node.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713
  1. /*
  2. * Basic Node interface support
  3. */
  4. #include <linux/module.h>
  5. #include <linux/init.h>
  6. #include <linux/mm.h>
  7. #include <linux/memory.h>
  8. #include <linux/vmstat.h>
  9. #include <linux/notifier.h>
  10. #include <linux/node.h>
  11. #include <linux/hugetlb.h>
  12. #include <linux/compaction.h>
  13. #include <linux/cpumask.h>
  14. #include <linux/topology.h>
  15. #include <linux/nodemask.h>
  16. #include <linux/cpu.h>
  17. #include <linux/device.h>
  18. #include <linux/swap.h>
  19. #include <linux/slab.h>
  20. static struct bus_type node_subsys = {
  21. .name = "node",
  22. .dev_name = "node",
  23. };
  24. static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
  25. {
  26. struct node *node_dev = to_node(dev);
  27. const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
  28. /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
  29. BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
  30. return cpumap_print_to_pagebuf(list, buf, mask);
  31. }
  32. static inline ssize_t node_read_cpumask(struct device *dev,
  33. struct device_attribute *attr, char *buf)
  34. {
  35. return node_read_cpumap(dev, false, buf);
  36. }
  37. static inline ssize_t node_read_cpulist(struct device *dev,
  38. struct device_attribute *attr, char *buf)
  39. {
  40. return node_read_cpumap(dev, true, buf);
  41. }
  42. static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
  43. static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
  44. #define K(x) ((x) << (PAGE_SHIFT - 10))
  45. static ssize_t node_read_meminfo(struct device *dev,
  46. struct device_attribute *attr, char *buf)
  47. {
  48. int n;
  49. int nid = dev->id;
  50. struct pglist_data *pgdat = NODE_DATA(nid);
  51. struct sysinfo i;
  52. si_meminfo_node(&i, nid);
  53. n = sprintf(buf,
  54. "Node %d MemTotal: %8lu kB\n"
  55. "Node %d MemFree: %8lu kB\n"
  56. "Node %d MemUsed: %8lu kB\n"
  57. "Node %d Active: %8lu kB\n"
  58. "Node %d Inactive: %8lu kB\n"
  59. "Node %d Active(anon): %8lu kB\n"
  60. "Node %d Inactive(anon): %8lu kB\n"
  61. "Node %d Active(file): %8lu kB\n"
  62. "Node %d Inactive(file): %8lu kB\n"
  63. "Node %d Unevictable: %8lu kB\n"
  64. "Node %d Mlocked: %8lu kB\n",
  65. nid, K(i.totalram),
  66. nid, K(i.freeram),
  67. nid, K(i.totalram - i.freeram),
  68. nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
  69. node_page_state(pgdat, NR_ACTIVE_FILE)),
  70. nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
  71. node_page_state(pgdat, NR_INACTIVE_FILE)),
  72. nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
  73. nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
  74. nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
  75. nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
  76. nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
  77. nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
  78. #ifdef CONFIG_HIGHMEM
  79. n += sprintf(buf + n,
  80. "Node %d HighTotal: %8lu kB\n"
  81. "Node %d HighFree: %8lu kB\n"
  82. "Node %d LowTotal: %8lu kB\n"
  83. "Node %d LowFree: %8lu kB\n",
  84. nid, K(i.totalhigh),
  85. nid, K(i.freehigh),
  86. nid, K(i.totalram - i.totalhigh),
  87. nid, K(i.freeram - i.freehigh));
  88. #endif
  89. n += sprintf(buf + n,
  90. "Node %d Dirty: %8lu kB\n"
  91. "Node %d Writeback: %8lu kB\n"
  92. "Node %d FilePages: %8lu kB\n"
  93. "Node %d Mapped: %8lu kB\n"
  94. "Node %d AnonPages: %8lu kB\n"
  95. "Node %d Shmem: %8lu kB\n"
  96. "Node %d KernelStack: %8lu kB\n"
  97. "Node %d PageTables: %8lu kB\n"
  98. "Node %d NFS_Unstable: %8lu kB\n"
  99. "Node %d Bounce: %8lu kB\n"
  100. "Node %d WritebackTmp: %8lu kB\n"
  101. "Node %d Slab: %8lu kB\n"
  102. "Node %d SReclaimable: %8lu kB\n"
  103. "Node %d SUnreclaim: %8lu kB\n"
  104. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  105. "Node %d AnonHugePages: %8lu kB\n"
  106. "Node %d ShmemHugePages: %8lu kB\n"
  107. "Node %d ShmemPmdMapped: %8lu kB\n"
  108. #endif
  109. ,
  110. nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
  111. nid, K(node_page_state(pgdat, NR_WRITEBACK)),
