blob: 7666dbd328df7595943efc6f5a7f54d35e29303d [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 Mannthey53947022006-09-30 23:27:08 -070029#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Yasunori Goto718127c2006-06-23 02:03:10 -070030static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
Dave Hansen3947be12005-10-29 18:16:54 -070031{
32 struct pglist_data *pgdat = zone->zone_pgdat;
33 int nr_pages = PAGES_PER_SECTION;
34 int nid = pgdat->node_id;
35 int zone_type;
36
37 zone_type = zone - pgdat->node_zones;
Yasunori Goto718127c2006-06-23 02:03:10 -070038 if (!populated_zone(zone)) {
39 int ret = 0;
40 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
41 if (ret < 0)
42 return ret;
43 }
Dave Hansen3947be12005-10-29 18:16:54 -070044 memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
45 zonetable_add(zone, nid, zone_type, phys_start_pfn, nr_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -070046 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -070047}
48
Dave Hansen3947be12005-10-29 18:16:54 -070049static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
50{
Dave Hansen3947be12005-10-29 18:16:54 -070051 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -070052 int ret;
53
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -070054 if (pfn_valid(phys_start_pfn))
55 return -EEXIST;
56
Dave Hansen0b0acbec2005-10-29 18:16:55 -070057 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -070058
59 if (ret < 0)
60 return ret;
61
Yasunori Goto718127c2006-06-23 02:03:10 -070062 ret = __add_zone(zone, phys_start_pfn);
63
64 if (ret < 0)
65 return ret;
66
Dave Hansen3947be12005-10-29 18:16:54 -070067 return register_new_memory(__pfn_to_section(phys_start_pfn));
68}
69
70/*
71 * Reasonably generic function for adding memory. It is
72 * expected that archs that support memory hotplug will
73 * call this function after deciding the zone to which to
74 * add the new pages.
75 */
76int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
77 unsigned long nr_pages)
78{
79 unsigned long i;
80 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -070081 int start_sec, end_sec;
82 /* during initialize mem_map, align hot-added range to section */
83 start_sec = pfn_to_section_nr(phys_start_pfn);
84 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -070085
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -070086 for (i = start_sec; i <= end_sec; i++) {
87 err = __add_section(zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -070088
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -070089 /*
90 * EEXIST is finally dealed with by ioresource collision
91 * check. see add_memory() => register_memory_resource()
92 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -070093 */
94 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -070095 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -070096 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -070097 }
98
99 return err;
100}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700101EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700102
103static void grow_zone_span(struct zone *zone,
104 unsigned long start_pfn, unsigned long end_pfn)
105{
106 unsigned long old_zone_end_pfn;
107
108 zone_span_writelock(zone);
109
110 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
111 if (start_pfn < zone->zone_start_pfn)
112 zone->zone_start_pfn = start_pfn;
113
Yasunori Goto25a6df92006-05-30 21:25:42 -0700114 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
115 zone->zone_start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700116
117 zone_span_writeunlock(zone);
118}
119
120static void grow_pgdat_span(struct pglist_data *pgdat,
121 unsigned long start_pfn, unsigned long end_pfn)
122{
123 unsigned long old_pgdat_end_pfn =
124 pgdat->node_start_pfn + pgdat->node_spanned_pages;
125
126 if (start_pfn < pgdat->node_start_pfn)
127 pgdat->node_start_pfn = start_pfn;
128
Yasunori Goto25a6df92006-05-30 21:25:42 -0700129 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
130 pgdat->node_start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700131}
132
133int online_pages(unsigned long pfn, unsigned long nr_pages)
134{
135 unsigned long i;
136 unsigned long flags;
137 unsigned long onlined_pages = 0;
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700138 struct resource res;
139 u64 section_end;
140 unsigned long start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700141 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700142 int need_zonelists_rebuild = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700143
144 /*
145 * This doesn't need a lock to do pfn_to_page().
146 * The section can't be removed here because of the
147 * memory_block->state_sem.
148 */
149 zone = page_zone(pfn_to_page(pfn));
150 pgdat_resize_lock(zone->zone_pgdat, &flags);
151 grow_zone_span(zone, pfn, pfn + nr_pages);
152 grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
153 pgdat_resize_unlock(zone->zone_pgdat, &flags);
154
Yasunori Goto68113782006-06-23 02:03:11 -0700155 /*
156 * If this zone is not populated, then it is not in zonelist.
157 * This means the page allocator ignores this zone.
158 * So, zonelist must be updated after online.
