Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/memory_hotplug.c |
| 3 | * |
| 4 | * Copyright (C) |
| 5 | */ |
| 6 | |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 7 | #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 Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 16 | #include <linux/writeback.h> |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 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> |
KAMEZAWA Hiroyuki | 0a54703 | 2006-06-27 02:53:35 -0700 | [diff] [blame] | 24 | #include <linux/ioport.h> |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 25 | #include <linux/delay.h> |
| 26 | #include <linux/migrate.h> |
| 27 | #include <linux/page-isolation.h> |
Badari Pulavarty | 7108878 | 2008-10-18 20:25:58 -0700 | [diff] [blame] | 28 | #include <linux/pfn.h> |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 29 | #include <linux/suspend.h> |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 30 | #include <linux/mm_inline.h> |
akpm@linux-foundation.org | d96ae53 | 2010-03-05 13:41:58 -0800 | [diff] [blame] | 31 | #include <linux/firmware-map.h> |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 32 | |
| 33 | #include <asm/tlbflush.h> |
| 34 | |
Adrian Bunk | 1e5ad9a | 2008-04-28 20:40:08 +0300 | [diff] [blame] | 35 | #include "internal.h" |
| 36 | |
Keith Mannthey | 45e0b78 | 2006-09-30 23:27:09 -0700 | [diff] [blame] | 37 | /* add this memory to iomem resource */ |
| 38 | static struct resource *register_memory_resource(u64 start, u64 size) |
| 39 | { |
| 40 | struct resource *res; |
| 41 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); |
| 42 | BUG_ON(!res); |
| 43 | |
| 44 | res->name = "System RAM"; |
| 45 | res->start = start; |
| 46 | res->end = start + size - 1; |
Yasunori Goto | 887c3cb | 2007-11-14 16:59:20 -0800 | [diff] [blame] | 47 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
Keith Mannthey | 45e0b78 | 2006-09-30 23:27:09 -0700 | [diff] [blame] | 48 | if (request_resource(&iomem_resource, res) < 0) { |
| 49 | printk("System RAM resource %llx - %llx cannot be added\n", |
| 50 | (unsigned long long)res->start, (unsigned long long)res->end); |
| 51 | kfree(res); |
| 52 | res = NULL; |
| 53 | } |
| 54 | return res; |
| 55 | } |
| 56 | |
| 57 | static void release_memory_resource(struct resource *res) |
| 58 | { |
| 59 | if (!res) |
| 60 | return; |
| 61 | release_resource(res); |
| 62 | kfree(res); |
| 63 | return; |
| 64 | } |
| 65 | |
Keith Mannthey | 5394702 | 2006-09-30 23:27:08 -0700 | [diff] [blame] | 66 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 67 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
Yasunori Goto | af370fb | 2008-07-23 21:28:17 -0700 | [diff] [blame] | 68 | static void get_page_bootmem(unsigned long info, struct page *page, int type) |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 69 | { |
Yasunori Goto | af370fb | 2008-07-23 21:28:17 -0700 | [diff] [blame] | 70 | atomic_set(&page->_mapcount, type); |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 71 | SetPagePrivate(page); |
| 72 | set_page_private(page, info); |
| 73 | atomic_inc(&page->_count); |
| 74 | } |
| 75 | |
Rakib Mullick | 23ce932 | 2009-12-14 17:59:44 -0800 | [diff] [blame] | 76 | /* reference to __meminit __free_pages_bootmem is valid |
| 77 | * so use __ref to tell modpost not to generate a warning */ |
| 78 | void __ref put_page_bootmem(struct page *page) |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 79 | { |
Yasunori Goto | af370fb | 2008-07-23 21:28:17 -0700 | [diff] [blame] | 80 | int type; |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 81 | |
Yasunori Goto | af370fb | 2008-07-23 21:28:17 -0700 | [diff] [blame] | 82 | type = atomic_read(&page->_mapcount); |
| 83 | BUG_ON(type >= -1); |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 84 | |
| 85 | if (atomic_dec_return(&page->_count) == 1) { |
| 86 | ClearPagePrivate(page); |
| 87 | set_page_private(page, 0); |
| 88 | reset_page_mapcount(page); |
| 89 | __free_pages_bootmem(page, 0); |
| 90 | } |
| 91 | |
| 92 | } |
| 93 | |
Adrian Bunk | d92bc31 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 94 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 95 | { |
| 96 | unsigned long *usemap, mapsize, section_nr, i; |
| 97 | struct mem_section *ms; |
| 98 | struct page *page, *memmap; |
| 99 | |
| 100 | if (!pfn_valid(start_pfn)) |
| 101 | return; |
| 102 | |
| 103 | section_nr = pfn_to_section_nr(start_pfn); |
| 104 | ms = __nr_to_section(section_nr); |
| 105 | |
| 106 | /* Get section's memmap address */ |
| 107 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); |
| 108 | |
| 109 | /* |
| 110 | * Get page for the memmap's phys address |
| 111 | * XXX: need more consideration for sparse_vmemmap... |
| 112 | */ |
| 113 | page = virt_to_page(memmap); |
| 114 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; |
| 115 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; |
| 116 | |
| 117 | /* remember memmap's page */ |
| 118 | for (i = 0; i < mapsize; i++, page++) |
| 119 | get_page_bootmem(section_nr, page, SECTION_INFO); |
| 120 | |
| 121 | usemap = __nr_to_section(section_nr)->pageblock_flags; |
