| 1 | /* |
| 2 | * Memory subsystem support |
| 3 | * |
| 4 | * Written by Matt Tolentino <matthew.e.tolentino@intel.com> |
| 5 | * Dave Hansen <haveblue@us.ibm.com> |
| 6 | * |
| 7 | * This file provides the necessary infrastructure to represent |
| 8 | * a SPARSEMEM-memory-model system's physical memory in /sysfs. |
| 9 | * All arch-independent code that assumes MEMORY_HOTPLUG requires |
| 10 | * SPARSEMEM should be contained here, or in mm/memory_hotplug.c. |
| 11 | */ |
| 12 | |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/topology.h> |
| 16 | #include <linux/capability.h> |
| 17 | #include <linux/device.h> |
| 18 | #include <linux/memory.h> |
| 19 | #include <linux/memory_hotplug.h> |
| 20 | #include <linux/mm.h> |
| 21 | #include <linux/mutex.h> |
| 22 | #include <linux/stat.h> |
| 23 | #include <linux/slab.h> |
| 24 | |
| 25 | #include <linux/atomic.h> |
| 26 | #include <linux/uaccess.h> |
| 27 | |
| 28 | static DEFINE_MUTEX(mem_sysfs_mutex); |
| 29 | |
| 30 | #define MEMORY_CLASS_NAME "memory" |
| 31 | |
| 32 | #define to_memory_block(dev) container_of(dev, struct memory_block, dev) |
| 33 | |
| 34 | static int sections_per_block; |
| 35 | |
| 36 | static inline int base_memory_block_id(int section_nr) |
| 37 | { |
| 38 | return section_nr / sections_per_block; |
| 39 | } |
| 40 | |
| 41 | static int memory_subsys_online(struct device *dev); |
| 42 | static int memory_subsys_offline(struct device *dev); |
| 43 | |
| 44 | static struct bus_type memory_subsys = { |
| 45 | .name = MEMORY_CLASS_NAME, |
| 46 | .dev_name = MEMORY_CLASS_NAME, |
| 47 | .online = memory_subsys_online, |
| 48 | .offline = memory_subsys_offline, |
| 49 | }; |
| 50 | |
| 51 | static BLOCKING_NOTIFIER_HEAD(memory_chain); |
| 52 | |
| 53 | int register_memory_notifier(struct notifier_block *nb) |
| 54 | { |
| 55 | return blocking_notifier_chain_register(&memory_chain, nb); |
| 56 | } |
| 57 | EXPORT_SYMBOL(register_memory_notifier); |
| 58 | |
| 59 | void unregister_memory_notifier(struct notifier_block *nb) |
| 60 | { |
| 61 | blocking_notifier_chain_unregister(&memory_chain, nb); |
| 62 | } |
| 63 | EXPORT_SYMBOL(unregister_memory_notifier); |
| 64 | |
| 65 | static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain); |
| 66 | |
| 67 | int register_memory_isolate_notifier(struct notifier_block *nb) |
| 68 | { |
| 69 | return atomic_notifier_chain_register(&memory_isolate_chain, nb); |
| 70 | } |
| 71 | EXPORT_SYMBOL(register_memory_isolate_notifier); |
| 72 | |
| 73 | void unregister_memory_isolate_notifier(struct notifier_block *nb) |
| 74 | { |
| 75 | atomic_notifier_chain_unregister(&memory_isolate_chain, nb); |
| 76 | } |
| 77 | EXPORT_SYMBOL(unregister_memory_isolate_notifier); |
| 78 | |
| 79 | static void memory_block_release(struct device *dev) |
| 80 | { |
| 81 | struct memory_block *mem = to_memory_block(dev); |
| 82 | |
| 83 | kfree(mem); |
| 84 | } |
| 85 | |
| 86 | unsigned long __weak memory_block_size_bytes(void) |
| 87 | { |
| 88 | return MIN_MEMORY_BLOCK_SIZE; |
| 89 | } |
| 90 | |
| 91 | static unsigned long get_memory_block_size(void) |
| 92 | { |
| 93 | unsigned long block_sz; |
| 94 | |
| 95 | block_sz = memory_block_size_bytes(); |
| 96 | |
| 97 | /* Validate blk_sz is a power of 2 and not less than section size */ |
| 98 | if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) { |
| 99 | WARN_ON(1); |
| 100 | block_sz = MIN_MEMORY_BLOCK_SIZE; |
| 101 | } |
| 102 | |
| 103 | return block_sz; |
| 104 | } |
| 105 | |
| 106 | /* |