159 */
160 if (!populated_zone(zone))
161 need_zonelists_rebuild = 1;
162
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700163 res.start = (u64)pfn << PAGE_SHIFT;
164 res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
165 res.flags = IORESOURCE_MEM; /* we just need system ram */
166 section_end = res.end;
167
KAMEZAWA Hiroyuki58c1b5b2006-08-05 12:15:01 -0700168 while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700169 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
170 nr_pages = (unsigned long)
171 ((res.end + 1 - res.start) >> PAGE_SHIFT);
172
173 if (PageReserved(pfn_to_page(start_pfn))) {
174 /* this region's page is not onlined now */
175 for (i = 0; i < nr_pages; i++) {
176 struct page *page = pfn_to_page(start_pfn + i);
177 online_page(page);
178 onlined_pages++;
179 }
180 }
181
182 res.start = res.end + 1;
183 res.end = section_end;
Dave Hansen3947be12005-10-29 18:16:54 -0700184 }
185 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800186 zone->zone_pgdat->node_present_pages += onlined_pages;
Dave Hansen3947be12005-10-29 18:16:54 -0700187
Dave Hansen61b13992005-10-29 18:16:56 -0700188 setup_per_zone_pages_min();
189
Yasunori Goto68113782006-06-23 02:03:11 -0700190 if (need_zonelists_rebuild)
191 build_all_zonelists();
KAMEZAWA Hiroyuki5a4d4362006-06-23 02:03:47 -0700192 vm_total_pages = nr_free_pagecache_pages();
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700193 writeback_set_ratelimit();
Dave Hansen3947be12005-10-29 18:16:54 -0700194 return 0;
195}
Keith Mannthey53947022006-09-30 23:27:08 -0700196#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700197
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700198static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
199{
200 struct pglist_data *pgdat;
201 unsigned long zones_size[MAX_NR_ZONES] = {0};
202 unsigned long zholes_size[MAX_NR_ZONES] = {0};
203 unsigned long start_pfn = start >> PAGE_SHIFT;
204
205 pgdat = arch_alloc_nodedata(nid);
206 if (!pgdat)
207 return NULL;
208
209 arch_refresh_nodedata(nid, pgdat);
210
211 /* we can use NODE_DATA(nid) from here */
212
213 /* init node's zones as empty zones, we don't have any present pages.*/
214 free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
215
216 return pgdat;
217}
218
219static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
220{
221 arch_refresh_nodedata(nid, NULL);
222 arch_free_nodedata(pgdat);
223 return;
224}
225
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700226/* add this memory to iomem resource */
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700227static struct resource *register_memory_resource(u64 start, u64 size)
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700228{
229 struct resource *res;
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700230 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
231 BUG_ON(!res);
232
233 res->name = "System RAM";
234 res->start = start;
235 res->end = start + size - 1;
236 res->flags = IORESOURCE_MEM;
237 if (request_resource(&iomem_resource, res) < 0) {
238 printk("System RAM resource %llx - %llx cannot be added\n",
239 (unsigned long long)res->start, (unsigned long long)res->end);
240 kfree(res);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700241 res = NULL;
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700242 }
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700243 return res;
244}
245
246static void release_memory_resource(struct resource *res)
247{
248 if (!res)
249 return;
250 release_resource(res);
251 kfree(res);
252 return;
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700253}
254
255
256
Yasunori Gotobc02af92006-06-27 02:53:30 -0700257int add_memory(int nid, u64 start, u64 size)
258{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700259 pg_data_t *pgdat = NULL;
260 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700261 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700262 int ret;
263
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700264 res = register_memory_resource(start, size);
265 if (!res)
266 return -EEXIST;
267
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700268 if (!node_online(nid)) {
269 pgdat = hotadd_new_pgdat(nid, start);
270 if (!pgdat)
271 return -ENOMEM;
272 new_pgdat = 1;
273 ret = kswapd_run(nid);
274 if (ret)
275 goto error;
276 }
277
Yasunori Gotobc02af92006-06-27 02:53:30 -0700278 /* call arch's memory hotadd */
279 ret = arch_add_memory(nid, start, size);
280
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700281 if (ret < 0)
282 goto error;
283
Yasunori Goto0fc44152006-06-27 02:53:38 -0700284 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700285 node_set_online(nid);
286
Paul Jackson38837fc2006-09-29 02:01:16 -0700287 cpuset_track_online_nodes();
288
Yasunori Goto0fc44152006-06-27 02:53:38 -0700289 if (new_pgdat) {
290 ret = register_one_node(nid);
291 /*
292 * If sysfs file of new node can't create, cpu on the node
293 * can't be hot-added. There is no rollback way now.
294 * So, check by BUG_ON() to catch it reluctantly..
295 */
296 BUG_ON(ret);
297 }
298
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700299 return ret;
300error:
301 /* rollback pgdat allocation and others */
302 if (new_pgdat)
303 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700304 if (res)
305 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700306
Yasunori Gotobc02af92006-06-27 02:53:30 -0700307 return ret;
308}
309EXPORT_SYMBOL_GPL(add_memory);