xref: /openbmc/linux/drivers/base/node.c (revision d4295e12)
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/swap.h>
21 #include <linux/slab.h>
22 
23 static struct bus_type node_subsys = {
24 	.name = "node",
25 	.dev_name = "node",
26 };
27 
28 
29 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
30 {
31 	ssize_t n;
32 	cpumask_var_t mask;
33 	struct node *node_dev = to_node(dev);
34 
35 	/* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
36 	BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
37 
38 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
39 		return 0;
40 
41 	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
42 	n = cpumap_print_to_pagebuf(list, buf, mask);
43 	free_cpumask_var(mask);
44 
45 	return n;
46 }
47 
48 static inline ssize_t node_read_cpumask(struct device *dev,
49 				struct device_attribute *attr, char *buf)
50 {
51 	return node_read_cpumap(dev, false, buf);
52 }
53 static inline ssize_t node_read_cpulist(struct device *dev,
54 				struct device_attribute *attr, char *buf)
55 {
56 	return node_read_cpumap(dev, true, buf);
57 }
58 
59 static DEVICE_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
60 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
61 
62 #define K(x) ((x) << (PAGE_SHIFT - 10))
63 static ssize_t node_read_meminfo(struct device *dev,
64 			struct device_attribute *attr, char *buf)
65 {
66 	int n;
67 	int nid = dev->id;
68 	struct pglist_data *pgdat = NODE_DATA(nid);
69 	struct sysinfo i;
70 	unsigned long sreclaimable, sunreclaimable;
71 
72 	si_meminfo_node(&i, nid);
73 	sreclaimable = node_page_state(pgdat, NR_SLAB_RECLAIMABLE);
74 	sunreclaimable = node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE);
75 	n = sprintf(buf,
76 		       "Node %d MemTotal:       %8lu kB\n"
77 		       "Node %d MemFree:        %8lu kB\n"
78 		       "Node %d MemUsed:        %8lu kB\n"
79 		       "Node %d Active:         %8lu kB\n"
80 		       "Node %d Inactive:       %8lu kB\n"
81 		       "Node %d Active(anon):   %8lu kB\n"
82 		       "Node %d Inactive(anon): %8lu kB\n"
83 		       "Node %d Active(file):   %8lu kB\n"
84 		       "Node %d Inactive(file): %8lu kB\n"
85 		       "Node %d Unevictable:    %8lu kB\n"
86 		       "Node %d Mlocked:        %8lu kB\n",
87 		       nid, K(i.totalram),
88 		       nid, K(i.freeram),
89 		       nid, K(i.totalram - i.freeram),
90 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
91 				node_page_state(pgdat, NR_ACTIVE_FILE)),
92 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
93 				node_page_state(pgdat, NR_INACTIVE_FILE)),
94 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
95 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
96 		       nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
97 		       nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
98 		       nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
99 		       nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
100 
101 #ifdef CONFIG_HIGHMEM
102 	n += sprintf(buf + n,
103 		       "Node %d HighTotal:      %8lu kB\n"
104 		       "Node %d HighFree:       %8lu kB\n"
105 		       "Node %d LowTotal:       %8lu kB\n"
106 		       "Node %d LowFree:        %8lu kB\n",
107 		       nid, K(i.totalhigh),
108 		       nid, K(i.freehigh),
109 		       nid, K(i.totalram - i.totalhigh),
110 		       nid, K(i.freeram - i.freehigh));
111 #endif
112 	n += sprintf(buf + n,
113 		       "Node %d Dirty:          %8lu kB\n"
114 		       "Node %d Writeback:      %8lu kB\n"
115 		       "Node %d FilePages:      %8lu kB\n"
116 		       "Node %d Mapped:         %8lu kB\n"
117 		       "Node %d AnonPages:      %8lu kB\n"
118 		       "Node %d Shmem:          %8lu kB\n"
119 		       "Node %d KernelStack:    %8lu kB\n"
120 		       "Node %d PageTables:     %8lu kB\n"
121 		       "Node %d NFS_Unstable:   %8lu kB\n"
122 		       "Node %d Bounce:         %8lu kB\n"
123 		       "Node %d WritebackTmp:   %8lu kB\n"
124 		       "Node %d KReclaimable:   %8lu kB\n"
125 		       "Node %d Slab:           %8lu kB\n"
126 		       "Node %d SReclaimable:   %8lu kB\n"
127 		       "Node %d SUnreclaim:     %8lu kB\n"
128 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
129 		       "Node %d AnonHugePages:  %8lu kB\n"
130 		       "Node %d ShmemHugePages: %8lu kB\n"
131 		       "Node %d ShmemPmdMapped: %8lu kB\n"
132 #endif
133 			,
