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, 0); 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, 0); 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 NFS_Unstable: %8lu kB\n" 437 "Node %d Bounce: %8lu kB\n" 438 "Node %d WritebackTmp: %8lu kB\n" 439 "Node %d KReclaimable: %8lu kB\n" 440 "Node %d Slab: %8lu kB\n" 441 "Node %d SReclaimable: %8lu kB\n" 442 "Node %d SUnreclaim: %8lu kB\n" 443 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 444 "Node %d AnonHugePages: %8lu kB\n" 445 "Node %d ShmemHugePages: %8lu kB\n" 446 "Node %d ShmemPmdMapped: %8lu kB\n" 447 "Node %d FileHugePages: %8lu kB\n" 448 "Node %d FilePmdMapped: %8lu kB\n" 449 #endif 450 , 451 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)), 452 nid, K(node_page_state(pgdat, NR_WRITEBACK)), 453 nid, K(node_page_state(pgdat, NR_FILE_PAGES)), 454 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)), 455 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)), 456 nid, K(i.sharedram), 457 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB), 458 #ifdef CONFIG_SHADOW_CALL_STACK 459 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB), 460 #endif 461 nid, K(node_page_state(pgdat, NR_PAGETABLE)), 462 nid, 0UL, 463 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)), 464 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), 465 nid, K(sreclaimable + 466 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)), 467 nid, K(sreclaimable + sunreclaimable), 468 nid, K(sreclaimable), 469 nid, K(sunreclaimable) 470 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 471 , 472 nid, K(node_page_state(pgdat, NR_ANON_THPS)), 473 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)), 474 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)), 475 nid, K(node_page_state(pgdat, NR_FILE_THPS)), 476 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED)) 477 #endif 478 ); 479 len += hugetlb_report_node_meminfo(buf, len, nid); 480 return len; 481 } 482 483 #undef K 484 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL); 485 486 static ssize_t node_read_numastat(struct device *dev, 487 struct device_attribute *attr, char *buf) 488 { 489 fold_vm_numa_events(); 490 return sysfs_emit(buf, 491 "numa_hit %lu\n" 492 "numa_miss %lu\n" 493 "numa_foreign %lu\n" 494 "interleave_hit %lu\n" 495 "local_node %lu\n" 496 "other_node %lu\n", 497 sum_zone_numa_event_state(dev->id, NUMA_HIT), 498 sum_zone_numa_event_state(dev->id, NUMA_MISS), 499 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN), 500 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT), 501 sum_zone_numa_event_state(dev->id, NUMA_LOCAL), 502 sum_zone_numa_event_state(dev->id, NUMA_OTHER)); 503 } 504 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL); 505 506 static ssize_t node_read_vmstat(struct device *dev, 507 struct device_attribute *attr, char *buf) 508 { 509 int nid = dev->id; 510 struct pglist_data *pgdat = NODE_DATA(nid); 511 int i; 512 int len = 0; 513 514 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) 515 len += sysfs_emit_at(buf, len, "%s %lu\n", 516 zone_stat_name(i), 517 sum_zone_node_page_state(nid, i)); 518 519 #ifdef CONFIG_NUMA 520 fold_vm_numa_events(); 521 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++) 522 len += sysfs_emit_at(buf, len, "%s %lu\n", 523 numa_stat_name(i), 524 sum_zone_numa_event_state(nid, i)); 525 526 #endif 527 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { 528 unsigned long pages = node_page_state_pages(pgdat, i); 529 530 if (vmstat_item_print_in_thp(i)) 531 pages /= HPAGE_PMD_NR; 532 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i), 533 pages); 534 } 535 536 return len; 537 } 538 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL); 539 540 static ssize_t node_read_distance(struct device *dev, 541 struct device_attribute *attr, char *buf) 542 { 543 int nid = dev->id; 544 int len = 0; 545 int i; 546 547 /* 548 * buf is currently PAGE_SIZE in length and each node needs 4 chars 549 * at the most (distance + space or newline). 