xref: /openbmc/linux/drivers/base/node.c (revision 74ce1896)
1 /*
2  * Basic Node interface support
3  */
4 
5 #include <linux/module.h>
6 #include <linux/init.h>
7 #include <linux/mm.h>
8 #include <linux/memory.h>
9 #include <linux/vmstat.h>
10 #include <linux/notifier.h>
11 #include <linux/node.h>
12 #include <linux/hugetlb.h>
13 #include <linux/compaction.h>
14 #include <linux/cpumask.h>
15 #include <linux/topology.h>
16 #include <linux/nodemask.h>
17 #include <linux/cpu.h>
18 #include <linux/device.h>
19 #include <linux/swap.h>
20 #include <linux/slab.h>
21 
22 static struct bus_type node_subsys = {
23 	.name = "node",
24 	.dev_name = "node",
25 };
26 
27 
28 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
29 {
30 	struct node *node_dev = to_node(dev);
31 	const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
32 
33 	/* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
34 	BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
35 
36 	return cpumap_print_to_pagebuf(list, buf, mask);
37 }
38 
39 static inline ssize_t node_read_cpumask(struct device *dev,
40 				struct device_attribute *attr, char *buf)
41 {
42 	return node_read_cpumap(dev, false, buf);
43 }
44 static inline ssize_t node_read_cpulist(struct device *dev,
45 				struct device_attribute *attr, char *buf)
46 {
47 	return node_read_cpumap(dev, true, buf);
48 }
49 
50 static DEVICE_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
51 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
52 
53 #define K(x) ((x) << (PAGE_SHIFT - 10))
54 static ssize_t node_read_meminfo(struct device *dev,
55 			struct device_attribute *attr, char *buf)
56 {
57 	int n;
58 	int nid = dev->id;
59 	struct pglist_data *pgdat = NODE_DATA(nid);
60 	struct sysinfo i;
61 
62 	si_meminfo_node(&i, nid);
63 	n = sprintf(buf,
64 		       "Node %d MemTotal:       %8lu kB\n"
65 		       "Node %d MemFree:        %8lu kB\n"
66 		       "Node %d MemUsed:        %8lu kB\n"
67 		       "Node %d Active:         %8lu kB\n"
68 		       "Node %d Inactive:       %8lu kB\n"
69 		       "Node %d Active(anon):   %8lu kB\n"
70 		       "Node %d Inactive(anon): %8lu kB\n"
71 		       "Node %d Active(file):   %8lu kB\n"
72 		       "Node %d Inactive(file): %8lu kB\n"
73 		       "Node %d Unevictable:    %8lu kB\n"
74 		       "Node %d Mlocked:        %8lu kB\n",
75 		       nid, K(i.totalram),
76 		       nid, K(i.freeram),
77 		       nid, K(i.totalram - i.freeram),
78 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
79 				node_page_state(pgdat, NR_ACTIVE_FILE)),
80 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
81 				node_page_state(pgdat, NR_INACTIVE_FILE)),
82 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
83 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
84 		       nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
85 		       nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
86 		       nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
87 		       nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
88 
89 #ifdef CONFIG_HIGHMEM
90 	n += sprintf(buf + n,
91 		       "Node %d HighTotal:      %8lu kB\n"
92 		       "Node %d HighFree:       %8lu kB\n"
93 		       "Node %d LowTotal:       %8lu kB\n"
94 		       "Node %d LowFree:        %8lu kB\n",
95 		       nid, K(i.totalhigh),
96 		       nid, K(i.freehigh),
97 		       nid, K(i.totalram - i.totalhigh),
98 		       nid, K(i.freeram - i.freehigh));
99 #endif
100 	n += sprintf(buf + n,
101 		       "Node %d Dirty:          %8lu kB\n"
102 		       "Node %d Writeback:      %8lu kB\n"
103 		       "Node %d FilePages:      %8lu kB\n"
104 		       "Node %d Mapped:         %8lu kB\n"
105 		       "Node %d AnonPages:      %8lu kB\n"
106 		       "Node %d Shmem:          %8lu kB\n"
107 		       "Node %d KernelStack:    %8lu kB\n"
108 		       "Node %d PageTables:     %8lu kB\n"
109 		       "Node %d NFS_Unstable:   %8lu kB\n"
110 		       "Node %d Bounce:         %8lu kB\n"
111 		       "Node %d WritebackTmp:   %8lu kB\n"
112 		       "Node %d Slab:           %8lu kB\n"
113 		       "Node %d SReclaimable:   %8lu kB\n"
114 		       "Node %d SUnreclaim:     %8lu kB\n"
115 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
116 		       "Node %d AnonHugePages:  %8lu kB\n"
117 		       "Node %d ShmemHugePages: %8lu kB\n"
118 		       "Node %d ShmemPmdMapped: %8lu kB\n"
119 #endif
120 			,
121 		       nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
122 		       nid, K(node_page_state(pgdat, NR_WRITEBACK)),
