blob: 0c055a090f4df19e042312e2a30cef3140fb8558 [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>
Paul Jackson38837fc2006-09-29 02:01:16 -070025#include <linux/cpuset.h>
Dave Hansen3947be12005-10-29 18:16:54 -070026
27#include <asm/tlbflush.h>
28
Keith Mannthey45e0b782006-09-30 23:27:09 -070029/* add this memory to iomem resource */
30static struct resource *register_memory_resource(u64 start, u64 size)
31{
32 struct resource *res;
33 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
34 BUG_ON(!res);
35
36 res->name = "System RAM";
37 res->start = start;
38 res->end = start + size - 1;
39 res->flags = IORESOURCE_MEM;
40 if (request_resource(&iomem_resource, res) < 0) {
41 printk("System RAM resource %llx - %llx cannot be added\n",
42 (unsigned long long)res->start, (unsigned long long)res->end);
43 kfree(res);
44 res = NULL;
45 }
46 return res;
47}
48
49static void release_memory_resource(struct resource *res)
50{
51 if (!res)
52 return;
53 release_resource(res);
54 kfree(res);
55 return;
56}
57
58
Keith Mannthey53947022006-09-30 23:27:08 -070059#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasunori Goto718127c2006-06-23 02:03:10 -070060static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -070061{
62 struct pglist_data *pgdat = zone->zone_pgdat;
63 int nr_pages = PAGES_PER_SECTION;
64 int nid = pgdat->node_id;
65 int zone_type;
66
67 zone_type = zone - pgdat->node_zones;
Yasunori Goto718127c2006-06-23 02:03:10 -070068 if (!populated_zone(zone)) {
69 int ret = 0;
70 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
71 if (ret < 0)
72 return ret;
73 }
Dave Hansen3947be12005-10-29 18:16:54 -070074 memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
Yasunori Goto718127c2006-06-23 02:03:10 -070075 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -070076}
77
Dave Hansen3947be12005-10-29 18:16:54 -070078static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
79{
Dave Hansen3947be12005-10-29 18:16:54 -070080 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -070081 int ret;
82
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -070083 if (pfn_valid(phys_start_pfn))
84 return -EEXIST;
85
Dave Hansen0b0acbe2005-10-29 18:16:55 -070086 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -070087
88 if (ret < 0)
89 return ret;
90
Yasunori Goto718127c2006-06-23 02:03:10 -070091 ret = __add_zone(zone, phys_start_pfn);
92
93 if (ret < 0)
94 return ret;
95
Dave Hansen3947be12005-10-29 18:16:54 -070096 return register_new_memory(__pfn_to_section(phys_start_pfn));
97}
98
99/*
100 * Reasonably generic function for adding memory. It is
101 * expected that archs that support memory hotplug will
102 * call this function after deciding the zone to which to
103 * add the new pages.
104 */
105int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
106 unsigned long nr_pages)
107{
108 unsigned long i;
109 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700110 int start_sec, end_sec;
111 /* during initialize mem_map, align hot-added range to section */
112 start_sec = pfn_to_section_nr(phys_start_pfn);
113 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700114
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700115 for (i = start_sec; i <= end_sec; i++) {
116 err = __add_section(zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700117
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700118 /*
119 * EEXIST is finally dealed with by ioresource collision
120 * check. see add_memory() => register_memory_resource()
121 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700122 */
123 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700124 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700125 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700126 }
127
128 return err;
129}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700130EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700131
132static void grow_zone_span(struct zone *zone,
133 unsigned long start_pfn, unsigned long end_pfn)
134{
135 unsigned long old_zone_end_pfn;
136
137 zone_span_writelock(zone);
138
139 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
140 if (start_pfn < zone->zone_start_pfn)
141 zone->zone_start_pfn = start_pfn;
142
Yasunori Goto25a6df92006-05-30 21:25:42 -0700143 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
144 zone->zone_start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700145
146 zone_span_writeunlock(zone);
147}
148
149static void grow_pgdat_span(struct pglist_data *pgdat,
150 unsigned long start_pfn, unsigned long end_pfn)
151{
152 unsigned long old_pgdat_end_pfn =
153 pgdat->node_start_pfn + pgdat->node_spanned_pages;
154
155 if (start_pfn < pgdat->node_start_pfn)
156 pgdat->node_start_pfn = start_pfn;
157
Yasunori Goto25a6df92006-05-30 21:25:42 -0700158 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
159 pgdat->node_start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700160}
161
162int online_pages(unsigned long pfn, unsigned long nr_pages)
163{
164 unsigned long i;
165 unsigned long flags;
166 unsigned long onlined_pages = 0;
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700167 struct resource res;
168 u64 section_end;
169 unsigned long start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700170 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700171 int need_zonelists_rebuild = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700172
173 /*
174 * This doesn't need a lock to do pfn_to_page().
