blob: 2c6523af54738faba3dd37eef938315503275de4 [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
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070085static void get_page_bootmem(unsigned long info, struct page *page, int type)
Yasunori Goto04753272008-04-28 02:13:31 -070086{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070087 atomic_set(&page->_mapcount, type);
Yasunori Goto04753272008-04-28 02:13:31 -070088 SetPagePrivate(page);
89 set_page_private(page, info);
90 atomic_inc(&page->_count);
91}
92
Rakib Mullick23ce9322009-12-14 17:59:44 -080093/* reference to __meminit __free_pages_bootmem is valid
94 * so use __ref to tell modpost not to generate a warning */
95void __ref put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -070096{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070097 int type;
Yasunori Goto04753272008-04-28 02:13:31 -070098
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070099 type = atomic_read(&page->_mapcount);
100 BUG_ON(type >= -1);
Yasunori Goto04753272008-04-28 02:13:31 -0700101
102 if (atomic_dec_return(&page->_count) == 1) {
103 ClearPagePrivate(page);
104 set_page_private(page, 0);
105 reset_page_mapcount(page);
106 __free_pages_bootmem(page, 0);
107 }
108
109}
110
Adrian Bunkd92bc312008-07-23 21:28:12 -0700111static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -0700112{
113 unsigned long *usemap, mapsize, section_nr, i;
114 struct mem_section *ms;
115 struct page *page, *memmap;
116
117 if (!pfn_valid(start_pfn))
118 return;
119
120 section_nr = pfn_to_section_nr(start_pfn);
121 ms = __nr_to_section(section_nr);
122
123 /* Get section's memmap address */
124 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
125
126 /*
127 * Get page for the memmap's phys address
128 * XXX: need more consideration for sparse_vmemmap...
129 */
130 page = virt_to_page(memmap);
131 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
132 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
133
134 /* remember memmap's page */
135 for (i = 0; i < mapsize; i++, page++)
136 get_page_bootmem(section_nr, page, SECTION_INFO);
137
138 usemap = __nr_to_section(section_nr)->pageblock_flags;
139 page = virt_to_page(usemap);
140
141 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
142
143 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700144 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700145
146}
147
148void register_page_bootmem_info_node(struct pglist_data *pgdat)
149{
150 unsigned long i, pfn, end_pfn, nr_pages;
151 int node = pgdat->node_id;
152 struct page *page;
153 struct zone *zone;
154
155 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
156 page = virt_to_page(pgdat);
157
158 for (i = 0; i < nr_pages; i++, page++)
159 get_page_bootmem(node, page, NODE_INFO);
160
161 zone = &pgdat->node_zones[0];
162 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
163 if (zone->wait_table) {
164 nr_pages = zone->wait_table_hash_nr_entries
165 * sizeof(wait_queue_head_t);
166 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
167 page = virt_to_page(zone->wait_table);
168
169 for (i = 0; i < nr_pages; i++, page++)
170 get_page_bootmem(node, page, NODE_INFO);
171 }
172 }
173
174 pfn = pgdat->node_start_pfn;
175 end_pfn = pfn + pgdat->node_spanned_pages;
176
177 /* register_section info */
178 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
179 register_page_bootmem_info_section(pfn);
180
181}
182#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
183
Heiko Carstens76cdd582008-05-14 16:05:52 -0700184static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
185 unsigned long end_pfn)
186{
187 unsigned long old_zone_end_pfn;
188
189 zone_span_writelock(zone);
190
191 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
192 if (start_pfn < zone->zone_start_pfn)
193 zone->zone_start_pfn = start_pfn;
194
195 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
196 zone->zone_start_pfn;
197
198 zone_span_writeunlock(zone);
199}
200
201static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
202 unsigned long end_pfn)
203{
204 unsigned long old_pgdat_end_pfn =
205 pgdat->node_start_pfn + pgdat->node_spanned_pages;
206
207 if (start_pfn < pgdat->node_start_pfn)
208 pgdat->node_start_pfn = start_pfn;
209
210 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
211 pgdat->node_start_pfn;
212}
213
Al Viro31168482008-11-22 17:33:24 +0000214static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700215{
216 struct pglist_data *pgdat = zone->zone_pgdat;
217 int nr_pages = PAGES_PER_SECTION;
218 int nid = pgdat->node_id;
219 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700220 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700221
