blob: 489f235502dbdde48f921d998896d78a56b0c1de [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>
Tang Chenc5320922013-11-12 15:08:10 -080034#include <linux/memblock.h>
Dave Hansen3947be12005-10-29 18:16:54 -070035
36#include <asm/tlbflush.h>
37
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030038#include "internal.h"
39
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070040/*
41 * online_page_callback contains pointer to current page onlining function.
42 * Initially it is generic_online_page(). If it is required it could be
43 * changed by calling set_online_page_callback() for callback registration
44 * and restore_online_page_callback() for generic callback restore.
45 */
46
47static void generic_online_page(struct page *page);
48
49static online_page_callback_t online_page_callback = generic_online_page;
50
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080051DEFINE_MUTEX(mem_hotplug_mutex);
52
53void lock_memory_hotplug(void)
54{
55 mutex_lock(&mem_hotplug_mutex);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080056}
57
58void unlock_memory_hotplug(void)
59{
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080060 mutex_unlock(&mem_hotplug_mutex);
61}
62
63
Keith Mannthey45e0b782006-09-30 23:27:09 -070064/* add this memory to iomem resource */
65static struct resource *register_memory_resource(u64 start, u64 size)
66{
67 struct resource *res;
68 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
69 BUG_ON(!res);
70
71 res->name = "System RAM";
72 res->start = start;
73 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080074 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070075 if (request_resource(&iomem_resource, res) < 0) {
Toshi Kani4996eed2013-07-03 15:02:39 -070076 pr_debug("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -070077 kfree(res);
78 res = NULL;
79 }
80 return res;
81}
82
83static void release_memory_resource(struct resource *res)
84{
85 if (!res)
86 return;
87 release_resource(res);
88 kfree(res);
89 return;
90}
91
Keith Mannthey53947022006-09-30 23:27:08 -070092#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -080093void get_page_bootmem(unsigned long info, struct page *page,
94 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -070095{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -080096 page->lru.next = (struct list_head *) type;
Yasunori Goto04753272008-04-28 02:13:31 -070097 SetPagePrivate(page);
98 set_page_private(page, info);
99 atomic_inc(&page->_count);
100}
101
Jiang Liu170a5a72013-07-03 15:03:17 -0700102void put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700103{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800104 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -0700105
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800106 type = (unsigned long) page->lru.next;
107 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
108 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700109
110 if (atomic_dec_return(&page->_count) == 1) {
111 ClearPagePrivate(page);
112 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800113 INIT_LIST_HEAD(&page->lru);
Jiang Liu170a5a72013-07-03 15:03:17 -0700114 free_reserved_page(page);
Yasunori Goto04753272008-04-28 02:13:31 -0700115 }
Yasunori Goto04753272008-04-28 02:13:31 -0700116}
117
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800118#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
119#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700120static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700121{
122 unsigned long *usemap, mapsize, section_nr, i;
123 struct mem_section *ms;
124 struct page *page, *memmap;
125
Yasunori Goto04753272008-04-28 02:13:31 -0700126 section_nr = pfn_to_section_nr(start_pfn);
127 ms = __nr_to_section(section_nr);
128
129 /* Get section's memmap address */
130 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
131
132 /*
133 * Get page for the memmap's phys address
134 * XXX: need more consideration for sparse_vmemmap...
135 */
136 page = virt_to_page(memmap);
137 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
138 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
139
140 /* remember memmap's page */
141 for (i = 0; i < mapsize; i++, page++)
142 get_page_bootmem(section_nr, page, SECTION_INFO);
143
144 usemap = __nr_to_section(section_nr)->pageblock_flags;
145 page = virt_to_page(usemap);
146
147 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
148
149 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700150 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700151
152}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800153#else /* CONFIG_SPARSEMEM_VMEMMAP */
154static void register_page_bootmem_info_section(unsigned long start_pfn)
155{
156 unsigned long *usemap, mapsize, section_nr, i;
157 struct mem_section *ms;
158 struct page *page, *memmap;
159
160 if (!pfn_valid(start_pfn))
161 return;
162
163 section_nr = pfn_to_section_nr(start_pfn);
164 ms = __nr_to_section(section_nr);
165
166 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
167
168 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
169
170 usemap = __nr_to_section(section_nr)->pageblock_flags;
171 page = virt_to_page(usemap);
172
173 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
174
175 for (i = 0; i < mapsize; i++, page++)
176 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
177}
178#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700179
180void register_page_bootmem_info_node(struct pglist_data *pgdat)
181{
182 unsigned long i, pfn, end_pfn, nr_pages;
183 int node = pgdat->node_id;
184 struct page *page;
185 struct zone *zone;
186
187 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
188 page = virt_to_page(pgdat);
189
190 for (i = 0; i < nr_pages; i++, page++)
191 get_page_bootmem(node, page, NODE_INFO);
192
193 zone = &pgdat->node_zones[0];
194 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
Xishi Qiu139c2d72013-09-11 14:21:46 -0700195 if (zone_is_initialized(zone)) {
Yasunori Goto04753272008-04-28 02:13:31 -0700196 nr_pages = zone->wait_table_hash_nr_entries
197 * sizeof(wait_queue_head_t);
198 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
199 page = virt_to_page(zone->wait_table);
200
201 for (i = 0; i < nr_pages; i++, page++)
202 get_page_bootmem(node, page, NODE_INFO);
203 }
204 }
205
206 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800207 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700208
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700209 /* register section info */
qiuxishif14851a2012-09-17 14:09:24 -0700210 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
211 /*
212 * Some platforms can assign the same pfn to multiple nodes - on
213 * node0 as well as nodeN. To avoid registering a pfn against
214 * multiple nodes we check that this pfn does not already
Tang Chen7e9f5eb2013-07-08 16:00:23 -0700215 * reside in some other nodes.
