xref: /openbmc/linux/drivers/base/node.c (revision 5e012745)
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/pm_runtime.h>
21 #include <linux/swap.h>
22 #include <linux/slab.h>
23 
24 static struct bus_type node_subsys = {
25 	.name = "node",
26 	.dev_name = "node",
27 };
28 
29 
30 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
31 {
32 	ssize_t n;
33 	cpumask_var_t mask;
34 	struct node *node_dev = to_node(dev);
35 
36 	/* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
37 	BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
38 
39 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
40 		return 0;
41 
42 	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
43 	n = cpumap_print_to_pagebuf(list, buf, mask);
44 	free_cpumask_var(mask);
45 
46 	return n;
47 }
48 
49 static inline ssize_t node_read_cpumask(struct device *dev,
50 				struct device_attribute *attr, char *buf)
51 {
52 	return node_read_cpumap(dev, false, buf);
53 }
54 static inline ssize_t node_read_cpulist(struct device *dev,
55 				struct device_attribute *attr, char *buf)
56 {
57 	return node_read_cpumap(dev, true, buf);
58 }
59 
60 static DEVICE_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
61 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
62 
63 /**
64  * struct node_access_nodes - Access class device to hold user visible
65  * 			      relationships to other nodes.
66  * @dev:	Device for this memory access class
67  * @list_node:	List element in the node's access list
68  * @access:	The access class rank
69  * @hmem_attrs: Heterogeneous memory performance attributes
70  */
71 struct node_access_nodes {
72 	struct device		dev;
73 	struct list_head	list_node;
74 	unsigned		access;
75 #ifdef CONFIG_HMEM_REPORTING
76 	struct node_hmem_attrs	hmem_attrs;
77 #endif
78 };
79 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
80 
81 static struct attribute *node_init_access_node_attrs[] = {
82 	NULL,
83 };
84 
85 static struct attribute *node_targ_access_node_attrs[] = {
86 	NULL,
87 };
88 
89 static const struct attribute_group initiators = {
90 	.name	= "initiators",
91 	.attrs	= node_init_access_node_attrs,
92 };
93 
94 static const struct attribute_group targets = {
95 	.name	= "targets",
96 	.attrs	= node_targ_access_node_attrs,
97 };
98 
99 static const struct attribute_group *node_access_node_groups[] = {
100 	&initiators,
101 	&targets,
102 	NULL,
103 };
104 
105 static void node_remove_accesses(struct node *node)
106 {
107 	struct node_access_nodes *c, *cnext;
108 
109 	list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
110 		list_del(&c->list_node);
111 		device_unregister(&c->dev);
112 	}
113 }
114 
115 static void node_access_release(struct device *dev)
116 {
117 	kfree(to_access_nodes(dev));
118 }
119 
120 static struct node_access_nodes *node_init_node_access(struct node *node,
121 						       unsigned access)
122 {
123 	struct node_access_nodes *access_node;
124 	struct device *dev;
125 
126 	list_for_each_entry(access_node, &node->access_list, list_node)
127 		if (access_node->access == access)
128 			return access_node;
129 
130 	access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
131 	if (!access_node)
132 		return NULL;
133 
134 	access_node->access = access;
135 	dev = &access_node->dev;
136 	dev->parent = &node->dev;
137 	dev->release = node_access_release;
138 	dev->groups = node_access_node_groups;
139 	if (dev_set_name(dev, "access%u", access))
140 		goto free;
141 
142 	if (device_register(dev))
143 		goto free_name;
144 
145 	pm_runtime_no_callbacks(dev);
146 	list_add_tail(&access_node->list_node, &node->access_list);
147 	return access_node;
148 free_name:
149 	kfree_const(dev->kobj.name);
150 free:
151 	kfree(access_node);
152 	return NULL;
153 }
154 
155 #ifdef CONFIG_HMEM_REPORTING
156 #define ACCESS_ATTR(name) 						   \
157 static ssize_t name##_show(struct device *dev,				   \
158 			   struct device_attribute *attr,		   \
159 			   char *buf)					   \
160 {									   \
