1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Basic Node interface support 4 */ 5 6 #include <linux/module.h> 7 #include <linux/init.h> 8 #include <linux/mm.h> 9 #include <linux/memory.h> 10 #include <linux/vmstat.h> 11 #include <linux/notifier.h> 12 #include <linux/node.h> 13 #include <linux/hugetlb.h> 14 #include <linux/compaction.h> 15 #include <linux/cpumask.h> 16 #include <linux/topology.h> 17 #include <linux/nodemask.h> 18 #include <linux/cpu.h> 19 #include <linux/device.h> 20 #include <linux/pm_runtime.h> 21 #include <linux/swap.h> 22 #include <linux/slab.h> 23 24 static struct bus_type node_subsys = { 25 .name = "node", 26 .dev_name = "node", 27 }; 28 29 static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj, 30 struct bin_attribute *attr, char *buf, 31 loff_t off, size_t count) 32 { 33 struct device *dev = kobj_to_dev(kobj); 34 struct node *node_dev = to_node(dev); 35 cpumask_var_t mask; 36 ssize_t n; 37 38 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 39 return 0; 40 41 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask); 42 n = cpumap_print_bitmask_to_buf(buf, mask, off, count); 43 free_cpumask_var(mask); 44 45 return n; 46 } 47 48 static BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES); 49 50 static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj, 51 struct bin_attribute *attr, char *buf, 52 loff_t off, size_t count) 53 { 54 struct device *dev = kobj_to_dev(kobj); 55 struct node *node_dev = to_node(dev); 56 cpumask_var_t mask; 57 ssize_t n; 58 59 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 60 return 0; 61 62 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask); 63 n = cpumap_print_list_to_buf(buf, mask, off, count); 64 free_cpumask_var(mask); 65 66 return n; 67 } 68 69 static BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES); 70 71 /** 72 * struct node_access_nodes - Access class device to hold user visible 73 * relationships to other nodes. 74 * @dev: Device for this memory access class 75 * @list_node: List element in the node's access list 76 * @access: The access class rank 77 * @hmem_attrs: Heterogeneous memory performance attributes 78 */ 79 struct node_access_nodes { 80 struct device dev; 81 struct list_head list_node; 82 unsigned int access; 83 #ifdef CONFIG_HMEM_REPORTING 84 struct node_hmem_attrs hmem_attrs; 85 #endif 86 }; 87 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev) 88 89 static struct attribute *node_init_access_node_attrs[] = { 90 NULL, 91 }; 92 93 static struct attribute *node_targ_access_node_attrs[] = { 94 NULL, 95 }; 96 97 static const struct attribute_group initiators = { 98 .name = "initiators", 99 .attrs = node_init_access_node_attrs, 100 }; 101 102 static const struct attribute_group targets = { 103 .name = "targets", 104 .attrs = node_targ_access_node_attrs, 105 }; 106 107 static const struct attribute_group *node_access_node_groups[] = { 108 &initiators, 109 &targets, 110 NULL, 111 }; 112 113 static void node_remove_accesses(struct node *node) 114 { 115 struct node_access_nodes *c, *cnext; 116 117 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) { 118 list_del(&c->list_node); 119 device_unregister(&c->dev); 120 } 121 } 122 123 static void node_access_release(struct device *dev) 124 { 125 kfree(to_access_nodes(dev)); 126 } 127 128 static struct node_access_nodes *node_init_node_access(struct node *node, 129 unsigned int access) 130 { 131 struct node_access_nodes *access_node; 132 struct device *dev; 133 134 list_for_each_entry(access_node, &node->access_list, list_node) 135 if (access_node->access == access) 136 return access_node; 137 138 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL); 139 if (!access_node) 140 return NULL; 141 142 access_node->access = access; 143 dev = &access_node->dev; 144 dev->parent = &node->dev; 145 dev->release = node_access_release; 146 dev->groups = node_access_node_groups; 147 if (dev_set_name(dev, "access%u", access)) 