 -0700194 if (zone_is_initialized(zone)) {
Yasunori Goto04753272008-04-28 02:13:31 -0700195 nr_pages = zone->wait_table_hash_nr_entries
196 * sizeof(wait_queue_head_t);
197 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
198 page = virt_to_page(zone->wait_table);
199
200 for (i = 0; i < nr_pages; i++, page++)
201 get_page_bootmem(node, page, NODE_INFO);
202 }
203 }
204
205 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800206 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700207
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700208 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700209 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
210 /*
211 * Some platforms can assign the same pfn to multiple nodes - on
212 * node0 as well as nodeN. To avoid registering a pfn against
213 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700214 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700215 */
216 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
217 register_page_bootmem_info_section(pfn);
218 }
Yasunori Goto04753272008-04-28 02:13:31 -0700219}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800220#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700221
Heiko Carstens76cdd582008-05-14 16:05:52 -0700222static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
223 unsigned long end_pfn)
224{
225 unsigned long old_zone_end_pfn;
226
227 zone_span_writelock(zone);
228
Xishi Qiuc33bc312013-09-11 14:21:44 -0700229 old_zone_end_pfn = zone_end_pfn(zone);
Xishi Qiu8080fc02013-09-11 14:21:45 -0700230 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700231 zone->zone_start_pfn = start_pfn;
232
233 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
234 zone->zone_start_pfn;
235
236 zone_span_writeunlock(zone);
237}
238
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800239static void resize_zone(struct zone *zone, unsigned long start_pfn,
240 unsigned long end_pfn)
241{
242 zone_span_writelock(zone);
243
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800244 if (end_pfn - start_pfn) {
245 zone->zone_start_pfn = start_pfn;
246 zone->spanned_pages = end_pfn - start_pfn;
247 } else {
248 /*
249 * make it consist as free_area_init_core(),
250 * if spanned_pages = 0, then keep start_pfn = 0
251 */
252 zone->zone_start_pfn = 0;
253 zone->spanned_pages = 0;
254 }
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800255
256 zone_span_writeunlock(zone);
257}
258
259static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
260 unsigned long end_pfn)
261{
262 enum zone_type zid = zone_idx(zone);
263 int nid = zone->zone_pgdat->node_id;
264 unsigned long pfn;
265
266 for (pfn = start_pfn; pfn < end_pfn; pfn++)
267 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
268}
269
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800270/* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
271 * alloc_bootmem_node_nopanic() */
272static int __ref ensure_zone_is_initialized(struct zone *zone,
273 unsigned long start_pfn, unsigned long num_pages)
274{
275 if (!zone_is_initialized(zone))
276 return init_currently_empty_zone(zone, start_pfn, num_pages,
277 MEMMAP_HOTPLUG);
278 return 0;
279}
280
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800281static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800282 unsigned long start_pfn, unsigned long end_pfn)
283{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800284 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800285 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800286 unsigned long z1_start_pfn;
287
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800288 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
289 if (ret)
290 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800291
292 pgdat_resize_lock(z1->zone_pgdat, &flags);
293
294 /* can't move pfns which are higher than @z2 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800295 if (end_pfn > zone_end_pfn(z2))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800296 goto out_fail;
Jiang Liu834405c2013-07-03 15:03:04 -0700297 /* the move out part must be at the left most of @z2 */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800298 if (start_pfn > z2->zone_start_pfn)
299 goto out_fail;
300 /* must included/overlap */
301 if (end_pfn <= z2->zone_start_pfn)
302 goto out_fail;
303
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800304 /* use start_pfn for z1's start_pfn if z1 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700305 if (!zone_is_empty(z1))
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800306 z1_start_pfn = z1->zone_start_pfn;
307 else
308 z1_start_pfn = start_pfn;
309
310 resize_zone(z1, z1_start_pfn, end_pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -0800311 resize_zone(z2, end_pfn, zone_end_pfn(z2));
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800312
313 pgdat_resize_unlock(z1->zone_pgdat, &flags);
314
315 fix_zone_id(z1, start_pfn, end_pfn);
316
317 return 0;
318out_fail:
319 pgdat_resize_unlock(z1->zone_pgdat, &flags);
320 return -1;
321}
322
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800323static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800324 unsigned long start_pfn, unsigned long end_pfn)
325{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800326 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800327 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800328 unsigned long z2_end_pfn;
329
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800330 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
331 if (ret)
332 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800333
334 pgdat_resize_lock(z1->zone_pgdat, &flags);
335
336 /* can't move pfns which are lower than @z1 */
337 if (z1->zone_start_pfn > start_pfn)
338 goto out_fail;
339 /* the move out part mast at the right most of @z1 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800340 if (zone_end_pfn(z1) > end_pfn)
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800341 goto out_fail;
342 /* must included/overlap */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800343 if (start_pfn >= zone_end_pfn(z1))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800344 goto out_fail;
345
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800346 /* use end_pfn for z2's end_pfn if z2 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700347 if (!zone_is_empty(z2))
Cody P Schafer108bcc92013-02-22 16:35:23 -0800348 z2_end_pfn = zone_end_pfn(z2);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800349 else
350 z2_end_pfn = end_pfn;
351
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800352 resize_zone(z1, z1->zone_start_pfn, start_pfn);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800353 resize_zone(z2, start_pfn, z2_end_pfn);
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800354
355 pgdat_resize_unlock(z1->zone_pgdat, &flags);
356
357 fix_zone_id(z2, start_pfn, end_pfn);
358
359 return 0;
360out_fail:
361 pgdat_resize_unlock(z1->zone_pgdat, &flags);
362 return -1;
363}
364
Heiko Carstens76cdd582008-05-14 16:05:52 -0700365static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
366 unsigned long end_pfn)
367{
Xishi Qiu83285c72013-11-12 15:07:19 -0800368 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
Heiko Carstens76cdd582008-05-14 16:05:52 -0700369
Tang Chen712cd382012-12-11 16:01:07 -0800370 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700371 pgdat->node_start_pfn = start_pfn;
372
373 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
374 pgdat->node_start_pfn;
375}
376
Al Viro31168482008-11-22 17:33:24 +0000377static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700378{
379 struct pglist_data *pgdat = zone->zone_pgdat;
380 int nr_pages = PAGES_PER_SECTION;
381 int nid = pgdat->node_id;
382 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700383 unsigned long flags;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800384 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700385
386 zone_type = zone - pgdat->node_zones;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800387 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
388 if (ret)
389 return ret;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700390
Heiko Carstens76cdd582008-05-14 16:05:52 -0700391 pgdat_resize_lock(zone->zone_pgdat, &flags);
392 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
393 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
394 phys_start_pfn + nr_pages);
395 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800396 memmap_init_zone(nr_pages, nid, zone_type,
397 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700398 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700399}
400
Gary Hadec04fc582009-01-06 14:39:14 -0800401static int __meminit __add_section(int nid, struct zone *zone,
402 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700403{
Dave Hansen3947be12005-10-29 18:16:54 -0700404 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700405 int ret;
406
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700407 if (pfn_valid(phys_start_pfn))
408 return -EEXIST;
409
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700410 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700411
412 if (ret < 0)
413 return ret;
414
Yasunori Goto718127c2006-06-23 02:03:10 -0700415 ret = __add_zone(zone, phys_start_pfn);
416
417 if (ret < 0)
418 return ret;
419
Gary Hadec04fc582009-01-06 14:39:14 -0800420 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700421}
422
David Rientjes4edd7ce2013-04-29 15:08:22 -0700423/*
424 * Reasonably generic function for adding memory. It is
425 * expected that archs that support memory hotplug will
426 * call this function after deciding the zone to which to
427 * add the new pages.
