| 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/sched/signal.h> |
| 10 | #include <linux/swap.h> |
| 11 | #include <linux/interrupt.h> |
| 12 | #include <linux/pagemap.h> |
| 13 | #include <linux/compiler.h> |
| 14 | #include <linux/export.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/memremap.h> |
| 22 | #include <linux/memory_hotplug.h> |
| 23 | #include <linux/highmem.h> |
| 24 | #include <linux/vmalloc.h> |
| 25 | #include <linux/ioport.h> |
| 26 | #include <linux/delay.h> |
| 27 | #include <linux/migrate.h> |
| 28 | #include <linux/page-isolation.h> |
| 29 | #include <linux/pfn.h> |
| 30 | #include <linux/suspend.h> |
| 31 | #include <linux/mm_inline.h> |
| 32 | #include <linux/firmware-map.h> |
| 33 | #include <linux/stop_machine.h> |
| 34 | #include <linux/hugetlb.h> |
| 35 | #include <linux/memblock.h> |
| 36 | #include <linux/bootmem.h> |
| 37 | #include <linux/compaction.h> |
| 38 | #include <linux/rmap.h> |
| 39 | |
| 40 | #include <asm/tlbflush.h> |
| 41 | |
| 42 | #include "internal.h" |
| 43 | |
| 44 | /* |
| 45 | * online_page_callback contains pointer to current page onlining function. |
| 46 | * Initially it is generic_online_page(). If it is required it could be |
| 47 | * changed by calling set_online_page_callback() for callback registration |
| 48 | * and restore_online_page_callback() for generic callback restore. |
| 49 | */ |
| 50 | |
| 51 | static void generic_online_page(struct page *page); |
| 52 | |
| 53 | static online_page_callback_t online_page_callback = generic_online_page; |
| 54 | static DEFINE_MUTEX(online_page_callback_lock); |
| 55 | |
| 56 | DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock); |
| 57 | |
| 58 | void get_online_mems(void) |
| 59 | { |
| 60 | percpu_down_read(&mem_hotplug_lock); |
| 61 | } |
| 62 | |
| 63 | void put_online_mems(void) |
| 64 | { |
| 65 | percpu_up_read(&mem_hotplug_lock); |
| 66 | } |
| 67 | |
| 68 | bool movable_node_enabled = false; |
| 69 | |
| 70 | #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE |
| 71 | bool memhp_auto_online; |
| 72 | #else |
| 73 | bool memhp_auto_online = true; |
| 74 | #endif |
| 75 | EXPORT_SYMBOL_GPL(memhp_auto_online); |
| 76 | |
| 77 | static int __init setup_memhp_default_state(char *str) |
| 78 | { |
| 79 | if (!strcmp(str, "online")) |
| 80 | memhp_auto_online = true; |
| 81 | else if (!strcmp(str, "offline")) |
| 82 | memhp_auto_online = false; |
| 83 | |
| 84 | return 1; |
| 85 | } |
| 86 | __setup("memhp_default_state=", setup_memhp_default_state); |
| 87 | |
| 88 | void mem_hotplug_begin(void) |
| 89 | { |
| 90 | cpus_read_lock(); |
| 91 | percpu_down_write(&mem_hotplug_lock); |
| 92 | } |
| 93 | |
| 94 | void mem_hotplug_done(void) |
| 95 | { |
| 96 | percpu_up_write(&mem_hotplug_lock); |
| 97 | cpus_read_unlock(); |
| 98 | } |
| 99 | |
| 100 | /* add this memory to iomem resource */ |
| 101 | static struct resource *register_memory_resource(u64 start, u64 size) |
| 102 | { |
| 103 | struct resource *res, *conflict; |
| 104 | res = kzalloc(sizeof(struct resource), GFP_KERNEL); |
| 105 | if (!res) |
| 106 | return ERR_PTR(-ENOMEM); |
| 107 | |
| 108 | res->name = "System RAM"; |
| 109 | res->start = start; |
| 110 | res->end = start + size - 1; |
| 111 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
| 112 | conflict = request_resource_conflict(&iomem_resource, res); |
| 113 | if (conflict) { |
| 114 | if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) { |
| 115 | pr_debug("Device unaddressable memory block " |
| 116 | "memory hotplug at %#010llx !\n", |
| 117 | (unsigned long long)start); |
| 118 | } |
| 119 | pr_debug("System RAM resource %pR cannot be added\n", res); |
| 120 | kfree(res); |
| 121 | return ERR_PTR(-EEXIST); |
| 122 | } |
| 123 | return res; |
| 124 | } |
| 125 | |
| 126 | static void release_memory_resource(struct resource *res) |
| 127 | { |
| 128 | if (!res) |
| 129 | return; |
| 130 | release_resource(res); |
| 131 | kfree(res); |
| 132 | return; |
| 133 | } |
| 134 | |
| 135 | #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE |
| 136 | void get_page_bootmem(unsigned long info, struct page *page, |
| 137 | unsigned long type) |
| 138 | { |
| 139 | page->freelist = (void *)type; |
| 140 | SetPagePrivate(page); |
| 141 | set_page_private(page, info); |
| 142 | page_ref_inc(page); |
| 143 | } |
| 144 | |
| 145 | void put_page_bootmem(struct page *page) |
| 146 | { |
| 147 | unsigned long type; |
| 148 | |
| 149 | type = (unsigned long) page->freelist; |
| 150 | BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE || |
| 151 | type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE); |
| 152 | |
| 153 | if (page_ref_dec_return(page) == 1) { |
| 154 | page->freelist = NULL; |
| 155 | ClearPagePrivate(page); |
| 156 | set_page_private(page, 0); |
| 157 | INIT_LIST_HEAD(&page->lru); |
| 158 | free_reserved_page(page); |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE |
| 163 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
| 164 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
| 165 | { |
| 166 | unsigned long *usemap, mapsize, section_nr, i; |
| 167 | struct mem_section *ms; |
| 168 | struct page *page, *memmap; |
| 169 | |
| 170 | section_nr = pfn_to_section_nr(start_pfn); |
| 171 | ms = __nr_to_section(section_nr); |
| 172 | |
| 173 | /* Get section's memmap address */ |
| 174 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); |
| 175 | |
| 176 | /* |
| 177 | * Get page for the memmap's phys address |
| 178 | * XXX: need more consideration for sparse_vmemmap... |
| 179 | */ |
| 180 | page = virt_to_page(memmap); |
| 181 | mapsize = sizeof(struct page) * PAGES_PER_SECTION; |
| 182 | mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT; |
| 183 | |
| 184 | /* remember memmap's page */ |
| 185 | for (i = 0; i < mapsize; i++, page++) |
| 186 | get_page_bootmem(section_nr, page, SECTION_INFO); |
| 187 | |
| 188 | usemap = __nr_to_section(section_nr)->pageblock_flags; |
| 189 | page = virt_to_page(usemap); |
| 190 | |
| 191 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; |
| 192 | |
| 193 | for (i = 0; i < mapsize; i++, page++) |
| 194 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
| 195 | |
| 196 | } |
| 197 | #else /* CONFIG_SPARSEMEM_VMEMMAP */ |
| 198 | static void register_page_bootmem_info_section(unsigned long start_pfn) |
| 199 | { |
| 200 | unsigned long *usemap, mapsize, section_nr, i; |
| 201 | struct mem_section *ms; |
| 202 | struct page *page, *memmap; |
| 203 | |
| 204 | if (!pfn_valid(start_pfn)) |
| 205 | return; |
| 206 | |
| 207 | section_nr = pfn_to_section_nr(start_pfn); |
| 208 | ms = __nr_to_section(section_nr); |
| 209 | |
| 210 | memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr); |
| 211 | |
| 212 | register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION); |
| 213 | |
| 214 | usemap = __nr_to_section(section_nr)->pageblock_flags; |
| 215 | page = virt_to_page(usemap); |
| 216 | |
| 217 | mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT; |
| 218 | |
| 219 | for (i = 0; i < mapsize; i++, page++) |
| 220 | get_page_bootmem(section_nr, page, MIX_SECTION_INFO); |
| 221 | } |
| 222 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ |
| 223 | |
| 224 | void __init register_page_bootmem_info_node(struct pglist_data *pgdat) |
| 225 | { |
| 226 | unsigned long i, pfn, end_pfn, nr_pages; |
| 227 | int node = pgdat->node_id; |
| 228 | struct page *page; |
| 229 | |
| 230 | nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT; |
| 231 | page = virt_to_page(pgdat); |
| 232 | |
| 233 | for (i = 0; i < nr_pages; i++, page++) |
| 234 | get_page_bootmem(node, page, NODE_INFO); |
| 235 | |
| 236 | pfn = pgdat->node_start_pfn; |
| 237 | end_pfn = pgdat_end_pfn(pgdat); |
| 238 | |
| 239 | /* register section info */ |
| 240 | for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
| 241 | /* |
| 242 | * Some platforms can assign the same pfn to multiple nodes - on |
| 243 | * node0 as well as nodeN. To avoid registering a pfn against |
| 244 | * multiple nodes we check that this pfn does not already |
| 245 | * reside in some other nodes. |
| 246 | */ |
| 247 | if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node)) |
| 248 | register_page_bootmem_info_section(pfn); |
| 249 | } |
| 250 | } |
| 251 | #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */ |
| 252 | |
| 253 | static int __meminit __add_section(int nid, unsigned long phys_start_pfn, |
| 254 | bool want_memblock) |
| 255 | { |
