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#include <linux/memblock.h>
  25
  26static const struct bus_type node_subsys = {
  27	.name = "node",
  28	.dev_name = "node",
  29};
  30
  31static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
  32				  const struct bin_attribute *attr, char *buf,
  33				  loff_t off, size_t count)
  34{
  35	struct device *dev = kobj_to_dev(kobj);
  36	struct node *node_dev = to_node(dev);
  37	cpumask_var_t mask;
  38	ssize_t n;
  39
  40	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
  41		return 0;
  42
  43	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
  44	n = cpumap_print_bitmask_to_buf(buf, mask, off, count);
  45	free_cpumask_var(mask);
  46
  47	return n;
  48}
  49
  50static const BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
  51
  52static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
  53				   const struct bin_attribute *attr, char *buf,
  54				   loff_t off, size_t count)
  55{
  56	struct device *dev = kobj_to_dev(kobj);
  57	struct node *node_dev = to_node(dev);
  58	cpumask_var_t mask;
  59	ssize_t n;
  60
  61	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
  62		return 0;
  63
  64	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
  65	n = cpumap_print_list_to_buf(buf, mask, off, count);
  66	free_cpumask_var(mask);
  67
  68	return n;
  69}
  70
  71static const BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
  72
  73/**
  74 * struct node_access_nodes - Access class device to hold user visible
  75 * 			      relationships to other nodes.
  76 * @dev:	Device for this memory access class
  77 * @list_node:	List element in the node's access list
  78 * @access:	The access class rank
  79 * @coord:	Heterogeneous memory performance coordinates
  80 */
  81struct node_access_nodes {
  82	struct device		dev;
  83	struct list_head	list_node;
  84	unsigned int		access;
  85#ifdef CONFIG_HMEM_REPORTING
  86	struct access_coordinate	coord;
  87#endif
  88};
  89#define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
  90
  91static struct attribute *node_init_access_node_attrs[] = {
  92	NULL,
  93};
  94
  95static struct attribute *node_targ_access_node_attrs[] = {
  96	NULL,
  97};
  98
  99static const struct attribute_group initiators = {
 100	.name	= "initiators",
 101	.attrs	= node_init_access_node_attrs,
 102};
 103
 104static const struct attribute_group targets = {
 105	.name	= "targets",
 106	.attrs	= node_targ_access_node_attrs,
 107};
 108
 109static const struct attribute_group *node_access_node_groups[] = {
 110	&initiators,
 111	&targets,
 112	NULL,
 113};
 114
 115#ifdef CONFIG_MEMORY_HOTPLUG
 116static BLOCKING_NOTIFIER_HEAD(node_chain);
 117
 118int register_node_notifier(struct notifier_block *nb)
 119{
 120	return blocking_notifier_chain_register(&node_chain, nb);
 121}
 122EXPORT_SYMBOL(register_node_notifier);
 123
 124void unregister_node_notifier(struct notifier_block *nb)
 125{
 126	blocking_notifier_chain_unregister(&node_chain, nb);
 127}
 128EXPORT_SYMBOL(unregister_node_notifier);
 129
 130int node_notify(unsigned long val, void *v)
 131{
 132	return blocking_notifier_call_chain(&node_chain, val, v);
 133}
 134#endif
 135
 136static void node_remove_accesses(struct node *node)
 137{
 138	struct node_access_nodes *c, *cnext;
 139
 140	list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
 141		list_del(&c->list_node);
 142		device_unregister(&c->dev);
 143	}
 144}
 145
 146static void node_access_release(struct device *dev)
 147{
 148	kfree(to_access_nodes(dev));
 149}
 150
 151static struct node_access_nodes *node_init_node_access(struct node *node,
 152						       enum access_coordinate_class access)
 153{
 154	struct node_access_nodes *access_node;
 155	struct device *dev;
 156
 157	list_for_each_entry(access_node, &node->access_list, list_node)
