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