blob: 02159c755136ea0463f57a2a951f9cad223aceb1 [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>
14#include <linux/module.h>
15#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
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -080037DEFINE_MUTEX(mem_hotplug_mutex);
38
39void lock_memory_hotplug(void)
40{
41 mutex_lock(&mem_hotplug_mutex);
42
43 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
44 lock_system_sleep();
45}
46
47void unlock_memory_hotplug(void)
48{
49 unlock_system_sleep();
50 mutex_unlock(&mem_hotplug_mutex);
51}
52
53
Keith Mannthey45e0b782006-09-30 23:27:09 -070054/* add this memory to iomem resource */
55static struct resource *register_memory_resource(u64 start, u64 size)
56{
57 struct resource *res;
58 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
59 BUG_ON(!res);
60
61 res->name = "System RAM";
62 res->start = start;
63 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080064 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070065 if (request_resource(&iomem_resource, res) < 0) {
66 printk("System RAM resource %llx - %llx cannot be added\n",
67 (unsigned long long)res->start, (unsigned long long)res->end);
68 kfree(res);
69 res = NULL;
70 }
71 return res;
72}
73
74static void release_memory_resource(struct resource *res)
75{
76 if (!res)
77 return;
78 release_resource(res);
79 kfree(res);
80 return;
81}
82
Keith Mannthey53947022006-09-30 23:27:08 -070083#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasunori Goto04753272008-04-28 02:13:31 -070084#ifndef CONFIG_SPARSEMEM_VMEMMAP
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -080085static void get_page_bootmem(unsigned long info, struct page *page,
86 unsigned long type)
Yasunori Goto04753272008-04-28 02:13:31 -070087{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -080088 page->lru.next = (struct list_head *) type;
Yasunori Goto04753272008-04-28 02:13:31 -070089 SetPagePrivate(page);
90 set_page_private(page, info);
91 atomic_inc(&page->_count);
92}
93
Rakib Mullick23ce9322009-12-14 17:59:44 -080094/* reference to __meminit __free_pages_bootmem is valid
95 * so use __ref to tell modpost not to generate a warning */
96void __ref put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -070097{
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -080098 unsigned long type;
Yasunori Goto04753272008-04-28 02:13:31 -070099
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800100 type = (unsigned long) page->lru.next;
101 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
102 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
Yasunori Goto04753272008-04-28 02:13:31 -0700103
104 if (atomic_dec_return(&page->_count) == 1) {
105 ClearPagePrivate(page);
106 set_page_private(page, 0);
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800107 INIT_LIST_HEAD(&page->lru);
Yasunori Goto04753272008-04-28 02:13:31 -0700108 __free_pages_bootmem(page, 0);
109 }
110
111}
112
Adrian Bunkd92bc312008-07-23 21:28:12 -0700113static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700114{
115 unsigned long *usemap, mapsize, section_nr, i;
116 struct mem_section *ms;
117 struct page *page, *memmap;
118
119 if (!pfn_valid(start_pfn))
120 return;
121
122 section_nr = pfn_to_section_nr(start_pfn);
123 ms = __nr_to_section(section_nr);
124
125 /* Get section's memmap address */
126 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
127
128 /*
129 * Get page for the memmap's phys address
130 * XXX: need more consideration for sparse_vmemmap...