148 goto free; 149 150 if (device_register(dev)) 151 goto free_name; 152 153 pm_runtime_no_callbacks(dev); 154 list_add_tail(&access_node->list_node, &node->access_list); 155 return access_node; 156 free_name: 157 kfree_const(dev->kobj.name); 158 free: 159 kfree(access_node); 160 return NULL; 161 } 162 163 #ifdef CONFIG_HMEM_REPORTING 164 #define ACCESS_ATTR(name) \ 165 static ssize_t name##_show(struct device *dev, \ 166 struct device_attribute *attr, \ 167 char *buf) \ 168 { \ 169 return sysfs_emit(buf, "%u\n", \ 170 to_access_nodes(dev)->hmem_attrs.name); \ 171 } \ 172 static DEVICE_ATTR_RO(name) 173 174 ACCESS_ATTR(read_bandwidth); 175 ACCESS_ATTR(read_latency); 176 ACCESS_ATTR(write_bandwidth); 177 ACCESS_ATTR(write_latency); 178 179 static struct attribute *access_attrs[] = { 180 &dev_attr_read_bandwidth.attr, 181 &dev_attr_read_latency.attr, 182 &dev_attr_write_bandwidth.attr, 183 &dev_attr_write_latency.attr, 184 NULL, 185 }; 186 187 /** 188 * node_set_perf_attrs - Set the performance values for given access class 189 * @nid: Node identifier to be set 190 * @hmem_attrs: Heterogeneous memory performance attributes 191 * @access: The access class the for the given attributes 192 */ 193 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs, 194 unsigned int access) 195 { 196 struct node_access_nodes *c; 197 struct node *node; 198 int i; 199 200 if (WARN_ON_ONCE(!node_online(nid))) 201 return; 202 203 node = node_devices[nid]; 204 c = node_init_node_access(node, access); 205 if (!c) 206 return; 207 208 c->hmem_attrs = *hmem_attrs; 209 for (i = 0; access_attrs[i] != NULL; i++) { 210 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i], 211 "initiators")) { 212 pr_info("failed to add performance attribute to node %d\n", 213 nid); 214 break; 215 } 216 } 217 } 218 219 /** 220 * struct node_cache_info - Internal tracking for memory node caches 221 * @dev: Device represeting the cache level 222 * @node: List element for tracking in the node 223 * @cache_attrs:Attributes for this cache level 224 */ 225 struct node_cache_info { 226 struct device dev; 227 struct list_head node; 228 struct node_cache_attrs cache_attrs; 229 }; 230 #define to_cache_info(device) container_of(device, struct node_cache_info, dev) 231 232 #define CACHE_ATTR(name, fmt) \ 233 static ssize_t name##_show(struct device *dev, \ 234 struct device_attribute *attr, \ 235 char *buf) \ 236 { \ 237 return sysfs_emit(buf, fmt "\n", \ 238 to_cache_info(dev)->cache_attrs.name); \ 239 } \ 240 static DEVICE_ATTR_RO(name); 241 242 CACHE_ATTR(size, "%llu") 243 CACHE_ATTR(line_size, "%u") 244 CACHE_ATTR(indexing, "%u") 245 CACHE_ATTR(write_policy, "%u") 246 247 static struct attribute *cache_attrs[] = { 248 &dev_attr_indexing.attr, 249 &dev_attr_size.attr, 250 &dev_attr_line_size.attr, 251 &dev_attr_write_policy.attr, 252 NULL, 253 }; 254 ATTRIBUTE_GROUPS(cache); 255 256 static void node_cache_release(struct device *dev) 257 { 258 kfree(dev); 259 } 260 261 static void node_cacheinfo_release(struct device *dev) 262 { 263 struct node_cache_info *info = to_cache_info(dev); 264 kfree(info); 265 } 266 267 static void node_init_cache_dev(struct node *node) 268 { 269 struct device *dev; 270 271 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 272 if (!dev) 273 return; 274 275 device_initialize(dev); 276 dev->parent = &node->dev; 277 dev->release = node_cache_release; 278 if (dev_set_name(dev, "memory_side_cache")) 279 goto put_device; 280 281 if (device_add(dev)) 282 goto put_device; 283 284 pm_runtime_no_callbacks(dev); 285 node->cache_dev = dev; 286 return; 287 put_device: 288 put_device(dev); 289 } 290 291 /** 292 * node_add_cache() - add cache attribute to a memory node 293 * @nid: Node identifier that has new cache attributes 294 * @cache_attrs: Attributes for the cache being added 295 */ 296 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs) 297 { 298 struct node_cache_info *info; 299 struct