• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
node_read_cpumap(struct device * dev,int type,char * buf)28 static ssize_t node_read_cpumap(struct device *dev, int type, char *buf)
29 {
30 	struct node *node_dev = to_node(dev);
31 	const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
32 	int len;
33 
34 	/* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
35 	BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
36 
37 	len = type?
38 		cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
39 		cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
40  	buf[len++] = '\n';
41  	buf[len] = '\0';
42 	return len;
43 }
44 
node_read_cpumask(struct device * dev,struct device_attribute * attr,char * buf)45 static inline ssize_t node_read_cpumask(struct device *dev,
46 				struct device_attribute *attr, char *buf)
47 {
48 	return node_read_cpumap(dev, 0, buf);
49 }
node_read_cpulist(struct device * dev,struct device_attribute * attr,char * buf)50 static inline ssize_t node_read_cpulist(struct device *dev,
51 				struct device_attribute *attr, char *buf)
52 {
53 	return node_read_cpumap(dev, 1, buf);
54 }
55 
56 static DEVICE_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
57 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
58 
59 #define K(x) ((x) << (PAGE_SHIFT - 10))
node_read_meminfo(struct device * dev,struct device_attribute * attr,char * buf)60 static ssize_t node_read_meminfo(struct device *dev,
61 			struct device_attribute *attr, char *buf)
62 {
63 	int n;
64 	int nid = dev->id;
65 	struct sysinfo i;
66 
67 	si_meminfo_node(&i, nid);
68 	n = sprintf(buf,
69 		       "Node %d MemTotal:       %8lu kB\n"
70 		       "Node %d MemFree:        %8lu kB\n"
71 		       "Node %d MemUsed:        %8lu kB\n"
72 		       "Node %d Active:         %8lu kB\n"
73 		       "Node %d Inactive:       %8lu kB\n"
74 		       "Node %d Active(anon):   %8lu kB\n"
75 		       "Node %d Inactive(anon): %8lu kB\n"
76 		       "Node %d Active(file):   %8lu kB\n"
77 		       "Node %d Inactive(file): %8lu kB\n"
78 		       "Node %d Unevictable:    %8lu kB\n"
79 		       "Node %d Mlocked:        %8lu kB\n",
80 		       nid, K(i.totalram),
81 		       nid, K(i.freeram),
82 		       nid, K(i.totalram - i.freeram),
83 		       nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
84 				node_page_state(nid, NR_ACTIVE_FILE)),
85 		       nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
86 				node_page_state(nid, NR_INACTIVE_FILE)),
87 		       nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
88 		       nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
89 		       nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
90 		       nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
91 		       nid, K(node_page_state(nid, NR_UNEVICTABLE)),
92 		       nid, K(node_page_state(nid, NR_MLOCK)));
93 
94 #ifdef CONFIG_HIGHMEM
95 	n += sprintf(buf + n,
96 		       "Node %d HighTotal:      %8lu kB\n"
97 		       "Node %d HighFree:       %8lu kB\n"
98 		       "Node %d LowTotal:       %8lu kB\n"
99 		       "Node %d LowFree:        %8lu kB\n",
100 		       nid, K(i.totalhigh),
101 		       nid, K(i.freehigh),
102 		       nid, K(i.totalram - i.totalhigh),
103 		       nid, K(i.freeram - i.freehigh));
104 #endif
105 	n += sprintf(buf + n,
106 		       "Node %d Dirty:          %8lu kB\n"
107 		       "Node %d Writeback:      %8lu kB\n"
108 		       "Node %d FilePages:      %8lu kB\n"
109 		       "Node %d Mapped:         %8lu kB\n"
110 		       "Node %d AnonPages:      %8lu kB\n"
111 		       "Node %d Shmem:          %8lu kB\n"
112 		       "Node %d KernelStack:    %8lu kB\n"
113 		       "Node %d PageTables:     %8lu kB\n"
114 		       "Node %d NFS_Unstable:   %8lu kB\n"
115 		       "Node %d Bounce:         %8lu kB\n"
116 		       "Node %d WritebackTmp:   %8lu kB\n"
117 		       "Node %d Slab:           %8lu kB\n"
118 		       "Node %d SReclaimable:   %8lu kB\n"