qiuxishif14851a2012-09-17 14:09:24 -0700216 */
217 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
218 register_page_bootmem_info_section(pfn);
219 }
Yasunori Goto04753272008-04-28 02:13:31 -0700220}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800221#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700222
Heiko Carstens76cdd582008-05-14 16:05:52 -0700223static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
224 unsigned long end_pfn)
225{
226 unsigned long old_zone_end_pfn;
227
228 zone_span_writelock(zone);
229
Xishi Qiuc33bc312013-09-11 14:21:44 -0700230 old_zone_end_pfn = zone_end_pfn(zone);
Xishi Qiu8080fc02013-09-11 14:21:45 -0700231 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700232 zone->zone_start_pfn = start_pfn;
233
234 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
235 zone->zone_start_pfn;
236
237 zone_span_writeunlock(zone);
238}
239
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800240static void resize_zone(struct zone *zone, unsigned long start_pfn,
241 unsigned long end_pfn)
242{
243 zone_span_writelock(zone);
244
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800245 if (end_pfn - start_pfn) {
246 zone->zone_start_pfn = start_pfn;
247 zone->spanned_pages = end_pfn - start_pfn;
248 } else {
249 /*
250 * make it consist as free_area_init_core(),
251 * if spanned_pages = 0, then keep start_pfn = 0
252 */
253 zone->zone_start_pfn = 0;
254 zone->spanned_pages = 0;
255 }
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800256
257 zone_span_writeunlock(zone);
258}
259
260static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
261 unsigned long end_pfn)
262{
263 enum zone_type zid = zone_idx(zone);
264 int nid = zone->zone_pgdat->node_id;
265 unsigned long pfn;
266
267 for (pfn = start_pfn; pfn < end_pfn; pfn++)
268 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
269}
270
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800271/* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
272 * alloc_bootmem_node_nopanic() */
273static int __ref ensure_zone_is_initialized(struct zone *zone,
274 unsigned long start_pfn, unsigned long num_pages)
275{
276 if (!zone_is_initialized(zone))
277 return init_currently_empty_zone(zone, start_pfn, num_pages,
278 MEMMAP_HOTPLUG);
279 return 0;
280}
281
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800282static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800283 unsigned long start_pfn, unsigned long end_pfn)
284{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800285 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800286 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800287 unsigned long z1_start_pfn;
288
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800289 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
290 if (ret)
291 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800292
293 pgdat_resize_lock(z1->zone_pgdat, &flags);
294
295 /* can't move pfns which are higher than @z2 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800296 if (end_pfn > zone_end_pfn(z2))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800297 goto out_fail;
Jiang Liu834405c2013-07-03 15:03:04 -0700298 /* the move out part must be at the left most of @z2 */
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800299 if (start_pfn > z2->zone_start_pfn)
300 goto out_fail;
301 /* must included/overlap */
302 if (end_pfn <= z2->zone_start_pfn)
303 goto out_fail;
304
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800305 /* use start_pfn for z1's start_pfn if z1 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700306 if (!zone_is_empty(z1))
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800307 z1_start_pfn = z1->zone_start_pfn;
308 else
309 z1_start_pfn = start_pfn;
310
311 resize_zone(z1, z1_start_pfn, end_pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -0800312 resize_zone(z2, end_pfn, zone_end_pfn(z2));
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800313
314 pgdat_resize_unlock(z1->zone_pgdat, &flags);
315
316 fix_zone_id(z1, start_pfn, end_pfn);
317
318 return 0;
319out_fail:
320 pgdat_resize_unlock(z1->zone_pgdat, &flags);
321 return -1;
322}
323
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800324static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800325 unsigned long start_pfn, unsigned long end_pfn)
326{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800327 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800328 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800329 unsigned long z2_end_pfn;
330
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800331 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
332 if (ret)
333 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800334
335 pgdat_resize_lock(z1->zone_pgdat, &flags);
336
337 /* can't move pfns which are lower than @z1 */
338 if (z1->zone_start_pfn > start_pfn)
339 goto out_fail;
340 /* the move out part mast at the right most of @z1 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800341 if (zone_end_pfn(z1) > end_pfn)
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800342 goto out_fail;
343 /* must included/overlap */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800344 if (start_pfn >= zone_end_pfn(z1))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800345 goto out_fail;
346
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800347 /* use end_pfn for z2's end_pfn if z2 is empty */
Xishi Qiu8080fc02013-09-11 14:21:45 -0700348 if (!zone_is_empty(z2))
Cody P Schafer108bcc92013-02-22 16:35:23 -0800349 z2_end_pfn = zone_end_pfn(z2);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800350 else
351 z2_end_pfn = end_pfn;
352
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800353 resize_zone(z1, z1->zone_start_pfn, start_pfn);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800354 resize_zone(z2, start_pfn, z2_end_pfn);
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800355
356 pgdat_resize_unlock(z1->zone_pgdat, &flags);
357
358 fix_zone_id(z2, start_pfn, end_pfn);
359
360 return 0;
361out_fail:
362 pgdat_resize_unlock(z1->zone_pgdat, &flags);