123 		       nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
124 		       nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
125 		       nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
126 		       nid, K(i.sharedram),
127 		       nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
128 		       nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
129 		       nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
130 		       nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
131 		       nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
132 		       nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE) +
133 			      node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
134 		       nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE)),
135 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
136 		       nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
137 		       nid, K(node_page_state(pgdat, NR_ANON_THPS) *
138 				       HPAGE_PMD_NR),
139 		       nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
140 				       HPAGE_PMD_NR),
141 		       nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
142 				       HPAGE_PMD_NR));
143 #else
144 		       nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)));
145 #endif
146 	n += hugetlb_report_node_meminfo(nid, buf + n);
147 	return n;
148 }
149 
150 #undef K
151 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
152 
153 static ssize_t node_read_numastat(struct device *dev,
154 				struct device_attribute *attr, char *buf)
155 {
156 	return sprintf(buf,
157 		       "numa_hit %lu\n"
158 		       "numa_miss %lu\n"
159 		       "numa_foreign %lu\n"
160 		       "interleave_hit %lu\n"
161 		       "local_node %lu\n"
162 		       "other_node %lu\n",
163 		       sum_zone_numa_state(dev->id, NUMA_HIT),
164 		       sum_zone_numa_state(dev->id, NUMA_MISS),
165 		       sum_zone_numa_state(dev->id, NUMA_FOREIGN),
166 		       sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
167 		       sum_zone_numa_state(dev->id, NUMA_LOCAL),
168 		       sum_zone_numa_state(dev->id, NUMA_OTHER));
169 }
170 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
171 
172 static ssize_t node_read_vmstat(struct device *dev,
173 				struct device_attribute *attr, char *buf)
174 {
175 	int nid = dev->id;
176 	struct pglist_data *pgdat = NODE_DATA(nid);
177 	int i;
178 	int n = 0;
179 
180 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
181 		n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
182 			     sum_zone_node_page_state(nid, i));
183 
184 #ifdef CONFIG_NUMA
185 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
186 		n += sprintf(buf+n, "%s %lu\n",
187 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
188 			     sum_zone_numa_state(nid, i));
189 #endif
190 
191 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
192 		n += sprintf(buf+n, "%s %lu\n",
193 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
194 			     NR_VM_NUMA_STAT_ITEMS],
195 			     node_page_state(pgdat, i));
196 
197 	return n;
198 }
199 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
200 
201 static ssize_t node_read_distance(struct device *dev,
202 			struct device_attribute *attr, char *buf)
203 {
204 	int nid = dev->id;
205 	int len = 0;
206 	int i;
207 
208 	/*
209 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
210 	 * at the most (distance + space or newline).
211 	 */
212 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
213 
214 	for_each_online_node(i)
215 		len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
216 
217 	len += sprintf(buf + len, "\n");
218 	return len;
219 }
220 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
221 
222 static struct attribute *node_dev_attrs[] = {
223 	&dev_attr_cpumap.attr,
224 	&dev_attr_cpulist.attr,
225 	&dev_attr_meminfo.attr,
226 	&dev_attr_numastat.attr,
227 	&dev_attr_distance.attr,
228 	&dev_attr_vmstat.attr,
229 	NULL
230 };
231 ATTRIBUTE_GROUPS(node_dev);
232 
233 #ifdef CONFIG_HUGETLBFS
234 /*
235  * hugetlbfs per node attributes registration interface:
236  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
237  * it will register its per node attributes for all online nodes with
238  * memory.  It will also call register_hugetlbfs_with_node(), below, to
239  * register its attribute registration functions with this node driver.
240  * Once these hooks have been initialized, the node driver will call into
241  * the hugetlb module to [un]register attributes for hot-plugged nodes.