119 		       "Node %d SUnreclaim:     %8lu kB\n"
120 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
121 		       "Node %d AnonHugePages:  %8lu kB\n"
122 #endif
123 			,
124 		       nid, K(node_page_state(nid, NR_FILE_DIRTY)),
125 		       nid, K(node_page_state(nid, NR_WRITEBACK)),
126 		       nid, K(node_page_state(nid, NR_FILE_PAGES)),
127 		       nid, K(node_page_state(nid, NR_FILE_MAPPED)),
128 		       nid, K(node_page_state(nid, NR_ANON_PAGES)),
129 		       nid, K(i.sharedram),
130 		       nid, node_page_state(nid, NR_KERNEL_STACK) *
131 				THREAD_SIZE / 1024,
132 		       nid, K(node_page_state(nid, NR_PAGETABLE)),
133 		       nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
134 		       nid, K(node_page_state(nid, NR_BOUNCE)),
135 		       nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
136 		       nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
137 				node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
138 		       nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
139 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
140 		       nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
141 			, nid,
142 			K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
143 			HPAGE_PMD_NR));
144 #else
145 		       nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
146 #endif
147 	n += hugetlb_report_node_meminfo(nid, buf + n);
148 	return n;
149 }
150 
151 #undef K
152 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
153 
node_read_numastat(struct device * dev,struct device_attribute * attr,char * buf)154 static ssize_t node_read_numastat(struct device *dev,
155 				struct device_attribute *attr, char *buf)
156 {
157 	return sprintf(buf,
158 		       "numa_hit %lu\n"
159 		       "numa_miss %lu\n"
160 		       "numa_foreign %lu\n"
161 		       "interleave_hit %lu\n"
162 		       "local_node %lu\n"
163 		       "other_node %lu\n",
164 		       node_page_state(dev->id, NUMA_HIT),
165 		       node_page_state(dev->id, NUMA_MISS),
166 		       node_page_state(dev->id, NUMA_FOREIGN),
167 		       node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
168 		       node_page_state(dev->id, NUMA_LOCAL),
169 		       node_page_state(dev->id, NUMA_OTHER));
170 }
171 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
172 
node_read_vmstat(struct device * dev,struct device_attribute * attr,char * buf)173 static ssize_t node_read_vmstat(struct device *dev,
174 				struct device_attribute *attr, char *buf)
175 {
176 	int nid = dev->id;
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 			     node_page_state(nid, i));
183 
184 	return n;
185 }
186 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
187 
node_read_distance(struct device * dev,struct device_attribute * attr,char * buf)188 static ssize_t node_read_distance(struct device *dev,
189 			struct device_attribute *attr, char * buf)
190 {
191 	int nid = dev->id;
192 	int len = 0;
193 	int i;
194 
195 	/*
196 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
197 	 * at the most (distance + space or newline).
198 	 */
199 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
200 
201 	for_each_online_node(i)
202 		len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
203 
204 	len += sprintf(buf + len, "\n");
205 	return len;
206 }
207 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
208 
209 #ifdef CONFIG_HUGETLBFS
210 /*
211  * hugetlbfs per node attributes registration interface:
212  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
213  * it will register its per node attributes for all online nodes with
214  * memory.  It will also call register_hugetlbfs_with_node(), below, to
215  * register its attribute registration functions with this node driver.
216  * Once these hooks have been initialized, the node driver will call into
217  * the hugetlb module to [un]register attributes for hot-plugged nodes.