222 zone_type = zone - pgdat->node_zones;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700223 if (!zone->wait_table) {
224 int ret;
225
226 ret = init_currently_empty_zone(zone, phys_start_pfn,
227 nr_pages, MEMMAP_HOTPLUG);
228 if (ret)
229 return ret;
230 }
231 pgdat_resize_lock(zone->zone_pgdat, &flags);
232 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
233 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
234 phys_start_pfn + nr_pages);
235 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa02007-01-10 23:15:30 -0800236 memmap_init_zone(nr_pages, nid, zone_type,
237 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700238 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700239}
240
Gary Hadec04fc582009-01-06 14:39:14 -0800241static int __meminit __add_section(int nid, struct zone *zone,
242 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700243{
Dave Hansen3947be12005-10-29 18:16:54 -0700244 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700245 int ret;
246
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700247 if (pfn_valid(phys_start_pfn))
248 return -EEXIST;
249
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700250 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700251
252 if (ret < 0)
253 return ret;
254
Yasunori Goto718127c2006-06-23 02:03:10 -0700255 ret = __add_zone(zone, phys_start_pfn);
256
257 if (ret < 0)
258 return ret;
259
Gary Hadec04fc582009-01-06 14:39:14 -0800260 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700261}
262
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700263#ifdef CONFIG_SPARSEMEM_VMEMMAP
264static int __remove_section(struct zone *zone, struct mem_section *ms)
265{
266 /*
267 * XXX: Freeing memmap with vmemmap is not implement yet.
268 * This should be removed later.
269 */
270 return -EBUSY;
271}
272#else
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700273static int __remove_section(struct zone *zone, struct mem_section *ms)
274{
275 unsigned long flags;
276 struct pglist_data *pgdat = zone->zone_pgdat;
277 int ret = -EINVAL;
278
279 if (!valid_section(ms))
280 return ret;
281
282 ret = unregister_memory_section(ms);
283 if (ret)
284 return ret;
285
286 pgdat_resize_lock(pgdat, &flags);
287 sparse_remove_one_section(zone, ms);
288 pgdat_resize_unlock(pgdat, &flags);
289 return 0;
290}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700291#endif
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700292
Dave Hansen3947be12005-10-29 18:16:54 -0700293/*
294 * Reasonably generic function for adding memory. It is
295 * expected that archs that support memory hotplug will
296 * call this function after deciding the zone to which to
297 * add the new pages.
298 */
Gary Hadec04fc582009-01-06 14:39:14 -0800299int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
300 unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700301{
302 unsigned long i;
303 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700304 int start_sec, end_sec;
305 /* during initialize mem_map, align hot-added range to section */
306 start_sec = pfn_to_section_nr(phys_start_pfn);
307 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700308
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700309 for (i = start_sec; i <= end_sec; i++) {
Gary Hadec04fc582009-01-06 14:39:14 -0800310 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700311
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700312 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200313 * EEXIST is finally dealt with by ioresource collision
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700314 * check. see add_memory() => register_memory_resource()
315 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700316 */
317 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700318 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700319 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700320 }
321
322 return err;
323}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700324EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700325
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700326/**
327 * __remove_pages() - remove sections of pages from a zone
328 * @zone: zone from which pages need to be removed
329 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
330 * @nr_pages: number of pages to remove (must be multiple of section size)
331 *
332 * Generic helper function to remove section mappings and sysfs entries
333 * for the section of the memory we are removing. Caller needs to make
334 * sure that pages are marked reserved and zones are adjust properly by
335 * calling offline_pages().