550 */ 551 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE); 552 553 for_each_online_node(i) { 554 len += sysfs_emit_at(buf, len, "%s%d", 555 i ? " " : "", node_distance(nid, i)); 556 } 557 558 len += sysfs_emit_at(buf, len, "\n"); 559 return len; 560 } 561 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL); 562 563 static struct attribute *node_dev_attrs[] = { 564 &dev_attr_meminfo.attr, 565 &dev_attr_numastat.attr, 566 &dev_attr_distance.attr, 567 &dev_attr_vmstat.attr, 568 NULL 569 }; 570 571 static struct bin_attribute *node_dev_bin_attrs[] = { 572 &bin_attr_cpumap, 573 &bin_attr_cpulist, 574 NULL 575 }; 576 577 static const struct attribute_group node_dev_group = { 578 .attrs = node_dev_attrs, 579 .bin_attrs = node_dev_bin_attrs 580 }; 581 582 static const struct attribute_group *node_dev_groups[] = { 583 &node_dev_group, 584 NULL 585 }; 586 587 #ifdef CONFIG_HUGETLBFS 588 /* 589 * hugetlbfs per node attributes registration interface: 590 * When/if hugetlb[fs] subsystem initializes [sometime after this module], 591 * it will register its per node attributes for all online nodes with 592 * memory. It will also call register_hugetlbfs_with_node(), below, to 593 * register its attribute registration functions with this node driver. 594 * Once these hooks have been initialized, the node driver will call into 595 * the hugetlb module to [un]register attributes for hot-plugged nodes. 596 */ 597 static node_registration_func_t __hugetlb_register_node; 598 static node_registration_func_t __hugetlb_unregister_node; 599 600 static inline bool hugetlb_register_node(struct node *node) 601 { 602 if (__hugetlb_register_node && 603 node_state(node->dev.id, N_MEMORY)) { 604 __hugetlb_register_node(node); 605 return true; 606 } 607 return false; 608 } 609 610 static inline void hugetlb_unregister_node(struct node *node) 611 { 612 if (__hugetlb_unregister_node) 613 __hugetlb_unregister_node(node); 614 } 615 616 void register_hugetlbfs_with_node(node_registration_func_t doregister, 617 node_registration_func_t unregister) 618 { 619 __hugetlb_register_node = doregister; 620 __hugetlb_unregister_node = unregister; 621 } 622 #else 623 static inline void hugetlb_register_node(struct node *node) {} 624 625 static inline void hugetlb_unregister_node(struct node *node) {} 626 #endif 627 628 static void node_device_release(struct device *dev) 629 { 630 struct node *node = to_node(dev); 631 632 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS) 633 /* 634 * We schedule the work only when a memory section is 635 * onlined/offlined on this node. When we come here, 636 * all the memory on this node has been offlined, 637 * so we won't enqueue new work to this work. 638 * 639 * The work is using node->node_work, so we should 640 * flush work before freeing the memory. 641 */ 642 flush_work(&node->node_work); 643 #endif 644 kfree(node); 645 } 646 647 /* 648 * register_node - Setup a sysfs device for a node. 649 * @num - Node number to use when creating the device. 650 * 651 * Initialize and register the node device. 652 */ 653 static int register_node(struct node *node, int num) 654 { 655 int error; 656 657 node->dev.id = num; 658 node->dev.bus = &node_subsys; 659 node->dev.release = node_device_release; 660 node->dev.groups = node_dev_groups; 661 error = device_register(&node->dev); 662 663 if (error) 664 put_device(&node->dev); 665 else { 666 hugetlb_register_node(node); 667 668 compaction_register_node(node); 669 } 670 return error; 671 } 672 673 /** 674 * unregister_node - unregister a node device 675 * @node: node going away 676 * 677 * Unregisters a node device @node. All the devices on the node must be 678 * unregistered before calling this function. 