428 */
429int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
430 unsigned long nr_pages)
431{
432 unsigned long i;
433 int err = 0;
434 int start_sec, end_sec;
435 /* during initialize mem_map, align hot-added range to section */
436 start_sec = pfn_to_section_nr(phys_start_pfn);
437 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
438
439 for (i = start_sec; i <= end_sec; i++) {
440 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
441
442 /*
443 * EEXIST is finally dealt with by ioresource collision
444 * check. see add_memory() => register_memory_resource()
445 * Warning will be printed if there is collision.
446 */
447 if (err && (err != -EEXIST))
448 break;
449 err = 0;
450 }
451
452 return err;
453}
454EXPORT_SYMBOL_GPL(__add_pages);
455
456#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800457/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
458static int find_smallest_section_pfn(int nid, struct zone *zone,
459 unsigned long start_pfn,
460 unsigned long end_pfn)
461{
462 struct mem_section *ms;
463
464 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
465 ms = __pfn_to_section(start_pfn);
466
467 if (unlikely(!valid_section(ms)))
468 continue;
469
470 if (unlikely(pfn_to_nid(start_pfn) != nid))
471 continue;
472
473 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
474 continue;
475
476 return start_pfn;
477 }
478
479 return 0;
480}
481
482/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
483static int find_biggest_section_pfn(int nid, struct zone *zone,
484 unsigned long start_pfn,
485 unsigned long end_pfn)
486{
487 struct mem_section *ms;
488 unsigned long pfn;
489
490 /* pfn is the end pfn of a memory section. */
491 pfn = end_pfn - 1;
492 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
493 ms = __pfn_to_section(pfn);
494
495 if (unlikely(!valid_section(ms)))
496 continue;
497
498 if (unlikely(pfn_to_nid(pfn) != nid))
499 continue;
500
501 if (zone && zone != page_zone(pfn_to_page(pfn)))
502 continue;
503
504 return pfn;
505 }
506
507 return 0;
508}
509
510static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
511 unsigned long end_pfn)
512{
Xishi Qiuc33bc312013-09-11 14:21:44 -0700513 unsigned long zone_start_pfn = zone->zone_start_pfn;
514 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
515 unsigned long zone_end_pfn = z;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800516 unsigned long pfn;
517 struct mem_section *ms;
518 int nid = zone_to_nid(zone);
519
520 zone_span_writelock(zone);
521 if (zone_start_pfn == start_pfn) {
522 /*
523 * If the section is smallest section in the zone, it need
524 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
525 * In this case, we find second smallest valid mem_section
526 * for shrinking zone.
527 */
528 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
529 zone_end_pfn);
530 if (pfn) {
531 zone->zone_start_pfn = pfn;
532 zone->spanned_pages = zone_end_pfn - pfn;
533 }
534 } else if (zone_end_pfn == end_pfn) {
535 /*
536 * If the section is biggest section in the zone, it need
537 * shrink zone->spanned_pages.
538 * In this case, we find second biggest valid mem_section for
539 * shrinking zone.
540 */
541 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
542 start_pfn);
543 if (pfn)
544 zone->spanned_pages = pfn - zone_start_pfn + 1;
545 }
546
547 /*
548 * The section is not biggest or smallest mem_section in the zone, it
549 * only creates a hole in the zone. So in this case, we need not
550 * change the zone. But perhaps, the zone has only hole data. Thus
551 * it check the zone has only hole or not.
552 */
553 pfn = zone_start_pfn;
554 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
555 ms = __pfn_to_section(pfn);
556
557 if (unlikely(!valid_section(ms)))
558 continue;
559
560 if (page_zone(pfn_to_page(pfn)) != zone)
561 continue;
562
563 /* If the section is current section, it continues the loop */
564 if (start_pfn == pfn)
565 continue;
566
567 /* If we find valid section, we have nothing to do */
568 zone_span_writeunlock(zone);
569 return;
570 }
571
572 /* The zone has no valid section */
573 zone->zone_start_pfn = 0;
574 zone->spanned_pages = 0;
575 zone_span_writeunlock(zone);
576}
577
578static void shrink_pgdat_span(struct pglist_data *pgdat,
579 unsigned long start_pfn, unsigned long end_pfn)
580{
Xishi Qiu83285c72013-11-12 15:07:19 -0800581 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
582 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
583 unsigned long pgdat_end_pfn = p;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800584 unsigned long pfn;
585 struct mem_section *ms;
586 int nid = pgdat->node_id;
587
588 if (pgdat_start_pfn == start_pfn) {
589 /*
590 * If the section is smallest section in the pgdat, it need
591 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
592 * In this case, we find second smallest valid mem_section
593 * for shrinking zone.
594 */
595 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
596 pgdat_end_pfn);
597 if (pfn) {
598 pgdat->node_start_pfn = pfn;
599 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
600 }
601 } else if (pgdat_end_pfn == end_pfn) {
602 /*
603 * If the section is biggest section in the pgdat, it need
604 * shrink pgdat->node_spanned_pages.
605 * In this case, we find second biggest valid mem_section for
606 * shrinking zone.
607 */
608 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
609 start_pfn);
610 if (pfn)
611 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
612 }
613
614 /*
615 * If the section is not biggest or smallest mem_section in the pgdat,
616 * it only creates a hole in the pgdat. So in this case, we need not
617 * change the pgdat.