| 107 | * use this as the physical section index that this memsection |
| 108 | * uses. |
| 109 | */ |
| 110 | |
| 111 | static ssize_t show_mem_start_phys_index(struct device *dev, |
| 112 | struct device_attribute *attr, char *buf) |
| 113 | { |
| 114 | struct memory_block *mem = to_memory_block(dev); |
| 115 | unsigned long phys_index; |
| 116 | |
| 117 | phys_index = mem->start_section_nr / sections_per_block; |
| 118 | return sprintf(buf, "%08lx\n", phys_index); |
| 119 | } |
| 120 | |
| 121 | /* |
| 122 | * Show whether the section of memory is likely to be hot-removable |
| 123 | */ |
| 124 | static ssize_t show_mem_removable(struct device *dev, |
| 125 | struct device_attribute *attr, char *buf) |
| 126 | { |
| 127 | unsigned long i, pfn; |
| 128 | int ret = 1; |
| 129 | struct memory_block *mem = to_memory_block(dev); |
| 130 | |
| 131 | if (mem->state != MEM_ONLINE) |
| 132 | goto out; |
| 133 | |
| 134 | for (i = 0; i < sections_per_block; i++) { |
| 135 | if (!present_section_nr(mem->start_section_nr + i)) |
| 136 | continue; |
| 137 | pfn = section_nr_to_pfn(mem->start_section_nr + i); |
| 138 | ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION); |
| 139 | } |
| 140 | |
| 141 | out: |
| 142 | return sprintf(buf, "%d\n", ret); |
| 143 | } |
| 144 | |
| 145 | /* |
| 146 | * online, offline, going offline, etc. |
| 147 | */ |
| 148 | static ssize_t show_mem_state(struct device *dev, |
| 149 | struct device_attribute *attr, char *buf) |
| 150 | { |
| 151 | struct memory_block *mem = to_memory_block(dev); |
| 152 | ssize_t len = 0; |
| 153 | |
| 154 | /* |
| 155 | * We can probably put these states in a nice little array |
| 156 | * so that they're not open-coded |
| 157 | */ |
| 158 | switch (mem->state) { |
| 159 | case MEM_ONLINE: |
| 160 | len = sprintf(buf, "online\n"); |
| 161 | break; |
| 162 | case MEM_OFFLINE: |
| 163 | len = sprintf(buf, "offline\n"); |
| 164 | break; |
| 165 | case MEM_GOING_OFFLINE: |
| 166 | len = sprintf(buf, "going-offline\n"); |
| 167 | break; |
| 168 | default: |
| 169 | len = sprintf(buf, "ERROR-UNKNOWN-%ld\n", |
| 170 | mem->state); |
| 171 | WARN_ON(1); |
| 172 | break; |
| 173 | } |
| 174 | |
| 175 | return len; |
| 176 | } |
| 177 | |
| 178 | int memory_notify(unsigned long val, void *v) |
| 179 | { |
| 180 | return blocking_notifier_call_chain(&memory_chain, val, v); |
| 181 | } |
| 182 | |
| 183 | int memory_isolate_notify(unsigned long val, void *v) |
| 184 | { |
| 185 | return atomic_notifier_call_chain(&memory_isolate_chain, val, v); |
| 186 | } |
| 187 | |
| 188 | /* |
| 189 | * The probe routines leave the pages reserved, just as the bootmem code does. |
| 190 | * Make sure they're still that way. |
| 191 | */ |
| 192 | static bool pages_correctly_reserved(unsigned long start_pfn) |
| 193 | { |
| 194 | int i, j; |
| 195 | struct page *page; |
| 196 | unsigned long pfn = start_pfn; |
| 197 | |
| 198 | /* |
| 199 | * memmap between sections is not contiguous except with |
| 200 | * SPARSEMEM_VMEMMAP. We lookup the page once per section |
| 201 | * and assume memmap is contiguous within each section |
| 202 | */ |
| 203 | for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) { |
| 204 | if (WARN_ON_ONCE(!pfn_valid(pfn))) |
| 205 | return false; |
| 206 | page = pfn_to_page(pfn); |
| 207 | |
| 208 | for (j = 0; j < PAGES_PER_SECTION; j++) { |
| 209 | if (PageReserved(page + j)) |
| 210 | continue; |
| 211 | |
| 212 | printk(KERN_WARNING "section number %ld page number %d " |