131 */
132 page = virt_to_page(memmap);
133 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
134 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
135
136 /* remember memmap's page */
137 for (i = 0; i < mapsize; i++, page++)
138 get_page_bootmem(section_nr, page, SECTION_INFO);
139
140 usemap = __nr_to_section(section_nr)->pageblock_flags;
141 page = virt_to_page(usemap);
142
143 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
144
145 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700146 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700147
148}
149
150void register_page_bootmem_info_node(struct pglist_data *pgdat)
151{
152 unsigned long i, pfn, end_pfn, nr_pages;
153 int node = pgdat->node_id;
154 struct page *page;
155 struct zone *zone;
156
157 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
158 page = virt_to_page(pgdat);
159
160 for (i = 0; i < nr_pages; i++, page++)
161 get_page_bootmem(node, page, NODE_INFO);
162
163 zone = &pgdat->node_zones[0];
164 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
165 if (zone->wait_table) {
166 nr_pages = zone->wait_table_hash_nr_entries
167 * sizeof(wait_queue_head_t);
168 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
169 page = virt_to_page(zone->wait_table);
170
171 for (i = 0; i < nr_pages; i++, page++)
172 get_page_bootmem(node, page, NODE_INFO);
173 }
174 }
175
176 pfn = pgdat->node_start_pfn;
177 end_pfn = pfn + pgdat->node_spanned_pages;
178
179 /* register_section info */
180 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
181 register_page_bootmem_info_section(pfn);
182
183}
184#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
185
Heiko Carstens76cdd582008-05-14 16:05:52 -0700186static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
187 unsigned long end_pfn)
188{
189 unsigned long old_zone_end_pfn;
190
191 zone_span_writelock(zone);
192
193 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
194 if (start_pfn < zone->zone_start_pfn)
195 zone->zone_start_pfn = start_pfn;
196
197 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
198 zone->zone_start_pfn;
199
200 zone_span_writeunlock(zone);
201}
202
203static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
204 unsigned long end_pfn)
205{
206 unsigned long old_pgdat_end_pfn =
207 pgdat->node_start_pfn + pgdat->node_spanned_pages;
208
209 if (start_pfn < pgdat->node_start_pfn)
210 pgdat->node_start_pfn = start_pfn;
211
212 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
213 pgdat->node_start_pfn;
214}
215
Al Viro31168482008-11-22 17:33:24 +0000216static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700217{
218 struct pglist_data *pgdat = zone->zone_pgdat;
219 int nr_pages = PAGES_PER_SECTION;
220 int nid = pgdat->node_id;
221 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700222 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700223
224 zone_type = zone - pgdat->node_zones;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700225 if (!zone->wait_table) {
226 int ret;
227
228 ret = init_currently_empty_zone(zone, phys_start_pfn,
229 nr_pages, MEMMAP_HOTPLUG);
230 if (ret)
231 return ret;
232 }
233 pgdat_resize_lock(zone->zone_pgdat, &flags);
234 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
235 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
236 phys_start_pfn + nr_pages);
237 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa022007-01-10 23:15:30 -0800238 memmap_init_zone(nr_pages, nid, zone_type,
239 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700240 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700241}
242
Gary Hadec04fc582009-01-06 14:39:14 -0800243static int __meminit __add_section(int nid, struct zone *zone,
244 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700245{
Dave Hansen3947be12005-10-29 18:16:54 -0700246 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700247 int ret;
248
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700249 if (pfn_valid(phys_start_pfn))
250 return -EEXIST;
251
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700252 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700253
254 if (ret < 0)
255 return ret;
256
Yasunori Goto718127c2006-06-23 02:03:10 -0700257 ret = __add_zone(zone, phys_start_pfn);
258
259 if (ret < 0)
260 return ret;
261
Gary Hadec04fc582009-01-06 14:39:14 -0800262 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700263}
264
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700265#ifdef CONFIG_SPARSEMEM_VMEMMAP
266static int __remove_section(struct zone *zone, struct mem_section *ms)
267{
268 /*
269 * XXX: Freeing memmap with vmemmap is not implement yet.
270 * This should be removed later.
271 */
272 return -EBUSY;
273}
274#else
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700275static int __remove_section(struct zone *zone, struct mem_section *ms)
276{
277 unsigned long flags;
278 struct pglist_data *pgdat = zone->zone_pgdat;
279 int ret = -EINVAL;
280
281 if (!valid_section(ms))
282 return ret;
283
284 ret = unregister_memory_section(ms);
285 if (ret)
286 return ret;
287
288 pgdat_resize_lock(pgdat, &flags);
289 sparse_remove_one_section(zone, ms);
290 pgdat_resize_unlock(pgdat, &flags);
291 return 0;
292}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700293#endif
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700294
Dave Hansen3947be12005-10-29 18:16:54 -0700295/*
296 * Reasonably generic function for adding memory. It is
297 * expected that archs that support memory hotplug will
298 * call this function after deciding the zone to which to
299 * add the new pages.