134 		       nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
135 		       nid, K(node_page_state(pgdat, NR_WRITEBACK)),
136 		       nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
137 		       nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
138 		       nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
139 		       nid, K(i.sharedram),
140 		       nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
141 		       nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
142 		       nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
143 		       nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
144 		       nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
145 		       nid, K(sreclaimable +
146 			      node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
147 		       nid, K(sreclaimable + sunreclaimable),
148 		       nid, K(sreclaimable),
149 		       nid, K(sunreclaimable)
150 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
151 		       ,
152 		       nid, K(node_page_state(pgdat, NR_ANON_THPS) *
153 				       HPAGE_PMD_NR),
154 		       nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
155 				       HPAGE_PMD_NR),
156 		       nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
157 				       HPAGE_PMD_NR)
158 #endif
159 		       );
160 	n += hugetlb_report_node_meminfo(nid, buf + n);
161 	return n;
162 }
163 
164 #undef K
165 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
166 
167 static ssize_t node_read_numastat(struct device *dev,
168 				struct device_attribute *attr, char *buf)
169 {
170 	return sprintf(buf,
171 		       "numa_hit %lu\n"
172 		       "numa_miss %lu\n"
173 		       "numa_foreign %lu\n"
174 		       "interleave_hit %lu\n"
175 		       "local_node %lu\n"
176 		       "other_node %lu\n",
177 		       sum_zone_numa_state(dev->id, NUMA_HIT),
178 		       sum_zone_numa_state(dev->id, NUMA_MISS),
179 		       sum_zone_numa_state(dev->id, NUMA_FOREIGN),
180 		       sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
181 		       sum_zone_numa_state(dev->id, NUMA_LOCAL),
182 		       sum_zone_numa_state(dev->id, NUMA_OTHER));
183 }
184 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
185 
186 static ssize_t node_read_vmstat(struct device *dev,
187 				struct device_attribute *attr, char *buf)
188 {
189 	int nid = dev->id;
190 	struct pglist_data *pgdat = NODE_DATA(nid);
191 	int i;
192 	int n = 0;
193 
194 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
195 		n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
196 			     sum_zone_node_page_state(nid, i));
197 
198 #ifdef CONFIG_NUMA
199 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
200 		n += sprintf(buf+n, "%s %lu\n",
201 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
202 			     sum_zone_numa_state(nid, i));
203 #endif
204 
205 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
206 		n += sprintf(buf+n, "%s %lu\n",
207 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
208 			     NR_VM_NUMA_STAT_ITEMS],
209 			     node_page_state(pgdat, i));
210 
211 	return n;
212 }
213 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
214 
215 static ssize_t node_read_distance(struct device *dev,
216 			struct device_attribute *attr, char *buf)
217 {
218 	int nid = dev->id;
219 	int len = 0;
220 	int i;
221 
222 	/*
223 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
224 	 * at the most (distance + space or newline).
225 	 */
226 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
227 
228 	for_each_online_node(i)
229 		len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
230 
231 	len += sprintf(buf + len, "\n");
232 	return len;
233 }
234 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
235 
236 static struct attribute *node_dev_attrs[] = {
237 	&dev_attr_cpumap.attr,
238 	&dev_attr_cpulist.attr,
239 	&dev_attr_meminfo.attr,
240 	&dev_attr_numastat.attr,
241 	&dev_attr_distance.attr,
242 	&dev_attr_vmstat.attr,
243 	NULL
244 };
245 ATTRIBUTE_GROUPS(node_dev);
246 
247 #ifdef CONFIG_HUGETLBFS
248 /*
249  * hugetlbfs per node attributes registration interface:
250  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
251  * it will register its per node attributes for all online nodes with
252  * memory.  It will also call register_hugetlbfs_with_node(), below, to
253  * register its attribute registration functions with this node driver.
