blob: 1ad92b46753edfe8d9f54fae81c5f3aa15b34110 [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>
Dave Hansen3947be12005-10-29 18:16:54 -070033
34#include <asm/tlbflush.h>
35
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030036#include "internal.h"
37
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070038/*
39 * online_page_callback contains pointer to current page onlining function.
40 * Initially it is generic_online_page(). If it is required it could be
41 * changed by calling set_online_page_callback() for callback registration
42 * and restore_online_page_callback() for generic callback restore.
43 */
44
45static void generic_online_page(struct page *page);
46
47static online_page_callback_t online_page_callback = generic_online_page;
48
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080049DEFINE_MUTEX(mem_hotplug_mutex);
50
51void lock_memory_hotplug(void)
52{
53 mutex_lock(&mem_hotplug_mutex);
54
55 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
56 lock_system_sleep();
57}
58
59void unlock_memory_hotplug(void)
60{
61 unlock_system_sleep();
62 mutex_unlock(&mem_hotplug_mutex);
63}
64
65
Keith Mannthey45e0b782006-09-30 23:27:09 -070066/* add this memory to iomem resource */
67static struct resource *register_memory_resource(u64 start, u64 size)
68{
69 struct resource *res;
70 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
71 BUG_ON(!res);
72
73 res->name = "System RAM";
74 res->start = start;
75 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080076 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070077 if (request_resource(&iomem_resource, res) < 0) {
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -070078 printk("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -070079 kfree(res);
80 res = NULL;
81 }
82 return res;
83}
84
85static void release_memory_resource(struct resource *res)
86{
87 if (!res)
88 return;
89 release_resource(res);
90 kfree(res);
91 return;
92}
93
Keith Mannthey53947022006-09-30 23:27:08 -070094#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -080095void get_page_bootmem(unsigned long info, struct page *page,
96 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -070097{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -080098 page->lru.next = (struct list_head *) type;
Yasunori Goto04753272008-04-28 02:13:31 -070099 SetPagePrivate(page);
100 set_page_private(page, info);
101 atomic_inc(&page->_count);
102}
103
Rakib Mullick23ce9322009-12-14 17:59:44 -0800104/* reference to __meminit __free_pages_bootmem is valid
105 * so use __ref to tell modpost not to generate a warning */
106void __ref put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -0700107{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800108 unsigned long type;
Jiang Liu9feedc92012-12-12 13:52:12 -0800109 static DEFINE_MUTEX(ppb_lock);
Yasunori Goto04753272008-04-28 02:13:31 -0700110
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800111 type = (unsigned long) page->lru.next;
112 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
113 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700114
115 if (atomic_dec_return(&page->_count) == 1) {
116 ClearPagePrivate(page);
117 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800118 INIT_LIST_HEAD(&page->lru);
Jiang Liu9feedc92012-12-12 13:52:12 -0800119
120 /*
121 * Please refer to comment for __free_pages_bootmem()
122 * for why we serialize here.
123 */
124 mutex_lock(&ppb_lock);
Yasunori Goto04753272008-04-28 02:13:31 -0700125 __free_pages_bootmem(page, 0);
Jiang Liu9feedc92012-12-12 13:52:12 -0800126 mutex_unlock(&ppb_lock);
Jiang Liuc60514b2013-02-22 16:33:56 -0800127 totalram_pages++;
Yasunori Goto04753272008-04-28 02:13:31 -0700128 }
129
130}
131
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800132#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
133#ifndef CONFIG_SPARSEMEM_VMEMMAP
Adrian Bunkd92bc312008-07-23 21:28:12 -0700134static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700135{
136 unsigned long *usemap, mapsize, section_nr, i;
137 struct mem_section *ms;
138 struct page *page, *memmap;
139
Yasunori Goto04753272008-04-28 02:13:31 -0700140 section_nr = pfn_to_section_nr(start_pfn);
141 ms = __nr_to_section(section_nr);
142
143 /* Get section's memmap address */
144 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
145
146 /*
147 * Get page for the memmap's phys address
148 * XXX: need more consideration for sparse_vmemmap...
149 */
150 page = virt_to_page(memmap);
151 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
152 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
153
154 /* remember memmap's page */
155 for (i = 0; i < mapsize; i++, page++)
156 get_page_bootmem(section_nr, page, SECTION_INFO);
157
158 usemap = __nr_to_section(section_nr)->pageblock_flags;
159 page = virt_to_page(usemap);
160
161 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
162
163 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700164 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700165
166}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800167#else /* CONFIG_SPARSEMEM_VMEMMAP */
168static void register_page_bootmem_info_section(unsigned long start_pfn)
169{
170 unsigned long *usemap, mapsize, section_nr, i;
171 struct mem_section *ms;
172 struct page *page, *memmap;
173
174 if (!pfn_valid(start_pfn))
175 return;
176
177 section_nr = pfn_to_section_nr(start_pfn);
178 ms = __nr_to_section(section_nr);
179
180 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
181
182 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
183
184 usemap = __nr_to_section(section_nr)->pageblock_flags;
185 page = virt_to_page(usemap);
186
187 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
188
189 for (i = 0; i < mapsize; i++, page++)
190 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
191}
192#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
Yasunori Goto04753272008-04-28 02:13:31 -0700193
194void register_page_bootmem_info_node(struct pglist_data *pgdat)
195{
196 unsigned long i, pfn, end_pfn, nr_pages;
197 int node = pgdat->node_id;
198 struct page *page;
199 struct zone *zone;
200
201 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
202 page = virt_to_page(pgdat);
203
204 for (i = 0; i < nr_pages; i++, page++)
205 get_page_bootmem(node, page, NODE_INFO);
206
207 zone = &pgdat->node_zones[0];
208 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
209 if (zone->wait_table) {
210 nr_pages = zone->wait_table_hash_nr_entries
211 * sizeof(wait_queue_head_t);
212 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
213 page = virt_to_page(zone->wait_table);
214
215 for (i = 0; i < nr_pages; i++, page++)
216 get_page_bootmem(node, page, NODE_INFO);
217 }
218 }
219
220 pfn = pgdat->node_start_pfn;
Cody P Schaferc1f19492013-02-22 16:35:32 -0800221 end_pfn = pgdat_end_pfn(pgdat);
Yasunori Goto04753272008-04-28 02:13:31 -0700222
223 /* register_section info */
qiuxishif14851a2012-09-17 14:09:24 -0700224 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
225 /*
226 * Some platforms can assign the same pfn to multiple nodes - on
227 * node0 as well as nodeN. To avoid registering a pfn against
228 * multiple nodes we check that this pfn does not already
229 * reside in some other node.