300 */
Gary Hadec04fc582009-01-06 14:39:14 -0800301int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
302 unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700303{
304 unsigned long i;
305 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700306 int start_sec, end_sec;
307 /* during initialize mem_map, align hot-added range to section */
308 start_sec = pfn_to_section_nr(phys_start_pfn);
309 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700310
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700311 for (i = start_sec; i <= end_sec; i++) {
Gary Hadec04fc582009-01-06 14:39:14 -0800312 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700313
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700314 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200315 * EEXIST is finally dealt with by ioresource collision
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700316 * check. see add_memory() => register_memory_resource()
317 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700318 */
319 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700320 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700321 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700322 }
323
324 return err;
325}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700326EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700327
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700328/**
329 * __remove_pages() - remove sections of pages from a zone
330 * @zone: zone from which pages need to be removed
331 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
332 * @nr_pages: number of pages to remove (must be multiple of section size)
333 *
334 * Generic helper function to remove section mappings and sysfs entries
335 * for the section of the memory we are removing. Caller needs to make
336 * sure that pages are marked reserved and zones are adjust properly by
337 * calling offline_pages().
338 */
339int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
340 unsigned long nr_pages)
341{
342 unsigned long i, ret = 0;
343 int sections_to_remove;
344
345 /*
346 * We can only remove entire sections
347 */
348 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
349 BUG_ON(nr_pages % PAGES_PER_SECTION);
350
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700351 sections_to_remove = nr_pages / PAGES_PER_SECTION;
352 for (i = 0; i < sections_to_remove; i++) {
353 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Nathan Fontenotde7f0cb2008-10-18 20:27:14 -0700354 release_mem_region(pfn << PAGE_SHIFT,
355 PAGES_PER_SECTION << PAGE_SHIFT);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700356 ret = __remove_section(zone, __pfn_to_section(pfn));
357 if (ret)
358 break;
359 }
360 return ret;
361}
362EXPORT_SYMBOL_GPL(__remove_pages);
363
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700364void online_page(struct page *page)
365{
Jan Beulich4738e1b2009-09-21 17:03:03 -0700366 unsigned long pfn = page_to_pfn(page);
367
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700368 totalram_pages++;
Jan Beulich4738e1b2009-09-21 17:03:03 -0700369 if (pfn >= num_physpages)
370 num_physpages = pfn + 1;
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700371
372#ifdef CONFIG_HIGHMEM
373 if (PageHighMem(page))
374 totalhigh_pages++;
375#endif
376
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700377 ClearPageReserved(page);
378 init_page_count(page);
379 __free_page(page);
380}
381
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700382static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
383 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700384{
385 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700386 unsigned long onlined_pages = *(unsigned long *)arg;
387 struct page *page;
388 if (PageReserved(pfn_to_page(start_pfn)))
389 for (i = 0; i < nr_pages; i++) {
390 page = pfn_to_page(start_pfn + i);
391 online_page(page);
392 onlined_pages++;
393 }
394 *(unsigned long *)arg = onlined_pages;
395 return 0;
396}
397
398
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -0700399int __ref online_pages(unsigned long pfn, unsigned long nr_pages)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700400{
Dave Hansen3947be12005-10-29 18:16:54 -0700401 unsigned long onlined_pages = 0;
402 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700403 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700404 int nid;
405 int ret;
406 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700407
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900408 lock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700409 arg.start_pfn = pfn;
410 arg.nr_pages = nr_pages;
411 arg.status_change_nid = -1;
412
413 nid = page_to_nid(pfn_to_page(pfn));
414 if (node_present_pages(nid) == 0)
415 arg.status_change_nid = nid;
416
417 ret = memory_notify(MEM_GOING_ONLINE, &arg);
418 ret = notifier_to_errno(ret);
419 if (ret) {
420 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900421 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700422 return ret;
423 }
Dave Hansen3947be12005-10-29 18:16:54 -0700424 /*
425 * This doesn't need a lock to do pfn_to_page().