device *dev; 300 struct node *node; 301 302 if (!node_online(nid) || !node_devices[nid]) 303 return; 304 305 node = node_devices[nid]; 306 list_for_each_entry(info, &node->cache_attrs, node) { 307 if (info->cache_attrs.level == cache_attrs->level) { 308 dev_warn(&node->dev, 309 "attempt to add duplicate cache level:%d\n", 310 cache_attrs->level); 311 return; 312 } 313 } 314 315 if (!node->cache_dev) 316 node_init_cache_dev(node); 317 if (!node->cache_dev) 318 return; 319 320 info = kzalloc(sizeof(*info), GFP_KERNEL); 321 if (!info) 322 return; 323 324 dev = &info->dev; 325 device_initialize(dev); 326 dev->parent = node->cache_dev; 327 dev->release = node_cacheinfo_release; 328 dev->groups = cache_groups; 329 if (dev_set_name(dev, "index%d", cache_attrs->level)) 330 goto put_device; 331 332 info->cache_attrs = *cache_attrs; 333 if (device_add(dev)) { 334 dev_warn(&node->dev, "failed to add cache level:%d\n", 335 cache_attrs->level); 336 goto put_device; 337 } 338 pm_runtime_no_callbacks(dev); 339 list_add_tail(&info->node, &node->cache_attrs); 340 return; 341 put_device: 342 put_device(dev); 343 } 344 345 static void node_remove_caches(struct node *node) 346 { 347 struct node_cache_info *info, *next; 348 349 if (!node->cache_dev) 350 return; 351 352 list_for_each_entry_safe(info, next, &node->cache_attrs, node) { 353 list_del(&info->node); 354 device_unregister(&info->dev); 355 } 356 device_unregister(node->cache_dev); 357 } 358 359 static void node_init_caches(unsigned int nid) 360 { 361 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs); 362 } 363 #else 364 static void node_init_caches(unsigned int nid) { } 365 static void node_remove_caches(struct node *node) { } 366 #endif 367 368 #define K(x) ((x) << (PAGE_SHIFT - 10)) 369 static ssize_t node_read_meminfo(struct device *dev, 370 struct device_attribute *attr, char *buf) 371 { 372 int len = 0; 373 int nid = dev->id; 374 struct pglist_data *pgdat = NODE_DATA(nid); 375 struct sysinfo i; 376 unsigned long sreclaimable, sunreclaimable; 377 unsigned long swapcached = 0; 378 379 si_meminfo_node(&i, nid); 380 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B); 381 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B); 382 #ifdef CONFIG_SWAP 383 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE); 384 #endif 385 len = sysfs_emit_at(buf, len, 386 "Node %d MemTotal: %8lu kB\n" 387 "Node %d MemFree: %8lu kB\n" 388 "Node %d MemUsed: %8lu kB\n" 389 "Node %d SwapCached: %8lu kB\n" 390 "Node %d Active: %8lu kB\n" 391 "Node %d Inactive: %8lu kB\n" 392 "Node %d Active(anon): %8lu kB\n" 393 "Node %d Inactive(anon): %8lu kB\n" 394 "Node %d Active(file): %8lu kB\n" 395 "Node %d Inactive(file): %8lu kB\n" 396 "Node %d Unevictable: %8lu kB\n" 397 "Node %d Mlocked: %8lu kB\n", 398 nid, K(i.totalram), 399 nid, K(i.freeram), 400 nid, K(i.totalram - i.freeram), 401 nid, K(swapcached), 402 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) + 403 node_page_state(pgdat, NR_ACTIVE_FILE)), 404 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) + 405 node_page_state(pgdat, NR_INACTIVE_FILE)), 406 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)), 407 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)), 408 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)), 409 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)), 410 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)), 411 nid, K(sum_zone_node_page_state(nid, NR_MLOCK))); 412 413 #ifdef CONFIG_HIGHMEM 414 len += sysfs_emit_at(buf, len, 415 "Node %d HighTotal: %8lu kB\n" 416 "Node %d HighFree: %8lu kB\n" 417 "Node %d LowTotal: %8lu kB\n" 418 "Node %d LowFree: %8lu kB\n", 419 nid, K(i.totalhigh), 420 nid, K(i.freehigh), 421 nid, K(i.totalram - i.totalhigh), 422 nid, K(i.freeram - i.freehigh)); 423 #endif 424 len += sysfs_emit_at(buf, len, 425 "Node %d Dirty: %8lu kB\n" 426 "Node %d Writeback: %8lu kB\n" 427 "Node %d FilePages: %8lu kB\n" 428 "Node %d