blob: 030ce8a5bb0e758e7ea04cc0c851955275356426 [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>
Dave Hansen3947be12005-10-29 18:16:54 -070031
32#include <asm/tlbflush.h>
33
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030034#include "internal.h"
35
Keith Mannthey45e0b782006-09-30 23:27:09 -070036/* add this memory to iomem resource */
37static struct resource *register_memory_resource(u64 start, u64 size)
38{
39 struct resource *res;
40 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
41 BUG_ON(!res);
42
43 res->name = "System RAM";
44 res->start = start;
45 res->end = start + size - 1;
Yasunori Goto887c3cb2007-11-14 16:59:20 -080046 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
Keith Mannthey45e0b782006-09-30 23:27:09 -070047 if (request_resource(&iomem_resource, res) < 0) {
48 printk("System RAM resource %llx - %llx cannot be added\n",
49 (unsigned long long)res->start, (unsigned long long)res->end);
50 kfree(res);
51 res = NULL;
52 }
53 return res;
54}
55
56static void release_memory_resource(struct resource *res)
57{
58 if (!res)
59 return;
60 release_resource(res);
61 kfree(res);
62 return;
63}
64
Keith Mannthey53947022006-09-30 23:27:08 -070065#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasunori Goto04753272008-04-28 02:13:31 -070066#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070067static void get_page_bootmem(unsigned long info, struct page *page, int type)
Yasunori Goto04753272008-04-28 02:13:31 -070068{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070069 atomic_set(&page->_mapcount, type);
Yasunori Goto04753272008-04-28 02:13:31 -070070 SetPagePrivate(page);
71 set_page_private(page, info);
72 atomic_inc(&page->_count);
73}
74
Rakib Mullick23ce9322009-12-14 17:59:44 -080075/* reference to __meminit __free_pages_bootmem is valid
76 * so use __ref to tell modpost not to generate a warning */
77void __ref put_page_bootmem(struct page *page)
Yasunori Goto04753272008-04-28 02:13:31 -070078{
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070079 int type;
Yasunori Goto04753272008-04-28 02:13:31 -070080
Yasunori Gotoaf370fb2008-07-23 21:28:17 -070081 type = atomic_read(&page->_mapcount);
82 BUG_ON(type >= -1);
Yasunori Goto04753272008-04-28 02:13:31 -070083
84 if (atomic_dec_return(&page->_count) == 1) {
85 ClearPagePrivate(page);
86 set_page_private(page, 0);
87 reset_page_mapcount(page);
88 __free_pages_bootmem(page, 0);
89 }
90
91}
92
Adrian Bunkd92bc312008-07-23 21:28:12 -070093static void register_page_bootmem_info_section(unsigned long start_pfn)
Yasunori Goto04753272008-04-28 02:13:31 -070094{
95 unsigned long *usemap, mapsize, section_nr, i;
96 struct mem_section *ms;
97 struct page *page, *memmap;
98
99 if (!pfn_valid(start_pfn))
100 return;
101
102 section_nr = pfn_to_section_nr(start_pfn);
103 ms = __nr_to_section(section_nr);
104
105 /* Get section's memmap address */
106 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
107
108 /*
109 * Get page for the memmap's phys address
110 * XXX: need more consideration for sparse_vmemmap...