336 */
337int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
338 unsigned long nr_pages)
339{
340 unsigned long i, ret = 0;
341 int sections_to_remove;
342
343 /*
344 * We can only remove entire sections
345 */
346 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
347 BUG_ON(nr_pages % PAGES_PER_SECTION);
348
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700349 sections_to_remove = nr_pages / PAGES_PER_SECTION;
350 for (i = 0; i < sections_to_remove; i++) {
351 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Nathan Fontenotde7f0cb2008-10-18 20:27:14 -0700352 release_mem_region(pfn << PAGE_SHIFT,
353 PAGES_PER_SECTION << PAGE_SHIFT);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700354 ret = __remove_section(zone, __pfn_to_section(pfn));
355 if (ret)
356 break;
357 }
358 return ret;
359}
360EXPORT_SYMBOL_GPL(__remove_pages);
361
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700362void online_page(struct page *page)
363{
Jan Beulich4738e1b2009-09-21 17:03:03 -0700364 unsigned long pfn = page_to_pfn(page);
365
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700366 totalram_pages++;
Jan Beulich4738e1b2009-09-21 17:03:03 -0700367 if (pfn >= num_physpages)
368 num_physpages = pfn + 1;
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700369
370#ifdef CONFIG_HIGHMEM
371 if (PageHighMem(page))
372 totalhigh_pages++;
373#endif
374
375#ifdef CONFIG_FLATMEM
376 max_mapnr = max(page_to_pfn(page), max_mapnr);
377#endif
378
379 ClearPageReserved(page);
380 init_page_count(page);
381 __free_page(page);
382}
383
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700384static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
385 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700386{
387 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700388 unsigned long onlined_pages = *(unsigned long *)arg;
389 struct page *page;
390 if (PageReserved(pfn_to_page(start_pfn)))
391 for (i = 0; i < nr_pages; i++) {
392 page = pfn_to_page(start_pfn + i);
393 online_page(page);
394 onlined_pages++;
395 }
396 *(unsigned long *)arg = onlined_pages;
397 return 0;
398}
399
400
401int online_pages(unsigned long pfn, unsigned long nr_pages)
402{
Dave Hansen3947be12005-10-29 18:16:54 -0700403 unsigned long onlined_pages = 0;
404 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700405 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700406 int nid;
407 int ret;
408 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700409
Yasunori Goto7b78d332007-10-21 16:41:36 -0700410 arg.start_pfn = pfn;
411 arg.nr_pages = nr_pages;
412 arg.status_change_nid = -1;
413
414 nid = page_to_nid(pfn_to_page(pfn));
415 if (node_present_pages(nid) == 0)
416 arg.status_change_nid = nid;
417
418 ret = memory_notify(MEM_GOING_ONLINE, &arg);
419 ret = notifier_to_errno(ret);
420 if (ret) {
421 memory_notify(MEM_CANCEL_ONLINE, &arg);
422 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);
446 return ret;
447 }
448
Dave Hansen3947be12005-10-29 18:16:54 -0700449 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800450 zone->zone_pgdat->node_present_pages += onlined_pages;
Haicheng Li1f522502010-05-24 14:32:51 -0700451 if (need_zonelists_rebuild)
452 build_all_zonelists(zone);
453 else
454 zone_pcp_update(zone);
Dave Hansen3947be12005-10-29 18:16:54 -0700455
Haicheng Li4eaf3f62010-05-24 14:32:52 -0700456 mutex_unlock(&zonelists_mutex);
Minchan Kimbc75d332009-06-16 15:32:48 -0700457 setup_per_zone_wmarks();
Minchan Kimbce73942009-06-16 15:32:50 -0700458 calculate_zone_inactive_ratio(zone);
Christoph Lameter7ea15302007-10-16 01:25:29 -0700459 if (onlined_pages) {
460 kswapd_run(zone_to_nid(zone));
461 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
462 }
Dave Hansen61b13992005-10-29 18:16:56 -0700463
Haicheng Li1f522502010-05-24 14:32:51 -0700464 vm_total_pages = nr_free_pagecache_pages();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700465
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700466 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700467
468 if (onlined_pages)
469 memory_notify(MEM_ONLINE, &arg);
470
Dave Hansen3947be12005-10-29 18:16:54 -0700471 return 0;
472}
Keith Mannthey53947022006-09-30 23:27:08 -0700473#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700474
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800475/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
476static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700477{
478 struct pglist_data *pgdat;
479 unsigned long zones_size[MAX_NR_ZONES] = {0};
480 unsigned long zholes_size[MAX_NR_ZONES] = {0};
481 unsigned long start_pfn = start >> PAGE_SHIFT;
482
483 pgdat = arch_alloc_nodedata(nid);
484 if (!pgdat)
485 return NULL;
486
487 arch_refresh_nodedata(nid, pgdat);
488
489 /* we can use NODE_DATA(nid) from here */
490
491 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -0700492 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700493
494 return pgdat;
495}
496
497static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
498{
499 arch_refresh_nodedata(nid, NULL);
500 arch_free_nodedata(pgdat);
501 return;
502}
503
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700504
minskey guocf234222010-05-24 14:32:41 -0700505/*
506 * called by cpu_up() to online a node without onlined memory.