618 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
619 * has only hole or not.
620 */
621 pfn = pgdat_start_pfn;
622 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
623 ms = __pfn_to_section(pfn);
624
625 if (unlikely(!valid_section(ms)))
626 continue;
627
628 if (pfn_to_nid(pfn) != nid)
629 continue;
630
631 /* If the section is current section, it continues the loop */
632 if (start_pfn == pfn)
633 continue;
634
635 /* If we find valid section, we have nothing to do */
636 return;
637 }
638
639 /* The pgdat has no valid section */
640 pgdat->node_start_pfn = 0;
641 pgdat->node_spanned_pages = 0;
642}
643
644static void __remove_zone(struct zone *zone, unsigned long start_pfn)
645{
646 struct pglist_data *pgdat = zone->zone_pgdat;
647 int nr_pages = PAGES_PER_SECTION;
648 int zone_type;
649 unsigned long flags;
650
651 zone_type = zone - pgdat->node_zones;
652
653 pgdat_resize_lock(zone->zone_pgdat, &flags);
654 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
655 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
656 pgdat_resize_unlock(zone->zone_pgdat, &flags);
657}
658
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700659static int __remove_section(struct zone *zone, struct mem_section *ms)
660{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800661 unsigned long start_pfn;
662 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700663 int ret = -EINVAL;
664
665 if (!valid_section(ms))
666 return ret;
667
668 ret = unregister_memory_section(ms);
669 if (ret)
670 return ret;
671
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800672 scn_nr = __section_nr(ms);
673 start_pfn = section_nr_to_pfn(scn_nr);
674 __remove_zone(zone, start_pfn);
675
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700676 sparse_remove_one_section(zone, ms);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700677 return 0;
678}
679
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700680/**
681 * __remove_pages() - remove sections of pages from a zone
682 * @zone: zone from which pages need to be removed
683 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
684 * @nr_pages: number of pages to remove (must be multiple of section size)
685 *
686 * Generic helper function to remove section mappings and sysfs entries
687 * for the section of the memory we are removing. Caller needs to make
688 * sure that pages are marked reserved and zones are adjust properly by
689 * calling offline_pages().
690 */
691int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
692 unsigned long nr_pages)
693{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700694 unsigned long i;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700695 int sections_to_remove;
Toshi Kanife74ebb2013-04-29 15:08:20 -0700696 resource_size_t start, size;
697 int ret = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700698
699 /*
700 * We can only remove entire sections
701 */
702 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
703 BUG_ON(nr_pages % PAGES_PER_SECTION);
704
Toshi Kanife74ebb2013-04-29 15:08:20 -0700705 start = phys_start_pfn << PAGE_SHIFT;
706 size = nr_pages * PAGE_SIZE;
707 ret = release_mem_region_adjustable(&iomem_resource, start, size);
Randy Dunlap348f9f02013-05-24 15:55:30 -0700708 if (ret) {
709 resource_size_t endres = start + size - 1;
710
711 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
712 &start, &endres, ret);
713 }
Yasuaki Ishimatsud760afd2012-10-08 16:34:14 -0700714
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700715 sections_to_remove = nr_pages / PAGES_PER_SECTION;
716 for (i = 0; i < sections_to_remove; i++) {
717 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
718 ret = __remove_section(zone, __pfn_to_section(pfn));
719 if (ret)
720 break;
721 }
722 return ret;
723}
724EXPORT_SYMBOL_GPL(__remove_pages);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700725#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700726
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700727int set_online_page_callback(online_page_callback_t callback)
728{
729 int rc = -EINVAL;
730
731 lock_memory_hotplug();
732
733 if (online_page_callback == generic_online_page) {
734 online_page_callback = callback;
735 rc = 0;
736 }
737
738 unlock_memory_hotplug();
739
740 return rc;
741}
742EXPORT_SYMBOL_GPL(set_online_page_callback);
743
744int restore_online_page_callback(online_page_callback_t callback)
745{
746 int rc = -EINVAL;
747
748 lock_memory_hotplug();
749
750 if (online_page_callback == callback) {
751 online_page_callback = generic_online_page;
752 rc = 0;
753 }
754
755 unlock_memory_hotplug();
756
757 return rc;
758}
759EXPORT_SYMBOL_GPL(restore_online_page_callback);
760
761void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700762{
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700763}
764EXPORT_SYMBOL_GPL(__online_page_set_limits);
765
766void __online_page_increment_counters(struct page *page)
767{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700768 adjust_managed_page_count(page, 1);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700769}
770EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700771
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700772void __online_page_free(struct page *page)
773{
Jiang Liu3dcc0572013-07-03 15:03:21 -0700774 __free_reserved_page(page);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700775}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700776EXPORT_SYMBOL_GPL(__online_page_free);
777
778static void generic_online_page(struct page *page)
779{
780 __online_page_set_limits(page);
781 __online_page_increment_counters(page);
782 __online_page_free(page);
783}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700784
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700785static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
786 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700787{
788 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700789 unsigned long onlined_pages = *(unsigned long *)arg;
790 struct page *page;
791 if (PageReserved(pfn_to_page(start_pfn)))
792 for (i = 0; i < nr_pages; i++) {
793 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700794 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700795 onlined_pages++;
796 }
797 *(unsigned long *)arg = onlined_pages;
798 return 0;
799}
800
Lai Jiangshan09285af2012-12-12 13:52:04 -0800801#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -0800802/*
803 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
804 * normal memory.
805 */
Lai Jiangshan09285af2012-12-12 13:52:04 -0800806static bool can_online_high_movable(struct zone *zone)
807{
808 return true;
809}
Tang Chen79a4dce2012-12-18 14:23:24 -0800810#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800811/* ensure every online node has NORMAL memory */
812static bool can_online_high_movable(struct zone *zone)
813{
814 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
815}
Tang Chen79a4dce2012-12-18 14:23:24 -0800816#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800817
Lai Jiangshand9713672012-12-11 16:01:03 -0800818/* check which state of node_states will be changed when online memory */
819static void node_states_check_changes_online(unsigned long nr_pages,
820 struct zone *zone, struct memory_notify *arg)
821{
822 int nid = zone_to_nid(zone);
823 enum zone_type zone_last = ZONE_NORMAL;
824
825 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800826 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
827 * contains nodes which have zones of 0...ZONE_NORMAL,
828 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -0800829 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800830 * If we don't have HIGHMEM nor movable node,
831 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
832 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -0800833 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800834 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -0800835 zone_last = ZONE_MOVABLE;
836
837 /*
838 * if the memory to be online is in a zone of 0...zone_last, and
839 * the zones of 0...zone_last don't have memory before online, we will
840 * need to set the node to node_states[N_NORMAL_MEMORY] after
841 * the memory is online.