161 	return sprintf(buf, "%u\n", to_access_nodes(dev)->hmem_attrs.name); \
162 }									   \
163 static DEVICE_ATTR_RO(name);
164 
165 ACCESS_ATTR(read_bandwidth)
166 ACCESS_ATTR(read_latency)
167 ACCESS_ATTR(write_bandwidth)
168 ACCESS_ATTR(write_latency)
169 
170 static struct attribute *access_attrs[] = {
171 	&dev_attr_read_bandwidth.attr,
172 	&dev_attr_read_latency.attr,
173 	&dev_attr_write_bandwidth.attr,
174 	&dev_attr_write_latency.attr,
175 	NULL,
176 };
177 
178 /**
179  * node_set_perf_attrs - Set the performance values for given access class
180  * @nid: Node identifier to be set
181  * @hmem_attrs: Heterogeneous memory performance attributes
182  * @access: The access class the for the given attributes
183  */
184 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
185 			 unsigned access)
186 {
187 	struct node_access_nodes *c;
188 	struct node *node;
189 	int i;
190 
191 	if (WARN_ON_ONCE(!node_online(nid)))
192 		return;
193 
194 	node = node_devices[nid];
195 	c = node_init_node_access(node, access);
196 	if (!c)
197 		return;
198 
199 	c->hmem_attrs = *hmem_attrs;
200 	for (i = 0; access_attrs[i] != NULL; i++) {
201 		if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
202 					    "initiators")) {
203 			pr_info("failed to add performance attribute to node %d\n",
204 				nid);
205 			break;
206 		}
207 	}
208 }
209 
210 /**
211  * struct node_cache_info - Internal tracking for memory node caches
212  * @dev:	Device represeting the cache level
213  * @node:	List element for tracking in the node
214  * @cache_attrs:Attributes for this cache level
215  */
216 struct node_cache_info {
217 	struct device dev;
218 	struct list_head node;
219 	struct node_cache_attrs cache_attrs;
220 };
221 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
222 
223 #define CACHE_ATTR(name, fmt) 						\
224 static ssize_t name##_show(struct device *dev,				\
225 			   struct device_attribute *attr,		\
226 			   char *buf)					\
227 {									\
228 	return sprintf(buf, fmt "\n", to_cache_info(dev)->cache_attrs.name);\
229 }									\
230 DEVICE_ATTR_RO(name);
231 
232 CACHE_ATTR(size, "%llu")
233 CACHE_ATTR(line_size, "%u")
234 CACHE_ATTR(indexing, "%u")
235 CACHE_ATTR(write_policy, "%u")
236 
237 static struct attribute *cache_attrs[] = {
238 	&dev_attr_indexing.attr,
239 	&dev_attr_size.attr,
240 	&dev_attr_line_size.attr,
241 	&dev_attr_write_policy.attr,
242 	NULL,
243 };
244 ATTRIBUTE_GROUPS(cache);
245 
246 static void node_cache_release(struct device *dev)
247 {
248 	kfree(dev);
249 }
250 
251 static void node_cacheinfo_release(struct device *dev)
252 {
253 	struct node_cache_info *info = to_cache_info(dev);
254 	kfree(info);
255 }
256 
257 static void node_init_cache_dev(struct node *node)
258 {
259 	struct device *dev;
260 
261 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
262 	if (!dev)
263 		return;
264 
265 	dev->parent = &node->dev;
266 	dev->release = node_cache_release;
267 	if (dev_set_name(dev, "memory_side_cache"))
268 		goto free_dev;
269 
270 	if (device_register(dev))
271 		goto free_name;
272 
273 	pm_runtime_no_callbacks(dev);
274 	node->cache_dev = dev;
275 	return;
276 free_name:
277 	kfree_const(dev->kobj.name);
278 free_dev:
279 	kfree(dev);
280 }
281 
282 /**
283  * node_add_cache() - add cache attribute to a memory node
284  * @nid: Node identifier that has new cache attributes
285  * @cache_attrs: Attributes for the cache being added
286  */
287 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
288 {
289 	struct node_cache_info *info;
290 	struct device *dev;
291 	struct node *node;
292 
293 	if (!node_online(nid) || !node_devices[nid])
294 		return;
295 
296 	node = node_devices[nid];
297 	list_for_each_entry(info, &node->cache_attrs, node) {
298 		if (info->cache_attrs.level == cache_attrs->level) {
299 			dev_warn(&node->dev,
300 				"attempt to add duplicate cache level:%d\n",
301 				cache_attrs->level);
302 			return;
303 		}
304 	}
305 
306 	if (!node->cache_dev)