230 */
231 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
232 register_page_bootmem_info_section(pfn);
233 }
Yasunori Goto04753272008-04-28 02:13:31 -0700234}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800235#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
Yasunori Goto04753272008-04-28 02:13:31 -0700236
Heiko Carstens76cdd582008-05-14 16:05:52 -0700237static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
238 unsigned long end_pfn)
239{
240 unsigned long old_zone_end_pfn;
241
242 zone_span_writelock(zone);
243
244 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
Tang Chen712cd382012-12-11 16:01:07 -0800245 if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700246 zone->zone_start_pfn = start_pfn;
247
248 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
249 zone->zone_start_pfn;
250
251 zone_span_writeunlock(zone);
252}
253
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800254static void resize_zone(struct zone *zone, unsigned long start_pfn,
255 unsigned long end_pfn)
256{
257 zone_span_writelock(zone);
258
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800259 if (end_pfn - start_pfn) {
260 zone->zone_start_pfn = start_pfn;
261 zone->spanned_pages = end_pfn - start_pfn;
262 } else {
263 /*
264 * make it consist as free_area_init_core(),
265 * if spanned_pages = 0, then keep start_pfn = 0
266 */
267 zone->zone_start_pfn = 0;
268 zone->spanned_pages = 0;
269 }
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800270
271 zone_span_writeunlock(zone);
272}
273
274static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
275 unsigned long end_pfn)
276{
277 enum zone_type zid = zone_idx(zone);
278 int nid = zone->zone_pgdat->node_id;
279 unsigned long pfn;
280
281 for (pfn = start_pfn; pfn < end_pfn; pfn++)
282 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
283}
284
Cody P Schaferf6bbb782013-02-22 16:35:30 -0800285/* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
286 * alloc_bootmem_node_nopanic() */
287static int __ref ensure_zone_is_initialized(struct zone *zone,
288 unsigned long start_pfn, unsigned long num_pages)
289{
290 if (!zone_is_initialized(zone))
291 return init_currently_empty_zone(zone, start_pfn, num_pages,
292 MEMMAP_HOTPLUG);
293 return 0;
294}
295
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800296static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800297 unsigned long start_pfn, unsigned long end_pfn)
298{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800299 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800300 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800301 unsigned long z1_start_pfn;
302
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800303 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
304 if (ret)
305 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800306
307 pgdat_resize_lock(z1->zone_pgdat, &flags);
308
309 /* can't move pfns which are higher than @z2 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800310 if (end_pfn > zone_end_pfn(z2))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800311 goto out_fail;
312 /* the move out part mast at the left most of @z2 */
313 if (start_pfn > z2->zone_start_pfn)
314 goto out_fail;
315 /* must included/overlap */
316 if (end_pfn <= z2->zone_start_pfn)
317 goto out_fail;
318
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800319 /* use start_pfn for z1's start_pfn if z1 is empty */
320 if (z1->spanned_pages)
321 z1_start_pfn = z1->zone_start_pfn;
322 else
323 z1_start_pfn = start_pfn;
324
325 resize_zone(z1, z1_start_pfn, end_pfn);
Cody P Schafer108bcc92013-02-22 16:35:23 -0800326 resize_zone(z2, end_pfn, zone_end_pfn(z2));
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800327
328 pgdat_resize_unlock(z1->zone_pgdat, &flags);
329
330 fix_zone_id(z1, start_pfn, end_pfn);
331
332 return 0;
333out_fail:
334 pgdat_resize_unlock(z1->zone_pgdat, &flags);
335 return -1;
336}
337
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800338static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800339 unsigned long start_pfn, unsigned long end_pfn)
340{
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800341 int ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800342 unsigned long flags;
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800343 unsigned long z2_end_pfn;
344
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800345 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
346 if (ret)
347 return ret;
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800348
349 pgdat_resize_lock(z1->zone_pgdat, &flags);
350
351 /* can't move pfns which are lower than @z1 */
352 if (z1->zone_start_pfn > start_pfn)
353 goto out_fail;
354 /* the move out part mast at the right most of @z1 */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800355 if (zone_end_pfn(z1) > end_pfn)
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800356 goto out_fail;
357 /* must included/overlap */
Cody P Schafer108bcc92013-02-22 16:35:23 -0800358 if (start_pfn >= zone_end_pfn(z1))
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800359 goto out_fail;
360
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800361 /* use end_pfn for z2's end_pfn if z2 is empty */
362 if (z2->spanned_pages)
Cody P Schafer108bcc92013-02-22 16:35:23 -0800363 z2_end_pfn = zone_end_pfn(z2);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800364 else
365 z2_end_pfn = end_pfn;
366
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800367 resize_zone(z1, z1->zone_start_pfn, start_pfn);
Lai Jiangshane455a9b2012-12-11 16:03:20 -0800368 resize_zone(z2, start_pfn, z2_end_pfn);
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800369
370 pgdat_resize_unlock(z1->zone_pgdat, &flags);
371
372 fix_zone_id(z2, start_pfn, end_pfn);
373
374 return 0;
375out_fail:
376 pgdat_resize_unlock(z1->zone_pgdat, &flags);
377 return -1;
378}
379
Heiko Carstens76cdd582008-05-14 16:05:52 -0700380static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
381 unsigned long end_pfn)
382{
383 unsigned long old_pgdat_end_pfn =
384 pgdat->node_start_pfn + pgdat->node_spanned_pages;
385
Tang Chen712cd382012-12-11 16:01:07 -0800386 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
Heiko Carstens76cdd582008-05-14 16:05:52 -0700387 pgdat->node_start_pfn = start_pfn;
388
389 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
390 pgdat->node_start_pfn;
391}
392
Al Viro31168482008-11-22 17:33:24 +0000393static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700394{
395 struct pglist_data *pgdat = zone->zone_pgdat;
396 int nr_pages = PAGES_PER_SECTION;
397 int nid = pgdat->node_id;
398 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700399 unsigned long flags;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800400 int ret;
Dave Hansen3947be12005-10-29 18:16:54 -0700401
402 zone_type = zone - pgdat->node_zones;
Cody P Schafer64dd1b22013-02-22 16:35:31 -0800403 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
404 if (ret)
405 return ret;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700406
Heiko Carstens76cdd582008-05-14 16:05:52 -0700407 pgdat_resize_lock(zone->zone_pgdat, &flags);
408 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
409 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
410 phys_start_pfn + nr_pages);
411 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800412 memmap_init_zone(nr_pages, nid, zone_type,
413 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700414 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700415}
416
Gary Hadec04fc582009-01-06 14:39:14 -0800417static int __meminit __add_section(int nid, struct zone *zone,
418 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700419{
Dave Hansen3947be12005-10-29 18:16:54 -0700420 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700421 int ret;
422
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700423 if (pfn_valid(phys_start_pfn))
424 return -EEXIST;
425
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700426 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700427
428 if (ret < 0)
429 return ret;
430
Yasunori Goto718127c2006-06-23 02:03:10 -0700431 ret = __add_zone(zone, phys_start_pfn);
432
433 if (ret < 0)
434 return ret;
435
Gary Hadec04fc582009-01-06 14:39:14 -0800436 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700437}
438
David Rientjes4edd7ce2013-04-29 15:08:22 -0700439/*
440 * Reasonably generic function for adding memory. It is
441 * expected that archs that support memory hotplug will
442 * call this function after deciding the zone to which to
443 * add the new pages.
444 */
445int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
446 unsigned long nr_pages)
447{
448 unsigned long i;
449 int err = 0;
450 int start_sec, end_sec;
451 /* during initialize mem_map, align hot-added range to section */
452 start_sec = pfn_to_section_nr(phys_start_pfn);
453 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
454
455 for (i = start_sec; i <= end_sec; i++) {
456 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
457
458 /*
459 * EEXIST is finally dealt with by ioresource collision
460 * check. see add_memory() => register_memory_resource()
461 * Warning will be printed if there is collision.