842 */
843 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
844 arg->status_change_nid_normal = nid;
845 else
846 arg->status_change_nid_normal = -1;
847
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800848#ifdef CONFIG_HIGHMEM
849 /*
850 * If we have movable node, node_states[N_HIGH_MEMORY]
851 * contains nodes which have zones of 0...ZONE_HIGHMEM,
852 * set zone_last to ZONE_HIGHMEM.
853 *
854 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
855 * contains nodes which have zones of 0...ZONE_MOVABLE,
856 * set zone_last to ZONE_MOVABLE.
857 */
858 zone_last = ZONE_HIGHMEM;
859 if (N_MEMORY == N_HIGH_MEMORY)
860 zone_last = ZONE_MOVABLE;
861
862 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
863 arg->status_change_nid_high = nid;
864 else
865 arg->status_change_nid_high = -1;
866#else
867 arg->status_change_nid_high = arg->status_change_nid_normal;
868#endif
869
Lai Jiangshand9713672012-12-11 16:01:03 -0800870 /*
871 * if the node don't have memory befor online, we will need to
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800872 * set the node to node_states[N_MEMORY] after the memory
Lai Jiangshand9713672012-12-11 16:01:03 -0800873 * is online.
874 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800875 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800876 arg->status_change_nid = nid;
877 else
878 arg->status_change_nid = -1;
879}
880
881static void node_states_set_node(int node, struct memory_notify *arg)
882{
883 if (arg->status_change_nid_normal >= 0)
884 node_set_state(node, N_NORMAL_MEMORY);
885
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800886 if (arg->status_change_nid_high >= 0)
887 node_set_state(node, N_HIGH_MEMORY);
888
889 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800890}
891
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700892
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800893int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700894{
Cody P Schaferaa472282013-07-03 15:02:10 -0700895 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700896 unsigned long onlined_pages = 0;
897 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700898 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700899 int nid;
900 int ret;
901 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700902
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900903 lock_memory_hotplug();
Lai Jiangshand9713672012-12-11 16:01:03 -0800904 /*
905 * This doesn't need a lock to do pfn_to_page().
906 * The section can't be removed here because of the
907 * memory_block->state_mutex.
908 */
909 zone = page_zone(pfn_to_page(pfn));
910
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800911 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
912 !can_online_high_movable(zone)) {
913 unlock_memory_hotplug();
Toshi Kani0a1be152013-07-08 16:00:41 -0700914 return -EINVAL;
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800915 }
916
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800917 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
918 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
919 unlock_memory_hotplug();
Toshi Kani0a1be152013-07-08 16:00:41 -0700920 return -EINVAL;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800921 }
922 }
923 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
924 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
925 unlock_memory_hotplug();
Toshi Kani0a1be152013-07-08 16:00:41 -0700926 return -EINVAL;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800927 }
928 }
929
930 /* Previous code may changed the zone of the pfn range */
931 zone = page_zone(pfn_to_page(pfn));
932
Yasunori Goto7b78d332007-10-21 16:41:36 -0700933 arg.start_pfn = pfn;
934 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -0800935 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700936
Xishi Qiu9c2606b2013-11-12 15:07:21 -0800937 nid = pfn_to_nid(pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700938
939 ret = memory_notify(MEM_GOING_ONLINE, &arg);
940 ret = notifier_to_errno(ret);
941 if (ret) {
942 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900943 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700944 return ret;
945 }
Dave Hansen3947be12005-10-29 18:16:54 -0700946 /*
Yasunori Goto68113782006-06-23 02:03:11 -0700947 * If this zone is not populated, then it is not in zonelist.
948 * This means the page allocator ignores this zone.
949 * So, zonelist must be updated after online.
950 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700951 mutex_lock(&zonelists_mutex);
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800952 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -0700953 need_zonelists_rebuild = 1;
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800954 build_all_zonelists(NULL, zone);
955 }
Yasunori Goto68113782006-06-23 02:03:11 -0700956
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700957 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700958 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700959 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800960 if (need_zonelists_rebuild)
961 zone_pcp_reset(zone);
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700962 mutex_unlock(&zonelists_mutex);
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -0700963 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
964 (unsigned long long) pfn << PAGE_SHIFT,
965 (((unsigned long long) pfn + nr_pages)
966 << PAGE_SHIFT) - 1);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700967 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900968 unlock_memory_hotplug();
Geoff Levandfd8a4222008-05-14 16:05:50 -0700969 return ret;
970 }
971
Dave Hansen3947be12005-10-29 18:16:54 -0700972 zone->present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700973
974 pgdat_resize_lock(zone->zone_pgdat, &flags);
Yasunori Gotof2937be2006-03-09 17:33:51 -0800975 zone->zone_pgdat->node_present_pages += onlined_pages;
Cody P Schaferaa472282013-07-03 15:02:10 -0700976 pgdat_resize_unlock(zone->zone_pgdat, &flags);
977
Jiang Liu08dff7b2012-07-31 16:43:30 -0700978 if (onlined_pages) {
Lai Jiangshand9713672012-12-11 16:01:03 -0800979 node_states_set_node(zone_to_nid(zone), &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -0700980 if (need_zonelists_rebuild)
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800981 build_all_zonelists(NULL, NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -0700982 else
983 zone_pcp_update(zone);
984 }
Dave Hansen3947be12005-10-29 18:16:54 -0700985
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700986 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700987
988 init_per_zone_wmark_min();
989
Jiang Liu08dff7b2012-07-31 16:43:30 -0700990 if (onlined_pages)
Christoph Lameter7ea15302007-10-16 01:25:29 -0700991 kswapd_run(zone_to_nid(zone));
Dave Hansen61b13992005-10-29 18:16:56 -0700992
Haicheng Li1f522502010-05-24 14:32:51 -0700993 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700994
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700995 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700996
997 if (onlined_pages)
998 memory_notify(MEM_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900999 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001000
Dave Hansen3947be12005-10-29 18:16:54 -07001001 return 0;
1002}
Keith Mannthey53947022006-09-30 23:27:08 -07001003#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -07001004
Hidetoshi Setoe1319332009-11-17 14:06:18 -08001005/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1006static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001007{
1008 struct pglist_data *pgdat;
1009 unsigned long zones_size[MAX_NR_ZONES] = {0};
1010 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1011 unsigned long start_pfn = start >> PAGE_SHIFT;
1012
Tang Chena1e565a2013-02-22 16:33:18 -08001013 pgdat = NODE_DATA(nid);
1014 if (!pgdat) {
1015 pgdat = arch_alloc_nodedata(nid);
1016 if (!pgdat)
1017 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001018
Tang Chena1e565a2013-02-22 16:33:18 -08001019 arch_refresh_nodedata(nid, pgdat);
1020 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001021
1022 /* we can use NODE_DATA(nid) from here */
1023
1024 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -07001025 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001026
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001027 /*
1028 * The node we allocated has no zone fallback lists. For avoiding
1029 * to access not-initialized zonelist, build here.