Mapped: %8lu kB\n" 429 "Node %d AnonPages: %8lu kB\n" 430 "Node %d Shmem: %8lu kB\n" 431 "Node %d KernelStack: %8lu kB\n" 432 #ifdef CONFIG_SHADOW_CALL_STACK 433 "Node %d ShadowCallStack:%8lu kB\n" 434 #endif 435 "Node %d PageTables: %8lu kB\n" 436 "Node %d SecPageTables: %8lu kB\n" 437 "Node %d NFS_Unstable: %8lu kB\n" 438 "Node %d Bounce: %8lu kB\n" 439 "Node %d WritebackTmp: %8lu kB\n" 440 "Node %d KReclaimable: %8lu kB\n" 441 "Node %d Slab: %8lu kB\n" 442 "Node %d SReclaimable: %8lu kB\n" 443 "Node %d SUnreclaim: %8lu kB\n" 444 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 445 "Node %d AnonHugePages: %8lu kB\n" 446 "Node %d ShmemHugePages: %8lu kB\n" 447 "Node %d ShmemPmdMapped: %8lu kB\n" 448 "Node %d FileHugePages: %8lu kB\n" 449 "Node %d FilePmdMapped: %8lu kB\n" 450 #endif 451 , 452 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)), 453 nid, K(node_page_state(pgdat, NR_WRITEBACK)), 454 nid, K(node_page_state(pgdat, NR_FILE_PAGES)), 455 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)), 456 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)), 457 nid, K(i.sharedram), 458 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB), 459 #ifdef CONFIG_SHADOW_CALL_STACK 460 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB), 461 #endif 462 nid, K(node_page_state(pgdat, NR_PAGETABLE)), 463 nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)), 464 nid, 0UL, 465 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)), 466 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), 467 nid, K(sreclaimable + 468 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)), 469 nid, K(sreclaimable + sunreclaimable), 470 nid, K(sreclaimable), 471 nid, K(sunreclaimable) 472 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 473 , 474 nid, K(node_page_state(pgdat, NR_ANON_THPS)), 475 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)), 476 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)), 477 nid, K(node_page_state(pgdat, NR_FILE_THPS)), 478 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED)) 479 #endif 480 ); 481 len += hugetlb_report_node_meminfo(buf, len, nid); 482 return len; 483 } 484 485 #undef K 486 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL); 487 488 static ssize_t node_read_numastat(struct device *dev, 489 struct device_attribute *attr, char *buf) 490 { 491 fold_vm_numa_events(); 492 return sysfs_emit(buf, 493 "numa_hit %lu\n" 494 "numa_miss %lu\n" 495 "numa_foreign %lu\n" 496 "interleave_hit %lu\n" 497 "local_node %lu\n" 498 "other_node %lu\n", 499 sum_zone_numa_event_state(dev->id, NUMA_HIT), 500 sum_zone_numa_event_state(dev->id, NUMA_MISS), 501 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN), 502 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT), 503 sum_zone_numa_event_state(dev->id, NUMA_LOCAL), 504 sum_zone_numa_event_state(dev->id, NUMA_OTHER)); 505 } 506 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL); 507 508 static ssize_t node_read_vmstat(struct device *dev, 509 struct device_attribute *attr, char *buf) 510 { 511 int nid = dev->id; 512 struct pglist_data *pgdat = NODE_DATA(nid); 513 int i; 514 int len = 0; 515 516 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) 517 len += sysfs_emit_at(buf, len, "%s %lu\n", 518 zone_stat_name(i), 519 sum_zone_node_page_state(nid, i)); 520 521 #ifdef CONFIG_NUMA 522 fold_vm_numa_events(); 523 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++) 524 len += sysfs_emit_at(buf, len, "%s %lu\n", 525 numa_stat_name(i), 526 sum_zone_numa_event_state(nid, i)); 527 528 #endif 529 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { 530 unsigned long pages = node_page_state_pages(pgdat, i); 531 532 if (vmstat_item_print_in_thp(i)) 533 pages /= HPAGE_PMD_NR; 534 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i), 535 pages); 536 } 537 538 return len; 539 } 540 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL); 541 542 static ssize_t node_read_distance(struct device *dev, 543 struct device_attribute *attr, char *buf) 544 { 545 int nid = dev->id; 546 int len = 0; 547 int i; 548 549 /* 550 * buf is currently PAGE_SIZE in length and each node needs 4 chars 551 * at the most (distance + space or newline). 