  112. nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
  113. nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
  114. nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
  115. nid, K(i.sharedram),
  116. nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
  117. nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
  118. nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
  119. nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
  120. nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
  121. nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE) +
  122. sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
  123. nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE)),
  124. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  125. nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
  126. nid, K(node_page_state(pgdat, NR_ANON_THPS) *
  127. HPAGE_PMD_NR),
  128. nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
  129. HPAGE_PMD_NR),
  130. nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
  131. HPAGE_PMD_NR));
  132. #else
  133. nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
  134. #endif
  135. n += hugetlb_report_node_meminfo(nid, buf + n);
  136. return n;
  137. }
  138. #undef K
  139. static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
  140. static ssize_t node_read_numastat(struct device *dev,
  141. struct device_attribute *attr, char *buf)
  142. {
  143. return sprintf(buf,
  144. "numa_hit %lu\n"
  145. "numa_miss %lu\n"
  146. "numa_foreign %lu\n"
  147. "interleave_hit %lu\n"
  148. "local_node %lu\n"
  149. "other_node %lu\n",
  150. sum_zone_node_page_state(dev->id, NUMA_HIT),
  151. sum_zone_node_page_state(dev->id, NUMA_MISS),
  152. sum_zone_node_page_state(dev->id, NUMA_FOREIGN),
  153. sum_zone_node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
  154. sum_zone_node_page_state(dev->id, NUMA_LOCAL),
  155. sum_zone_node_page_state(dev->id, NUMA_OTHER));
  156. }
  157. static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
  158. static ssize_t node_read_vmstat(struct device *dev,
  159. struct device_attribute *attr, char *buf)
  160. {
  161. int nid = dev->id;
  162. struct pglist_data *pgdat = NODE_DATA(nid);
  163. int i;
  164. int n = 0;
  165. for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
  166. n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
  167. sum_zone_node_page_state(nid, i));
  168. for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
  169. n += sprintf(buf+n, "%s %lu\n",
  170. vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
  171. node_page_state(pgdat, i));
  172. return n;
  173. }
  174. static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
  175. static ssize_t node_read_distance(struct device *dev,
  176. struct device_attribute *attr, char *buf)
  177. {
  178. int nid = dev->id;
  179. int len = 0;
  180. int i;
  181. /*
  182. * buf is currently PAGE_SIZE in length and each node needs 4 chars
  183. * at the most (distance + space or newline).
  184. */
  185. BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
  186. for_each_online_node(i)
  187. len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
  188. len += sprintf(buf + len, "\n");
  189. return len;
  190. }
  191. static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
  192. static struct attribute *node_dev_attrs[] = {
  193. &dev_attr_cpumap.attr,
  194. &dev_attr_cpulist.attr,
  195. &dev_attr_meminfo.attr,
  196. &dev_attr_numastat.attr,
  197. &dev_attr_distance.attr,
  198. &dev_attr_vmstat.attr,
  199. NULL
  200. };
  201. ATTRIBUTE_GROUPS(node_dev);
  202. #ifdef CONFIG_HUGETLBFS
  203. /*
  204. * hugetlbfs per node attributes registration interface:
  205. * When/if hugetlb[fs] subsystem initializes [sometime after this module],
  206. * it will register its per node attributes for all online nodes with
  207. * memory. It will also call register_hugetlbfs_with_node(), below, to
  208. * register its attribute registration functions with this node driver.
  209. * Once these hooks have been initialized, the node driver will call into
  210. * the hugetlb module to [un]register attributes for hot-plugged nodes.