218  */
219 static node_registration_func_t __hugetlb_register_node;
220 static node_registration_func_t __hugetlb_unregister_node;
221 
hugetlb_register_node(struct node * node)222 static inline bool hugetlb_register_node(struct node *node)
223 {
224 	if (__hugetlb_register_node &&
225 			node_state(node->dev.id, N_MEMORY)) {
226 		__hugetlb_register_node(node);
227 		return true;
228 	}
229 	return false;
230 }
231 
hugetlb_unregister_node(struct node * node)232 static inline void hugetlb_unregister_node(struct node *node)
233 {
234 	if (__hugetlb_unregister_node)
235 		__hugetlb_unregister_node(node);
236 }
237 
register_hugetlbfs_with_node(node_registration_func_t doregister,node_registration_func_t unregister)238 void register_hugetlbfs_with_node(node_registration_func_t doregister,
239 				  node_registration_func_t unregister)
240 {
241 	__hugetlb_register_node   = doregister;
242 	__hugetlb_unregister_node = unregister;
243 }
244 #else
hugetlb_register_node(struct node * node)245 static inline void hugetlb_register_node(struct node *node) {}
246 
hugetlb_unregister_node(struct node * node)247 static inline void hugetlb_unregister_node(struct node *node) {}
248 #endif
249 
node_device_release(struct device * dev)250 static void node_device_release(struct device *dev)
251 {
252 	struct node *node = to_node(dev);
253 
254 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
255 	/*
256 	 * We schedule the work only when a memory section is
257 	 * onlined/offlined on this node. When we come here,
258 	 * all the memory on this node has been offlined,
259 	 * so we won't enqueue new work to this work.
260 	 *
261 	 * The work is using node->node_work, so we should
262 	 * flush work before freeing the memory.
263 	 */
264 	flush_work(&node->node_work);
265 #endif
266 	kfree(node);
267 }
268 
269 /*
270  * register_node - Setup a sysfs device for a node.
271  * @num - Node number to use when creating the device.
272  *
273  * Initialize and register the node device.
274  */
register_node(struct node * node,int num,struct node * parent)275 static int register_node(struct node *node, int num, struct node *parent)
276 {
277 	int error;
278 
279 	node->dev.id = num;
280 	node->dev.bus = &node_subsys;
281 	node->dev.release = node_device_release;
282 	error = device_register(&node->dev);
283 
284 	if (!error){
285 		device_create_file(&node->dev, &dev_attr_cpumap);
286 		device_create_file(&node->dev, &dev_attr_cpulist);
287 		device_create_file(&node->dev, &dev_attr_meminfo);
288 		device_create_file(&node->dev, &dev_attr_numastat);
289 		device_create_file(&node->dev, &dev_attr_distance);
290 		device_create_file(&node->dev, &dev_attr_vmstat);
291 
292 		hugetlb_register_node(node);
293 
294 		compaction_register_node(node);
295 	}
296 	return error;
297 }
298 
299 /**
300  * unregister_node - unregister a node device
301  * @node: node going away
302  *
303  * Unregisters a node device @node.  All the devices on the node must be
304  * unregistered before calling this function.
