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
  30 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
  31 {
  32         ssize_t n;
  33         cpumask_var_t mask;
  34         struct node *node_dev = to_node(dev);
  35
  36         /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
  37         BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
  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_to_pagebuf(list, buf, mask);
  44         free_cpumask_var(mask);
  45
  46         return n;
  47 }
  48
  49 static inline ssize_t cpumap_show(struct device *dev,
  50                                   struct device_attribute *attr,
  51                                   char *buf)
  52 {
  53         return node_read_cpumap(dev, false, buf);
  54 }
  55
  56 static DEVICE_ATTR_RO(cpumap);
  57
  58 static inline ssize_t cpulist_show(struct device *dev,
  59                                    struct device_attribute *attr,
  60                                    char *buf)
  61 {
  62         return node_read_cpumap(dev, true, buf);
  63 }
  64
  65 static DEVICE_ATTR_RO(cpulist);
  66
  67 /**
  68  * struct node_access_nodes - Access class device to hold user visible
  69  *                            relationships to other nodes.
  70  * @dev:        Device for this memory access class
  71  * @list_node:  List element in the node's access list
  72  * @access:     The access class rank
  73  * @hmem_attrs: Heterogeneous memory performance attributes
  74  */
  75 struct node_access_nodes {
  76         struct device           dev;
  77         struct list_head        list_node;
  78         unsigned                access;
  79 #ifdef CONFIG_HMEM_REPORTING
  80         struct node_hmem_attrs  hmem_attrs;
  81 #endif
  82 };
  83 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
  84
  85 static struct attribute *node_init_access_node_attrs[] = {
  86         NULL,
  87 };
  88
  89 static struct attribute *node_targ_access_node_attrs[] = {
  90         NULL,
  91 };
  92
  93 static const struct attribute_group initiators = {
  94         .name   = "initiators",
  95         .attrs  = node_init_access_node_attrs,
  96 };
  97
  98 static const struct attribute_group targets = {
  99         .name   = "targets",
 100         .attrs  = node_targ_access_node_attrs,
 101 };
 102
 103 static const struct attribute_group *node_access_node_groups[] = {
 104         &initiators,
 105         &targets,
 106         NULL,
 107 };
 108
 109 static void node_remove_accesses(struct node *node)
 110 {
 111         struct node_access_nodes *c, *cnext;
 112
 113         list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
 114                 list_del(&c->list_node);
 115                 device_unregister(&c->dev);
 116         }
 117 }
 118
 119 static void node_access_release(struct device *dev)
 120 {
 121         kfree(to_access_nodes(dev));
 122 }
 123
 124 static struct node_access_nodes *node_init_node_access(struct node *node,
 125                                                        unsigned access)
 126 {
 127         struct node_access_nodes *access_node;
 128         struct device *dev;
 129
 130         list_for_each_entry(access_node, &node->access_list, list_node)
 131                 if (access_node->access == access)
 132                         return access_node;
 133
 134         access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
 135         if (!access_node)
 136                 return NULL;
 137
 138         access_node->access = access;
 139         dev = &access_node->dev;
 140         dev->parent = &node->dev;
 141         dev->release = node_access_release;
 142         dev->groups = node_access_node_groups;
 143         if (dev_set_name(dev, "access%u", access))
 144                 goto free;
 145
 146         if (device_register(dev))
 147                 goto free_name;
 148
 149         pm_runtime_no_callbacks(dev);
 150         list_add_tail(&access_node->list_node, &node->access_list);
 151         return access_node;
 152 free_name:
 153         kfree_const(dev->kobj.name);
 154 free:
 155         kfree(access_node);
 156         return NULL;
 157 }
 158
 159 #ifdef CONFIG_HMEM_REPORTING
 160 #define ACCESS_ATTR(name)                                               \
 161 static ssize_t name##_show(struct device *dev,                          \
 162                            struct device_attribute *attr,               \
 163                            char *buf)                                   \
 164 {                                                                       \
 165         return sysfs_emit(buf, "%u\n",                                  \
 166                           to_access_nodes(dev)->hmem_attrs.name);       \
 167 }                                                                       \
 168 static DEVICE_ATTR_RO(name)
 169
 170 ACCESS_ATTR(read_bandwidth);
 171 ACCESS_ATTR(read_latency);
 172 ACCESS_ATTR(write_bandwidth);
 173 ACCESS_ATTR(write_latency);
 174
 175 static struct attribute *access_attrs[] = {
 176         &dev_attr_read_bandwidth.attr,
 177         &dev_attr_read_latency.attr,
 178         &dev_attr_write_bandwidth.attr,
 179         &dev_attr_write_latency.attr,
 180         NULL,
 181 };
 182
 183 /**