307 		node_init_cache_dev(node);
308 	if (!node->cache_dev)
309 		return;
310 
311 	info = kzalloc(sizeof(*info), GFP_KERNEL);
312 	if (!info)
313 		return;
314 
315 	dev = &info->dev;
316 	dev->parent = node->cache_dev;
317 	dev->release = node_cacheinfo_release;
318 	dev->groups = cache_groups;
319 	if (dev_set_name(dev, "index%d", cache_attrs->level))
320 		goto free_cache;
321 
322 	info->cache_attrs = *cache_attrs;
323 	if (device_register(dev)) {
324 		dev_warn(&node->dev, "failed to add cache level:%d\n",
325 			 cache_attrs->level);
326 		goto free_name;
327 	}
328 	pm_runtime_no_callbacks(dev);
329 	list_add_tail(&info->node, &node->cache_attrs);
330 	return;
331 free_name:
332 	kfree_const(dev->kobj.name);
333 free_cache:
334 	kfree(info);
335 }
336 
337 static void node_remove_caches(struct node *node)
338 {
339 	struct node_cache_info *info, *next;
340 
341 	if (!node->cache_dev)
342 		return;
343 
344 	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
345 		list_del(&info->node);
346 		device_unregister(&info->dev);
347 	}
348 	device_unregister(node->cache_dev);
349 }
350 
351 static void node_init_caches(unsigned int nid)
352 {
353 	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
354 }
355 #else
356 static void node_init_caches(unsigned int nid) { }
357 static void node_remove_caches(struct node *node) { }
358 #endif
359 
360 #define K(x) ((x) << (PAGE_SHIFT - 10))
361 static ssize_t node_read_meminfo(struct device *dev,
362 			struct device_attribute *attr, char *buf)
363 {
364 	int n;
365 	int nid = dev->id;
366 	struct pglist_data *pgdat = NODE_DATA(nid);
367 	struct sysinfo i;
368 	unsigned long sreclaimable, sunreclaimable;
369 
370 	si_meminfo_node(&i, nid);
371 	sreclaimable = node_page_state(pgdat, NR_SLAB_RECLAIMABLE);
372 	sunreclaimable = node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE);
373 	n = sprintf(buf,
374 		       "Node %d MemTotal:       %8lu kB\n"
375 		       "Node %d MemFree:        %8lu kB\n"
376 		       "Node %d MemUsed:        %8lu kB\n"
377 		       "Node %d Active:         %8lu kB\n"
378 		       "Node %d Inactive:       %8lu kB\n"
379 		       "Node %d Active(anon):   %8lu kB\n"
380 		       "Node %d Inactive(anon): %8lu kB\n"
381 		       "Node %d Active(file):   %8lu kB\n"
382 		       "Node %d Inactive(file): %8lu kB\n"
383 		       "Node %d Unevictable:    %8lu kB\n"
384 		       "Node %d Mlocked:        %8lu kB\n",
385 		       nid, K(i.totalram),
386 		       nid, K(i.freeram),
387 		       nid, K(i.totalram - i.freeram),
388 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
389 				node_page_state(pgdat, NR_ACTIVE_FILE)),
390 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
391 				node_page_state(pgdat, NR_INACTIVE_FILE)),
392 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
393 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
394 		       nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
395 		       nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
396 		       nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
397 		       nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
398 
399 #ifdef CONFIG_HIGHMEM
400 	n += sprintf(buf + n,
401 		       "Node %d HighTotal:      %8lu kB\n"
402 		       "Node %d HighFree:       %8lu kB\n"
403 		       "Node %d LowTotal:       %8lu kB\n"
404 		       "Node %d LowFree:        %8lu kB\n",
405 		       nid, K(i.totalhigh),
406 		       nid, K(i.freehigh),
407 		       nid, K(i.totalram - i.totalhigh),
408 		       nid, K(i.freeram - i.freehigh));
409 #endif
410 	n += sprintf(buf + n,
411 		       "Node %d Dirty:          %8lu kB\n"
412 		       "Node %d Writeback:      %8lu kB\n"
413 		       "Node %d FilePages:      %8lu kB\n"
414 		       "Node %d Mapped:         %8lu kB\n"
415 		       "Node %d AnonPages:      %8lu kB\n"
416 		       "Node %d Shmem:          %8lu kB\n"
417 		       "Node %d KernelStack:    %8lu kB\n"
418 		       "Node %d PageTables:     %8lu kB\n"