 158		if (access_node->access == access)
 159			return access_node;
 160
 161	access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
 162	if (!access_node)
 163		return NULL;
 164
 165	access_node->access = access;
 166	dev = &access_node->dev;
 167	dev->parent = &node->dev;
 168	dev->release = node_access_release;
 169	dev->groups = node_access_node_groups;
 170	if (dev_set_name(dev, "access%u", access))
 171		goto free;
 172
 173	if (device_register(dev))
 174		goto free_name;
 175
 176	pm_runtime_no_callbacks(dev);
 177	list_add_tail(&access_node->list_node, &node->access_list);
 178	return access_node;
 179free_name:
 180	kfree_const(dev->kobj.name);
 181free:
 182	kfree(access_node);
 183	return NULL;
 184}
 185
 186#ifdef CONFIG_HMEM_REPORTING
 187#define ACCESS_ATTR(property)						\
 188static ssize_t property##_show(struct device *dev,			\
 189			   struct device_attribute *attr,		\
 190			   char *buf)					\
 191{									\
 192	return sysfs_emit(buf, "%u\n",					\
 193			  to_access_nodes(dev)->coord.property);	\
 194}									\
 195static DEVICE_ATTR_RO(property)
 196
 197ACCESS_ATTR(read_bandwidth);
 198ACCESS_ATTR(read_latency);
 199ACCESS_ATTR(write_bandwidth);
 200ACCESS_ATTR(write_latency);
 201
 202static struct attribute *access_attrs[] = {
 203	&dev_attr_read_bandwidth.attr,
 204	&dev_attr_read_latency.attr,
 205	&dev_attr_write_bandwidth.attr,
 206	&dev_attr_write_latency.attr,
 207	NULL,
 208};
 209
 210/**
 211 * node_set_perf_attrs - Set the performance values for given access class
 212 * @nid: Node identifier to be set
 213 * @coord: Heterogeneous memory performance coordinates
 214 * @access: The access class the for the given attributes
 215 */
 216void node_set_perf_attrs(unsigned int nid, struct access_coordinate *coord,
 217			 enum access_coordinate_class access)
 218{
 219	struct node_access_nodes *c;
 220	struct node *node;
 221	int i;
 222
 223	if (WARN_ON_ONCE(!node_online(nid)))
 224		return;
 225
 226	node = node_devices[nid];
 227	c = node_init_node_access(node, access);
 228	if (!c)
 229		return;
 230
 231	c->coord = *coord;
 232	for (i = 0; access_attrs[i] != NULL; i++) {
 233		if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
 234					    "initiators")) {
 235			pr_info("failed to add performance attribute to node %d\n",
 236				nid);
 237			break;
 238		}
 239	}
 240
 241	/* When setting CPU access coordinates, update mempolicy */
 242	if (access == ACCESS_COORDINATE_CPU) {
 243		if (mempolicy_set_node_perf(nid, coord)) {
 244			pr_info("failed to set mempolicy attrs for node %d\n",
 245				nid);
 246		}
 247	}
 248}
 249EXPORT_SYMBOL_GPL(node_set_perf_attrs);
 250
 251/**
 252 * node_update_perf_attrs - Update the performance values for given access class
 253 * @nid: Node identifier to be updated
 254 * @coord: Heterogeneous memory performance coordinates
 255 * @access: The access class for the given attributes
 256 */
 257void node_update_perf_attrs(unsigned int nid, struct access_coordinate *coord,
 258			    enum access_coordinate_class access)
 259{
 260	struct node_access_nodes *access_node;
 261	struct node *node;
 262	int i;
 263
 264	if (WARN_ON_ONCE(!node_online(nid)))
 265		return;
 266
 267	node = node_devices[nid];
 268	list_for_each_entry(access_node, &node->access_list, list_node) {
 269		if (access_node->access != access)
 270			continue;
 271
 272		access_node->coord = *coord;
 273		for (i = 0; access_attrs[i]; i++) {
 274			sysfs_notify(&access_node->dev.kobj,
 275				     NULL, access_attrs[i]->name);
 276		}
 277		break;
 278	}
 279
 280	/* When setting CPU access coordinates, update mempolicy */
 281	if (access != ACCESS_COORDINATE_CPU)
 282		return;
 283
 284	if (mempolicy_set_node_perf(nid, coord))
 285		pr_info("failed to set mempolicy attrs for node %d\n", nid);
 286}
 287EXPORT_SYMBOL_GPL(node_update_perf_attrs);