242  */
243 static node_registration_func_t __hugetlb_register_node;
244 static node_registration_func_t __hugetlb_unregister_node;
245 
246 static inline bool hugetlb_register_node(struct node *node)
247 {
248 	if (__hugetlb_register_node &&
249 			node_state(node->dev.id, N_MEMORY)) {
250 		__hugetlb_register_node(node);
251 		return true;
252 	}
253 	return false;
254 }
255 
256 static inline void hugetlb_unregister_node(struct node *node)
257 {
258 	if (__hugetlb_unregister_node)
259 		__hugetlb_unregister_node(node);
260 }
261 
262 void register_hugetlbfs_with_node(node_registration_func_t doregister,
263 				  node_registration_func_t unregister)
264 {
265 	__hugetlb_register_node   = doregister;
266 	__hugetlb_unregister_node = unregister;
267 }
268 #else
269 static inline void hugetlb_register_node(struct node *node) {}
270 
271 static inline void hugetlb_unregister_node(struct node *node) {}
272 #endif
273 
274 static void node_device_release(struct device *dev)
275 {
276 	struct node *node = to_node(dev);
277 
278 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
279 	/*
280 	 * We schedule the work only when a memory section is
281 	 * onlined/offlined on this node. When we come here,
282 	 * all the memory on this node has been offlined,
283 	 * so we won't enqueue new work to this work.
284 	 *
285 	 * The work is using node->node_work, so we should
286 	 * flush work before freeing the memory.
287 	 */
288 	flush_work(&node->node_work);
289 #endif
290 	kfree(node);
291 }
292 
293 /*
294  * register_node - Setup a sysfs device for a node.
295  * @num - Node number to use when creating the device.
296  *
297  * Initialize and register the node device.
298  */
299 static int register_node(struct node *node, int num)
300 {
301 	int error;
302 
303 	node->dev.id = num;
304 	node->dev.bus = &node_subsys;
305 	node->dev.release = node_device_release;
306 	node->dev.groups = node_dev_groups;
307 	error = device_register(&node->dev);
308 
309 	if (!error){
310 		hugetlb_register_node(node);
311 
312 		compaction_register_node(node);
313 	}
314 	return error;
315 }
316 
317 /**
318  * unregister_node - unregister a node device
319  * @node: node going away
320  *
321  * Unregisters a node device @node.  All the devices on the node must be
322  * unregistered before calling this function.
323  */
324 void unregister_node(struct node *node)
325 {
326 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
327 
328 	device_unregister(&node->dev);
329 }
330 
331 struct node *node_devices[MAX_NUMNODES];
332 
333 /*
334  * register cpu under node
335  */
336 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
337 {
338 	int ret;
339 	struct device *obj;
340 
341 	if (!node_online(nid))
342 		return 0;
343 
344 	obj = get_cpu_device(cpu);
345 	if (!obj)
346 		return 0;
347 
348 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
349 				&obj->kobj,
350 				kobject_name(&obj->kobj));
351 	if (ret)
352 		return ret;
353 
354 	return sysfs_create_link(&obj->kobj,
355 				 &node_devices[nid]->dev.kobj,
356 				 kobject_name(&node_devices[nid]->dev.kobj));
357 }
358 
359 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
360 {
361 	struct device *obj;
362 
363 	if (!node_online(nid))
364 		return 0;
365 
366 	obj = get_cpu_device(cpu);
367 	if (!obj)
368 		return 0;
369 
370 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
371 			  kobject_name(&obj->kobj));
372 	sysfs_remove_link(&obj->kobj,
373 			  kobject_name(&node_devices[nid]->dev.kobj));
374 
375 	return 0;
376 }
377 
378 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
379 static int __ref get_nid_for_pfn(unsigned long pfn)
380 {
381 	if (!pfn_valid_within(pfn))
382 		return -1;
383 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
384 	if (system_state < SYSTEM_RUNNING)
385 		return early_pfn_to_nid(pfn);
386 #endif
387 	return pfn_to_nid(pfn);
388 }
389 
390 /* register memory section under specified node if it spans that node */
391 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
392 {
393 	int ret;
394 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
395 
396 	if (!mem_blk)
397 		return -EFAULT;
398 	if (!node_online(nid))
399 		return 0;
400 
401 	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
402 	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
403 	sect_end_pfn += PAGES_PER_SECTION - 1;
404 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
405 		int page_nid;
406 
407 		/*
408 		 * memory block could have several absent sections from start.