254  * Once these hooks have been initialized, the node driver will call into
255  * the hugetlb module to [un]register attributes for hot-plugged nodes.
256  */
257 static node_registration_func_t __hugetlb_register_node;
258 static node_registration_func_t __hugetlb_unregister_node;
259 
260 static inline bool hugetlb_register_node(struct node *node)
261 {
262 	if (__hugetlb_register_node &&
263 			node_state(node->dev.id, N_MEMORY)) {
264 		__hugetlb_register_node(node);
265 		return true;
266 	}
267 	return false;
268 }
269 
270 static inline void hugetlb_unregister_node(struct node *node)
271 {
272 	if (__hugetlb_unregister_node)
273 		__hugetlb_unregister_node(node);
274 }
275 
276 void register_hugetlbfs_with_node(node_registration_func_t doregister,
277 				  node_registration_func_t unregister)
278 {
279 	__hugetlb_register_node   = doregister;
280 	__hugetlb_unregister_node = unregister;
281 }
282 #else
283 static inline void hugetlb_register_node(struct node *node) {}
284 
285 static inline void hugetlb_unregister_node(struct node *node) {}
286 #endif
287 
288 static void node_device_release(struct device *dev)
289 {
290 	struct node *node = to_node(dev);
291 
292 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
293 	/*
294 	 * We schedule the work only when a memory section is
295 	 * onlined/offlined on this node. When we come here,
296 	 * all the memory on this node has been offlined,
297 	 * so we won't enqueue new work to this work.
298 	 *
299 	 * The work is using node->node_work, so we should
300 	 * flush work before freeing the memory.
301 	 */
302 	flush_work(&node->node_work);
303 #endif
304 	kfree(node);
305 }
306 
307 /*
308  * register_node - Setup a sysfs device for a node.
309  * @num - Node number to use when creating the device.
310  *
311  * Initialize and register the node device.
312  */
313 static int register_node(struct node *node, int num)
314 {
315 	int error;
316 
317 	node->dev.id = num;
318 	node->dev.bus = &node_subsys;
319 	node->dev.release = node_device_release;
320 	node->dev.groups = node_dev_groups;
321 	error = device_register(&node->dev);
322 
323 	if (error)
324 		put_device(&node->dev);
325 	else {
326 		hugetlb_register_node(node);
327 
328 		compaction_register_node(node);
329 	}
330 	return error;
331 }
332 
333 /**
334  * unregister_node - unregister a node device
335  * @node: node going away
336  *
337  * Unregisters a node device @node.  All the devices on the node must be
338  * unregistered before calling this function.
339  */
340 void unregister_node(struct node *node)
341 {
342 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
343 
344 	device_unregister(&node->dev);
345 }
346 
347 struct node *node_devices[MAX_NUMNODES];
348 
349 /*
350  * register cpu under node
351  */
352 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
353 {
354 	int ret;
355 	struct device *obj;
356 
357 	if (!node_online(nid))
358 		return 0;
359 
360 	obj = get_cpu_device(cpu);
361 	if (!obj)
362 		return 0;
363 
364 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
365 				&obj->kobj,
366 				kobject_name(&obj->kobj));
367 	if (ret)
368 		return ret;
369 
370 	return sysfs_create_link(&obj->kobj,
371 				 &node_devices[nid]->dev.kobj,
372 				 kobject_name(&node_devices[nid]->dev.kobj));
373 }
374 
375 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
376 {
377 	struct device *obj;
378 
379 	if (!node_online(nid))
380 		return 0;
381 
382 	obj = get_cpu_device(cpu);
383 	if (!obj)
384 		return 0;
385 
386 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
387 			  kobject_name(&obj->kobj));
388 	sysfs_remove_link(&obj->kobj,
389 			  kobject_name(&node_devices[nid]->dev.kobj));
390 
391 	return 0;
392 }
393 
394 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
395 static int __ref get_nid_for_pfn(unsigned long pfn)
396 {
397 	if (!pfn_valid_within(pfn))
398 		return -1;
399 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
400 	if (system_state < SYSTEM_RUNNING)
401 		return early_pfn_to_nid(pfn);
402 #endif
403 	return pfn_to_nid(pfn);
404 }
405 
406 /* register memory section under specified node if it spans that node */
407 int register_mem_sect_under_node(struct memory_block *mem_blk, void *arg)
408 {
409 	int ret, nid = *(int *)arg;
410 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
411 
412 	mem_blk->nid = nid;
413 
414 	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
415 	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
416 	sect_end_pfn += PAGES_PER_SECTION - 1;
417 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
418 		int page_nid;
419 
420 		/*
421 		 * memory block could have several absent sections from start.
