• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
node_read_cpumap(struct device * dev,bool list,char * buf)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 
node_read_cpumask(struct device * dev,struct device_attribute * attr,char * buf)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 }
node_read_cpulist(struct device * dev,struct device_attribute * attr,char * buf)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))
node_read_meminfo(struct device * dev,struct device_attribute * attr,char * buf)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 
71 	si_meminfo_node(&i, nid);
72 	n = sprintf(buf,
73 		       "Node %d MemTotal:       %8lu kB\n"
74 		       "Node %d MemFree:        %8lu kB\n"
75 		       "Node %d MemUsed:        %8lu kB\n"
76 		       "Node %d Active:         %8lu kB\n"
77 		       "Node %d Inactive:       %8lu kB\n"
78 		       "Node %d Active(anon):   %8lu kB\n"
79 		       "Node %d Inactive(anon): %8lu kB\n"
80 		       "Node %d Active(file):   %8lu kB\n"
81 		       "Node %d Inactive(file): %8lu kB\n"
82 		       "Node %d Unevictable:    %8lu kB\n"
83 		       "Node %d Mlocked:        %8lu kB\n",
84 		       nid, K(i.totalram),
85 		       nid, K(i.freeram),
86 		       nid, K(i.totalram - i.freeram),
87 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
88 				node_page_state(pgdat, NR_ACTIVE_FILE)),
89 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
90 				node_page_state(pgdat, NR_INACTIVE_FILE)),
91 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
92 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
93 		       nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
94 		       nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
95 		       nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
96 		       nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
97 
98 #ifdef CONFIG_HIGHMEM
99 	n += sprintf(buf + n,
100 		       "Node %d HighTotal:      %8lu kB\n"
101 		       "Node %d HighFree:       %8lu kB\n"
102 		       "Node %d LowTotal:       %8lu kB\n"
103 		       "Node %d LowFree:        %8lu kB\n",
104 		       nid, K(i.totalhigh),
105 		       nid, K(i.freehigh),
106 		       nid, K(i.totalram - i.totalhigh),
107 		       nid, K(i.freeram - i.freehigh));
108 #endif
109 	n += sprintf(buf + n,
110 		       "Node %d Dirty:          %8lu kB\n"
111 		       "Node %d Writeback:      %8lu kB\n"
112 		       "Node %d FilePages:      %8lu kB\n"
113 		       "Node %d Mapped:         %8lu kB\n"
114 		       "Node %d AnonPages:      %8lu kB\n"
115 		       "Node %d Shmem:          %8lu kB\n"
116 		       "Node %d KernelStack:    %8lu kB\n"
117 		       "Node %d PageTables:     %8lu kB\n"
118 		       "Node %d NFS_Unstable:   %8lu kB\n"
119 		       "Node %d Bounce:         %8lu kB\n"
120 		       "Node %d WritebackTmp:   %8lu kB\n"
121 		       "Node %d Slab:           %8lu kB\n"
122 		       "Node %d SReclaimable:   %8lu kB\n"
123 		       "Node %d SUnreclaim:     %8lu kB\n"
124 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
125 		       "Node %d AnonHugePages:  %8lu kB\n"
126 		       "Node %d ShmemHugePages: %8lu kB\n"
127 		       "Node %d ShmemPmdMapped: %8lu kB\n"
128 #endif
129 			,
130 		       nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
131 		       nid, K(node_page_state(pgdat, NR_WRITEBACK)),
132 		       nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
133 		       nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
134 		       nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
135 		       nid, K(i.sharedram),
136 		       nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
137 		       nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
138 		       nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
139 		       nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
140 		       nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
141 		       nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE) +
142 			      node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
143 		       nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE)),
144 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
145 		       nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
146 		       nid, K(node_page_state(pgdat, NR_ANON_THPS) *
147 				       HPAGE_PMD_NR),
148 		       nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
149 				       HPAGE_PMD_NR),
150 		       nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
151 				       HPAGE_PMD_NR));
152 #else
153 		       nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)));
154 #endif
155 	n += hugetlb_report_node_meminfo(nid, buf + n);
156 	return n;
157 }
158 
159 #undef K
160 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
161 
node_read_numastat(struct device * dev,struct device_attribute * attr,char * buf)162 static ssize_t node_read_numastat(struct device *dev,
163 				struct device_attribute *attr, char *buf)
164 {
165 	return sprintf(buf,
166 		       "numa_hit %lu\n"
167 		       "numa_miss %lu\n"
168 		       "numa_foreign %lu\n"
169 		       "interleave_hit %lu\n"
170 		       "local_node %lu\n"
171 		       "other_node %lu\n",
172 		       sum_zone_numa_state(dev->id, NUMA_HIT),
173 		       sum_zone_numa_state(dev->id, NUMA_MISS),