507 */
508int mem_online_node(int nid)
509{
510 pg_data_t *pgdat;
511 int ret;
512
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800513 lock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -0700514 pgdat = hotadd_new_pgdat(nid, 0);
515 if (pgdat) {
516 ret = -ENOMEM;
517 goto out;
518 }
519 node_set_online(nid);
520 ret = register_one_node(nid);
521 BUG_ON(ret);
522
523out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800524 unlock_memory_hotplug();
minskey guocf234222010-05-24 14:32:41 -0700525 return ret;
526}
527
Al Viro31168482008-11-22 17:33:24 +0000528/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
529int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -0700530{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700531 pg_data_t *pgdat = NULL;
532 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700533 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700534 int ret;
535
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800536 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -0800537
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700538 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800539 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700540 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800541 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700542
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700543 if (!node_online(nid)) {
544 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800545 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700546 if (!pgdat)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800547 goto out;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700548 new_pgdat = 1;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700549 }
550
Yasunori Gotobc02af92006-06-27 02:53:30 -0700551 /* call arch's memory hotadd */
552 ret = arch_add_memory(nid, start, size);
553
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700554 if (ret < 0)
555 goto error;
556
Yasunori Goto0fc44152006-06-27 02:53:38 -0700557 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700558 node_set_online(nid);
559
Yasunori Goto0fc44152006-06-27 02:53:38 -0700560 if (new_pgdat) {
561 ret = register_one_node(nid);
562 /*
563 * If sysfs file of new node can't create, cpu on the node
564 * can't be hot-added. There is no rollback way now.
565 * So, check by BUG_ON() to catch it reluctantly..
566 */
567 BUG_ON(ret);
568 }
569
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -0800570 /* create new memmap entry */
571 firmware_map_add_hotplug(start, start + size, "System RAM");
572
Andi Kleen6ad696d2009-11-17 14:06:22 -0800573 goto out;
574
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700575error:
576 /* rollback pgdat allocation and others */
577 if (new_pgdat)
578 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700579 if (res)
580 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700581
Andi Kleen6ad696d2009-11-17 14:06:22 -0800582out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800583 unlock_memory_hotplug();
Yasunori Gotobc02af92006-06-27 02:53:30 -0700584 return ret;
585}
586EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700587
588#ifdef CONFIG_MEMORY_HOTREMOVE
589/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700590 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
591 * set and the size of the free page is given by page_order(). Using this,
592 * the function determines if the pageblock contains only free pages.
593 * Due to buddy contraints, a free page at least the size of a pageblock will
594 * be located at the start of the pageblock
595 */
596static inline int pageblock_free(struct page *page)
597{
598 return PageBuddy(page) && page_order(page) >= pageblock_order;
599}
600
601/* Return the start of the next active pageblock after a given page */
602static struct page *next_active_pageblock(struct page *page)
603{
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700604 /* Ensure the starting page is pageblock-aligned */
605 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
606
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700607 /* If the entire pageblock is free, move to the end of free page */
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700608 if (pageblock_free(page)) {
609 int order;
610 /* be careful. we don't have locks, page_order can be changed.*/
611 order = page_order(page);
612 if ((order < MAX_ORDER) && (order >= pageblock_order))
613 return page + (1 << order);
614 }
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700615
KAMEZAWA Hiroyuki0dcc48c2010-09-09 16:38:01 -0700616 return page + pageblock_nr_pages;
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700617}
618
619/* Checks if this range of memory is likely to be hot-removable. */
620int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
621{
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700622 struct page *page = pfn_to_page(start_pfn);
623 struct page *end_page = page + nr_pages;
624
625 /* Check the starting page of each pageblock within the range */
626 for (; page < end_page; page = next_active_pageblock(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -0700627 if (!is_pageblock_removable_nolock(page))
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700628 return 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -0700629 cond_resched();
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700630 }
631
632 /* All pageblocks in the memory block are likely to be hot-removable */
633 return 1;
634}
635
636/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700637 * Confirm all pages in a range [start, end) is belongs to the same zone.
