blob: ec2f199cc5f711b9a5130606ce1ee16f368e4f14 [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>
Dave Hansen3947be12005-10-29 18:16:54 -070032
33#include <asm/tlbflush.h>
34
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030035#include "internal.h"
36
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -070037/*
38 * online_page_callback contains pointer to current page onlining function.
39 * Initially it is generic_online_page(). If it is required it could be
40 * changed by calling set_online_page_callback() for callback registration
41 * and restore_online_page_callback() for generic callback restore.
42 */
43
44static void generic_online_page(struct page *page);
45
46static online_page_callback_t online_page_callback = generic_online_page;
47
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080048DEFINE_MUTEX(mem_hotplug_mutex);
49
50void lock_memory_hotplug(void)
51{
52 mutex_lock(&mem_hotplug_mutex);
53
54 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
55 lock_system_sleep();
56}
57
58void unlock_memory_hotplug(void)
59{
60 unlock_system_sleep();
61 mutex_unlock(&mem_hotplug_mutex);
62}
63
64
Keith Mannthey45e0b782006-09-30 23:27:09 -070065/* add this memory to iomem resource */
66static struct resource *register_memory_resource(u64 start, u64 size)
67{
68 struct resource *res;
69 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
70 BUG_ON(!res);
71
72 res->name = "System RAM";
73 res->start = start;
74 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080075 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070076 if (request_resource(&iomem_resource, res) < 0) {
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -070077 printk("System RAM resource %pR cannot be added\n", res);
Keith Mannthey45e0b782006-09-30 23:27:09 -070078 kfree(res);
79 res = NULL;
80 }
81 return res;
82}
83
84static void release_memory_resource(struct resource *res)
85{
86 if (!res)
87 return;
88 release_resource(res);
89 kfree(res);
90 return;
91}
92
Keith Mannthey53947022006-09-30 23:27:08 -070093#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasunori Goto04753272008-04-28 02:13:31 -070094#ifndef CONFIG_SPARSEMEM_VMEMMAP
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -080095static void 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;
Yasunori Goto04753272008-04-28 02:13:31 -0700109
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800110 type = (unsigned long) page->lru.next;
111 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
112 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700113
114 if (atomic_dec_return(&page->_count) == 1) {
115 ClearPagePrivate(page);
116 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800117 INIT_LIST_HEAD(&page->lru);
Yasunori Goto04753272008-04-28 02:13:31 -0700118 __free_pages_bootmem(page, 0);
119 }
120
121}
122
Adrian Bunkd92bc312008-07-23 21:28:12 -0700123static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700124{
125 unsigned long *usemap, mapsize, section_nr, i;
126 struct mem_section *ms;
127 struct page *page, *memmap;
128
Yasunori Goto04753272008-04-28 02:13:31 -0700129 section_nr = pfn_to_section_nr(start_pfn);
130 ms = __nr_to_section(section_nr);
131
132 /* Get section's memmap address */
133 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
134
135 /*
136 * Get page for the memmap's phys address
137 * XXX: need more consideration for sparse_vmemmap...
138 */
139 page = virt_to_page(memmap);
140 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
141 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
142
143 /* remember memmap's page */
144 for (i = 0; i < mapsize; i++, page++)
145 get_page_bootmem(section_nr, page, SECTION_INFO);
146
147 usemap = __nr_to_section(section_nr)->pageblock_flags;
148 page = virt_to_page(usemap);
149
150 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
151
152 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700153 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700154
155}
156
157void register_page_bootmem_info_node(struct pglist_data *pgdat)
158{
159 unsigned long i, pfn, end_pfn, nr_pages;
160 int node = pgdat->node_id;
161 struct page *page;
162 struct zone *zone;
163
164 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
165 page = virt_to_page(pgdat);
166
167 for (i = 0; i < nr_pages; i++, page++)
168 get_page_bootmem(node, page, NODE_INFO);
169
170 zone = &pgdat->node_zones[0];
171 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
172 if (zone->wait_table) {
173 nr_pages = zone->wait_table_hash_nr_entries
174 * sizeof(wait_queue_head_t);
175 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
176 page = virt_to_page(zone->wait_table);
177
178 for (i = 0; i < nr_pages; i++, page++)
179 get_page_bootmem(node, page, NODE_INFO);
180 }
181 }
182
183 pfn = pgdat->node_start_pfn;
184 end_pfn = pfn + pgdat->node_spanned_pages;
185
186 /* register_section info */
qiuxishif14851a2012-09-17 14:09:24 -0700187 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
188 /*
189 * Some platforms can assign the same pfn to multiple nodes - on
190 * node0 as well as nodeN. To avoid registering a pfn against
191 * multiple nodes we check that this pfn does not already
192 * reside in some other node.