 288
 289/**
 290 * struct node_cache_info - Internal tracking for memory node caches
 291 * @dev:	Device represeting the cache level
 292 * @node:	List element for tracking in the node
 293 * @cache_attrs:Attributes for this cache level
 294 */
 295struct node_cache_info {
 296	struct device dev;
 297	struct list_head node;
 298	struct node_cache_attrs cache_attrs;
 299};
 300#define to_cache_info(device) container_of(device, struct node_cache_info, dev)
 301
 302#define CACHE_ATTR(name, fmt) 						\
 303static ssize_t name##_show(struct device *dev,				\
 304			   struct device_attribute *attr,		\
 305			   char *buf)					\
 306{									\
 307	return sysfs_emit(buf, fmt "\n",				\
 308			  to_cache_info(dev)->cache_attrs.name);	\
 309}									\
 310static DEVICE_ATTR_RO(name);
 311
 312CACHE_ATTR(size, "%llu")
 313CACHE_ATTR(line_size, "%u")
 314CACHE_ATTR(indexing, "%u")
 315CACHE_ATTR(write_policy, "%u")
 316CACHE_ATTR(address_mode, "%#x")
 317
 318static struct attribute *cache_attrs[] = {
 319	&dev_attr_indexing.attr,
 320	&dev_attr_size.attr,
 321	&dev_attr_line_size.attr,
 322	&dev_attr_write_policy.attr,
 323	&dev_attr_address_mode.attr,
 324	NULL,
 325};
 326ATTRIBUTE_GROUPS(cache);
 327
 328static void node_cache_release(struct device *dev)
 329{
 330	kfree(dev);
 331}
 332
 333static void node_cacheinfo_release(struct device *dev)
 334{
 335	struct node_cache_info *info = to_cache_info(dev);
 336	kfree(info);
 337}
 338
 339static void node_init_cache_dev(struct node *node)
 340{
 341	struct device *dev;
 342
 343	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 344	if (!dev)
 345		return;
 346
 347	device_initialize(dev);
 348	dev->parent = &node->dev;
 349	dev->release = node_cache_release;
 350	if (dev_set_name(dev, "memory_side_cache"))
 351		goto put_device;
 352
 353	if (device_add(dev))
 354		goto put_device;
 355
 356	pm_runtime_no_callbacks(dev);
 357	node->cache_dev = dev;
 358	return;
 359put_device:
 360	put_device(dev);
 361}
 362
 363/**
 364 * node_add_cache() - add cache attribute to a memory node
 365 * @nid: Node identifier that has new cache attributes
 366 * @cache_attrs: Attributes for the cache being added
 367 */
 368void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
 369{
 370	struct node_cache_info *info;
 371	struct device *dev;
 372	struct node *node;
 373
 374	if (!node_online(nid) || !node_devices[nid])
 375		return;
 376
 377	node = node_devices[nid];
 378	list_for_each_entry(info, &node->cache_attrs, node) {
 379		if (info->cache_attrs.level == cache_attrs->level) {
 380			dev_warn(&node->dev,
 381				"attempt to add duplicate cache level:%d\n",
 382				cache_attrs->level);
 383			return;
 384		}
 385	}
 386
 387	if (!node->cache_dev)
 388		node_init_cache_dev(node);
 389	if (!node->cache_dev)
 390		return;
 391
 392	info = kzalloc(sizeof(*info), GFP_KERNEL);
 393	if (!info)
 394		return;
 395
 396	dev = &info->dev;
 397	device_initialize(dev);
 398	dev->parent = node->cache_dev;
 399	dev->release = node_cacheinfo_release;
 400	dev->groups = cache_groups;
 401	if (dev_set_name(dev, "index%d", cache_attrs->level))
 402		goto put_device;
 403
 404	info->cache_attrs = *cache_attrs;
 405	if (device_add(dev)) {
 406		dev_warn(&node->dev, "failed to add cache level:%d\n",
 407			 cache_attrs->level);
 408		goto put_device;
 409	}
 410	pm_runtime_no_callbacks(dev);
 411	list_add_tail(&info->node, &node->cache_attrs);
 412	return;
 413put_device:
 414	put_device(dev);
 415}
 416
 417static void node_remove_caches(struct node *node)
 418{
 419	struct node_cache_info *info, *next;
 420
 421	if (!node->cache_dev)
 422		return;
 423
 424	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
 425		list_del(&info->node);
 426		device_unregister(&info->dev);
 427	}
 428	device_unregister(node->cache_dev);
 429}
 430
