drivers/base/node.c

   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(property)                                           \
 165 static ssize_t property##_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.property);   \
 171 }                                                                       \
 172 static DEVICE_ATTR_RO(property)
 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 #ifdef CONFIG_UNACCEPTED_MEMORY
 452                              "Node %d Unaccepted:     %8lu kB\n"
 453 #endif
 454                              ,
 455                              nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
 456                              nid, K(node_page_state(pgdat, NR_WRITEBACK)),
 457                              nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
 458                              nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
 459                              nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
 460                              nid, K(i.sharedram),
 461                              nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
 462 #ifdef CONFIG_SHADOW_CALL_STACK
 463                              nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
 464 #endif
 465                              nid, K(node_page_state(pgdat, NR_PAGETABLE)),
 466                              nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
 467                              nid, 0UL,
 468                              nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
 469                              nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
 470                              nid, K(sreclaimable +
 471                                     node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
 472                              nid, K(sreclaimable + sunreclaimable),
 473                              nid, K(sreclaimable),
 474                              nid, K(sunreclaimable)
 475 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 476                              ,
 477                              nid, K(node_page_state(pgdat, NR_ANON_THPS)),
 478                              nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
 479                              nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
 480                              nid, K(node_page_state(pgdat, NR_FILE_THPS)),
 481                              nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
 482 #endif
 483 #ifdef CONFIG_UNACCEPTED_MEMORY
 484                              ,
 485                              nid, K(sum_zone_node_page_state(nid, NR_UNACCEPTED))
 486 #endif
 487                             );
 488         len += hugetlb_report_node_meminfo(buf, len, nid);
 489         return len;
 490 }
 491
 492 #undef K
 493 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
 494
 495 static ssize_t node_read_numastat(struct device *dev,
 496                                   struct device_attribute *attr, char *buf)
 497 {
 498         fold_vm_numa_events();
 499         return sysfs_emit(buf,
 500                           "numa_hit %lu\n"
 501                           "numa_miss %lu\n"
 502                           "numa_foreign %lu\n"
 503                           "interleave_hit %lu\n"
 504                           "local_node %lu\n"
 505                           "other_node %lu\n",
 506                           sum_zone_numa_event_state(dev->id, NUMA_HIT),
 507                           sum_zone_numa_event_state(dev->id, NUMA_MISS),
 508                           sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
 509                           sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
 510                           sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
 511                           sum_zone_numa_event_state(dev->id, NUMA_OTHER));
 512 }
 513 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
 514
 515 static ssize_t node_read_vmstat(struct device *dev,
 516                                 struct device_attribute *attr, char *buf)
 517 {
 518         int nid = dev->id;
 519         struct pglist_data *pgdat = NODE_DATA(nid);
 520         int i;
 521         int len = 0;
 522
 523         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 524                 len += sysfs_emit_at(buf, len, "%s %lu\n",
 525                                      zone_stat_name(i),
 526                                      sum_zone_node_page_state(nid, i));
 527
 528 #ifdef CONFIG_NUMA
 529         fold_vm_numa_events();
 530         for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
 531                 len += sysfs_emit_at(buf, len, "%s %lu\n",
 532                                      numa_stat_name(i),
 533                                      sum_zone_numa_event_state(nid, i));
 534
 535 #endif
 536         for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
 537                 unsigned long pages = node_page_state_pages(pgdat, i);
 538
 539                 if (vmstat_item_print_in_thp(i))
 540                         pages /= HPAGE_PMD_NR;
 541                 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
 542                                      pages);
 543         }
 544
 545         return len;
 546 }
 547 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
 548
 549 static ssize_t node_read_distance(struct device *dev,
 550                                   struct device_attribute *attr, char *buf)
 551 {
 552         int nid = dev->id;
 553         int len = 0;
 554         int i;
 555
 556         /*
 557          * buf is currently PAGE_SIZE in length and each node needs 4 chars
 558          * at the most (distance + space or newline).