193 */
194 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
195 register_page_bootmem_info_section(pfn);
196 }
Yasunori Goto04753272008-04-28 02:13:31 -0700197}
198#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
199
Heiko Carstens76cdd582008-05-14 16:05:52 -0700200static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
201 unsigned long end_pfn)
202{
203 unsigned long old_zone_end_pfn;
204
205 zone_span_writelock(zone);
206
207 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
208 if (start_pfn < zone->zone_start_pfn)
209 zone->zone_start_pfn = start_pfn;
210
211 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
212 zone->zone_start_pfn;
213
214 zone_span_writeunlock(zone);
215}
216
217static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
218 unsigned long end_pfn)
219{
220 unsigned long old_pgdat_end_pfn =
221 pgdat->node_start_pfn + pgdat->node_spanned_pages;
222
223 if (start_pfn < pgdat->node_start_pfn)
224 pgdat->node_start_pfn = start_pfn;
225
226 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
227 pgdat->node_start_pfn;
228}
229
Al Viro31168482008-11-22 17:33:24 +0000230static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700231{
232 struct pglist_data *pgdat = zone->zone_pgdat;
233 int nr_pages = PAGES_PER_SECTION;
234 int nid = pgdat->node_id;
235 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700236 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700237
238 zone_type = zone - pgdat->node_zones;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700239 if (!zone->wait_table) {
240 int ret;
241
242 ret = init_currently_empty_zone(zone, phys_start_pfn,
243 nr_pages, MEMMAP_HOTPLUG);
244 if (ret)
245 return ret;
246 }
247 pgdat_resize_lock(zone->zone_pgdat, &flags);
248 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
249 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
250 phys_start_pfn + nr_pages);
251 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800252 memmap_init_zone(nr_pages, nid, zone_type,
253 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700254 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700255}
256
Gary Hadec04fc582009-01-06 14:39:14 -0800257static int __meminit __add_section(int nid, struct zone *zone,
258 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700259{
Dave Hansen3947be12005-10-29 18:16:54 -0700260 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700261 int ret;
262
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700263 if (pfn_valid(phys_start_pfn))
264 return -EEXIST;
265
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700266 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700267
268 if (ret < 0)
269 return ret;
270
Yasunori Goto718127c2006-06-23 02:03:10 -0700271 ret = __add_zone(zone, phys_start_pfn);
272
273 if (ret < 0)
274 return ret;
275
Gary Hadec04fc582009-01-06 14:39:14 -0800276 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700277}
278
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700279#ifdef CONFIG_SPARSEMEM_VMEMMAP
280static int __remove_section(struct zone *zone, struct mem_section *ms)
281{
282 /*
283 * XXX: Freeing memmap with vmemmap is not implement yet.
284 * This should be removed later.