111 */
112 page = virt_to_page(memmap);
113 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
114 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
115
116 /* remember memmap's page */
117 for (i = 0; i < mapsize; i++, page++)
118 get_page_bootmem(section_nr, page, SECTION_INFO);
119
120 usemap = __nr_to_section(section_nr)->pageblock_flags;
121 page = virt_to_page(usemap);
122
123 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
124
125 for (i = 0; i < mapsize; i++, page++)
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700126 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
Yasunori Goto04753272008-04-28 02:13:31 -0700127
128}
129
130void register_page_bootmem_info_node(struct pglist_data *pgdat)
131{
132 unsigned long i, pfn, end_pfn, nr_pages;
133 int node = pgdat->node_id;
134 struct page *page;
135 struct zone *zone;
136
137 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
138 page = virt_to_page(pgdat);
139
140 for (i = 0; i < nr_pages; i++, page++)
141 get_page_bootmem(node, page, NODE_INFO);
142
143 zone = &pgdat->node_zones[0];
144 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
145 if (zone->wait_table) {
146 nr_pages = zone->wait_table_hash_nr_entries
147 * sizeof(wait_queue_head_t);
148 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
149 page = virt_to_page(zone->wait_table);
150
151 for (i = 0; i < nr_pages; i++, page++)
152 get_page_bootmem(node, page, NODE_INFO);
153 }
154 }
155
156 pfn = pgdat->node_start_pfn;
157 end_pfn = pfn + pgdat->node_spanned_pages;
158
159 /* register_section info */
160 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
161 register_page_bootmem_info_section(pfn);
162
163}
164#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
165
Heiko Carstens76cdd582008-05-14 16:05:52 -0700166static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
167 unsigned long end_pfn)
168{
169 unsigned long old_zone_end_pfn;
170
171 zone_span_writelock(zone);
172
173 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
174 if (start_pfn < zone->zone_start_pfn)
175 zone->zone_start_pfn = start_pfn;
176
177 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
178 zone->zone_start_pfn;
179
180 zone_span_writeunlock(zone);
181}
182
183static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
184 unsigned long end_pfn)
185{
186 unsigned long old_pgdat_end_pfn =
187 pgdat->node_start_pfn + pgdat->node_spanned_pages;
188
189 if (start_pfn < pgdat->node_start_pfn)
190 pgdat->node_start_pfn = start_pfn;
191
192 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
193 pgdat->node_start_pfn;
194}
195
Al Viro31168482008-11-22 17:33:24 +0000196static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700197{
198 struct pglist_data *pgdat = zone->zone_pgdat;
199 int nr_pages = PAGES_PER_SECTION;
200 int nid = pgdat->node_id;
201 int zone_type;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700202 unsigned long flags;
Dave Hansen3947be12005-10-29 18:16:54 -0700203
204 zone_type = zone - pgdat->node_zones;
Heiko Carstens76cdd582008-05-14 16:05:52 -0700205 if (!zone->wait_table) {
206 int ret;
207
208 ret = init_currently_empty_zone(zone, phys_start_pfn,
209 nr_pages, MEMMAP_HOTPLUG);
210 if (ret)
211 return ret;
212 }
213 pgdat_resize_lock(zone->zone_pgdat, &flags);
214 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
215 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
216 phys_start_pfn + nr_pages);
217 pgdat_resize_unlock(zone->zone_pgdat, &flags);
Dave Hansena2f3aa02007-01-10 23:15:30 -0800218 memmap_init_zone(nr_pages, nid, zone_type,
219 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -0700220 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700221}
222
Gary Hadec04fc582009-01-06 14:39:14 -0800223static int __meminit __add_section(int nid, struct zone *zone,
224 unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -0700225{
Dave Hansen3947be12005-10-29 18:16:54 -0700226 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -0700227 int ret;
228
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700229 if (pfn_valid(phys_start_pfn))
230 return -EEXIST;
231
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700232 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700233
234 if (ret < 0)
235 return ret;
236
Yasunori Goto718127c2006-06-23 02:03:10 -0700237 ret = __add_zone(zone, phys_start_pfn);
238
239 if (ret < 0)
240 return ret;
241
Gary Hadec04fc582009-01-06 14:39:14 -0800242 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
Dave Hansen3947be12005-10-29 18:16:54 -0700243}
244
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700245#ifdef CONFIG_SPARSEMEM_VMEMMAP
246static int __remove_section(struct zone *zone, struct mem_section *ms)
247{
248 /*
249 * XXX: Freeing memmap with vmemmap is not implement yet.
250 * This should be removed later.
251 */
252 return -EBUSY;
253}
254#else
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700255static int __remove_section(struct zone *zone, struct mem_section *ms)
256{
257 unsigned long flags;
258 struct pglist_data *pgdat = zone->zone_pgdat;
259 int ret = -EINVAL;
260
261 if (!valid_section(ms))
262 return ret;
263
264 ret = unregister_memory_section(ms);
265 if (ret)
266 return ret;
267
268 pgdat_resize_lock(pgdat, &flags);
269 sparse_remove_one_section(zone, ms);
270 pgdat_resize_unlock(pgdat, &flags);
271 return 0;
272}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700273#endif
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700274
Dave Hansen3947be12005-10-29 18:16:54 -0700275/*
276 * Reasonably generic function for adding memory. It is
277 * expected that archs that support memory hotplug will
278 * call this function after deciding the zone to which to
279 * add the new pages.