| 256 | int ret; |
| 257 | int i; |
| 258 | |
| 259 | if (pfn_valid(phys_start_pfn)) |
| 260 | return -EEXIST; |
| 261 | |
| 262 | ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn); |
| 263 | if (ret < 0) |
| 264 | return ret; |
| 265 | |
| 266 | /* |
| 267 | * Make all the pages reserved so that nobody will stumble over half |
| 268 | * initialized state. |
| 269 | * FIXME: We also have to associate it with a node because pfn_to_node |
| 270 | * relies on having page with the proper node. |
| 271 | */ |
| 272 | for (i = 0; i < PAGES_PER_SECTION; i++) { |
| 273 | unsigned long pfn = phys_start_pfn + i; |
| 274 | struct page *page; |
| 275 | if (!pfn_valid(pfn)) |
| 276 | continue; |
| 277 | |
| 278 | page = pfn_to_page(pfn); |
| 279 | set_page_node(page, nid); |
| 280 | SetPageReserved(page); |
| 281 | } |
| 282 | |
| 283 | if (!want_memblock) |
| 284 | return 0; |
| 285 | |
| 286 | return register_new_memory(nid, __pfn_to_section(phys_start_pfn)); |
| 287 | } |
| 288 | |
| 289 | /* |
| 290 | * Reasonably generic function for adding memory. It is |
| 291 | * expected that archs that support memory hotplug will |
| 292 | * call this function after deciding the zone to which to |
| 293 | * add the new pages. |
| 294 | */ |
| 295 | int __ref __add_pages(int nid, unsigned long phys_start_pfn, |
| 296 | unsigned long nr_pages, bool want_memblock) |
| 297 | { |
| 298 | unsigned long i; |
| 299 | int err = 0; |
| 300 | int start_sec, end_sec; |
| 301 | struct vmem_altmap *altmap; |
| 302 | |
| 303 | /* during initialize mem_map, align hot-added range to section */ |
| 304 | start_sec = pfn_to_section_nr(phys_start_pfn); |
| 305 | end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1); |
| 306 | |
| 307 | altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn)); |
| 308 | if (altmap) { |
| 309 | /* |
| 310 | * Validate altmap is within bounds of the total request |
| 311 | */ |
| 312 | if (altmap->base_pfn != phys_start_pfn |
| 313 | || vmem_altmap_offset(altmap) > nr_pages) { |
| 314 | pr_warn_once("memory add fail, invalid altmap\n"); |
| 315 | err = -EINVAL; |
| 316 | goto out; |
| 317 | } |
| 318 | altmap->alloc = 0; |
| 319 | } |
| 320 | |
| 321 | for (i = start_sec; i <= end_sec; i++) { |
| 322 | err = __add_section(nid, section_nr_to_pfn(i), want_memblock); |
| 323 | |
| 324 | /* |
| 325 | * EEXIST is finally dealt with by ioresource collision |
| 326 | * check. see add_memory() => register_memory_resource() |
| 327 | * Warning will be printed if there is collision. |
| 328 | */ |
| 329 | if (err && (err != -EEXIST)) |
| 330 | break; |
| 331 | err = 0; |
| 332 | cond_resched(); |
| 333 | } |
| 334 | vmemmap_populate_print_last(); |
| 335 | out: |
| 336 | return err; |
| 337 | } |
| 338 | EXPORT_SYMBOL_GPL(__add_pages); |
| 339 | |
| 340 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 341 | /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ |
| 342 | static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, |
| 343 | unsigned long start_pfn, |
| 344 | unsigned long end_pfn) |
| 345 | { |
| 346 | struct mem_section *ms; |
| 347 | |
| 348 | for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) { |
| 349 | ms = __pfn_to_section(start_pfn); |
| 350 | |
| 351 | if (unlikely(!valid_section(ms))) |
| 352 | continue; |
| 353 | |
| 354 | if (unlikely(pfn_to_nid(start_pfn) != nid)) |
| 355 | continue; |
| 356 | |
| 357 | if (zone && zone != page_zone(pfn_to_page(start_pfn))) |
| 358 | continue; |
| 359 | |
| 360 | return start_pfn; |
| 361 | } |
| 362 | |
| 363 | return 0; |
| 364 | } |
| 365 | |
| 366 | /* find the biggest valid pfn in the range [start_pfn, end_pfn). */ |
| 367 | static unsigned long find_biggest_section_pfn(int nid, struct zone *zone, |
| 368 | unsigned long start_pfn, |
| 369 | unsigned long end_pfn) |
| 370 | { |
| 371 | struct mem_section *ms; |
| 372 | unsigned long pfn; |
| 373 | |
| 374 | /* pfn is the end pfn of a memory section. */ |
| 375 | pfn = end_pfn - 1; |
| 376 | for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) { |
| 377 | ms = __pfn_to_section(pfn); |
| 378 | |
| 379 | if (unlikely(!valid_section(ms))) |
| 380 | continue; |
| 381 | |
| 382 | if (unlikely(pfn_to_nid(pfn) != nid)) |
| 383 | continue; |
| 384 | |
| 385 | if (zone && zone != page_zone(pfn_to_page(pfn))) |
| 386 | continue; |
| 387 | |
| 388 | return pfn; |
| 389 | } |
| 390 | |
| 391 | return 0; |
| 392 | } |
| 393 | |
| 394 | static void shrink_zone_span(struct zone *zone, unsigned long start_pfn, |
| 395 | unsigned long end_pfn) |
| 396 | { |
| 397 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
| 398 | unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */ |
| 399 | unsigned long zone_end_pfn = z; |
| 400 | unsigned long pfn; |
| 401 | struct mem_section *ms; |
| 402 | int nid = zone_to_nid(zone); |
| 403 | |
| 404 | zone_span_writelock(zone); |
| 405 | if (zone_start_pfn == start_pfn) { |
| 406 | /* |
| 407 | * If the section is smallest section in the zone, it need |
| 408 | * shrink zone->zone_start_pfn and zone->zone_spanned_pages. |
| 409 | * In this case, we find second smallest valid mem_section |
| 410 | * for shrinking zone. |
| 411 | */ |
| 412 | pfn = find_smallest_section_pfn(nid, zone, end_pfn, |
| 413 | zone_end_pfn); |
| 414 | if (pfn) { |
| 415 | zone->zone_start_pfn = pfn; |
| 416 | zone->spanned_pages = zone_end_pfn - pfn; |
| 417 | } |
| 418 | } else if (zone_end_pfn == end_pfn) { |
| 419 | /* |
| 420 | * If the section is biggest section in the zone, it need |
| 421 | * shrink zone->spanned_pages. |
| 422 | * In this case, we find second biggest valid mem_section for |
| 423 | * shrinking zone. |
| 424 | */ |
| 425 | pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn, |
| 426 | start_pfn); |
| 427 | if (pfn) |
| 428 | zone->spanned_pages = pfn - zone_start_pfn + 1; |
| 429 | } |
| 430 | |
| 431 | /* |
| 432 | * The section is not biggest or smallest mem_section in the zone, it |
| 433 | * only creates a hole in the zone. So in this case, we need not |
| 434 | * change the zone. But perhaps, the zone has only hole data. Thus |
| 435 | * it check the zone has only hole or not. |
| 436 | */ |
| 437 | pfn = zone_start_pfn; |
| 438 | for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) { |
| 439 | ms = __pfn_to_section(pfn); |
| 440 | |
| 441 | if (unlikely(!valid_section(ms))) |
| 442 | continue; |
| 443 | |
| 444 | if (page_zone(pfn_to_page(pfn)) != zone) |
| 445 | continue; |
| 446 | |
| 447 | /* If the section is current section, it continues the loop */ |
| 448 | if (start_pfn == pfn) |
| 449 | continue; |
| 450 | |
| 451 | /* If we find valid section, we have nothing to do */ |
| 452 | zone_span_writeunlock(zone); |
| 453 | return; |
| 454 | } |
| 455 | |
| 456 | /* The zone has no valid section */ |
| 457 | zone->zone_start_pfn = 0; |
| 458 | zone->spanned_pages = 0; |
| 459 | zone_span_writeunlock(zone); |
| 460 | } |
| 461 | |
| 462 | static void shrink_pgdat_span(struct pglist_data *pgdat, |
| 463 | unsigned long start_pfn, unsigned long end_pfn) |
| 464 | { |
| 465 | unsigned long pgdat_start_pfn = pgdat->node_start_pfn; |
| 466 | unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */ |
| 467 | unsigned long pgdat_end_pfn = p; |
| 468 | unsigned long pfn; |
| 469 | struct mem_section *ms; |
| 470 | int nid = pgdat->node_id; |
| 471 | |
| 472 | if (pgdat_start_pfn == start_pfn) { |
| 473 | /* |
| 474 | * If the section is smallest section in the pgdat, it need |
| 475 | * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages. |
| 476 | * In this case, we find second smallest valid mem_section |
| 477 | * for shrinking zone. |
| 478 | */ |
| 479 | pfn = find_smallest_section_pfn(nid, NULL, end_pfn, |
| 480 | pgdat_end_pfn); |
| 481 | if (pfn) { |
| 482 | pgdat->node_start_pfn = pfn; |
| 483 | pgdat->node_spanned_pages = pgdat_end_pfn - pfn; |
| 484 | } |
| 485 | } else if (pgdat_end_pfn == end_pfn) { |
| 486 | /* |
| 487 | * If the section is biggest section in the pgdat, it need |
| 488 | * shrink pgdat->node_spanned_pages. |
| 489 | * In this case, we find second biggest valid mem_section for |
| 490 | * shrinking zone. |
| 491 | */ |
| 492 | pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn, |
| 493 | start_pfn); |
| 494 | if (pfn) |
| 495 | pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1; |
| 496 | } |
| 497 | |
| 498 | /* |
| 499 | * If the section is not biggest or smallest mem_section in the pgdat, |
| 500 | * it only creates a hole in the pgdat. So in this case, we need not |