| 213 | "not reserved, was it already online?\n", |
| 214 | pfn_to_section_nr(pfn), j); |
| 215 | |
| 216 | return false; |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | return true; |
| 221 | } |
| 222 | |
| 223 | /* |
| 224 | * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is |
| 225 | * OK to have direct references to sparsemem variables in here. |
| 226 | * Must already be protected by mem_hotplug_begin(). |
| 227 | */ |
| 228 | static int |
| 229 | memory_block_action(unsigned long phys_index, unsigned long action, int online_type) |
| 230 | { |
| 231 | unsigned long start_pfn; |
| 232 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
| 233 | int ret; |
| 234 | |
| 235 | start_pfn = section_nr_to_pfn(phys_index); |
| 236 | |
| 237 | switch (action) { |
| 238 | case MEM_ONLINE: |
| 239 | if (!pages_correctly_reserved(start_pfn)) |
| 240 | return -EBUSY; |
| 241 | |
| 242 | ret = online_pages(start_pfn, nr_pages, online_type); |
| 243 | break; |
| 244 | case MEM_OFFLINE: |
| 245 | ret = offline_pages(start_pfn, nr_pages); |
| 246 | break; |
| 247 | default: |
| 248 | WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: " |
| 249 | "%ld\n", __func__, phys_index, action, action); |
| 250 | ret = -EINVAL; |
| 251 | } |
| 252 | |
| 253 | return ret; |
| 254 | } |
| 255 | |
| 256 | static int memory_block_change_state(struct memory_block *mem, |
| 257 | unsigned long to_state, unsigned long from_state_req) |
| 258 | { |
| 259 | int ret = 0; |
| 260 | |
| 261 | if (mem->state != from_state_req) |
| 262 | return -EINVAL; |
| 263 | |
| 264 | if (to_state == MEM_OFFLINE) |
| 265 | mem->state = MEM_GOING_OFFLINE; |
| 266 | |
| 267 | ret = memory_block_action(mem->start_section_nr, to_state, |
| 268 | mem->online_type); |
| 269 | |
| 270 | mem->state = ret ? from_state_req : to_state; |
| 271 | |
| 272 | return ret; |
| 273 | } |
| 274 | |
| 275 | /* The device lock serializes operations on memory_subsys_[online|offline] */ |
| 276 | static int memory_subsys_online(struct device *dev) |
| 277 | { |
| 278 | struct memory_block *mem = to_memory_block(dev); |
| 279 | int ret; |
| 280 | |
| 281 | if (mem->state == MEM_ONLINE) |
| 282 | return 0; |
| 283 | |
| 284 | /* |
| 285 | * If we are called from store_mem_state(), online_type will be |
| 286 | * set >= 0 Otherwise we were called from the device online |
| 287 | * attribute and need to set the online_type. |
| 288 | */ |
| 289 | if (mem->online_type < 0) |
| 290 | mem->online_type = MMOP_ONLINE_KEEP; |
| 291 | |
| 292 | /* Already under protection of mem_hotplug_begin() */ |
| 293 | ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE); |
| 294 | |
| 295 | /* clear online_type */ |
| 296 | mem->online_type = -1; |
| 297 | |
| 298 | return ret; |
| 299 | } |
| 300 | |
| 301 | static int memory_subsys_offline(struct device *dev) |
| 302 | { |
| 303 | struct memory_block *mem = to_memory_block(dev); |
| 304 | |
| 305 | if (mem->state == MEM_OFFLINE) |
| 306 | return 0; |
| 307 | |
| 308 | /* Can't offline block with non-present sections */ |
| 309 | if (mem->section_count != sections_per_block) |
| 310 | return -EINVAL; |
| 311 | |
| 312 | return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE); |
| 313 | } |
| 314 | |
| 315 | static ssize_t |
| 316 | store_mem_state(struct device *dev, |
| 317 | struct device_attribute *attr, const char *buf, size_t count) |
| 318 | { |
| 319 | struct memory_block *mem = to_memory_block(dev); |
| 320 | int ret, online_type; |
| 321 | |