 184  * node_set_perf_attrs - Set the performance values for given access class
 185  * @nid: Node identifier to be set
 186  * @hmem_attrs: Heterogeneous memory performance attributes
 187  * @access: The access class the for the given attributes
 188  */
 189 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
 190                          unsigned access)
 191 {
 192         struct node_access_nodes *c;
 193         struct node *node;
 194         int i;
 195
 196         if (WARN_ON_ONCE(!node_online(nid)))
 197                 return;
 198
 199         node = node_devices[nid];
 200         c = node_init_node_access(node, access);
 201         if (!c)
 202                 return;
 203
 204         c->hmem_attrs = *hmem_attrs;
 205         for (i = 0; access_attrs[i] != NULL; i++) {
 206                 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
 207                                             "initiators")) {
 208                         pr_info("failed to add performance attribute to node %d\n",
 209                                 nid);
 210                         break;
 211                 }
 212         }
 213 }
 214
 215 /**
 216  * struct node_cache_info - Internal tracking for memory node caches
 217  * @dev:        Device represeting the cache level
 218  * @node:       List element for tracking in the node
 219  * @cache_attrs:Attributes for this cache level
 220  */
 221 struct node_cache_info {
 222         struct device dev;
 223         struct list_head node;
 224         struct node_cache_attrs cache_attrs;
 225 };
 226 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
 227
 228 #define CACHE_ATTR(name, fmt)                                           \
 229 static ssize_t name##_show(struct device *dev,                          \
 230                            struct device_attribute *attr,               \
 231                            char *buf)                                   \
 232 {                                                                       \
 233         return sysfs_emit(buf, fmt "\n",                                \
 234                           to_cache_info(dev)->cache_attrs.name);        \
 235 }                                                                       \
 236 DEVICE_ATTR_RO(name);
 237
 238 CACHE_ATTR(size, "%llu")
 239 CACHE_ATTR(line_size, "%u")
 240 CACHE_ATTR(indexing, "%u")
 241 CACHE_ATTR(write_policy, "%u")
 242
 243 static struct attribute *cache_attrs[] = {
 244         &dev_attr_indexing.attr,
 245         &dev_attr_size.attr,
 246         &dev_attr_line_size.attr,
 247         &dev_attr_write_policy.attr,
 248         NULL,
 249 };
 250 ATTRIBUTE_GROUPS(cache);
 251
 252 static void node_cache_release(struct device *dev)
 253 {
 254         kfree(dev);
 255 }
 256
 257 static void node_cacheinfo_release(struct device *dev)
 258 {
 259         struct node_cache_info *info = to_cache_info(dev);
 260         kfree(info);
 261 }
 262
 263 static void node_init_cache_dev(struct node *node)
 264 {
 265         struct device *dev;
 266
 267         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 268         if (!dev)
 269                 return;
 270
 271         device_initialize(dev);
 272         dev->parent = &node->dev;
 273         dev->release = node_cache_release;
 274         if (dev_set_name(dev, "memory_side_cache"))
 275                 goto put_device;
 276
 277         if (device_add(dev))
 278                 goto put_device;
 279
 280         pm_runtime_no_callbacks(dev);
 281         node->cache_dev = dev;
 282         return;
 283 put_device:
 284         put_device(dev);
 285 }
 286
 287 /**
 288  * node_add_cache() - add cache attribute to a memory node
 289  * @nid: Node identifier that has new cache attributes
 290  * @cache_attrs: Attributes for the cache being added
 291  */
 292 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
 293 {
 294         struct node_cache_info *info;
 295         struct device *dev;
 296         struct node *node;
 297
 298         if (!node_online(nid) || !node_devices[nid])
 299                 return;
 300
 301         node = node_devices[nid];
 302         list_for_each_entry(info, &node->cache_attrs, node) {
 303                 if (info->cache_attrs.level == cache_attrs->level) {
 304                         dev_warn(&node->dev,
 305                                 "attempt to add duplicate cache level:%d\n",
 306                                 cache_attrs->level);
 307                         return;
 308                 }
 309         }
 310
 311         if (!node->cache_dev)
 312                 node_init_cache_dev(node);
 313         if (!node->cache_dev)
 314                 return;
 315
 316         info = kzalloc(sizeof(*info), GFP_KERNEL);
 317         if (!info)
 318                 return;
 319
 320         dev = &info->dev;
 321         device_initialize(dev);
 322         dev->parent = node->cache_dev;
 323         dev->release = node_cacheinfo_release;
 324         dev->groups = cache_groups;
 325         if (dev_set_name(dev, "index%d", cache_attrs->level))
 326                 goto put_device;
 327
 328         info->cache_attrs = *cache_attrs;
 329         if (device_add(dev)) {
 330                 dev_warn(&node->dev, "failed to add cache level:%d\n",
 331                          cache_attrs->level);