| 501 | * change the pgdat. |
| 502 | * But perhaps, the pgdat has only hole data. Thus it check the pgdat |
| 503 | * has only hole or not. |
| 504 | */ |
| 505 | pfn = pgdat_start_pfn; |
| 506 | for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) { |
| 507 | ms = __pfn_to_section(pfn); |
| 508 | |
| 509 | if (unlikely(!valid_section(ms))) |
| 510 | continue; |
| 511 | |
| 512 | if (pfn_to_nid(pfn) != nid) |
| 513 | continue; |
| 514 | |
| 515 | /* If the section is current section, it continues the loop */ |
| 516 | if (start_pfn == pfn) |
| 517 | continue; |
| 518 | |
| 519 | /* If we find valid section, we have nothing to do */ |
| 520 | return; |
| 521 | } |
| 522 | |
| 523 | /* The pgdat has no valid section */ |
| 524 | pgdat->node_start_pfn = 0; |
| 525 | pgdat->node_spanned_pages = 0; |
| 526 | } |
| 527 | |
| 528 | static void __remove_zone(struct zone *zone, unsigned long start_pfn) |
| 529 | { |
| 530 | struct pglist_data *pgdat = zone->zone_pgdat; |
| 531 | int nr_pages = PAGES_PER_SECTION; |
| 532 | unsigned long flags; |
| 533 | |
| 534 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
| 535 | shrink_zone_span(zone, start_pfn, start_pfn + nr_pages); |
| 536 | shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages); |
| 537 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
| 538 | } |
| 539 | |
| 540 | static int __remove_section(struct zone *zone, struct mem_section *ms, |
| 541 | unsigned long map_offset) |
| 542 | { |
| 543 | unsigned long start_pfn; |
| 544 | int scn_nr; |
| 545 | int ret = -EINVAL; |
| 546 | |
| 547 | if (!valid_section(ms)) |
| 548 | return ret; |
| 549 | |
| 550 | ret = unregister_memory_section(ms); |
| 551 | if (ret) |
| 552 | return ret; |
| 553 | |
| 554 | scn_nr = __section_nr(ms); |
| 555 | start_pfn = section_nr_to_pfn((unsigned long)scn_nr); |
| 556 | __remove_zone(zone, start_pfn); |
| 557 | |
| 558 | sparse_remove_one_section(zone, ms, map_offset); |
| 559 | return 0; |
| 560 | } |
| 561 | |
| 562 | /** |
| 563 | * __remove_pages() - remove sections of pages from a zone |
| 564 | * @zone: zone from which pages need to be removed |
| 565 | * @phys_start_pfn: starting pageframe (must be aligned to start of a section) |
| 566 | * @nr_pages: number of pages to remove (must be multiple of section size) |
| 567 | * |
| 568 | * Generic helper function to remove section mappings and sysfs entries |
| 569 | * for the section of the memory we are removing. Caller needs to make |
| 570 | * sure that pages are marked reserved and zones are adjust properly by |
| 571 | * calling offline_pages(). |
| 572 | */ |
| 573 | int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, |
| 574 | unsigned long nr_pages) |
| 575 | { |
| 576 | unsigned long i; |
| 577 | unsigned long map_offset = 0; |
| 578 | int sections_to_remove, ret = 0; |
| 579 | |
| 580 | /* In the ZONE_DEVICE case device driver owns the memory region */ |
| 581 | if (is_dev_zone(zone)) { |
| 582 | struct page *page = pfn_to_page(phys_start_pfn); |
| 583 | struct vmem_altmap *altmap; |
| 584 | |
| 585 | altmap = to_vmem_altmap((unsigned long) page); |
| 586 | if (altmap) |
| 587 | map_offset = vmem_altmap_offset(altmap); |
| 588 | } else { |
| 589 | resource_size_t start, size; |
| 590 | |
| 591 | start = phys_start_pfn << PAGE_SHIFT; |
| 592 | size = nr_pages * PAGE_SIZE; |
| 593 | |
| 594 | ret = release_mem_region_adjustable(&iomem_resource, start, |
| 595 | size); |
| 596 | if (ret) { |
| 597 | resource_size_t endres = start + size - 1; |
| 598 | |
| 599 | pr_warn("Unable to release resource <%pa-%pa> (%d)\n", |
| 600 | &start, &endres, ret); |
| 601 | } |
| 602 | } |
| 603 | |
| 604 | clear_zone_contiguous(zone); |
| 605 | |
| 606 | /* |
| 607 | * We can only remove entire sections |
| 608 | */ |
| 609 | BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK); |
| 610 | BUG_ON(nr_pages % PAGES_PER_SECTION); |
| 611 | |
| 612 | sections_to_remove = nr_pages / PAGES_PER_SECTION; |
| 613 | for (i = 0; i < sections_to_remove; i++) { |
| 614 | unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; |
| 615 | |
| 616 | ret = __remove_section(zone, __pfn_to_section(pfn), map_offset); |
| 617 | map_offset = 0; |
| 618 | if (ret) |
| 619 | break; |
| 620 | } |
| 621 | |
| 622 | set_zone_contiguous(zone); |
| 623 | |
| 624 | return ret; |
| 625 | } |
| 626 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
| 627 | |
| 628 | int set_online_page_callback(online_page_callback_t callback) |
| 629 | { |
| 630 | int rc = -EINVAL; |
| 631 | |
| 632 | get_online_mems(); |
| 633 | mutex_lock(&online_page_callback_lock); |
| 634 | |
| 635 | if (online_page_callback == generic_online_page) { |
| 636 | online_page_callback = callback; |
| 637 | rc = 0; |
| 638 | } |
| 639 | |
| 640 | mutex_unlock(&online_page_callback_lock); |
| 641 | put_online_mems(); |
| 642 | |
| 643 | return rc; |
| 644 | } |
| 645 | EXPORT_SYMBOL_GPL(set_online_page_callback); |
| 646 | |
| 647 | int restore_online_page_callback(online_page_callback_t callback) |
| 648 | { |
| 649 | int rc = -EINVAL; |
| 650 | |
| 651 | get_online_mems(); |
| 652 | mutex_lock(&online_page_callback_lock); |
| 653 | |
| 654 | if (online_page_callback == callback) { |
| 655 | online_page_callback = generic_online_page; |
| 656 | rc = 0; |
| 657 | } |
| 658 | |
| 659 | mutex_unlock(&online_page_callback_lock); |
| 660 | put_online_mems(); |
| 661 | |
| 662 | return rc; |
| 663 | } |
| 664 | EXPORT_SYMBOL_GPL(restore_online_page_callback); |
| 665 | |
| 666 | void __online_page_set_limits(struct page *page) |
| 667 | { |
| 668 | } |
| 669 | EXPORT_SYMBOL_GPL(__online_page_set_limits); |
| 670 | |
| 671 | void __online_page_increment_counters(struct page *page) |
| 672 | { |
| 673 | adjust_managed_page_count(page, 1); |
| 674 | } |
| 675 | EXPORT_SYMBOL_GPL(__online_page_increment_counters); |
| 676 | |
| 677 | void __online_page_free(struct page *page) |
| 678 | { |
| 679 | __free_reserved_page(page); |
| 680 | } |
| 681 | EXPORT_SYMBOL_GPL(__online_page_free); |
| 682 | |
| 683 | static void generic_online_page(struct page *page) |
| 684 | { |
| 685 | __online_page_set_limits(page); |
| 686 | __online_page_increment_counters(page); |
| 687 | __online_page_free(page); |
| 688 | } |
| 689 | |
| 690 | static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages, |
| 691 | void *arg) |
| 692 | { |
| 693 | unsigned long i; |
| 694 | unsigned long onlined_pages = *(unsigned long *)arg; |
| 695 | struct page *page; |
| 696 | |
| 697 | if (PageReserved(pfn_to_page(start_pfn))) |
| 698 | for (i = 0; i < nr_pages; i++) { |
| 699 | page = pfn_to_page(start_pfn + i); |
| 700 | (*online_page_callback)(page); |
| 701 | onlined_pages++; |
| 702 | } |
| 703 | |
| 704 | online_mem_sections(start_pfn, start_pfn + nr_pages); |
| 705 | |
| 706 | *(unsigned long *)arg = onlined_pages; |
| 707 | return 0; |
| 708 | } |
| 709 | |
| 710 | /* check which state of node_states will be changed when online memory */ |
| 711 | static void node_states_check_changes_online(unsigned long nr_pages, |
| 712 | struct zone *zone, struct memory_notify *arg) |
| 713 | { |
| 714 | int nid = zone_to_nid(zone); |
| 715 | enum zone_type zone_last = ZONE_NORMAL; |
| 716 | |
| 717 | /* |
| 718 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
| 719 | * contains nodes which have zones of 0...ZONE_NORMAL, |
| 720 | * set zone_last to ZONE_NORMAL. |
| 721 | * |
| 722 | * If we don't have HIGHMEM nor movable node, |
| 723 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of |
| 724 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. |
| 725 | */ |
| 726 | if (N_MEMORY == N_NORMAL_MEMORY) |
| 727 | zone_last = ZONE_MOVABLE; |
| 728 | |
| 729 | /* |
| 730 | * if the memory to be online is in a zone of 0...zone_last, and |
| 731 | * the zones of 0...zone_last don't have memory before online, we will |
| 732 | * need to set the node to node_states[N_NORMAL_MEMORY] after |
| 733 | * the memory is online. |
| 734 | */ |
| 735 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY)) |
| 736 | arg->status_change_nid_normal = nid; |
| 737 | else |
| 738 | arg->status_change_nid_normal = -1; |
| 739 | |
| 740 | #ifdef CONFIG_HIGHMEM |
| 741 | /* |
| 742 | * If we have movable node, node_states[N_HIGH_MEMORY] |
| 743 | * contains nodes which have zones of 0...ZONE_HIGHMEM, |
| 744 | * set zone_last to ZONE_HIGHMEM. |
| 745 | * |
| 746 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] |