285 */
286 return -EBUSY;
287}
288#else
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700289static int __remove_section(struct zone *zone, struct mem_section *ms)
290{
291 unsigned long flags;
292 struct pglist_data *pgdat = zone->zone_pgdat;
293 int ret = -EINVAL;
294
295 if (!valid_section(ms))
296 return ret;
297
298 ret = unregister_memory_section(ms);
299 if (ret)
300 return ret;
301
302 pgdat_resize_lock(pgdat, &flags);
303 sparse_remove_one_section(zone, ms);
304 pgdat_resize_unlock(pgdat, &flags);
305 return 0;
306}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700307#endif
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700308
Dave Hansen3947be12005-10-29 18:16:54 -0700309/*
310 * Reasonably generic function for adding memory. It is
311 * expected that archs that support memory hotplug will
312 * call this function after deciding the zone to which to
313 * add the new pages.
314 */
Gary Hadec04fc582009-01-06 14:39:14 -0800315int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
316 unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700317{
318 unsigned long i;
319 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700320 int start_sec, end_sec;
321 /* during initialize mem_map, align hot-added range to section */
322 start_sec = pfn_to_section_nr(phys_start_pfn);
323 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700324
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700325 for (i = start_sec; i <= end_sec; i++) {
Gary Hadec04fc582009-01-06 14:39:14 -0800326 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700327
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700328 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200329 * EEXIST is finally dealt with by ioresource collision
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700330 * check. see add_memory() => register_memory_resource()
331 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700332 */
333 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700334 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700335 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700336 }
337
338 return err;
339}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700340EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700341
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700342/**
343 * __remove_pages() - remove sections of pages from a zone
344 * @zone: zone from which pages need to be removed
345 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
346 * @nr_pages: number of pages to remove (must be multiple of section size)
347 *
348 * Generic helper function to remove section mappings and sysfs entries
349 * for the section of the memory we are removing. Caller needs to make
350 * sure that pages are marked reserved and zones are adjust properly by
351 * calling offline_pages().
352 */
353int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
354 unsigned long nr_pages)
355{
356 unsigned long i, ret = 0;
357 int sections_to_remove;
358
359 /*
360 * We can only remove entire sections
361 */
362 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
363 BUG_ON(nr_pages % PAGES_PER_SECTION);
364
Yasuaki Ishimatsud760afd2012-10-08 16:34:14 -0700365 release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
366
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700367 sections_to_remove = nr_pages / PAGES_PER_SECTION;
368 for (i = 0; i < sections_to_remove; i++) {
369 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
370 ret = __remove_section(zone, __pfn_to_section(pfn));
371 if (ret)
372 break;
373 }
374 return ret;
375}
376EXPORT_SYMBOL_GPL(__remove_pages);
377
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700378int set_online_page_callback(online_page_callback_t callback)
379{
380 int rc = -EINVAL;
381
382 lock_memory_hotplug();
383
384 if (online_page_callback == generic_online_page) {
385 online_page_callback = callback;
386 rc = 0;
387 }
388
389 unlock_memory_hotplug();
390
391 return rc;
392}
393EXPORT_SYMBOL_GPL(set_online_page_callback);
394
395int restore_online_page_callback(online_page_callback_t callback)
396{
397 int rc = -EINVAL;
398
399 lock_memory_hotplug();
400
401 if (online_page_callback == callback) {
402 online_page_callback = generic_online_page;
403 rc = 0;
404 }
405
406 unlock_memory_hotplug();
407
408 return rc;
409}
410EXPORT_SYMBOL_GPL(restore_online_page_callback);
411
412void __online_page_set_limits(struct page *page)
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700413{
Jan Beulich4738e1b2009-09-21 17:03:03 -0700414 unsigned long pfn = page_to_pfn(page);
415
Jan Beulich4738e1b2009-09-21 17:03:03 -0700416 if (pfn >= num_physpages)
417 num_physpages = pfn + 1;
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700418}
419EXPORT_SYMBOL_GPL(__online_page_set_limits);
420
421void __online_page_increment_counters(struct page *page)
422{
423 totalram_pages++;
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700424
425#ifdef CONFIG_HIGHMEM
426 if (PageHighMem(page))
427 totalhigh_pages++;
428#endif
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700429}
430EXPORT_SYMBOL_GPL(__online_page_increment_counters);
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700431
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700432void __online_page_free(struct page *page)
433{
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700434 ClearPageReserved(page);
435 init_page_count(page);
436 __free_page(page);
437}
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700438EXPORT_SYMBOL_GPL(__online_page_free);
439
440static void generic_online_page(struct page *page)
441{
442 __online_page_set_limits(page);
443 __online_page_increment_counters(page);
444 __online_page_free(page);
445}
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700446