422 		 * skip pfn range from absent section
423 		 */
424 		if (!pfn_present(pfn)) {
425 			pfn = round_down(pfn + PAGES_PER_SECTION,
426 					 PAGES_PER_SECTION) - 1;
427 			continue;
428 		}
429 
430 		/*
431 		 * We need to check if page belongs to nid only for the boot
432 		 * case, during hotplug we know that all pages in the memory
433 		 * block belong to the same node.
434 		 */
435 		if (system_state == SYSTEM_BOOTING) {
436 			page_nid = get_nid_for_pfn(pfn);
437 			if (page_nid < 0)
438 				continue;
439 			if (page_nid != nid)
440 				continue;
441 		}
442 		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
443 					&mem_blk->dev.kobj,
444 					kobject_name(&mem_blk->dev.kobj));
445 		if (ret)
446 			return ret;
447 
448 		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
449 				&node_devices[nid]->dev.kobj,
450 				kobject_name(&node_devices[nid]->dev.kobj));
451 	}
452 	/* mem section does not span the specified node */
453 	return 0;
454 }
455 
456 /* unregister memory section under all nodes that it spans */
457 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
458 				    unsigned long phys_index)
459 {
460 	NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
461 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
462 
463 	if (!mem_blk) {
464 		NODEMASK_FREE(unlinked_nodes);
465 		return -EFAULT;
466 	}
467 	if (!unlinked_nodes)
468 		return -ENOMEM;
469 	nodes_clear(*unlinked_nodes);
470 
471 	sect_start_pfn = section_nr_to_pfn(phys_index);
472 	sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
473 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
474 		int nid;
475 
476 		nid = get_nid_for_pfn(pfn);
477 		if (nid < 0)
478 			continue;
479 		if (!node_online(nid))
480 			continue;
481 		if (node_test_and_set(nid, *unlinked_nodes))
482 			continue;
483 		sysfs_remove_link(&node_devices[nid]->dev.kobj,
484 			 kobject_name(&mem_blk->dev.kobj));
485 		sysfs_remove_link(&mem_blk->dev.kobj,
486 			 kobject_name(&node_devices[nid]->dev.kobj));
487 	}
488 	NODEMASK_FREE(unlinked_nodes);
489 	return 0;
490 }
491 
492 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn)
493 {
494 	return walk_memory_range(start_pfn, end_pfn, (void *)&nid,
495 					register_mem_sect_under_node);
496 }
497 
498 #ifdef CONFIG_HUGETLBFS
499 /*
500  * Handle per node hstate attribute [un]registration on transistions
501  * to/from memoryless state.
502  */
503 static void node_hugetlb_work(struct work_struct *work)
504 {
505 	struct node *node = container_of(work, struct node, node_work);
506 
507 	/*
508 	 * We only get here when a node transitions to/from memoryless state.
509 	 * We can detect which transition occurred by examining whether the
510 	 * node has memory now.  hugetlb_register_node() already check this
511 	 * so we try to register the attributes.  If that fails, then the
512 	 * node has transitioned to memoryless, try to unregister the
513 	 * attributes.
514 	 */
515 	if (!hugetlb_register_node(node))
516 		hugetlb_unregister_node(node);
517 }
518 
519 static void init_node_hugetlb_work(int nid)
520 {
521 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
522 }
523 
524 static int node_memory_callback(struct notifier_block *self,
525 				unsigned long action, void *arg)
526 {
527 	struct memory_notify *mnb = arg;
528 	int nid = mnb->status_change_nid;
529 
530 	switch (action) {
531 	case MEM_ONLINE:
532 	case MEM_OFFLINE:
533 		/*
534 		 * offload per node hstate [un]registration to a work thread
535 		 * when transitioning to/from memoryless state.