419 		       "Node %d NFS_Unstable:   %8lu kB\n"
420 		       "Node %d Bounce:         %8lu kB\n"
421 		       "Node %d WritebackTmp:   %8lu kB\n"
422 		       "Node %d KReclaimable:   %8lu kB\n"
423 		       "Node %d Slab:           %8lu kB\n"
424 		       "Node %d SReclaimable:   %8lu kB\n"
425 		       "Node %d SUnreclaim:     %8lu kB\n"
426 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
427 		       "Node %d AnonHugePages:  %8lu kB\n"
428 		       "Node %d ShmemHugePages: %8lu kB\n"
429 		       "Node %d ShmemPmdMapped: %8lu kB\n"
430 #endif
431 			,
432 		       nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
433 		       nid, K(node_page_state(pgdat, NR_WRITEBACK)),
434 		       nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
435 		       nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
436 		       nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
437 		       nid, K(i.sharedram),
438 		       nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
439 		       nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
440 		       nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
441 		       nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
442 		       nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
443 		       nid, K(sreclaimable +
444 			      node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
445 		       nid, K(sreclaimable + sunreclaimable),
446 		       nid, K(sreclaimable),
447 		       nid, K(sunreclaimable)
448 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
449 		       ,
450 		       nid, K(node_page_state(pgdat, NR_ANON_THPS) *
451 				       HPAGE_PMD_NR),
452 		       nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
453 				       HPAGE_PMD_NR),
454 		       nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
455 				       HPAGE_PMD_NR)
456 #endif
457 		       );
458 	n += hugetlb_report_node_meminfo(nid, buf + n);
459 	return n;
460 }
461 
462 #undef K
463 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
464 
465 static ssize_t node_read_numastat(struct device *dev,
466 				struct device_attribute *attr, char *buf)
467 {
468 	return sprintf(buf,
469 		       "numa_hit %lu\n"
470 		       "numa_miss %lu\n"
471 		       "numa_foreign %lu\n"
472 		       "interleave_hit %lu\n"
473 		       "local_node %lu\n"
474 		       "other_node %lu\n",
475 		       sum_zone_numa_state(dev->id, NUMA_HIT),
476 		       sum_zone_numa_state(dev->id, NUMA_MISS),
477 		       sum_zone_numa_state(dev->id, NUMA_FOREIGN),
478 		       sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
479 		       sum_zone_numa_state(dev->id, NUMA_LOCAL),
480 		       sum_zone_numa_state(dev->id, NUMA_OTHER));
481 }
482 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
483 
484 static ssize_t node_read_vmstat(struct device *dev,
485 				struct device_attribute *attr, char *buf)
486 {
487 	int nid = dev->id;
488 	struct pglist_data *pgdat = NODE_DATA(nid);
489 	int i;
490 	int n = 0;
491 
492 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
493 		n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
494 			     sum_zone_node_page_state(nid, i));
495 
496 #ifdef CONFIG_NUMA
497 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
498 		n += sprintf(buf+n, "%s %lu\n",
499 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
500 			     sum_zone_numa_state(nid, i));
501 #endif
502 
503 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
504 		n += sprintf(buf+n, "%s %lu\n",
505 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
506 			     NR_VM_NUMA_STAT_ITEMS],
507 			     node_page_state(pgdat, i));
508 
509 	return n;
510 }
511 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
512 
513 static ssize_t node_read_distance(struct device *dev,
514 			struct device_attribute *attr, char *buf)
515 {
516 	int nid = dev->id;
517 	int len = 0;
518 	int i;
519 
520 	/*
521 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
522 	 * at the most (distance + space or newline).