305  */
unregister_node(struct node * node)306 void unregister_node(struct node *node)
307 {
308 	device_remove_file(&node->dev, &dev_attr_cpumap);
309 	device_remove_file(&node->dev, &dev_attr_cpulist);
310 	device_remove_file(&node->dev, &dev_attr_meminfo);
311 	device_remove_file(&node->dev, &dev_attr_numastat);
312 	device_remove_file(&node->dev, &dev_attr_distance);
313 	device_remove_file(&node->dev, &dev_attr_vmstat);
314 
315 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
316 
317 	device_unregister(&node->dev);
318 }
319 
320 struct node *node_devices[MAX_NUMNODES];
321 
322 /*
323  * register cpu under node
324  */
register_cpu_under_node(unsigned int cpu,unsigned int nid)325 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
326 {
327 	int ret;
328 	struct device *obj;
329 
330 	if (!node_online(nid))
331 		return 0;
332 
333 	obj = get_cpu_device(cpu);
334 	if (!obj)
335 		return 0;
336 
337 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
338 				&obj->kobj,
339 				kobject_name(&obj->kobj));
340 	if (ret)
341 		return ret;
342 
343 	return sysfs_create_link(&obj->kobj,
344 				 &node_devices[nid]->dev.kobj,
345 				 kobject_name(&node_devices[nid]->dev.kobj));
346 }
347 
unregister_cpu_under_node(unsigned int cpu,unsigned int nid)348 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
349 {
350 	struct device *obj;
351 
352 	if (!node_online(nid))
353 		return 0;
354 
355 	obj = get_cpu_device(cpu);
356 	if (!obj)
357 		return 0;
358 
359 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
360 			  kobject_name(&obj->kobj));
361 	sysfs_remove_link(&obj->kobj,
362 			  kobject_name(&node_devices[nid]->dev.kobj));
363 
364 	return 0;
365 }
366 
367 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
368 #define page_initialized(page)  (page->lru.next)
369 
get_nid_for_pfn(unsigned long pfn)370 static int get_nid_for_pfn(unsigned long pfn)
371 {
372 	struct page *page;
373 
374 	if (!pfn_valid_within(pfn))
375 		return -1;
376 	page = pfn_to_page(pfn);
377 	if (!page_initialized(page))
378 		return -1;
379 	return pfn_to_nid(pfn);
380 }
381 
382 /* register memory section under specified node if it spans that node */
register_mem_sect_under_node(struct memory_block * mem_blk,int nid)383 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
384 {
385 	int ret;
386 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
387 
388 	if (!mem_blk)
389 		return -EFAULT;
390 	if (!node_online(nid))
391 		return 0;
392 
393 	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
394 	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
395 	sect_end_pfn += PAGES_PER_SECTION - 1;
396 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
397 		int page_nid;
398 
399 		/*
400 		 * memory block could have several absent sections from start.
401 		 * skip pfn range from absent section
402 		 */
403 		if (!pfn_present(pfn)) {
404 			pfn = round_down(pfn + PAGES_PER_SECTION,
405 					 PAGES_PER_SECTION) - 1;
406 			continue;
407 		}
408 
409 		page_nid = get_nid_for_pfn(pfn);
410 		if (page_nid < 0)
411 			continue;
412 		if (page_nid != nid)
413 			continue;
414 		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
415 					&mem_blk->dev.kobj,
416 					kobject_name(&mem_blk->dev.kobj));
417 		if (ret)
418 			return ret;
419 
420 		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
421 				&node_devices[nid]->dev.kobj,
422 				kobject_name(&node_devices[nid]->dev.kobj));
423 	}
424 	/* mem section does not span the specified node */
425 	return 0;
426 }
427 
428 /* unregister memory section under all nodes that it spans */
unregister_mem_sect_under_nodes(struct memory_block * mem_blk,unsigned long phys_index)429 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
430 				    unsigned long phys_index)
431 {
432 	NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
433 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
434 
435 	if (!mem_blk) {
436 		NODEMASK_FREE(unlinked_nodes);
437 		return -EFAULT;