| 122 | page = virt_to_page(usemap); |
| 123 | |
| 124 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; |
| 125 | |
| 126 | for (i = 0; i < mapsize; i++, page++) |
Yasunori Goto | af370fb | 2008-07-23 21:28:17 -0700 | [diff] [blame] | 127 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
Yasunori Goto | 0475327 | 2008-04-28 02:13:31 -0700 | [diff] [blame] | 128 | |
| 129 | } |
| 130 | |
| 131 | void register_page_bootmem_info_node(struct pglist_data *pgdat) |
| 132 | { |
| 133 | unsigned long i, pfn, end_pfn, nr_pages; |
| 134 | int node = pgdat->node_id; |
| 135 | struct page *page; |
| 136 | struct zone *zone; |
| 137 | |
| 138 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; |
| 139 | page = virt_to_page(pgdat); |
| 140 | |
| 141 | for (i = 0; i < nr_pages; i++, page++) |
| 142 | get_page_bootmem(node, page, NODE_INFO); |
| 143 | |
| 144 | zone = &pgdat->node_zones[0]; |
| 145 | for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) { |
| 146 | if (zone->wait_table) { |
| 147 | nr_pages = zone->wait_table_hash_nr_entries |
| 148 | * sizeof(wait_queue_head_t); |
| 149 | nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT; |
| 150 | page = virt_to_page(zone->wait_table); |
| 151 | |
| 152 | for (i = 0; i < nr_pages; i++, page++) |
| 153 | get_page_bootmem(node, page, NODE_INFO); |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | pfn = pgdat->node_start_pfn; |
| 158 | end_pfn = pfn + pgdat->node_spanned_pages; |
| 159 | |
| 160 | /* register_section info */ |
| 161 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) |
| 162 | register_page_bootmem_info_section(pfn); |
| 163 | |
| 164 | } |
| 165 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ |
| 166 | |
Heiko Carstens | 76cdd58 | 2008-05-14 16:05:52 -0700 | [diff] [blame] | 167 | static void grow_zone_span(struct zone *zone, unsigned long start_pfn, |
| 168 | unsigned long end_pfn) |
| 169 | { |
| 170 | unsigned long old_zone_end_pfn; |
| 171 | |
| 172 | zone_span_writelock(zone); |
| 173 | |
| 174 | old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; |
| 175 | if (start_pfn < zone->zone_start_pfn) |
| 176 | zone->zone_start_pfn = start_pfn; |
| 177 | |
| 178 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - |
| 179 | zone->zone_start_pfn; |
| 180 | |
| 181 | zone_span_writeunlock(zone); |
| 182 | } |
| 183 | |
| 184 | static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn, |
| 185 | unsigned long end_pfn) |
| 186 | { |
| 187 | unsigned long old_pgdat_end_pfn = |
| 188 | pgdat->node_start_pfn + pgdat->node_spanned_pages; |
| 189 | |
| 190 | if (start_pfn < pgdat->node_start_pfn) |
| 191 | pgdat->node_start_pfn = start_pfn; |
| 192 | |
| 193 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - |
| 194 | pgdat->node_start_pfn; |
| 195 | } |
| 196 | |
Al Viro | 3116848 | 2008-11-22 17:33:24 +0000 | [diff] [blame] | 197 | static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 198 | { |
| 199 | struct pglist_data *pgdat = zone->zone_pgdat; |
| 200 | int nr_pages = PAGES_PER_SECTION; |
| 201 | int nid = pgdat->node_id; |
| 202 | int zone_type; |
Heiko Carstens | 76cdd58 | 2008-05-14 16:05:52 -0700 | [diff] [blame] | 203 | unsigned long flags; |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 204 | |
| 205 | zone_type = zone - pgdat->node_zones; |
Heiko Carstens | 76cdd58 | 2008-05-14 16:05:52 -0700 | [diff] [blame] | 206 | if (!zone->wait_table) { |
| 207 | int ret; |
| 208 | |
| 209 | ret = init_currently_empty_zone(zone, phys_start_pfn, |
| 210 | nr_pages, MEMMAP_HOTPLUG); |
| 211 | if (ret) |
| 212 | return ret; |
| 213 | } |
| 214 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
| 215 | grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages); |
| 216 | grow_pgdat_span(zone->zone_pgdat, phys_start_pfn, |
| 217 | phys_start_pfn + nr_pages); |
| 218 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 219 | memmap_init_zone(nr_pages, nid, zone_type, |
| 220 | phys_start_pfn, MEMMAP_HOTPLUG); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 221 | return 0; |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 222 | } |
| 223 | |
Gary Hade | c04fc58 | 2009-01-06 14:39:14 -0800 | [diff] [blame] | 224 | static int __meminit __add_section(int nid, struct zone *zone, |
| 225 | unsigned long phys_start_pfn) |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 226 | { |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 227 | int nr_pages = PAGES_PER_SECTION; |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 228 | int ret; |
| 229 | |
KAMEZAWA Hiroyuki | ebd1530 | 2006-08-05 12:15:06 -0700 | [diff] [blame] | 230 | if (pfn_valid(phys_start_pfn)) |
| 231 | return -EEXIST; |
| 232 | |
Dave Hansen | 0b0acbec | 2005-10-29 18:16:55 -0700 | [diff] [blame] | 233 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 234 | |
| 235 | if (ret < 0) |
| 236 | return ret; |
| 237 | |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 238 | ret = __add_zone(zone, phys_start_pfn); |
| 239 | |
| 240 | if (ret < 0) |
| 241 | return ret; |
| 242 | |
Gary Hade | c04fc58 | 2009-01-06 14:39:14 -0800 | [diff] [blame] | 243 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 244 | } |
| 245 | |