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700447static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
448 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700449{
450 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700451 unsigned long onlined_pages = *(unsigned long *)arg;
452 struct page *page;
453 if (PageReserved(pfn_to_page(start_pfn)))
454 for (i = 0; i < nr_pages; i++) {
455 page = pfn_to_page(start_pfn + i);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700456 (*online_page_callback)(page);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700457 onlined_pages++;
458 }
459 *(unsigned long *)arg = onlined_pages;
460 return 0;
461}
462
463
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -0700464int __ref online_pages(unsigned long pfn, unsigned long nr_pages)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700465{
Dave Hansen3947be12005-10-29 18:16:54 -0700466 unsigned long onlined_pages = 0;
467 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700468 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700469 int nid;
470 int ret;
471 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700472
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900473 lock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700474 arg.start_pfn = pfn;
475 arg.nr_pages = nr_pages;
476 arg.status_change_nid = -1;
477
478 nid = page_to_nid(pfn_to_page(pfn));
479 if (node_present_pages(nid) == 0)
480 arg.status_change_nid = nid;
481
482 ret = memory_notify(MEM_GOING_ONLINE, &arg);
483 ret = notifier_to_errno(ret);
484 if (ret) {
485 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900486 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700487 return ret;
488 }
Dave Hansen3947be12005-10-29 18:16:54 -0700489 /*
490 * This doesn't need a lock to do pfn_to_page().
491 * The section can't be removed here because of the
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800492 * memory_block->state_mutex.
Dave Hansen3947be12005-10-29 18:16:54 -0700493 */
494 zone = page_zone(pfn_to_page(pfn));
Yasunori Goto68113782006-06-23 02:03:11 -0700495 /*
496 * If this zone is not populated, then it is not in zonelist.
497 * This means the page allocator ignores this zone.
498 * So, zonelist must be updated after online.
499 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700500 mutex_lock(&zonelists_mutex);
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800501 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -0700502 need_zonelists_rebuild = 1;
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800503 build_all_zonelists(NULL, zone);
504 }
Yasunori Goto68113782006-06-23 02:03:11 -0700505
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700506 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700507 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700508 if (ret) {
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800509 if (need_zonelists_rebuild)
510 zone_pcp_reset(zone);
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700511 mutex_unlock(&zonelists_mutex);
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -0700512 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
513 (unsigned long long) pfn << PAGE_SHIFT,
514 (((unsigned long long) pfn + nr_pages)
515 << PAGE_SHIFT) - 1);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700516 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900517 unlock_memory_hotplug();
Geoff Levandfd8a4222008-05-14 16:05:50 -0700518 return ret;
519 }
520
Dave Hansen3947be12005-10-29 18:16:54 -0700521 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800522 zone->zone_pgdat->node_present_pages += onlined_pages;
Jiang Liu08dff7b2012-07-31 16:43:30 -0700523 if (onlined_pages) {
524 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
525 if (need_zonelists_rebuild)
Wen Congyang6dcd73d2012-12-11 16:01:01 -0800526 build_all_zonelists(NULL, NULL);
Jiang Liu08dff7b2012-07-31 16:43:30 -0700527 else
528 zone_pcp_update(zone);
529 }
Dave Hansen3947be12005-10-29 18:16:54 -0700530
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700531 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700532
533 init_per_zone_wmark_min();
534
Jiang Liu08dff7b2012-07-31 16:43:30 -0700535 if (onlined_pages)
Christoph Lameter7ea15302007-10-16 01:25:29 -0700536 kswapd_run(zone_to_nid(zone));
Dave Hansen61b13992005-10-29 18:16:56 -0700537
Haicheng Li1f522502010-05-24 14:32:51 -0700538 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700539
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700540 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700541
542 if (onlined_pages)
543 memory_notify(MEM_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900544 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700545
Dave Hansen3947be12005-10-29 18:16:54 -0700546 return 0;
547}
Keith Mannthey53947022006-09-30 23:27:08 -0700548#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700549
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800550/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
551static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700552{
553 struct pglist_data *pgdat;
554 unsigned long zones_size[MAX_NR_ZONES] = {0};
555 unsigned long zholes_size[MAX_NR_ZONES] = {0};
556 unsigned long start_pfn = start >> PAGE_SHIFT;
557
558 pgdat = arch_alloc_nodedata(nid);
559 if (!pgdat)
560 return NULL;
561
562 arch_refresh_nodedata(nid, pgdat);
563
564 /* we can use NODE_DATA(nid) from here */
565
566 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -0700567 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700568
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700569 /*
570 * The node we allocated has no zone fallback lists. For avoiding
571 * to access not-initialized zonelist, build here.