426 * The section can't be removed here because of the
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800427 * memory_block->state_mutex.
Dave Hansen3947be12005-10-29 18:16:54 -0700428 */
429 zone = page_zone(pfn_to_page(pfn));
Yasunori Goto68113782006-06-23 02:03:11 -0700430 /*
431 * If this zone is not populated, then it is not in zonelist.
432 * This means the page allocator ignores this zone.
433 * So, zonelist must be updated after online.
434 */
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700435 mutex_lock(&zonelists_mutex);
Yasunori Goto68113782006-06-23 02:03:11 -0700436 if (!populated_zone(zone))
437 need_zonelists_rebuild = 1;
438
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700439 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700440 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700441 if (ret) {
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700442 mutex_unlock(&zonelists_mutex);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700443 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
444 nr_pages, pfn);
445 memory_notify(MEM_CANCEL_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900446 unlock_memory_hotplug();
Geoff Levandfd8a4222008-05-14 16:05:50 -0700447 return ret;
448 }
449
Dave Hansen3947be12005-10-29 18:16:54 -0700450 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800451 zone->zone_pgdat->node_present_pages += onlined_pages;
Haicheng Li1f522502010-05-24 14:32:51 -0700452 if (need_zonelists_rebuild)
453 build_all_zonelists(zone);
454 else
455 zone_pcp_update(zone);
Dave Hansen3947be12005-10-29 18:16:54 -0700456
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700457 mutex_unlock(&zonelists_mutex);
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700458
459 init_per_zone_wmark_min();
460
Christoph Lameter7ea15302007-10-16 01:25:29 -0700461 if (onlined_pages) {
462 kswapd_run(zone_to_nid(zone));
463 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
464 }
Dave Hansen61b13992005-10-29 18:16:56 -0700465
Haicheng Li1f522502010-05-24 14:32:51 -0700466 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700467
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700468 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700469
470 if (onlined_pages)
471 memory_notify(MEM_ONLINE, &arg);
KAMEZAWA Hiroyuki925268a2011-01-11 16:44:01 +0900472 unlock_memory_hotplug();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700473
Dave Hansen3947be12005-10-29 18:16:54 -0700474 return 0;
475}
Keith Mannthey53947022006-09-30 23:27:08 -0700476#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700477
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800478/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
479static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700480{
481 struct pglist_data *pgdat;
482 unsigned long zones_size[MAX_NR_ZONES] = {0};
483 unsigned long zholes_size[MAX_NR_ZONES] = {0};
484 unsigned long start_pfn = start >> PAGE_SHIFT;
485
486 pgdat = arch_alloc_nodedata(nid);
487 if (!pgdat)
488 return NULL;
489
490 arch_refresh_nodedata(nid, pgdat);
491
492 /* we can use NODE_DATA(nid) from here */
493
494 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -0700495 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700496
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -0700497 /*
498 * The node we allocated has no zone fallback lists. For avoiding
499 * to access not-initialized zonelist, build here.
500 */
501 build_all_zonelists(NULL);
502
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700503 return pgdat;
504}
505
506static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
507{
508 arch_refresh_nodedata(nid, NULL);
509 arch_free_nodedata(pgdat);
510 return;
511}
512
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700513
minskey guocf234222010-05-24 14:32:41 -0700514/*
515 * called by cpu_up() to online a node without onlined memory.