| 747 | * contains nodes which have zones of 0...ZONE_MOVABLE, |
| 748 | * set zone_last to ZONE_MOVABLE. |
| 749 | */ |
| 750 | zone_last = ZONE_HIGHMEM; |
| 751 | if (N_MEMORY == N_HIGH_MEMORY) |
| 752 | zone_last = ZONE_MOVABLE; |
| 753 | |
| 754 | if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY)) |
| 755 | arg->status_change_nid_high = nid; |
| 756 | else |
| 757 | arg->status_change_nid_high = -1; |
| 758 | #else |
| 759 | arg->status_change_nid_high = arg->status_change_nid_normal; |
| 760 | #endif |
| 761 | |
| 762 | /* |
| 763 | * if the node don't have memory befor online, we will need to |
| 764 | * set the node to node_states[N_MEMORY] after the memory |
| 765 | * is online. |
| 766 | */ |
| 767 | if (!node_state(nid, N_MEMORY)) |
| 768 | arg->status_change_nid = nid; |
| 769 | else |
| 770 | arg->status_change_nid = -1; |
| 771 | } |
| 772 | |
| 773 | static void node_states_set_node(int node, struct memory_notify *arg) |
| 774 | { |
| 775 | if (arg->status_change_nid_normal >= 0) |
| 776 | node_set_state(node, N_NORMAL_MEMORY); |
| 777 | |
| 778 | if (arg->status_change_nid_high >= 0) |
| 779 | node_set_state(node, N_HIGH_MEMORY); |
| 780 | |
| 781 | node_set_state(node, N_MEMORY); |
| 782 | } |
| 783 | |
| 784 | static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn, |
| 785 | unsigned long nr_pages) |
| 786 | { |
| 787 | unsigned long old_end_pfn = zone_end_pfn(zone); |
| 788 | |
| 789 | if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn) |
| 790 | zone->zone_start_pfn = start_pfn; |
| 791 | |
| 792 | zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn; |
| 793 | } |
| 794 | |
| 795 | static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn, |
| 796 | unsigned long nr_pages) |
| 797 | { |
| 798 | unsigned long old_end_pfn = pgdat_end_pfn(pgdat); |
| 799 | |
| 800 | if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn) |
| 801 | pgdat->node_start_pfn = start_pfn; |
| 802 | |
| 803 | pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn; |
| 804 | } |
| 805 | |
| 806 | void __ref move_pfn_range_to_zone(struct zone *zone, |
| 807 | unsigned long start_pfn, unsigned long nr_pages) |
| 808 | { |
| 809 | struct pglist_data *pgdat = zone->zone_pgdat; |
| 810 | int nid = pgdat->node_id; |
| 811 | unsigned long flags; |
| 812 | |
| 813 | if (zone_is_empty(zone)) |
| 814 | init_currently_empty_zone(zone, start_pfn, nr_pages); |
| 815 | |
| 816 | clear_zone_contiguous(zone); |
| 817 | |
| 818 | /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */ |
| 819 | pgdat_resize_lock(pgdat, &flags); |
| 820 | zone_span_writelock(zone); |
| 821 | resize_zone_range(zone, start_pfn, nr_pages); |
| 822 | zone_span_writeunlock(zone); |
| 823 | resize_pgdat_range(pgdat, start_pfn, nr_pages); |
| 824 | pgdat_resize_unlock(pgdat, &flags); |
| 825 | |
| 826 | /* |
| 827 | * TODO now we have a visible range of pages which are not associated |
| 828 | * with their zone properly. Not nice but set_pfnblock_flags_mask |
| 829 | * expects the zone spans the pfn range. All the pages in the range |
| 830 | * are reserved so nobody should be touching them so we should be safe |
| 831 | */ |
| 832 | memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG); |
| 833 | |
| 834 | set_zone_contiguous(zone); |
| 835 | } |
| 836 | |
| 837 | /* |
| 838 | * Returns a default kernel memory zone for the given pfn range. |
| 839 | * If no kernel zone covers this pfn range it will automatically go |
| 840 | * to the ZONE_NORMAL. |
| 841 | */ |
| 842 | static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn, |
| 843 | unsigned long nr_pages) |
| 844 | { |
| 845 | struct pglist_data *pgdat = NODE_DATA(nid); |
| 846 | int zid; |
| 847 | |
| 848 | for (zid = 0; zid <= ZONE_NORMAL; zid++) { |
| 849 | struct zone *zone = &pgdat->node_zones[zid]; |
| 850 | |
| 851 | if (zone_intersects(zone, start_pfn, nr_pages)) |
| 852 | return zone; |
| 853 | } |
| 854 | |
| 855 | return &pgdat->node_zones[ZONE_NORMAL]; |
| 856 | } |
| 857 | |
| 858 | static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn, |
| 859 | unsigned long nr_pages) |
| 860 | { |
| 861 | struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn, |
| 862 | nr_pages); |
| 863 | struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; |
| 864 | bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages); |
| 865 | bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages); |
| 866 | |
| 867 | /* |
| 868 | * We inherit the existing zone in a simple case where zones do not |
| 869 | * overlap in the given range |
| 870 | */ |
| 871 | if (in_kernel ^ in_movable) |
| 872 | return (in_kernel) ? kernel_zone : movable_zone; |
| 873 | |
| 874 | /* |
| 875 | * If the range doesn't belong to any zone or two zones overlap in the |
| 876 | * given range then we use movable zone only if movable_node is |
| 877 | * enabled because we always online to a kernel zone by default. |
| 878 | */ |
| 879 | return movable_node_enabled ? movable_zone : kernel_zone; |
| 880 | } |
| 881 | |
| 882 | struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn, |
| 883 | unsigned long nr_pages) |
| 884 | { |
| 885 | if (online_type == MMOP_ONLINE_KERNEL) |
| 886 | return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages); |
| 887 | |
| 888 | if (online_type == MMOP_ONLINE_MOVABLE) |
| 889 | return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE]; |
| 890 | |
| 891 | return default_zone_for_pfn(nid, start_pfn, nr_pages); |
| 892 | } |
| 893 | |
| 894 | /* |
| 895 | * Associates the given pfn range with the given node and the zone appropriate |
| 896 | * for the given online type. |
| 897 | */ |
| 898 | static struct zone * __meminit move_pfn_range(int online_type, int nid, |
| 899 | unsigned long start_pfn, unsigned long nr_pages) |
| 900 | { |
| 901 | struct zone *zone; |
| 902 | |
| 903 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); |
| 904 | move_pfn_range_to_zone(zone, start_pfn, nr_pages); |
| 905 | return zone; |
| 906 | } |
| 907 | |
| 908 | /* Must be protected by mem_hotplug_begin() or a device_lock */ |
| 909 | int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type) |
| 910 | { |
| 911 | unsigned long flags; |
| 912 | unsigned long onlined_pages = 0; |
| 913 | struct zone *zone; |
| 914 | int need_zonelists_rebuild = 0; |
| 915 | int nid; |
| 916 | int ret; |
| 917 | struct memory_notify arg; |
| 918 | |
| 919 | nid = pfn_to_nid(pfn); |
| 920 | /* associate pfn range with the zone */ |
| 921 | zone = move_pfn_range(online_type, nid, pfn, nr_pages); |
| 922 | |
| 923 | arg.start_pfn = pfn; |
| 924 | arg.nr_pages = nr_pages; |
| 925 | node_states_check_changes_online(nr_pages, zone, &arg); |
| 926 | |
| 927 | ret = memory_notify(MEM_GOING_ONLINE, &arg); |
| 928 | ret = notifier_to_errno(ret); |
| 929 | if (ret) |
| 930 | goto failed_addition; |
| 931 | |
| 932 | /* |
| 933 | * If this zone is not populated, then it is not in zonelist. |
| 934 | * This means the page allocator ignores this zone. |
| 935 | * So, zonelist must be updated after online. |
| 936 | */ |
| 937 | if (!populated_zone(zone)) { |
| 938 | need_zonelists_rebuild = 1; |
| 939 | setup_zone_pageset(zone); |
| 940 | } |
| 941 | |
| 942 | ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages, |
| 943 | online_pages_range); |
| 944 | if (ret) { |
| 945 | if (need_zonelists_rebuild) |
| 946 | zone_pcp_reset(zone); |
| 947 | goto failed_addition; |
| 948 | } |
| 949 | |
| 950 | zone->present_pages += onlined_pages; |
| 951 | |
| 952 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
| 953 | zone->zone_pgdat->node_present_pages += onlined_pages; |
| 954 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
| 955 | |
| 956 | if (onlined_pages) { |
| 957 | node_states_set_node(nid, &arg); |
| 958 | if (need_zonelists_rebuild) |
| 959 | build_all_zonelists(NULL); |
| 960 | else |
| 961 | zone_pcp_update(zone); |
| 962 | } |
| 963 | |
| 964 | init_per_zone_wmark_min(); |
| 965 | |
| 966 | if (onlined_pages) { |
| 967 | kswapd_run(nid); |
| 968 | kcompactd_run(nid); |
| 969 | } |
| 970 | |
| 971 | vm_total_pages = nr_free_pagecache_pages(); |
| 972 | |
| 973 | writeback_set_ratelimit(); |
| 974 | |
| 975 | if (onlined_pages) |
| 976 | memory_notify(MEM_ONLINE, &arg); |
| 977 | return 0; |
| 978 | |
| 979 | failed_addition: |
| 980 | pr_debug("online_pages [mem %#010llx-%#010llx] failed\n", |
| 981 | (unsigned long long) pfn << PAGE_SHIFT, |
| 982 | (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1); |