536 		 */
537 		if (nid != NUMA_NO_NODE)
538 			schedule_work(&node_devices[nid]->node_work);
539 		break;
540 
541 	case MEM_GOING_ONLINE:
542 	case MEM_GOING_OFFLINE:
543 	case MEM_CANCEL_ONLINE:
544 	case MEM_CANCEL_OFFLINE:
545 	default:
546 		break;
547 	}
548 
549 	return NOTIFY_OK;
550 }
551 #endif	/* CONFIG_HUGETLBFS */
552 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
553 
554 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
555     !defined(CONFIG_HUGETLBFS)
556 static inline int node_memory_callback(struct notifier_block *self,
557 				unsigned long action, void *arg)
558 {
559 	return NOTIFY_OK;
560 }
561 
562 static void init_node_hugetlb_work(int nid) { }
563 
564 #endif
565 
566 int __register_one_node(int nid)
567 {
568 	int error;
569 	int cpu;
570 
571 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
572 	if (!node_devices[nid])
573 		return -ENOMEM;
574 
575 	error = register_node(node_devices[nid], nid);
576 
577 	/* link cpu under this node */
578 	for_each_present_cpu(cpu) {
579 		if (cpu_to_node(cpu) == nid)
580 			register_cpu_under_node(cpu, nid);
581 	}
582 
583 	/* initialize work queue for memory hot plug */
584 	init_node_hugetlb_work(nid);
585 
586 	return error;
587 }
588 
589 void unregister_one_node(int nid)
590 {
591 	if (!node_devices[nid])
592 		return;
593 
594 	unregister_node(node_devices[nid]);
595 	node_devices[nid] = NULL;
596 }
597 
598 /*
599  * node states attributes
600  */
601 
602 static ssize_t print_nodes_state(enum node_states state, char *buf)
603 {
604 	int n;
605 
606 	n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
607 		      nodemask_pr_args(&node_states[state]));
608 	buf[n++] = '\n';
609 	buf[n] = '\0';
610 	return n;
611 }
612 
613 struct node_attr {
614 	struct device_attribute attr;
615 	enum node_states state;
616 };
617 
618 static ssize_t show_node_state(struct device *dev,
619 			       struct device_attribute *attr, char *buf)
620 {
621 	struct node_attr *na = container_of(attr, struct node_attr, attr);
622 	return print_nodes_state(na->state, buf);
623 }
624 
625 #define _NODE_ATTR(name, state) \
626 	{ __ATTR(name, 0444, show_node_state, NULL), state }
627 
628 static struct node_attr node_state_attr[] = {
629 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
630 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
631 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
632 #ifdef CONFIG_HIGHMEM
633 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
634 #endif
635 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
636 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
637 };
638 
639 static struct attribute *node_state_attrs[] = {
640 	&node_state_attr[N_POSSIBLE].attr.attr,
641 	&node_state_attr[N_ONLINE].attr.attr,
642 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
643 #ifdef CONFIG_HIGHMEM
644 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
645 #endif
646 	&node_state_attr[N_MEMORY].attr.attr,
647 	&node_state_attr[N_CPU].attr.attr,
648 	NULL
649 };
650 
651 static struct attribute_group memory_root_attr_group = {
652 	.attrs = node_state_attrs,
653 };
654 
655 static const struct attribute_group *cpu_root_attr_groups[] = {
656 	&memory_root_attr_group,
657 	NULL,
658 };
659 
660 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
661 static int __init register_node_type(void)
662 {
663 	int ret;
664 
665  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
666  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
667 
668 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
669 	if (!ret) {
670 		static struct notifier_block node_memory_callback_nb = {
671 			.notifier_call = node_memory_callback,
672 			.priority = NODE_CALLBACK_PRI,
673 		};
674 		register_hotmemory_notifier(&node_memory_callback_nb);
675 	}
676 
677 	/*
678 	 * Note:  we're not going to unregister the node class if we fail
679 	 * to register the node state class attribute files.
680 	 */
681 	return ret;
682 }
683 postcore_initcall(register_node_type);
684