679 */ 680 void unregister_node(struct node *node) 681 { 682 hugetlb_unregister_node(node); /* no-op, if memoryless node */ 683 node_remove_accesses(node); 684 node_remove_caches(node); 685 device_unregister(&node->dev); 686 } 687 688 struct node *node_devices[MAX_NUMNODES]; 689 690 /* 691 * register cpu under node 692 */ 693 int register_cpu_under_node(unsigned int cpu, unsigned int nid) 694 { 695 int ret; 696 struct device *obj; 697 698 if (!node_online(nid)) 699 return 0; 700 701 obj = get_cpu_device(cpu); 702 if (!obj) 703 return 0; 704 705 ret = sysfs_create_link(&node_devices[nid]->dev.kobj, 706 &obj->kobj, 707 kobject_name(&obj->kobj)); 708 if (ret) 709 return ret; 710 711 return sysfs_create_link(&obj->kobj, 712 &node_devices[nid]->dev.kobj, 713 kobject_name(&node_devices[nid]->dev.kobj)); 714 } 715 716 /** 717 * register_memory_node_under_compute_node - link memory node to its compute 718 * node for a given access class. 719 * @mem_nid: Memory node number 720 * @cpu_nid: Cpu node number 721 * @access: Access class to register 722 * 723 * Description: 724 * For use with platforms that may have separate memory and compute nodes. 725 * This function will export node relationships linking which memory 726 * initiator nodes can access memory targets at a given ranked access 727 * class. 728 */ 729 int register_memory_node_under_compute_node(unsigned int mem_nid, 730 unsigned int cpu_nid, 731 unsigned int access) 732 { 733 struct node *init_node, *targ_node; 734 struct node_access_nodes *initiator, *target; 735 int ret; 736 737 if (!node_online(cpu_nid) || !node_online(mem_nid)) 738 return -ENODEV; 739 740 init_node = node_devices[cpu_nid]; 741 targ_node = node_devices[mem_nid]; 742 initiator = node_init_node_access(init_node, access); 743 target = node_init_node_access(targ_node, access); 744 if (!initiator || !target) 745 return -ENOMEM; 746 747 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets", 748 &targ_node->dev.kobj, 749 dev_name(&targ_node->dev)); 750 if (ret) 751 return ret; 752 753 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators", 754 &init_node->dev.kobj, 755 dev_name(&init_node->dev)); 756 if (ret) 757 goto err; 758 759 return 0; 760 err: 761 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets", 762 dev_name(&targ_node->dev)); 763 return ret; 764 } 765 766 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid) 767 { 768 struct device *obj; 769 770 if (!node_online(nid)) 771 return 0; 772 773 obj = get_cpu_device(cpu); 774 if (!obj) 775 return 0; 776 777 sysfs_remove_link(&node_devices[nid]->dev.kobj, 778 kobject_name(&obj->kobj)); 779 sysfs_remove_link(&obj->kobj, 780 kobject_name(&node_devices[nid]->dev.kobj)); 781 782 return 0; 783 } 784 785 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE 786 static int __ref get_nid_for_pfn(unsigned long pfn) 787 { 788 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 789 if (system_state < SYSTEM_RUNNING) 790 return early_pfn_to_nid(pfn); 791 #endif 792 return pfn_to_nid(pfn); 793 } 794 795 static void do_register_memory_block_under_node(int nid, 796 struct memory_block *mem_blk) 797 { 798 int ret; 799 800 /* 801 * If this memory block spans multiple nodes, we only indicate 802 * the last processed node. 803 */ 804 mem_blk->nid = nid; 805 806 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj, 807 &mem_blk->dev.kobj, 808 kobject_name(&mem_blk->dev.kobj)); 809 if (ret && ret != -EEXIST) 810 dev_err_ratelimited(&node_devices[nid]->dev, 811 "can't create link to %s in sysfs (%d)\n", 812 kobject_name(&mem_blk->dev.kobj), ret); 813 814 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj, 815 &node_devices[nid]->dev.kobj, 816 kobject_name(&node_devices[nid]->dev.kobj)); 817 if (ret && ret != -EEXIST) 818 dev_err_ratelimited(&mem_blk->dev, 819 "can't create link to %s in