| 983 | memory_notify(MEM_CANCEL_ONLINE, &arg); |
| 984 | return ret; |
| 985 | } |
| 986 | #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ |
| 987 | |
| 988 | static void reset_node_present_pages(pg_data_t *pgdat) |
| 989 | { |
| 990 | struct zone *z; |
| 991 | |
| 992 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) |
| 993 | z->present_pages = 0; |
| 994 | |
| 995 | pgdat->node_present_pages = 0; |
| 996 | } |
| 997 | |
| 998 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
| 999 | static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start) |
| 1000 | { |
| 1001 | struct pglist_data *pgdat; |
| 1002 | unsigned long zones_size[MAX_NR_ZONES] = {0}; |
| 1003 | unsigned long zholes_size[MAX_NR_ZONES] = {0}; |
| 1004 | unsigned long start_pfn = PFN_DOWN(start); |
| 1005 | |
| 1006 | pgdat = NODE_DATA(nid); |
| 1007 | if (!pgdat) { |
| 1008 | pgdat = arch_alloc_nodedata(nid); |
| 1009 | if (!pgdat) |
| 1010 | return NULL; |
| 1011 | |
| 1012 | arch_refresh_nodedata(nid, pgdat); |
| 1013 | } else { |
| 1014 | /* |
| 1015 | * Reset the nr_zones, order and classzone_idx before reuse. |
| 1016 | * Note that kswapd will init kswapd_classzone_idx properly |
| 1017 | * when it starts in the near future. |
| 1018 | */ |
| 1019 | pgdat->nr_zones = 0; |
| 1020 | pgdat->kswapd_order = 0; |
| 1021 | pgdat->kswapd_classzone_idx = 0; |
| 1022 | } |
| 1023 | |
| 1024 | /* we can use NODE_DATA(nid) from here */ |
| 1025 | |
| 1026 | /* init node's zones as empty zones, we don't have any present pages.*/ |
| 1027 | free_area_init_node(nid, zones_size, start_pfn, zholes_size); |
| 1028 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); |
| 1029 | |
| 1030 | /* |
| 1031 | * The node we allocated has no zone fallback lists. For avoiding |
| 1032 | * to access not-initialized zonelist, build here. |
| 1033 | */ |
| 1034 | build_all_zonelists(pgdat); |
| 1035 | |
| 1036 | /* |
| 1037 | * zone->managed_pages is set to an approximate value in |
| 1038 | * free_area_init_core(), which will cause |
| 1039 | * /sys/device/system/node/nodeX/meminfo has wrong data. |
| 1040 | * So reset it to 0 before any memory is onlined. |
| 1041 | */ |
| 1042 | reset_node_managed_pages(pgdat); |
| 1043 | |
| 1044 | /* |
| 1045 | * When memory is hot-added, all the memory is in offline state. So |
| 1046 | * clear all zones' present_pages because they will be updated in |
| 1047 | * online_pages() and offline_pages(). |
| 1048 | */ |
| 1049 | reset_node_present_pages(pgdat); |
| 1050 | |
| 1051 | return pgdat; |
| 1052 | } |
| 1053 | |
| 1054 | static void rollback_node_hotadd(int nid, pg_data_t *pgdat) |
| 1055 | { |
| 1056 | arch_refresh_nodedata(nid, NULL); |
| 1057 | free_percpu(pgdat->per_cpu_nodestats); |
| 1058 | arch_free_nodedata(pgdat); |
| 1059 | return; |
| 1060 | } |
| 1061 | |
| 1062 | |
| 1063 | /** |
| 1064 | * try_online_node - online a node if offlined |
| 1065 | * |
| 1066 | * called by cpu_up() to online a node without onlined memory. |
| 1067 | */ |
| 1068 | int try_online_node(int nid) |
| 1069 | { |
| 1070 | pg_data_t *pgdat; |
| 1071 | int ret; |
| 1072 | |
| 1073 | if (node_online(nid)) |
| 1074 | return 0; |
| 1075 | |
| 1076 | mem_hotplug_begin(); |
| 1077 | pgdat = hotadd_new_pgdat(nid, 0); |
| 1078 | if (!pgdat) { |
| 1079 | pr_err("Cannot online node %d due to NULL pgdat\n", nid); |
| 1080 | ret = -ENOMEM; |
| 1081 | goto out; |
| 1082 | } |
| 1083 | node_set_online(nid); |
| 1084 | ret = register_one_node(nid); |
| 1085 | BUG_ON(ret); |
| 1086 | out: |
| 1087 | mem_hotplug_done(); |
| 1088 | return ret; |
| 1089 | } |
| 1090 | |
| 1091 | static int check_hotplug_memory_range(u64 start, u64 size) |
| 1092 | { |
| 1093 | u64 start_pfn = PFN_DOWN(start); |
| 1094 | u64 nr_pages = size >> PAGE_SHIFT; |
| 1095 | |
| 1096 | /* Memory range must be aligned with section */ |
| 1097 | if ((start_pfn & ~PAGE_SECTION_MASK) || |
| 1098 | (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) { |
| 1099 | pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n", |
| 1100 | (unsigned long long)start, |
| 1101 | (unsigned long long)size); |
| 1102 | return -EINVAL; |
| 1103 | } |
| 1104 | |
| 1105 | return 0; |
| 1106 | } |
| 1107 | |
| 1108 | static int online_memory_block(struct memory_block *mem, void *arg) |
| 1109 | { |
| 1110 | return device_online(&mem->dev); |
| 1111 | } |
| 1112 | |
| 1113 | /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ |
| 1114 | int __ref add_memory_resource(int nid, struct resource *res, bool online) |
| 1115 | { |
| 1116 | u64 start, size; |
| 1117 | pg_data_t *pgdat = NULL; |
| 1118 | bool new_pgdat; |
| 1119 | bool new_node; |
| 1120 | int ret; |
| 1121 | |
| 1122 | start = res->start; |
| 1123 | size = resource_size(res); |
| 1124 | |
| 1125 | ret = check_hotplug_memory_range(start, size); |
| 1126 | if (ret) |
| 1127 | return ret; |
| 1128 | |
| 1129 | { /* Stupid hack to suppress address-never-null warning */ |
| 1130 | void *p = NODE_DATA(nid); |
| 1131 | new_pgdat = !p; |
| 1132 | } |
| 1133 | |
| 1134 | mem_hotplug_begin(); |
| 1135 | |
| 1136 | /* |
| 1137 | * Add new range to memblock so that when hotadd_new_pgdat() is called |
| 1138 | * to allocate new pgdat, get_pfn_range_for_nid() will be able to find |
| 1139 | * this new range and calculate total pages correctly. The range will |
| 1140 | * be removed at hot-remove time. |
| 1141 | */ |
| 1142 | memblock_add_node(start, size, nid); |
| 1143 | |
| 1144 | new_node = !node_online(nid); |
| 1145 | if (new_node) { |
| 1146 | pgdat = hotadd_new_pgdat(nid, start); |
| 1147 | ret = -ENOMEM; |
| 1148 | if (!pgdat) |
| 1149 | goto error; |
| 1150 | } |
| 1151 | |
| 1152 | /* call arch's memory hotadd */ |
| 1153 | ret = arch_add_memory(nid, start, size, true); |
| 1154 | |
| 1155 | if (ret < 0) |
| 1156 | goto error; |
| 1157 | |
| 1158 | /* we online node here. we can't roll back from here. */ |
| 1159 | node_set_online(nid); |
| 1160 | |
| 1161 | if (new_node) { |
| 1162 | unsigned long start_pfn = start >> PAGE_SHIFT; |
| 1163 | unsigned long nr_pages = size >> PAGE_SHIFT; |
| 1164 | |
| 1165 | ret = __register_one_node(nid); |
| 1166 | if (ret) |
| 1167 | goto register_fail; |
| 1168 | |
| 1169 | /* |
| 1170 | * link memory sections under this node. This is already |
| 1171 | * done when creatig memory section in register_new_memory |
| 1172 | * but that depends to have the node registered so offline |
| 1173 | * nodes have to go through register_node. |
| 1174 | * TODO clean up this mess. |
| 1175 | */ |
| 1176 | ret = link_mem_sections(nid, start_pfn, nr_pages); |
| 1177 | register_fail: |
| 1178 | /* |
| 1179 | * If sysfs file of new node can't create, cpu on the node |
| 1180 | * can't be hot-added. There is no rollback way now. |
| 1181 | * So, check by BUG_ON() to catch it reluctantly.. |
| 1182 | */ |
| 1183 | BUG_ON(ret); |
| 1184 | } |
| 1185 | |
| 1186 | /* create new memmap entry */ |
| 1187 | firmware_map_add_hotplug(start, start + size, "System RAM"); |
| 1188 | |
| 1189 | /* online pages if requested */ |
| 1190 | if (online) |
| 1191 | walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), |
| 1192 | NULL, online_memory_block); |
| 1193 | |
| 1194 | goto out; |
| 1195 | |
| 1196 | error: |
| 1197 | /* rollback pgdat allocation and others */ |
| 1198 | if (new_pgdat && pgdat) |
| 1199 | rollback_node_hotadd(nid, pgdat); |
| 1200 | memblock_remove(start, size); |
| 1201 | |
| 1202 | out: |
| 1203 | mem_hotplug_done(); |
| 1204 | return ret; |
| 1205 | } |
| 1206 | EXPORT_SYMBOL_GPL(add_memory_resource); |
| 1207 | |
| 1208 | int __ref add_memory(int nid, u64 start, u64 size) |
| 1209 | { |
| 1210 | struct resource *res; |
| 1211 | int ret; |
| 1212 | |
| 1213 | res = register_memory_resource(start, size); |
| 1214 | if (IS_ERR(res)) |
| 1215 | return PTR_ERR(res); |
| 1216 | |
| 1217 | ret = add_memory_resource(nid, res, memhp_auto_online); |
| 1218 | if (ret < 0) |
| 1219 | release_memory_resource(res); |
| 1220 | return ret; |
| 1221 | } |
| 1222 | EXPORT_SYMBOL_GPL(add_memory); |
| 1223 | |
| 1224 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 1225 | /* |
| 1226 | * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy |
| 1227 | * set and the size of the free page is given by page_order(). Using this, |
| 1228 | * the function determines if the pageblock contains only free pages. |
| 1229 | * Due to buddy contraints, a free page at least the size of a pageblock will |
| 1230 | * be located at the start of the pageblock |