| 322 | ret = lock_device_hotplug_sysfs(); |
| 323 | if (ret) |
| 324 | return ret; |
| 325 | |
| 326 | if (sysfs_streq(buf, "online_kernel")) |
| 327 | online_type = MMOP_ONLINE_KERNEL; |
| 328 | else if (sysfs_streq(buf, "online_movable")) |
| 329 | online_type = MMOP_ONLINE_MOVABLE; |
| 330 | else if (sysfs_streq(buf, "online")) |
| 331 | online_type = MMOP_ONLINE_KEEP; |
| 332 | else if (sysfs_streq(buf, "offline")) |
| 333 | online_type = MMOP_OFFLINE; |
| 334 | else { |
| 335 | ret = -EINVAL; |
| 336 | goto err; |
| 337 | } |
| 338 | |
| 339 | /* |
| 340 | * Memory hotplug needs to hold mem_hotplug_begin() for probe to find |
| 341 | * the correct memory block to online before doing device_online(dev), |
| 342 | * which will take dev->mutex. Take the lock early to prevent an |
| 343 | * inversion, memory_subsys_online() callbacks will be implemented by |
| 344 | * assuming it's already protected. |
| 345 | */ |
| 346 | mem_hotplug_begin(); |
| 347 | |
| 348 | switch (online_type) { |
| 349 | case MMOP_ONLINE_KERNEL: |
| 350 | case MMOP_ONLINE_MOVABLE: |
| 351 | case MMOP_ONLINE_KEEP: |
| 352 | mem->online_type = online_type; |
| 353 | ret = device_online(&mem->dev); |
| 354 | break; |
| 355 | case MMOP_OFFLINE: |
| 356 | ret = device_offline(&mem->dev); |
| 357 | break; |
| 358 | default: |
| 359 | ret = -EINVAL; /* should never happen */ |
| 360 | } |
| 361 | |
| 362 | mem_hotplug_done(); |
| 363 | err: |
| 364 | unlock_device_hotplug(); |
| 365 | |
| 366 | if (ret < 0) |
| 367 | return ret; |
| 368 | if (ret) |
| 369 | return -EINVAL; |
| 370 | |
| 371 | return count; |
| 372 | } |
| 373 | |
| 374 | /* |
| 375 | * phys_device is a bad name for this. What I really want |
| 376 | * is a way to differentiate between memory ranges that |
| 377 | * are part of physical devices that constitute |
| 378 | * a complete removable unit or fru. |
| 379 | * i.e. do these ranges belong to the same physical device, |
| 380 | * s.t. if I offline all of these sections I can then |
| 381 | * remove the physical device? |
| 382 | */ |
| 383 | static ssize_t show_phys_device(struct device *dev, |
| 384 | struct device_attribute *attr, char *buf) |
| 385 | { |
| 386 | struct memory_block *mem = to_memory_block(dev); |
| 387 | return sprintf(buf, "%d\n", mem->phys_device); |
| 388 | } |
| 389 | |
| 390 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 391 | static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn, |
| 392 | unsigned long nr_pages, int online_type, |
| 393 | struct zone *default_zone) |
| 394 | { |
| 395 | struct zone *zone; |
| 396 | |
| 397 | zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages); |
| 398 | if (zone != default_zone) { |
| 399 | strcat(buf, " "); |
| 400 | strcat(buf, zone->name); |
| 401 | } |
| 402 | } |
| 403 | |
| 404 | static ssize_t show_valid_zones(struct device *dev, |
| 405 | struct device_attribute *attr, char *buf) |
| 406 | { |
| 407 | struct memory_block *mem = to_memory_block(dev); |
| 408 | unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr); |
| 409 | unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block; |
| 410 | unsigned long valid_start_pfn, valid_end_pfn; |
| 411 | struct zone *default_zone; |
| 412 | int nid; |
| 413 | |
| 414 | /* |
| 415 | * The block contains more than one zone can not be offlined. |
| 416 | * This can happen e.g. for ZONE_DMA and ZONE_DMA32 |
| 417 | */ |
| 418 | if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages, &valid_start_pfn, &valid_end_pfn)) |