 431static void node_init_caches(unsigned int nid)
 432{
 433	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
 434}
 435#else
 436static void node_init_caches(unsigned int nid) { }
 437static void node_remove_caches(struct node *node) { }
 438#endif
 439
 440#define K(x) ((x) << (PAGE_SHIFT - 10))
 441static ssize_t node_read_meminfo(struct device *dev,
 442			struct device_attribute *attr, char *buf)
 443{
 444	int len = 0;
 445	int nid = dev->id;
 446	struct pglist_data *pgdat = NODE_DATA(nid);
 447	struct sysinfo i;
 448	unsigned long sreclaimable, sunreclaimable;
 449	unsigned long swapcached = 0;
 450
 451	si_meminfo_node(&i, nid);
 452	sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
 453	sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
 454#ifdef CONFIG_SWAP
 455	swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
 456#endif
 457	len = sysfs_emit_at(buf, len,
 458			    "Node %d MemTotal:       %8lu kB\n"
 459			    "Node %d MemFree:        %8lu kB\n"
 460			    "Node %d MemUsed:        %8lu kB\n"
 461			    "Node %d SwapCached:     %8lu kB\n"
 462			    "Node %d Active:         %8lu kB\n"
 463			    "Node %d Inactive:       %8lu kB\n"
 464			    "Node %d Active(anon):   %8lu kB\n"
 465			    "Node %d Inactive(anon): %8lu kB\n"
 466			    "Node %d Active(file):   %8lu kB\n"
 467			    "Node %d Inactive(file): %8lu kB\n"
 468			    "Node %d Unevictable:    %8lu kB\n"
 469			    "Node %d Mlocked:        %8lu kB\n",
 470			    nid, K(i.totalram),
 471			    nid, K(i.freeram),
 472			    nid, K(i.totalram - i.freeram),
 473			    nid, K(swapcached),
 474			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
 475				   node_page_state(pgdat, NR_ACTIVE_FILE)),
 476			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
 477				   node_page_state(pgdat, NR_INACTIVE_FILE)),
 478			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
 479			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
 480			    nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
 481			    nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
 482			    nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
 483			    nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
 484
 485#ifdef CONFIG_HIGHMEM
 486	len += sysfs_emit_at(buf, len,
 487			     "Node %d HighTotal:      %8lu kB\n"
 488			     "Node %d HighFree:       %8lu kB\n"
 489			     "Node %d LowTotal:       %8lu kB\n"
 490			     "Node %d LowFree:        %8lu kB\n",
 491			     nid, K(i.totalhigh),
 492			     nid, K(i.freehigh),
 493			     nid, K(i.totalram - i.totalhigh),
 494			     nid, K(i.freeram - i.freehigh));
 495#endif
 496	len += sysfs_emit_at(buf, len,
 497			     "Node %d Dirty:          %8lu kB\n"
 498			     "Node %d Writeback:      %8lu kB\n"
 499			     "Node %d FilePages:      %8lu kB\n"
 500			     "Node %d Mapped:         %8lu kB\n"
 501			     "Node %d AnonPages:      %8lu kB\n"
 502			     "Node %d Shmem:          %8lu kB\n"
 503			     "Node %d KernelStack:    %8lu kB\n"
 504#ifdef CONFIG_SHADOW_CALL_STACK
 505			     "Node %d ShadowCallStack:%8lu kB\n"
 506#endif
 507			     "Node %d PageTables:     %8lu kB\n"
 508			     "Node %d SecPageTables:  %8lu kB\n"
 509			     "Node %d NFS_Unstable:   %8lu kB\n"
 510			     "Node %d Bounce:         %8lu kB\n"
 511			     "Node %d WritebackTmp:   %8lu kB\n"
 512			     "Node %d KReclaimable:   %8lu kB\n"
 513			     "Node %d Slab:           %8lu kB\n"
 514			     "Node %d SReclaimable:   %8lu kB\n"
 515			     "Node %d SUnreclaim:     %8lu kB\n"
 516#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 517			     "Node %d AnonHugePages:  %8lu kB\n"
 518			     "Node %d ShmemHugePages: %8lu kB\n"
 519			     "Node %d ShmemPmdMapped: %8lu kB\n"
 520			     "Node %d FileHugePages:  %8lu kB\n"
 521			     "Node %d FilePmdMapped:  %8lu kB\n"
 522#endif
 523#ifdef CONFIG_UNACCEPTED_MEMORY
