| 1 | /* |
| 2 | * linux/mm/memory_hotplug.c |
| 3 | * |
| 4 | * Copyright (C) |
| 5 | */ |
| 6 | |
| 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> |
| 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> |
| 26 | |
| 27 | #include <asm/tlbflush.h> |
| 28 | |
| 29 | /* add this memory to iomem resource */ |
| 30 | static 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 | |
| 49 | static 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 | |
| 59 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
| 60 | static int __add_zone(struct zone *zone, unsigned long phys_start_pfn) |
| 61 | { |
| 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; |
| 68 | if (!populated_zone(zone)) { |
| 69 | int ret = 0; |
| 70 | ret = init_currently_empty_zone(zone, phys_start_pfn, |
| 71 | nr_pages, MEMMAP_HOTPLUG); |
| 72 | if (ret < 0) |
| 73 | return ret; |
| 74 | } |
| 75 | memmap_init_zone(nr_pages, nid, zone_type, |
| 76 | phys_start_pfn, MEMMAP_HOTPLUG); |
| 77 | return 0; |
| 78 | } |
| 79 | |
| 80 | static int __add_section(struct zone *zone, unsigned long phys_start_pfn) |
| 81 | { |
| 82 | int nr_pages = PAGES_PER_SECTION; |
| 83 | int ret; |
| 84 | |
| 85 | if (pfn_valid(phys_start_pfn)) |
| 86 | return -EEXIST; |
| 87 | |
| 88 | ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages); |
| 89 | |
| 90 | if (ret < 0) |
| 91 | return ret; |
| 92 | |
| 93 | ret = __add_zone(zone, phys_start_pfn); |
| 94 | |
| 95 | if (ret < 0) |
| 96 | return ret; |
| 97 | |
| 98 | 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 | */ |
| 107 | int __add_pages(struct zone *zone, unsigned long phys_start_pfn, |
| 108 | unsigned long nr_pages) |
| 109 | { |
| 110 | unsigned long i; |
| 111 | int err = 0; |
| 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); |
| 116 | |
| 117 | for (i = start_sec; i <= end_sec; i++) { |
| 118 | err = __add_section(zone, i << PFN_SECTION_SHIFT); |
| 119 | |
| 120 | /* |
| 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. |
| 124 | */ |
| 125 | if (err && (err != -EEXIST)) |
| 126 | break; |
| 127 | err = 0; |
| 128 | } |
| 129 | |
| 130 | return err; |
| 131 | } |
| 132 | EXPORT_SYMBOL_GPL(__add_pages); |
| 133 | |
| 134 | static 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 | |
| 145 | zone->spanned_pages = max(old_zone_end_pfn, end_pfn) - |
| 146 | zone->zone_start_pfn; |
| 147 | |
| 148 | zone_span_writeunlock(zone); |
| 149 | } |
| 150 | |
| 151 | static 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 | |
| 160 | pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) - |
| 161 | pgdat->node_start_pfn; |
| 162 | } |
| 163 | |
| 164 | int online_pages(unsigned long pfn, unsigned long nr_pages) |
| 165 | { |
| 166 | unsigned long i; |
| 167 | unsigned long flags; |
| 168 | unsigned long onlined_pages = 0; |
| 169 | struct resource res; |
| 170 | u64 section_end; |
| 171 | unsigned long start_pfn; |
| 172 | struct zone *zone; |
| 173 | int need_zonelists_rebuild = 0; |
| 174 | |
| 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 | |
| 186 | /* |
| 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 | |
| 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; |
| 198 | |
| 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); |
| 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; |
| 215 | } |
| 216 | zone->present_pages += onlined_pages; |
| 217 | zone->zone_pgdat->node_present_pages += onlined_pages; |
| 218 | |
| 219 | setup_per_zone_pages_min(); |
| 220 | |
| 221 | if (need_zonelists_rebuild) |
| 222 | build_all_zonelists(); |
| 223 | vm_total_pages = nr_free_pagecache_pages(); |
| 224 | writeback_set_ratelimit(); |
| 225 | return 0; |
| 226 | } |
| 227 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
| 228 | |
| 229 | static 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 | |
| 250 | static 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 | |
| 257 | |
| 258 | int add_memory(int nid, u64 start, u64 size) |
| 259 | { |
| 260 | pg_data_t *pgdat = NULL; |
| 261 | int new_pgdat = 0; |
| 262 | struct resource *res; |
| 263 | int ret; |
| 264 | |
| 265 | res = register_memory_resource(start, size); |
| 266 | if (!res) |
| 267 | return -EEXIST; |
| 268 | |
| 269 | 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 | |
| 279 | /* call arch's memory hotadd */ |
| 280 | ret = arch_add_memory(nid, start, size); |
| 281 | |
| 282 | if (ret < 0) |
| 283 | goto error; |
| 284 | |
| 285 | /* we online node here. we can't roll back from here. */ |
| 286 | node_set_online(nid); |
| 287 | |
| 288 | cpuset_track_online_nodes(); |
| 289 | |
| 290 | 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 | |
| 300 | return ret; |
| 301 | error: |
| 302 | /* rollback pgdat allocation and others */ |
| 303 | if (new_pgdat) |
| 304 | rollback_node_hotadd(nid, pgdat); |
| 305 | if (res) |
| 306 | release_memory_resource(res); |
| 307 | |
| 308 | return ret; |
| 309 | } |
| 310 | EXPORT_SYMBOL_GPL(add_memory); |