572 */
David Rientjesf957db42011-06-22 18:13:04 -0700573 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -0700574 build_all_zonelists(pgdat, NULL);
David Rientjesf957db42011-06-22 18:13:04 -0700575 mutex_unlock(&zonelists_mutex);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700576
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700577 return pgdat;
578}
579
580static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
581{
582 arch_refresh_nodedata(nid, NULL);
583 arch_free_nodedata(pgdat);
584 return;
585}
586
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700587
minskey guocf234222010-05-24 14:32:41 -0700588/*
589 * called by cpu_up() to online a node without onlined memory.
590 */
591int mem_online_node(int nid)
592{
593 pg_data_t *pgdat;
594 int ret;
595
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800596 lock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -0700597 pgdat = hotadd_new_pgdat(nid, 0);
David Rientjes7553e8f2011-06-22 18:13:01 -0700598 if (!pgdat) {
minskey guocf234222010-05-24 14:32:41 -0700599 ret = -ENOMEM;
600 goto out;
601 }
602 node_set_online(nid);
603 ret = register_one_node(nid);
604 BUG_ON(ret);
605
606out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800607 unlock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -0700608 return ret;
609}
610
Al Viro31168482008-11-22 17:33:24 +0000611/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
612int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -0700613{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700614 pg_data_t *pgdat = NULL;
615 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700616 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700617 int ret;
618
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800619 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -0800620
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700621 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800622 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700623 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800624 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700625
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700626 if (!node_online(nid)) {
627 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800628 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700629 if (!pgdat)
Wen Congyang41b9e2d2012-07-11 14:02:31 -0700630 goto error;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700631 new_pgdat = 1;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700632 }
633
Yasunori Gotobc02af92006-06-27 02:53:30 -0700634 /* call arch's memory hotadd */
635 ret = arch_add_memory(nid, start, size);
636
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700637 if (ret < 0)
638 goto error;
639
Yasunori Goto0fc44152006-06-27 02:53:38 -0700640 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700641 node_set_online(nid);
642
Yasunori Goto0fc44152006-06-27 02:53:38 -0700643 if (new_pgdat) {
644 ret = register_one_node(nid);
645 /*
646 * If sysfs file of new node can't create, cpu on the node
647 * can't be hot-added. There is no rollback way now.
648 * So, check by BUG_ON() to catch it reluctantly..
649 */
650 BUG_ON(ret);
651 }
652
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -0800653 /* create new memmap entry */
654 firmware_map_add_hotplug(start, start + size, "System RAM");
655
Andi Kleen6ad696d2009-11-17 14:06:22 -0800656 goto out;
657
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700658error:
659 /* rollback pgdat allocation and others */
660 if (new_pgdat)
661 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700662 if (res)
663 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700664
Andi Kleen6ad696d2009-11-17 14:06:22 -0800665out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800666 unlock_memory_hotplug();
Yasunori Gotobc02af92006-06-27 02:53:30 -0700667 return ret;
668}
669EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700670
671#ifdef CONFIG_MEMORY_HOTREMOVE
672/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700673 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
674 * set and the size of the free page is given by page_order(). Using this,
675 * the function determines if the pageblock contains only free pages.