| 1231 | */ |
| 1232 | static inline int pageblock_free(struct page *page) |
| 1233 | { |
| 1234 | return PageBuddy(page) && page_order(page) >= pageblock_order; |
| 1235 | } |
| 1236 | |
| 1237 | /* Return the start of the next active pageblock after a given page */ |
| 1238 | static struct page *next_active_pageblock(struct page *page) |
| 1239 | { |
| 1240 | /* Ensure the starting page is pageblock-aligned */ |
| 1241 | BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1)); |
| 1242 | |
| 1243 | /* If the entire pageblock is free, move to the end of free page */ |
| 1244 | if (pageblock_free(page)) { |
| 1245 | int order; |
| 1246 | /* be careful. we don't have locks, page_order can be changed.*/ |
| 1247 | order = page_order(page); |
| 1248 | if ((order < MAX_ORDER) && (order >= pageblock_order)) |
| 1249 | return page + (1 << order); |
| 1250 | } |
| 1251 | |
| 1252 | return page + pageblock_nr_pages; |
| 1253 | } |
| 1254 | |
| 1255 | /* Checks if this range of memory is likely to be hot-removable. */ |
| 1256 | bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages) |
| 1257 | { |
| 1258 | struct page *page = pfn_to_page(start_pfn); |
| 1259 | unsigned long end_pfn = min(start_pfn + nr_pages, zone_end_pfn(page_zone(page))); |
| 1260 | struct page *end_page = pfn_to_page(end_pfn); |
| 1261 | |
| 1262 | /* Check the starting page of each pageblock within the range */ |
| 1263 | for (; page < end_page; page = next_active_pageblock(page)) { |
| 1264 | if (!is_pageblock_removable_nolock(page)) |
| 1265 | return false; |
| 1266 | cond_resched(); |
| 1267 | } |
| 1268 | |
| 1269 | /* All pageblocks in the memory block are likely to be hot-removable */ |
| 1270 | return true; |
| 1271 | } |
| 1272 | |
| 1273 | /* |
| 1274 | * Confirm all pages in a range [start, end) belong to the same zone. |
| 1275 | * When true, return its valid [start, end). |
| 1276 | */ |
| 1277 | int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn, |
| 1278 | unsigned long *valid_start, unsigned long *valid_end) |
| 1279 | { |
| 1280 | unsigned long pfn, sec_end_pfn; |
| 1281 | unsigned long start, end; |
| 1282 | struct zone *zone = NULL; |
| 1283 | struct page *page; |
| 1284 | int i; |
| 1285 | for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1); |
| 1286 | pfn < end_pfn; |
| 1287 | pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) { |
| 1288 | /* Make sure the memory section is present first */ |
| 1289 | if (!present_section_nr(pfn_to_section_nr(pfn))) |
| 1290 | continue; |
| 1291 | for (; pfn < sec_end_pfn && pfn < end_pfn; |
| 1292 | pfn += MAX_ORDER_NR_PAGES) { |
| 1293 | i = 0; |
| 1294 | /* This is just a CONFIG_HOLES_IN_ZONE check.*/ |
| 1295 | while ((i < MAX_ORDER_NR_PAGES) && |
| 1296 | !pfn_valid_within(pfn + i)) |
| 1297 | i++; |
| 1298 | if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn) |
| 1299 | continue; |
| 1300 | /* Check if we got outside of the zone */ |
| 1301 | if (zone && !zone_spans_pfn(zone, pfn + i)) |
| 1302 | return 0; |
| 1303 | page = pfn_to_page(pfn + i); |
| 1304 | if (zone && page_zone(page) != zone) |
| 1305 | return 0; |
| 1306 | if (!zone) |
| 1307 | start = pfn + i; |
| 1308 | zone = page_zone(page); |
| 1309 | end = pfn + MAX_ORDER_NR_PAGES; |
| 1310 | } |
| 1311 | } |
| 1312 | |
| 1313 | if (zone) { |
| 1314 | *valid_start = start; |
| 1315 | *valid_end = min(end, end_pfn); |
| 1316 | return 1; |
| 1317 | } else { |
| 1318 | return 0; |
| 1319 | } |
| 1320 | } |
| 1321 | |
| 1322 | /* |
| 1323 | * Scan pfn range [start,end) to find movable/migratable pages (LRU pages, |
| 1324 | * non-lru movable pages and hugepages). We scan pfn because it's much |
| 1325 | * easier than scanning over linked list. This function returns the pfn |
| 1326 | * of the first found movable page if it's found, otherwise 0. |
| 1327 | */ |
| 1328 | static unsigned long scan_movable_pages(unsigned long start, unsigned long end) |
| 1329 | { |
| 1330 | unsigned long pfn; |
| 1331 | struct page *page; |
| 1332 | for (pfn = start; pfn < end; pfn++) { |
| 1333 | if (pfn_valid(pfn)) { |
| 1334 | page = pfn_to_page(pfn); |
| 1335 | if (PageLRU(page)) |
| 1336 | return pfn; |
| 1337 | if (__PageMovable(page)) |
| 1338 | return pfn; |
| 1339 | if (PageHuge(page)) { |
| 1340 | if (page_huge_active(page)) |
| 1341 | return pfn; |
| 1342 | else |
| 1343 | pfn = round_up(pfn + 1, |
| 1344 | 1 << compound_order(page)) - 1; |
| 1345 | } |
| 1346 | } |
| 1347 | } |
| 1348 | return 0; |
| 1349 | } |
| 1350 | |
| 1351 | static struct page *new_node_page(struct page *page, unsigned long private, |
| 1352 | int **result) |
| 1353 | { |
| 1354 | int nid = page_to_nid(page); |
| 1355 | nodemask_t nmask = node_states[N_MEMORY]; |
| 1356 | |
| 1357 | /* |
| 1358 | * try to allocate from a different node but reuse this node if there |
| 1359 | * are no other online nodes to be used (e.g. we are offlining a part |
| 1360 | * of the only existing node) |
| 1361 | */ |
| 1362 | node_clear(nid, nmask); |
| 1363 | if (nodes_empty(nmask)) |
| 1364 | node_set(nid, nmask); |
| 1365 | |
| 1366 | return new_page_nodemask(page, nid, &nmask); |
| 1367 | } |
| 1368 | |
| 1369 | #define NR_OFFLINE_AT_ONCE_PAGES (256) |
| 1370 | static int |
| 1371 | do_migrate_range(unsigned long start_pfn, unsigned long end_pfn) |
| 1372 | { |
| 1373 | unsigned long pfn; |
| 1374 | struct page *page; |
| 1375 | int move_pages = NR_OFFLINE_AT_ONCE_PAGES; |
| 1376 | int not_managed = 0; |
| 1377 | int ret = 0; |
| 1378 | LIST_HEAD(source); |
| 1379 | |
| 1380 | for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) { |
| 1381 | if (!pfn_valid(pfn)) |
| 1382 | continue; |
| 1383 | page = pfn_to_page(pfn); |
| 1384 | |
| 1385 | if (PageHuge(page)) { |
| 1386 | struct page *head = compound_head(page); |
| 1387 | pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1; |
| 1388 | if (compound_order(head) > PFN_SECTION_SHIFT) { |
| 1389 | ret = -EBUSY; |
| 1390 | break; |
| 1391 | } |
| 1392 | if (isolate_huge_page(page, &source)) |
| 1393 | move_pages -= 1 << compound_order(head); |
| 1394 | continue; |
| 1395 | } else if (thp_migration_supported() && PageTransHuge(page)) |
| 1396 | pfn = page_to_pfn(compound_head(page)) |
| 1397 | + hpage_nr_pages(page) - 1; |
| 1398 | |
| 1399 | /* |
| 1400 | * HWPoison pages have elevated reference counts so the migration would |
| 1401 | * fail on them. It also doesn't make any sense to migrate them in the |
| 1402 | * first place. Still try to unmap such a page in case it is still mapped |
| 1403 | * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep |
| 1404 | * the unmap as the catch all safety net). |
| 1405 | */ |
| 1406 | if (PageHWPoison(page)) { |
| 1407 | if (WARN_ON(PageLRU(page))) |
| 1408 | isolate_lru_page(page); |
| 1409 | if (page_mapped(page)) |
| 1410 | try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS); |
| 1411 | continue; |
| 1412 | } |
| 1413 | |
| 1414 | if (!get_page_unless_zero(page)) |
| 1415 | continue; |
| 1416 | /* |
| 1417 | * We can skip free pages. And we can deal with pages on |
| 1418 | * LRU and non-lru movable pages. |
| 1419 | */ |
| 1420 | if (PageLRU(page)) |
| 1421 | ret = isolate_lru_page(page); |
| 1422 | else |
| 1423 | ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE); |
| 1424 | if (!ret) { /* Success */ |
| 1425 | put_page(page); |
| 1426 | list_add_tail(&page->lru, &source); |
| 1427 | move_pages--; |
| 1428 | if (!__PageMovable(page)) |
| 1429 | inc_node_page_state(page, NR_ISOLATED_ANON + |
| 1430 | page_is_file_cache(page)); |
| 1431 | |
| 1432 | } else { |
| 1433 | #ifdef CONFIG_DEBUG_VM |
| 1434 | pr_alert("failed to isolate pfn %lx\n", pfn); |
| 1435 | dump_page(page, "isolation failed"); |
| 1436 | #endif |
| 1437 | put_page(page); |
| 1438 | /* Because we don't have big zone->lock. we should |
| 1439 | check this again here. */ |
| 1440 | if (page_count(page)) { |
| 1441 | not_managed++; |
| 1442 | ret = -EBUSY; |
| 1443 | break; |
| 1444 | } |
| 1445 | } |
| 1446 | } |
| 1447 | if (!list_empty(&source)) { |
| 1448 | if (not_managed) { |
| 1449 | putback_movable_pages(&source); |
| 1450 | goto out; |
| 1451 | } |
| 1452 | |
| 1453 | /* Allocate a new page from the nearest neighbor node */ |
| 1454 | ret = migrate_pages(&source, new_node_page, NULL, 0, |
| 1455 | MIGRATE_SYNC, MR_MEMORY_HOTPLUG); |
| 1456 | if (ret) |
| 1457 | putback_movable_pages(&source); |
| 1458 | } |
| 1459 | out: |
| 1460 | return ret; |
| 1461 | } |
| 1462 | |
| 1463 | /* |