363 return -1;
364}
365
Heiko Carstens76cdd582008-05-14 16:05:52 -0700366static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
367 unsigned long end_pfn)
368{
Xishi Qiu83285c72013-11-12 15:07:19 -0800369 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
Heiko Carstens76cdd582008-05-14 16:05:52 -0700370
Tang Chen712cd382012-12-11 16:01:07 -0800371 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700372 pgdat->node_start_pfn = start_pfn;
373
374 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
375 pgdat->node_start_pfn;
376}
377
Al Viro31168482008-11-22 17:33:24 +0000378static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700379{
380 struct pglist_data *pgdat = zone->zone_pgdat;
381 int nr_pages = PAGES_PER_SECTION;
382 int nid = pgdat->node_id;
383 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700384 unsigned long flags;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800385 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700386
387 zone_type = zone - pgdat->node_zones;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800388 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
389 if (ret)
390 return ret;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700391
Heiko Carstens76cdd582008-05-14 16:05:52 -0700392 pgdat_resize_lock(zone->zone_pgdat, &flags);
393 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
394 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
395 phys_start_pfn + nr_pages);
396 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800397 memmap_init_zone(nr_pages, nid, zone_type,
398 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700399 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700400}
401
Gary Hadec04fc582009-01-06 14:39:14 -0800402static int __meminit __add_section(int nid, struct zone *zone,
403 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700404{
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
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800410 ret = sparse_add_one_section(zone, phys_start_pfn);
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
Toshi Kani01b0f192013-11-12 15:07:25 -08001046/**
1047 * try_online_node - online a node if offlined
1048 *
minskey guocf234222010-05-24 14:32:41 -07001049 * called by cpu_up() to online a node without onlined memory.
1050 */
Toshi Kani01b0f192013-11-12 15:07:25 -08001051int try_online_node(int nid)
minskey guocf234222010-05-24 14:32:41 -07001052{
1053 pg_data_t *pgdat;
1054 int ret;
1055
Toshi Kani01b0f192013-11-12 15:07:25 -08001056 if (node_online(nid))
1057 return 0;
1058
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001059 lock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -07001060 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001061 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001062 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001063 ret = -ENOMEM;
1064 goto out;
1065 }
1066 node_set_online(nid);
1067 ret = register_one_node(nid);
1068 BUG_ON(ret);
1069
Toshi Kani01b0f192013-11-12 15:07:25 -08001070 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1071 mutex_lock(&zonelists_mutex);
1072 build_all_zonelists(NULL, NULL);
1073 mutex_unlock(&zonelists_mutex);
1074 }
1075
minskey guocf234222010-05-24 14:32:41 -07001076out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001077 unlock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -07001078 return ret;
1079}
1080
Toshi Kani27356f52013-09-11 14:21:49 -07001081static int check_hotplug_memory_range(u64 start, u64 size)
1082{
1083 u64 start_pfn = start >> PAGE_SHIFT;
1084 u64 nr_pages = size >> PAGE_SHIFT;
1085
1086 /* Memory range must be aligned with section */
1087 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1088 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1089 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1090 (unsigned long long)start,
1091 (unsigned long long)size);
1092 return -EINVAL;
1093 }
1094
1095 return 0;
1096}
1097
Al Viro31168482008-11-22 17:33:24 +00001098/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1099int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001100{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001101 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001102 bool new_pgdat;
1103 bool new_node;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001104 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001105 int ret;
1106
Toshi Kani27356f52013-09-11 14:21:49 -07001107 ret = check_hotplug_memory_range(start, size);
1108 if (ret)
1109 return ret;
1110
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001111 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -08001112
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001113 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001114 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001115 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -08001116 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001117
Tang Chena1e565a2013-02-22 16:33:18 -08001118 { /* Stupid hack to suppress address-never-null warning */
1119 void *p = NODE_DATA(nid);
1120 new_pgdat = !p;
1121 }
1122 new_node = !node_online(nid);
1123 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001124 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001125 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001126 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001127 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001128 }
1129
Yasunori Gotobc02af92006-06-27 02:53:30 -07001130 /* call arch's memory hotadd */
1131 ret = arch_add_memory(nid, start, size);
1132
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001133 if (ret < 0)
1134 goto error;
1135
Yasunori Goto0fc44152006-06-27 02:53:38 -07001136 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001137 node_set_online(nid);
1138
Tang Chena1e565a2013-02-22 16:33:18 -08001139 if (new_node) {
Yasunori Goto0fc44152006-06-27 02:53:38 -07001140 ret = register_one_node(nid);
1141 /*
1142 * If sysfs file of new node can't create, cpu on the node
1143 * can't be hot-added. There is no rollback way now.