174 		       sum_zone_numa_state(dev->id, NUMA_FOREIGN),
175 		       sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
176 		       sum_zone_numa_state(dev->id, NUMA_LOCAL),
177 		       sum_zone_numa_state(dev->id, NUMA_OTHER));
178 }
179 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
180 
node_read_vmstat(struct device * dev,struct device_attribute * attr,char * buf)181 static ssize_t node_read_vmstat(struct device *dev,
182 				struct device_attribute *attr, char *buf)
183 {
184 	int nid = dev->id;
185 	struct pglist_data *pgdat = NODE_DATA(nid);
186 	int i;
187 	int n = 0;
188 
189 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
190 		n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
191 			     sum_zone_node_page_state(nid, i));
192 
193 #ifdef CONFIG_NUMA
194 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
195 		n += sprintf(buf+n, "%s %lu\n",
196 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
197 			     sum_zone_numa_state(nid, i));
198 #endif
199 
200 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
201 		/* Skip hidden vmstat items. */
202 		if (*vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
203 				 NR_VM_NUMA_STAT_ITEMS] == '\0')
204 			continue;
205 		n += sprintf(buf+n, "%s %lu\n",
206 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
207 			     NR_VM_NUMA_STAT_ITEMS],
208 			     node_page_state(pgdat, i));
209 	}
210 
211 	return n;
212 }
213 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
214 
node_read_distance(struct device * dev,struct device_attribute * attr,char * buf)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 
hugetlb_register_node(struct node * node)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 
hugetlb_unregister_node(struct node * node)270 static inline void hugetlb_unregister_node(struct node *node)
271 {
272 	if (__hugetlb_unregister_node)
273 		__hugetlb_unregister_node(node);
274 }
275 
register_hugetlbfs_with_node(node_registration_func_t doregister,node_registration_func_t unregister)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
hugetlb_register_node(struct node * node)283 static inline void hugetlb_register_node(struct node *node) {}
284 
hugetlb_unregister_node(struct node * node)285 static inline void hugetlb_unregister_node(struct node *node) {}
286 #endif
287 
node_device_release(struct device * dev)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  */
register_node(struct node * node,int num)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 		hugetlb_register_node(node);
325 
326 		compaction_register_node(node);
327 	}
328 	return error;
329 }
330 
331 /**
332  * unregister_node - unregister a node device
333  * @node: node going away
334  *
335  * Unregisters a node device @node.  All the devices on the node must be
336  * unregistered before calling this function.
337  */
unregister_node(struct node * node)338 void unregister_node(struct node *node)
339 {
340 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
341 
342 	device_unregister(&node->dev);
343 }
344 
345 struct node *node_devices[MAX_NUMNODES];
346 
347 /*
348  * register cpu under node
349  */
register_cpu_under_node(unsigned int cpu,unsigned int nid)350 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
351 {
352 	int ret;
353 	struct device *obj;
354 
355 	if (!node_online(nid))
356 		return 0;
357 
358 	obj = get_cpu_device(cpu);
359 	if (!obj)
360 		return 0;
361 
362 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
363 				&obj->kobj,
364 				kobject_name(&obj->kobj));
365 	if (ret)
366 		return ret;
367 
368 	return sysfs_create_link(&obj->kobj,
369 				 &node_devices[nid]->dev.kobj,
370 				 kobject_name(&node_devices[nid]->dev.kobj));
371 }
372 
unregister_cpu_under_node(unsigned int cpu,unsigned int nid)373 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
374 {
375 	struct device *obj;
376 
377 	if (!node_online(nid))
378 		return 0;
379 
380 	obj = get_cpu_device(cpu);
381 	if (!obj)
382 		return 0;
383 
384 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
385 			  kobject_name(&obj->kobj));
386 	sysfs_remove_link(&obj->kobj,
387 			  kobject_name(&node_devices[nid]->dev.kobj));
388 
389 	return 0;
390 }
391 
392 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
get_nid_for_pfn(unsigned long pfn)393 static int __ref get_nid_for_pfn(unsigned long pfn)
394 {
395 	if (!pfn_valid_within(pfn))
396 		return -1;
397 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
398 	if (system_state < SYSTEM_RUNNING)
399 		return early_pfn_to_nid(pfn);
400 #endif
401 	return pfn_to_nid(pfn);
402 }
403 
404 /* register memory section under specified node if it spans that node */
register_mem_sect_under_node(struct memory_block * mem_blk,int nid)405 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
406 {
407 	int ret;
408 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
409 
410 	if (!mem_blk)
411 		return -EFAULT;
412 	if (!node_online(nid))
413 		return 0;
414 
415 	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
416 	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
417 	sect_end_pfn += PAGES_PER_SECTION - 1;
418 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
419 		int page_nid;
420 
421 		/*
422 		 * memory block could have several absent sections from start.
423 		 * skip pfn range from absent section
424 		 */
425 		if (!pfn_present(pfn)) {
426 			pfn = round_down(pfn + PAGES_PER_SECTION,
427 					 PAGES_PER_SECTION) - 1;
428 			continue;
429 		}
430 
431 		page_nid = get_nid_for_pfn(pfn);
432 		if (page_nid < 0)
433 			continue;
434 		if (page_nid != nid)
435 			continue;
436 		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
437 					&mem_blk->dev.kobj,
438 					kobject_name(&mem_blk->dev.kobj));
439 		if (ret)
440 			return ret;
441 
442 		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
443 				&node_devices[nid]->dev.kobj,
444 				kobject_name(&node_devices[nid]->dev.kobj));
445 	}
446 	/* mem section does not span the specified node */
447 	return 0;
448 }
449 
450 /* unregister memory section under all nodes that it spans */
unregister_mem_sect_under_nodes(struct memory_block * mem_blk,unsigned long phys_index)451 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
452 				    unsigned long phys_index)
453 {
454 	NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
455 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
456 
457 	if (!mem_blk) {
458 		NODEMASK_FREE(unlinked_nodes);
459 		return -EFAULT;
460 	}
461 	if (!unlinked_nodes)
462 		return -ENOMEM;
463 	nodes_clear(*unlinked_nodes);
464 
465 	sect_start_pfn = section_nr_to_pfn(phys_index);
466 	sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
467 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
468 		int nid;
469 
470 		nid = get_nid_for_pfn(pfn);
471 		if (nid < 0)
472 			continue;
473 		if (!node_online(nid))
474 			continue;
475 		if (node_test_and_set(nid, *unlinked_nodes))
476 			continue;
477 		sysfs_remove_link(&node_devices[nid]->dev.kobj,
478 			 kobject_name(&mem_blk->dev.kobj));
479 		sysfs_remove_link(&mem_blk->dev.kobj,
480 			 kobject_name(&node_devices[nid]->dev.kobj));
481 	}
482 	NODEMASK_FREE(unlinked_nodes);
483 	return 0;
484 }
485 
link_mem_sections(int nid,unsigned long start_pfn,unsigned long nr_pages)486 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long nr_pages)
487 {
488 	unsigned long end_pfn = start_pfn + nr_pages;
489 	unsigned long pfn;
490 	struct memory_block *mem_blk = NULL;
491 	int err = 0;
492 
493 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
494 		unsigned long section_nr = pfn_to_section_nr(pfn);
495 		struct mem_section *mem_sect;
496 		int ret;
497 
498 		if (!present_section_nr(section_nr))
499 			continue;
500 		mem_sect = __nr_to_section(section_nr);
501 
502 		/* same memblock ? */
503 		if (mem_blk)
504 			if ((section_nr >= mem_blk->start_section_nr) &&
505 			    (section_nr <= mem_blk->end_section_nr))
506 				continue;
507 
508 		mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
509 
510 		ret = register_mem_sect_under_node(mem_blk, nid);
511 		if (!err)
512 			err = ret;
513 
514 		/* discard ref obtained in find_memory_block() */
515 	}
516 
517 	if (mem_blk)
518 		kobject_put(&mem_blk->dev.kobj);
519 	return err;
520 }
521 
522 #ifdef CONFIG_HUGETLBFS
523 /*
524  * Handle per node hstate attribute [un]registration on transistions
525  * to/from memoryless state.
526  */
node_hugetlb_work(struct work_struct * work)527 static void node_hugetlb_work(struct work_struct *work)
528 {
529 	struct node *node = container_of(work, struct node, node_work);
530 
531 	/*
532 	 * We only get here when a node transitions to/from memoryless state.
533 	 * We can detect which transition occurred by examining whether the
534 	 * node has memory now.  hugetlb_register_node() already check this
535 	 * so we try to register the attributes.  If that fails, then the
536 	 * node has transitioned to memoryless, try to unregister the
537 	 * attributes.