638 */
639static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
640{
641 unsigned long pfn;
642 struct zone *zone = NULL;
643 struct page *page;
644 int i;
645 for (pfn = start_pfn;
646 pfn < end_pfn;
647 pfn += MAX_ORDER_NR_PAGES) {
648 i = 0;
649 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
650 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
651 i++;
652 if (i == MAX_ORDER_NR_PAGES)
653 continue;
654 page = pfn_to_page(pfn + i);
655 if (zone && page_zone(page) != zone)
656 return 0;
657 zone = page_zone(page);
658 }
659 return 1;
660}
661
662/*
663 * Scanning pfn is much easier than scanning lru list.
664 * Scan pfn from start to end and Find LRU page.
665 */
Andrew Morton7bbc0902010-10-26 14:22:05 -0700666static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700667{
668 unsigned long pfn;
669 struct page *page;
670 for (pfn = start; pfn < end; pfn++) {
671 if (pfn_valid(pfn)) {
672 page = pfn_to_page(pfn);
673 if (PageLRU(page))
674 return pfn;
675 }
676 }
677 return 0;
678}
679
680static struct page *
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800681hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700682{
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800683 /* This should be improooooved!! */
684 return alloc_page(GFP_HIGHUSER_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700685}
686
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700687#define NR_OFFLINE_AT_ONCE_PAGES (256)
688static int
689do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
690{
691 unsigned long pfn;
692 struct page *page;
693 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
694 int not_managed = 0;
695 int ret = 0;
696 LIST_HEAD(source);
697
698 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
699 if (!pfn_valid(pfn))
700 continue;
701 page = pfn_to_page(pfn);
702 if (!page_count(page))
703 continue;
704 /*
705 * We can skip free pages. And we can only deal with pages on
706 * LRU.
707 */
Nick Piggin62695a82008-10-18 20:26:09 -0700708 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700709 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -0700710 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700711 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800712 inc_zone_page_state(page, NR_ISOLATED_ANON +
713 page_is_file_cache(page));
714
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700715 } else {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700716#ifdef CONFIG_DEBUG_VM
Wu Fengguang718a3822010-03-10 15:20:43 -0800717 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
718 pfn);
719 dump_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700720#endif
Bob Liu809c4442010-10-26 14:22:10 -0700721 /* Becasue we don't have big zone->lock. we should
722 check this again here. */
723 if (page_count(page)) {
724 not_managed++;
Bob Liuf3ab2632010-10-26 14:22:10 -0700725 ret = -EBUSY;
Bob Liu809c4442010-10-26 14:22:10 -0700726 break;
727 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700728 }
729 }
Bob Liuf3ab2632010-10-26 14:22:10 -0700730 if (!list_empty(&source)) {
731 if (not_managed) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700732 putback_lru_pages(&source);
Bob Liuf3ab2632010-10-26 14:22:10 -0700733 goto out;
734 }
735 /* this function returns # of failed pages */
736 ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
737 if (ret)
738 putback_lru_pages(&source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700739 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700740out:
741 return ret;
742}
743
744/*
745 * remove from free_area[] and mark all as Reserved.
746 */
747static int
748offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
749 void *data)
750{
751 __offline_isolated_pages(start, start + nr_pages);
752 return 0;
753}
754
755static void
756offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
757{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700758 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700759 offline_isolated_pages_cb);
760}
761
762/*
763 * Check all pages in range, recoreded as memory resource, are isolated.
764 */
765static int
766check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
767 void *data)
768{
769 int ret;
770 long offlined = *(long *)data;
771 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
772 offlined = nr_pages;
773 if (!ret)
774 *(long *)data += offlined;
775 return ret;
776}
777
778static long
779check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
780{
781 long offlined = 0;
782 int ret;
783
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700784 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700785 check_pages_isolated_cb);
786 if (ret < 0)
787 offlined = (long)ret;
788 return offlined;
789}
790
Andrew Mortonb4e655a2009-12-14 17:59:35 -0800791static int offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700792 unsigned long end_pfn, unsigned long timeout)
793{
794 unsigned long pfn, nr_pages, expire;
795 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700796 int ret, drain, retry_max, node;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700797 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700798 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700799
800 BUG_ON(start_pfn >= end_pfn);
801 /* at least, alignment against pageblock is necessary */
802 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
803 return -EINVAL;
804 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
805 return -EINVAL;
806 /* This makes hotplug much easier...and readable.