280 */
Gary Hadec04fc582009-01-06 14:39:14 -0800281int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
282 unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700283{
284 unsigned long i;
285 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700286 int start_sec, end_sec;
287 /* during initialize mem_map, align hot-added range to section */
288 start_sec = pfn_to_section_nr(phys_start_pfn);
289 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700290
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700291 for (i = start_sec; i <= end_sec; i++) {
Gary Hadec04fc582009-01-06 14:39:14 -0800292 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700293
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700294 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200295 * EEXIST is finally dealt with by ioresource collision
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700296 * check. see add_memory() => register_memory_resource()
297 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700298 */
299 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700300 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700301 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700302 }
303
304 return err;
305}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700306EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700307
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700308/**
309 * __remove_pages() - remove sections of pages from a zone
310 * @zone: zone from which pages need to be removed
311 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
312 * @nr_pages: number of pages to remove (must be multiple of section size)
313 *
314 * Generic helper function to remove section mappings and sysfs entries
315 * for the section of the memory we are removing. Caller needs to make
316 * sure that pages are marked reserved and zones are adjust properly by
317 * calling offline_pages().
318 */
319int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
320 unsigned long nr_pages)
321{
322 unsigned long i, ret = 0;
323 int sections_to_remove;
324
325 /*
326 * We can only remove entire sections
327 */
328 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
329 BUG_ON(nr_pages % PAGES_PER_SECTION);
330
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700331 sections_to_remove = nr_pages / PAGES_PER_SECTION;
332 for (i = 0; i < sections_to_remove; i++) {
333 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
Nathan Fontenotde7f0cb2008-10-18 20:27:14 -0700334 release_mem_region(pfn << PAGE_SHIFT,
335 PAGES_PER_SECTION << PAGE_SHIFT);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700336 ret = __remove_section(zone, __pfn_to_section(pfn));
337 if (ret)
338 break;
339 }
340 return ret;
341}
342EXPORT_SYMBOL_GPL(__remove_pages);
343
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700344void online_page(struct page *page)
345{
Jan Beulich4738e1b2009-09-21 17:03:03 -0700346 unsigned long pfn = page_to_pfn(page);
347
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700348 totalram_pages++;
Jan Beulich4738e1b2009-09-21 17:03:03 -0700349 if (pfn >= num_physpages)
350 num_physpages = pfn + 1;
Jeremy Fitzhardinge180c06e2008-04-28 02:12:03 -0700351
352#ifdef CONFIG_HIGHMEM
353 if (PageHighMem(page))
354 totalhigh_pages++;
355#endif
356
357#ifdef CONFIG_FLATMEM
358 max_mapnr = max(page_to_pfn(page), max_mapnr);
359#endif
360
361 ClearPageReserved(page);
362 init_page_count(page);
363 __free_page(page);
364}
365
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700366static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
367 void *arg)
Dave Hansen3947be12005-10-29 18:16:54 -0700368{
369 unsigned long i;
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700370 unsigned long onlined_pages = *(unsigned long *)arg;
371 struct page *page;
372 if (PageReserved(pfn_to_page(start_pfn)))
373 for (i = 0; i < nr_pages; i++) {
374 page = pfn_to_page(start_pfn + i);
375 online_page(page);
376 onlined_pages++;
377 }
378 *(unsigned long *)arg = onlined_pages;
379 return 0;
380}
381
382
383int online_pages(unsigned long pfn, unsigned long nr_pages)
384{
Dave Hansen3947be12005-10-29 18:16:54 -0700385 unsigned long onlined_pages = 0;
386 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700387 int need_zonelists_rebuild = 0;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700388 int nid;
389 int ret;
390 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -0700391
Yasunori Goto7b78d332007-10-21 16:41:36 -0700392 arg.start_pfn = pfn;
393 arg.nr_pages = nr_pages;
394 arg.status_change_nid = -1;
395
396 nid = page_to_nid(pfn_to_page(pfn));
397 if (node_present_pages(nid) == 0)
398 arg.status_change_nid = nid;
399
400 ret = memory_notify(MEM_GOING_ONLINE, &arg);
401 ret = notifier_to_errno(ret);
402 if (ret) {
403 memory_notify(MEM_CANCEL_ONLINE, &arg);
404 return ret;
405 }
Dave Hansen3947be12005-10-29 18:16:54 -0700406 /*
407 * This doesn't need a lock to do pfn_to_page().