516 */
517int mem_online_node(int nid)
518{
519 pg_data_t *pgdat;
520 int ret;
521
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800522 lock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -0700523 pgdat = hotadd_new_pgdat(nid, 0);
524 if (pgdat) {
525 ret = -ENOMEM;
526 goto out;
527 }
528 node_set_online(nid);
529 ret = register_one_node(nid);
530 BUG_ON(ret);
531
532out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800533 unlock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -0700534 return ret;
535}
536
Al Viro31168482008-11-22 17:33:24 +0000537/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
538int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -0700539{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700540 pg_data_t *pgdat = NULL;
541 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700542 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700543 int ret;
544
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800545 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -0800546
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700547 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800548 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700549 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800550 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700551
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700552 if (!node_online(nid)) {
553 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800554 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700555 if (!pgdat)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800556 goto out;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700557 new_pgdat = 1;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700558 }
559
Yasunori Gotobc02af92006-06-27 02:53:30 -0700560 /* call arch's memory hotadd */
561 ret = arch_add_memory(nid, start, size);
562
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700563 if (ret < 0)
564 goto error;
565
Yasunori Goto0fc44152006-06-27 02:53:38 -0700566 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700567 node_set_online(nid);
568
Yasunori Goto0fc44152006-06-27 02:53:38 -0700569 if (new_pgdat) {
570 ret = register_one_node(nid);
571 /*
572 * If sysfs file of new node can't create, cpu on the node
573 * can't be hot-added. There is no rollback way now.
574 * So, check by BUG_ON() to catch it reluctantly..
575 */
576 BUG_ON(ret);
577 }
578
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -0800579 /* create new memmap entry */
580 firmware_map_add_hotplug(start, start + size, "System RAM");
581
Andi Kleen6ad696d2009-11-17 14:06:22 -0800582 goto out;
583
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700584error:
585 /* rollback pgdat allocation and others */
586 if (new_pgdat)
587 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700588 if (res)
589 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700590
Andi Kleen6ad696d2009-11-17 14:06:22 -0800591out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800592 unlock_memory_hotplug();
Yasunori Gotobc02af92006-06-27 02:53:30 -0700593 return ret;
594}
595EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700596
597#ifdef CONFIG_MEMORY_HOTREMOVE
598/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700599 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
600 * set and the size of the free page is given by page_order(). Using this,
601 * the function determines if the pageblock contains only free pages.
602 * Due to buddy contraints, a free page at least the size of a pageblock will
603 * be located at the start of the pageblock
604 */
605static inline int pageblock_free(struct page *page)
606{
607 return PageBuddy(page) && page_order(page) >= pageblock_order;
608}
609
610/* Return the start of the next active pageblock after a given page */
611static struct page *next_active_pageblock(struct page *page)
612{
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700613 /* Ensure the starting page is pageblock-aligned */
614 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
615
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700616 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700617 if (pageblock_free(page)) {
618 int order;
619 /* be careful. we don't have locks, page_order can be changed.*/
620 order = page_order(page);
621 if ((order < MAX_ORDER) && (order >= pageblock_order))
622 return page + (1 << order);
623 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700624
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700625 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700626}
627
628/* Checks if this range of memory is likely to be hot-removable. */
629int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
630{
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700631 struct page *page = pfn_to_page(start_pfn);
632 struct page *end_page = page + nr_pages;
633
634 /* Check the starting page of each pageblock within the range */
635 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -0700636 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700637 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -0700638 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700639 }
640
641 /* All pageblocks in the memory block are likely to be hot-removable */
642 return 1;
643}
644
645/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700646 * Confirm all pages in a range [start, end) is belongs to the same zone.
647 */
648static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
649{
650 unsigned long pfn;
651 struct zone *zone = NULL;
652 struct page *page;
653 int i;
654 for (pfn = start_pfn;
655 pfn < end_pfn;
656 pfn += MAX_ORDER_NR_PAGES) {
657 i = 0;
658 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
659 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
660 i++;
661 if (i == MAX_ORDER_NR_PAGES)
662 continue;
663 page = pfn_to_page(pfn + i);
664 if (zone && page_zone(page) != zone)
665 return 0;
666 zone = page_zone(page);
667 }
668 return 1;
669}
670
671/*
672 * Scanning pfn is much easier than scanning lru list.