| 1464 | * remove from free_area[] and mark all as Reserved. |
| 1465 | */ |
| 1466 | static int |
| 1467 | offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages, |
| 1468 | void *data) |
| 1469 | { |
| 1470 | __offline_isolated_pages(start, start + nr_pages); |
| 1471 | return 0; |
| 1472 | } |
| 1473 | |
| 1474 | static void |
| 1475 | offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
| 1476 | { |
| 1477 | walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL, |
| 1478 | offline_isolated_pages_cb); |
| 1479 | } |
| 1480 | |
| 1481 | /* |
| 1482 | * Check all pages in range, recoreded as memory resource, are isolated. |
| 1483 | */ |
| 1484 | static int |
| 1485 | check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages, |
| 1486 | void *data) |
| 1487 | { |
| 1488 | int ret; |
| 1489 | long offlined = *(long *)data; |
| 1490 | ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true); |
| 1491 | offlined = nr_pages; |
| 1492 | if (!ret) |
| 1493 | *(long *)data += offlined; |
| 1494 | return ret; |
| 1495 | } |
| 1496 | |
| 1497 | static long |
| 1498 | check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn) |
| 1499 | { |
| 1500 | long offlined = 0; |
| 1501 | int ret; |
| 1502 | |
| 1503 | ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined, |
| 1504 | check_pages_isolated_cb); |
| 1505 | if (ret < 0) |
| 1506 | offlined = (long)ret; |
| 1507 | return offlined; |
| 1508 | } |
| 1509 | |
| 1510 | static int __init cmdline_parse_movable_node(char *p) |
| 1511 | { |
| 1512 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 1513 | movable_node_enabled = true; |
| 1514 | #else |
| 1515 | pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n"); |
| 1516 | #endif |
| 1517 | return 0; |
| 1518 | } |
| 1519 | early_param("movable_node", cmdline_parse_movable_node); |
| 1520 | |
| 1521 | /* check which state of node_states will be changed when offline memory */ |
| 1522 | static void node_states_check_changes_offline(unsigned long nr_pages, |
| 1523 | struct zone *zone, struct memory_notify *arg) |
| 1524 | { |
| 1525 | struct pglist_data *pgdat = zone->zone_pgdat; |
| 1526 | unsigned long present_pages = 0; |
| 1527 | enum zone_type zt, zone_last = ZONE_NORMAL; |
| 1528 | |
| 1529 | /* |
| 1530 | * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY] |
| 1531 | * contains nodes which have zones of 0...ZONE_NORMAL, |
| 1532 | * set zone_last to ZONE_NORMAL. |
| 1533 | * |
| 1534 | * If we don't have HIGHMEM nor movable node, |
| 1535 | * node_states[N_NORMAL_MEMORY] contains nodes which have zones of |
| 1536 | * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE. |
| 1537 | */ |
| 1538 | if (N_MEMORY == N_NORMAL_MEMORY) |
| 1539 | zone_last = ZONE_MOVABLE; |
| 1540 | |
| 1541 | /* |
| 1542 | * check whether node_states[N_NORMAL_MEMORY] will be changed. |
| 1543 | * If the memory to be offline is in a zone of 0...zone_last, |
| 1544 | * and it is the last present memory, 0...zone_last will |
| 1545 | * become empty after offline , thus we can determind we will |
| 1546 | * need to clear the node from node_states[N_NORMAL_MEMORY]. |
| 1547 | */ |
| 1548 | for (zt = 0; zt <= zone_last; zt++) |
| 1549 | present_pages += pgdat->node_zones[zt].present_pages; |
| 1550 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) |
| 1551 | arg->status_change_nid_normal = zone_to_nid(zone); |
| 1552 | else |
| 1553 | arg->status_change_nid_normal = -1; |
| 1554 | |
| 1555 | #ifdef CONFIG_HIGHMEM |
| 1556 | /* |
| 1557 | * If we have movable node, node_states[N_HIGH_MEMORY] |
| 1558 | * contains nodes which have zones of 0...ZONE_HIGHMEM, |
| 1559 | * set zone_last to ZONE_HIGHMEM. |
| 1560 | * |
| 1561 | * If we don't have movable node, node_states[N_NORMAL_MEMORY] |
| 1562 | * contains nodes which have zones of 0...ZONE_MOVABLE, |
| 1563 | * set zone_last to ZONE_MOVABLE. |
| 1564 | */ |
| 1565 | zone_last = ZONE_HIGHMEM; |
| 1566 | if (N_MEMORY == N_HIGH_MEMORY) |
| 1567 | zone_last = ZONE_MOVABLE; |
| 1568 | |
| 1569 | for (; zt <= zone_last; zt++) |
| 1570 | present_pages += pgdat->node_zones[zt].present_pages; |
| 1571 | if (zone_idx(zone) <= zone_last && nr_pages >= present_pages) |
| 1572 | arg->status_change_nid_high = zone_to_nid(zone); |
| 1573 | else |
| 1574 | arg->status_change_nid_high = -1; |
| 1575 | #else |
| 1576 | arg->status_change_nid_high = arg->status_change_nid_normal; |
| 1577 | #endif |
| 1578 | |
| 1579 | /* |
| 1580 | * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE |
| 1581 | */ |
| 1582 | zone_last = ZONE_MOVABLE; |
| 1583 | |
| 1584 | /* |
| 1585 | * check whether node_states[N_HIGH_MEMORY] will be changed |
| 1586 | * If we try to offline the last present @nr_pages from the node, |
| 1587 | * we can determind we will need to clear the node from |
| 1588 | * node_states[N_HIGH_MEMORY]. |
| 1589 | */ |
| 1590 | for (; zt <= zone_last; zt++) |
| 1591 | present_pages += pgdat->node_zones[zt].present_pages; |
| 1592 | if (nr_pages >= present_pages) |
| 1593 | arg->status_change_nid = zone_to_nid(zone); |
| 1594 | else |
| 1595 | arg->status_change_nid = -1; |
| 1596 | } |
| 1597 | |
| 1598 | static void node_states_clear_node(int node, struct memory_notify *arg) |
| 1599 | { |
| 1600 | if (arg->status_change_nid_normal >= 0) |
| 1601 | node_clear_state(node, N_NORMAL_MEMORY); |
| 1602 | |
| 1603 | if ((N_MEMORY != N_NORMAL_MEMORY) && |
| 1604 | (arg->status_change_nid_high >= 0)) |
| 1605 | node_clear_state(node, N_HIGH_MEMORY); |
| 1606 | |
| 1607 | if ((N_MEMORY != N_HIGH_MEMORY) && |
| 1608 | (arg->status_change_nid >= 0)) |
| 1609 | node_clear_state(node, N_MEMORY); |
| 1610 | } |
| 1611 | |
| 1612 | static int __ref __offline_pages(unsigned long start_pfn, |
| 1613 | unsigned long end_pfn, unsigned long timeout) |
| 1614 | { |
| 1615 | unsigned long pfn, nr_pages, expire; |
| 1616 | long offlined_pages; |
| 1617 | int ret, drain, retry_max, node; |
| 1618 | unsigned long flags; |
| 1619 | unsigned long valid_start, valid_end; |
| 1620 | struct zone *zone; |
| 1621 | struct memory_notify arg; |
| 1622 | |
| 1623 | /* at least, alignment against pageblock is necessary */ |
| 1624 | if (!IS_ALIGNED(start_pfn, pageblock_nr_pages)) |
| 1625 | return -EINVAL; |
| 1626 | if (!IS_ALIGNED(end_pfn, pageblock_nr_pages)) |
| 1627 | return -EINVAL; |
| 1628 | /* This makes hotplug much easier...and readable. |
| 1629 | we assume this for now. .*/ |
| 1630 | if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end)) |
| 1631 | return -EINVAL; |
| 1632 | |
| 1633 | zone = page_zone(pfn_to_page(valid_start)); |
| 1634 | node = zone_to_nid(zone); |
| 1635 | nr_pages = end_pfn - start_pfn; |
| 1636 | |
| 1637 | /* set above range as isolated */ |
| 1638 | ret = start_isolate_page_range(start_pfn, end_pfn, |
| 1639 | MIGRATE_MOVABLE, true); |
| 1640 | if (ret) |
| 1641 | return ret; |
| 1642 | |
| 1643 | arg.start_pfn = start_pfn; |
| 1644 | arg.nr_pages = nr_pages; |
| 1645 | node_states_check_changes_offline(nr_pages, zone, &arg); |
| 1646 | |
| 1647 | ret = memory_notify(MEM_GOING_OFFLINE, &arg); |
| 1648 | ret = notifier_to_errno(ret); |
| 1649 | if (ret) |
| 1650 | goto failed_removal; |
| 1651 | |
| 1652 | pfn = start_pfn; |
| 1653 | expire = jiffies + timeout; |
| 1654 | drain = 0; |
| 1655 | retry_max = 5; |
| 1656 | repeat: |
| 1657 | /* start memory hot removal */ |
| 1658 | ret = -EAGAIN; |
| 1659 | if (time_after(jiffies, expire)) |
| 1660 | goto failed_removal; |
| 1661 | ret = -EINTR; |
| 1662 | if (signal_pending(current)) |
| 1663 | goto failed_removal; |
| 1664 | ret = 0; |
| 1665 | if (drain) { |
| 1666 | lru_add_drain_all_cpuslocked(); |
| 1667 | cond_resched(); |
| 1668 | drain_all_pages(zone); |
| 1669 | } |
| 1670 | |
| 1671 | pfn = scan_movable_pages(start_pfn, end_pfn); |
| 1672 | if (pfn) { /* We have movable pages */ |
| 1673 | ret = do_migrate_range(pfn, end_pfn); |
| 1674 | if (!ret) { |
| 1675 | drain = 1; |
| 1676 | goto repeat; |
| 1677 | } else { |
| 1678 | if (ret < 0) |
| 1679 | if (--retry_max == 0) |
| 1680 | goto failed_removal; |
| 1681 | yield(); |
| 1682 | drain = 1; |
| 1683 | goto repeat; |
| 1684 | } |
| 1685 | } |
| 1686 | /* drain all zone's lru pagevec, this is asynchronous... */ |
| 1687 | lru_add_drain_all_cpuslocked(); |
| 1688 | yield(); |
| 1689 | /* drain pcp pages, this is synchronous. */ |
| 1690 | drain_all_pages(zone); |
| 1691 | /* |
| 1692 | * dissolve free hugepages in the memory block before doing offlining |
| 1693 | * actually in order to make hugetlbfs's object counting consistent. |