 559          */
 560         BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
 561
 562         for_each_online_node(i) {
 563                 len += sysfs_emit_at(buf, len, "%s%d",
 564                                      i ? " " : "", node_distance(nid, i));
 565         }
 566
 567         len += sysfs_emit_at(buf, len, "\n");
 568         return len;
 569 }
 570 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
 571
 572 static struct attribute *node_dev_attrs[] = {
 573         &dev_attr_meminfo.attr,
 574         &dev_attr_numastat.attr,
 575         &dev_attr_distance.attr,
 576         &dev_attr_vmstat.attr,
 577         NULL
 578 };
 579
 580 static struct bin_attribute *node_dev_bin_attrs[] = {
 581         &bin_attr_cpumap,
 582         &bin_attr_cpulist,
 583         NULL
 584 };
 585
 586 static const struct attribute_group node_dev_group = {
 587         .attrs = node_dev_attrs,
 588         .bin_attrs = node_dev_bin_attrs
 589 };
 590
 591 static const struct attribute_group *node_dev_groups[] = {
 592         &node_dev_group,
 593 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
 594         &arch_node_dev_group,
 595 #endif
 596 #ifdef CONFIG_MEMORY_FAILURE
 597         &memory_failure_attr_group,
 598 #endif
 599         NULL
 600 };
 601
 602 static void node_device_release(struct device *dev)
 603 {
 604         kfree(to_node(dev));
 605 }
 606
 607 /*
 608  * register_node - Setup a sysfs device for a node.
 609  * @num - Node number to use when creating the device.
 610  *
 611  * Initialize and register the node device.
 612  */
 613 static int register_node(struct node *node, int num)
 614 {
 615         int error;
 616
 617         node->dev.id = num;
 618         node->dev.bus = &node_subsys;
 619         node->dev.release = node_device_release;
 620         node->dev.groups = node_dev_groups;
 621         error = device_register(&node->dev);
 622
 623         if (error) {
 624                 put_device(&node->dev);
 625         } else {
 626                 hugetlb_register_node(node);
 627                 compaction_register_node(node);
 628         }
 629
 630         return error;
 631 }
 632
 633 /**
 634  * unregister_node - unregister a node device
 635  * @node: node going away
 636  *
 637  * Unregisters a node device @node.  All the devices on the node must be
 638  * unregistered before calling this function.
 639  */
 640 void unregister_node(struct node *node)
 641 {
 642         hugetlb_unregister_node(node);
 643         compaction_unregister_node(node);
 644         node_remove_accesses(node);
 645         node_remove_caches(node);
 646         device_unregister(&node->dev);
 647 }
 648
 649 struct node *node_devices[MAX_NUMNODES];
 650
 651 /*
 652  * register cpu under node
 653  */
 654 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
 655 {
 656         int ret;
 657         struct device *obj;
 658
 659         if (!node_online(nid))
 660                 return 0;
 661
 662         obj = get_cpu_device(cpu);
 663         if (!obj)
 664                 return 0;
 665
 666         ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
 667                                 &obj->kobj,
 668                                 kobject_name(&obj->kobj));
 669         if (ret)
 670                 return ret;
 671
 672         return sysfs_create_link(&obj->kobj,
 673                                  &node_devices[nid]->dev.kobj,
 674                                  kobject_name(&node_devices[nid]->dev.kobj));
 675 }
 676
 677 /**
 678  * register_memory_node_under_compute_node - link memory node to its compute
 679  *                                           node for a given access class.
 680  * @mem_nid:    Memory node number
 681  * @cpu_nid:    Cpu  node number
 682  * @access:     Access class to register
 683  *
 684  * Description:
 685  *      For use with platforms that may have separate memory and compute nodes.
 686  *      This function will export node relationships linking which memory
 687  *      initiator nodes can access memory targets at a given ranked access
 688  *      class.