1144 * So, check by BUG_ON() to catch it reluctantly..
1145 */
1146 BUG_ON(ret);
1147 }
1148
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001149 /* create new memmap entry */
1150 firmware_map_add_hotplug(start, start + size, "System RAM");
1151
Andi Kleen6ad696d2009-11-17 14:06:22 -08001152 goto out;
1153
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001154error:
1155 /* rollback pgdat allocation and others */
1156 if (new_pgdat)
1157 rollback_node_hotadd(nid, pgdat);
Sasha Levina864b9d2013-02-22 16:32:48 -08001158 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001159
Andi Kleen6ad696d2009-11-17 14:06:22 -08001160out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001161 unlock_memory_hotplug();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001162 return ret;
1163}
1164EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001165
1166#ifdef CONFIG_MEMORY_HOTREMOVE
1167/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001168 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1169 * set and the size of the free page is given by page_order(). Using this,
1170 * the function determines if the pageblock contains only free pages.
1171 * Due to buddy contraints, a free page at least the size of a pageblock will
1172 * be located at the start of the pageblock
1173 */
1174static inline int pageblock_free(struct page *page)
1175{
1176 return PageBuddy(page) && page_order(page) >= pageblock_order;
1177}
1178
1179/* Return the start of the next active pageblock after a given page */
1180static struct page *next_active_pageblock(struct page *page)
1181{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001182 /* Ensure the starting page is pageblock-aligned */
1183 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1184
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001185 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001186 if (pageblock_free(page)) {
1187 int order;
1188 /* be careful. we don't have locks, page_order can be changed.*/
1189 order = page_order(page);
1190 if ((order < MAX_ORDER) && (order >= pageblock_order))
1191 return page + (1 << order);
1192 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001193
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001194 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001195}
1196
1197/* Checks if this range of memory is likely to be hot-removable. */
1198int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1199{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001200 struct page *page = pfn_to_page(start_pfn);
1201 struct page *end_page = page + nr_pages;
1202
1203 /* Check the starting page of each pageblock within the range */
1204 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001205 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001206 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001207 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001208 }
1209
1210 /* All pageblocks in the memory block are likely to be hot-removable */
1211 return 1;
1212}
1213
1214/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001215 * Confirm all pages in a range [start, end) is belongs to the same zone.
1216 */
1217static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1218{
1219 unsigned long pfn;
1220 struct zone *zone = NULL;
1221 struct page *page;
1222 int i;
1223 for (pfn = start_pfn;
1224 pfn < end_pfn;
1225 pfn += MAX_ORDER_NR_PAGES) {
1226 i = 0;
1227 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1228 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1229 i++;
1230 if (i == MAX_ORDER_NR_PAGES)
1231 continue;
1232 page = pfn_to_page(pfn + i);
1233 if (zone && page_zone(page) != zone)
1234 return 0;
1235 zone = page_zone(page);
1236 }
1237 return 1;
1238}
1239
1240/*
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001241 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1242 * and hugepages). We scan pfn because it's much easier than scanning over
1243 * linked list. This function returns the pfn of the first found movable
1244 * page if it's found, otherwise 0.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001245 */
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001246static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001247{
1248 unsigned long pfn;
1249 struct page *page;
1250 for (pfn = start; pfn < end; pfn++) {
1251 if (pfn_valid(pfn)) {
1252 page = pfn_to_page(pfn);
1253 if (PageLRU(page))
1254 return pfn;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001255 if (PageHuge(page)) {
1256 if (is_hugepage_active(page))
1257 return pfn;
1258 else
1259 pfn = round_up(pfn + 1,
1260 1 << compound_order(page)) - 1;
1261 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001262 }
1263 }
1264 return 0;
1265}
1266
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001267#define NR_OFFLINE_AT_ONCE_PAGES (256)
1268static int
1269do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1270{
1271 unsigned long pfn;
1272 struct page *page;
1273 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1274 int not_managed = 0;
1275 int ret = 0;
1276 LIST_HEAD(source);
1277
1278 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1279 if (!pfn_valid(pfn))
1280 continue;
1281 page = pfn_to_page(pfn);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001282
1283 if (PageHuge(page)) {
1284 struct page *head = compound_head(page);
1285 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1286 if (compound_order(head) > PFN_SECTION_SHIFT) {
1287 ret = -EBUSY;
1288 break;
1289 }
1290 if (isolate_huge_page(page, &source))
1291 move_pages -= 1 << compound_order(head);
1292 continue;
1293 }
1294
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001295 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001296 continue;
1297 /*
1298 * We can skip free pages. And we can only deal with pages on
1299 * LRU.