 332                 goto put_device;
 333         }
 334         pm_runtime_no_callbacks(dev);
 335         list_add_tail(&info->node, &node->cache_attrs);
 336         return;
 337 put_device:
 338         put_device(dev);
 339 }
 340
 341 static void node_remove_caches(struct node *node)
 342 {
 343         struct node_cache_info *info, *next;
 344
 345         if (!node->cache_dev)
 346                 return;
 347
 348         list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
 349                 list_del(&info->node);
 350                 device_unregister(&info->dev);
 351         }
 352         device_unregister(node->cache_dev);
 353 }
 354
 355 static void node_init_caches(unsigned int nid)
 356 {
 357         INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
 358 }
 359 #else
 360 static void node_init_caches(unsigned int nid) { }
 361 static void node_remove_caches(struct node *node) { }
 362 #endif
 363
 364 #define K(x) ((x) << (PAGE_SHIFT - 10))
 365 static ssize_t node_read_meminfo(struct device *dev,
 366                         struct device_attribute *attr, char *buf)
 367 {
 368         int len = 0;
 369         int nid = dev->id;
 370         struct pglist_data *pgdat = NODE_DATA(nid);
 371         struct sysinfo i;
 372         unsigned long sreclaimable, sunreclaimable;
 373
 374         si_meminfo_node(&i, nid);
 375         sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
 376         sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
 377         len = sysfs_emit_at(buf, len,
 378                             "Node %d MemTotal:       %8lu kB\n"
 379                             "Node %d MemFree:        %8lu kB\n"
 380                             "Node %d MemUsed:        %8lu kB\n"
 381                             "Node %d Active:         %8lu kB\n"
 382                             "Node %d Inactive:       %8lu kB\n"
 383                             "Node %d Active(anon):   %8lu kB\n"
 384                             "Node %d Inactive(anon): %8lu kB\n"
 385                             "Node %d Active(file):   %8lu kB\n"
 386                             "Node %d Inactive(file): %8lu kB\n"
 387                             "Node %d Unevictable:    %8lu kB\n"
 388                             "Node %d Mlocked:        %8lu kB\n",
 389                             nid, K(i.totalram),
 390                             nid, K(i.freeram),
 391                             nid, K(i.totalram - i.freeram),
 392                             nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
 393                                    node_page_state(pgdat, NR_ACTIVE_FILE)),
 394                             nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
 395                                    node_page_state(pgdat, NR_INACTIVE_FILE)),
 396                             nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
 397                             nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
 398                             nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
 399                             nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
 400                             nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
 401                             nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
 402
 403 #ifdef CONFIG_HIGHMEM
 404         len += sysfs_emit_at(buf, len,
 405                              "Node %d HighTotal:      %8lu kB\n"
 406                              "Node %d HighFree:       %8lu kB\n"
 407                              "Node %d LowTotal:       %8lu kB\n"
 408                              "Node %d LowFree:        %8lu kB\n",
 409                              nid, K(i.totalhigh),
 410                              nid, K(i.freehigh),
 411                              nid, K(i.totalram - i.totalhigh),
 412                              nid, K(i.freeram - i.freehigh));
 413 #endif
 414         len += sysfs_emit_at(buf, len,
 415                              "Node %d Dirty:          %8lu kB\n"
 416                              "Node %d Writeback:      %8lu kB\n"
 417                              "Node %d FilePages:      %8lu kB\n"
 418                              "Node %d Mapped:         %8lu kB\n"
 419                              "Node %d AnonPages:      %8lu kB\n"
 420                              "Node %d Shmem:          %8lu kB\n"
 421                              "Node %d KernelStack:    %8lu kB\n"
 422 #ifdef CONFIG_SHADOW_CALL_STACK
 423                              "Node %d ShadowCallStack:%8lu kB\n"
 424 #endif
 425                              "Node %d PageTables:     %8lu kB\n"
 426                              "Node %d NFS_Unstable:   %8lu kB\n"
 427                              "Node %d Bounce:         %8lu kB\n"
 428                              "Node %d WritebackTmp:   %8lu kB\n"
 429                              "Node %d KReclaimable:   %8lu kB\n"
 430                              "Node %d Slab:           %8lu kB\n"
 431                              "Node %d SReclaimable:   %8lu kB\n"
 432                              "Node %d SUnreclaim:     %8lu kB\n"
 433 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 434                              "Node %d AnonHugePages:  %8lu kB\n"
 435                              "Node %d ShmemHugePages: %8lu kB\n"
 436                              "Node %d ShmemPmdMapped: %8lu kB\n"
 437                              "Node %d FileHugePages: %8lu kB\n"
 438                              "Node %d FilePmdMapped: %8lu kB\n"
 439 #endif
 440                              ,
 441                              nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
 442                              nid, K(node_page_state(pgdat, NR_WRITEBACK)),
 443                              nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
 444                              nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
 445                              nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
 446                              nid, K(i.sharedram),
 447                              nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
 448 #ifdef CONFIG_SHADOW_CALL_STACK
 449                              nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
 450 #endif
 451                              nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
 452                              nid, 0UL,
 453                              nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
 454                              nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
 455                              nid, K(sreclaimable +
 456                                     node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
 457                              nid, K(sreclaimable + sunreclaimable),
 458                              nid, K(sreclaimable),
 459                              nid, K(sunreclaimable)
 460 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 461                              ,
 462                              nid, K(node_page_state(pgdat, NR_ANON_THPS) *
 463                                     HPAGE_PMD_NR),
 464                              nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
 465                                     HPAGE_PMD_NR),
 466                              nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
 467                                     HPAGE_PMD_NR),
 468                              nid, K(node_page_state(pgdat, NR_FILE_THPS) *
 469                                     HPAGE_PMD_NR),
 470                              nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED) *
 471                                     HPAGE_PMD_NR)
 472 #endif
 473                             );
 474         len += hugetlb_report_node_meminfo(buf, len, nid);
 475         return len;
 476 }
 477
 478 #undef K
 479 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
 480
 481 static ssize_t node_read_numastat(struct device *dev,
 482                                   struct device_attribute *attr, char *buf)
 483 {
 484         return sysfs_emit(buf,
 485                           "numa_hit %lu\n"
 486                           "numa_miss %lu\n"
 487                           "numa_foreign %lu\n"
 488                           "interleave_hit %lu\n"
 489                           "local_node %lu\n"
 490                           "other_node %lu\n",
 491                           sum_zone_numa_state(dev->id, NUMA_HIT),
 492                           sum_zone_numa_state(dev->id, NUMA_MISS),
 493                           sum_zone_numa_state(dev->id, NUMA_FOREIGN),
 494                           sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
 495                           sum_zone_numa_state(dev->id, NUMA_LOCAL),
 496                           sum_zone_numa_state(dev->id, NUMA_OTHER));
 497 }
 498 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
 499
 500 static ssize_t node_read_vmstat(struct device *dev,
 501                                 struct device_attribute *attr, char *buf)
 502 {
 503         int nid = dev->id;
 504         struct pglist_data *pgdat = NODE_DATA(nid);
 505         int i;
 506         int len = 0;
 507
 508         for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 509                 len += sysfs_emit_at(buf, len, "%s %lu\n",
 510                                      zone_stat_name(i),
 511                                      sum_zone_node_page_state(nid, i));
 512
 513 #ifdef CONFIG_NUMA
 514         for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
 515                 len += sysfs_emit_at(buf, len, "%s %lu\n",
 516                                      numa_stat_name(i),
 517                                      sum_zone_numa_state(nid, i));
 518
 519 #endif
 520         for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
 521                 len += sysfs_emit_at(buf, len, "%s %lu\n",
 522                                      node_stat_name(i),
 523                                      node_page_state_pages(pgdat, i));
 524
 525         return len;
 526 }
 527 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
 528
 529 static ssize_t node_read_distance(struct device *dev,
 530                                   struct device_attribute *attr, char *buf)
 531 {
 532         int nid = dev->id;
 533         int len = 0;
 534         int i;
 535
 536         /*
 537          * buf is currently PAGE_SIZE in length and each node needs 4 chars
 538          * at the most (distance + space or newline).
 539          */
 540         BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
 541
 542         for_each_online_node(i) {
 543                 len += sysfs_emit_at(buf, len, "%s%d",
 544                                      i ? " " : "", node_distance(nid, i));
 545         }
 546
 547         len += sysfs_emit_at(buf, len, "\n");
 548         return len;
 549 }
 550 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
 551
 552 static struct attribute *node_dev_attrs[] = {
 553         &dev_attr_cpumap.attr,
 554         &dev_attr_cpulist.attr,
 555         &dev_attr_meminfo.attr,
 556         &dev_attr_numastat.attr,
 557         &dev_attr_distance.attr,
 558         &dev_attr_vmstat.attr,
 559         NULL
 560 };
 561 ATTRIBUTE_GROUPS(node_dev);
 562
 563 #ifdef CONFIG_HUGETLBFS
 564 /*
 565  * hugetlbfs per node attributes registration interface:
 566  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
 567  * it will register its per node attributes for all online nodes with
 568  * memory.  It will also call register_hugetlbfs_with_node(), below, to
 569  * register its attribute registration functions with this node driver.
 570  * Once these hooks have been initialized, the node driver will call into
 571  * the hugetlb module to [un]register attributes for hot-plugged nodes.
 572  */
 573 static node_registration_func_t __hugetlb_register_node;
 574 static node_registration_func_t __hugetlb_unregister_node;
 575
 576 static inline bool hugetlb_register_node(struct node *node)
 577 {
 578         if (__hugetlb_register_node &&
 579                         node_state(node->dev.id, N_MEMORY)) {
 580                 __hugetlb_register_node(node);
 581                 return true;
 582         }
 583         return false;
 584 }
 585
 586 static inline void hugetlb_unregister_node(struct node *node)
 587 {
 588         if (__hugetlb_unregister_node)
 589                 __hugetlb_unregister_node(node);
 590 }
 591
 592 void register_hugetlbfs_with_node(node_registration_func_t doregister,
 593                                   node_registration_func_t unregister)
 594 {
 595         __hugetlb_register_node   = doregister;
 596         __hugetlb_unregister_node = unregister;
 597 }
 598 #else
 599 static inline void hugetlb_register_node(struct node *node) {}
 600
 601 static inline void hugetlb_unregister_node(struct node *node) {}
 602 #endif
 603
 604 static void node_device_release(struct device *dev)
 605 {
 606         struct node *node = to_node(dev);
 607
 608 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
 609         /*
 610          * We schedule the work only when a memory section is
 611          * onlined/offlined on this node. When we come here,
 612          * all the memory on this node has been offlined,
 613          * so we won't enqueue new work to this work.
 614          *
 615          * The work is using node->node_work, so we should
 616          * flush work before freeing the memory.
 617          */
 618         flush_work(&node->node_work);
 619 #endif
 620         kfree(node);
 621 }
 622
 623 /*
 624  * register_node - Setup a sysfs device for a node.
 625  * @num - Node number to use when creating the device.
 626  *
 627  * Initialize and register the node device.
 628  */
 629 static int register_node(struct node *node, int num)
 630 {
 631         int error;
 632
 633         node->dev.id = num;
 634         node->dev.bus = &node_subsys;
 635         node->dev.release = node_device_release;
 636         node->dev.groups = node_dev_groups;
 637         error = device_register(&node->dev);
 638
 639         if (error)
 640                 put_device(&node->dev);
 641         else {
 642                 hugetlb_register_node(node);
 643
 644                 compaction_register_node(node);
 645         }
 646         return error;
 647 }
 648
 649 /**
 650  * unregister_node - unregister a node device
 651  * @node: node going away
 652  *
 653  * Unregisters a node device @node.  All the devices on the node must be
 654  * unregistered before calling this function.
 655  */
 656 void unregister_node(struct node *node)
 657 {
 658         compaction_unregister_node(node);
 659         hugetlb_unregister_node(node);          /* no-op, if memoryless node */
 660         node_remove_accesses(node);
 661         node_remove_caches(node);
 662         device_unregister(&node->dev);
 663 }
 664
 665 struct node *node_devices[MAX_NUMNODES];
 666
 667 /*
 668  * register cpu under node
 669  */
 670 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
 671 {
 672         int ret;
 673         struct device *obj;
 674
 675         if (!node_online(nid))
 676                 return 0;
 677
 678         obj = get_cpu_device(cpu);
 679         if (!obj)
 680                 return 0;
 681
 682         ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
 683                                 &obj->kobj,
 684                                 kobject_name(&obj->kobj));
 685         if (ret)
 686                 return ret;
 687
 688         return sysfs_create_link(&obj->kobj,
 689                                  &node_devices[nid]->dev.kobj,
 690                                  kobject_name(&node_devices[nid]->dev.kobj));
 691 }
 692
 693 /**
 694  * register_memory_node_under_compute_node - link memory node to its compute
 695  *                                           node for a given access class.
 696  * @mem_nid:    Memory node number
 697  * @cpu_nid:    Cpu  node number
 698  * @access:     Access class to register
 699  *
 700  * Description:
 701  *      For use with platforms that may have separate memory and compute nodes.
 702  *      This function will export node relationships linking which memory
 703  *      initiator nodes can access memory targets at a given ranked access
 704  *      class.
 705  */
 706 int register_memory_node_under_compute_node(unsigned int mem_nid,
 707                                             unsigned int cpu_nid,
 708                                             unsigned access)
 709 {
 710         struct node *init_node, *targ_node;
 711         struct node_access_nodes *initiator, *target;
 712         int ret;
 713
 714         if (!node_online(cpu_nid) || !node_online(mem_nid))
 715                 return -ENODEV;
 716
 717         init_node = node_devices[cpu_nid];
 718         targ_node = node_devices[mem_nid];
 719         initiator = node_init_node_access(init_node, access);
 720         target = node_init_node_access(targ_node, access);
 721         if (!initiator || !target)
 722                 return -ENOMEM;
 723
 724         ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
 725                                       &targ_node->dev.kobj,
 726                                       dev_name(&targ_node->dev));
 727         if (ret)
 728                 return ret;
 729
 730         ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
 731                                       &init_node->dev.kobj,
 732                                       dev_name(&init_node->dev));
 733         if (ret)
 734                 goto err;
 735
 736         return 0;
 737  err:
 738         sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
 739                                      dev_name(&targ_node->dev));
 740         return ret;
 741 }
 742
 743 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
 744 {
 745         struct device *obj;
 746
 747         if (!node_online(nid))
 748                 return 0;
 749
 750         obj = get_cpu_device(cpu);
 751         if (!obj)
 752                 return 0;
 753
 754         sysfs_remove_link(&node_devices[nid]->dev.kobj,
 755                           kobject_name(&obj->kobj));
 756         sysfs_remove_link(&obj->kobj,
 757                           kobject_name(&node_devices[nid]->dev.kobj));
 758
 759         return 0;
 760 }
 761
 762 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 763 static int __ref get_nid_for_pfn(unsigned long pfn)
 764 {
 765         if (!pfn_valid_within(pfn))
 766                 return -1;
 767 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
 768         if (system_state < SYSTEM_RUNNING)
 769                 return early_pfn_to_nid(pfn);
 770 #endif
 771         return pfn_to_nid(pfn);
 772 }
 773
 774 static void do_register_memory_block_under_node(int nid,
 775                                                 struct memory_block *mem_blk)
 776 {
 777         int ret;
 778
 779         /*
 780          * If this memory block spans multiple nodes, we only indicate
 781          * the last processed node.
 782          */
 783         mem_blk->nid = nid;
 784
 785         ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
 786                                        &mem_blk->dev.kobj,
 787                                        kobject_name(&mem_blk->dev.kobj));
 788         if (ret && ret != -EEXIST)
 789                 dev_err_ratelimited(&node_devices[nid]->dev,
 790                                     "can't create link to %s in sysfs (%d)\n",
 791                                     kobject_name(&mem_blk->dev.kobj), ret);
 792
 793         ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
 794                                 &node_devices[nid]->dev.kobj,
 795                                 kobject_name(&node_devices[nid]->dev.kobj));
 796         if (ret && ret != -EEXIST)
 797                 dev_err_ratelimited(&mem_blk->dev,
 798                                     "can't create link to %s in sysfs (%d)\n",
 799                                     kobject_name(&node_devices[nid]->dev.kobj),
 800                                     ret);
 801 }
 802
 803 /* register memory section under specified node if it spans that node */
 804 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
 805                                                void *arg)
 806 {
 807         unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
 808         unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
 809         unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
 810         int nid = *(int *)arg;
 811         unsigned long pfn;
 812
 813         for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
 814                 int page_nid;
 815
 816                 /*
 817                  * memory block could have several absent sections from start.
 818                  * skip pfn range from absent section
 819                  */
 820                 if (!pfn_in_present_section(pfn)) {
 821                         pfn = round_down(pfn + PAGES_PER_SECTION,
 822                                          PAGES_PER_SECTION) - 1;
 823                         continue;
 824                 }
 825
 826                 /*
 827                  * We need to check if page belongs to nid only at the boot
 828                  * case because node's ranges can be interleaved.
 829                  */
 830                 page_nid = get_nid_for_pfn(pfn);
 831                 if (page_nid < 0)
 832                         continue;
 833                 if (page_nid != nid)
 834                         continue;
 835
 836                 do_register_memory_block_under_node(nid, mem_blk);
 837                 return 0;
 838         }
 839         /* mem section does not span the specified node */
 840         return 0;
 841 }
 842
 843 /*
 844  * During hotplug we know that all pages in the memory block belong to the same
 845  * node.
 846  */
 847 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
 848                                                  void *arg)
 849 {
 850         int nid = *(int *)arg;
 851
 852         do_register_memory_block_under_node(nid, mem_blk);
 853         return 0;
 854 }
 855
 856 /*
 857  * Unregister a memory block device under the node it spans. Memory blocks
 858  * with multiple nodes cannot be offlined and therefore also never be removed.
 859  */
 860 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
 861 {
 862         if (mem_blk->nid == NUMA_NO_NODE)
 863                 return;
 864
 865         sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
 866                           kobject_name(&mem_blk->dev.kobj));
 867         sysfs_remove_link(&mem_blk->dev.kobj,
 868                           kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
 869 }
 870
 871 void link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn,
 872                        enum meminit_context context)
 873 {
 874         walk_memory_blocks_func_t func;
 875
 876         if (context == MEMINIT_HOTPLUG)
 877                 func = register_mem_block_under_node_hotplug;
 878         else
 879                 func = register_mem_block_under_node_early;
 880
 881         walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
 882                            (void *)&nid, func);
 883         return;
 884 }
 885
 886 #ifdef CONFIG_HUGETLBFS
 887 /*
 888  * Handle per node hstate attribute [un]registration on transistions
 889  * to/from memoryless state.
 890  */
 891 static void node_hugetlb_work(struct work_struct *work)
 892 {
 893         struct node *node = container_of(work, struct node, node_work);
 894
 895         /*
 896          * We only get here when a node transitions to/from memoryless state.
 897          * We can detect which transition occurred by examining whether the
 898          * node has memory now.  hugetlb_register_node() already check this
 899          * so we try to register the attributes.  If that fails, then the
 900          * node has transitioned to memoryless, try to unregister the
 901          * attributes.
 902          */
 903         if (!hugetlb_register_node(node))
 904                 hugetlb_unregister_node(node);
 905 }
 906
 907 static void init_node_hugetlb_work(int nid)
 908 {
 909         INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
 910 }
 911
 912 static int node_memory_callback(struct notifier_block *self,
 913                                 unsigned long action, void *arg)
 914 {
 915         struct memory_notify *mnb = arg;
 916         int nid = mnb->status_change_nid;
 917
 918         switch (action) {
 919         case MEM_ONLINE:
 920         case MEM_OFFLINE:
 921                 /*
 922                  * offload per node hstate [un]registration to a work thread
 923                  * when transitioning to/from memoryless state.