462 */
463 if (err && (err != -EEXIST))
464 break;
465 err = 0;
466 }
467
468 return err;
469}
470EXPORT_SYMBOL_GPL(__add_pages);
471
472#ifdef CONFIG_MEMORY_HOTREMOVE
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800473/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
474static int find_smallest_section_pfn(int nid, struct zone *zone,
475 unsigned long start_pfn,
476 unsigned long end_pfn)
477{
478 struct mem_section *ms;
479
480 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
481 ms = __pfn_to_section(start_pfn);
482
483 if (unlikely(!valid_section(ms)))
484 continue;
485
486 if (unlikely(pfn_to_nid(start_pfn) != nid))
487 continue;
488
489 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
490 continue;
491
492 return start_pfn;
493 }
494
495 return 0;
496}
497
498/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
499static int find_biggest_section_pfn(int nid, struct zone *zone,
500 unsigned long start_pfn,
501 unsigned long end_pfn)
502{
503 struct mem_section *ms;
504 unsigned long pfn;
505
506 /* pfn is the end pfn of a memory section. */
507 pfn = end_pfn - 1;
508 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
509 ms = __pfn_to_section(pfn);
510
511 if (unlikely(!valid_section(ms)))
512 continue;
513
514 if (unlikely(pfn_to_nid(pfn) != nid))
515 continue;
516
517 if (zone && zone != page_zone(pfn_to_page(pfn)))
518 continue;
519
520 return pfn;
521 }
522
523 return 0;
524}
525
526static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
527 unsigned long end_pfn)
528{
529 unsigned long zone_start_pfn = zone->zone_start_pfn;
530 unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
531 unsigned long pfn;
532 struct mem_section *ms;
533 int nid = zone_to_nid(zone);
534
535 zone_span_writelock(zone);
536 if (zone_start_pfn == start_pfn) {
537 /*
538 * If the section is smallest section in the zone, it need
539 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
540 * In this case, we find second smallest valid mem_section
541 * for shrinking zone.
542 */
543 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
544 zone_end_pfn);
545 if (pfn) {
546 zone->zone_start_pfn = pfn;
547 zone->spanned_pages = zone_end_pfn - pfn;
548 }
549 } else if (zone_end_pfn == end_pfn) {
550 /*
551 * If the section is biggest section in the zone, it need
552 * shrink zone->spanned_pages.
553 * In this case, we find second biggest valid mem_section for
554 * shrinking zone.
555 */
556 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
557 start_pfn);
558 if (pfn)
559 zone->spanned_pages = pfn - zone_start_pfn + 1;
560 }
561
562 /*
563 * The section is not biggest or smallest mem_section in the zone, it
564 * only creates a hole in the zone. So in this case, we need not
565 * change the zone. But perhaps, the zone has only hole data. Thus
566 * it check the zone has only hole or not.
567 */
568 pfn = zone_start_pfn;
569 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
570 ms = __pfn_to_section(pfn);
571
572 if (unlikely(!valid_section(ms)))
573 continue;
574
575 if (page_zone(pfn_to_page(pfn)) != zone)
576 continue;
577
578 /* If the section is current section, it continues the loop */
579 if (start_pfn == pfn)
580 continue;
581
582 /* If we find valid section, we have nothing to do */
583 zone_span_writeunlock(zone);
584 return;
585 }
586
587 /* The zone has no valid section */
588 zone->zone_start_pfn = 0;
589 zone->spanned_pages = 0;
590 zone_span_writeunlock(zone);
591}
592
593static void shrink_pgdat_span(struct pglist_data *pgdat,
594 unsigned long start_pfn, unsigned long end_pfn)
595{
596 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
597 unsigned long pgdat_end_pfn =
598 pgdat->node_start_pfn + pgdat->node_spanned_pages;
599 unsigned long pfn;
600 struct mem_section *ms;
601 int nid = pgdat->node_id;
602
603 if (pgdat_start_pfn == start_pfn) {
604 /*
605 * If the section is smallest section in the pgdat, it need
606 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
607 * In this case, we find second smallest valid mem_section
608 * for shrinking zone.
609 */
610 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
611 pgdat_end_pfn);
612 if (pfn) {
613 pgdat->node_start_pfn = pfn;
614 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
615 }
616 } else if (pgdat_end_pfn == end_pfn) {
617 /*
618 * If the section is biggest section in the pgdat, it need
619 * shrink pgdat->node_spanned_pages.
620 * In this case, we find second biggest valid mem_section for
621 * shrinking zone.
622 */
623 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
624 start_pfn);
625 if (pfn)
626 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
627 }
628
629 /*
630 * If the section is not biggest or smallest mem_section in the pgdat,
631 * it only creates a hole in the pgdat. So in this case, we need not
632 * change the pgdat.
633 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
634 * has only hole or not.
635 */
636 pfn = pgdat_start_pfn;
637 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
638 ms = __pfn_to_section(pfn);
639
640 if (unlikely(!valid_section(ms)))
641 continue;
642
643 if (pfn_to_nid(pfn) != nid)
644 continue;
645
646 /* If the section is current section, it continues the loop */
647 if (start_pfn == pfn)
648 continue;
649
650 /* If we find valid section, we have nothing to do */
651 return;
652 }
653
654 /* The pgdat has no valid section */
655 pgdat->node_start_pfn = 0;
656 pgdat->node_spanned_pages = 0;
657}
658
659static void __remove_zone(struct zone *zone, unsigned long start_pfn)
660{
661 struct pglist_data *pgdat = zone->zone_pgdat;
662 int nr_pages = PAGES_PER_SECTION;
663 int zone_type;
664 unsigned long flags;
665
666 zone_type = zone - pgdat->node_zones;
667
668 pgdat_resize_lock(zone->zone_pgdat, &flags);
669 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
670 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
671 pgdat_resize_unlock(zone->zone_pgdat, &flags);
672}
673
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700674static int __remove_section(struct zone *zone, struct mem_section *ms)
675{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800676 unsigned long start_pfn;
677 int scn_nr;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700678 int ret = -EINVAL;
679
680 if (!valid_section(ms))
681 return ret;
682
683 ret = unregister_memory_section(ms);
684 if (ret)
685 return ret;
686
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800687 scn_nr = __section_nr(ms);
688 start_pfn = section_nr_to_pfn(scn_nr);
689 __remove_zone(zone, start_pfn);
690
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700691 sparse_remove_one_section(zone, ms);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700692 return 0;
693}
694
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700695/**
696 * __remove_pages() - remove sections of pages from a zone
697 * @zone: zone from which pages need to be removed
698 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
699 * @nr_pages: number of pages to remove (must be multiple of section size)
700 *
701 * Generic helper function to remove section mappings and sysfs entries
702 * for the section of the memory we are removing. Caller needs to make
703 * sure that pages are marked reserved and zones are adjust properly by
704 * calling offline_pages().