409 		 * skip pfn range from absent section
410 		 */
411 		if (!pfn_present(pfn)) {
412 			pfn = round_down(pfn + PAGES_PER_SECTION,
413 					 PAGES_PER_SECTION) - 1;
414 			continue;
415 		}
416 
417 		page_nid = get_nid_for_pfn(pfn);
418 		if (page_nid < 0)
419 			continue;
420 		if (page_nid != nid)
421 			continue;
422 		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
423 					&mem_blk->dev.kobj,
424 					kobject_name(&mem_blk->dev.kobj));
425 		if (ret)
426 			return ret;
427 
428 		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
429 				&node_devices[nid]->dev.kobj,
430 				kobject_name(&node_devices[nid]->dev.kobj));
431 	}
432 	/* mem section does not span the specified node */
433 	return 0;
434 }
435 
436 /* unregister memory section under all nodes that it spans */
437 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
438 				    unsigned long phys_index)
439 {
440 	NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
441 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
442 
443 	if (!mem_blk) {
444 		NODEMASK_FREE(unlinked_nodes);
445 		return -EFAULT;
446 	}
447 	if (!unlinked_nodes)
448 		return -ENOMEM;
449 	nodes_clear(*unlinked_nodes);
450 
451 	sect_start_pfn = section_nr_to_pfn(phys_index);
452 	sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
453 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
454 		int nid;
455 
456 		nid = get_nid_for_pfn(pfn);
457 		if (nid < 0)
458 			continue;
459 		if (!node_online(nid))
460 			continue;
461 		if (node_test_and_set(nid, *unlinked_nodes))
462 			continue;
463 		sysfs_remove_link(&node_devices[nid]->dev.kobj,
464 			 kobject_name(&mem_blk->dev.kobj));
465 		sysfs_remove_link(&mem_blk->dev.kobj,
466 			 kobject_name(&node_devices[nid]->dev.kobj));
467 	}
468 	NODEMASK_FREE(unlinked_nodes);
469 	return 0;
470 }
471 
472 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long nr_pages)
473 {
474 	unsigned long end_pfn = start_pfn + nr_pages;
475 	unsigned long pfn;
476 	struct memory_block *mem_blk = NULL;
477 	int err = 0;
478 
479 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
480 		unsigned long section_nr = pfn_to_section_nr(pfn);
481 		struct mem_section *mem_sect;
482 		int ret;
483 
484 		if (!present_section_nr(section_nr))
485 			continue;
486 		mem_sect = __nr_to_section(section_nr);
487 
488 		/* same memblock ? */
489 		if (mem_blk)
490 			if ((section_nr >= mem_blk->start_section_nr) &&
491 			    (section_nr <= mem_blk->end_section_nr))
492 				continue;
493 
494 		mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
495 
496 		ret = register_mem_sect_under_node(mem_blk, nid);
497 		if (!err)
498 			err = ret;
499 
500 		/* discard ref obtained in find_memory_block() */
501 	}
502 
503 	if (mem_blk)
504 		kobject_put(&mem_blk->dev.kobj);
505 	return err;
506 }
507 
508 #ifdef CONFIG_HUGETLBFS
509 /*
510  * Handle per node hstate attribute [un]registration on transistions
511  * to/from memoryless state.
512  */
513 static void node_hugetlb_work(struct work_struct *work)
514 {
515 	struct node *node = container_of(work, struct node, node_work);
516 
517 	/*
518 	 * We only get here when a node transitions to/from memoryless state.
519 	 * We can detect which transition occurred by examining whether the
520 	 * node has memory now.  hugetlb_register_node() already check this
521 	 * so we try to register the attributes.  If that fails, then the
522 	 * node has transitioned to memoryless, try to unregister the
523 	 * attributes.
524 	 */
525 	if (!hugetlb_register_node(node))
526 		hugetlb_unregister_node(node);
527 }
528 
529 static void init_node_hugetlb_work(int nid)
530 {
531 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
532 }
533 
534 static int node_memory_callback(struct notifier_block *self,
535 				unsigned long action, void *arg)
536 {
537 	struct memory_notify *mnb = arg;
538 	int nid = mnb->status_change_nid;
539 
540 	switch (action) {
541 	case MEM_ONLINE:
542 	case MEM_OFFLINE:
543 		/*
544 		 * offload per node hstate [un]registration to a work thread
545 		 * when transitioning to/from memoryless state.