673 * Scan pfn from start to end and Find LRU page.
674 */
Andrew Morton7bbc0902010-10-26 14:22:05 -0700675static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700676{
677 unsigned long pfn;
678 struct page *page;
679 for (pfn = start; pfn < end; pfn++) {
680 if (pfn_valid(pfn)) {
681 page = pfn_to_page(pfn);
682 if (PageLRU(page))
683 return pfn;
684 }
685 }
686 return 0;
687}
688
689static struct page *
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800690hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700691{
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800692 /* This should be improooooved!! */
693 return alloc_page(GFP_HIGHUSER_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700694}
695
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700696#define NR_OFFLINE_AT_ONCE_PAGES (256)
697static int
698do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
699{
700 unsigned long pfn;
701 struct page *page;
702 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
703 int not_managed = 0;
704 int ret = 0;
705 LIST_HEAD(source);
706
707 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
708 if (!pfn_valid(pfn))
709 continue;
710 page = pfn_to_page(pfn);
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -0700711 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700712 continue;
713 /*
714 * We can skip free pages. And we can only deal with pages on
715 * LRU.
716 */
Nick Piggin62695a82008-10-18 20:26:09 -0700717 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700718 if (!ret) { /* Success */
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -0700719 put_page(page);
Nick Piggin62695a82008-10-18 20:26:09 -0700720 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700721 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800722 inc_zone_page_state(page, NR_ISOLATED_ANON +
723 page_is_file_cache(page));
724
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700725 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700726#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -0800727 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
728 pfn);
729 dump_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700730#endif
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -0700731 put_page(page);
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300732 /* Because we don't have big zone->lock. we should
Bob Liu809c4442010-10-26 14:22:10 -0700733 check this again here. */
734 if (page_count(page)) {
735 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -0700736 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -0700737 break;
738 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700739 }
740 }
Bob Liuf3ab2632010-10-26 14:22:10 -0700741 if (!list_empty(&source)) {
742 if (not_managed) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700743 putback_lru_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -0700744 goto out;
745 }
746 /* this function returns # of failed pages */
Mel Gorman77f1fe62011-01-13 15:45:57 -0800747 ret = migrate_pages(&source, hotremove_migrate_alloc, 0,
Mel Gorman7f0f2492011-01-13 15:45:58 -0800748 true, true);
Bob Liuf3ab2632010-10-26 14:22:10 -0700749 if (ret)
750 putback_lru_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700751 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700752out:
753 return ret;
754}
755
756/*
757 * remove from free_area[] and mark all as Reserved.
758 */
759static int
760offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
761 void *data)
762{
763 __offline_isolated_pages(start, start + nr_pages);
764 return 0;
765}
766
767static void
768offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
769{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700770 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700771 offline_isolated_pages_cb);
772}
773
774/*
775 * Check all pages in range, recoreded as memory resource, are isolated.
776 */
777static int
778check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
779 void *data)
780{
781 int ret;
782 long offlined = *(long *)data;
783 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
784 offlined = nr_pages;
785 if (!ret)
786 *(long *)data += offlined;
787 return ret;
788}
789
790static long
791check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
792{
793 long offlined = 0;
794 int ret;
795
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700796 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700797 check_pages_isolated_cb);
798 if (ret < 0)
799 offlined = (long)ret;
800 return offlined;
801}
802
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -0700803static int __ref offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700804 unsigned long end_pfn, unsigned long timeout)
805{
806 unsigned long pfn, nr_pages, expire;
807 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700808 int ret, drain, retry_max, node;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700809 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700810 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700811
812 BUG_ON(start_pfn >= end_pfn);
813 /* at least, alignment against pageblock is necessary */
814 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
815 return -EINVAL;
816 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
817 return -EINVAL;
818 /* This makes hotplug much easier...and readable.