705 */
706int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
707 unsigned long nr_pages)
708{
Toshi Kanife74ebb2013-04-29 15:08:20 -0700709 unsigned long i;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700710 int sections_to_remove;
Toshi Kanife74ebb2013-04-29 15:08:20 -0700711 resource_size_t start, size;
712 int ret = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700713
714 /*
715 * We can only remove entire sections
716 */
717 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
718 BUG_ON(nr_pages % PAGES_PER_SECTION);
719
Toshi Kanife74ebb2013-04-29 15:08:20 -0700720 start = phys_start_pfn << PAGE_SHIFT;
721 size = nr_pages * PAGE_SIZE;
722 ret = release_mem_region_adjustable(&iomem_resource, start, size);
Randy Dunlap348f9f02013-05-24 15:55:30 -0700723 if (ret) {
724 resource_size_t endres = start + size - 1;
725
726 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
727 &start, &endres, ret);
728 }
Yasuaki Ishimatsud760afd2012-10-08 16:34:14 -0700729
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700730 sections_to_remove = nr_pages / PAGES_PER_SECTION;
731 for (i = 0; i < sections_to_remove; i++) {
732 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
733 ret = __remove_section(zone, __pfn_to_section(pfn));
734 if (ret)
735 break;
736 }
737 return ret;
738}
739EXPORT_SYMBOL_GPL(__remove_pages);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700740#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700741
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700742int set_online_page_callback(online_page_callback_t callback)
743{
744 int rc = -EINVAL;
745
746 lock_memory_hotplug();
747
748 if (online_page_callback == generic_online_page) {
749 online_page_callback = callback;
750 rc = 0;
751 }
752
753 unlock_memory_hotplug();
754
755 return rc;
756}
757EXPORT_SYMBOL_GPL(set_online_page_callback);
758
759int restore_online_page_callback(online_page_callback_t callback)
760{
761 int rc = -EINVAL;
762
763 lock_memory_hotplug();
764
765 if (online_page_callback == callback) {
766 online_page_callback = generic_online_page;
767 rc = 0;
768 }
769
770 unlock_memory_hotplug();
771
772 return rc;
773}
774EXPORT_SYMBOL_GPL(restore_online_page_callback);
775
776void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700777{
Jan Beulich4738e1b2009-09-21 17:03:03 -0700778 unsigned long pfn = page_to_pfn(page);
779
Jan Beulich4738e1b2009-09-21 17:03:03 -0700780 if (pfn >= num_physpages)
781 num_physpages = pfn + 1;
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700782}
783EXPORT_SYMBOL_GPL(__online_page_set_limits);
784
785void __online_page_increment_counters(struct page *page)
786{
787 totalram_pages++;
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700788
789#ifdef CONFIG_HIGHMEM
790 if (PageHighMem(page))
791 totalhigh_pages++;
792#endif
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700793}
794EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700795
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700796void __online_page_free(struct page *page)
797{
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700798 ClearPageReserved(page);
799 init_page_count(page);
800 __free_page(page);
801}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700802EXPORT_SYMBOL_GPL(__online_page_free);
803
804static void generic_online_page(struct page *page)
805{
806 __online_page_set_limits(page);
807 __online_page_increment_counters(page);
808 __online_page_free(page);
809}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700810
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700811static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
812 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700813{
814 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700815 unsigned long onlined_pages = *(unsigned long *)arg;
816 struct page *page;
817 if (PageReserved(pfn_to_page(start_pfn)))
818 for (i = 0; i < nr_pages; i++) {
819 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700820 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700821 onlined_pages++;
822 }
823 *(unsigned long *)arg = onlined_pages;
824 return 0;
825}
826
Lai Jiangshan09285af2012-12-12 13:52:04 -0800827#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -0800828/*
829 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
830 * normal memory.
831 */
Lai Jiangshan09285af2012-12-12 13:52:04 -0800832static bool can_online_high_movable(struct zone *zone)
833{
834 return true;
835}
Tang Chen79a4dce2012-12-18 14:23:24 -0800836#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800837/* ensure every online node has NORMAL memory */
838static bool can_online_high_movable(struct zone *zone)
839{
840 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
841}
Tang Chen79a4dce2012-12-18 14:23:24 -0800842#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800843
Lai Jiangshand9713672012-12-11 16:01:03 -0800844/* check which state of node_states will be changed when online memory */
845static void node_states_check_changes_online(unsigned long nr_pages,
846 struct zone *zone, struct memory_notify *arg)
847{
848 int nid = zone_to_nid(zone);
849 enum zone_type zone_last = ZONE_NORMAL;
850
851 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800852 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
853 * contains nodes which have zones of 0...ZONE_NORMAL,
854 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -0800855 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800856 * If we don't have HIGHMEM nor movable node,
857 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
858 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -0800859 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800860 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -0800861 zone_last = ZONE_MOVABLE;
862
863 /*
864 * if the memory to be online is in a zone of 0...zone_last, and
865 * the zones of 0...zone_last don't have memory before online, we will
866 * need to set the node to node_states[N_NORMAL_MEMORY] after
867 * the memory is online.
868 */
869 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
870 arg->status_change_nid_normal = nid;
871 else
872 arg->status_change_nid_normal = -1;
873
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800874#ifdef CONFIG_HIGHMEM
875 /*
876 * If we have movable node, node_states[N_HIGH_MEMORY]
877 * contains nodes which have zones of 0...ZONE_HIGHMEM,
878 * set zone_last to ZONE_HIGHMEM.
879 *
880 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
881 * contains nodes which have zones of 0...ZONE_MOVABLE,
882 * set zone_last to ZONE_MOVABLE.
883 */
884 zone_last = ZONE_HIGHMEM;
885 if (N_MEMORY == N_HIGH_MEMORY)
886 zone_last = ZONE_MOVABLE;
887
888 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
889 arg->status_change_nid_high = nid;
890 else
891 arg->status_change_nid_high = -1;
892#else
893 arg->status_change_nid_high = arg->status_change_nid_normal;
894#endif
895
Lai Jiangshand9713672012-12-11 16:01:03 -0800896 /*
897 * if the node don't have memory befor online, we will need to
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800898 * set the node to node_states[N_MEMORY] after the memory
Lai Jiangshand9713672012-12-11 16:01:03 -0800899 * is online.
900 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800901 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800902 arg->status_change_nid = nid;
903 else
904 arg->status_change_nid = -1;
905}
906
907static void node_states_set_node(int node, struct memory_notify *arg)
908{
909 if (arg->status_change_nid_normal >= 0)
910 node_set_state(node, N_NORMAL_MEMORY);
911
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800912 if (arg->status_change_nid_high >= 0)
913 node_set_state(node, N_HIGH_MEMORY);
914
915 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800916}
917
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700918
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800919int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700920{
Dave Hansen3947be12005-10-29 18:16:54 -0700921 unsigned long onlined_pages = 0;
922 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700923 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700924 int nid;
925 int ret;
926 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700927
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900928 lock_memory_hotplug();
Lai Jiangshand9713672012-12-11 16:01:03 -0800929 /*
930 * This doesn't need a lock to do pfn_to_page().
931 * The section can't be removed here because of the
932 * memory_block->state_mutex.
933 */
934 zone = page_zone(pfn_to_page(pfn));
935
Lai Jiangshan74d42d82012-12-11 16:03:23 -0800936 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
937 !can_online_high_movable(zone)) {
938 unlock_memory_hotplug();
939 return -1;
940 }
941
Lai Jiangshan511c2ab2012-12-11 16:03:16 -0800942 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
943 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
944 unlock_memory_hotplug();
945 return -1;
946 }
947 }
948 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
949 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
950 unlock_memory_hotplug();
951 return -1;
952 }
953 }
954
955 /* Previous code may changed the zone of the pfn range */
956 zone = page_zone(pfn_to_page(pfn));
957
Yasunori Goto7b78d332007-10-21 16:41:36 -0700958 arg.start_pfn = pfn;
959 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -0800960 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700961
962 nid = page_to_nid(pfn_to_page(pfn));
Yasunori Goto7b78d332007-10-21 16:41:36 -0700963
964 ret = memory_notify(MEM_GOING_ONLINE, &arg);
965 ret = notifier_to_errno(ret);
966 if (ret) {
967 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900968 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700969 return ret;
970 }
Dave Hansen3947be12005-10-29 18:16:54 -0700971 /*
Yasunori Goto68113782006-06-23 02:03:11 -0700972 * If this zone is not populated, then it is not in zonelist.
973 * This means the page allocator ignores this zone.
974 * So, zonelist must be updated after online.