523 	 */
524 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
525 
526 	for_each_online_node(i)
527 		len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
528 
529 	len += sprintf(buf + len, "\n");
530 	return len;
531 }
532 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
533 
534 static struct attribute *node_dev_attrs[] = {
535 	&dev_attr_cpumap.attr,
536 	&dev_attr_cpulist.attr,
537 	&dev_attr_meminfo.attr,
538 	&dev_attr_numastat.attr,
539 	&dev_attr_distance.attr,
540 	&dev_attr_vmstat.attr,
541 	NULL
542 };
543 ATTRIBUTE_GROUPS(node_dev);
544 
545 #ifdef CONFIG_HUGETLBFS
546 /*
547  * hugetlbfs per node attributes registration interface:
548  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
549  * it will register its per node attributes for all online nodes with
550  * memory.  It will also call register_hugetlbfs_with_node(), below, to
551  * register its attribute registration functions with this node driver.
552  * Once these hooks have been initialized, the node driver will call into
553  * the hugetlb module to [un]register attributes for hot-plugged nodes.
554  */
555 static node_registration_func_t __hugetlb_register_node;
556 static node_registration_func_t __hugetlb_unregister_node;
557 
558 static inline bool hugetlb_register_node(struct node *node)
559 {
560 	if (__hugetlb_register_node &&
561 			node_state(node->dev.id, N_MEMORY)) {
562 		__hugetlb_register_node(node);
563 		return true;
564 	}
565 	return false;
566 }
567 
568 static inline void hugetlb_unregister_node(struct node *node)
569 {
570 	if (__hugetlb_unregister_node)
571 		__hugetlb_unregister_node(node);
572 }
573 
574 void register_hugetlbfs_with_node(node_registration_func_t doregister,
575 				  node_registration_func_t unregister)
576 {
577 	__hugetlb_register_node   = doregister;
578 	__hugetlb_unregister_node = unregister;
579 }
580 #else
581 static inline void hugetlb_register_node(struct node *node) {}
582 
583 static inline void hugetlb_unregister_node(struct node *node) {}
584 #endif
585 
586 static void node_device_release(struct device *dev)
587 {
588 	struct node *node = to_node(dev);
589 
590 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
591 	/*
592 	 * We schedule the work only when a memory section is
593 	 * onlined/offlined on this node. When we come here,
594 	 * all the memory on this node has been offlined,
595 	 * so we won't enqueue new work to this work.
596 	 *
597 	 * The work is using node->node_work, so we should
598 	 * flush work before freeing the memory.
599 	 */
600 	flush_work(&node->node_work);
601 #endif
602 	kfree(node);
603 }
604 
605 /*
606  * register_node - Setup a sysfs device for a node.
607  * @num - Node number to use when creating the device.
608  *
609  * Initialize and register the node device.