538 	 */
539 	if (!hugetlb_register_node(node))
540 		hugetlb_unregister_node(node);
541 }
542 
init_node_hugetlb_work(int nid)543 static void init_node_hugetlb_work(int nid)
544 {
545 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
546 }
547 
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)548 static int node_memory_callback(struct notifier_block *self,
549 				unsigned long action, void *arg)
550 {
551 	struct memory_notify *mnb = arg;
552 	int nid = mnb->status_change_nid;
553 
554 	switch (action) {
555 	case MEM_ONLINE:
556 	case MEM_OFFLINE:
557 		/*
558 		 * offload per node hstate [un]registration to a work thread
559 		 * when transitioning to/from memoryless state.
560 		 */
561 		if (nid != NUMA_NO_NODE)
562 			schedule_work(&node_devices[nid]->node_work);
563 		break;
564 
565 	case MEM_GOING_ONLINE:
566 	case MEM_GOING_OFFLINE:
567 	case MEM_CANCEL_ONLINE:
568 	case MEM_CANCEL_OFFLINE:
569 	default:
570 		break;
571 	}
572 
573 	return NOTIFY_OK;
574 }
575 #endif	/* CONFIG_HUGETLBFS */
576 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
577 
578 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
579     !defined(CONFIG_HUGETLBFS)
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)580 static inline int node_memory_callback(struct notifier_block *self,
581 				unsigned long action, void *arg)
582 {
583 	return NOTIFY_OK;
584 }
585 
init_node_hugetlb_work(int nid)586 static void init_node_hugetlb_work(int nid) { }
587 
588 #endif
589 
__register_one_node(int nid)590 int __register_one_node(int nid)
591 {
592 	int error;
593 	int cpu;
594 
595 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
596 	if (!node_devices[nid])
597 		return -ENOMEM;
598 
599 	error = register_node(node_devices[nid], nid);
600 
601 	/* link cpu under this node */
602 	for_each_present_cpu(cpu) {
603 		if (cpu_to_node(cpu) == nid)
604 			register_cpu_under_node(cpu, nid);
605 	}
606 
607 	/* initialize work queue for memory hot plug */
608 	init_node_hugetlb_work(nid);
609 
610 	return error;
611 }
612 
unregister_one_node(int nid)613 void unregister_one_node(int nid)
614 {
615 	if (!node_devices[nid])
616 		return;
617 
618 	unregister_node(node_devices[nid]);
619 	node_devices[nid] = NULL;
620 }
621 
622 /*
623  * node states attributes
624  */
625 
print_nodes_state(enum node_states state,char * buf)626 static ssize_t print_nodes_state(enum node_states state, char *buf)
627 {
628 	int n;
629 
630 	n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
631 		      nodemask_pr_args(&node_states[state]));
632 	buf[n++] = '\n';
633 	buf[n] = '\0';
634 	return n;
635 }
636 
637 struct node_attr {
638 	struct device_attribute attr;
639 	enum node_states state;
640 };
641 
show_node_state(struct device * dev,struct device_attribute * attr,char * buf)642 static ssize_t show_node_state(struct device *dev,
643 			       struct device_attribute *attr, char *buf)
644 {
645 	struct node_attr *na = container_of(attr, struct node_attr, attr);
646 	return print_nodes_state(na->state, buf);
647 }
648 
649 #define _NODE_ATTR(name, state) \
650 	{ __ATTR(name, 0444, show_node_state, NULL), state }
651 
652 static struct node_attr node_state_attr[] = {
653 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
654 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
655 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
656 #ifdef CONFIG_HIGHMEM
657 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
658 #endif
659 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
660 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
661 };
662 
663 static struct attribute *node_state_attrs[] = {
664 	&node_state_attr[N_POSSIBLE].attr.attr,
665 	&node_state_attr[N_ONLINE].attr.attr,
666 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
667 #ifdef CONFIG_HIGHMEM
668 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
669 #endif
670 	&node_state_attr[N_MEMORY].attr.attr,
671 	&node_state_attr[N_CPU].attr.attr,
672 	NULL
673 };
674 
675 static struct attribute_group memory_root_attr_group = {
676 	.attrs = node_state_attrs,
677 };
678 
679 static const struct attribute_group *cpu_root_attr_groups[] = {
680 	&memory_root_attr_group,
681 	NULL,
682 };
683 
684 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
register_node_type(void)685 static int __init register_node_type(void)
686 {
687 	int ret;
688 
689  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
690  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
691 
692 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
693 	if (!ret) {
694 		static struct notifier_block node_memory_callback_nb = {
695 			.notifier_call = node_memory_callback,
696 			.priority = NODE_CALLBACK_PRI,
697 		};
698 		register_hotmemory_notifier(&node_memory_callback_nb);
699 	}
700 
701 	/*
702 	 * Note:  we're not going to unregister the node class if we fail
703 	 * to register the node state class attribute files.
704 	 */
705 	return ret;
706 }
707 postcore_initcall(register_node_type);
708