Yasunori Goto | 0c0a4a5 | 2008-04-28 02:13:34 -0700 | [diff] [blame] | 246 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
| 247 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
| 248 | { |
| 249 | /* |
| 250 | * XXX: Freeing memmap with vmemmap is not implement yet. |
| 251 | * This should be removed later. |
| 252 | */ |
| 253 | return -EBUSY; |
| 254 | } |
| 255 | #else |
Badari Pulavarty | ea01ea9 | 2008-04-28 02:12:01 -0700 | [diff] [blame] | 256 | static int __remove_section(struct zone *zone, struct mem_section *ms) |
| 257 | { |
| 258 | unsigned long flags; |
| 259 | struct pglist_data *pgdat = zone->zone_pgdat; |
| 260 | int ret = -EINVAL; |
| 261 | |
| 262 | if (!valid_section(ms)) |
| 263 | return ret; |
| 264 | |
| 265 | ret = unregister_memory_section(ms); |
| 266 | if (ret) |
| 267 | return ret; |
| 268 | |
| 269 | pgdat_resize_lock(pgdat, &flags); |
| 270 | sparse_remove_one_section(zone, ms); |
| 271 | pgdat_resize_unlock(pgdat, &flags); |
| 272 | return 0; |
| 273 | } |
Yasunori Goto | 0c0a4a5 | 2008-04-28 02:13:34 -0700 | [diff] [blame] | 274 | #endif |
Badari Pulavarty | ea01ea9 | 2008-04-28 02:12:01 -0700 | [diff] [blame] | 275 | |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 276 | /* |
| 277 | * Reasonably generic function for adding memory. It is |
| 278 | * expected that archs that support memory hotplug will |
| 279 | * call this function after deciding the zone to which to |
| 280 | * add the new pages. |
| 281 | */ |
Gary Hade | c04fc58 | 2009-01-06 14:39:14 -0800 | [diff] [blame] | 282 | int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn, |
| 283 | unsigned long nr_pages) |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 284 | { |
| 285 | unsigned long i; |
| 286 | int err = 0; |
KAMEZAWA Hiroyuki | 6f71271 | 2006-08-05 12:14:58 -0700 | [diff] [blame] | 287 | int start_sec, end_sec; |
| 288 | /* during initialize mem_map, align hot-added range to section */ |
| 289 | start_sec = pfn_to_section_nr(phys_start_pfn); |
| 290 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 291 | |
KAMEZAWA Hiroyuki | 6f71271 | 2006-08-05 12:14:58 -0700 | [diff] [blame] | 292 | for (i = start_sec; i <= end_sec; i++) { |
Gary Hade | c04fc58 | 2009-01-06 14:39:14 -0800 | [diff] [blame] | 293 | err = __add_section(nid, zone, i << PFN_SECTION_SHIFT); |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 294 | |
KAMEZAWA Hiroyuki | 6f71271 | 2006-08-05 12:14:58 -0700 | [diff] [blame] | 295 | /* |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 296 | * EEXIST is finally dealt with by ioresource collision |
KAMEZAWA Hiroyuki | 6f71271 | 2006-08-05 12:14:58 -0700 | [diff] [blame] | 297 | * check. see add_memory() => register_memory_resource() |
| 298 | * Warning will be printed if there is collision. |
Joel H Schopp | bed120c | 2006-05-01 12:16:11 -0700 | [diff] [blame] | 299 | */ |
| 300 | if (err && (err != -EEXIST)) |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 301 | break; |
KAMEZAWA Hiroyuki | 6f71271 | 2006-08-05 12:14:58 -0700 | [diff] [blame] | 302 | err = 0; |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 303 | } |
| 304 | |
| 305 | return err; |
| 306 | } |
Joel H Schopp | bed120c | 2006-05-01 12:16:11 -0700 | [diff] [blame] | 307 | EXPORT_SYMBOL_GPL(__add_pages); |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 308 | |
Badari Pulavarty | ea01ea9 | 2008-04-28 02:12:01 -0700 | [diff] [blame] | 309 | /** |
| 310 | * __remove_pages() - remove sections of pages from a zone |
| 311 | * @zone: zone from which pages need to be removed |
| 312 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) |
| 313 | * @nr_pages: number of pages to remove (must be multiple of section size) |
| 314 | * |
| 315 | * Generic helper function to remove section mappings and sysfs entries |
| 316 | * for the section of the memory we are removing. Caller needs to make |
| 317 | * sure that pages are marked reserved and zones are adjust properly by |
| 318 | * calling offline_pages(). |
| 319 | */ |
| 320 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, |
| 321 | unsigned long nr_pages) |
| 322 | { |
| 323 | unsigned long i, ret = 0; |
| 324 | int sections_to_remove; |
| 325 | |
| 326 | /* |
| 327 | * We can only remove entire sections |
| 328 | */ |
| 329 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); |
| 330 | BUG_ON(nr_pages % PAGES_PER_SECTION); |
| 331 | |
Badari Pulavarty | ea01ea9 | 2008-04-28 02:12:01 -0700 | [diff] [blame] | 332 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
| 333 | for (i = 0; i < sections_to_remove; i++) { |
| 334 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; |
Nathan Fontenot | de7f0cb | 2008-10-18 20:27:14 -0700 | [diff] [blame] | 335 | release_mem_region(pfn << PAGE_SHIFT, |
| 336 | PAGES_PER_SECTION << PAGE_SHIFT); |
Badari Pulavarty | ea01ea9 | 2008-04-28 02:12:01 -0700 | [diff] [blame] | 337 | ret = __remove_section(zone, __pfn_to_section(pfn)); |
| 338 | if (ret) |
| 339 | break; |
| 340 | } |
| 341 | return ret; |
| 342 | } |
| 343 | EXPORT_SYMBOL_GPL(__remove_pages); |
| 344 | |
Jeremy Fitzhardinge | 180c06e | 2008-04-28 02:12:03 -0700 | [diff] [blame] | 345 | void online_page(struct page *page) |