610  */
611 static int register_node(struct node *node, int num)
612 {
613 	int error;
614 
615 	node->dev.id = num;
616 	node->dev.bus = &node_subsys;
617 	node->dev.release = node_device_release;
618 	node->dev.groups = node_dev_groups;
619 	error = device_register(&node->dev);
620 
621 	if (error)
622 		put_device(&node->dev);
623 	else {
624 		hugetlb_register_node(node);
625 
626 		compaction_register_node(node);
627 	}
628 	return error;
629 }
630 
631 /**
632  * unregister_node - unregister a node device
633  * @node: node going away
634  *
635  * Unregisters a node device @node.  All the devices on the node must be
636  * unregistered before calling this function.
637  */
638 void unregister_node(struct node *node)
639 {
640 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
641 	node_remove_accesses(node);
642 	node_remove_caches(node);
643 	device_unregister(&node->dev);
644 }
645 
646 struct node *node_devices[MAX_NUMNODES];
647 
648 /*
649  * register cpu under node
650  */
651 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
652 {
653 	int ret;
654 	struct device *obj;
655 
656 	if (!node_online(nid))
657 		return 0;
658 
659 	obj = get_cpu_device(cpu);
660 	if (!obj)
661 		return 0;
662 
663 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
664 				&obj->kobj,
665 				kobject_name(&obj->kobj));
666 	if (ret)
667 		return ret;
668 
669 	return sysfs_create_link(&obj->kobj,
670 				 &node_devices[nid]->dev.kobj,
671 				 kobject_name(&node_devices[nid]->dev.kobj));
672 }
673 
674 /**
675  * register_memory_node_under_compute_node - link memory node to its compute
676  *					     node for a given access class.
677  * @mem_nid:	Memory node number
678  * @cpu_nid:	Cpu  node number
679  * @access:	Access class to register
680  *
681  * Description:
682  * 	For use with platforms that may have separate memory and compute nodes.
683  * 	This function will export node relationships linking which memory
684  * 	initiator nodes can access memory targets at a given ranked access
685  * 	class.
686  */
687 int register_memory_node_under_compute_node(unsigned int mem_nid,
688 					    unsigned int cpu_nid,
689 					    unsigned access)
690 {
691 	struct node *init_node, *targ_node;
692 	struct node_access_nodes *initiator, *target;
693 	int ret;
694 
695 	if (!node_online(cpu_nid) || !node_online(mem_nid))
696 		return -ENODEV;
697 
698 	init_node = node_devices[cpu_nid];
699 	targ_node = node_devices[mem_nid];
700 	initiator = node_init_node_access(init_node, access);
701 	target = node_init_node_access(targ_node, access);
702 	if (!initiator || !target)
703 		return -ENOMEM;
704 
705 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
706 				      &targ_node->dev.kobj,
707 				      dev_name(&targ_node->dev));
708 	if (ret)
709 		return ret;
710 
711 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
712 				      &init_node->dev.kobj,
713 				      dev_name(&init_node->dev));
714 	if (ret)
715 		goto err;
716 
717 	return 0;
718  err:
719 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
720 				     dev_name(&targ_node->dev));
721 	return ret;
722 }
723 
724 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
725 {
726 	struct device *obj;
727 
728 	if (!node_online(nid))
729 		return 0;
730 
731 	obj = get_cpu_device(cpu);
732 	if (!obj)
733 		return 0;
734 
735 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
736 			  kobject_name(&obj->kobj));
737 	sysfs_remove_link(&obj->kobj,
738 			  kobject_name(&node_devices[nid]->dev.kobj));
739 
740 	return 0;
741 }
742 
743 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
744 static int __ref get_nid_for_pfn(unsigned long pfn)
745 {
746 	if (!pfn_valid_within(pfn))
747 		return -1;
748 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
749 	if (system_state < SYSTEM_RUNNING)
750 		return early_pfn_to_nid(pfn);
751 #endif
752 	return pfn_to_nid(pfn);
753 }
754 
755 /* register memory section under specified node if it spans that node */
756 static int register_mem_sect_under_node(struct memory_block *mem_blk,
757 					 void *arg)
758 {
759 	int ret, nid = *(int *)arg;
760 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
761 
762 	mem_blk->nid = nid;
763 
764 	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
765 	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
766 	sect_end_pfn += PAGES_PER_SECTION - 1;
767 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
768 		int page_nid;
769 
770 		/*
771 		 * memory block could have several absent sections from start.