 924                  */
 925                 if (nid != NUMA_NO_NODE)
 926                         schedule_work(&node_devices[nid]->node_work);
 927                 break;
 928
 929         case MEM_GOING_ONLINE:
 930         case MEM_GOING_OFFLINE:
 931         case MEM_CANCEL_ONLINE:
 932         case MEM_CANCEL_OFFLINE:
 933         default:
 934                 break;
 935         }
 936
 937         return NOTIFY_OK;
 938 }
 939 #endif  /* CONFIG_HUGETLBFS */
 940 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 941
 942 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
 943     !defined(CONFIG_HUGETLBFS)
 944 static inline int node_memory_callback(struct notifier_block *self,
 945                                 unsigned long action, void *arg)
 946 {
 947         return NOTIFY_OK;
 948 }
 949
 950 static void init_node_hugetlb_work(int nid) { }
 951
 952 #endif
 953
 954 int __register_one_node(int nid)
 955 {
 956         int error;
 957         int cpu;
 958
 959         node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
 960         if (!node_devices[nid])
 961                 return -ENOMEM;
 962
 963         error = register_node(node_devices[nid], nid);
 964
 965         /* link cpu under this node */
 966         for_each_present_cpu(cpu) {
 967                 if (cpu_to_node(cpu) == nid)
 968                         register_cpu_under_node(cpu, nid);
 969         }
 970
 971         INIT_LIST_HEAD(&node_devices[nid]->access_list);
 972         /* initialize work queue for memory hot plug */
 973         init_node_hugetlb_work(nid);
 974         node_init_caches(nid);
 975
 976         return error;
 977 }
 978
 979 void unregister_one_node(int nid)
 980 {
 981         if (!node_devices[nid])
 982                 return;
 983
 984         unregister_node(node_devices[nid]);
 985         node_devices[nid] = NULL;
 986 }
 987
 988 /*
 989  * node states attributes
 990  */
 991
 992 struct node_attr {
 993         struct device_attribute attr;
 994         enum node_states state;
 995 };
 996
 997 static ssize_t show_node_state(struct device *dev,
 998                                struct device_attribute *attr, char *buf)
 999 {
1000         struct node_attr *na = container_of(attr, struct node_attr, attr);
1001
1002         return sysfs_emit(buf, "%*pbl\n",
1003                           nodemask_pr_args(&node_states[na->state]));
1004 }
1005
1006 #define _NODE_ATTR(name, state) \
1007         { __ATTR(name, 0444, show_node_state, NULL), state }
1008
1009 static struct node_attr node_state_attr[] = {
1010         [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1011         [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1012         [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1013 #ifdef CONFIG_HIGHMEM
1014         [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1015 #endif
1016         [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1017         [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1018         [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
1019                                            N_GENERIC_INITIATOR),
1020 };
1021
1022 static struct attribute *node_state_attrs[] = {
1023         &node_state_attr[N_POSSIBLE].attr.attr,
1024         &node_state_attr[N_ONLINE].attr.attr,
1025         &node_state_attr[N_NORMAL_MEMORY].attr.attr,
1026 #ifdef CONFIG_HIGHMEM
1027         &node_state_attr[N_HIGH_MEMORY].attr.attr,
1028 #endif
1029         &node_state_attr[N_MEMORY].attr.attr,
1030         &node_state_attr[N_CPU].attr.attr,
1031         &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
1032         NULL
1033 };
1034
1035 static struct attribute_group memory_root_attr_group = {
1036         .attrs = node_state_attrs,
1037 };
1038
1039 static const struct attribute_group *cpu_root_attr_groups[] = {
1040         &memory_root_attr_group,
1041         NULL,
1042 };
1043
1044 #define NODE_CALLBACK_PRI       2       /* lower than SLAB */
1045 static int __init register_node_type(void)
1046 {
1047         int ret;
1048
1049         BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1050         BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1051
1052         ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1053         if (!ret) {
1054                 static struct notifier_block node_memory_callback_nb = {
1055                         .notifier_call = node_memory_callback,
1056                         .priority = NODE_CALLBACK_PRI,
1057                 };
1058                 register_hotmemory_notifier(&node_memory_callback_nb);
1059         }
1060
1061         /*
1062          * Note:  we're not going to unregister the node class if we fail
1063          * to register the node state class attribute files.
1064          */
1065         return ret;
1066 }
1067 postcore_initcall(register_node_type);