| 346 | { |
Jan Beulich | 4738e1b | 2009-09-21 17:03:03 -0700 | [diff] [blame] | 347 | unsigned long pfn = page_to_pfn(page); |
| 348 | |
Jeremy Fitzhardinge | 180c06e | 2008-04-28 02:12:03 -0700 | [diff] [blame] | 349 | totalram_pages++; |
Jan Beulich | 4738e1b | 2009-09-21 17:03:03 -0700 | [diff] [blame] | 350 | if (pfn >= num_physpages) |
| 351 | num_physpages = pfn + 1; |
Jeremy Fitzhardinge | 180c06e | 2008-04-28 02:12:03 -0700 | [diff] [blame] | 352 | |
| 353 | #ifdef CONFIG_HIGHMEM |
| 354 | if (PageHighMem(page)) |
| 355 | totalhigh_pages++; |
| 356 | #endif |
| 357 | |
| 358 | #ifdef CONFIG_FLATMEM |
| 359 | max_mapnr = max(page_to_pfn(page), max_mapnr); |
| 360 | #endif |
| 361 | |
| 362 | ClearPageReserved(page); |
| 363 | init_page_count(page); |
| 364 | __free_page(page); |
| 365 | } |
| 366 | |
KAMEZAWA Hiroyuki | 75884fb | 2007-10-16 01:26:10 -0700 | [diff] [blame] | 367 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
| 368 | void *arg) |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 369 | { |
| 370 | unsigned long i; |
KAMEZAWA Hiroyuki | 75884fb | 2007-10-16 01:26:10 -0700 | [diff] [blame] | 371 | unsigned long onlined_pages = *(unsigned long *)arg; |
| 372 | struct page *page; |
| 373 | if (PageReserved(pfn_to_page(start_pfn))) |
| 374 | for (i = 0; i < nr_pages; i++) { |
| 375 | page = pfn_to_page(start_pfn + i); |
| 376 | online_page(page); |
| 377 | onlined_pages++; |
| 378 | } |
| 379 | *(unsigned long *)arg = onlined_pages; |
| 380 | return 0; |
| 381 | } |
| 382 | |
| 383 | |
| 384 | int online_pages(unsigned long pfn, unsigned long nr_pages) |
| 385 | { |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 386 | unsigned long onlined_pages = 0; |
| 387 | struct zone *zone; |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 388 | int need_zonelists_rebuild = 0; |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 389 | int nid; |
| 390 | int ret; |
| 391 | struct memory_notify arg; |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 392 | |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 393 | arg.start_pfn = pfn; |
| 394 | arg.nr_pages = nr_pages; |
| 395 | arg.status_change_nid = -1; |
| 396 | |
| 397 | nid = page_to_nid(pfn_to_page(pfn)); |
| 398 | if (node_present_pages(nid) == 0) |
| 399 | arg.status_change_nid = nid; |
| 400 | |
| 401 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
| 402 | ret = notifier_to_errno(ret); |
| 403 | if (ret) { |
| 404 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
| 405 | return ret; |
| 406 | } |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 407 | /* |
| 408 | * This doesn't need a lock to do pfn_to_page(). |
| 409 | * The section can't be removed here because of the |
Daniel Walker | da19cbc | 2008-02-04 23:35:47 -0800 | [diff] [blame] | 410 | * memory_block->state_mutex. |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 411 | */ |
| 412 | zone = page_zone(pfn_to_page(pfn)); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 413 | /* |
| 414 | * If this zone is not populated, then it is not in zonelist. |
| 415 | * This means the page allocator ignores this zone. |
| 416 | * So, zonelist must be updated after online. |
| 417 | */ |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 418 | mutex_lock(&zonelists_mutex); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 419 | if (!populated_zone(zone)) |
| 420 | need_zonelists_rebuild = 1; |
| 421 | |
KAMEZAWA Hiroyuki | 908eedc | 2009-09-22 16:45:46 -0700 | [diff] [blame] | 422 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
KAMEZAWA Hiroyuki | 75884fb | 2007-10-16 01:26:10 -0700 | [diff] [blame] | 423 | online_pages_range); |
Geoff Levand | fd8a422 | 2008-05-14 16:05:50 -0700 | [diff] [blame] | 424 | if (ret) { |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 425 | mutex_unlock(&zonelists_mutex); |
Geoff Levand | fd8a422 | 2008-05-14 16:05:50 -0700 | [diff] [blame] | 426 | printk(KERN_DEBUG "online_pages %lx at %lx failed\n", |
| 427 | nr_pages, pfn); |
| 428 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
| 429 | return ret; |
| 430 | } |
| 431 | |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 432 | zone->present_pages += onlined_pages; |
Yasunori Goto | f2937be | 2006-03-09 17:33:51 -0800 | [diff] [blame] | 433 | zone->zone_pgdat->node_present_pages += onlined_pages; |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 434 | if (need_zonelists_rebuild) |
| 435 | build_all_zonelists(zone); |
| 436 | else |
| 437 | zone_pcp_update(zone); |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 438 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 439 | mutex_unlock(&zonelists_mutex); |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 440 | setup_per_zone_wmarks(); |
Minchan Kim | bce7394 | 2009-06-16 15:32:50 -0700 | [diff] [blame] | 441 | calculate_zone_inactive_ratio(zone); |
Christoph Lameter | 7ea1530 | 2007-10-16 01:25:29 -0700 | [diff] [blame] | 442 | if (onlined_pages) { |
| 443 | kswapd_run(zone_to_nid(zone)); |
| 444 | node_set_state(zone_to_nid(zone), N_HIGH_MEMORY); |
| 445 | } |