975 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700976 mutex_lock(&zonelists_mutex);
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800977 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -0700978 need_zonelists_rebuild = 1;
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800979 build_all_zonelists(NULL, zone);
980 }
Yasunori Goto68113782006-06-23 02:03:11 -0700981
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700982 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700983 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700984 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800985 if (need_zonelists_rebuild)
986 zone_pcp_reset(zone);
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700987 mutex_unlock(&zonelists_mutex);
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -0700988 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
989 (unsigned long long) pfn << PAGE_SHIFT,
990 (((unsigned long long) pfn + nr_pages)
991 << PAGE_SHIFT) - 1);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700992 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900993 unlock_memory_hotplug();
Geoff Levandfd8a4222008-05-14 16:05:50 -0700994 return ret;
995 }
996
Jiang Liu9feedc92012-12-12 13:52:12 -0800997 zone->managed_pages += onlined_pages;
Dave Hansen3947be12005-10-29 18:16:54 -0700998 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800999 zone->zone_pgdat->node_present_pages += onlined_pages;
Jiang Liu08dff7b2012-07-31 16:43:30 -07001000 if (onlined_pages) {
Lai Jiangshand9713672012-12-11 16:01:03 -08001001 node_states_set_node(zone_to_nid(zone), &arg);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001002 if (need_zonelists_rebuild)
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001003 build_all_zonelists(NULL, NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -07001004 else
1005 zone_pcp_update(zone);
1006 }
Dave Hansen3947be12005-10-29 18:16:54 -07001007
Haicheng Li4eaf3f62010-05-24 14:32:52 -07001008 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001009
1010 init_per_zone_wmark_min();
1011
Jiang Liu08dff7b2012-07-31 16:43:30 -07001012 if (onlined_pages)
Christoph Lameter7ea15302007-10-16 01:25:29 -07001013 kswapd_run(zone_to_nid(zone));
Dave Hansen61b13992005-10-29 18:16:56 -07001014
Haicheng Li1f522502010-05-24 14:32:51 -07001015 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -07001016
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -07001017 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001018
1019 if (onlined_pages)
1020 memory_notify(MEM_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +09001021 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001022
Dave Hansen3947be12005-10-29 18:16:54 -07001023 return 0;
1024}
Keith Mannthey53947022006-09-30 23:27:08 -07001025#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -07001026
Hidetoshi Setoe1319332009-11-17 14:06:18 -08001027/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1028static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001029{
1030 struct pglist_data *pgdat;
1031 unsigned long zones_size[MAX_NR_ZONES] = {0};
1032 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1033 unsigned long start_pfn = start >> PAGE_SHIFT;
1034
Tang Chena1e565a2013-02-22 16:33:18 -08001035 pgdat = NODE_DATA(nid);
1036 if (!pgdat) {
1037 pgdat = arch_alloc_nodedata(nid);
1038 if (!pgdat)
1039 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001040
Tang Chena1e565a2013-02-22 16:33:18 -08001041 arch_refresh_nodedata(nid, pgdat);
1042 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001043
1044 /* we can use NODE_DATA(nid) from here */
1045
1046 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -07001047 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001048
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001049 /*
1050 * The node we allocated has no zone fallback lists. For avoiding
1051 * to access not-initialized zonelist, build here.
1052 */
David Rientjesf957db42011-06-22 18:13:04 -07001053 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07001054 build_all_zonelists(pgdat, NULL);
David Rientjesf957db42011-06-22 18:13:04 -07001055 mutex_unlock(&zonelists_mutex);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001056
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001057 return pgdat;
1058}
1059
1060static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1061{
1062 arch_refresh_nodedata(nid, NULL);
1063 arch_free_nodedata(pgdat);
1064 return;
1065}
1066
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001067
minskey guocf234222010-05-24 14:32:41 -07001068/*
1069 * called by cpu_up() to online a node without onlined memory.
1070 */
1071int mem_online_node(int nid)
1072{
1073 pg_data_t *pgdat;
1074 int ret;
1075
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001076 lock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -07001077 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -07001078 if (!pgdat) {
minskey guocf234222010-05-24 14:32:41 -07001079 ret = -ENOMEM;
1080 goto out;
1081 }
1082 node_set_online(nid);
1083 ret = register_one_node(nid);
1084 BUG_ON(ret);
1085
1086out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001087 unlock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -07001088 return ret;
1089}
1090
Al Viro31168482008-11-22 17:33:24 +00001091/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1092int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001093{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001094 pg_data_t *pgdat = NULL;
Tang Chena1e565a2013-02-22 16:33:18 -08001095 bool new_pgdat;
1096 bool new_node;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001097 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001098 int ret;
1099
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001100 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -08001101
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001102 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001103 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001104 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -08001105 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -07001106
Tang Chena1e565a2013-02-22 16:33:18 -08001107 { /* Stupid hack to suppress address-never-null warning */
1108 void *p = NODE_DATA(nid);
1109 new_pgdat = !p;
1110 }
1111 new_node = !node_online(nid);
1112 if (new_node) {
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001113 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -08001114 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001115 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -07001116 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001117 }
1118
Yasunori Gotobc02af92006-06-27 02:53:30 -07001119 /* call arch's memory hotadd */
1120 ret = arch_add_memory(nid, start, size);
1121
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001122 if (ret < 0)
1123 goto error;
1124
Yasunori Goto0fc44152006-06-27 02:53:38 -07001125 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001126 node_set_online(nid);
1127
Tang Chena1e565a2013-02-22 16:33:18 -08001128 if (new_node) {
Yasunori Goto0fc44152006-06-27 02:53:38 -07001129 ret = register_one_node(nid);
1130 /*
1131 * If sysfs file of new node can't create, cpu on the node
1132 * can't be hot-added. There is no rollback way now.
1133 * So, check by BUG_ON() to catch it reluctantly..
1134 */
1135 BUG_ON(ret);
1136 }
1137
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001138 /* create new memmap entry */
1139 firmware_map_add_hotplug(start, start + size, "System RAM");
1140
Andi Kleen6ad696d2009-11-17 14:06:22 -08001141 goto out;
1142
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001143error:
1144 /* rollback pgdat allocation and others */
1145 if (new_pgdat)
1146 rollback_node_hotadd(nid, pgdat);
Sasha Levina864b9d2013-02-22 16:32:48 -08001147 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001148
Andi Kleen6ad696d2009-11-17 14:06:22 -08001149out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001150 unlock_memory_hotplug();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001151 return ret;
1152}
1153EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001154
1155#ifdef CONFIG_MEMORY_HOTREMOVE
1156/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001157 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1158 * set and the size of the free page is given by page_order(). Using this,
1159 * the function determines if the pageblock contains only free pages.
1160 * Due to buddy contraints, a free page at least the size of a pageblock will
1161 * be located at the start of the pageblock
1162 */
1163static inline int pageblock_free(struct page *page)
1164{
1165 return PageBuddy(page) && page_order(page) >= pageblock_order;
1166}
1167
1168/* Return the start of the next active pageblock after a given page */
1169static struct page *next_active_pageblock(struct page *page)
1170{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001171 /* Ensure the starting page is pageblock-aligned */
1172 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1173
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001174 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001175 if (pageblock_free(page)) {
1176 int order;
1177 /* be careful. we don't have locks, page_order can be changed.*/
1178 order = page_order(page);
1179 if ((order < MAX_ORDER) && (order >= pageblock_order))
1180 return page + (1 << order);
1181 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001182
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -07001183 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001184}
1185
1186/* Checks if this range of memory is likely to be hot-removable. */
1187int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1188{
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001189 struct page *page = pfn_to_page(start_pfn);
1190 struct page *end_page = page + nr_pages;
1191
1192 /* Check the starting page of each pageblock within the range */
1193 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001194 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001195 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07001196 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -07001197 }
1198
1199 /* All pageblocks in the memory block are likely to be hot-removable */
1200 return 1;
1201}
1202
1203/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001204 * Confirm all pages in a range [start, end) is belongs to the same zone.