408 * The section can't be removed here because of the
Daniel Walkerda19cbc2008-02-04 23:35:47 -0800409 * memory_block->state_mutex.
Dave Hansen3947be12005-10-29 18:16:54 -0700410 */
411 zone = page_zone(pfn_to_page(pfn));
Yasunori Goto68113782006-06-23 02:03:11 -0700412 /*
413 * If this zone is not populated, then it is not in zonelist.
414 * This means the page allocator ignores this zone.
415 * So, zonelist must be updated after online.
416 */
417 if (!populated_zone(zone))
418 need_zonelists_rebuild = 1;
419
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700420 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700421 online_pages_range);
Geoff Levandfd8a4222008-05-14 16:05:50 -0700422 if (ret) {
423 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
424 nr_pages, pfn);
425 memory_notify(MEM_CANCEL_ONLINE, &arg);
426 return ret;
427 }
428
Dave Hansen3947be12005-10-29 18:16:54 -0700429 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800430 zone->zone_pgdat->node_present_pages += onlined_pages;
Dave Hansen3947be12005-10-29 18:16:54 -0700431
Shaohua Li112067f2009-09-21 17:01:16 -0700432 zone_pcp_update(zone);
Minchan Kimbc75d332009-06-16 15:32:48 -0700433 setup_per_zone_wmarks();
Minchan Kimbce73942009-06-16 15:32:50 -0700434 calculate_zone_inactive_ratio(zone);
Christoph Lameter7ea15302007-10-16 01:25:29 -0700435 if (onlined_pages) {
436 kswapd_run(zone_to_nid(zone));
437 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
438 }
Dave Hansen61b13992005-10-29 18:16:56 -0700439
Yasunori Goto68113782006-06-23 02:03:11 -0700440 if (need_zonelists_rebuild)
441 build_all_zonelists();
Kent Liu2f7f24e2008-07-23 21:28:18 -0700442 else
443 vm_total_pages = nr_free_pagecache_pages();
444
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700445 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700446
447 if (onlined_pages)
448 memory_notify(MEM_ONLINE, &arg);
449
Dave Hansen3947be12005-10-29 18:16:54 -0700450 return 0;
451}
Keith Mannthey53947022006-09-30 23:27:08 -0700452#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700453
Hidetoshi Setoe1319332009-11-17 14:06:18 -0800454/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
455static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700456{
457 struct pglist_data *pgdat;
458 unsigned long zones_size[MAX_NR_ZONES] = {0};
459 unsigned long zholes_size[MAX_NR_ZONES] = {0};
460 unsigned long start_pfn = start >> PAGE_SHIFT;
461
462 pgdat = arch_alloc_nodedata(nid);
463 if (!pgdat)
464 return NULL;
465
466 arch_refresh_nodedata(nid, pgdat);
467
468 /* we can use NODE_DATA(nid) from here */
469
470 /* init node's zones as empty zones, we don't have any present pages.*/
Johannes Weiner9109fb72008-07-23 21:27:20 -0700471 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700472
473 return pgdat;
474}
475
476static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
477{
478 arch_refresh_nodedata(nid, NULL);
479 arch_free_nodedata(pgdat);
480 return;
481}
482
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700483
Al Viro31168482008-11-22 17:33:24 +0000484/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
485int __ref add_memory(int nid, u64 start, u64 size)
Yasunori Gotobc02af92006-06-27 02:53:30 -0700486{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700487 pg_data_t *pgdat = NULL;
488 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700489 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700490 int ret;
491
Andi Kleen6ad696d2009-11-17 14:06:22 -0800492 lock_system_sleep();
493
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700494 res = register_memory_resource(start, size);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800495 ret = -EEXIST;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700496 if (!res)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800497 goto out;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700498
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700499 if (!node_online(nid)) {
500 pgdat = hotadd_new_pgdat(nid, start);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800501 ret = -ENOMEM;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700502 if (!pgdat)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800503 goto out;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700504 new_pgdat = 1;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700505 }
506
Yasunori Gotobc02af92006-06-27 02:53:30 -0700507 /* call arch's memory hotadd */
508 ret = arch_add_memory(nid, start, size);
509
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700510 if (ret < 0)
511 goto error;
512
Yasunori Goto0fc44152006-06-27 02:53:38 -0700513 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700514 node_set_online(nid);
515
Yasunori Goto0fc44152006-06-27 02:53:38 -0700516 if (new_pgdat) {
517 ret = register_one_node(nid);
518 /*
519 * If sysfs file of new node can't create, cpu on the node
520 * can't be hot-added. There is no rollback way now.