1300 */
Nick Piggin62695a82008-10-18 20:26:09 -07001301 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001302 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001303 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001304 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001305 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001306 inc_zone_page_state(page, NR_ISOLATED_ANON +
1307 page_is_file_cache(page));
1308
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001309 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001310#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -08001311 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1312 pfn);
1313 dump_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001314#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001315 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001316 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001317 check this again here. */
1318 if (page_count(page)) {
1319 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001320 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001321 break;
1322 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001323 }
1324 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001325 if (!list_empty(&source)) {
1326 if (not_managed) {
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001327 putback_movable_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001328 goto out;
1329 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001330
1331 /*
1332 * alloc_migrate_target should be improooooved!!
1333 * migrate_pages returns # of failed pages.
1334 */
1335 ret = migrate_pages(&source, alloc_migrate_target, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001336 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001337 if (ret)
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001338 putback_movable_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001339 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001340out:
1341 return ret;
1342}
1343
1344/*
1345 * remove from free_area[] and mark all as Reserved.
1346 */
1347static int
1348offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1349 void *data)
1350{
1351 __offline_isolated_pages(start, start + nr_pages);
1352 return 0;
1353}
1354
1355static void
1356offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1357{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001358 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001359 offline_isolated_pages_cb);
1360}
1361
1362/*
1363 * Check all pages in range, recoreded as memory resource, are isolated.
1364 */
1365static int
1366check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1367 void *data)
1368{
1369 int ret;
1370 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001371 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001372 offlined = nr_pages;
1373 if (!ret)
1374 *(long *)data += offlined;
1375 return ret;
1376}
1377
1378static long
1379check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1380{
1381 long offlined = 0;
1382 int ret;
1383
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001384 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001385 check_pages_isolated_cb);
1386 if (ret < 0)
1387 offlined = (long)ret;
1388 return offlined;
1389}
1390
Lai Jiangshan09285af2012-12-12 13:52:04 -08001391#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -08001392/*
1393 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1394 * normal memory.
1395 */
Lai Jiangshan09285af2012-12-12 13:52:04 -08001396static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1397{
1398 return true;
1399}
Tang Chen79a4dce2012-12-18 14:23:24 -08001400#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001401/* ensure the node has NORMAL memory if it is still online */
1402static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1403{
1404 struct pglist_data *pgdat = zone->zone_pgdat;
1405 unsigned long present_pages = 0;
1406 enum zone_type zt;
1407
1408 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1409 present_pages += pgdat->node_zones[zt].present_pages;
1410
1411 if (present_pages > nr_pages)
1412 return true;
1413
1414 present_pages = 0;
1415 for (; zt <= ZONE_MOVABLE; zt++)
1416 present_pages += pgdat->node_zones[zt].present_pages;
1417
1418 /*
1419 * we can't offline the last normal memory until all
1420 * higher memory is offlined.
1421 */
1422 return present_pages == 0;
1423}
Tang Chen79a4dce2012-12-18 14:23:24 -08001424#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001425
Tang Chenc5320922013-11-12 15:08:10 -08001426static int __init cmdline_parse_movable_node(char *p)
1427{
1428#ifdef CONFIG_MOVABLE_NODE
1429 /*
1430 * Memory used by the kernel cannot be hot-removed because Linux
1431 * cannot migrate the kernel pages. When memory hotplug is
1432 * enabled, we should prevent memblock from allocating memory
1433 * for the kernel.
1434 *
1435 * ACPI SRAT records all hotpluggable memory ranges. But before
1436 * SRAT is parsed, we don't know about it.
1437 *
1438 * The kernel image is loaded into memory at very early time. We
1439 * cannot prevent this anyway. So on NUMA system, we set any
1440 * node the kernel resides in as un-hotpluggable.
1441 *
1442 * Since on modern servers, one node could have double-digit
1443 * gigabytes memory, we can assume the memory around the kernel
1444 * image is also un-hotpluggable. So before SRAT is parsed, just
1445 * allocate memory near the kernel image to try the best to keep
1446 * the kernel away from hotpluggable memory.
1447 */
1448 memblock_set_bottom_up(true);
1449#else
1450 pr_warn("movable_node option not supported\n");
1451#endif
1452 return 0;
1453}
1454early_param("movable_node", cmdline_parse_movable_node);
1455
Lai Jiangshand9713672012-12-11 16:01:03 -08001456/* check which state of node_states will be changed when offline memory */
1457static void node_states_check_changes_offline(unsigned long nr_pages,
1458 struct zone *zone, struct memory_notify *arg)
1459{
1460 struct pglist_data *pgdat = zone->zone_pgdat;
1461 unsigned long present_pages = 0;
1462 enum zone_type zt, zone_last = ZONE_NORMAL;
1463
1464 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001465 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1466 * contains nodes which have zones of 0...ZONE_NORMAL,
1467 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001468 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001469 * If we don't have HIGHMEM nor movable node,
1470 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1471 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001472 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001473 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001474 zone_last = ZONE_MOVABLE;
1475
1476 /*
1477 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1478 * If the memory to be offline is in a zone of 0...zone_last,
1479 * and it is the last present memory, 0...zone_last will
1480 * become empty after offline , thus we can determind we will
1481 * need to clear the node from node_states[N_NORMAL_MEMORY].