1205 */
1206static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1207{
1208 unsigned long pfn;
1209 struct zone *zone = NULL;
1210 struct page *page;
1211 int i;
1212 for (pfn = start_pfn;
1213 pfn < end_pfn;
1214 pfn += MAX_ORDER_NR_PAGES) {
1215 i = 0;
1216 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1217 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1218 i++;
1219 if (i == MAX_ORDER_NR_PAGES)
1220 continue;
1221 page = pfn_to_page(pfn + i);
1222 if (zone && page_zone(page) != zone)
1223 return 0;
1224 zone = page_zone(page);
1225 }
1226 return 1;
1227}
1228
1229/*
1230 * Scanning pfn is much easier than scanning lru list.
1231 * Scan pfn from start to end and Find LRU page.
1232 */
Andrew Morton7bbc0902010-10-26 14:22:05 -07001233static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001234{
1235 unsigned long pfn;
1236 struct page *page;
1237 for (pfn = start; pfn < end; pfn++) {
1238 if (pfn_valid(pfn)) {
1239 page = pfn_to_page(pfn);
1240 if (PageLRU(page))
1241 return pfn;
1242 }
1243 }
1244 return 0;
1245}
1246
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001247#define NR_OFFLINE_AT_ONCE_PAGES (256)
1248static int
1249do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1250{
1251 unsigned long pfn;
1252 struct page *page;
1253 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1254 int not_managed = 0;
1255 int ret = 0;
1256 LIST_HEAD(source);
1257
1258 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1259 if (!pfn_valid(pfn))
1260 continue;
1261 page = pfn_to_page(pfn);
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001262 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001263 continue;
1264 /*
1265 * We can skip free pages. And we can only deal with pages on
1266 * LRU.
1267 */
Nick Piggin62695a82008-10-18 20:26:09 -07001268 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001269 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001270 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001271 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001272 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001273 inc_zone_page_state(page, NR_ISOLATED_ANON +
1274 page_is_file_cache(page));
1275
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001276 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001277#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -08001278 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1279 pfn);
1280 dump_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001281#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001282 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001283 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -07001284 check this again here. */
1285 if (page_count(page)) {
1286 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -07001287 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -07001288 break;
1289 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001290 }
1291 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001292 if (!list_empty(&source)) {
1293 if (not_managed) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001294 putback_lru_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -07001295 goto out;
1296 }
Minchan Kim74c08f92012-10-08 16:32:54 -07001297
1298 /*
1299 * alloc_migrate_target should be improooooved!!
1300 * migrate_pages returns # of failed pages.
1301 */
1302 ret = migrate_pages(&source, alloc_migrate_target, 0,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001303 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Bob Liuf3ab2632010-10-26 14:22:10 -07001304 if (ret)
1305 putback_lru_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001306 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001307out:
1308 return ret;
1309}
1310
1311/*
1312 * remove from free_area[] and mark all as Reserved.
1313 */
1314static int
1315offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1316 void *data)
1317{
1318 __offline_isolated_pages(start, start + nr_pages);
1319 return 0;
1320}
1321
1322static void
1323offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1324{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001325 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001326 offline_isolated_pages_cb);
1327}
1328
1329/*
1330 * Check all pages in range, recoreded as memory resource, are isolated.
1331 */
1332static int
1333check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1334 void *data)
1335{
1336 int ret;
1337 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -08001338 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001339 offlined = nr_pages;
1340 if (!ret)
1341 *(long *)data += offlined;
1342 return ret;
1343}
1344
1345static long
1346check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1347{
1348 long offlined = 0;
1349 int ret;
1350
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -07001351 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001352 check_pages_isolated_cb);
1353 if (ret < 0)
1354 offlined = (long)ret;
1355 return offlined;
1356}
1357
Lai Jiangshan09285af2012-12-12 13:52:04 -08001358#ifdef CONFIG_MOVABLE_NODE
Tang Chen79a4dce2012-12-18 14:23:24 -08001359/*
1360 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1361 * normal memory.
1362 */
Lai Jiangshan09285af2012-12-12 13:52:04 -08001363static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1364{
1365 return true;
1366}
Tang Chen79a4dce2012-12-18 14:23:24 -08001367#else /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001368/* ensure the node has NORMAL memory if it is still online */
1369static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1370{
1371 struct pglist_data *pgdat = zone->zone_pgdat;
1372 unsigned long present_pages = 0;
1373 enum zone_type zt;
1374
1375 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1376 present_pages += pgdat->node_zones[zt].present_pages;
1377
1378 if (present_pages > nr_pages)
1379 return true;
1380
1381 present_pages = 0;
1382 for (; zt <= ZONE_MOVABLE; zt++)
1383 present_pages += pgdat->node_zones[zt].present_pages;
1384
1385 /*
1386 * we can't offline the last normal memory until all
1387 * higher memory is offlined.
1388 */
1389 return present_pages == 0;
1390}
Tang Chen79a4dce2012-12-18 14:23:24 -08001391#endif /* CONFIG_MOVABLE_NODE */
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001392
Lai Jiangshand9713672012-12-11 16:01:03 -08001393/* check which state of node_states will be changed when offline memory */
1394static void node_states_check_changes_offline(unsigned long nr_pages,
1395 struct zone *zone, struct memory_notify *arg)
1396{
1397 struct pglist_data *pgdat = zone->zone_pgdat;
1398 unsigned long present_pages = 0;
1399 enum zone_type zt, zone_last = ZONE_NORMAL;
1400
1401 /*
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001402 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1403 * contains nodes which have zones of 0...ZONE_NORMAL,
1404 * set zone_last to ZONE_NORMAL.
Lai Jiangshand9713672012-12-11 16:01:03 -08001405 *
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001406 * If we don't have HIGHMEM nor movable node,
1407 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1408 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
Lai Jiangshand9713672012-12-11 16:01:03 -08001409 */
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001410 if (N_MEMORY == N_NORMAL_MEMORY)
Lai Jiangshand9713672012-12-11 16:01:03 -08001411 zone_last = ZONE_MOVABLE;
1412
1413 /*
1414 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1415 * If the memory to be offline is in a zone of 0...zone_last,
1416 * and it is the last present memory, 0...zone_last will
1417 * become empty after offline , thus we can determind we will
1418 * need to clear the node from node_states[N_NORMAL_MEMORY].
1419 */
1420 for (zt = 0; zt <= zone_last; zt++)
1421 present_pages += pgdat->node_zones[zt].present_pages;
1422 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1423 arg->status_change_nid_normal = zone_to_nid(zone);
1424 else
1425 arg->status_change_nid_normal = -1;
1426
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001427#ifdef CONFIG_HIGHMEM
1428 /*
1429 * If we have movable node, node_states[N_HIGH_MEMORY]
1430 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1431 * set zone_last to ZONE_HIGHMEM.
1432 *
1433 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1434 * contains nodes which have zones of 0...ZONE_MOVABLE,
1435 * set zone_last to ZONE_MOVABLE.
1436 */
1437 zone_last = ZONE_HIGHMEM;
1438 if (N_MEMORY == N_HIGH_MEMORY)
1439 zone_last = ZONE_MOVABLE;
1440
1441 for (; 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_high = zone_to_nid(zone);
1445 else
1446 arg->status_change_nid_high = -1;
1447#else
1448 arg->status_change_nid_high = arg->status_change_nid_normal;
1449#endif
1450
Lai Jiangshand9713672012-12-11 16:01:03 -08001451 /*
1452 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1453 */
1454 zone_last = ZONE_MOVABLE;
1455
1456 /*
1457 * check whether node_states[N_HIGH_MEMORY] will be changed
1458 * If we try to offline the last present @nr_pages from the node,
1459 * we can determind we will need to clear the node from
1460 * node_states[N_HIGH_MEMORY].