819 we assume this for now. .*/
820 if (!test_pages_in_a_zone(start_pfn, end_pfn))
821 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700822
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800823 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -0800824
Yasunori Goto7b78d332007-10-21 16:41:36 -0700825 zone = page_zone(pfn_to_page(start_pfn));
826 node = zone_to_nid(zone);
827 nr_pages = end_pfn - start_pfn;
828
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700829 /* set above range as isolated */
830 ret = start_isolate_page_range(start_pfn, end_pfn);
831 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800832 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700833
834 arg.start_pfn = start_pfn;
835 arg.nr_pages = nr_pages;
836 arg.status_change_nid = -1;
837 if (nr_pages >= node_present_pages(node))
838 arg.status_change_nid = node;
839
840 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
841 ret = notifier_to_errno(ret);
842 if (ret)
843 goto failed_removal;
844
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700845 pfn = start_pfn;
846 expire = jiffies + timeout;
847 drain = 0;
848 retry_max = 5;
849repeat:
850 /* start memory hot removal */
851 ret = -EAGAIN;
852 if (time_after(jiffies, expire))
853 goto failed_removal;
854 ret = -EINTR;
855 if (signal_pending(current))
856 goto failed_removal;
857 ret = 0;
858 if (drain) {
859 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700860 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800861 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700862 }
863
864 pfn = scan_lru_pages(start_pfn, end_pfn);
865 if (pfn) { /* We have page on LRU */
866 ret = do_migrate_range(pfn, end_pfn);
867 if (!ret) {
868 drain = 1;
869 goto repeat;
870 } else {
871 if (ret < 0)
872 if (--retry_max == 0)
873 goto failed_removal;
874 yield();
875 drain = 1;
876 goto repeat;
877 }
878 }
879 /* drain all zone's lru pagevec, this is asyncronous... */
880 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700881 yield();
882 /* drain pcp pages , this is synchrouns. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800883 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700884 /* check again */
885 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
886 if (offlined_pages < 0) {
887 ret = -EBUSY;
888 goto failed_removal;
889 }
890 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
891 /* Ok, all of our target is islaoted.
892 We cannot do rollback at this point. */
893 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -0800894 /* reset pagetype flags and makes migrate type to be MOVABLE */
895 undo_isolate_page_range(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700896 /* removal success */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700897 zone->present_pages -= offlined_pages;
898 zone->zone_pgdat->node_present_pages -= offlined_pages;
899 totalram_pages -= offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700900
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -0700901 init_per_zone_wmark_min();
902
David Rientjes8fe23e02009-12-14 17:58:33 -0800903 if (!node_present_pages(node)) {
904 node_clear_state(node, N_HIGH_MEMORY);
905 kswapd_stop(node);
906 }
Minchan Kimbce73942009-06-16 15:32:50 -0700907
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700908 vm_total_pages = nr_free_pagecache_pages();
909 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700910
911 memory_notify(MEM_OFFLINE, &arg);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800912 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700913 return 0;
914
915failed_removal:
916 printk(KERN_INFO "memory offlining %lx to %lx failed\n",
917 start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700918 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700919 /* pushback to free area */
920 undo_isolate_page_range(start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700921
Andi Kleen6ad696d2009-11-17 14:06:22 -0800922out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800923 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700924 return ret;
925}
Badari Pulavarty71088782008-10-18 20:25:58 -0700926
927int remove_memory(u64 start, u64 size)
928{
929 unsigned long start_pfn, end_pfn;
930
931 start_pfn = PFN_DOWN(start);
932 end_pfn = start_pfn + PFN_DOWN(size);
933 return offline_pages(start_pfn, end_pfn, 120 * HZ);
934}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -0700935#else
936int remove_memory(u64 start, u64 size)
937{
938 return -EINVAL;
939}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700940#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavarty71088782008-10-18 20:25:58 -0700941EXPORT_SYMBOL_GPL(remove_memory);