 524			     "Node %d Unaccepted:     %8lu kB\n"
 525#endif
 526			     ,
 527			     nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
 528			     nid, K(node_page_state(pgdat, NR_WRITEBACK)),
 529			     nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
 530			     nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
 531			     nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
 532			     nid, K(i.sharedram),
 533			     nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
 534#ifdef CONFIG_SHADOW_CALL_STACK
 535			     nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
 536#endif
 537			     nid, K(node_page_state(pgdat, NR_PAGETABLE)),
 538			     nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
 539			     nid, 0UL,
 540			     nid, 0UL,
 541			     nid, 0UL,
 542			     nid, K(sreclaimable +
 543				    node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
 544			     nid, K(sreclaimable + sunreclaimable),
 545			     nid, K(sreclaimable),
 546			     nid, K(sunreclaimable)
 547#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 548			     ,
 549			     nid, K(node_page_state(pgdat, NR_ANON_THPS)),
 550			     nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
 551			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
 552			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
 553			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
 554#endif
 555#ifdef CONFIG_UNACCEPTED_MEMORY
 556			     ,
 557			     nid, K(sum_zone_node_page_state(nid, NR_UNACCEPTED))
 558#endif
 559			    );
 560	len += hugetlb_report_node_meminfo(buf, len, nid);
 561	return len;
 562}
 563
 564#undef K
 565static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
 566
 567static ssize_t node_read_numastat(struct device *dev,
 568				  struct device_attribute *attr, char *buf)
 569{
 570	fold_vm_numa_events();
 571	return sysfs_emit(buf,
 572			  "numa_hit %lu\n"
 573			  "numa_miss %lu\n"
 574			  "numa_foreign %lu\n"
 575			  "interleave_hit %lu\n"
 576			  "local_node %lu\n"
 577			  "other_node %lu\n",
 578			  sum_zone_numa_event_state(dev->id, NUMA_HIT),
 579			  sum_zone_numa_event_state(dev->id, NUMA_MISS),
 580			  sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
 581			  sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
 582			  sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
 583			  sum_zone_numa_event_state(dev->id, NUMA_OTHER));
 584}
 585static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
 586
 587static ssize_t node_read_vmstat(struct device *dev,
 588				struct device_attribute *attr, char *buf)
 589{
 590	int nid = dev->id;
 591	struct pglist_data *pgdat = NODE_DATA(nid);
 592	int i;
 593	int len = 0;
 594
 595	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
 596		len += sysfs_emit_at(buf, len, "%s %lu\n",
 597				     zone_stat_name(i),
 598				     sum_zone_node_page_state(nid, i));
 599
 600#ifdef CONFIG_NUMA
 601	fold_vm_numa_events();
 602	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
 603		len += sysfs_emit_at(buf, len, "%s %lu\n",
 604				     numa_stat_name(i),
 605				     sum_zone_numa_event_state(nid, i));
 606
 607#endif
 608	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
 609		unsigned long pages = node_page_state_pages(pgdat, i);
 610
 611		if (vmstat_item_print_in_thp(i))
 612			pages /= HPAGE_PMD_NR;
 613		len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
 614				     pages);
 615	}
 616
 617	return len;
 618}
 619static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
 620
 621static ssize_t node_read_distance(struct device *dev,
 622				  struct device_attribute *attr, char *buf)
 623{
 624	int nid = dev->id;
 625	int len = 0;
 626	int i;
 627
 628	/*
 629	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
 630	 * at the most (distance + space or newline).