  211. */
  212. static node_registration_func_t __hugetlb_register_node;
  213. static node_registration_func_t __hugetlb_unregister_node;
  214. static inline bool hugetlb_register_node(struct node *node)
  215. {
  216. if (__hugetlb_register_node &&
  217. node_state(node->dev.id, N_MEMORY)) {
  218. __hugetlb_register_node(node);
  219. return true;
  220. }
  221. return false;
  222. }
  223. static inline void hugetlb_unregister_node(struct node *node)
  224. {
  225. if (__hugetlb_unregister_node)
  226. __hugetlb_unregister_node(node);
  227. }
  228. void register_hugetlbfs_with_node(node_registration_func_t doregister,
  229. node_registration_func_t unregister)
  230. {
  231. __hugetlb_register_node = doregister;
  232. __hugetlb_unregister_node = unregister;
  233. }
  234. #else
  235. static inline void hugetlb_register_node(struct node *node) {}
  236. static inline void hugetlb_unregister_node(struct node *node) {}
  237. #endif
  238. static void node_device_release(struct device *dev)
  239. {
  240. struct node *node = to_node(dev);
  241. #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
  242. /*
  243. * We schedule the work only when a memory section is
  244. * onlined/offlined on this node. When we come here,
  245. * all the memory on this node has been offlined,
  246. * so we won't enqueue new work to this work.
  247. *
  248. * The work is using node->node_work, so we should
  249. * flush work before freeing the memory.
  250. */
  251. flush_work(&node->node_work);
  252. #endif
  253. kfree(node);
  254. }
  255. /*
  256. * register_node - Setup a sysfs device for a node.
  257. * @num - Node number to use when creating the device.
  258. *
  259. * Initialize and register the node device.
  260. */
  261. static int register_node(struct node *node, int num, struct node *parent)
  262. {
  263. int error;
  264. node->dev.id = num;
  265. node->dev.bus = &node_subsys;
  266. node->dev.release = node_device_release;
  267. node->dev.groups = node_dev_groups;
  268. error = device_register(&node->dev);
  269. if (!error){
  270. hugetlb_register_node(node);
  271. compaction_register_node(node);
  272. }
  273. return error;
  274. }
  275. /**
  276. * unregister_node - unregister a node device
  277. * @node: node going away
  278. *
  279. * Unregisters a node device @node. All the devices on the node must be
  280. * unregistered before calling this function.
  281. */
  282. void unregister_node(struct node *node)
  283. {
  284. hugetlb_unregister_node(node); /* no-op, if memoryless node */
  285. device_unregister(&node->dev);
  286. }
  287. struct node *node_devices[MAX_NUMNODES];
  288. /*
  289. * register cpu under node
  290. */
  291. int register_cpu_under_node(unsigned int cpu, unsigned int nid)
  292. {
  293. int ret;
  294. struct device *obj;
  295. if (!node_online(nid))
  296. return 0;
  297. obj = get_cpu_device(cpu);
  298. if (!obj)
  299. return 0;
  300. ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
  301. &obj->kobj,
  302. kobject_name(&obj->kobj));
  303. if (ret)
  304. return ret;
  305. return sysfs_create_link(&obj->kobj,
  306. &node_devices[nid]->dev.kobj,
  307. kobject_name(&node_devices[nid]->dev.kobj));
  308. }
  309. int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
  310. {
  311. struct device *obj;
  312. if (!node_online(nid))
  313. return 0;
  314. obj = get_cpu_device(cpu);
  315. if (!obj)
  316. return 0;
  317. sysfs_remove_link(&node_devices[nid]->dev.kobj,
  318. kobject_name(&obj->kobj));
  319. sysfs_remove_link(&obj->kobj,
  320. kobject_name(&node_devices[nid]->dev.kobj));
  321. return 0;
  322. }
  323. #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  324. #define page_initialized(page) (page->lru.next)
  325. static int __ref get_nid_for_pfn(unsigned long pfn)
  326. {
  327. struct page *page;
  328. if (!pfn_valid_within(pfn))
  329. return -1;
  330. #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
  331. if (system_state == SYSTEM_BOOTING)
  332. return early_pfn_to_nid(pfn);
  333. #endif
  334. page = pfn_to_page(pfn);
  335. if (!page_initialized(page))
  336. return -1;
  337. return pfn_to_nid(pfn);
  338. }
  339. /* register memory section under specified node if it spans that node */
  340. int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
  341. {
  342. int ret;
  343. unsigned long pfn, sect_start_pfn, sect_end_pfn;
  344. if (!mem_blk)
  345. return -EFAULT;
  346. if (!node_online(nid))
  347. return 0;
  348. sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
  349. sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
  350. sect_end_pfn += PAGES_PER_SECTION - 1;
  351. for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
  352. int page_nid;
  353. /*
  354. * memory block could have several absent sections from start.