772 		 * skip pfn range from absent section
773 		 */
774 		if (!pfn_present(pfn)) {
775 			pfn = round_down(pfn + PAGES_PER_SECTION,
776 					 PAGES_PER_SECTION) - 1;
777 			continue;
778 		}
779 
780 		/*
781 		 * We need to check if page belongs to nid only for the boot
782 		 * case, during hotplug we know that all pages in the memory
783 		 * block belong to the same node.
784 		 */
785 		if (system_state == SYSTEM_BOOTING) {
786 			page_nid = get_nid_for_pfn(pfn);
787 			if (page_nid < 0)
788 				continue;
789 			if (page_nid != nid)
790 				continue;
791 		}
792 		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
793 					&mem_blk->dev.kobj,
794 					kobject_name(&mem_blk->dev.kobj));
795 		if (ret)
796 			return ret;
797 
798 		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
799 				&node_devices[nid]->dev.kobj,
800 				kobject_name(&node_devices[nid]->dev.kobj));
801 	}
802 	/* mem section does not span the specified node */
803 	return 0;
804 }
805 
806 /*
807  * Unregister memory block device under all nodes that it spans.
808  * Has to be called with mem_sysfs_mutex held (due to unlinked_nodes).
809  */
810 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
811 {
812 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
813 	static nodemask_t unlinked_nodes;
814 
815 	nodes_clear(unlinked_nodes);
816 	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
817 	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
818 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
819 		int nid;
820 
821 		nid = get_nid_for_pfn(pfn);
822 		if (nid < 0)
823 			continue;
824 		if (!node_online(nid))
825 			continue;
826 		if (node_test_and_set(nid, unlinked_nodes))
827 			continue;
828 		sysfs_remove_link(&node_devices[nid]->dev.kobj,
829 			 kobject_name(&mem_blk->dev.kobj));
830 		sysfs_remove_link(&mem_blk->dev.kobj,
831 			 kobject_name(&node_devices[nid]->dev.kobj));
832 	}
833 }
834 
835 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn)
836 {
837 	return walk_memory_blocks(PFN_PHYS(start_pfn),
838 				  PFN_PHYS(end_pfn - start_pfn), (void *)&nid,
839 				  register_mem_sect_under_node);
840 }
841 
842 #ifdef CONFIG_HUGETLBFS
843 /*
844  * Handle per node hstate attribute [un]registration on transistions
845  * to/from memoryless state.
846  */
847 static void node_hugetlb_work(struct work_struct *work)
848 {
849 	struct node *node = container_of(work, struct node, node_work);
850 
851 	/*
852 	 * We only get here when a node transitions to/from memoryless state.
853 	 * We can detect which transition occurred by examining whether the
854 	 * node has memory now.  hugetlb_register_node() already check this
855 	 * so we try to register the attributes.  If that fails, then the
856 	 * node has transitioned to memoryless, try to unregister the
857 	 * attributes.
858 	 */
859 	if (!hugetlb_register_node(node))
860 		hugetlb_unregister_node(node);
861 }
862 
863 static void init_node_hugetlb_work(int nid)
864 {
865 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
866 }
867 
868 static int node_memory_callback(struct notifier_block *self,
869 				unsigned long action, void *arg)
870 {
871 	struct memory_notify *mnb = arg;
872 	int nid = mnb->status_change_nid;
873 
874 	switch (action) {
875 	case MEM_ONLINE:
876 	case MEM_OFFLINE:
877 		/*
878 		 * offload per node hstate [un]registration to a work thread
879 		 * when transitioning to/from memoryless state.