552 */ 553 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE); 554 555 for_each_online_node(i) { 556 len += sysfs_emit_at(buf, len, "%s%d", 557 i ? " " : "", node_distance(nid, i)); 558 } 559 560 len += sysfs_emit_at(buf, len, "\n"); 561 return len; 562 } 563 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL); 564 565 static struct attribute *node_dev_attrs[] = { 566 &dev_attr_meminfo.attr, 567 &dev_attr_numastat.attr, 568 &dev_attr_distance.attr, 569 &dev_attr_vmstat.attr, 570 NULL 571 }; 572 573 static struct bin_attribute *node_dev_bin_attrs[] = { 574 &bin_attr_cpumap, 575 &bin_attr_cpulist, 576 NULL 577 }; 578 579 static const struct attribute_group node_dev_group = { 580 .attrs = node_dev_attrs, 581 .bin_attrs = node_dev_bin_attrs 582 }; 583 584 static const struct attribute_group *node_dev_groups[] = { 585 &node_dev_group, 586 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP 587 &arch_node_dev_group, 588 #endif 589 NULL 590 }; 591 592 #ifdef CONFIG_HUGETLBFS 593 /* 594 * hugetlbfs per node attributes registration interface: 595 * When/if hugetlb[fs] subsystem initializes [sometime after this module], 596 * it will register its per node attributes for all online nodes with 597 * memory. It will also call register_hugetlbfs_with_node(), below, to 598 * register its attribute registration functions with this node driver. 599 * Once these hooks have been initialized, the node driver will call into 600 * the hugetlb module to [un]register attributes for hot-plugged nodes. 601 */ 602 static node_registration_func_t __hugetlb_register_node; 603 static node_registration_func_t __hugetlb_unregister_node; 604 605 static inline bool hugetlb_register_node(struct node *node) 606 { 607 if (__hugetlb_register_node && 608 node_state(node->dev.id, N_MEMORY)) { 609 __hugetlb_register_node(node); 610 return true; 611 } 612 return false; 613 } 614 615 static inline void hugetlb_unregister_node(struct node *node) 616 { 617 if (__hugetlb_unregister_node) 618 __hugetlb_unregister_node(node); 619 } 620 621 void register_hugetlbfs_with_node(node_registration_func_t doregister, 622 node_registration_func_t unregister) 623 { 624 __hugetlb_register_node = doregister; 625 __hugetlb_unregister_node = unregister; 626 } 627 #else 628 static inline void hugetlb_register_node(struct node *node) {} 629 630 static inline void hugetlb_unregister_node(struct node *node) {} 631 #endif 632 633 static void node_device_release(struct device *dev) 634 { 635 struct node *node = to_node(dev); 636 637 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_HUGETLBFS) 638 /* 639 * We schedule the work only when a memory section is 640 * onlined/offlined on this node. When we come here, 641 * all the memory on this node has been offlined, 642 * so we won't enqueue new work to this work. 643 * 644 * The work is using node->node_work, so we should 645 * flush work before freeing the memory. 646 */ 647 flush_work(&node->node_work); 648 #endif 649 kfree(node); 650 } 651 652 /* 653 * register_node - Setup a sysfs device for a node. 654 * @num - Node number to use when creating the device. 655 * 656 * Initialize and register the node device. 657 */ 658 static int register_node(struct node *node, int num) 659 { 660 int error; 661 662 node->dev.id = num; 663 node->dev.bus = &node_subsys; 664 node->dev.release = node_device_release; 665 node->dev.groups = node_dev_groups; 666 error = device_register(&node->dev); 667 668 if (error) 669 put_device(&node->dev); 670 else { 671 hugetlb_register_node(node); 672 673 compaction_register_node(node); 674 } 675 return error; 676 } 677 678 /** 679 * unregister_node - unregister a node device 680 * @node: node going away 681 * 682 * Unregisters a node device @node. All the devices on the node must be 683 * unregistered before calling this function. 