1461 */
1462 for (; zt <= zone_last; zt++)
1463 present_pages += pgdat->node_zones[zt].present_pages;
1464 if (nr_pages >= present_pages)
1465 arg->status_change_nid = zone_to_nid(zone);
1466 else
1467 arg->status_change_nid = -1;
1468}
1469
1470static void node_states_clear_node(int node, struct memory_notify *arg)
1471{
1472 if (arg->status_change_nid_normal >= 0)
1473 node_clear_state(node, N_NORMAL_MEMORY);
1474
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001475 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1476 (arg->status_change_nid_high >= 0))
Lai Jiangshand9713672012-12-11 16:01:03 -08001477 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001478
1479 if ((N_MEMORY != N_HIGH_MEMORY) &&
1480 (arg->status_change_nid >= 0))
1481 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001482}
1483
Wen Congyanga16cee12012-10-08 16:33:58 -07001484static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001485 unsigned long end_pfn, unsigned long timeout)
1486{
1487 unsigned long pfn, nr_pages, expire;
1488 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001489 int ret, drain, retry_max, node;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001490 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001491 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001492
1493 BUG_ON(start_pfn >= end_pfn);
1494 /* at least, alignment against pageblock is necessary */
1495 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1496 return -EINVAL;
1497 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1498 return -EINVAL;
1499 /* This makes hotplug much easier...and readable.
1500 we assume this for now. .*/
1501 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1502 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001503
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001504 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -08001505
Yasunori Goto7b78d332007-10-21 16:41:36 -07001506 zone = page_zone(pfn_to_page(start_pfn));
1507 node = zone_to_nid(zone);
1508 nr_pages = end_pfn - start_pfn;
1509
Lai Jiangshan74d42d82012-12-11 16:03:23 -08001510 ret = -EINVAL;
1511 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1512 goto out;
1513
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001514 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001515 ret = start_isolate_page_range(start_pfn, end_pfn,
1516 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001517 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -08001518 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001519
1520 arg.start_pfn = start_pfn;
1521 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001522 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001523
1524 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1525 ret = notifier_to_errno(ret);
1526 if (ret)
1527 goto failed_removal;
1528
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001529 pfn = start_pfn;
1530 expire = jiffies + timeout;
1531 drain = 0;
1532 retry_max = 5;
1533repeat:
1534 /* start memory hot removal */
1535 ret = -EAGAIN;
1536 if (time_after(jiffies, expire))
1537 goto failed_removal;
1538 ret = -EINTR;
1539 if (signal_pending(current))
1540 goto failed_removal;
1541 ret = 0;
1542 if (drain) {
1543 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001544 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001545 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001546 }
1547
1548 pfn = scan_lru_pages(start_pfn, end_pfn);
1549 if (pfn) { /* We have page on LRU */
1550 ret = do_migrate_range(pfn, end_pfn);
1551 if (!ret) {
1552 drain = 1;
1553 goto repeat;
1554 } else {
1555 if (ret < 0)
1556 if (--retry_max == 0)
1557 goto failed_removal;
1558 yield();
1559 drain = 1;
1560 goto repeat;
1561 }
1562 }
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001563 /* drain all zone's lru pagevec, this is asynchronous... */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001564 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001565 yield();
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001566 /* drain pcp pages, this is synchronous. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001567 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001568 /* check again */
1569 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1570 if (offlined_pages < 0) {
1571 ret = -EBUSY;
1572 goto failed_removal;
1573 }
1574 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
Adam Buchbinderb3834be2012-09-19 21:48:02 -04001575 /* Ok, all of our target is isolated.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001576 We cannot do rollback at this point. */
1577 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -08001578 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001579 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001580 /* removal success */
Jiang Liu9feedc92012-12-12 13:52:12 -08001581 zone->managed_pages -= offlined_pages;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001582 zone->present_pages -= offlined_pages;
1583 zone->zone_pgdat->node_present_pages -= offlined_pages;
1584 totalram_pages -= offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001585
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001586 init_per_zone_wmark_min();
1587
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001588 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001589 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001590 mutex_lock(&zonelists_mutex);
1591 build_all_zonelists(NULL, NULL);
1592 mutex_unlock(&zonelists_mutex);
1593 } else
1594 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -07001595
Lai Jiangshand9713672012-12-11 16:01:03 -08001596 node_states_clear_node(node, &arg);
1597 if (arg.status_change_nid >= 0)
David Rientjes8fe23e02009-12-14 17:58:33 -08001598 kswapd_stop(node);
Minchan Kimbce73942009-06-16 15:32:50 -07001599
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001600 vm_total_pages = nr_free_pagecache_pages();
1601 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001602
1603 memory_notify(MEM_OFFLINE, &arg);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001604 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001605 return 0;
1606
1607failed_removal:
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07001608 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1609 (unsigned long long) start_pfn << PAGE_SHIFT,
1610 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001611 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001612 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001613 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001614
Andi Kleen6ad696d2009-11-17 14:06:22 -08001615out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001616 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001617 return ret;
1618}
Badari Pulavarty71088782008-10-18 20:25:58 -07001619
Wen Congyanga16cee12012-10-08 16:33:58 -07001620int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1621{
1622 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1623}
1624
Wen Congyangbbc76be2013-02-22 16:32:54 -08001625/**
1626 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1627 * @start_pfn: start pfn of the memory range
Toshi Kanie05c4bb2013-04-29 15:06:16 -07001628 * @end_pfn: end pfn of the memory range
Wen Congyangbbc76be2013-02-22 16:32:54 -08001629 * @arg: argument passed to func
1630 * @func: callback for each memory section walked
1631 *
1632 * This function walks through all present mem sections in range
1633 * [start_pfn, end_pfn) and call func on each mem section.
1634 *
1635 * Returns the return value of func.
1636 */
1637static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1638 void *arg, int (*func)(struct memory_block *, void *))
Badari Pulavarty71088782008-10-18 20:25:58 -07001639{
Wen Congyange90bdb72012-10-08 16:34:01 -07001640 struct memory_block *mem = NULL;
1641 struct mem_section *section;
Wen Congyange90bdb72012-10-08 16:34:01 -07001642 unsigned long pfn, section_nr;
1643 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001644
Wen Congyange90bdb72012-10-08 16:34:01 -07001645 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1646 section_nr = pfn_to_section_nr(pfn);
1647 if (!present_section_nr(section_nr))
1648 continue;
1649
1650 section = __nr_to_section(section_nr);
1651 /* same memblock? */
1652 if (mem)
1653 if ((section_nr >= mem->start_section_nr) &&
1654 (section_nr <= mem->end_section_nr))
1655 continue;
1656
1657 mem = find_memory_block_hinted(section, mem);
1658 if (!mem)
1659 continue;
1660
Wen Congyangbbc76be2013-02-22 16:32:54 -08001661 ret = func(mem, arg);
Wen Congyange90bdb72012-10-08 16:34:01 -07001662 if (ret) {
Wen Congyangbbc76be2013-02-22 16:32:54 -08001663 kobject_put(&mem->dev.kobj);
1664 return ret;
Wen Congyange90bdb72012-10-08 16:34:01 -07001665 }
1666 }
1667
1668 if (mem)
1669 kobject_put(&mem->dev.kobj);
1670
Wen Congyangbbc76be2013-02-22 16:32:54 -08001671 return 0;
1672}
1673
1674/**
1675 * offline_memory_block_cb - callback function for offlining memory block
1676 * @mem: the memory block to be offlined
1677 * @arg: buffer to hold error msg
1678 *
1679 * Always return 0, and put the error msg in arg if any.