676 * Due to buddy contraints, a free page at least the size of a pageblock will
677 * be located at the start of the pageblock
678 */
679static inline int pageblock_free(struct page *page)
680{
681 return PageBuddy(page) && page_order(page) >= pageblock_order;
682}
683
684/* Return the start of the next active pageblock after a given page */
685static struct page *next_active_pageblock(struct page *page)
686{
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700687 /* Ensure the starting page is pageblock-aligned */
688 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
689
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700690 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700691 if (pageblock_free(page)) {
692 int order;
693 /* be careful. we don't have locks, page_order can be changed.*/
694 order = page_order(page);
695 if ((order < MAX_ORDER) && (order >= pageblock_order))
696 return page + (1 << order);
697 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700698
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700699 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700700}
701
702/* Checks if this range of memory is likely to be hot-removable. */
703int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
704{
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700705 struct page *page = pfn_to_page(start_pfn);
706 struct page *end_page = page + nr_pages;
707
708 /* Check the starting page of each pageblock within the range */
709 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -0700710 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700711 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -0700712 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700713 }
714
715 /* All pageblocks in the memory block are likely to be hot-removable */
716 return 1;
717}
718
719/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700720 * Confirm all pages in a range [start, end) is belongs to the same zone.
721 */
722static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
723{
724 unsigned long pfn;
725 struct zone *zone = NULL;
726 struct page *page;
727 int i;
728 for (pfn = start_pfn;
729 pfn < end_pfn;
730 pfn += MAX_ORDER_NR_PAGES) {
731 i = 0;
732 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
733 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
734 i++;
735 if (i == MAX_ORDER_NR_PAGES)
736 continue;
737 page = pfn_to_page(pfn + i);
738 if (zone && page_zone(page) != zone)
739 return 0;
740 zone = page_zone(page);
741 }
742 return 1;
743}
744
745/*
746 * Scanning pfn is much easier than scanning lru list.
747 * Scan pfn from start to end and Find LRU page.
748 */
Andrew Morton7bbc0902010-10-26 14:22:05 -0700749static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700750{
751 unsigned long pfn;
752 struct page *page;
753 for (pfn = start; pfn < end; pfn++) {
754 if (pfn_valid(pfn)) {
755 page = pfn_to_page(pfn);
756 if (PageLRU(page))
757 return pfn;
758 }
759 }
760 return 0;
761}
762
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700763#define NR_OFFLINE_AT_ONCE_PAGES (256)
764static int
765do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
766{
767 unsigned long pfn;
768 struct page *page;
769 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
770 int not_managed = 0;
771 int ret = 0;
772 LIST_HEAD(source);
773
774 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
775 if (!pfn_valid(pfn))
776 continue;
777 page = pfn_to_page(pfn);
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -0700778 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700779 continue;
780 /*
781 * We can skip free pages. And we can only deal with pages on
782 * LRU.
783 */
Nick Piggin62695a82008-10-18 20:26:09 -0700784 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700785 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -0700786 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -0700787 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700788 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800789 inc_zone_page_state(page, NR_ISOLATED_ANON +
790 page_is_file_cache(page));
791
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700792 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700793#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -0800794 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
795 pfn);
796 dump_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700797#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -0700798 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300799 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -0700800 check this again here. */
801 if (page_count(page)) {
802 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -0700803 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -0700804 break;
805 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700806 }
807 }
Bob Liuf3ab2632010-10-26 14:22:10 -0700808 if (!list_empty(&source)) {
809 if (not_managed) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700810 putback_lru_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -0700811 goto out;
812 }
Minchan Kim74c08f92012-10-08 16:32:54 -0700813
814 /*
815 * alloc_migrate_target should be improooooved!!