684 */ 685 void unregister_node(struct node *node) 686 { 687 compaction_unregister_node(node); 688 hugetlb_unregister_node(node); /* no-op, if memoryless node */ 689 node_remove_accesses(node); 690 node_remove_caches(node); 691 device_unregister(&node->dev); 692 } 693 694 struct node *node_devices[MAX_NUMNODES]; 695 696 /* 697 * register cpu under node 698 */ 699 int register_cpu_under_node(unsigned int cpu, unsigned int nid) 700 { 701 int ret; 702 struct device *obj; 703 704 if (!node_online(nid)) 705 return 0; 706 707 obj = get_cpu_device(cpu); 708 if (!obj) 709 return 0; 710 711 ret = sysfs_create_link(&node_devices[nid]->dev.kobj, 712 &obj->kobj, 713 kobject_name(&obj->kobj)); 714 if (ret) 715 return ret; 716 717 return sysfs_create_link(&obj->kobj, 718 &node_devices[nid]->dev.kobj, 719 kobject_name(&node_devices[nid]->dev.kobj)); 720 } 721 722 /** 723 * register_memory_node_under_compute_node - link memory node to its compute 724 * node for a given access class. 725 * @mem_nid: Memory node number 726 * @cpu_nid: Cpu node number 727 * @access: Access class to register 728 * 729 * Description: 730 * For use with platforms that may have separate memory and compute nodes. 731 * This function will export node relationships linking which memory 732 * initiator nodes can access memory targets at a given ranked access 733 * class. 734 */ 735 int register_memory_node_under_compute_node(unsigned int mem_nid, 736 unsigned int cpu_nid, 737 unsigned int access) 738 { 739 struct node *init_node, *targ_node; 740 struct node_access_nodes *initiator, *target; 741 int ret; 742 743 if (!node_online(cpu_nid) || !node_online(mem_nid)) 744 return -ENODEV; 745 746 init_node = node_devices[cpu_nid]; 747 targ_node = node_devices[mem_nid]; 748 initiator = node_init_node_access(init_node, access); 749 target = node_init_node_access(targ_node, access); 750 if (!initiator || !target) 751 return -ENOMEM; 752 753 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets", 754 &targ_node->dev.kobj, 755 dev_name(&targ_node->dev)); 756 if (ret) 757 return ret; 758 759 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators", 760 &init_node->dev.kobj, 761 dev_name(&init_node->dev)); 762 if (ret) 763 goto err; 764 765 return 0; 766 err: 767 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets", 768 dev_name(&targ_node->dev)); 769 return ret; 770 } 771 772 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid) 773 { 774 struct device *obj; 775 776 if (!node_online(nid)) 777 return 0; 778 779 obj = get_cpu_device(cpu); 780 if (!obj) 781 return 0; 782 783 sysfs_remove_link(&node_devices[nid]->dev.kobj, 784 kobject_name(&obj->kobj)); 785 sysfs_remove_link(&obj->kobj, 786 kobject_name(&node_devices[nid]->dev.kobj)); 787 788 return 0; 789 } 790 791 #ifdef CONFIG_MEMORY_HOTPLUG 792 static int __ref get_nid_for_pfn(unsigned long pfn) 793 { 794 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 795 if (system_state < SYSTEM_RUNNING) 796 return early_pfn_to_nid(pfn); 797 #endif 798 return pfn_to_nid(pfn); 799 } 800 801 static void do_register_memory_block_under_node(int nid, 802 struct memory_block *mem_blk, 803 enum meminit_context context) 804 { 805 int ret; 806 807 memory_block_add_nid(mem_blk, nid, context); 808 809 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj, 810 &mem_blk->dev.kobj, 811 kobject_name(&mem_blk->dev.kobj)); 812 if (ret && ret != -EEXIST) 813 dev_err_ratelimited(&node_devices[nid]->dev, 814 "can't create link to %s in sysfs (%d)\n", 815 kobject_name(&mem_blk->dev.kobj), ret); 816 817 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj, 818 &node_devices[nid]->dev.kobj, 819 kobject_name(&node_devices[nid]->dev.kobj)); 820 if (ret && ret != -EEXIST) 821 dev_err_ratelimited(&mem_blk->dev, 822 "can't create link to %s