1030 */
David Rientjesf957db42011-06-22 18:13:04 -07001031 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07001032 build_all_zonelists(pgdat, NULL);
David Rientjesf957db42011-06-22 18:13:04 -07001033 mutex_unlock(&zonelists_mutex);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001034
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001035 return pgdat;
1036}
1037
1038static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1039{
1040 arch_refresh_nodedata(nid, NULL);
1041 arch_free_nodedata(pgdat);
1042 return;
1043}
1044
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001045
minskey guocf234222010-05-24 14:32:41 -07001046/*
1047 * called by cpu_up() to online a node without onlined memory.
1048 */
1049int mem_online_node(int nid)
1050{
1051 pg_data_t *pgdat;
1052 int ret;
1053
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001054 lock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -07001055 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001056 if (!pgdat) {
minskey guocf234222010-05-24 14:32:41 -07001057 ret = -ENOMEM;
1058 goto out;
1059 }
1060 node_set_online(nid);
1061 ret = register_one_node(nid);
1062 BUG_ON(ret);
1063
1064out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001065 unlock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -07001066 return ret;
1067}
1068
Toshi Kani27356f52013-09-11 14:21:49 -07001069static int check_hotplug_memory_range(u64 start, u64 size)
1070{
1071 u64 start_pfn = start >> PAGE_SHIFT;
1072 u64 nr_pages = size >> PAGE_SHIFT;
1073
1074 /* Memory range must be aligned with section */
1075 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1076 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1077 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1078 (unsigned long long)start,
1079 (unsigned long long)size);
1080 return -EINVAL;
1081 }
1082
1083 return 0;
1084}
1085
Al Viro31168482008-11-22 17:33:24 +00001086/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1087int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001088{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001089 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001090 bool new_pgdat;
1091 bool new_node;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001092 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001093 int ret;
1094
Toshi Kani27356f52013-09-11 14:21:49 -07001095 ret = check_hotplug_memory_range(start, size);
1096 if (ret)
1097 return ret;
1098
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001099 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -08001100
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001101 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001102 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001103 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -08001104 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001105
Tang Chena1e565a2013-02-22 16:33:18 -08001106 { /* Stupid hack to suppress address-never-null warning */
1107 void *p = NODE_DATA(nid);
1108 new_pgdat = !p;
1109 }
1110 new_node = !node_online(nid);
1111 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001112 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001113 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001114 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001115 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001116 }
1117
Yasunori Gotobc02af92006-06-27 02:53:30 -07001118 /* call arch's memory hotadd */
1119 ret = arch_add_memory(nid, start, size);
1120
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001121 if (ret < 0)
1122 goto error;
1123
Yasunori Goto0fc44152006-06-27 02:53:38 -07001124 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001125 node_set_online(nid);
1126
Tang Chena1e565a2013-02-22 16:33:18 -08001127 if (new_node) {
Yasunori Goto0fc44152006-06-27 02:53:38 -07001128 ret = register_one_node(nid);
1129 /*
1130 * If sysfs file of new node can't create, cpu on the node
1131 * can't be hot-added. There is no rollback way now.
1132 * So, check by BUG_ON() to catch it reluctantly..
1133 */
1134 BUG_ON(ret);
1135 }
1136
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001137 /* create new memmap entry */
1138 firmware_map_add_hotplug(start, start + size, "System RAM");
1139
Andi Kleen6ad696d2009-11-17 14:06:22 -08001140 goto out;
1141
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001142error:
1143 /* rollback pgdat allocation and others */
1144 if (new_pgdat)
1145 rollback_node_hotadd(nid, pgdat);
Sasha Levina864b9d2013-02-22 16:32:48 -08001146 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001147
Andi Kleen6ad696d2009-11-17 14:06:22 -08001148out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001149 unlock_memory_hotplug();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001150 return ret;
1151}
1152EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001153
1154#ifdef CONFIG_MEMORY_HOTREMOVE
1155/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001156 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1157 * set and the size of the free page is given by page_order(). Using this,
1158 * the function determines if the pageblock contains only free pages.
1159 * Due to buddy contraints, a free page at least the size of a pageblock will
1160 * be located at the start of the pageblock
1161 */
1162static inline int pageblock_free(struct page *page)
1163{
1164 return PageBuddy(page) && page_order(page) >= pageblock_order;
1165}
1166
1167/* Return the start of the next active pageblock after a given page */
1168static struct page *next_active_pageblock(struct page *page)
1169{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001170 /* Ensure the starting page is pageblock-aligned */
1171 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1172
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001173 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001174 if (pageblock_free(page)) {
1175 int order;
1176 /* be careful. we don't have locks, page_order can be changed.*/
1177 order = page_order(page);
1178 if ((order < MAX_ORDER) && (order >= pageblock_order))
1179 return page + (1 << order);
1180 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001181
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001182 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001183}
1184
1185/* Checks if this range of memory is likely to be hot-removable. */
1186int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1187{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001188 struct page *page = pfn_to_page(start_pfn);
1189 struct page *end_page = page + nr_pages;
1190
1191 /* Check the starting page of each pageblock within the range */
1192 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001193 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001194 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001195 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001196 }
1197
1198 /* All pageblocks in the memory block are likely to be hot-removable */
1199 return 1;
1200}
1201
1202/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001203 * Confirm all pages in a range [start, end) is belongs to the same zone.