438 	}
439 	if (!unlinked_nodes)
440 		return -ENOMEM;
441 	nodes_clear(*unlinked_nodes);
442 
443 	sect_start_pfn = section_nr_to_pfn(phys_index);
444 	sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
445 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
446 		int nid;
447 
448 		nid = get_nid_for_pfn(pfn);
449 		if (nid < 0)
450 			continue;
451 		if (!node_online(nid))
452 			continue;
453 		if (node_test_and_set(nid, *unlinked_nodes))
454 			continue;
455 		sysfs_remove_link(&node_devices[nid]->dev.kobj,
456 			 kobject_name(&mem_blk->dev.kobj));
457 		sysfs_remove_link(&mem_blk->dev.kobj,
458 			 kobject_name(&node_devices[nid]->dev.kobj));
459 	}
460 	NODEMASK_FREE(unlinked_nodes);
461 	return 0;
462 }
463 
link_mem_sections(int nid)464 static int link_mem_sections(int nid)
465 {
466 	unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
467 	unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
468 	unsigned long pfn;
469 	struct memory_block *mem_blk = NULL;
470 	int err = 0;
471 
472 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
473 		unsigned long section_nr = pfn_to_section_nr(pfn);
474 		struct mem_section *mem_sect;
475 		int ret;
476 
477 		if (!present_section_nr(section_nr))
478 			continue;
479 		mem_sect = __nr_to_section(section_nr);
480 
481 		/* same memblock ? */
482 		if (mem_blk)
483 			if ((section_nr >= mem_blk->start_section_nr) &&
484 			    (section_nr <= mem_blk->end_section_nr))
485 				continue;
486 
487 		mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
488 
489 		ret = register_mem_sect_under_node(mem_blk, nid);
490 		if (!err)
491 			err = ret;
492 
493 		/* discard ref obtained in find_memory_block() */
494 	}
495 
496 	if (mem_blk)
497 		kobject_put(&mem_blk->dev.kobj);
498 	return err;
499 }
500 
501 #ifdef CONFIG_HUGETLBFS
502 /*
503  * Handle per node hstate attribute [un]registration on transistions
504  * to/from memoryless state.
505  */
node_hugetlb_work(struct work_struct * work)506 static void node_hugetlb_work(struct work_struct *work)
507 {
508 	struct node *node = container_of(work, struct node, node_work);
509 
510 	/*
511 	 * We only get here when a node transitions to/from memoryless state.
512 	 * We can detect which transition occurred by examining whether the
513 	 * node has memory now.  hugetlb_register_node() already check this
514 	 * so we try to register the attributes.  If that fails, then the
515 	 * node has transitioned to memoryless, try to unregister the
516 	 * attributes.
517 	 */
518 	if (!hugetlb_register_node(node))
519 		hugetlb_unregister_node(node);
520 }
521 
init_node_hugetlb_work(int nid)522 static void init_node_hugetlb_work(int nid)
523 {
524 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
525 }
526 
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)527 static int node_memory_callback(struct notifier_block *self,
528 				unsigned long action, void *arg)
529 {
530 	struct memory_notify *mnb = arg;
531 	int nid = mnb->status_change_nid;
532 
533 	switch (action) {
534 	case MEM_ONLINE:
535 	case MEM_OFFLINE:
536 		/*
537 		 * offload per node hstate [un]registration to a work thread
538 		 * when transitioning to/from memoryless state.
539 		 */
540 		if (nid != NUMA_NO_NODE)
541 			schedule_work(&node_devices[nid]->node_work);
542 		break;
543 
544 	case MEM_GOING_ONLINE:
545 	case MEM_GOING_OFFLINE:
546 	case MEM_CANCEL_ONLINE:
547 	case MEM_CANCEL_OFFLINE:
548 	default:
549 		break;
550 	}
551 
552 	return NOTIFY_OK;
553 }
554 #endif	/* CONFIG_HUGETLBFS */
555 #else	/* !CONFIG_MEMORY_HOTPLUG_SPARSE */
556 
link_mem_sections(int nid)557 static int link_mem_sections(int nid) { return 0; }
558 #endif	/* CONFIG_MEMORY_HOTPLUG_SPARSE */
559 
560 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
561     !defined(CONFIG_HUGETLBFS)
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)562 static inline int node_memory_callback(struct notifier_block *self,
563 				unsigned long action, void *arg)
564 {
565 	return NOTIFY_OK;
566 }
567 
init_node_hugetlb_work(int nid)568 static void init_node_hugetlb_work(int nid) { }
569 
570 #endif
571 
register_one_node(int nid)572 int register_one_node(int nid)
573 {
574 	int error = 0;
575 	int cpu;
576 
577 	if (node_online(nid)) {
578 		int p_node = parent_node(nid);
579 		struct node *parent = NULL;
580 
581 		if (p_node != nid)
582 			parent = node_devices[p_node];
583 
584 		node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
585 		if (!node_devices[nid])
586 			return -ENOMEM;
587 
588 		error = register_node(node_devices[nid], nid, parent);
589 
590 		/* link cpu under this node */
591 		for_each_present_cpu(cpu) {
592 			if (cpu_to_node(cpu) == nid)
593 				register_cpu_under_node(cpu, nid);
594 		}
595 
596 		/* link memory sections under this node */
597 		error = link_mem_sections(nid);
598 
599 		/* initialize work queue for memory hot plug */
600 		init_node_hugetlb_work(nid);
601 	}
602 
603 	return error;
604 
605 }
606 
unregister_one_node(int nid)607 void unregister_one_node(int nid)
608 {
609 	if (!node_devices[nid])
610 		return;
611 
612 	unregister_node(node_devices[nid]);
613 	node_devices[nid] = NULL;
614 }
615 
616 /*
617  * node states attributes
618  */
619 
print_nodes_state(enum node_states state,char * buf)620 static ssize_t print_nodes_state(enum node_states state, char *buf)
621 {
622 	int n;
623 
624 	n = nodelist_scnprintf(buf, PAGE_SIZE-2, node_states[state]);
625 	buf[n++] = '\n';
626 	buf[n] = '\0';
627 	return n;
628 }
629 
630 struct node_attr {
631 	struct device_attribute attr;
632 	enum node_states state;
633 };
634 
show_node_state(struct device * dev,struct device_attribute * attr,char * buf)635 static ssize_t show_node_state(struct device *dev,
636 			       struct device_attribute *attr, char *buf)
637 {
638 	struct node_attr *na = container_of(attr, struct node_attr, attr);
639 	return print_nodes_state(na->state, buf);
640 }
641 
642 #define _NODE_ATTR(name, state) \
643 	{ __ATTR(name, 0444, show_node_state, NULL), state }
644 
645 static struct node_attr node_state_attr[] = {
646 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
647 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
648 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
649 #ifdef CONFIG_HIGHMEM
650 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
651 #endif
652 #ifdef CONFIG_MOVABLE_NODE
653 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
654 #endif
655 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
656 };
657 
658 static struct attribute *node_state_attrs[] = {
659 	&node_state_attr[N_POSSIBLE].attr.attr,
660 	&node_state_attr[N_ONLINE].attr.attr,
661 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
662 #ifdef CONFIG_HIGHMEM
663 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
664 #endif
665 #ifdef CONFIG_MOVABLE_NODE
666 	&node_state_attr[N_MEMORY].attr.attr,
667 #endif
668 	&node_state_attr[N_CPU].attr.attr,
669 	NULL
670 };
671 
672 static struct attribute_group memory_root_attr_group = {
673 	.attrs = node_state_attrs,
674 };
675 
676 static const struct attribute_group *cpu_root_attr_groups[] = {
677 	&memory_root_attr_group,
678 	NULL,
679 };
680 
681 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
register_node_type(void)682 static int __init register_node_type(void)
683 {
684 	int ret;
685 
686  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
687  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
688 
689 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
690 	if (!ret) {
691 		static struct notifier_block node_memory_callback_nb = {
692 			.notifier_call = node_memory_callback,
693 			.priority = NODE_CALLBACK_PRI,
694 		};
695 		register_hotmemory_notifier(&node_memory_callback_nb);
696 	}
697 
698 	/*
699 	 * Note:  we're not going to unregister the node class if we fail
700 	 * to register the node state class attribute files.
701 	 */
702 	return ret;
703 }
704 postcore_initcall(register_node_type);
705