in sysfs (%d)\n", 823 kobject_name(&node_devices[nid]->dev.kobj), 824 ret); 825 } 826 827 /* register memory section under specified node if it spans that node */ 828 static int register_mem_block_under_node_early(struct memory_block *mem_blk, 829 void *arg) 830 { 831 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE; 832 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr); 833 unsigned long end_pfn = start_pfn + memory_block_pfns - 1; 834 int nid = *(int *)arg; 835 unsigned long pfn; 836 837 for (pfn = start_pfn; pfn <= end_pfn; pfn++) { 838 int page_nid; 839 840 /* 841 * memory block could have several absent sections from start. 842 * skip pfn range from absent section 843 */ 844 if (!pfn_in_present_section(pfn)) { 845 pfn = round_down(pfn + PAGES_PER_SECTION, 846 PAGES_PER_SECTION) - 1; 847 continue; 848 } 849 850 /* 851 * We need to check if page belongs to nid only at the boot 852 * case because node's ranges can be interleaved. 853 */ 854 page_nid = get_nid_for_pfn(pfn); 855 if (page_nid < 0) 856 continue; 857 if (page_nid != nid) 858 continue; 859 860 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY); 861 return 0; 862 } 863 /* mem section does not span the specified node */ 864 return 0; 865 } 866 867 /* 868 * During hotplug we know that all pages in the memory block belong to the same 869 * node. 870 */ 871 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk, 872 void *arg) 873 { 874 int nid = *(int *)arg; 875 876 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG); 877 return 0; 878 } 879 880 /* 881 * Unregister a memory block device under the node it spans. Memory blocks 882 * with multiple nodes cannot be offlined and therefore also never be removed. 883 */ 884 void unregister_memory_block_under_nodes(struct memory_block *mem_blk) 885 { 886 if (mem_blk->nid == NUMA_NO_NODE) 887 return; 888 889 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj, 890 kobject_name(&mem_blk->dev.kobj)); 891 sysfs_remove_link(&mem_blk->dev.kobj, 892 kobject_name(&node_devices[mem_blk->nid]->dev.kobj)); 893 } 894 895 void register_memory_blocks_under_node(int nid, unsigned long start_pfn, 896 unsigned long end_pfn, 897 enum meminit_context context) 898 { 899 walk_memory_blocks_func_t func; 900 901 if (context == MEMINIT_HOTPLUG) 902 func = register_mem_block_under_node_hotplug; 903 else 904 func = register_mem_block_under_node_early; 905 906 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn), 907 (void *)&nid, func); 908 return; 909 } 910 911 #ifdef CONFIG_HUGETLBFS 912 /* 913 * Handle per node hstate attribute [un]registration on transistions 914 * to/from memoryless state. 915 */ 916 static void node_hugetlb_work(struct work_struct *work) 917 { 918 struct node *node = container_of(work, struct node, node_work); 919 920 /* 921 * We only get here when a node transitions to/from memoryless state. 922 * We can detect which transition occurred by examining whether the 923 * node has memory now. hugetlb_register_node() already check this 924 * so we try to register the attributes. If that fails, then the 925 * node has transitioned to memoryless, try to unregister the 926 * attributes. 927 */ 928 if (!hugetlb_register_node(node)) 929 hugetlb_unregister_node(node); 930 } 931 932 static void init_node_hugetlb_work(int nid) 933 { 934 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work); 935 } 936 937 static int node_memory_callback(struct notifier_block *self, 938 unsigned long action, void *arg) 939 { 940 struct memory_notify *mnb = arg; 941 int nid = mnb->status_change_nid; 942 943 switch (action) { 944 case MEM_ONLINE: 945 case MEM_OFFLINE: 946 /* 947 * offload per node hstate [un]registration to a work thread 948 * when transitioning to/from memoryless state. 