Dave Hansen | 61b1399 | 2005-10-29 18:16:56 -0700 | [diff] [blame] | 446 | |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 447 | vm_total_pages = nr_free_pagecache_pages(); |
Kent Liu | 2f7f24e | 2008-07-23 21:28:18 -0700 | [diff] [blame] | 448 | |
Chandra Seetharaman | 2d1d43f | 2006-09-29 02:01:25 -0700 | [diff] [blame] | 449 | writeback_set_ratelimit(); |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 450 | |
| 451 | if (onlined_pages) |
| 452 | memory_notify(MEM_ONLINE, &arg); |
| 453 | |
Dave Hansen | 3947be1 | 2005-10-29 18:16:54 -0700 | [diff] [blame] | 454 | return 0; |
| 455 | } |
Keith Mannthey | 5394702 | 2006-09-30 23:27:08 -0700 | [diff] [blame] | 456 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 457 | |
Hidetoshi Seto | e131933 | 2009-11-17 14:06:18 -0800 | [diff] [blame] | 458 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
| 459 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 460 | { |
| 461 | struct pglist_data *pgdat; |
| 462 | unsigned long zones_size[MAX_NR_ZONES] = {0}; |
| 463 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; |
| 464 | unsigned long start_pfn = start >> PAGE_SHIFT; |
| 465 | |
| 466 | pgdat = arch_alloc_nodedata(nid); |
| 467 | if (!pgdat) |
| 468 | return NULL; |
| 469 | |
| 470 | arch_refresh_nodedata(nid, pgdat); |
| 471 | |
| 472 | /* we can use NODE_DATA(nid) from here */ |
| 473 | |
| 474 | /* init node's zones as empty zones, we don't have any present pages.*/ |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 475 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 476 | |
| 477 | return pgdat; |
| 478 | } |
| 479 | |
| 480 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) |
| 481 | { |
| 482 | arch_refresh_nodedata(nid, NULL); |
| 483 | arch_free_nodedata(pgdat); |
| 484 | return; |
| 485 | } |
| 486 | |
KAMEZAWA Hiroyuki | 0a54703 | 2006-06-27 02:53:35 -0700 | [diff] [blame] | 487 | |
minskey guo | cf23422 | 2010-05-24 14:32:41 -0700 | [diff] [blame] | 488 | /* |
| 489 | * called by cpu_up() to online a node without onlined memory. |
| 490 | */ |
| 491 | int mem_online_node(int nid) |
| 492 | { |
| 493 | pg_data_t *pgdat; |
| 494 | int ret; |
| 495 | |
| 496 | lock_system_sleep(); |
| 497 | pgdat = hotadd_new_pgdat(nid, 0); |
| 498 | if (pgdat) { |
| 499 | ret = -ENOMEM; |
| 500 | goto out; |
| 501 | } |
| 502 | node_set_online(nid); |
| 503 | ret = register_one_node(nid); |
| 504 | BUG_ON(ret); |
| 505 | |
| 506 | out: |
| 507 | unlock_system_sleep(); |
| 508 | return ret; |
| 509 | } |
| 510 | |
Al Viro | 3116848 | 2008-11-22 17:33:24 +0000 | [diff] [blame] | 511 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
| 512 | int __ref add_memory(int nid, u64 start, u64 size) |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 513 | { |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 514 | pg_data_t *pgdat = NULL; |
| 515 | int new_pgdat = 0; |
KAMEZAWA Hiroyuki | ebd1530 | 2006-08-05 12:15:06 -0700 | [diff] [blame] | 516 | struct resource *res; |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 517 | int ret; |
| 518 | |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 519 | lock_system_sleep(); |
| 520 | |
KAMEZAWA Hiroyuki | ebd1530 | 2006-08-05 12:15:06 -0700 | [diff] [blame] | 521 | res = register_memory_resource(start, size); |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 522 | ret = -EEXIST; |
KAMEZAWA Hiroyuki | ebd1530 | 2006-08-05 12:15:06 -0700 | [diff] [blame] | 523 | if (!res) |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 524 | goto out; |
KAMEZAWA Hiroyuki | ebd1530 | 2006-08-05 12:15:06 -0700 | [diff] [blame] | 525 | |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 526 | if (!node_online(nid)) { |
| 527 | pgdat = hotadd_new_pgdat(nid, start); |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 528 | ret = -ENOMEM; |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 529 | if (!pgdat) |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 530 | goto out; |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 531 | new_pgdat = 1; |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 532 | } |
| 533 | |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 534 | /* call arch's memory hotadd */ |
| 535 | ret = arch_add_memory(nid, start, size); |
| 536 | |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 537 | if (ret < 0) |
| 538 | goto error; |
| 539 | |
Yasunori Goto | 0fc4415 | 2006-06-27 02:53:38 -0700 | [diff] [blame] | 540 | /* we online node here. we can't roll back from here. */ |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 541 | node_set_online(nid); |
| 542 | |
Yasunori Goto | 0fc4415 | 2006-06-27 02:53:38 -0700 | [diff] [blame] | 543 | if (new_pgdat) { |
| 544 | ret = register_one_node(nid); |
| 545 | /* |
| 546 | * If sysfs file of new node can't create, cpu on the node |
| 547 | * can't be hot-added. There is no rollback way now. |
| 548 | * So, check by BUG_ON() to catch it reluctantly.. |
| 549 | */ |
| 550 | BUG_ON(ret); |
| 551 | } |
| 552 | |
akpm@linux-foundation.org | d96ae53 | 2010-03-05 13:41:58 -0800 | [diff] [blame] | 553 | /* create new memmap entry */ |
| 554 | firmware_map_add_hotplug(start, start + size, "System RAM"); |
| 555 | |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 556 | goto out; |