 631	 */
 632	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
 633
 634	for_each_online_node(i) {
 635		len += sysfs_emit_at(buf, len, "%s%d",
 636				     i ? " " : "", node_distance(nid, i));
 637	}
 638
 639	len += sysfs_emit_at(buf, len, "\n");
 640	return len;
 641}
 642static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
 643
 644static struct attribute *node_dev_attrs[] = {
 645	&dev_attr_meminfo.attr,
 646	&dev_attr_numastat.attr,
 647	&dev_attr_distance.attr,
 648	&dev_attr_vmstat.attr,
 649	NULL
 650};
 651
 652static const struct bin_attribute *node_dev_bin_attrs[] = {
 653	&bin_attr_cpumap,
 654	&bin_attr_cpulist,
 655	NULL
 656};
 657
 658static const struct attribute_group node_dev_group = {
 659	.attrs = node_dev_attrs,
 660	.bin_attrs = node_dev_bin_attrs,
 661};
 662
 663static const struct attribute_group *node_dev_groups[] = {
 664	&node_dev_group,
 665#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
 666	&arch_node_dev_group,
 667#endif
 668#ifdef CONFIG_MEMORY_FAILURE
 669	&memory_failure_attr_group,
 670#endif
 671	NULL
 672};
 673
 674static void node_device_release(struct device *dev)
 675{
 676	kfree(to_node(dev));
 677}
 678
 679struct node *node_devices[MAX_NUMNODES];
 680
 681/*
 682 * register cpu under node
 683 */
 684int register_cpu_under_node(unsigned int cpu, unsigned int nid)
 685{
 686	int ret;
 687	struct device *obj;
 688
 689	if (!node_online(nid))
 690		return 0;
 691
 692	obj = get_cpu_device(cpu);
 693	if (!obj)
 694		return 0;
 695
 696	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
 697				&obj->kobj,
 698				kobject_name(&obj->kobj));
 699	if (ret)
 700		return ret;
 701
 702	return sysfs_create_link(&obj->kobj,
 703				 &node_devices[nid]->dev.kobj,
 704				 kobject_name(&node_devices[nid]->dev.kobj));
 705}
 706
 707/**
 708 * register_memory_node_under_compute_node - link memory node to its compute
 709 *					     node for a given access class.
 710 * @mem_nid:	Memory node number
 711 * @cpu_nid:	Cpu  node number
 712 * @access:	Access class to register
 713 *
 714 * Description:
 715 * 	For use with platforms that may have separate memory and compute nodes.
 716 * 	This function will export node relationships linking which memory
 717 * 	initiator nodes can access memory targets at a given ranked access
 718 * 	class.