 689  */
 690 int register_memory_node_under_compute_node(unsigned int mem_nid,
 691                                             unsigned int cpu_nid,
 692                                             unsigned int access)
 693 {
 694         struct node *init_node, *targ_node;
 695         struct node_access_nodes *initiator, *target;
 696         int ret;
 697
 698         if (!node_online(cpu_nid) || !node_online(mem_nid))
 699                 return -ENODEV;
 700
 701         init_node = node_devices[cpu_nid];
 702         targ_node = node_devices[mem_nid];
 703         initiator = node_init_node_access(init_node, access);
 704         target = node_init_node_access(targ_node, access);
 705         if (!initiator || !target)
 706                 return -ENOMEM;
 707
 708         ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
 709                                       &targ_node->dev.kobj,
 710                                       dev_name(&targ_node->dev));
 711         if (ret)
 712                 return ret;
 713
 714         ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
 715                                       &init_node->dev.kobj,
 716                                       dev_name(&init_node->dev));
 717         if (ret)
 718                 goto err;
 719
 720         return 0;
 721  err:
 722         sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
 723                                      dev_name(&targ_node->dev));
 724         return ret;
 725 }
 726
 727 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
 728 {
 729         struct device *obj;
 730
 731         if (!node_online(nid))
 732                 return 0;
 733
 734         obj = get_cpu_device(cpu);
 735         if (!obj)
 736                 return 0;
 737
 738         sysfs_remove_link(&node_devices[nid]->dev.kobj,
 739                           kobject_name(&obj->kobj));
 740         sysfs_remove_link(&obj->kobj,
 741                           kobject_name(&node_devices[nid]->dev.kobj));
 742
 743         return 0;
 744 }
 745
 746 #ifdef CONFIG_MEMORY_HOTPLUG
 747 static int __ref get_nid_for_pfn(unsigned long pfn)
 748 {
 749 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
 750         if (system_state < SYSTEM_RUNNING)
 751                 return early_pfn_to_nid(pfn);
 752 #endif
 753         return pfn_to_nid(pfn);
 754 }
 755
 756 static void do_register_memory_block_under_node(int nid,
 757                                                 struct memory_block *mem_blk,
 758                                                 enum meminit_context context)
 759 {
 760         int ret;
 761
 762         memory_block_add_nid(mem_blk, nid, context);
 763
 764         ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
 765                                        &mem_blk->dev.kobj,
 766                                        kobject_name(&mem_blk->dev.kobj));
 767         if (ret && ret != -EEXIST)
 768                 dev_err_ratelimited(&node_devices[nid]->dev,
 769                                     "can't create link to %s in sysfs (%d)\n",
 770                                     kobject_name(&mem_blk->dev.kobj), ret);
 771
 772         ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
 773                                 &node_devices[nid]->dev.kobj,
 774                                 kobject_name(&node_devices[nid]->dev.kobj));
 775         if (ret && ret != -EEXIST)
 776                 dev_err_ratelimited(&mem_blk->dev,
 777                                     "can't create link to %s in sysfs (%d)\n",
 778                                     kobject_name(&node_devices[nid]->dev.kobj),
 779                                     ret);
 780 }
 781
 782 /* register memory section under specified node if it spans that node */
 783 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
 784                                                void *arg)
 785 {
 786         unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
 787         unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
 788         unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
 789         int nid = *(int *)arg;
 790         unsigned long pfn;
 791
 792         for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
 793                 int page_nid;
 794
 795                 /*
 796                  * memory block could have several absent sections from start.
 797                  * skip pfn range from absent section
 798                  */
 799                 if (!pfn_in_present_section(pfn)) {
 800                         pfn = round_down(pfn + PAGES_PER_SECTION,
 801                                          PAGES_PER_SECTION) - 1;
 802                         continue;
 803                 }
 804
 805                 /*
 806                  * We need to check if page belongs to nid only at the boot
 807                  * case because node's ranges can be interleaved.
 808                  */
 809                 page_nid = get_nid_for_pfn(pfn);
 810                 if (page_nid < 0)
 811                         continue;
 812                 if (page_nid != nid)
 813                         continue;
 814
 815                 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
 816                 return 0;
 817         }
 818         /* mem section does not span the specified node */
 819         return 0;
 820 }
 821
 822 /*
 823  * During hotplug we know that all pages in the memory block belong to the same
 824  * node.
 825  */
 826 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
 827                                                  void *arg)
 828 {
 829         int nid = *(int *)arg;
 830
 831         do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
 832         return 0;
 833 }
 834
 835 /*
 836  * Unregister a memory block device under the node it spans. Memory blocks
 837  * with multiple nodes cannot be offlined and therefore also never be removed.