807 we assume this for now. .*/
808 if (!test_pages_in_a_zone(start_pfn, end_pfn))
809 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700810
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800811 lock_memory_hotplug();
Andi Kleen6ad696d2009-11-17 14:06:22 -0800812
Yasunori Goto7b78d332007-10-21 16:41:36 -0700813 zone = page_zone(pfn_to_page(start_pfn));
814 node = zone_to_nid(zone);
815 nr_pages = end_pfn - start_pfn;
816
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700817 /* set above range as isolated */
818 ret = start_isolate_page_range(start_pfn, end_pfn);
819 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800820 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700821
822 arg.start_pfn = start_pfn;
823 arg.nr_pages = nr_pages;
824 arg.status_change_nid = -1;
825 if (nr_pages >= node_present_pages(node))
826 arg.status_change_nid = node;
827
828 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
829 ret = notifier_to_errno(ret);
830 if (ret)
831 goto failed_removal;
832
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700833 pfn = start_pfn;
834 expire = jiffies + timeout;
835 drain = 0;
836 retry_max = 5;
837repeat:
838 /* start memory hot removal */
839 ret = -EAGAIN;
840 if (time_after(jiffies, expire))
841 goto failed_removal;
842 ret = -EINTR;
843 if (signal_pending(current))
844 goto failed_removal;
845 ret = 0;
846 if (drain) {
847 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700848 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800849 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700850 }
851
852 pfn = scan_lru_pages(start_pfn, end_pfn);
853 if (pfn) { /* We have page on LRU */
854 ret = do_migrate_range(pfn, end_pfn);
855 if (!ret) {
856 drain = 1;
857 goto repeat;
858 } else {
859 if (ret < 0)
860 if (--retry_max == 0)
861 goto failed_removal;
862 yield();
863 drain = 1;
864 goto repeat;
865 }
866 }
867 /* drain all zone's lru pagevec, this is asyncronous... */
868 lru_add_drain_all();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700869 yield();
870 /* drain pcp pages , this is synchrouns. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800871 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700872 /* check again */
873 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
874 if (offlined_pages < 0) {
875 ret = -EBUSY;
876 goto failed_removal;
877 }
878 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
879 /* Ok, all of our target is islaoted.
880 We cannot do rollback at this point. */
881 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -0800882 /* reset pagetype flags and makes migrate type to be MOVABLE */
883 undo_isolate_page_range(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700884 /* removal success */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700885 zone->present_pages -= offlined_pages;
886 zone->zone_pgdat->node_present_pages -= offlined_pages;
887 totalram_pages -= offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700888
Minchan Kimbce73942009-06-16 15:32:50 -0700889 setup_per_zone_wmarks();
890 calculate_zone_inactive_ratio(zone);
David Rientjes8fe23e02009-12-14 17:58:33 -0800891 if (!node_present_pages(node)) {
892 node_clear_state(node, N_HIGH_MEMORY);
893 kswapd_stop(node);
894 }
Minchan Kimbce73942009-06-16 15:32:50 -0700895
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700896 vm_total_pages = nr_free_pagecache_pages();
897 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700898
899 memory_notify(MEM_OFFLINE, &arg);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800900 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700901 return 0;
902
903failed_removal:
904 printk(KERN_INFO "memory offlining %lx to %lx failed\n",
905 start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700906 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700907 /* pushback to free area */
908 undo_isolate_page_range(start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700909
Andi Kleen6ad696d2009-11-17 14:06:22 -0800910out:
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800911 unlock_memory_hotplug();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700912 return ret;
913}
Badari Pulavarty71088782008-10-18 20:25:58 -0700914
915int remove_memory(u64 start, u64 size)
916{
917 unsigned long start_pfn, end_pfn;
918
919 start_pfn = PFN_DOWN(start);
920 end_pfn = start_pfn + PFN_DOWN(size);
921 return offline_pages(start_pfn, end_pfn, 120 * HZ);
922}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -0700923#else
924int remove_memory(u64 start, u64 size)
925{
926 return -EINVAL;
927}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700928#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavarty71088782008-10-18 20:25:58 -0700929EXPORT_SYMBOL_GPL(remove_memory);