  355. * skip pfn range from absent section
  356. */
  357. if (!pfn_present(pfn)) {
  358. pfn = round_down(pfn + PAGES_PER_SECTION,
  359. PAGES_PER_SECTION) - 1;
  360. continue;
  361. }
  362. page_nid = get_nid_for_pfn(pfn);
  363. if (page_nid < 0)
  364. continue;
  365. if (page_nid != nid)
  366. continue;
  367. ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
  368. &mem_blk->dev.kobj,
  369. kobject_name(&mem_blk->dev.kobj));
  370. if (ret)
  371. return ret;
  372. return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
  373. &node_devices[nid]->dev.kobj,
  374. kobject_name(&node_devices[nid]->dev.kobj));
  375. }
  376. /* mem section does not span the specified node */
  377. return 0;
  378. }
  379. /* unregister memory section under all nodes that it spans */
  380. int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
  381. unsigned long phys_index)
  382. {
  383. NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
  384. unsigned long pfn, sect_start_pfn, sect_end_pfn;
  385. if (!mem_blk) {
  386. NODEMASK_FREE(unlinked_nodes);
  387. return -EFAULT;
  388. }
  389. if (!unlinked_nodes)
  390. return -ENOMEM;
  391. nodes_clear(*unlinked_nodes);
  392. sect_start_pfn = section_nr_to_pfn(phys_index);
  393. sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
  394. for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
  395. int nid;
  396. nid = get_nid_for_pfn(pfn);
  397. if (nid < 0)
  398. continue;
  399. if (!node_online(nid))
  400. continue;
  401. if (node_test_and_set(nid, *unlinked_nodes))
  402. continue;
  403. sysfs_remove_link(&node_devices[nid]->dev.kobj,
  404. kobject_name(&mem_blk->dev.kobj));
  405. sysfs_remove_link(&mem_blk->dev.kobj,
  406. kobject_name(&node_devices[nid]->dev.kobj));
  407. }
  408. NODEMASK_FREE(unlinked_nodes);
  409. return 0;
  410. }
  411. static int link_mem_sections(int nid)
  412. {
  413. unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
  414. unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
  415. unsigned long pfn;
  416. struct memory_block *mem_blk = NULL;
  417. int err = 0;
  418. for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
  419. unsigned long section_nr = pfn_to_section_nr(pfn);
  420. struct mem_section *mem_sect;
  421. int ret;
  422. if (!present_section_nr(section_nr))
  423. continue;
  424. mem_sect = __nr_to_section(section_nr);
  425. /* same memblock ? */
  426. if (mem_blk)
  427. if ((section_nr >= mem_blk->start_section_nr) &&
  428. (section_nr <= mem_blk->end_section_nr))
  429. continue;
  430. mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
  431. ret = register_mem_sect_under_node(mem_blk, nid);
  432. if (!err)
  433. err = ret;
  434. /* discard ref obtained in find_memory_block() */
  435. }
  436. if (mem_blk)
  437. kobject_put(&mem_blk->dev.kobj);
  438. return err;
  439. }
  440. #ifdef CONFIG_HUGETLBFS
  441. /*
  442. * Handle per node hstate attribute [un]registration on transistions
  443. * to/from memoryless state.
  444. */
  445. static void node_hugetlb_work(struct work_struct *work)
  446. {
  447. struct node *node = container_of(work, struct node, node_work);
  448. /*
  449. * We only get here when a node transitions to/from memoryless state.
  450. * We can detect which transition occurred by examining whether the
  451. * node has memory now. hugetlb_register_node() already check this
  452. * so we try to register the attributes. If that fails, then the
  453. * node has transitioned to memoryless, try to unregister the
  454. * attributes.
  455. */
  456. if (!hugetlb_register_node(node))
  457. hugetlb_unregister_node(node);
  458. }
  459. static void init_node_hugetlb_work(int nid)
  460. {
  461. INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
  462. }
  463. static int node_memory_callback(struct notifier_block *self,
  464. unsigned long action, void *arg)
  465. {
  466. struct memory_notify *mnb = arg;
  467. int nid = mnb->status_change_nid;
  468. switch (action) {
  469. case MEM_ONLINE:
  470. case MEM_OFFLINE:
  471. /*
  472. * offload per node hstate [un]registration to a work thread
  473. * when transitioning to/from memoryless state.