| 557 | |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 558 | error: |
| 559 | /* rollback pgdat allocation and others */ |
| 560 | if (new_pgdat) |
| 561 | rollback_node_hotadd(nid, pgdat); |
KAMEZAWA Hiroyuki | ebd1530 | 2006-08-05 12:15:06 -0700 | [diff] [blame] | 562 | if (res) |
| 563 | release_memory_resource(res); |
Yasunori Goto | 9af3c2d | 2006-06-27 02:53:34 -0700 | [diff] [blame] | 564 | |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 565 | out: |
| 566 | unlock_system_sleep(); |
Yasunori Goto | bc02af9 | 2006-06-27 02:53:30 -0700 | [diff] [blame] | 567 | return ret; |
| 568 | } |
| 569 | EXPORT_SYMBOL_GPL(add_memory); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 570 | |
| 571 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 572 | /* |
Badari Pulavarty | 5c755e9 | 2008-07-23 21:28:19 -0700 | [diff] [blame] | 573 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy |
| 574 | * set and the size of the free page is given by page_order(). Using this, |
| 575 | * the function determines if the pageblock contains only free pages. |
| 576 | * Due to buddy contraints, a free page at least the size of a pageblock will |
| 577 | * be located at the start of the pageblock |
| 578 | */ |
| 579 | static inline int pageblock_free(struct page *page) |
| 580 | { |
| 581 | return PageBuddy(page) && page_order(page) >= pageblock_order; |
| 582 | } |
| 583 | |
| 584 | /* Return the start of the next active pageblock after a given page */ |
| 585 | static struct page *next_active_pageblock(struct page *page) |
| 586 | { |
| 587 | int pageblocks_stride; |
| 588 | |
| 589 | /* Ensure the starting page is pageblock-aligned */ |
| 590 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); |
| 591 | |
| 592 | /* Move forward by at least 1 * pageblock_nr_pages */ |
| 593 | pageblocks_stride = 1; |
| 594 | |
| 595 | /* If the entire pageblock is free, move to the end of free page */ |
| 596 | if (pageblock_free(page)) |
| 597 | pageblocks_stride += page_order(page) - pageblock_order; |
| 598 | |
| 599 | return page + (pageblocks_stride * pageblock_nr_pages); |
| 600 | } |
| 601 | |
| 602 | /* Checks if this range of memory is likely to be hot-removable. */ |
| 603 | int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
| 604 | { |
| 605 | int type; |
| 606 | struct page *page = pfn_to_page(start_pfn); |
| 607 | struct page *end_page = page + nr_pages; |
| 608 | |
| 609 | /* Check the starting page of each pageblock within the range */ |
| 610 | for (; page < end_page; page = next_active_pageblock(page)) { |
| 611 | type = get_pageblock_migratetype(page); |
| 612 | |
| 613 | /* |
| 614 | * A pageblock containing MOVABLE or free pages is considered |
| 615 | * removable |
| 616 | */ |
| 617 | if (type != MIGRATE_MOVABLE && !pageblock_free(page)) |
| 618 | return 0; |
| 619 | |
| 620 | /* |
| 621 | * A pageblock starting with a PageReserved page is not |
| 622 | * considered removable. |
| 623 | */ |
| 624 | if (PageReserved(page)) |
| 625 | return 0; |
| 626 | } |
| 627 | |
| 628 | /* All pageblocks in the memory block are likely to be hot-removable */ |
| 629 | return 1; |
| 630 | } |
| 631 | |
| 632 | /* |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 633 | * Confirm all pages in a range [start, end) is belongs to the same zone. |
| 634 | */ |
| 635 | static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn) |
| 636 | { |
| 637 | unsigned long pfn; |
| 638 | struct zone *zone = NULL; |
| 639 | struct page *page; |
| 640 | int i; |
| 641 | for (pfn = start_pfn; |
| 642 | pfn < end_pfn; |
| 643 | pfn += MAX_ORDER_NR_PAGES) { |
| 644 | i = 0; |
| 645 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ |
| 646 | while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i)) |
| 647 | i++; |
| 648 | if (i == MAX_ORDER_NR_PAGES) |
| 649 | continue; |
| 650 | page = pfn_to_page(pfn + i); |
| 651 | if (zone && page_zone(page) != zone) |
| 652 | return 0; |
| 653 | zone = page_zone(page); |
| 654 | } |
| 655 | return 1; |
| 656 | } |
| 657 | |
| 658 | /* |
| 659 | * Scanning pfn is much easier than scanning lru list. |
| 660 | * Scan pfn from start to end and Find LRU page. |
| 661 | */ |
| 662 | int scan_lru_pages(unsigned long start, unsigned long end) |
| 663 | { |
| 664 | unsigned long pfn; |
| 665 | struct page *page; |
| 666 | for (pfn = start; pfn < end; pfn++) { |
| 667 | if (pfn_valid(pfn)) { |
| 668 | page = pfn_to_page(pfn); |
| 669 | if (PageLRU(page)) |
| 670 | return pfn; |
| 671 | } |
| 672 | } |
| 673 | return 0; |
| 674 | } |
| 675 | |
| 676 | static struct page * |
Hugh Dickins | 3c1d437 | 2009-01-06 14:39:23 -0800 | [diff] [blame] | 677 | hotremove_migrate_alloc(struct page *page, unsigned long private, int **x) |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 678 | { |
Hugh Dickins | 3c1d437 | 2009-01-06 14:39:23 -0800 | [diff] [blame] | 679 | /* This should be improooooved!! */ |
| 680 | return alloc_page(GFP_HIGHUSER_MOVABLE); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 681 | } |
| 682 | |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 683 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