| 419 | return sprintf(buf, "none\n"); |
| 420 | |
| 421 | start_pfn = valid_start_pfn; |
| 422 | nr_pages = valid_end_pfn - start_pfn; |
| 423 | |
| 424 | /* |
| 425 | * Check the existing zone. Make sure that we do that only on the |
| 426 | * online nodes otherwise the page_zone is not reliable |
| 427 | */ |
| 428 | if (mem->state == MEM_ONLINE) { |
| 429 | strcat(buf, page_zone(pfn_to_page(start_pfn))->name); |
| 430 | goto out; |
| 431 | } |
| 432 | |
| 433 | nid = pfn_to_nid(start_pfn); |
| 434 | default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages); |
| 435 | strcat(buf, default_zone->name); |
| 436 | |
| 437 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL, |
| 438 | default_zone); |
| 439 | print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE, |
| 440 | default_zone); |
| 441 | out: |
| 442 | strcat(buf, "\n"); |
| 443 | |
| 444 | return strlen(buf); |
| 445 | } |
| 446 | static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL); |
| 447 | #endif |
| 448 | |
| 449 | static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL); |
| 450 | static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state); |
| 451 | static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL); |
| 452 | static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL); |
| 453 | |
| 454 | /* |
| 455 | * Block size attribute stuff |
| 456 | */ |
| 457 | static ssize_t |
| 458 | print_block_size(struct device *dev, struct device_attribute *attr, |
| 459 | char *buf) |
| 460 | { |
| 461 | return sprintf(buf, "%lx\n", get_memory_block_size()); |
| 462 | } |
| 463 | |
| 464 | static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL); |
| 465 | |
| 466 | /* |
| 467 | * Memory auto online policy. |
| 468 | */ |
| 469 | |
| 470 | static ssize_t |
| 471 | show_auto_online_blocks(struct device *dev, struct device_attribute *attr, |
| 472 | char *buf) |
| 473 | { |
| 474 | if (memhp_auto_online) |
| 475 | return sprintf(buf, "online\n"); |
| 476 | else |
| 477 | return sprintf(buf, "offline\n"); |
| 478 | } |
| 479 | |
| 480 | static ssize_t |
| 481 | store_auto_online_blocks(struct device *dev, struct device_attribute *attr, |
| 482 | const char *buf, size_t count) |
| 483 | { |
| 484 | if (sysfs_streq(buf, "online")) |
| 485 | memhp_auto_online = true; |
| 486 | else if (sysfs_streq(buf, "offline")) |
| 487 | memhp_auto_online = false; |
| 488 | else |
| 489 | return -EINVAL; |
| 490 | |
| 491 | return count; |
| 492 | } |
| 493 | |
| 494 | static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks, |
| 495 | store_auto_online_blocks); |
| 496 | |
| 497 | /* |
| 498 | * Some architectures will have custom drivers to do this, and |
| 499 | * will not need to do it from userspace. The fake hot-add code |
| 500 | * as well as ppc64 will do all of their discovery in userspace |
| 501 | * and will require this interface. |
| 502 | */ |
| 503 | #ifdef CONFIG_ARCH_MEMORY_PROBE |
| 504 | static ssize_t |
| 505 | memory_probe_store(struct device *dev, struct device_attribute *attr, |
| 506 | const char *buf, size_t count) |
| 507 | { |
| 508 | u64 phys_addr; |
| 509 | int nid, ret; |
| 510 | unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block; |
| 511 | |
| 512 | ret = kstrtoull(buf, 0, &phys_addr); |
| 513 | if (ret) |
| 514 | return ret; |
| 515 | |
| 516 | if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1)) |
| 517 | return -EINVAL; |
| 518 | |
| 519 | nid = memory_add_physaddr_to_nid(phys_addr); |
| 520 | ret = add_memory(nid, phys_addr, |