 838  */
 839 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
 840 {
 841         if (mem_blk->nid == NUMA_NO_NODE)
 842                 return;
 843
 844         sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
 845                           kobject_name(&mem_blk->dev.kobj));
 846         sysfs_remove_link(&mem_blk->dev.kobj,
 847                           kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
 848 }
 849
 850 void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
 851                                        unsigned long end_pfn,
 852                                        enum meminit_context context)
 853 {
 854         walk_memory_blocks_func_t func;
 855
 856         if (context == MEMINIT_HOTPLUG)
 857                 func = register_mem_block_under_node_hotplug;
 858         else
 859                 func = register_mem_block_under_node_early;
 860
 861         walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
 862                            (void *)&nid, func);
 863         return;
 864 }
 865 #endif /* CONFIG_MEMORY_HOTPLUG */
 866
 867 int __register_one_node(int nid)
 868 {
 869         int error;
 870         int cpu;
 871         struct node *node;
 872
 873         node = kzalloc(sizeof(struct node), GFP_KERNEL);
 874         if (!node)
 875                 return -ENOMEM;
 876
 877         INIT_LIST_HEAD(&node->access_list);
 878         node_devices[nid] = node;
 879
 880         error = register_node(node_devices[nid], nid);
 881
 882         /* link cpu under this node */
 883         for_each_present_cpu(cpu) {
 884                 if (cpu_to_node(cpu) == nid)
 885                         register_cpu_under_node(cpu, nid);
 886         }
 887
 888         node_init_caches(nid);
 889
 890         return error;
 891 }
 892
 893 void unregister_one_node(int nid)
 894 {
 895         if (!node_devices[nid])
 896                 return;
 897
 898         unregister_node(node_devices[nid]);
 899         node_devices[nid] = NULL;
 900 }
 901
 902 /*
 903  * node states attributes
 904  */
 905
 906 struct node_attr {
 907         struct device_attribute attr;
 908         enum node_states state;
 909 };
 910
 911 static ssize_t show_node_state(struct device *dev,
 912                                struct device_attribute *attr, char *buf)
 913 {
 914         struct node_attr *na = container_of(attr, struct node_attr, attr);
 915
 916         return sysfs_emit(buf, "%*pbl\n",
 917                           nodemask_pr_args(&node_states[na->state]));
 918 }
 919
 920 #define _NODE_ATTR(name, state) \
 921         { __ATTR(name, 0444, show_node_state, NULL), state }
 922
 923 static struct node_attr node_state_attr[] = {
 924         [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
 925         [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
 926         [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
 927 #ifdef CONFIG_HIGHMEM
 928         [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
 929 #endif
 930         [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
 931         [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
 932         [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
 933                                            N_GENERIC_INITIATOR),
 934 };
 935
 936 static struct attribute *node_state_attrs[] = {
 937         &node_state_attr[N_POSSIBLE].attr.attr,
 938         &node_state_attr[N_ONLINE].attr.attr,
 939         &node_state_attr[N_NORMAL_MEMORY].attr.attr,
 940 #ifdef CONFIG_HIGHMEM
 941         &node_state_attr[N_HIGH_MEMORY].attr.attr,
 942 #endif
 943         &node_state_attr[N_MEMORY].attr.attr,
 944         &node_state_attr[N_CPU].attr.attr,
 945         &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
 946         NULL
 947 };
 948
 949 static const struct attribute_group memory_root_attr_group = {
 950         .attrs = node_state_attrs,
 951 };
 952
 953 static const struct attribute_group *cpu_root_attr_groups[] = {
 954         &memory_root_attr_group,
 955         NULL,
 956 };
 957
 958 void __init node_dev_init(void)
 959 {
 960         int ret, i;
 961
 962         BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
 963         BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
 964
 965         ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
 966         if (ret)
 967                 panic("%s() failed to register subsystem: %d\n", __func__, ret);
 968
 969         /*
 970          * Create all node devices, which will properly link the node
 971          * to applicable memory block devices and already created cpu devices.
 972          */
 973         for_each_online_node(i) {
 974                 ret = register_one_node(i);
 975                 if (ret)
 976                         panic("%s() failed to add node: %d\n", __func__, ret);
 977         }
 978 }