521 * So, check by BUG_ON() to catch it reluctantly..
522 */
523 BUG_ON(ret);
524 }
525
Andi Kleen6ad696d2009-11-17 14:06:22 -0800526 goto out;
527
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700528error:
529 /* rollback pgdat allocation and others */
530 if (new_pgdat)
531 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700532 if (res)
533 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700534
Andi Kleen6ad696d2009-11-17 14:06:22 -0800535out:
536 unlock_system_sleep();
Yasunori Gotobc02af92006-06-27 02:53:30 -0700537 return ret;
538}
539EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700540
541#ifdef CONFIG_MEMORY_HOTREMOVE
542/*
Badari Pulavarty5c755e92008-07-23 21:28:19 -0700543 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
544 * set and the size of the free page is given by page_order(). Using this,
545 * the function determines if the pageblock contains only free pages.
546 * Due to buddy contraints, a free page at least the size of a pageblock will
547 * be located at the start of the pageblock
548 */
549static inline int pageblock_free(struct page *page)
550{
551 return PageBuddy(page) && page_order(page) >= pageblock_order;
552}
553
554/* Return the start of the next active pageblock after a given page */
555static struct page *next_active_pageblock(struct page *page)
556{
557 int pageblocks_stride;
558
559 /* Ensure the starting page is pageblock-aligned */
560 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
561
562 /* Move forward by at least 1 * pageblock_nr_pages */
563 pageblocks_stride = 1;
564
565 /* If the entire pageblock is free, move to the end of free page */
566 if (pageblock_free(page))
567 pageblocks_stride += page_order(page) - pageblock_order;
568
569 return page + (pageblocks_stride * pageblock_nr_pages);
570}
571
572/* Checks if this range of memory is likely to be hot-removable. */
573int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
574{
575 int type;
576 struct page *page = pfn_to_page(start_pfn);
577 struct page *end_page = page + nr_pages;
578
579 /* Check the starting page of each pageblock within the range */
580 for (; page < end_page; page = next_active_pageblock(page)) {
581 type = get_pageblock_migratetype(page);
582
583 /*
584 * A pageblock containing MOVABLE or free pages is considered
585 * removable
586 */
587 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
588 return 0;
589
590 /*
591 * A pageblock starting with a PageReserved page is not
592 * considered removable.
593 */
594 if (PageReserved(page))
595 return 0;
596 }
597
598 /* All pageblocks in the memory block are likely to be hot-removable */
599 return 1;
600}
601
602/*
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700603 * Confirm all pages in a range [start, end) is belongs to the same zone.
604 */
605static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
606{
607 unsigned long pfn;
608 struct zone *zone = NULL;
609 struct page *page;
610 int i;
611 for (pfn = start_pfn;
612 pfn < end_pfn;
613 pfn += MAX_ORDER_NR_PAGES) {
614 i = 0;
615 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
616 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
617 i++;
618 if (i == MAX_ORDER_NR_PAGES)
619 continue;
620 page = pfn_to_page(pfn + i);
621 if (zone && page_zone(page) != zone)
622 return 0;
623 zone = page_zone(page);
624 }
625 return 1;
626}
627
628/*
629 * Scanning pfn is much easier than scanning lru list.
630 * Scan pfn from start to end and Find LRU page.
