blob: 84279127fcd3ab935f922b4c7e1156bb205d9964 [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;
Dave Hansena2f3aa022007-01-10 23:15:30 -080070 ret = init_currently_empty_zone(zone, phys_start_pfn,
71 nr_pages, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -070072 if (ret < 0)
73 return ret;
74 }
Dave Hansena2f3aa022007-01-10 23:15:30 -080075 memmap_init_zone(nr_pages, nid, zone_type,
76 phys_start_pfn, MEMMAP_HOTPLUG);
Yasunori Goto718127c2006-06-23 02:03:10 -070077 return 0;
Dave Hansen3947be12005-10-29 18:16:54 -070078}
79
Dave Hansen3947be12005-10-29 18:16:54 -070080static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
81{
Dave Hansen3947be12005-10-29 18:16:54 -070082 int nr_pages = PAGES_PER_SECTION;
Dave Hansen3947be12005-10-29 18:16:54 -070083 int ret;
84
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -070085 if (pfn_valid(phys_start_pfn))
86 return -EEXIST;
87
Dave Hansen0b0acbe2005-10-29 18:16:55 -070088 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
Dave Hansen3947be12005-10-29 18:16:54 -070089
90 if (ret < 0)
91 return ret;
92
Yasunori Goto718127c2006-06-23 02:03:10 -070093 ret = __add_zone(zone, phys_start_pfn);
94
95 if (ret < 0)
96 return ret;
97
Dave Hansen3947be12005-10-29 18:16:54 -070098 return register_new_memory(__pfn_to_section(phys_start_pfn));
99}
100
101/*
102 * Reasonably generic function for adding memory. It is
103 * expected that archs that support memory hotplug will
104 * call this function after deciding the zone to which to
105 * add the new pages.
106 */
107int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
108 unsigned long nr_pages)
109{
110 unsigned long i;
111 int err = 0;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700112 int start_sec, end_sec;
113 /* during initialize mem_map, align hot-added range to section */
114 start_sec = pfn_to_section_nr(phys_start_pfn);
115 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
Dave Hansen3947be12005-10-29 18:16:54 -0700116
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700117 for (i = start_sec; i <= end_sec; i++) {
118 err = __add_section(zone, i << PFN_SECTION_SHIFT);
Dave Hansen3947be12005-10-29 18:16:54 -0700119
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700120 /*
121 * EEXIST is finally dealed with by ioresource collision
122 * check. see add_memory() => register_memory_resource()
123 * Warning will be printed if there is collision.
Joel H Schoppbed120c2006-05-01 12:16:11 -0700124 */
125 if (err && (err != -EEXIST))
Dave Hansen3947be12005-10-29 18:16:54 -0700126 break;
KAMEZAWA Hiroyuki6f712712006-08-05 12:14:58 -0700127 err = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700128 }
129
130 return err;
131}
Joel H Schoppbed120c2006-05-01 12:16:11 -0700132EXPORT_SYMBOL_GPL(__add_pages);
Dave Hansen3947be12005-10-29 18:16:54 -0700133
134static void grow_zone_span(struct zone *zone,
135 unsigned long start_pfn, unsigned long end_pfn)
136{
137 unsigned long old_zone_end_pfn;
138
139 zone_span_writelock(zone);
140
141 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
142 if (start_pfn < zone->zone_start_pfn)
143 zone->zone_start_pfn = start_pfn;
144
Yasunori Goto25a6df92006-05-30 21:25:42 -0700145 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
146 zone->zone_start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700147
148 zone_span_writeunlock(zone);
149}
150
151static void grow_pgdat_span(struct pglist_data *pgdat,
152 unsigned long start_pfn, unsigned long end_pfn)
153{
154 unsigned long old_pgdat_end_pfn =
155 pgdat->node_start_pfn + pgdat->node_spanned_pages;
156
157 if (start_pfn < pgdat->node_start_pfn)
158 pgdat->node_start_pfn = start_pfn;
159
Yasunori Goto25a6df92006-05-30 21:25:42 -0700160 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
161 pgdat->node_start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700162}
163
164int online_pages(unsigned long pfn, unsigned long nr_pages)
165{
166 unsigned long i;
167 unsigned long flags;
168 unsigned long onlined_pages = 0;
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700169 struct resource res;
170 u64 section_end;
171 unsigned long start_pfn;
Dave Hansen3947be12005-10-29 18:16:54 -0700172 struct zone *zone;
Yasunori Goto68113782006-06-23 02:03:11 -0700173 int need_zonelists_rebuild = 0;
Dave Hansen3947be12005-10-29 18:16:54 -0700174
175 /*
176 * This doesn't need a lock to do pfn_to_page().
177 * The section can't be removed here because of the
178 * memory_block->state_sem.
179 */
180 zone = page_zone(pfn_to_page(pfn));
181 pgdat_resize_lock(zone->zone_pgdat, &flags);
182 grow_zone_span(zone, pfn, pfn + nr_pages);
183 grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
184 pgdat_resize_unlock(zone->zone_pgdat, &flags);
185
Yasunori Goto68113782006-06-23 02:03:11 -0700186 /*
187 * If this zone is not populated, then it is not in zonelist.