816 * migrate_pages returns # of failed pages.
817 */
818 ret = migrate_pages(&source, alloc_migrate_target, 0,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800819 true, MIGRATE_SYNC);
Bob Liuf3ab2632010-10-26 14:22:10 -0700820 if (ret)
821 putback_lru_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700822 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700823out:
824 return ret;
825}
826
827/*
828 * remove from free_area[] and mark all as Reserved.
829 */
830static int
831offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
832 void *data)
833{
834 __offline_isolated_pages(start, start + nr_pages);
835 return 0;
836}
837
838static void
839offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
840{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700841 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700842 offline_isolated_pages_cb);
843}
844
845/*
846 * Check all pages in range, recoreded as memory resource, are isolated.
847 */
848static int
849check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
850 void *data)
851{
852 int ret;
853 long offlined = *(long *)data;
Wen Congyangb023f462012-12-11 16:00:45 -0800854 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700855 offlined = nr_pages;
856 if (!ret)
857 *(long *)data += offlined;
858 return ret;
859}
860
861static long
862check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
863{
864 long offlined = 0;
865 int ret;
866
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700867 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700868 check_pages_isolated_cb);
869 if (ret < 0)
870 offlined = (long)ret;
871 return offlined;
872}
873
Wen Congyanga16cee12012-10-08 16:33:58 -0700874static int __ref __offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700875 unsigned long end_pfn, unsigned long timeout)
876{
877 unsigned long pfn, nr_pages, expire;
878 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700879 int ret, drain, retry_max, node;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700880 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700881 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700882
883 BUG_ON(start_pfn >= end_pfn);
884 /* at least, alignment against pageblock is necessary */
885 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
886 return -EINVAL;
887 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
888 return -EINVAL;
889 /* This makes hotplug much easier...and readable.
890 we assume this for now. .*/
891 if (!test_pages_in_a_zone(start_pfn, end_pfn))
892 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700893
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800894 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -0800895
Yasunori Goto7b78d332007-10-21 16:41:36 -0700896 zone = page_zone(pfn_to_page(start_pfn));
897 node = zone_to_nid(zone);
898 nr_pages = end_pfn - start_pfn;
899
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700900 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -0800901 ret = start_isolate_page_range(start_pfn, end_pfn,
902 MIGRATE_MOVABLE, true);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700903 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800904 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700905
906 arg.start_pfn = start_pfn;
907 arg.nr_pages = nr_pages;
908 arg.status_change_nid = -1;
909 if (nr_pages >= node_present_pages(node))
910 arg.status_change_nid = node;
911
912 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
913 ret = notifier_to_errno(ret);
914 if (ret)
915 goto failed_removal;
916
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700917 pfn = start_pfn;
918 expire = jiffies + timeout;
919 drain = 0;
920 retry_max = 5;
921repeat:
922 /* start memory hot removal */
923 ret = -EAGAIN;
924 if (time_after(jiffies, expire))
925 goto failed_removal;
926 ret = -EINTR;
927 if (signal_pending(current))
928 goto failed_removal;
929 ret = 0;
930 if (drain) {
931 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700932 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800933 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700934 }
935
936 pfn = scan_lru_pages(start_pfn, end_pfn);
937 if (pfn) { /* We have page on LRU */
938 ret = do_migrate_range(pfn, end_pfn);
939 if (!ret) {
940 drain = 1;
941 goto repeat;
942 } else {
943 if (ret < 0)
944 if (--retry_max == 0)
945 goto failed_removal;
946 yield();
947 drain = 1;
948 goto repeat;
949 }
950 }
951 /* drain all zone's lru pagevec, this is asyncronous... */
952 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700953 yield();
954 /* drain pcp pages , this is synchrouns. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800955 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700956 /* check again */
957 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
958 if (offlined_pages < 0) {
959 ret = -EBUSY;
960 goto failed_removal;
961 }
962 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
963 /* Ok, all of our target is islaoted.