| 684 | static int |
| 685 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) |
| 686 | { |
| 687 | unsigned long pfn; |
| 688 | struct page *page; |
| 689 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; |
| 690 | int not_managed = 0; |
| 691 | int ret = 0; |
| 692 | LIST_HEAD(source); |
| 693 | |
| 694 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { |
| 695 | if (!pfn_valid(pfn)) |
| 696 | continue; |
| 697 | page = pfn_to_page(pfn); |
| 698 | if (!page_count(page)) |
| 699 | continue; |
| 700 | /* |
| 701 | * We can skip free pages. And we can only deal with pages on |
| 702 | * LRU. |
| 703 | */ |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 704 | ret = isolate_lru_page(page); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 705 | if (!ret) { /* Success */ |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 706 | list_add_tail(&page->lru, &source); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 707 | move_pages--; |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 708 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
| 709 | page_is_file_cache(page)); |
| 710 | |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 711 | } else { |
| 712 | /* Becasue we don't have big zone->lock. we should |
| 713 | check this again here. */ |
| 714 | if (page_count(page)) |
| 715 | not_managed++; |
| 716 | #ifdef CONFIG_DEBUG_VM |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 717 | printk(KERN_ALERT "removing pfn %lx from LRU failed\n", |
| 718 | pfn); |
| 719 | dump_page(page); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 720 | #endif |
| 721 | } |
| 722 | } |
| 723 | ret = -EBUSY; |
| 724 | if (not_managed) { |
| 725 | if (!list_empty(&source)) |
| 726 | putback_lru_pages(&source); |
| 727 | goto out; |
| 728 | } |
| 729 | ret = 0; |
| 730 | if (list_empty(&source)) |
| 731 | goto out; |
| 732 | /* this function returns # of failed pages */ |
Hugh Dickins | 62b61f6 | 2009-12-14 17:59:33 -0800 | [diff] [blame] | 733 | ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 734 | |
| 735 | out: |
| 736 | return ret; |
| 737 | } |
| 738 | |
| 739 | /* |
| 740 | * remove from free_area[] and mark all as Reserved. |
| 741 | */ |
| 742 | static int |
| 743 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, |
| 744 | void *data) |
| 745 | { |
| 746 | __offline_isolated_pages(start, start + nr_pages); |
| 747 | return 0; |
| 748 | } |
| 749 | |
| 750 | static void |
| 751 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
| 752 | { |
KAMEZAWA Hiroyuki | 908eedc | 2009-09-22 16:45:46 -0700 | [diff] [blame] | 753 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 754 | offline_isolated_pages_cb); |
| 755 | } |
| 756 | |
| 757 | /* |
| 758 | * Check all pages in range, recoreded as memory resource, are isolated. |
| 759 | */ |
| 760 | static int |
| 761 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, |
| 762 | void *data) |
| 763 | { |
| 764 | int ret; |
| 765 | long offlined = *(long *)data; |
| 766 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages); |
| 767 | offlined = nr_pages; |
| 768 | if (!ret) |
| 769 | *(long *)data += offlined; |
| 770 | return ret; |
| 771 | } |
| 772 | |
| 773 | static long |
| 774 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) |
| 775 | { |
| 776 | long offlined = 0; |
| 777 | int ret; |
| 778 | |
KAMEZAWA Hiroyuki | 908eedc | 2009-09-22 16:45:46 -0700 | [diff] [blame] | 779 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 780 | check_pages_isolated_cb); |
| 781 | if (ret < 0) |
| 782 | offlined = (long)ret; |
| 783 | return offlined; |
| 784 | } |
| 785 | |
Andrew Morton | b4e655a | 2009-12-14 17:59:35 -0800 | [diff] [blame] | 786 | static int offline_pages(unsigned long start_pfn, |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 787 | unsigned long end_pfn, unsigned long timeout) |
| 788 | { |
| 789 | unsigned long pfn, nr_pages, expire; |
| 790 | long offlined_pages; |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 791 | int ret, drain, retry_max, node; |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 792 | struct zone *zone; |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 793 | struct memory_notify arg; |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 794 | |
| 795 | BUG_ON(start_pfn >= end_pfn); |
| 796 | /* at least, alignment against pageblock is necessary */ |
| 797 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) |
| 798 | return -EINVAL; |
| 799 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) |
| 800 | return -EINVAL; |
| 801 | /* This makes hotplug much easier...and readable. |
| 802 | we assume this for now. .*/ |
| 803 | if (!test_pages_in_a_zone(start_pfn, end_pfn)) |
| 804 | return -EINVAL; |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 805 | |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 806 | lock_system_sleep(); |
| 807 | |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 808 | zone = page_zone(pfn_to_page(start_pfn)); |
| 809 | node = zone_to_nid(zone); |
| 810 | nr_pages = end_pfn - start_pfn; |
| 811 | |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 812 | /* set above range as isolated */ |
