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/writeback.h>
17 #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>
24 #include <linux/ioport.h>
25 #include <linux/cpuset.h>
27 #include <asm/tlbflush.h>
29 /* add this memory to iomem resource */
30 static struct resource *register_memory_resource(u64 start, u64 size)
33 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
36 res->name = "System RAM";
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);
49 static void release_memory_resource(struct resource *res)
53 release_resource(res);
59 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
60 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
62 struct pglist_data *pgdat = zone->zone_pgdat;
63 int nr_pages = PAGES_PER_SECTION;
64 int nid = pgdat->node_id;
67 zone_type = zone - pgdat->node_zones;
68 if (!zone->wait_table) {
70 ret = init_currently_empty_zone(zone, phys_start_pfn,
71 nr_pages, MEMMAP_HOTPLUG);
75 memmap_init_zone(nr_pages, nid, zone_type,
76 phys_start_pfn, MEMMAP_HOTPLUG);
80 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
82 int nr_pages = PAGES_PER_SECTION;
85 if (pfn_valid(phys_start_pfn))
88 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
93 ret = __add_zone(zone, phys_start_pfn);
98 return register_new_memory(__pfn_to_section(phys_start_pfn));
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
107 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
108 unsigned long nr_pages)
112 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);
117 for (i = start_sec; i <= end_sec; i++) {
118 err = __add_section(zone, i << PFN_SECTION_SHIFT);
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.
125 if (err && (err != -EEXIST))
132 EXPORT_SYMBOL_GPL(__add_pages);
134 static void grow_zone_span(struct zone *zone,
135 unsigned long start_pfn, unsigned long end_pfn)
137 unsigned long old_zone_end_pfn;
139 zone_span_writelock(zone);
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;
145 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
146 zone->zone_start_pfn;
148 zone_span_writeunlock(zone);
151 static void grow_pgdat_span(struct pglist_data *pgdat,
152 unsigned long start_pfn, unsigned long end_pfn)
154 unsigned long old_pgdat_end_pfn =
155 pgdat->node_start_pfn + pgdat->node_spanned_pages;
157 if (start_pfn < pgdat->node_start_pfn)
158 pgdat->node_start_pfn = start_pfn;
160 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
161 pgdat->node_start_pfn;
164 int online_pages(unsigned long pfn, unsigned long nr_pages)
168 unsigned long onlined_pages = 0;
171 unsigned long start_pfn;
173 int need_zonelists_rebuild = 0;
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.
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);
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.
191 if (!populated_zone(zone))
192 need_zonelists_rebuild = 1;
194 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;
199 while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
200 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
201 nr_pages = (unsigned long)
202 ((res.end + 1 - res.start) >> PAGE_SHIFT);
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);
213 res.start = res.end + 1;
214 res.end = section_end;
216 zone->present_pages += onlined_pages;
217 zone->zone_pgdat->node_present_pages += onlined_pages;
219 setup_per_zone_pages_min();
221 kswapd_run(zone_to_nid(zone));
222 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
225 if (need_zonelists_rebuild)
226 build_all_zonelists();
227 vm_total_pages = nr_free_pagecache_pages();
228 writeback_set_ratelimit();
231 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
233 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
235 struct pglist_data *pgdat;
236 unsigned long zones_size[MAX_NR_ZONES] = {0};
237 unsigned long zholes_size[MAX_NR_ZONES] = {0};
238 unsigned long start_pfn = start >> PAGE_SHIFT;
240 pgdat = arch_alloc_nodedata(nid);
244 arch_refresh_nodedata(nid, pgdat);
246 /* we can use NODE_DATA(nid) from here */
248 /* init node's zones as empty zones, we don't have any present pages.*/
249 free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
254 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
256 arch_refresh_nodedata(nid, NULL);
257 arch_free_nodedata(pgdat);
262 int add_memory(int nid, u64 start, u64 size)
264 pg_data_t *pgdat = NULL;
266 struct resource *res;
269 res = register_memory_resource(start, size);
273 if (!node_online(nid)) {
274 pgdat = hotadd_new_pgdat(nid, start);
280 /* call arch's memory hotadd */
281 ret = arch_add_memory(nid, start, size);
286 /* we online node here. we can't roll back from here. */
287 node_set_online(nid);
289 cpuset_track_online_nodes();
292 ret = register_one_node(nid);
294 * If sysfs file of new node can't create, cpu on the node
295 * can't be hot-added. There is no rollback way now.
296 * So, check by BUG_ON() to catch it reluctantly..
303 /* rollback pgdat allocation and others */
305 rollback_node_hotadd(nid, pgdat);
307 release_memory_resource(res);
311 EXPORT_SYMBOL_GPL(add_memory);