964 We cannot do rollback at this point. */
965 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -0800966 /* reset pagetype flags and makes migrate type to be MOVABLE */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200967 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700968 /* removal success */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700969 zone->present_pages -= offlined_pages;
970 zone->zone_pgdat->node_present_pages -= offlined_pages;
971 totalram_pages -= offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700972
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700973 init_per_zone_wmark_min();
974
Xishi Qiu1e8537b2012-10-08 16:31:51 -0700975 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -0700976 zone_pcp_reset(zone);
Xishi Qiu1e8537b2012-10-08 16:31:51 -0700977 mutex_lock(&zonelists_mutex);
978 build_all_zonelists(NULL, NULL);
979 mutex_unlock(&zonelists_mutex);
980 } else
981 zone_pcp_update(zone);
Jiang Liu340175b2012-07-31 16:43:32 -0700982
David Rientjes8fe23e02009-12-14 17:58:33 -0800983 if (!node_present_pages(node)) {
984 node_clear_state(node, N_HIGH_MEMORY);
985 kswapd_stop(node);
986 }
Minchan Kimbce73942009-06-16 15:32:50 -0700987
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700988 vm_total_pages = nr_free_pagecache_pages();
989 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700990
991 memory_notify(MEM_OFFLINE, &arg);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800992 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700993 return 0;
994
995failed_removal:
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -0700996 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
997 (unsigned long long) start_pfn << PAGE_SHIFT,
998 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700999 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001000 /* pushback to free area */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02001001 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001002
Andi Kleen6ad696d2009-11-17 14:06:22 -08001003out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -08001004 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001005 return ret;
1006}
Badari Pulavarty71088782008-10-18 20:25:58 -07001007
Wen Congyanga16cee12012-10-08 16:33:58 -07001008int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1009{
1010 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1011}
1012
Badari Pulavarty71088782008-10-18 20:25:58 -07001013int remove_memory(u64 start, u64 size)
1014{
Wen Congyange90bdb72012-10-08 16:34:01 -07001015 struct memory_block *mem = NULL;
1016 struct mem_section *section;
Badari Pulavarty71088782008-10-18 20:25:58 -07001017 unsigned long start_pfn, end_pfn;
Wen Congyange90bdb72012-10-08 16:34:01 -07001018 unsigned long pfn, section_nr;
1019 int ret;
Badari Pulavarty71088782008-10-18 20:25:58 -07001020
1021 start_pfn = PFN_DOWN(start);
1022 end_pfn = start_pfn + PFN_DOWN(size);
Wen Congyange90bdb72012-10-08 16:34:01 -07001023
1024 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1025 section_nr = pfn_to_section_nr(pfn);
1026 if (!present_section_nr(section_nr))
1027 continue;
1028
1029 section = __nr_to_section(section_nr);
1030 /* same memblock? */
1031 if (mem)
1032 if ((section_nr >= mem->start_section_nr) &&
1033 (section_nr <= mem->end_section_nr))
1034 continue;
1035
1036 mem = find_memory_block_hinted(section, mem);
1037 if (!mem)
1038 continue;
1039
1040 ret = offline_memory_block(mem);
1041 if (ret) {
1042 kobject_put(&mem->dev.kobj);
1043 return ret;
1044 }
1045 }
1046
1047 if (mem)
1048 kobject_put(&mem->dev.kobj);
1049
1050 return 0;
Badari Pulavarty71088782008-10-18 20:25:58 -07001051}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -07001052#else
Wen Congyanga16cee12012-10-08 16:33:58 -07001053int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1054{
1055 return -EINVAL;
1056}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -07001057int remove_memory(u64 start, u64 size)
1058{
1059 return -EINVAL;
1060}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001061#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavarty71088782008-10-18 20:25:58 -07001062EXPORT_SYMBOL_GPL(remove_memory);