631 */
632int scan_lru_pages(unsigned long start, unsigned long end)
633{
634 unsigned long pfn;
635 struct page *page;
636 for (pfn = start; pfn < end; pfn++) {
637 if (pfn_valid(pfn)) {
638 page = pfn_to_page(pfn);
639 if (PageLRU(page))
640 return pfn;
641 }
642 }
643 return 0;
644}
645
646static struct page *
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800647hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700648{
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800649 /* This should be improooooved!! */
650 return alloc_page(GFP_HIGHUSER_MOVABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700651}
652
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700653#define NR_OFFLINE_AT_ONCE_PAGES (256)
654static int
655do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
656{
657 unsigned long pfn;
658 struct page *page;
659 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
660 int not_managed = 0;
661 int ret = 0;
662 LIST_HEAD(source);
663
664 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
665 if (!pfn_valid(pfn))
666 continue;
667 page = pfn_to_page(pfn);
668 if (!page_count(page))
669 continue;
670 /*
671 * We can skip free pages. And we can only deal with pages on
672 * LRU.
673 */
Nick Piggin62695a82008-10-18 20:26:09 -0700674 ret = isolate_lru_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700675 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -0700676 list_add_tail(&page->lru, &source);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700677 move_pages--;
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800678 inc_zone_page_state(page, NR_ISOLATED_ANON +
679 page_is_file_cache(page));
680
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700681 } else {
682 /* Becasue we don't have big zone->lock. we should
683 check this again here. */
684 if (page_count(page))
685 not_managed++;
686#ifdef CONFIG_DEBUG_VM
687 printk(KERN_INFO "removing from LRU failed"
688 " %lx/%d/%lx\n",
689 pfn, page_count(page), page->flags);
690#endif
691 }
692 }
693 ret = -EBUSY;
694 if (not_managed) {
695 if (!list_empty(&source))
696 putback_lru_pages(&source);
697 goto out;
698 }
699 ret = 0;
700 if (list_empty(&source))
701 goto out;
702 /* this function returns # of failed pages */
Hugh Dickins62b61f62009-12-14 17:59:33 -0800703 ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700704
705out:
706 return ret;
707}
708
709/*
710 * remove from free_area[] and mark all as Reserved.
711 */
712static int
713offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
714 void *data)
715{
716 __offline_isolated_pages(start, start + nr_pages);
717 return 0;
718}
719
720static void
721offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
722{
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700723 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700724 offline_isolated_pages_cb);
725}
726
727/*
728 * Check all pages in range, recoreded as memory resource, are isolated.
729 */
730static int
731check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
732 void *data)
733{
734 int ret;
735 long offlined = *(long *)data;
736 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
737 offlined = nr_pages;
738 if (!ret)
739 *(long *)data += offlined;
740 return ret;
741}
742
743static long
744check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
745{
746 long offlined = 0;
747 int ret;
748
KAMEZAWA Hiroyuki908eedc2009-09-22 16:45:46 -0700749 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700750 check_pages_isolated_cb);
751 if (ret < 0)
752 offlined = (long)ret;
753 return offlined;
754}
755
Andrew Mortonb4e655a2009-12-14 17:59:35 -0800756static int offline_pages(unsigned long start_pfn,
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700757 unsigned long end_pfn, unsigned long timeout)
758{
759 unsigned long pfn, nr_pages, expire;
760 long offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700761 int ret, drain, retry_max, node;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700762 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700763 struct memory_notify arg;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700764
765 BUG_ON(start_pfn >= end_pfn);
766 /* at least, alignment against pageblock is necessary */
767 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
768 return -EINVAL;
769 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
770 return -EINVAL;
771 /* This makes hotplug much easier...and readable.