546 		 */
547 		if (nid != NUMA_NO_NODE)
548 			schedule_work(&node_devices[nid]->node_work);
549 		break;
550 
551 	case MEM_GOING_ONLINE:
552 	case MEM_GOING_OFFLINE:
553 	case MEM_CANCEL_ONLINE:
554 	case MEM_CANCEL_OFFLINE:
555 	default:
556 		break;
557 	}
558 
559 	return NOTIFY_OK;
560 }
561 #endif	/* CONFIG_HUGETLBFS */
562 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
563 
564 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
565     !defined(CONFIG_HUGETLBFS)
566 static inline int node_memory_callback(struct notifier_block *self,
567 				unsigned long action, void *arg)
568 {
569 	return NOTIFY_OK;
570 }
571 
572 static void init_node_hugetlb_work(int nid) { }
573 
574 #endif
575 
576 int __register_one_node(int nid)
577 {
578 	int error;
579 	int cpu;
580 
581 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
582 	if (!node_devices[nid])
583 		return -ENOMEM;
584 
585 	error = register_node(node_devices[nid], nid);
586 
587 	/* link cpu under this node */
588 	for_each_present_cpu(cpu) {
589 		if (cpu_to_node(cpu) == nid)
590 			register_cpu_under_node(cpu, nid);
591 	}
592 
593 	/* initialize work queue for memory hot plug */
594 	init_node_hugetlb_work(nid);
595 
596 	return error;
597 }
598 
599 void unregister_one_node(int nid)
600 {
601 	if (!node_devices[nid])
602 		return;
603 
604 	unregister_node(node_devices[nid]);
605 	node_devices[nid] = NULL;
606 }
607 
608 /*
609  * node states attributes
610  */
611 
612 static ssize_t print_nodes_state(enum node_states state, char *buf)
613 {
614 	int n;
615 
616 	n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
617 		      nodemask_pr_args(&node_states[state]));
618 	buf[n++] = '\n';
619 	buf[n] = '\0';
620 	return n;
621 }
622 
623 struct node_attr {
624 	struct device_attribute attr;
625 	enum node_states state;
626 };
627 
628 static ssize_t show_node_state(struct device *dev,
629 			       struct device_attribute *attr, char *buf)
630 {
631 	struct node_attr *na = container_of(attr, struct node_attr, attr);
632 	return print_nodes_state(na->state, buf);
633 }
634 
635 #define _NODE_ATTR(name, state) \
636 	{ __ATTR(name, 0444, show_node_state, NULL), state }
637 
638 static struct node_attr node_state_attr[] = {
639 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
640 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
641 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
642 #ifdef CONFIG_HIGHMEM
643 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
644 #endif
645 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
646 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
647 };
648 
649 static struct attribute *node_state_attrs[] = {
650 	&node_state_attr[N_POSSIBLE].attr.attr,
651 	&node_state_attr[N_ONLINE].attr.attr,
652 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
653 #ifdef CONFIG_HIGHMEM
654 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
655 #endif
656 	&node_state_attr[N_MEMORY].attr.attr,
657 	&node_state_attr[N_CPU].attr.attr,
658 	NULL
659 };
660 
661 static struct attribute_group memory_root_attr_group = {
662 	.attrs = node_state_attrs,
663 };
664 
665 static const struct attribute_group *cpu_root_attr_groups[] = {
666 	&memory_root_attr_group,
667 	NULL,
668 };
669 
670 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
671 static int __init register_node_type(void)
672 {
673 	int ret;
674 
675  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
676  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
677 
678 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
679 	if (!ret) {
680 		static struct notifier_block node_memory_callback_nb = {
681 			.notifier_call = node_memory_callback,
682 			.priority = NODE_CALLBACK_PRI,
683 		};
684 		register_hotmemory_notifier(&node_memory_callback_nb);
685 	}
686 
687 	/*
688 	 * Note:  we're not going to unregister the node class if we fail
689 	 * to register the node state class attribute files.
690 	 */
691 	return ret;
692 }
693 postcore_initcall(register_node_type);
694