949 */ 950 if (nid != NUMA_NO_NODE) 951 schedule_work(&node_devices[nid]->node_work); 952 break; 953 954 case MEM_GOING_ONLINE: 955 case MEM_GOING_OFFLINE: 956 case MEM_CANCEL_ONLINE: 957 case MEM_CANCEL_OFFLINE: 958 default: 959 break; 960 } 961 962 return NOTIFY_OK; 963 } 964 #endif /* CONFIG_HUGETLBFS */ 965 #endif /* CONFIG_MEMORY_HOTPLUG */ 966 967 #if !defined(CONFIG_MEMORY_HOTPLUG) || !defined(CONFIG_HUGETLBFS) 968 static inline int node_memory_callback(struct notifier_block *self, 969 unsigned long action, void *arg) 970 { 971 return NOTIFY_OK; 972 } 973 974 static void init_node_hugetlb_work(int nid) { } 975 976 #endif 977 978 int __register_one_node(int nid) 979 { 980 int error; 981 int cpu; 982 983 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL); 984 if (!node_devices[nid]) 985 return -ENOMEM; 986 987 error = register_node(node_devices[nid], nid); 988 989 /* link cpu under this node */ 990 for_each_present_cpu(cpu) { 991 if (cpu_to_node(cpu) == nid) 992 register_cpu_under_node(cpu, nid); 993 } 994 995 INIT_LIST_HEAD(&node_devices[nid]->access_list); 996 /* initialize work queue for memory hot plug */ 997 init_node_hugetlb_work(nid); 998 node_init_caches(nid); 999 1000 return error; 1001 } 1002 1003 void unregister_one_node(int nid) 1004 { 1005 if (!node_devices[nid]) 1006 return; 1007 1008 unregister_node(node_devices[nid]); 1009 node_devices[nid] = NULL; 1010 } 1011 1012 /* 1013 * node states attributes 1014 */ 1015 1016 struct node_attr { 1017 struct device_attribute attr; 1018 enum node_states state; 1019 }; 1020 1021 static ssize_t show_node_state(struct device *dev, 1022 struct device_attribute *attr, char *buf) 1023 { 1024 struct node_attr *na = container_of(attr, struct node_attr, attr); 1025 1026 return sysfs_emit(buf, "%*pbl\n", 1027 nodemask_pr_args(&node_states[na->state])); 1028 } 1029 1030 #define _NODE_ATTR(name, state) \ 1031 { __ATTR(name, 0444, show_node_state, NULL), state } 1032 1033 static struct node_attr node_state_attr[] = { 1034 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE), 1035 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE), 1036 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY), 1037 #ifdef CONFIG_HIGHMEM 1038 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY), 1039 #endif 1040 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), 1041 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), 1042 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator, 1043 N_GENERIC_INITIATOR), 1044 }; 1045 1046 static struct attribute *node_state_attrs[] = { 1047 &node_state_attr[N_POSSIBLE].attr.attr, 1048 &node_state_attr[N_ONLINE].attr.attr, 1049 &node_state_attr[N_NORMAL_MEMORY].attr.attr, 1050 #ifdef CONFIG_HIGHMEM 1051 &node_state_attr[N_HIGH_MEMORY].attr.attr, 1052 #endif 1053 &node_state_attr[N_MEMORY].attr.attr, 1054 &node_state_attr[N_CPU].attr.attr, 1055 &node_state_attr[N_GENERIC_INITIATOR].attr.attr, 1056 NULL 1057 }; 1058 1059 static const struct attribute_group memory_root_attr_group = { 1060 .attrs = node_state_attrs, 1061 }; 1062 1063 static const struct attribute_group *cpu_root_attr_groups[] = { 1064 &memory_root_attr_group, 1065 NULL, 1066 }; 1067 1068 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */ 1069 void __init node_dev_init(void) 1070 { 1071 static struct notifier_block node_memory_callback_nb = { 1072 .notifier_call = node_memory_callback, 1073 .priority = NODE_CALLBACK_PRI, 1074 }; 1075 int ret, i; 1076 1077 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES); 1078 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES); 1079 1080 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups); 1081 if (ret) 1082 panic("%s() failed to register subsystem: %d\n", __func__, ret); 1083 1084 register_hotmemory_notifier(&node_memory_callback_nb); 1085 1086 /* 1087 * Create all node devices, which will properly link the node 1088 * to applicable memory block devices and already created cpu devices. 1089 */ 1090 for_each_online_node(i) { 1091 ret = register_one_node(i); 1092 if (ret) 1093 panic("%s() failed to add node: %d\n", __func__, ret); 1094 } 1095 } 1096