772 we assume this for now. .*/
773 if (!test_pages_in_a_zone(start_pfn, end_pfn))
774 return -EINVAL;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700775
Andi Kleen6ad696d2009-11-17 14:06:22 -0800776 lock_system_sleep();
777
Yasunori Goto7b78d332007-10-21 16:41:36 -0700778 zone = page_zone(pfn_to_page(start_pfn));
779 node = zone_to_nid(zone);
780 nr_pages = end_pfn - start_pfn;
781
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700782 /* set above range as isolated */
783 ret = start_isolate_page_range(start_pfn, end_pfn);
784 if (ret)
Andi Kleen6ad696d2009-11-17 14:06:22 -0800785 goto out;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700786
787 arg.start_pfn = start_pfn;
788 arg.nr_pages = nr_pages;
789 arg.status_change_nid = -1;
790 if (nr_pages >= node_present_pages(node))
791 arg.status_change_nid = node;
792
793 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
794 ret = notifier_to_errno(ret);
795 if (ret)
796 goto failed_removal;
797
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700798 pfn = start_pfn;
799 expire = jiffies + timeout;
800 drain = 0;
801 retry_max = 5;
802repeat:
803 /* start memory hot removal */
804 ret = -EAGAIN;
805 if (time_after(jiffies, expire))
806 goto failed_removal;
807 ret = -EINTR;
808 if (signal_pending(current))
809 goto failed_removal;
810 ret = 0;
811 if (drain) {
812 lru_add_drain_all();
813 flush_scheduled_work();
814 cond_resched();
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800815 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700816 }
817
818 pfn = scan_lru_pages(start_pfn, end_pfn);
819 if (pfn) { /* We have page on LRU */
820 ret = do_migrate_range(pfn, end_pfn);
821 if (!ret) {
822 drain = 1;
823 goto repeat;
824 } else {
825 if (ret < 0)
826 if (--retry_max == 0)
827 goto failed_removal;
828 yield();
829 drain = 1;
830 goto repeat;
831 }
832 }
833 /* drain all zone's lru pagevec, this is asyncronous... */
834 lru_add_drain_all();
835 flush_scheduled_work();
836 yield();
837 /* drain pcp pages , this is synchrouns. */
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800838 drain_all_pages();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700839 /* check again */
840 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
841 if (offlined_pages < 0) {
842 ret = -EBUSY;
843 goto failed_removal;
844 }
845 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
846 /* Ok, all of our target is islaoted.
847 We cannot do rollback at this point. */
848 offline_isolated_pages(start_pfn, end_pfn);
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -0800849 /* reset pagetype flags and makes migrate type to be MOVABLE */
850 undo_isolate_page_range(start_pfn, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700851 /* removal success */
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700852 zone->present_pages -= offlined_pages;
853 zone->zone_pgdat->node_present_pages -= offlined_pages;
854 totalram_pages -= offlined_pages;
Yasunori Goto7b78d332007-10-21 16:41:36 -0700855
Minchan Kimbce73942009-06-16 15:32:50 -0700856 setup_per_zone_wmarks();
857 calculate_zone_inactive_ratio(zone);
David Rientjes8fe23e02009-12-14 17:58:33 -0800858 if (!node_present_pages(node)) {
859 node_clear_state(node, N_HIGH_MEMORY);
860 kswapd_stop(node);
861 }
Minchan Kimbce73942009-06-16 15:32:50 -0700862
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700863 vm_total_pages = nr_free_pagecache_pages();
864 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -0700865
866 memory_notify(MEM_OFFLINE, &arg);
Andi Kleen6ad696d2009-11-17 14:06:22 -0800867 unlock_system_sleep();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700868 return 0;
869
870failed_removal:
871 printk(KERN_INFO "memory offlining %lx to %lx failed\n",
872 start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700873 memory_notify(MEM_CANCEL_OFFLINE, &arg);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700874 /* pushback to free area */
875 undo_isolate_page_range(start_pfn, end_pfn);
Yasunori Goto7b78d332007-10-21 16:41:36 -0700876
Andi Kleen6ad696d2009-11-17 14:06:22 -0800877out:
878 unlock_system_sleep();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700879 return ret;
880}
Badari Pulavarty71088782008-10-18 20:25:58 -0700881
882int remove_memory(u64 start, u64 size)
883{
884 unsigned long start_pfn, end_pfn;
885
886 start_pfn = PFN_DOWN(start);
887 end_pfn = start_pfn + PFN_DOWN(size);
888 return offline_pages(start_pfn, end_pfn, 120 * HZ);
889}
KAMEZAWA Hiroyuki48e94192007-10-16 01:26:14 -0700890#else
891int remove_memory(u64 start, u64 size)
892{
893 return -EINVAL;
894}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -0700895#endif /* CONFIG_MEMORY_HOTREMOVE */
Badari Pulavarty71088782008-10-18 20:25:58 -0700896EXPORT_SYMBOL_GPL(remove_memory);