 719 */
 720int register_memory_node_under_compute_node(unsigned int mem_nid,
 721					    unsigned int cpu_nid,
 722					    enum access_coordinate_class access)
 723{
 724	struct node *init_node, *targ_node;
 725	struct node_access_nodes *initiator, *target;
 726	int ret;
 727
 728	if (!node_online(cpu_nid) || !node_online(mem_nid))
 729		return -ENODEV;
 730
 731	init_node = node_devices[cpu_nid];
 732	targ_node = node_devices[mem_nid];
 733	initiator = node_init_node_access(init_node, access);
 734	target = node_init_node_access(targ_node, access);
 735	if (!initiator || !target)
 736		return -ENOMEM;
 737
 738	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
 739				      &targ_node->dev.kobj,
 740				      dev_name(&targ_node->dev));
 741	if (ret)
 742		return ret;
 743
 744	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
 745				      &init_node->dev.kobj,
 746				      dev_name(&init_node->dev));
 747	if (ret)
 748		goto err;
 749
 750	return 0;
 751 err:
 752	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
 753				     dev_name(&targ_node->dev));
 754	return ret;
 755}
 756
 757int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
 758{
 759	struct device *obj;
 760
 761	if (!node_online(nid))
 762		return 0;
 763
 764	obj = get_cpu_device(cpu);
 765	if (!obj)
 766		return 0;
 767
 768	sysfs_remove_link(&node_devices[nid]->dev.kobj,
 769			  kobject_name(&obj->kobj));
 770	sysfs_remove_link(&obj->kobj,
 771			  kobject_name(&node_devices[nid]->dev.kobj));
 772
 773	return 0;
 774}
 775
 776#ifdef CONFIG_MEMORY_HOTPLUG
 777static void do_register_memory_block_under_node(int nid,
 778						struct memory_block *mem_blk)
 779{
 780	int ret;
 781
 782	ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
 783				       &mem_blk->dev.kobj,
 784				       kobject_name(&mem_blk->dev.kobj));
 785	if (ret && ret != -EEXIST)
 786		dev_err_ratelimited(&node_devices[nid]->dev,
 787				    "can't create link to %s in sysfs (%d)\n",
 788				    kobject_name(&mem_blk->dev.kobj), ret);
 789
 790	ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
 791				&node_devices[nid]->dev.kobj,
 792				kobject_name(&node_devices[nid]->dev.kobj));
 793	if (ret && ret != -EEXIST)
 794		dev_err_ratelimited(&mem_blk->dev,
 795				    "can't create link to %s in sysfs (%d)\n",
 796				    kobject_name(&node_devices[nid]->dev.kobj),
 797				    ret);
 798}
 799
 800/*
 801 * During hotplug we know that all pages in the memory block belong to the same
 802 * node.
 803 */
 804static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
 805						 void *arg)
 806{
 807	int nid = *(int *)arg;
 808
 809	do_register_memory_block_under_node(nid, mem_blk);
 810	return 0;
 811}
 812
 813/*
 814 * Unregister a memory block device under the node it spans. Memory blocks
 815 * with multiple nodes cannot be offlined and therefore also never be removed.
 816 */
 817void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
 818{
 819	if (mem_blk->nid == NUMA_NO_NODE)
 820		return;
 821
 822	sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
 823			  kobject_name(&mem_blk->dev.kobj));
 824	sysfs_remove_link(&mem_blk->dev.kobj,
 825			  kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
 826}
 827
 828/* register all memory blocks under the corresponding nodes */
 829static void register_memory_blocks_under_nodes(void)
 830{
 831	struct memblock_region *r;
 832
 833	for_each_mem_region(r) {
 834		const unsigned long start_block_id = phys_to_block_id(r->base);
 835		const unsigned long end_block_id = phys_to_block_id(r->base + r->size - 1);
 836		const int nid = memblock_get_region_node(r);
 837		unsigned long block_id;
 838
 839		if (!node_online(nid))
 840			continue;
 841
 842		for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
 843			struct memory_block *mem;
 844
 845			mem = find_memory_block_by_id(block_id);
 846			if (!mem)
 847				continue;
 848
 849			memory_block_add_nid_early(mem, nid);
 850			do_register_memory_block_under_node(nid, mem);
 851			put_device(&mem->dev);
 852		}
 853
 854	}
 855}
 856
 857void register_memory_blocks_under_node_hotplug(int nid, unsigned long start_pfn,
 858					       unsigned long end_pfn)
 859{
 860	walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
 861			   (void *)&nid, register_mem_block_under_node_hotplug);
 862	return;
 863}
 864#endif /* CONFIG_MEMORY_HOTPLUG */
 865
 866/**
 867 * register_node - Initialize and register the node device.
 868 * @nid: Node number to use when creating the device.