| 1694 | */ |
| 1695 | ret = dissolve_free_huge_pages(start_pfn, end_pfn); |
| 1696 | if (ret) |
| 1697 | goto failed_removal; |
| 1698 | /* check again */ |
| 1699 | offlined_pages = check_pages_isolated(start_pfn, end_pfn); |
| 1700 | if (offlined_pages < 0) { |
| 1701 | ret = -EBUSY; |
| 1702 | goto failed_removal; |
| 1703 | } |
| 1704 | pr_info("Offlined Pages %ld\n", offlined_pages); |
| 1705 | /* Ok, all of our target is isolated. |
| 1706 | We cannot do rollback at this point. */ |
| 1707 | offline_isolated_pages(start_pfn, end_pfn); |
| 1708 | /* reset pagetype flags and makes migrate type to be MOVABLE */ |
| 1709 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
| 1710 | /* removal success */ |
| 1711 | adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages); |
| 1712 | zone->present_pages -= offlined_pages; |
| 1713 | |
| 1714 | pgdat_resize_lock(zone->zone_pgdat, &flags); |
| 1715 | zone->zone_pgdat->node_present_pages -= offlined_pages; |
| 1716 | pgdat_resize_unlock(zone->zone_pgdat, &flags); |
| 1717 | |
| 1718 | init_per_zone_wmark_min(); |
| 1719 | |
| 1720 | if (!populated_zone(zone)) { |
| 1721 | zone_pcp_reset(zone); |
| 1722 | build_all_zonelists(NULL); |
| 1723 | } else |
| 1724 | zone_pcp_update(zone); |
| 1725 | |
| 1726 | node_states_clear_node(node, &arg); |
| 1727 | if (arg.status_change_nid >= 0) { |
| 1728 | kswapd_stop(node); |
| 1729 | kcompactd_stop(node); |
| 1730 | } |
| 1731 | |
| 1732 | vm_total_pages = nr_free_pagecache_pages(); |
| 1733 | writeback_set_ratelimit(); |
| 1734 | |
| 1735 | memory_notify(MEM_OFFLINE, &arg); |
| 1736 | return 0; |
| 1737 | |
| 1738 | failed_removal: |
| 1739 | pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n", |
| 1740 | (unsigned long long) start_pfn << PAGE_SHIFT, |
| 1741 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); |
| 1742 | memory_notify(MEM_CANCEL_OFFLINE, &arg); |
| 1743 | /* pushback to free area */ |
| 1744 | undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE); |
| 1745 | return ret; |
| 1746 | } |
| 1747 | |
| 1748 | /* Must be protected by mem_hotplug_begin() or a device_lock */ |
| 1749 | int offline_pages(unsigned long start_pfn, unsigned long nr_pages) |
| 1750 | { |
| 1751 | return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ); |
| 1752 | } |
| 1753 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
| 1754 | |
| 1755 | /** |
| 1756 | * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn) |
| 1757 | * @start_pfn: start pfn of the memory range |
| 1758 | * @end_pfn: end pfn of the memory range |
| 1759 | * @arg: argument passed to func |
| 1760 | * @func: callback for each memory section walked |
| 1761 | * |
| 1762 | * This function walks through all present mem sections in range |
| 1763 | * [start_pfn, end_pfn) and call func on each mem section. |
| 1764 | * |
| 1765 | * Returns the return value of func. |
| 1766 | */ |
| 1767 | int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn, |
| 1768 | void *arg, int (*func)(struct memory_block *, void *)) |
| 1769 | { |
| 1770 | struct memory_block *mem = NULL; |
| 1771 | struct mem_section *section; |
| 1772 | unsigned long pfn, section_nr; |
| 1773 | int ret; |
| 1774 | |
| 1775 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
| 1776 | section_nr = pfn_to_section_nr(pfn); |
| 1777 | if (!present_section_nr(section_nr)) |
| 1778 | continue; |
| 1779 | |
| 1780 | section = __nr_to_section(section_nr); |
| 1781 | /* same memblock? */ |
| 1782 | if (mem) |
| 1783 | if ((section_nr >= mem->start_section_nr) && |
| 1784 | (section_nr <= mem->end_section_nr)) |
| 1785 | continue; |
| 1786 | |
| 1787 | mem = find_memory_block_hinted(section, mem); |
| 1788 | if (!mem) |
| 1789 | continue; |
| 1790 | |
| 1791 | ret = func(mem, arg); |
| 1792 | if (ret) { |
| 1793 | kobject_put(&mem->dev.kobj); |
| 1794 | return ret; |
| 1795 | } |
| 1796 | } |
| 1797 | |
| 1798 | if (mem) |
| 1799 | kobject_put(&mem->dev.kobj); |
| 1800 | |
| 1801 | return 0; |
| 1802 | } |
| 1803 | |
| 1804 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 1805 | static int check_memblock_offlined_cb(struct memory_block *mem, void *arg) |
| 1806 | { |
| 1807 | int ret = !is_memblock_offlined(mem); |
| 1808 | |
| 1809 | if (unlikely(ret)) { |
| 1810 | phys_addr_t beginpa, endpa; |
| 1811 | |
| 1812 | beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)); |
| 1813 | endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1; |
| 1814 | pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n", |
| 1815 | &beginpa, &endpa); |
| 1816 | } |
| 1817 | |
| 1818 | return ret; |
| 1819 | } |
| 1820 | |
| 1821 | static int check_cpu_on_node(pg_data_t *pgdat) |
| 1822 | { |
| 1823 | int cpu; |
| 1824 | |
| 1825 | for_each_present_cpu(cpu) { |
| 1826 | if (cpu_to_node(cpu) == pgdat->node_id) |
| 1827 | /* |
| 1828 | * the cpu on this node isn't removed, and we can't |
| 1829 | * offline this node. |
| 1830 | */ |
| 1831 | return -EBUSY; |
| 1832 | } |
| 1833 | |
| 1834 | return 0; |
| 1835 | } |
| 1836 | |
| 1837 | static void unmap_cpu_on_node(pg_data_t *pgdat) |
| 1838 | { |
| 1839 | #ifdef CONFIG_ACPI_NUMA |
| 1840 | int cpu; |
| 1841 | |
| 1842 | for_each_possible_cpu(cpu) |
| 1843 | if (cpu_to_node(cpu) == pgdat->node_id) |
| 1844 | numa_clear_node(cpu); |
| 1845 | #endif |
| 1846 | } |
| 1847 | |
| 1848 | static int check_and_unmap_cpu_on_node(pg_data_t *pgdat) |
| 1849 | { |
| 1850 | int ret; |
| 1851 | |
| 1852 | ret = check_cpu_on_node(pgdat); |
| 1853 | if (ret) |
| 1854 | return ret; |
| 1855 | |
| 1856 | /* |
| 1857 | * the node will be offlined when we come here, so we can clear |
| 1858 | * the cpu_to_node() now. |
| 1859 | */ |
| 1860 | |
| 1861 | unmap_cpu_on_node(pgdat); |
| 1862 | return 0; |
| 1863 | } |
| 1864 | |
| 1865 | /** |
| 1866 | * try_offline_node |
| 1867 | * |
| 1868 | * Offline a node if all memory sections and cpus of the node are removed. |
| 1869 | * |
| 1870 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug |
| 1871 | * and online/offline operations before this call. |
| 1872 | */ |
| 1873 | void try_offline_node(int nid) |
| 1874 | { |
| 1875 | pg_data_t *pgdat = NODE_DATA(nid); |
| 1876 | unsigned long start_pfn = pgdat->node_start_pfn; |
| 1877 | unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages; |
| 1878 | unsigned long pfn; |
| 1879 | |
| 1880 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
| 1881 | unsigned long section_nr = pfn_to_section_nr(pfn); |
| 1882 | |
| 1883 | if (!present_section_nr(section_nr)) |
| 1884 | continue; |
| 1885 | |
| 1886 | if (pfn_to_nid(pfn) != nid) |
| 1887 | continue; |
| 1888 | |
| 1889 | /* |
| 1890 | * some memory sections of this node are not removed, and we |
| 1891 | * can't offline node now. |
| 1892 | */ |
| 1893 | return; |
| 1894 | } |
| 1895 | |
| 1896 | if (check_and_unmap_cpu_on_node(pgdat)) |
| 1897 | return; |
| 1898 | |
| 1899 | /* |
| 1900 | * all memory/cpu of this node are removed, we can offline this |
| 1901 | * node now. |
| 1902 | */ |
| 1903 | node_set_offline(nid); |
| 1904 | unregister_one_node(nid); |
| 1905 | } |
| 1906 | EXPORT_SYMBOL(try_offline_node); |
| 1907 | |
| 1908 | /** |
| 1909 | * remove_memory |
| 1910 | * |
| 1911 | * NOTE: The caller must call lock_device_hotplug() to serialize hotplug |
| 1912 | * and online/offline operations before this call, as required by |
| 1913 | * try_offline_node(). |
| 1914 | */ |
| 1915 | void __ref remove_memory(int nid, u64 start, u64 size) |
| 1916 | { |
| 1917 | int ret; |
| 1918 | |
| 1919 | BUG_ON(check_hotplug_memory_range(start, size)); |
| 1920 | |
| 1921 | mem_hotplug_begin(); |
| 1922 | |
| 1923 | /* |
| 1924 | * All memory blocks must be offlined before removing memory. Check |
| 1925 | * whether all memory blocks in question are offline and trigger a BUG() |
| 1926 | * if this is not the case. |
| 1927 | */ |
| 1928 | ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL, |
| 1929 | check_memblock_offlined_cb); |
| 1930 | if (ret) |
| 1931 | BUG(); |
| 1932 | |
| 1933 | /* remove memmap entry */ |
| 1934 | firmware_map_remove(start, start + size, "System RAM"); |
| 1935 | memblock_free(start, size); |
| 1936 | memblock_remove(start, size); |
| 1937 | |
| 1938 | arch_remove_memory(start, size); |
| 1939 | |
| 1940 | try_offline_node(nid); |
| 1941 | |
| 1942 | mem_hotplug_done(); |
| 1943 | } |
| 1944 | EXPORT_SYMBOL_GPL(remove_memory); |
| 1945 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |