2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.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>
16 #include <linux/slab.h>
17 #include <linux/sysctl.h>
18 #include <linux/cpu.h>
19 #include <linux/memory.h>
20 #include <linux/memory_hotplug.h>
21 #include <linux/highmem.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ioport.h>
24 #include <linux/cpuset.h>
26 #include <asm/tlbflush.h>
28 extern void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
30 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
32 struct pglist_data *pgdat = zone->zone_pgdat;
33 int nr_pages = PAGES_PER_SECTION;
34 int nid = pgdat->node_id;
37 zone_type = zone - pgdat->node_zones;
38 if (!populated_zone(zone)) {
40 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
44 memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
45 zonetable_add(zone, nid, zone_type, phys_start_pfn, nr_pages);
49 extern int sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
51 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
53 int nr_pages = PAGES_PER_SECTION;
56 if (pfn_valid(phys_start_pfn))
59 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
64 ret = __add_zone(zone, phys_start_pfn);
69 return register_new_memory(__pfn_to_section(phys_start_pfn));
73 * Reasonably generic function for adding memory. It is
74 * expected that archs that support memory hotplug will
75 * call this function after deciding the zone to which to
78 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
79 unsigned long nr_pages)
83 int start_sec, end_sec;
84 /* during initialize mem_map, align hot-added range to section */
85 start_sec = pfn_to_section_nr(phys_start_pfn);
86 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
88 for (i = start_sec; i <= end_sec; i++) {
89 err = __add_section(zone, i << PFN_SECTION_SHIFT);
92 * EEXIST is finally dealed with by ioresource collision
93 * check. see add_memory() => register_memory_resource()
94 * Warning will be printed if there is collision.
96 if (err && (err != -EEXIST))
103 EXPORT_SYMBOL_GPL(__add_pages);
105 static void grow_zone_span(struct zone *zone,
106 unsigned long start_pfn, unsigned long end_pfn)
108 unsigned long old_zone_end_pfn;
110 zone_span_writelock(zone);
112 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
113 if (start_pfn < zone->zone_start_pfn)
114 zone->zone_start_pfn = start_pfn;
116 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
117 zone->zone_start_pfn;
119 zone_span_writeunlock(zone);
122 static void grow_pgdat_span(struct pglist_data *pgdat,
123 unsigned long start_pfn, unsigned long end_pfn)
125 unsigned long old_pgdat_end_pfn =
126 pgdat->node_start_pfn + pgdat->node_spanned_pages;
128 if (start_pfn < pgdat->node_start_pfn)
129 pgdat->node_start_pfn = start_pfn;
131 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
132 pgdat->node_start_pfn;
135 int online_pages(unsigned long pfn, unsigned long nr_pages)
139 unsigned long onlined_pages = 0;
142 unsigned long start_pfn;
144 int need_zonelists_rebuild = 0;
147 * This doesn't need a lock to do pfn_to_page().
148 * The section can't be removed here because of the
149 * memory_block->state_sem.
151 zone = page_zone(pfn_to_page(pfn));
152 pgdat_resize_lock(zone->zone_pgdat, &flags);
153 grow_zone_span(zone, pfn, pfn + nr_pages);
154 grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
155 pgdat_resize_unlock(zone->zone_pgdat, &flags);
158 * If this zone is not populated, then it is not in zonelist.
159 * This means the page allocator ignores this zone.
160 * So, zonelist must be updated after online.
162 if (!populated_zone(zone))
163 need_zonelists_rebuild = 1;
165 res.start = (u64)pfn << PAGE_SHIFT;
166 res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
167 res.flags = IORESOURCE_MEM; /* we just need system ram */
168 section_end = res.end;
170 while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
171 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
172 nr_pages = (unsigned long)
173 ((res.end + 1 - res.start) >> PAGE_SHIFT);
175 if (PageReserved(pfn_to_page(start_pfn))) {
176 /* this region's page is not onlined now */
177 for (i = 0; i < nr_pages; i++) {
178 struct page *page = pfn_to_page(start_pfn + i);
184 res.start = res.end + 1;
185 res.end = section_end;
187 zone->present_pages += onlined_pages;
188 zone->zone_pgdat->node_present_pages += onlined_pages;
190 setup_per_zone_pages_min();
192 if (need_zonelists_rebuild)
193 build_all_zonelists();
194 vm_total_pages = nr_free_pagecache_pages();
198 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
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;
205 pgdat = arch_alloc_nodedata(nid);
209 arch_refresh_nodedata(nid, pgdat);
211 /* we can use NODE_DATA(nid) from here */
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);
219 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
221 arch_refresh_nodedata(nid, NULL);
222 arch_free_nodedata(pgdat);
226 /* add this memory to iomem resource */
227 static struct resource *register_memory_resource(u64 start, u64 size)
229 struct resource *res;
230 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
233 res->name = "System RAM";
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);
246 static void release_memory_resource(struct resource *res)
250 release_resource(res);
257 int add_memory(int nid, u64 start, u64 size)
259 pg_data_t *pgdat = NULL;
261 struct resource *res;
264 res = register_memory_resource(start, size);
268 if (!node_online(nid)) {
269 pgdat = hotadd_new_pgdat(nid, start);
273 ret = kswapd_run(nid);
278 /* call arch's memory hotadd */
279 ret = arch_add_memory(nid, start, size);
284 /* we online node here. we can't roll back from here. */
285 node_set_online(nid);
287 cpuset_track_online_nodes();
290 ret = register_one_node(nid);
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..
301 /* rollback pgdat allocation and others */
303 rollback_node_hotadd(nid, pgdat);
305 release_memory_resource(res);
309 EXPORT_SYMBOL_GPL(add_memory);