sysfs (%d)\n", 820 kobject_name(&node_devices[nid]->dev.kobj), 821 ret); 822 } 823 824 /* register memory section under specified node if it spans that node */ 825 static int register_mem_block_under_node_early(struct memory_block *mem_blk, 826 void *arg) 827 { 828 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE; 829 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr); 830 unsigned long end_pfn = start_pfn + memory_block_pfns - 1; 831 int nid = *(int *)arg; 832 unsigned long pfn; 833 834 for (pfn = start_pfn; pfn <= end_pfn; pfn++) { 835 int page_nid; 836 837 /* 838 * memory block could have several absent sections from start. 839 * skip pfn range from absent section 840 */ 841 if (!pfn_in_present_section(pfn)) { 842 pfn = round_down(pfn + PAGES_PER_SECTION, 843 PAGES_PER_SECTION) - 1; 844 continue; 845 } 846 847 /* 848 * We need to check if page belongs to nid only at the boot 849 * case because node's ranges can be interleaved. 850 */ 851 page_nid = get_nid_for_pfn(pfn); 852 if (page_nid < 0) 853 continue; 854 if (page_nid != nid) 855 continue; 856 857 do_register_memory_block_under_node(nid, mem_blk); 858 return 0; 859 } 860 /* mem section does not span the specified node */ 861 return 0; 862 } 863 864 /* 865 * During hotplug we know that all pages in the memory block belong to the same 866 * node. 867 */ 868 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk, 869 void *arg) 870 { 871 int nid = *(int *)arg; 872 873 do_register_memory_block_under_node(nid, mem_blk); 874 return 0; 875 } 876 877 /* 878 * Unregister a memory block device under the node it spans. Memory blocks 879 * with multiple nodes cannot be offlined and therefore also never be removed. 880 */ 881 void unregister_memory_block_under_nodes(struct memory_block *mem_blk) 882 { 883 if (mem_blk->nid == NUMA_NO_NODE) 884 return; 885 886 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj, 887 kobject_name(&mem_blk->dev.kobj)); 888 sysfs_remove_link(&mem_blk->dev.kobj, 889 kobject_name(&node_devices[mem_blk->nid]->dev.kobj)); 890 } 891 892 void link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn, 893 enum meminit_context context) 894 { 895 walk_memory_blocks_func_t func; 896 897 if (context == MEMINIT_HOTPLUG) 898 func = register_mem_block_under_node_hotplug; 899 else 900 func = register_mem_block_under_node_early; 901 902 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn), 903 (void *)&nid, func); 904 return; 905 } 906 907 #ifdef CONFIG_HUGETLBFS 908 /* 909 * Handle per node hstate attribute [un]registration on transistions 910 * to/from memoryless state. 911 */ 912 static void node_hugetlb_work(struct work_struct *work) 913 { 914 struct node *node = container_of(work, struct node, node_work); 915 916 /* 917 * We only get here when a node transitions to/from memoryless state. 918 * We can detect which transition occurred by examining whether the 919 * node has memory now. hugetlb_register_node() already check this 920 * so we try to register the attributes. If that fails, then the 921 * node has transitioned to memoryless, try to unregister the 922 * attributes. 923 */ 924 if (!hugetlb_register_node(node)) 925 hugetlb_unregister_node(node); 926 } 927 928 static void init_node_hugetlb_work(int nid) 929 { 930 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work); 931 } 932 933 static int node_memory_callback(struct notifier_block *self, 934 unsigned long action, void *arg) 935 { 936 struct memory_notify *mnb = arg; 937 int nid = mnb->status_change_nid; 938 939 switch (action) { 940 case MEM_ONLINE: 941 case MEM_OFFLINE: 942 /* 943 * offload per node hstate [un]registration to a work thread 944 * when transitioning to/from memoryless state. 