175 * The section can't be removed here because of the
176 * memory_block->state_sem.
177 */
178 zone = page_zone(pfn_to_page(pfn));
179 pgdat_resize_lock(zone->zone_pgdat, &flags);
180 grow_zone_span(zone, pfn, pfn + nr_pages);
181 grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
182 pgdat_resize_unlock(zone->zone_pgdat, &flags);
183
Yasunori Goto68113782006-06-23 02:03:11 -0700184 /*
185 * If this zone is not populated, then it is not in zonelist.
186 * This means the page allocator ignores this zone.
187 * So, zonelist must be updated after online.
188 */
189 if (!populated_zone(zone))
190 need_zonelists_rebuild = 1;
191
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700192 res.start = (u64)pfn << PAGE_SHIFT;
193 res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
194 res.flags = IORESOURCE_MEM; /* we just need system ram */
195 section_end = res.end;
196
KAMEZAWA Hiroyuki58c1b5b2006-08-05 12:15:01 -0700197 while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700198 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
199 nr_pages = (unsigned long)
200 ((res.end + 1 - res.start) >> PAGE_SHIFT);
201
202 if (PageReserved(pfn_to_page(start_pfn))) {
203 /* this region's page is not onlined now */
204 for (i = 0; i < nr_pages; i++) {
205 struct page *page = pfn_to_page(start_pfn + i);
206 online_page(page);
207 onlined_pages++;
208 }
209 }
210
211 res.start = res.end + 1;
212 res.end = section_end;
Dave Hansen3947be12005-10-29 18:16:54 -0700213 }
214 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800215 zone->zone_pgdat->node_present_pages += onlined_pages;
Dave Hansen3947be12005-10-29 18:16:54 -0700216
Dave Hansen61b13992005-10-29 18:16:56 -0700217 setup_per_zone_pages_min();
218
Yasunori Goto68113782006-06-23 02:03:11 -0700219 if (need_zonelists_rebuild)
220 build_all_zonelists();
KAMEZAWA Hiroyuki5a4d4362006-06-23 02:03:47 -0700221 vm_total_pages = nr_free_pagecache_pages();
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700222 writeback_set_ratelimit();
Dave Hansen3947be12005-10-29 18:16:54 -0700223 return 0;
224}
Keith Mannthey53947022006-09-30 23:27:08 -0700225#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700226
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700227static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
228{
229 struct pglist_data *pgdat;
230 unsigned long zones_size[MAX_NR_ZONES] = {0};
231 unsigned long zholes_size[MAX_NR_ZONES] = {0};
232 unsigned long start_pfn = start >> PAGE_SHIFT;
233
234 pgdat = arch_alloc_nodedata(nid);
235 if (!pgdat)
236 return NULL;
237
238 arch_refresh_nodedata(nid, pgdat);
239
240 /* we can use NODE_DATA(nid) from here */
241
242 /* init node's zones as empty zones, we don't have any present pages.*/
243 free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
244
245 return pgdat;
246}
247
248static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
249{
250 arch_refresh_nodedata(nid, NULL);
251 arch_free_nodedata(pgdat);
252 return;
253}
254
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700255
Yasunori Gotobc02af92006-06-27 02:53:30 -0700256int add_memory(int nid, u64 start, u64 size)
257{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700258 pg_data_t *pgdat = NULL;
259 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700260 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700261 int ret;
262
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700263 res = register_memory_resource(start, size);
264 if (!res)
265 return -EEXIST;
266
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700267 if (!node_online(nid)) {
268 pgdat = hotadd_new_pgdat(nid, start);
269 if (!pgdat)
270 return -ENOMEM;
271 new_pgdat = 1;
272 ret = kswapd_run(nid);
273 if (ret)
274 goto error;
275 }
276
Yasunori Gotobc02af92006-06-27 02:53:30 -0700277 /* call arch's memory hotadd */
278 ret = arch_add_memory(nid, start, size);
279
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700280 if (ret < 0)
281 goto error;
282
Yasunori Goto0fc44152006-06-27 02:53:38 -0700283 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700284 node_set_online(nid);
285
Paul Jackson38837fc2006-09-29 02:01:16 -0700286 cpuset_track_online_nodes();
287
Yasunori Goto0fc44152006-06-27 02:53:38 -0700288 if (new_pgdat) {
289 ret = register_one_node(nid);
290 /*
291 * If sysfs file of new node can't create, cpu on the node
292 * can't be hot-added. There is no rollback way now.
293 * So, check by BUG_ON() to catch it reluctantly..
294 */
295 BUG_ON(ret);
296 }
297
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700298 return ret;
299error:
300 /* rollback pgdat allocation and others */
301 if (new_pgdat)
302 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700303 if (res)
304 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700305
Yasunori Gotobc02af92006-06-27 02:53:30 -0700306 return ret;
307}
308EXPORT_SYMBOL_GPL(add_memory);