| 521 | MIN_MEMORY_BLOCK_SIZE * sections_per_block); |
| 522 | |
| 523 | if (ret) |
| 524 | goto out; |
| 525 | |
| 526 | ret = count; |
| 527 | out: |
| 528 | return ret; |
| 529 | } |
| 530 | |
| 531 | static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store); |
| 532 | #endif |
| 533 | |
| 534 | #ifdef CONFIG_MEMORY_FAILURE |
| 535 | /* |
| 536 | * Support for offlining pages of memory |
| 537 | */ |
| 538 | |
| 539 | /* Soft offline a page */ |
| 540 | static ssize_t |
| 541 | store_soft_offline_page(struct device *dev, |
| 542 | struct device_attribute *attr, |
| 543 | const char *buf, size_t count) |
| 544 | { |
| 545 | int ret; |
| 546 | u64 pfn; |
| 547 | if (!capable(CAP_SYS_ADMIN)) |
| 548 | return -EPERM; |
| 549 | if (kstrtoull(buf, 0, &pfn) < 0) |
| 550 | return -EINVAL; |
| 551 | pfn >>= PAGE_SHIFT; |
| 552 | if (!pfn_valid(pfn)) |
| 553 | return -ENXIO; |
| 554 | ret = soft_offline_page(pfn_to_page(pfn), 0); |
| 555 | return ret == 0 ? count : ret; |
| 556 | } |
| 557 | |
| 558 | /* Forcibly offline a page, including killing processes. */ |
| 559 | static ssize_t |
| 560 | store_hard_offline_page(struct device *dev, |
| 561 | struct device_attribute *attr, |
| 562 | const char *buf, size_t count) |
| 563 | { |
| 564 | int ret; |
| 565 | u64 pfn; |
| 566 | if (!capable(CAP_SYS_ADMIN)) |
| 567 | return -EPERM; |
| 568 | if (kstrtoull(buf, 0, &pfn) < 0) |
| 569 | return -EINVAL; |
| 570 | pfn >>= PAGE_SHIFT; |
| 571 | ret = memory_failure(pfn, 0, 0); |
| 572 | return ret ? ret : count; |
| 573 | } |
| 574 | |
| 575 | static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page); |
| 576 | static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page); |
| 577 | #endif |
| 578 | |
| 579 | /* |
| 580 | * Note that phys_device is optional. It is here to allow for |
| 581 | * differentiation between which *physical* devices each |
| 582 | * section belongs to... |
| 583 | */ |
| 584 | int __weak arch_get_memory_phys_device(unsigned long start_pfn) |
| 585 | { |
| 586 | return 0; |
| 587 | } |
| 588 | |
| 589 | /* |
| 590 | * A reference for the returned object is held and the reference for the |
| 591 | * hinted object is released. |
| 592 | */ |
| 593 | struct memory_block *find_memory_block_hinted(struct mem_section *section, |
| 594 | struct memory_block *hint) |
| 595 | { |
| 596 | int block_id = base_memory_block_id(__section_nr(section)); |
| 597 | struct device *hintdev = hint ? &hint->dev : NULL; |
| 598 | struct device *dev; |
| 599 | |
| 600 | dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev); |
| 601 | if (hint) |
| 602 | put_device(&hint->dev); |
| 603 | if (!dev) |
| 604 | return NULL; |
| 605 | return to_memory_block(dev); |
| 606 | } |
| 607 | |
| 608 | /* |
| 609 | * For now, we have a linear search to go find the appropriate |
| 610 | * memory_block corresponding to a particular phys_index. If |
| 611 | * this gets to be a real problem, we can always use a radix |
| 612 | * tree or something here. |
| 613 | * |
| 614 | * This could be made generic for all device subsystems. |
| 615 | */ |
| 616 | struct memory_block *find_memory_block(struct mem_section *section) |
| 617 | { |
| 618 | return find_memory_block_hinted(section, NULL); |
| 619 | } |
| 620 | |
| 621 | static struct attribute *memory_memblk_attrs[] = { |
| 622 | &dev_attr_phys_index.attr, |
| 623 | &dev_attr_state.attr, |
| 624 | &dev_attr_phys_device.attr, |
| 625 | &dev_attr_removable.attr, |