1482 */
1483 for (zt = 0; zt <= zone_last; zt++)
1484 present_pages += pgdat->node_zones[zt].present_pages;
1485 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1486 arg->status_change_nid_normal = zone_to_nid(zone);
1487 else
1488 arg->status_change_nid_normal = -1;
1489
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001490#ifdef CONFIG_HIGHMEM
1491 /*
1492 * If we have movable node, node_states[N_HIGH_MEMORY]
1493 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1494 * set zone_last to ZONE_HIGHMEM.
1495 *
1496 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1497 * contains nodes which have zones of 0...ZONE_MOVABLE,
1498 * set zone_last to ZONE_MOVABLE.
1499 */
1500 zone_last = ZONE_HIGHMEM;
1501 if (N_MEMORY == N_HIGH_MEMORY)
1502 zone_last = ZONE_MOVABLE;
1503
1504 for (; zt <= zone_last; zt++)
1505 present_pages += pgdat->node_zones[zt].present_pages;
1506 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1507 arg->status_change_nid_high = zone_to_nid(zone);
1508 else
1509 arg->status_change_nid_high = -1;
1510#else
1511 arg->status_change_nid_high = arg->status_change_nid_normal;
1512#endif
1513
Lai Jiangshand9713672012-12-11 16:01:03 -08001514 /*
1515 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1516 */
1517 zone_last = ZONE_MOVABLE;
1518
1519 /*
1520 * check whether node_states[N_HIGH_MEMORY] will be changed
1521 * If we try to offline the last present @nr_pages from the node,
1522 * we can determind we will need to clear the node from
1523 * node_states[N_HIGH_MEMORY].
1524 */
1525 for (; zt <= zone_last; zt++)
1526 present_pages += pgdat->node_zones[zt].present_pages;
1527 if (nr_pages >= present_pages)
1528 arg->status_change_nid = zone_to_nid(zone);
1529 else
1530 arg->status_change_nid = -1;
1531}
1532
1533static void node_states_clear_node(int node, struct memory_notify *arg)
1534{
1535 if (arg->status_change_nid_normal >= 0)
1536 node_clear_state(node, N_NORMAL_MEMORY);
1537
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001538 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1539 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001540 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001541
1542 if ((N_MEMORY != N_HIGH_MEMORY) &&
1543 (arg->status_change_nid >= 0))
1544 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001545}
1546
Wen Congyanga16cee12012-10-08 16:33:58 -07001547static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001548 unsigned long end_pfn, unsigned long timeout)
1549{
1550 unsigned long pfn, nr_pages, expire;
1551 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001552 int ret, drain, retry_max, node;
Cody P Schaferd7029092013-07-03 15:02:11 -07001553 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001554 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001555 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001556
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001557 /* at least, alignment against pageblock is necessary */
1558 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1559 return -EINVAL;
1560 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1561 return -EINVAL;
1562 /* This makes hotplug much easier...and readable.
1563 we assume this for now. .*/
1564 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1565 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001566
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001567 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -08001568
Yasunori Goto7b78d332007-10-21 16:41:36 -07001569 zone = page_zone(pfn_to_page(start_pfn));
1570 node = zone_to_nid(zone);
1571 nr_pages = end_pfn - start_pfn;
1572
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001573 ret = -EINVAL;
1574 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1575 goto out;
1576
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001577 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001578 ret = start_isolate_page_range(start_pfn, end_pfn,
1579 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001580 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -08001581 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001582
1583 arg.start_pfn = start_pfn;
1584 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001585 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001586
1587 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1588 ret = notifier_to_errno(ret);
1589 if (ret)
1590 goto failed_removal;
1591
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001592 pfn = start_pfn;
1593 expire = jiffies + timeout;
1594 drain = 0;
1595 retry_max = 5;
1596repeat:
1597 /* start memory hot removal */
1598 ret = -EAGAIN;
1599 if (time_after(jiffies, expire))
1600 goto failed_removal;
1601 ret = -EINTR;
1602 if (signal_pending(current))
1603 goto failed_removal;
1604 ret = 0;
1605 if (drain) {
1606 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001607 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001608 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001609 }
1610
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001611 pfn = scan_movable_pages(start_pfn, end_pfn);
1612 if (pfn) { /* We have movable pages */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001613 ret = do_migrate_range(pfn, end_pfn);
1614 if (!ret) {
1615 drain = 1;
1616 goto repeat;
1617 } else {
1618 if (ret < 0)
1619 if (--retry_max == 0)
1620 goto failed_removal;
1621 yield();
1622 drain = 1;
1623 goto repeat;
1624 }
1625 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001626 /* drain all zone's lru pagevec, this is asynchronous... */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001627 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001628 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001629 /* drain pcp pages, this is synchronous. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001630 drain_all_pages();
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001631 /*
1632 * dissolve free hugepages in the memory block before doing offlining
1633 * actually in order to make hugetlbfs's object counting consistent.
1634 */
1635 dissolve_free_huge_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001636 /* check again */
1637 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1638 if (offlined_pages < 0) {
1639 ret = -EBUSY;
1640 goto failed_removal;
1641 }
1642 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001643 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001644 We cannot do rollback at this point. */
1645 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001646 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001647 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001648 /* removal success */
Jiang Liu3dcc0572013-07-03 15:03:21 -07001649 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001650 zone->present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001651
1652 pgdat_resize_lock(zone->zone_pgdat, &flags);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001653 zone->zone_pgdat->node_present_pages -= offlined_pages;
Cody P Schaferd7029092013-07-03 15:02:11 -07001654 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001655
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001656 init_per_zone_wmark_min();
1657
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001658 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001659 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001660 mutex_lock(&zonelists_mutex);
1661 build_all_zonelists(NULL, NULL);
1662 mutex_unlock(&zonelists_mutex);
1663 } else
1664 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001665
Lai Jiangshand9713672012-12-11 16:01:03 -08001666 node_states_clear_node(node, &arg);
1667 if (arg.status_change_nid >= 0)
David Rientjes8fe23e02009-12-14 17:58:33 -08001668 kswapd_stop(node);
Minchan Kimbce73942009-06-16 15:32:50 -07001669
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001670 vm_total_pages = nr_free_pagecache_pages();
1671 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001672
1673 memory_notify(MEM_OFFLINE, &arg);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001674 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001675 return 0;
1676
1677failed_removal:
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07001678 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1679 (unsigned long long) start_pfn << PAGE_SHIFT,
1680 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001681 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001682 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001683 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001684
Andi Kleen6ad696d2009-11-17 14:06:22 -08001685out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001686 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001687 return ret;
1688}
Badari Pulavarty71088782008-10-18 20:25:58 -07001689
Wen Congyanga16cee12012-10-08 16:33:58 -07001690int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1691{
1692 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1693}
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001694#endif /* CONFIG_MEMORY_HOTREMOVE */
Wen Congyanga16cee12012-10-08 16:33:58 -07001695
Wen Congyangbbc76be2013-02-22 16:32:54 -08001696/**
1697 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1698 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001699 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001700 * @arg: argument passed to func
1701 * @func: callback for each memory section walked
1702 *
1703 * This function walks through all present mem sections in range
1704 * [start_pfn, end_pfn) and call func on each mem section.
1705 *
1706 * Returns the return value of func.
1707 */
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001708int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
Wen Congyangbbc76be2013-02-22 16:32:54 -08001709 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001710{
Wen Congyange90bdb72012-10-08 16:34:01 -07001711 struct memory_block *mem = NULL;
1712 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001713 unsigned long pfn, section_nr;
1714 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001715
Wen Congyange90bdb72012-10-08 16:34:01 -07001716 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1717 section_nr = pfn_to_section_nr(pfn);
1718 if (!present_section_nr(section_nr))
1719 continue;
1720
1721 section = __nr_to_section(section_nr);
1722 /* same memblock? */
1723 if (mem)
1724 if ((section_nr >= mem->start_section_nr) &&
1725 (section_nr <= mem->end_section_nr))
1726 continue;
1727
1728 mem = find_memory_block_hinted(section, mem);
1729 if (!mem)
1730 continue;
1731
Wen Congyangbbc76be2013-02-22 16:32:54 -08001732 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001733 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001734 kobject_put(&mem->dev.kobj);
1735 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001736 }
1737 }
1738
1739 if (mem)
1740 kobject_put(&mem->dev.kobj);
1741
Wen Congyangbbc76be2013-02-22 16:32:54 -08001742 return 0;
1743}
1744
Rafael J. Wysockie2ff3942013-05-08 00:29:49 +02001745#ifdef CONFIG_MEMORY_HOTREMOVE
Xishi Qiud6de9d52013-11-12 15:07:20 -08001746static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001747{
1748 int ret = !is_memblock_offlined(mem);
1749
Randy Dunlap349daa02013-04-29 15:08:49 -07001750 if (unlikely(ret)) {
1751 phys_addr_t beginpa, endpa;
1752
1753 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1754 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Wen Congyangbbc76be2013-02-22 16:32:54 -08001755 pr_warn("removing memory fails, because memory "
Randy Dunlap349daa02013-04-29 15:08:49 -07001756 "[%pa-%pa] is onlined\n",
1757 &beginpa, &endpa);
1758 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001759
1760 return ret;
1761}
1762
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001763static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08001764{
Tang Chen60a5a192013-02-22 16:33:14 -08001765 int cpu;
1766
1767 for_each_present_cpu(cpu) {
1768 if (cpu_to_node(cpu) == pgdat->node_id)
1769 /*
1770 * the cpu on this node isn't removed, and we can't
1771 * offline this node.