188 * This means the page allocator ignores this zone.
189 * So, zonelist must be updated after online.
190 */
191 if (!populated_zone(zone))
192 need_zonelists_rebuild = 1;
193
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700194 res.start = (u64)pfn << PAGE_SHIFT;
195 res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
196 res.flags = IORESOURCE_MEM; /* we just need system ram */
197 section_end = res.end;
198
KAMEZAWA Hiroyuki58c1b5b2006-08-05 12:15:01 -0700199 while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
KAMEZAWA Hiroyuki2842f112006-06-27 02:53:36 -0700200 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
201 nr_pages = (unsigned long)
202 ((res.end + 1 - res.start) >> PAGE_SHIFT);
203
204 if (PageReserved(pfn_to_page(start_pfn))) {
205 /* this region's page is not onlined now */
206 for (i = 0; i < nr_pages; i++) {
207 struct page *page = pfn_to_page(start_pfn + i);
208 online_page(page);
209 onlined_pages++;
210 }
211 }
212
213 res.start = res.end + 1;
214 res.end = section_end;
Dave Hansen3947be12005-10-29 18:16:54 -0700215 }
216 zone->present_pages += onlined_pages;
Yasunori Gotof2937be2006-03-09 17:33:51 -0800217 zone->zone_pgdat->node_present_pages += onlined_pages;
Dave Hansen3947be12005-10-29 18:16:54 -0700218
Dave Hansen61b13992005-10-29 18:16:56 -0700219 setup_per_zone_pages_min();
220
Yasunori Goto68113782006-06-23 02:03:11 -0700221 if (need_zonelists_rebuild)
222 build_all_zonelists();
KAMEZAWA Hiroyuki5a4d4362006-06-23 02:03:47 -0700223 vm_total_pages = nr_free_pagecache_pages();
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700224 writeback_set_ratelimit();
Dave Hansen3947be12005-10-29 18:16:54 -0700225 return 0;
226}
Keith Mannthey53947022006-09-30 23:27:08 -0700227#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700228
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700229static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
230{
231 struct pglist_data *pgdat;
232 unsigned long zones_size[MAX_NR_ZONES] = {0};
233 unsigned long zholes_size[MAX_NR_ZONES] = {0};
234 unsigned long start_pfn = start >> PAGE_SHIFT;
235
236 pgdat = arch_alloc_nodedata(nid);
237 if (!pgdat)
238 return NULL;
239
240 arch_refresh_nodedata(nid, pgdat);
241
242 /* we can use NODE_DATA(nid) from here */
243
244 /* init node's zones as empty zones, we don't have any present pages.*/
245 free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
246
247 return pgdat;
248}
249
250static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
251{
252 arch_refresh_nodedata(nid, NULL);
253 arch_free_nodedata(pgdat);
254 return;
255}
256
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -0700257
Yasunori Gotobc02af92006-06-27 02:53:30 -0700258int add_memory(int nid, u64 start, u64 size)
259{
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700260 pg_data_t *pgdat = NULL;
261 int new_pgdat = 0;
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700262 struct resource *res;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700263 int ret;
264
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700265 res = register_memory_resource(start, size);
266 if (!res)
267 return -EEXIST;
268
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700269 if (!node_online(nid)) {
270 pgdat = hotadd_new_pgdat(nid, start);
271 if (!pgdat)
272 return -ENOMEM;
273 new_pgdat = 1;
274 ret = kswapd_run(nid);
275 if (ret)
276 goto error;
277 }
278
Yasunori Gotobc02af92006-06-27 02:53:30 -0700279 /* call arch's memory hotadd */
280 ret = arch_add_memory(nid, start, size);
281
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700282 if (ret < 0)
283 goto error;
284
Yasunori Goto0fc44152006-06-27 02:53:38 -0700285 /* we online node here. we can't roll back from here. */
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700286 node_set_online(nid);
287
Paul Jackson38837fc2006-09-29 02:01:16 -0700288 cpuset_track_online_nodes();
289
Yasunori Goto0fc44152006-06-27 02:53:38 -0700290 if (new_pgdat) {
291 ret = register_one_node(nid);
292 /*
293 * If sysfs file of new node can't create, cpu on the node
294 * can't be hot-added. There is no rollback way now.
295 * So, check by BUG_ON() to catch it reluctantly..
296 */
297 BUG_ON(ret);
298 }
299
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700300 return ret;
301error:
302 /* rollback pgdat allocation and others */
303 if (new_pgdat)
304 rollback_node_hotadd(nid, pgdat);
KAMEZAWA Hiroyukiebd15302006-08-05 12:15:06 -0700305 if (res)
306 release_memory_resource(res);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -0700307
Yasunori Gotobc02af92006-06-27 02:53:30 -0700308 return ret;
309}
310EXPORT_SYMBOL_GPL(add_memory);