| 626 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 627 | &dev_attr_valid_zones.attr, |
| 628 | #endif |
| 629 | NULL |
| 630 | }; |
| 631 | |
| 632 | static struct attribute_group memory_memblk_attr_group = { |
| 633 | .attrs = memory_memblk_attrs, |
| 634 | }; |
| 635 | |
| 636 | static const struct attribute_group *memory_memblk_attr_groups[] = { |
| 637 | &memory_memblk_attr_group, |
| 638 | NULL, |
| 639 | }; |
| 640 | |
| 641 | /* |
| 642 | * register_memory - Setup a sysfs device for a memory block |
| 643 | */ |
| 644 | static |
| 645 | int register_memory(struct memory_block *memory) |
| 646 | { |
| 647 | memory->dev.bus = &memory_subsys; |
| 648 | memory->dev.id = memory->start_section_nr / sections_per_block; |
| 649 | memory->dev.release = memory_block_release; |
| 650 | memory->dev.groups = memory_memblk_attr_groups; |
| 651 | memory->dev.offline = memory->state == MEM_OFFLINE; |
| 652 | |
| 653 | return device_register(&memory->dev); |
| 654 | } |
| 655 | |
| 656 | static int init_memory_block(struct memory_block **memory, |
| 657 | struct mem_section *section, unsigned long state) |
| 658 | { |
| 659 | struct memory_block *mem; |
| 660 | unsigned long start_pfn; |
| 661 | int scn_nr; |
| 662 | int ret = 0; |
| 663 | |
| 664 | mem = kzalloc(sizeof(*mem), GFP_KERNEL); |
| 665 | if (!mem) |
| 666 | return -ENOMEM; |
| 667 | |
| 668 | scn_nr = __section_nr(section); |
| 669 | mem->start_section_nr = |
| 670 | base_memory_block_id(scn_nr) * sections_per_block; |
| 671 | mem->end_section_nr = mem->start_section_nr + sections_per_block - 1; |
| 672 | mem->state = state; |
| 673 | start_pfn = section_nr_to_pfn(mem->start_section_nr); |
| 674 | mem->phys_device = arch_get_memory_phys_device(start_pfn); |
| 675 | |
| 676 | ret = register_memory(mem); |
| 677 | |
| 678 | *memory = mem; |
| 679 | return ret; |
| 680 | } |
| 681 | |
| 682 | static int add_memory_block(int base_section_nr) |
| 683 | { |
| 684 | struct memory_block *mem; |
| 685 | int i, ret, section_count = 0, section_nr; |
| 686 | |
| 687 | for (i = base_section_nr; |
| 688 | (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS; |
| 689 | i++) { |
| 690 | if (!present_section_nr(i)) |
| 691 | continue; |
| 692 | if (section_count == 0) |
| 693 | section_nr = i; |
| 694 | section_count++; |
| 695 | } |
| 696 | |
| 697 | if (section_count == 0) |
| 698 | return 0; |
| 699 | ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE); |
| 700 | if (ret) |
| 701 | return ret; |
| 702 | mem->section_count = section_count; |
| 703 | return 0; |
| 704 | } |
| 705 | |
| 706 | /* |
| 707 | * need an interface for the VM to add new memory regions, |
| 708 | * but without onlining it. |
| 709 | */ |
| 710 | int register_new_memory(int nid, struct mem_section *section) |
| 711 | { |
| 712 | int ret = 0; |
| 713 | struct memory_block *mem; |
| 714 | |
| 715 | mutex_lock(&mem_sysfs_mutex); |
| 716 | |
| 717 | mem = find_memory_block(section); |
| 718 | if (mem) { |
| 719 | mem->section_count++; |
| 720 | put_device(&mem->dev); |
| 721 | } else { |
| 722 | ret = init_memory_block(&mem, section, MEM_OFFLINE); |
| 723 | if (ret) |
| 724 | goto out; |
| 725 | mem->section_count++; |
| 726 | } |
| 727 | |
| 728 | if (mem->section_count == sections_per_block) |
| 729 | ret = register_mem_sect_under_node(mem, nid); |
| 730 | out: |
| 731 | mutex_unlock(&mem_sysfs_mutex); |
| 732 | return ret; |
| 733 | } |
| 734 | |
| 735 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 736 | static void |