| 813 | ret = start_isolate_page_range(start_pfn, end_pfn); |
| 814 | if (ret) |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 815 | goto out; |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 816 | |
| 817 | arg.start_pfn = start_pfn; |
| 818 | arg.nr_pages = nr_pages; |
| 819 | arg.status_change_nid = -1; |
| 820 | if (nr_pages >= node_present_pages(node)) |
| 821 | arg.status_change_nid = node; |
| 822 | |
| 823 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); |
| 824 | ret = notifier_to_errno(ret); |
| 825 | if (ret) |
| 826 | goto failed_removal; |
| 827 | |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 828 | pfn = start_pfn; |
| 829 | expire = jiffies + timeout; |
| 830 | drain = 0; |
| 831 | retry_max = 5; |
| 832 | repeat: |
| 833 | /* start memory hot removal */ |
| 834 | ret = -EAGAIN; |
| 835 | if (time_after(jiffies, expire)) |
| 836 | goto failed_removal; |
| 837 | ret = -EINTR; |
| 838 | if (signal_pending(current)) |
| 839 | goto failed_removal; |
| 840 | ret = 0; |
| 841 | if (drain) { |
| 842 | lru_add_drain_all(); |
| 843 | flush_scheduled_work(); |
| 844 | cond_resched(); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 845 | drain_all_pages(); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 846 | } |
| 847 | |
| 848 | pfn = scan_lru_pages(start_pfn, end_pfn); |
| 849 | if (pfn) { /* We have page on LRU */ |
| 850 | ret = do_migrate_range(pfn, end_pfn); |
| 851 | if (!ret) { |
| 852 | drain = 1; |
| 853 | goto repeat; |
| 854 | } else { |
| 855 | if (ret < 0) |
| 856 | if (--retry_max == 0) |
| 857 | goto failed_removal; |
| 858 | yield(); |
| 859 | drain = 1; |
| 860 | goto repeat; |
| 861 | } |
| 862 | } |
| 863 | /* drain all zone's lru pagevec, this is asyncronous... */ |
| 864 | lru_add_drain_all(); |
| 865 | flush_scheduled_work(); |
| 866 | yield(); |
| 867 | /* drain pcp pages , this is synchrouns. */ |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 868 | drain_all_pages(); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 869 | /* check again */ |
| 870 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); |
| 871 | if (offlined_pages < 0) { |
| 872 | ret = -EBUSY; |
| 873 | goto failed_removal; |
| 874 | } |
| 875 | printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages); |
| 876 | /* Ok, all of our target is islaoted. |
| 877 | We cannot do rollback at this point. */ |
| 878 | offline_isolated_pages(start_pfn, end_pfn); |
KAMEZAWA Hiroyuki | dbc0e4c | 2007-11-14 16:59:12 -0800 | [diff] [blame] | 879 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
| 880 | undo_isolate_page_range(start_pfn, end_pfn); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 881 | /* removal success */ |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 882 | zone->present_pages -= offlined_pages; |
| 883 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
| 884 | totalram_pages -= offlined_pages; |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 885 | |
Minchan Kim | bce7394 | 2009-06-16 15:32:50 -0700 | [diff] [blame] | 886 | setup_per_zone_wmarks(); |
| 887 | calculate_zone_inactive_ratio(zone); |
David Rientjes | 8fe23e0 | 2009-12-14 17:58:33 -0800 | [diff] [blame] | 888 | if (!node_present_pages(node)) { |
| 889 | node_clear_state(node, N_HIGH_MEMORY); |
| 890 | kswapd_stop(node); |
| 891 | } |
Minchan Kim | bce7394 | 2009-06-16 15:32:50 -0700 | [diff] [blame] | 892 | |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 893 | vm_total_pages = nr_free_pagecache_pages(); |
| 894 | writeback_set_ratelimit(); |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 895 | |
| 896 | memory_notify(MEM_OFFLINE, &arg); |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 897 | unlock_system_sleep(); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 898 | return 0; |
| 899 | |
| 900 | failed_removal: |
| 901 | printk(KERN_INFO "memory offlining %lx to %lx failed\n", |
| 902 | start_pfn, end_pfn); |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 903 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 904 | /* pushback to free area */ |
| 905 | undo_isolate_page_range(start_pfn, end_pfn); |
Yasunori Goto | 7b78d33 | 2007-10-21 16:41:36 -0700 | [diff] [blame] | 906 | |
Andi Kleen | 6ad696d | 2009-11-17 14:06:22 -0800 | [diff] [blame] | 907 | out: |
| 908 | unlock_system_sleep(); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 909 | return ret; |
| 910 | } |
Badari Pulavarty | 7108878 | 2008-10-18 20:25:58 -0700 | [diff] [blame] | 911 | |
| 912 | int remove_memory(u64 start, u64 size) |
| 913 | { |
| 914 | unsigned long start_pfn, end_pfn; |
| 915 | |
| 916 | start_pfn = PFN_DOWN(start); |
| 917 | end_pfn = start_pfn + PFN_DOWN(size); |
| 918 | return offline_pages(start_pfn, end_pfn, 120 * HZ); |
| 919 | } |
KAMEZAWA Hiroyuki | 48e9419 | 2007-10-16 01:26:14 -0700 | [diff] [blame] | 920 | #else |
| 921 | int remove_memory(u64 start, u64 size) |
| 922 | { |
| 923 | return -EINVAL; |
| 924 | } |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 925 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
Badari Pulavarty | 7108878 | 2008-10-18 20:25:58 -0700 | [diff] [blame] | 926 | EXPORT_SYMBOL_GPL(remove_memory); |