1204 */
1205static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1206{
1207 unsigned long pfn;
1208 struct zone *zone = NULL;
1209 struct page *page;
1210 int i;
1211 for (pfn = start_pfn;
1212 pfn < end_pfn;
1213 pfn += MAX_ORDER_NR_PAGES) {
1214 i = 0;
1215 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1216 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1217 i++;
1218 if (i == MAX_ORDER_NR_PAGES)
1219 continue;
1220 page = pfn_to_page(pfn + i);
1221 if (zone && page_zone(page) != zone)
1222 return 0;
1223 zone = page_zone(page);
1224 }
1225 return 1;
1226}
1227
1228/*
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001229 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1230 * and hugepages). We scan pfn because it's much easier than scanning over
1231 * linked list. This function returns the pfn of the first found movable
1232 * page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001233 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001234static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001235{
1236 unsigned long pfn;
1237 struct page *page;
1238 for (pfn = start; pfn < end; pfn++) {
1239 if (pfn_valid(pfn)) {
1240 page = pfn_to_page(pfn);
1241 if (PageLRU(page))
1242 return pfn;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001243 if (PageHuge(page)) {
1244 if (is_hugepage_active(page))
1245 return pfn;
1246 else
1247 pfn = round_up(pfn + 1,
1248 1 << compound_order(page)) - 1;
1249 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001250 }
1251 }
1252 return 0;
1253}
1254
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001255#define NR_OFFLINE_AT_ONCE_PAGES (256)
1256static int
1257do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1258{
1259 unsigned long pfn;
1260 struct page *page;
1261 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1262 int not_managed = 0;
1263 int ret = 0;
1264 LIST_HEAD(source);
1265
1266 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1267 if (!pfn_valid(pfn))
1268 continue;
1269 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001270
1271 if (PageHuge(page)) {
1272 struct page *head = compound_head(page);
1273 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1274 if (compound_order(head) > PFN_SECTION_SHIFT) {
1275 ret = -EBUSY;
1276 break;
1277 }
1278 if (isolate_huge_page(page, &source))
1279 move_pages -= 1 << compound_order(head);
1280 continue;
1281 }
1282
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001283 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001284 continue;
1285 /*
1286 * We can skip free pages. And we can only deal with pages on
1287 * LRU.
1288 */
Nick Piggin62695a82008-10-18 20:26:09 -07001289 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001290 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001291 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001292 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001293 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001294 inc_zone_page_state(page, NR_ISOLATED_ANON +
1295 page_is_file_cache(page));
1296
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001297 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001298#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -08001299 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1300 pfn);
1301 dump_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001302#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001303 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001304 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001305 check this again here. */
1306 if (page_count(page)) {
1307 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001308 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001309 break;
1310 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001311 }
1312 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001313 if (!list_empty(&source)) {
1314 if (not_managed) {
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001315 putback_movable_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001316 goto out;
1317 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001318
1319 /*
1320 * alloc_migrate_target should be improooooved!!
1321 * migrate_pages returns # of failed pages.
1322 */
1323 ret = migrate_pages(&source, alloc_migrate_target, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001324 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001325 if (ret)
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001326 putback_movable_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001327 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001328out:
1329 return ret;
1330}
1331
1332/*
1333 * remove from free_area[] and mark all as Reserved.
1334 */
1335static int
1336offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1337 void *data)
1338{
1339 __offline_isolated_pages(start, start + nr_pages);
1340 return 0;
1341}
1342
1343static void
1344offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1345{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001346 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001347 offline_isolated_pages_cb);
1348}
1349
1350/*
1351 * Check all pages in range, recoreded as memory resource, are isolated.
1352 */
1353static int
1354check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1355 void *data)
1356{
1357 int ret;
1358 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001359 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001360 offlined = nr_pages;
1361 if (!ret)
1362 *(long *)data += offlined;
1363 return ret;
1364}
1365
1366static long
1367check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1368{
1369 long offlined = 0;
1370 int ret;
1371
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001372 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001373 check_pages_isolated_cb);
1374 if (ret < 0)
1375 offlined = (long)ret;
1376 return offlined;
1377}
1378
Lai Jiangshan09285af2012-12-12 13:52:04 -08001379#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -08001380/*
1381 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1382 * normal memory.
1383 */
Lai Jiangshan09285af2012-12-12 13:52:04 -08001384static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1385{
1386 return true;
1387}
Tang Chen79a4dce2012-12-18 14:23:24 -08001388#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001389/* ensure the node has NORMAL memory if it is still online */
1390static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1391{
1392 struct pglist_data *pgdat = zone->zone_pgdat;
1393 unsigned long present_pages = 0;
1394 enum zone_type zt;
1395
1396 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1397 present_pages += pgdat->node_zones[zt].present_pages;
1398
1399 if (present_pages > nr_pages)
1400 return true;
1401
1402 present_pages = 0;
1403 for (; zt <= ZONE_MOVABLE; zt++)
1404 present_pages += pgdat->node_zones[zt].present_pages;
1405
1406 /*
1407 * we can't offline the last normal memory until all
1408 * higher memory is offlined.
1409 */
1410 return present_pages == 0;
1411}
Tang Chen79a4dce2012-12-18 14:23:24 -08001412#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001413
Lai Jiangshand9713672012-12-11 16:01:03 -08001414/* check which state of node_states will be changed when offline memory */
1415static void node_states_check_changes_offline(unsigned long nr_pages,
1416 struct zone *zone, struct memory_notify *arg)
1417{
1418 struct pglist_data *pgdat = zone->zone_pgdat;
1419 unsigned long present_pages = 0;
1420 enum zone_type zt, zone_last = ZONE_NORMAL;
1421
1422 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001423 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1424 * contains nodes which have zones of 0...ZONE_NORMAL,
1425 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001426 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001427 * If we don't have HIGHMEM nor movable node,
1428 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1429 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001430 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001431 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001432 zone_last = ZONE_MOVABLE;
1433
1434 /*
1435 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1436 * If the memory to be offline is in a zone of 0...zone_last,
1437 * and it is the last present memory, 0...zone_last will
1438 * become empty after offline , thus we can determind we will
1439 * need to clear the node from node_states[N_NORMAL_MEMORY].
1440 */
1441 for (zt = 0; zt <= zone_last; zt++)
1442 present_pages += pgdat->node_zones[zt].present_pages;
1443 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1444 arg->status_change_nid_normal = zone_to_nid(zone);
1445 else
1446 arg->status_change_nid_normal = -1;
1447
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001448#ifdef CONFIG_HIGHMEM
1449 /*
1450 * If we have movable node, node_states[N_HIGH_MEMORY]
1451 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1452 * set zone_last to ZONE_HIGHMEM.
1453 *
1454 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1455 * contains nodes which have zones of 0...ZONE_MOVABLE,
1456 * set zone_last to ZONE_MOVABLE.