| 737 | unregister_memory(struct memory_block *memory) |
| 738 | { |
| 739 | BUG_ON(memory->dev.bus != &memory_subsys); |
| 740 | |
| 741 | /* drop the ref. we got in remove_memory_block() */ |
| 742 | put_device(&memory->dev); |
| 743 | device_unregister(&memory->dev); |
| 744 | } |
| 745 | |
| 746 | static int remove_memory_section(unsigned long node_id, |
| 747 | struct mem_section *section, int phys_device) |
| 748 | { |
| 749 | struct memory_block *mem; |
| 750 | |
| 751 | mutex_lock(&mem_sysfs_mutex); |
| 752 | |
| 753 | /* |
| 754 | * Some users of the memory hotplug do not want/need memblock to |
| 755 | * track all sections. Skip over those. |
| 756 | */ |
| 757 | mem = find_memory_block(section); |
| 758 | if (!mem) |
| 759 | goto out_unlock; |
| 760 | |
| 761 | unregister_mem_sect_under_nodes(mem, __section_nr(section)); |
| 762 | |
| 763 | mem->section_count--; |
| 764 | if (mem->section_count == 0) |
| 765 | unregister_memory(mem); |
| 766 | else |
| 767 | put_device(&mem->dev); |
| 768 | |
| 769 | out_unlock: |
| 770 | mutex_unlock(&mem_sysfs_mutex); |
| 771 | return 0; |
| 772 | } |
| 773 | |
| 774 | int unregister_memory_section(struct mem_section *section) |
| 775 | { |
| 776 | if (!present_section(section)) |
| 777 | return -EINVAL; |
| 778 | |
| 779 | return remove_memory_section(0, section, 0); |
| 780 | } |
| 781 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
| 782 | |
| 783 | /* return true if the memory block is offlined, otherwise, return false */ |
| 784 | bool is_memblock_offlined(struct memory_block *mem) |
| 785 | { |
| 786 | return mem->state == MEM_OFFLINE; |
| 787 | } |
| 788 | |
| 789 | static struct attribute *memory_root_attrs[] = { |
| 790 | #ifdef CONFIG_ARCH_MEMORY_PROBE |
| 791 | &dev_attr_probe.attr, |
| 792 | #endif |
| 793 | |
| 794 | #ifdef CONFIG_MEMORY_FAILURE |
| 795 | &dev_attr_soft_offline_page.attr, |
| 796 | &dev_attr_hard_offline_page.attr, |
| 797 | #endif |
| 798 | |
| 799 | &dev_attr_block_size_bytes.attr, |
| 800 | &dev_attr_auto_online_blocks.attr, |
| 801 | NULL |
| 802 | }; |
| 803 | |
| 804 | static struct attribute_group memory_root_attr_group = { |
| 805 | .attrs = memory_root_attrs, |
| 806 | }; |
| 807 | |
| 808 | static const struct attribute_group *memory_root_attr_groups[] = { |
| 809 | &memory_root_attr_group, |
| 810 | NULL, |
| 811 | }; |
| 812 | |
| 813 | /* |
| 814 | * Initialize the sysfs support for memory devices... |
| 815 | */ |
| 816 | int __init memory_dev_init(void) |
| 817 | { |
| 818 | unsigned int i; |
| 819 | int ret; |
| 820 | int err; |
| 821 | unsigned long block_sz; |
| 822 | |
| 823 | ret = subsys_system_register(&memory_subsys, memory_root_attr_groups); |
| 824 | if (ret) |
| 825 | goto out; |
| 826 | |
| 827 | block_sz = get_memory_block_size(); |
| 828 | sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE; |
| 829 | |
| 830 | /* |
| 831 | * Create entries for memory sections that were found |
| 832 | * during boot and have been initialized |
| 833 | */ |
| 834 | mutex_lock(&mem_sysfs_mutex); |
| 835 | for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) { |
| 836 | /* Don't iterate over sections we know are !present: */ |
| 837 | if (i > __highest_present_section_nr) |
| 838 | break; |
| 839 | |
| 840 | err = add_memory_block(i); |
| 841 | if (!ret) |
| 842 | ret = err; |
| 843 | } |
| 844 | mutex_unlock(&mem_sysfs_mutex); |
| 845 | |
| 846 | out: |
| 847 | if (ret) |
| 848 | printk(KERN_ERR "%s() failed: %d\n", __func__, ret); |
| 849 | return ret; |
| 850 | } |