1680 */
1681static int offline_memory_block_cb(struct memory_block *mem, void *arg)
1682{
1683 int *ret = arg;
1684 int error = offline_memory_block(mem);
1685
1686 if (error != 0 && *ret == 0)
1687 *ret = error;
1688
1689 return 0;
1690}
1691
1692static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1693{
1694 int ret = !is_memblock_offlined(mem);
1695
Randy Dunlap349daa02013-04-29 15:08:49 -07001696 if (unlikely(ret)) {
1697 phys_addr_t beginpa, endpa;
1698
1699 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1700 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
Wen Congyangbbc76be2013-02-22 16:32:54 -08001701 pr_warn("removing memory fails, because memory "
Randy Dunlap349daa02013-04-29 15:08:49 -07001702 "[%pa-%pa] is onlined\n",
1703 &beginpa, &endpa);
1704 }
Wen Congyangbbc76be2013-02-22 16:32:54 -08001705
1706 return ret;
1707}
1708
Tang Chen60a5a192013-02-22 16:33:14 -08001709static int check_cpu_on_node(void *data)
1710{
1711 struct pglist_data *pgdat = data;
1712 int cpu;
1713
1714 for_each_present_cpu(cpu) {
1715 if (cpu_to_node(cpu) == pgdat->node_id)
1716 /*
1717 * the cpu on this node isn't removed, and we can't
1718 * offline this node.
1719 */
1720 return -EBUSY;
1721 }
1722
1723 return 0;
1724}
1725
Wen Congyange13fe862013-02-22 16:33:31 -08001726static void unmap_cpu_on_node(void *data)
1727{
1728#ifdef CONFIG_ACPI_NUMA
1729 struct pglist_data *pgdat = data;
1730 int cpu;
1731
1732 for_each_possible_cpu(cpu)
1733 if (cpu_to_node(cpu) == pgdat->node_id)
1734 numa_clear_node(cpu);
1735#endif
1736}
1737
1738static int check_and_unmap_cpu_on_node(void *data)
1739{
1740 int ret = check_cpu_on_node(data);
1741
1742 if (ret)
1743 return ret;
1744
1745 /*
1746 * the node will be offlined when we come here, so we can clear
1747 * the cpu_to_node() now.
1748 */
1749
1750 unmap_cpu_on_node(data);
1751 return 0;
1752}
1753
Tang Chen60a5a192013-02-22 16:33:14 -08001754/* offline the node if all memory sections of this node are removed */
Wen Congyang90b30cd2013-02-22 16:33:27 -08001755void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08001756{
Wen Congyangd822b862013-02-22 16:33:16 -08001757 pg_data_t *pgdat = NODE_DATA(nid);
1758 unsigned long start_pfn = pgdat->node_start_pfn;
1759 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
Tang Chen60a5a192013-02-22 16:33:14 -08001760 unsigned long pfn;
Wen Congyangd822b862013-02-22 16:33:16 -08001761 struct page *pgdat_page = virt_to_page(pgdat);
1762 int i;
Tang Chen60a5a192013-02-22 16:33:14 -08001763
1764 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1765 unsigned long section_nr = pfn_to_section_nr(pfn);
1766
1767 if (!present_section_nr(section_nr))
1768 continue;
1769
1770 if (pfn_to_nid(pfn) != nid)
1771 continue;
1772
1773 /*
1774 * some memory sections of this node are not removed, and we
1775 * can't offline node now.
1776 */
1777 return;
1778 }
1779
Wen Congyange13fe862013-02-22 16:33:31 -08001780 if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL))
Tang Chen60a5a192013-02-22 16:33:14 -08001781 return;
1782
1783 /*
1784 * all memory/cpu of this node are removed, we can offline this
1785 * node now.
1786 */
1787 node_set_offline(nid);
1788 unregister_one_node(nid);
Wen Congyangd822b862013-02-22 16:33:16 -08001789
1790 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1791 /* node data is allocated from boot memory */
1792 return;
1793
1794 /* free waittable in each zone */
1795 for (i = 0; i < MAX_NR_ZONES; i++) {
1796 struct zone *zone = pgdat->node_zones + i;
1797
Jianguo Wuca4b3f32013-03-22 15:04:50 -07001798 /*
1799 * wait_table may be allocated from boot memory,
1800 * here only free if it's allocated by vmalloc.
1801 */
1802 if (is_vmalloc_addr(zone->wait_table))
Wen Congyangd822b862013-02-22 16:33:16 -08001803 vfree(zone->wait_table);
1804 }
1805
1806 /*
1807 * Since there is no way to guarentee the address of pgdat/zone is not
1808 * on stack of any kernel threads or used by other kernel objects
1809 * without reference counting or other symchronizing method, do not
1810 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1811 * the memory when the node is online again.
1812 */
1813 memset(pgdat, 0, sizeof(*pgdat));
Tang Chen60a5a192013-02-22 16:33:14 -08001814}
Wen Congyang90b30cd2013-02-22 16:33:27 -08001815EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08001816
1817int __ref remove_memory(int nid, u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001818{
1819 unsigned long start_pfn, end_pfn;
1820 int ret = 0;
1821 int retry = 1;
1822
1823 start_pfn = PFN_DOWN(start);
Toshi Kanif8749452013-03-13 14:59:31 -07001824 end_pfn = PFN_UP(start + size - 1);
Wen Congyangbbc76be2013-02-22 16:32:54 -08001825
1826 /*
1827 * When CONFIG_MEMCG is on, one memory block may be used by other
1828 * blocks to store page cgroup when onlining pages. But we don't know
1829 * in what order pages are onlined. So we iterate twice to offline
1830 * memory:
1831 * 1st iterate: offline every non primary memory block.
1832 * 2nd iterate: offline primary (i.e. first added) memory block.
1833 */
1834repeat:
1835 walk_memory_range(start_pfn, end_pfn, &ret,
1836 offline_memory_block_cb);
1837 if (ret) {
1838 if (!retry)
1839 return ret;
1840
1841 retry = 0;
1842 ret = 0;
Wen Congyang993c1aa2013-02-22 16:32:50 -08001843 goto repeat;
1844 }
1845
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001846 lock_memory_hotplug();
1847
1848 /*
1849 * we have offlined all memory blocks like this:
1850 * 1. lock memory hotplug
1851 * 2. offline a memory block
1852 * 3. unlock memory hotplug
1853 *
1854 * repeat step1-3 to offline the memory block. All memory blocks
1855 * must be offlined before removing memory. But we don't hold the
1856 * lock in the whole operation. So we should check whether all
1857 * memory blocks are offlined.
1858 */
1859
Wen Congyangbbc76be2013-02-22 16:32:54 -08001860 ret = walk_memory_range(start_pfn, end_pfn, NULL,
1861 is_memblock_offlined_cb);
1862 if (ret) {
1863 unlock_memory_hotplug();
1864 return ret;
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001865 }
1866
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08001867 /* remove memmap entry */
1868 firmware_map_remove(start, start + size, "System RAM");
1869
Wen Congyang24d335c2013-02-22 16:32:58 -08001870 arch_remove_memory(start, size);
1871
Tang Chen60a5a192013-02-22 16:33:14 -08001872 try_offline_node(nid);
1873
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08001874 unlock_memory_hotplug();
1875
Wen Congyange90bdb72012-10-08 16:34:01 -07001876 return 0;
Badari Pulavarty71088782008-10-18 20:25:58 -07001877}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -07001878#else
Wen Congyanga16cee12012-10-08 16:33:58 -07001879int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1880{
1881 return -EINVAL;
1882}
Tang Chen60a5a192013-02-22 16:33:14 -08001883int remove_memory(int nid, u64 start, u64 size)
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -07001884{
1885 return -EINVAL;
1886}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001887#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavarty71088782008-10-18 20:25:58 -07001888EXPORT_SYMBOL_GPL(remove_memory);