1457 */
1458 zone_last = ZONE_HIGHMEM;
1459 if (N_MEMORY == N_HIGH_MEMORY)
1460 zone_last = ZONE_MOVABLE;
1461
1462 for (; zt <= zone_last; zt++)
1463 present_pages += pgdat->node_zones[zt].present_pages;
1464 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1465 arg->status_change_nid_high = zone_to_nid(zone);
1466 else
1467 arg->status_change_nid_high = -1;
1468#else
1469 arg->status_change_nid_high = arg->status_change_nid_normal;
1470#endif
1471
Lai Jiangshand9713672012-12-11 16:01:03 -08001472 /*
1473 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1474 */
1475 zone_last = ZONE_MOVABLE;
1476
1477 /*
1478 * check whether node_states[N_HIGH_MEMORY] will be changed
1479 * If we try to offline the last present @nr_pages from the node,
1480 * we can determind we will need to clear the node from
1481 * node_states[N_HIGH_MEMORY].
1482 */
1483 for (; zt <= zone_last; zt++)
1484 present_pages += pgdat->node_zones[zt].present_pages;
1485 if (nr_pages >= present_pages)
1486 arg->status_change_nid = zone_to_nid(zone);
1487 else
1488 arg->status_change_nid = -1;
1489}
1490
1491static void node_states_clear_node(int node, struct memory_notify *arg)
1492{
1493 if (arg->status_change_nid_normal >= 0)
1494 node_clear_state(node, N_NORMAL_MEMORY);
1495
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001496 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1497 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001498 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001499
1500 if ((N_MEMORY != N_HIGH_MEMORY) &&
1501 (arg->status_change_nid >= 0))
1502 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001503}
1504
Wen Congyanga16cee12012-10-08 16:33:58 -07001505static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001506 unsigned long end_pfn, unsigned long timeout)
1507{
1508 unsigned long pfn, nr_pages, expire;
1509 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001510 int ret, drain, retry_max, node;
Cody P Schaferd7029092013-07-03 15:02:11 -07001511 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001512 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001513 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001514
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001515 /* at least, alignment against pageblock is necessary */
1516 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1517 return -EINVAL;
1518 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1519 return -EINVAL;
1520 /* This makes hotplug much easier...and readable.
1521 we assume this for now. .*/
1522 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1523 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001524
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001525 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -08001526
Yasunori Goto7b78d332007-10-21 16:41:36 -07001527 zone = page_zone(pfn_to_page(start_pfn));
1528 node = zone_to_nid(zone);
1529 nr_pages = end_pfn - start_pfn;
1530
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001531 ret = -EINVAL;
1532 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1533 goto out;
1534
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001535 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001536 ret = start_isolate_page_range(start_pfn, end_pfn,
1537 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001538 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -08001539 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001540
1541 arg.start_pfn = start_pfn;
1542 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001543 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001544
1545 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1546 ret = notifier_to_errno(ret);
1547 if (ret)
1548 goto failed_removal;
1549
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001550 pfn = start_pfn;
1551 expire = jiffies + timeout;
1552 drain = 0;
1553 retry_max = 5;
1554repeat:
1555 /* start memory hot removal */
1556 ret = -EAGAIN;
1557 if (time_after(jiffies, expire))
1558 goto failed_removal;
1559 ret = -EINTR;
1560 if (signal_pending(current))
1561 goto failed_removal;
1562 ret = 0;
1563 if (drain) {
1564 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001565 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001566 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001567 }
1568
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001569 pfn = scan_movable_pages(start_pfn, end_pfn);
1570 if (pfn) { /* We have movable pages */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001571 ret = do_migrate_range(pfn, end_pfn);
1572 if (!ret) {
1573 drain = 1;
1574 goto repeat;
1575 } else {
1576 if (ret < 0)
1577 if (--retry_max == 0)
1578 goto failed_removal;
1579 yield();
1580 drain = 1;
1581 goto repeat;
1582 }
1583 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001584 /* drain all zone's lru pagevec, this is asynchronous... */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001585 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001586 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001587 /* drain pcp pages, this is synchronous. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001588 drain_all_pages();
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001589 /*
1590 * dissolve free hugepages in the memory block before doing offlining
1591 * actually in order to make hugetlbfs's object counting consistent.
1592 */
1593 dissolve_free_huge_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001594 /* check again */
1595 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1596 if (offlined_pages < 0) {
1597 ret = -EBUSY;
1598 goto failed_removal;
1599 }
1600 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001601 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001602 We cannot do rollback at this point. */
1603 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001604 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001605 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001606 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001607 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001608 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001609
1610 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001611 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001612 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001613
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001614 init_per_zone_wmark_min();
1615
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001616 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001617 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001618 mutex_lock(&zonelists_mutex);
1619 build_all_zonelists(NULL, NULL);
1620 mutex_unlock(&zonelists_mutex);
1621 } else
1622 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001623
Lai Jiangshand9713672012-12-11 16:01:03 -08001624 node_states_clear_node(node, &arg);
1625 if (arg.status_change_nid >= 0)
David Rientjes8fe23e02009-12-14 17:58:33 -08001626 kswapd_stop(node);
Minchan Kimbce73942009-06-16 15:32:50 -07001627
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001628 vm_total_pages = nr_free_pagecache_pages();
1629 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001630
1631 memory_notify(MEM_OFFLINE, &arg);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001632 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001633 return 0;
1634
1635failed_removal:
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07001636 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1637 (unsigned long long) start_pfn << PAGE_SHIFT,
1638 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001639 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001640 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001641 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001642
Andi Kleen6ad696d2009-11-17 14:06:22 -08001643out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001644 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001645 return ret;
1646}
Badari Pulavarty71088782008-10-18 20:25:58 -07001647
Wen Congyanga16cee12012-10-08 16:33:58 -07001648int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1649{
1650 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1651}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001652#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07001653
Wen Congyangbbc76be2013-02-22 16:32:54 -08001654/**
1655 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1656 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001657 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001658 * @arg: argument passed to func
1659 * @func: callback for each memory section walked
1660 *
1661 * This function walks through all present mem sections in range
1662 * [start_pfn, end_pfn) and call func on each mem section.
1663 *
1664 * Returns the return value of func.