945 */ 946 if (nid != NUMA_NO_NODE) 947 schedule_work(&node_devices[nid]->node_work); 948 break; 949 950 case MEM_GOING_ONLINE: 951 case MEM_GOING_OFFLINE: 952 case MEM_CANCEL_ONLINE: 953 case MEM_CANCEL_OFFLINE: 954 default: 955 break; 956 } 957 958 return NOTIFY_OK; 959 } 960 #endif /* CONFIG_HUGETLBFS */ 961 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */ 962 963 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \ 964 !defined(CONFIG_HUGETLBFS) 965 static inline int node_memory_callback(struct notifier_block *self, 966 unsigned long action, void *arg) 967 { 968 return NOTIFY_OK; 969 } 970 971 static void init_node_hugetlb_work(int nid) { } 972 973 #endif 974 975 int __register_one_node(int nid) 976 { 977 int error; 978 int cpu; 979 980 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL); 981 if (!node_devices[nid]) 982 return -ENOMEM; 983 984 error = register_node(node_devices[nid], nid); 985 986 /* link cpu under this node */ 987 for_each_present_cpu(cpu) { 988 if (cpu_to_node(cpu) == nid) 989 register_cpu_under_node(cpu, nid); 990 } 991 992 INIT_LIST_HEAD(&node_devices[nid]->access_list); 993 /* initialize work queue for memory hot plug */ 994 init_node_hugetlb_work(nid); 995 node_init_caches(nid); 996 997 return error; 998 } 999 1000 void unregister_one_node(int nid) 1001 { 1002 if (!node_devices[nid]) 1003 return; 1004 1005 unregister_node(node_devices[nid]); 1006 node_devices[nid] = NULL; 1007 } 1008 1009 /* 1010 * node states attributes 1011 */ 1012 1013 struct node_attr { 1014 struct device_attribute attr; 1015 enum node_states state; 1016 }; 1017 1018 static ssize_t show_node_state(struct device *dev, 1019 struct device_attribute *attr, char *buf) 1020 { 1021 struct node_attr *na = container_of(attr, struct node_attr, attr); 1022 1023 return sysfs_emit(buf, "%*pbl\n", 1024 nodemask_pr_args(&node_states[na->state])); 1025 } 1026 1027 #define _NODE_ATTR(name, state) \ 1028 { __ATTR(name, 0444, show_node_state, NULL), state } 1029 1030 static struct node_attr node_state_attr[] = { 1031 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE), 1032 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE), 1033 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY), 1034 #ifdef CONFIG_HIGHMEM 1035 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY), 1036 #endif 1037 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), 1038 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), 1039 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator, 1040 N_GENERIC_INITIATOR), 1041 }; 1042 1043 static struct attribute *node_state_attrs[] = { 1044 &node_state_attr[N_POSSIBLE].attr.attr, 1045 &node_state_attr[N_ONLINE].attr.attr, 1046 &node_state_attr[N_NORMAL_MEMORY].attr.attr, 1047 #ifdef CONFIG_HIGHMEM 1048 &node_state_attr[N_HIGH_MEMORY].attr.attr, 1049 #endif 1050 &node_state_attr[N_MEMORY].attr.attr, 1051 &node_state_attr[N_CPU].attr.attr, 1052 &node_state_attr[N_GENERIC_INITIATOR].attr.attr, 1053 NULL 1054 }; 1055 1056 static const struct attribute_group memory_root_attr_group = { 1057 .attrs = node_state_attrs, 1058 }; 1059 1060 static const struct attribute_group *cpu_root_attr_groups[] = { 1061 &memory_root_attr_group, 1062 NULL, 1063 }; 1064 1065 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */ 1066 static int __init register_node_type(void) 1067 { 1068 int ret; 1069 1070 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES); 1071 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES); 1072 1073 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups); 1074 if (!ret) { 1075 static struct notifier_block node_memory_callback_nb = { 1076 .notifier_call = node_memory_callback, 1077 .priority = NODE_CALLBACK_PRI, 1078 }; 1079 register_hotmemory_notifier(&node_memory_callback_nb); 1080 } 1081 1082 /* 1083 * Note: we're not going to unregister the node class if we fail 1084 * to register the node state class attribute files. 1085 */ 1086 return ret; 1087 } 1088 postcore_initcall(register_node_type); 1089