880 		 */
881 		if (nid != NUMA_NO_NODE)
882 			schedule_work(&node_devices[nid]->node_work);
883 		break;
884 
885 	case MEM_GOING_ONLINE:
886 	case MEM_GOING_OFFLINE:
887 	case MEM_CANCEL_ONLINE:
888 	case MEM_CANCEL_OFFLINE:
889 	default:
890 		break;
891 	}
892 
893 	return NOTIFY_OK;
894 }
895 #endif	/* CONFIG_HUGETLBFS */
896 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
897 
898 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
899     !defined(CONFIG_HUGETLBFS)
900 static inline int node_memory_callback(struct notifier_block *self,
901 				unsigned long action, void *arg)
902 {
903 	return NOTIFY_OK;
904 }
905 
906 static void init_node_hugetlb_work(int nid) { }
907 
908 #endif
909 
910 int __register_one_node(int nid)
911 {
912 	int error;
913 	int cpu;
914 
915 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
916 	if (!node_devices[nid])
917 		return -ENOMEM;
918 
919 	error = register_node(node_devices[nid], nid);
920 
921 	/* link cpu under this node */
922 	for_each_present_cpu(cpu) {
923 		if (cpu_to_node(cpu) == nid)
924 			register_cpu_under_node(cpu, nid);
925 	}
926 
927 	INIT_LIST_HEAD(&node_devices[nid]->access_list);
928 	/* initialize work queue for memory hot plug */
929 	init_node_hugetlb_work(nid);
930 	node_init_caches(nid);
931 
932 	return error;
933 }
934 
935 void unregister_one_node(int nid)
936 {
937 	if (!node_devices[nid])
938 		return;
939 
940 	unregister_node(node_devices[nid]);
941 	node_devices[nid] = NULL;
942 }
943 
944 /*
945  * node states attributes
946  */
947 
948 static ssize_t print_nodes_state(enum node_states state, char *buf)
949 {
950 	int n;
951 
952 	n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
953 		      nodemask_pr_args(&node_states[state]));
954 	buf[n++] = '\n';
955 	buf[n] = '\0';
956 	return n;
957 }
958 
959 struct node_attr {
960 	struct device_attribute attr;
961 	enum node_states state;
962 };
963 
964 static ssize_t show_node_state(struct device *dev,
965 			       struct device_attribute *attr, char *buf)
966 {
967 	struct node_attr *na = container_of(attr, struct node_attr, attr);
968 	return print_nodes_state(na->state, buf);
969 }
970 
971 #define _NODE_ATTR(name, state) \
972 	{ __ATTR(name, 0444, show_node_state, NULL), state }
973 
974 static struct node_attr node_state_attr[] = {
975 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
976 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
977 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
978 #ifdef CONFIG_HIGHMEM
979 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
980 #endif
981 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
982 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
983 };
984 
985 static struct attribute *node_state_attrs[] = {
986 	&node_state_attr[N_POSSIBLE].attr.attr,
987 	&node_state_attr[N_ONLINE].attr.attr,
988 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
989 #ifdef CONFIG_HIGHMEM
990 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
991 #endif
992 	&node_state_attr[N_MEMORY].attr.attr,
993 	&node_state_attr[N_CPU].attr.attr,
994 	NULL
995 };
996 
997 static struct attribute_group memory_root_attr_group = {
998 	.attrs = node_state_attrs,
999 };
1000 
1001 static const struct attribute_group *cpu_root_attr_groups[] = {
1002 	&memory_root_attr_group,
1003 	NULL,
1004 };
1005 
1006 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
1007 static int __init register_node_type(void)
1008 {
1009 	int ret;
1010 
1011  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1012  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1013 
1014 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1015 	if (!ret) {
1016 		static struct notifier_block node_memory_callback_nb = {
1017 			.notifier_call = node_memory_callback,
1018 			.priority = NODE_CALLBACK_PRI,
1019 		};
1020 		register_hotmemory_notifier(&node_memory_callback_nb);
1021 	}
1022 
1023 	/*
1024 	 * Note:  we're not going to unregister the node class if we fail
1025 	 * to register the node state class attribute files.
1026 	 */
1027 	return ret;
1028 }
1029 postcore_initcall(register_node_type);
1030