1772 */
1773 return -EBUSY;
1774 }
1775
1776 return 0;
1777}
1778
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001779static void unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001780{
1781#ifdef CONFIG_ACPI_NUMA
Wen Congyange13fe862013-02-22 16:33:31 -08001782 int cpu;
1783
1784 for_each_possible_cpu(cpu)
1785 if (cpu_to_node(cpu) == pgdat->node_id)
1786 numa_clear_node(cpu);
1787#endif
1788}
1789
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001790static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
Wen Congyange13fe862013-02-22 16:33:31 -08001791{
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001792 int ret;
Wen Congyange13fe862013-02-22 16:33:31 -08001793
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001794 ret = check_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001795 if (ret)
1796 return ret;
1797
1798 /*
1799 * the node will be offlined when we come here, so we can clear
1800 * the cpu_to_node() now.
1801 */
1802
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001803 unmap_cpu_on_node(pgdat);
Wen Congyange13fe862013-02-22 16:33:31 -08001804 return 0;
1805}
1806
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001807/**
1808 * try_offline_node
1809 *
1810 * Offline a node if all memory sections and cpus of the node are removed.
1811 *
1812 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1813 * and online/offline operations before this call.
1814 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001815void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001816{
Wen Congyangd822b862013-02-22 16:33:16 -08001817 pg_data_t *pgdat = NODE_DATA(nid);
1818 unsigned long start_pfn = pgdat->node_start_pfn;
1819 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08001820 unsigned long pfn;
Wen Congyangd822b862013-02-22 16:33:16 -08001821 struct page *pgdat_page = virt_to_page(pgdat);
1822 int i;
Tang Chen60a5a192013-02-22 16:33:14 -08001823
1824 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1825 unsigned long section_nr = pfn_to_section_nr(pfn);
1826
1827 if (!present_section_nr(section_nr))
1828 continue;
1829
1830 if (pfn_to_nid(pfn) != nid)
1831 continue;
1832
1833 /*
1834 * some memory sections of this node are not removed, and we
1835 * can't offline node now.
1836 */
1837 return;
1838 }
1839
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001840 if (check_and_unmap_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08001841 return;
1842
1843 /*
1844 * all memory/cpu of this node are removed, we can offline this
1845 * node now.
1846 */
1847 node_set_offline(nid);
1848 unregister_one_node(nid);
Wen Congyangd822b862013-02-22 16:33:16 -08001849
1850 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1851 /* node data is allocated from boot memory */
1852 return;
1853
1854 /* free waittable in each zone */
1855 for (i = 0; i < MAX_NR_ZONES; i++) {
1856 struct zone *zone = pgdat->node_zones + i;
1857
Jianguo Wuca4b3f32013-03-22 15:04:50 -07001858 /*
1859 * wait_table may be allocated from boot memory,
1860 * here only free if it's allocated by vmalloc.
1861 */
1862 if (is_vmalloc_addr(zone->wait_table))
Wen Congyangd822b862013-02-22 16:33:16 -08001863 vfree(zone->wait_table);
1864 }
1865
1866 /*
1867 * Since there is no way to guarentee the address of pgdat/zone is not
1868 * on stack of any kernel threads or used by other kernel objects
1869 * without reference counting or other symchronizing method, do not
1870 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1871 * the memory when the node is online again.
1872 */
1873 memset(pgdat, 0, sizeof(*pgdat));
Tang Chen60a5a192013-02-22 16:33:14 -08001874}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001875EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001876
Toshi Kani0f1cfe92013-09-11 14:21:50 -07001877/**
1878 * remove_memory
1879 *
1880 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1881 * and online/offline operations before this call, as required by
1882 * try_offline_node().
1883 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001884void __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001885{
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001886 int ret;
Wen Congyang993c1aa2013-02-22 16:32:50 -08001887
Toshi Kani27356f52013-09-11 14:21:49 -07001888 BUG_ON(check_hotplug_memory_range(start, size));
1889
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001890 lock_memory_hotplug();
1891
1892 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001893 * All memory blocks must be offlined before removing memory. Check
1894 * whether all memory blocks in question are offline and trigger a BUG()
1895 * if this is not the case.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001896 */
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001897 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
Xishi Qiud6de9d52013-11-12 15:07:20 -08001898 check_memblock_offlined_cb);
Wen Congyangbbc76be2013-02-22 16:32:54 -08001899 if (ret) {
1900 unlock_memory_hotplug();
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02001901 BUG();
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001902 }
1903
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001904 /* remove memmap entry */
1905 firmware_map_remove(start, start + size, "System RAM");
1906
Wen Congyang24d335c2013-02-22 16:32:58 -08001907 arch_remove_memory(start, size);
1908
Tang Chen60a5a192013-02-22 16:33:14 -08001909 try_offline_node(nid);
1910
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001911 unlock_memory_hotplug();
Badari Pulavarty71088782008-10-18 20:25:58 -07001912}
Badari Pulavarty71088782008-10-18 20:25:58 -07001913EXPORT_SYMBOL_GPL(remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02001914#endif /* CONFIG_MEMORY_HOTREMOVE */