1665 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001666int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08001667 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001668{
Wen Congyange90bdb72012-10-08 16:34:01 -07001669 struct memory_block *mem = NULL;
1670 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001671 unsigned long pfn, section_nr;
1672 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001673
Wen Congyange90bdb72012-10-08 16:34:01 -07001674 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1675 section_nr = pfn_to_section_nr(pfn);
1676 if (!present_section_nr(section_nr))
1677 continue;
1678
1679 section = __nr_to_section(section_nr);
1680 /* same memblock? */
1681 if (mem)
1682 if ((section_nr >= mem->start_section_nr) &&
1683 (section_nr <= mem->end_section_nr))
1684 continue;
1685
1686 mem = find_memory_block_hinted(section, mem);
1687 if (!mem)
1688 continue;
1689
Wen Congyangbbc76be2013-02-22 16:32:54 -08001690 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001691 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001692 kobject_put(&mem->dev.kobj);
1693 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001694 }
1695 }
1696
1697 if (mem)
1698 kobject_put(&mem->dev.kobj);
1699
Wen Congyangbbc76be2013-02-22 16:32:54 -08001700 return 0;
1701}
1702
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001703#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08001704static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001705{
1706 int ret = !is_memblock_offlined(mem);
1707
Randy Dunlap349daa02013-04-29 15:08:49 -07001708 if (unlikely(ret)) {
1709 phys_addr_t beginpa, endpa;
1710
1711 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1712 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Wen Congyangbbc76be2013-02-22 16:32:54 -08001713 pr_warn("removing memory fails, because memory "
Randy Dunlap349daa02013-04-29 15:08:49 -07001714 "[%pa-%pa] is onlined\n",
1715 &beginpa, &endpa);
1716 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001717
1718 return ret;
1719}
1720
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001721static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001722{
Tang Chen60a5a192013-02-22 16:33:14 -08001723 int cpu;
1724
1725 for_each_present_cpu(cpu) {
1726 if (cpu_to_node(cpu) == pgdat->node_id)
1727 /*
1728 * the cpu on this node isn't removed, and we can't
1729 * offline this node.
1730 */
1731 return -EBUSY;
1732 }
1733
1734 return 0;
1735}
1736
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001737static void unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001738{
1739#ifdef CONFIG_ACPI_NUMA
Wen Congyange13fe862013-02-22 16:33:31 -08001740 int cpu;
1741
1742 for_each_possible_cpu(cpu)
1743 if (cpu_to_node(cpu) == pgdat->node_id)
1744 numa_clear_node(cpu);
1745#endif
1746}
1747
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001748static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001749{
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001750 int ret;
Wen Congyange13fe862013-02-22 16:33:31 -08001751
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001752 ret = check_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001753 if (ret)
1754 return ret;
1755
1756 /*
1757 * the node will be offlined when we come here, so we can clear
1758 * the cpu_to_node() now.
1759 */
1760
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001761 unmap_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001762 return 0;
1763}
1764
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001765/**
1766 * try_offline_node
1767 *
1768 * Offline a node if all memory sections and cpus of the node are removed.
1769 *
1770 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1771 * and online/offline operations before this call.
1772 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001773void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001774{
Wen Congyangd822b862013-02-22 16:33:16 -08001775 pg_data_t *pgdat = NODE_DATA(nid);
1776 unsigned long start_pfn = pgdat->node_start_pfn;
1777 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08001778 unsigned long pfn;
Wen Congyangd822b862013-02-22 16:33:16 -08001779 struct page *pgdat_page = virt_to_page(pgdat);
1780 int i;
Tang Chen60a5a192013-02-22 16:33:14 -08001781
1782 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1783 unsigned long section_nr = pfn_to_section_nr(pfn);
1784
1785 if (!present_section_nr(section_nr))
1786 continue;
1787
1788 if (pfn_to_nid(pfn) != nid)
1789 continue;
1790
1791 /*
1792 * some memory sections of this node are not removed, and we
1793 * can't offline node now.
1794 */
1795 return;
1796 }
1797
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001798 if (check_and_unmap_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08001799 return;
1800
1801 /*
1802 * all memory/cpu of this node are removed, we can offline this
1803 * node now.
1804 */
1805 node_set_offline(nid);
1806 unregister_one_node(nid);
Wen Congyangd822b862013-02-22 16:33:16 -08001807
1808 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1809 /* node data is allocated from boot memory */
1810 return;
1811
1812 /* free waittable in each zone */
1813 for (i = 0; i < MAX_NR_ZONES; i++) {
1814 struct zone *zone = pgdat->node_zones + i;
1815
Jianguo Wuca4b3f32013-03-22 15:04:50 -07001816 /*
1817 * wait_table may be allocated from boot memory,
1818 * here only free if it's allocated by vmalloc.
1819 */
1820 if (is_vmalloc_addr(zone->wait_table))
Wen Congyangd822b862013-02-22 16:33:16 -08001821 vfree(zone->wait_table);
1822 }
1823
1824 /*
1825 * Since there is no way to guarentee the address of pgdat/zone is not
1826 * on stack of any kernel threads or used by other kernel objects
1827 * without reference counting or other symchronizing method, do not
1828 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1829 * the memory when the node is online again.
1830 */
1831 memset(pgdat, 0, sizeof(*pgdat));
Tang Chen60a5a192013-02-22 16:33:14 -08001832}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001833EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001834
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001835/**
1836 * remove_memory
1837 *
1838 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1839 * and online/offline operations before this call, as required by
1840 * try_offline_node().
1841 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001842void __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001843{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001844 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08001845
Toshi Kani27356f52013-09-11 14:21:49 -07001846 BUG_ON(check_hotplug_memory_range(start, size));
1847
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001848 lock_memory_hotplug();
1849
1850 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001851 * All memory blocks must be offlined before removing memory. Check
1852 * whether all memory blocks in question are offline and trigger a BUG()
1853 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001854 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001855 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08001856 check_memblock_offlined_cb);
Wen Congyangbbc76be2013-02-22 16:32:54 -08001857 if (ret) {
1858 unlock_memory_hotplug();
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001859 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001860 }
1861
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001862 /* remove memmap entry */
1863 firmware_map_remove(start, start + size, "System RAM");
1864
Wen Congyang24d335c2013-02-22 16:32:58 -08001865 arch_remove_memory(start, size);
1866
Tang Chen60a5a192013-02-22 16:33:14 -08001867 try_offline_node(nid);
1868
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001869 unlock_memory_hotplug();
Badari Pulavarty71088782008-10-18 20:25:58 -07001870}
Badari Pulavarty71088782008-10-18 20:25:58 -07001871EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02001872#endif /* CONFIG_MEMORY_HOTREMOVE */