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