 869 *
 870 * Return: 0 on success, -errno otherwise
 871 */
 872int register_node(int nid)
 873{
 874	int error;
 875	int cpu;
 876	struct node *node;
 877
 878	node = kzalloc(sizeof(struct node), GFP_KERNEL);
 879	if (!node)
 880		return -ENOMEM;
 881
 882	INIT_LIST_HEAD(&node->access_list);
 883
 884	node->dev.id = nid;
 885	node->dev.bus = &node_subsys;
 886	node->dev.release = node_device_release;
 887	node->dev.groups = node_dev_groups;
 888
 889	error = device_register(&node->dev);
 890	if (error) {
 891		put_device(&node->dev);
 892		return error;
 893	}
 894
 895	node_devices[nid] = node;
 896	hugetlb_register_node(node);
 897	compaction_register_node(node);
 898	reclaim_register_node(node);
 899
 900	/* link cpu under this node */
 901	for_each_present_cpu(cpu) {
 902		if (cpu_to_node(cpu) == nid)
 903			register_cpu_under_node(cpu, nid);
 904	}
 905
 906	node_init_caches(nid);
 907
 908	return error;
 909}
 910/**
 911 * unregister_node - unregister a node device
 912 * @nid: nid of the node going away
 913 *
 914 * Unregisters the node device at node id @nid. All the devices on the
 915 * node must be unregistered before calling this function.
 916 */
 917void unregister_node(int nid)
 918{
 919	struct node *node = node_devices[nid];
 920
 921	if (!node)
 922		return;
 923
 924	hugetlb_unregister_node(node);
 925	compaction_unregister_node(node);
 926	reclaim_unregister_node(node);
 927	node_remove_accesses(node);
 928	node_remove_caches(node);
 929	device_unregister(&node->dev);
 930	node_devices[nid] = NULL;
 931}
 932
 933/*
 934 * node states attributes
 935 */
 936
 937struct node_attr {
 938	struct device_attribute attr;
 939	enum node_states state;
 940};
 941
 942static ssize_t show_node_state(struct device *dev,
 943			       struct device_attribute *attr, char *buf)
 944{
 945	struct node_attr *na = container_of(attr, struct node_attr, attr);
 946
 947	return sysfs_emit(buf, "%*pbl\n",
 948			  nodemask_pr_args(&node_states[na->state]));
 949}
 950
 951#define _NODE_ATTR(name, state) \
 952	{ __ATTR(name, 0444, show_node_state, NULL), state }
 953
 954static struct node_attr node_state_attr[] = {
 955	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
 956	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
 957	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
 958#ifdef CONFIG_HIGHMEM
 959	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
 960#endif
 961	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
 962	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
 963	[N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
 964					   N_GENERIC_INITIATOR),
 965};
 966
 967static struct attribute *node_state_attrs[] = {
 968	&node_state_attr[N_POSSIBLE].attr.attr,
 969	&node_state_attr[N_ONLINE].attr.attr,
 970	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
 971#ifdef CONFIG_HIGHMEM
 972	&node_state_attr[N_HIGH_MEMORY].attr.attr,
 973#endif
 974	&node_state_attr[N_MEMORY].attr.attr,
 975	&node_state_attr[N_CPU].attr.attr,
 976	&node_state_attr[N_GENERIC_INITIATOR].attr.attr,
 977	NULL
 978};
 979
 980static const struct attribute_group memory_root_attr_group = {
 981	.attrs = node_state_attrs,
 982};
 983
 984static const struct attribute_group *cpu_root_attr_groups[] = {
 985	&memory_root_attr_group,
 986	NULL,
 987};
 988
 989void __init node_dev_init(void)
 990{
 991	int ret, i;
 992
 993 	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
 994 	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
 995
 996	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
 997	if (ret)
 998		panic("%s() failed to register subsystem: %d\n", __func__, ret);
 999
1000	/*
1001	 * Create all node devices, which will properly link the node
1002	 * to already created cpu devices.
1003	 */
1004	for_each_online_node(i) {
1005		ret =  register_node(i);
1006		if (ret)
1007			panic("%s() failed to add node: %d\n", __func__, ret);
1008	}
1009
1010	register_memory_blocks_under_nodes();
1011}