  474. */
  475. if (nid != NUMA_NO_NODE)
  476. schedule_work(&node_devices[nid]->node_work);
  477. break;
  478. case MEM_GOING_ONLINE:
  479. case MEM_GOING_OFFLINE:
  480. case MEM_CANCEL_ONLINE:
  481. case MEM_CANCEL_OFFLINE:
  482. default:
  483. break;
  484. }
  485. return NOTIFY_OK;
  486. }
  487. #endif /* CONFIG_HUGETLBFS */
  488. #else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
  489. static int link_mem_sections(int nid) { return 0; }
  490. #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
  491. #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
  492. !defined(CONFIG_HUGETLBFS)
  493. static inline int node_memory_callback(struct notifier_block *self,
  494. unsigned long action, void *arg)
  495. {
  496. return NOTIFY_OK;
  497. }
  498. static void init_node_hugetlb_work(int nid) { }
  499. #endif
  500. int register_one_node(int nid)
  501. {
  502. int error = 0;
  503. int cpu;
  504. if (node_online(nid)) {
  505. int p_node = parent_node(nid);
  506. struct node *parent = NULL;
  507. if (p_node != nid)
  508. parent = node_devices[p_node];
  509. node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
  510. if (!node_devices[nid])
  511. return -ENOMEM;
  512. error = register_node(node_devices[nid], nid, parent);
  513. /* link cpu under this node */
  514. for_each_present_cpu(cpu) {
  515. if (cpu_to_node(cpu) == nid)
  516. register_cpu_under_node(cpu, nid);
  517. }
  518. /* link memory sections under this node */
  519. error = link_mem_sections(nid);
  520. /* initialize work queue for memory hot plug */
  521. init_node_hugetlb_work(nid);
  522. }
  523. return error;
  524. }
  525. void unregister_one_node(int nid)
  526. {
  527. if (!node_devices[nid])
  528. return;
  529. unregister_node(node_devices[nid]);
  530. node_devices[nid] = NULL;
  531. }
  532. /*
  533. * node states attributes
  534. */
  535. static ssize_t print_nodes_state(enum node_states state, char *buf)
  536. {
  537. int n;
  538. n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
  539. nodemask_pr_args(&node_states[state]));
  540. buf[n++] = '\n';
  541. buf[n] = '\0';
  542. return n;
  543. }
  544. struct node_attr {
  545. struct device_attribute attr;
  546. enum node_states state;
  547. };
  548. static ssize_t show_node_state(struct device *dev,
  549. struct device_attribute *attr, char *buf)
  550. {
  551. struct node_attr *na = container_of(attr, struct node_attr, attr);
  552. return print_nodes_state(na->state, buf);
  553. }
  554. #define _NODE_ATTR(name, state) \
  555. { __ATTR(name, 0444, show_node_state, NULL), state }
  556. static struct node_attr node_state_attr[] = {
  557. [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
  558. [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
  559. [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
  560. #ifdef CONFIG_HIGHMEM
  561. [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
  562. #endif
  563. #ifdef CONFIG_MOVABLE_NODE
  564. [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
  565. #endif
  566. [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
  567. };
  568. static struct attribute *node_state_attrs[] = {
  569. &node_state_attr[N_POSSIBLE].attr.attr,
  570. &node_state_attr[N_ONLINE].attr.attr,
  571. &node_state_attr[N_NORMAL_MEMORY].attr.attr,
  572. #ifdef CONFIG_HIGHMEM
  573. &node_state_attr[N_HIGH_MEMORY].attr.attr,
  574. #endif
  575. #ifdef CONFIG_MOVABLE_NODE
  576. &node_state_attr[N_MEMORY].attr.attr,
  577. #endif
  578. &node_state_attr[N_CPU].attr.attr,
  579. NULL
  580. };
  581. static struct attribute_group memory_root_attr_group = {
  582. .attrs = node_state_attrs,
  583. };
  584. static const struct attribute_group *cpu_root_attr_groups[] = {
  585. &memory_root_attr_group,
  586. NULL,
  587. };
  588. #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
  589. static int __init register_node_type(void)
  590. {
  591. int ret;
  592. BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
  593. BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
  594. ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
  595. if (!ret) {
  596. static struct notifier_block node_memory_callback_nb = {
  597. .notifier_call = node_memory_callback,
  598. .priority = NODE_CALLBACK_PRI,
  599. };
  600. register_hotmemory_notifier(&node_memory_callback_nb);
  601. }
  602. /*
  603. * Note: we're not going to unregister the node class if we fail
  604. * to register the node state class attribute files.
  605. */
  606. return ret;
  607. }
  608. postcore_initcall(register_node_type);