xref: /openbmc/linux/drivers/base/node.c (revision b9df3997)
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 		       "Node %d FileHugePages: %8lu kB\n"
431 		       "Node %d FilePmdMapped: %8lu kB\n"
432 #endif
433 			,
434 		       nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
435 		       nid, K(node_page_state(pgdat, NR_WRITEBACK)),
436 		       nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
437 		       nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
438 		       nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
439 		       nid, K(i.sharedram),
440 		       nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
441 		       nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
442 		       nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
443 		       nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
444 		       nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
445 		       nid, K(sreclaimable +
446 			      node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
447 		       nid, K(sreclaimable + sunreclaimable),
448 		       nid, K(sreclaimable),
449 		       nid, K(sunreclaimable)
450 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
451 		       ,
452 		       nid, K(node_page_state(pgdat, NR_ANON_THPS) *
453 				       HPAGE_PMD_NR),
454 		       nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
455 				       HPAGE_PMD_NR),
456 		       nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
457 				       HPAGE_PMD_NR),
458 		       nid, K(node_page_state(pgdat, NR_FILE_THPS) *
459 				       HPAGE_PMD_NR),
460 		       nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED) *
461 				       HPAGE_PMD_NR)
462 #endif
463 		       );
464 	n += hugetlb_report_node_meminfo(nid, buf + n);
465 	return n;
466 }
467 
468 #undef K
469 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
470 
471 static ssize_t node_read_numastat(struct device *dev,
472 				struct device_attribute *attr, char *buf)
473 {
474 	return sprintf(buf,
475 		       "numa_hit %lu\n"
476 		       "numa_miss %lu\n"
477 		       "numa_foreign %lu\n"
478 		       "interleave_hit %lu\n"
479 		       "local_node %lu\n"
480 		       "other_node %lu\n",
481 		       sum_zone_numa_state(dev->id, NUMA_HIT),
482 		       sum_zone_numa_state(dev->id, NUMA_MISS),
483 		       sum_zone_numa_state(dev->id, NUMA_FOREIGN),
484 		       sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
485 		       sum_zone_numa_state(dev->id, NUMA_LOCAL),
486 		       sum_zone_numa_state(dev->id, NUMA_OTHER));
487 }
488 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
489 
490 static ssize_t node_read_vmstat(struct device *dev,
491 				struct device_attribute *attr, char *buf)
492 {
493 	int nid = dev->id;
494 	struct pglist_data *pgdat = NODE_DATA(nid);
495 	int i;
496 	int n = 0;
497 
498 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
499 		n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
500 			     sum_zone_node_page_state(nid, i));
501 
502 #ifdef CONFIG_NUMA
503 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
504 		n += sprintf(buf+n, "%s %lu\n",
505 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
506 			     sum_zone_numa_state(nid, i));
507 #endif
508 
509 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
510 		n += sprintf(buf+n, "%s %lu\n",
511 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
512 			     NR_VM_NUMA_STAT_ITEMS],
513 			     node_page_state(pgdat, i));
514 
515 	return n;
516 }
517 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
518 
519 static ssize_t node_read_distance(struct device *dev,
520 			struct device_attribute *attr, char *buf)
521 {
522 	int nid = dev->id;
523 	int len = 0;
524 	int i;
525 
526 	/*
527 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
528 	 * at the most (distance + space or newline).
529 	 */
530 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
531 
532 	for_each_online_node(i)
533 		len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
534 
535 	len += sprintf(buf + len, "\n");
536 	return len;
537 }
538 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
539 
540 static struct attribute *node_dev_attrs[] = {
541 	&dev_attr_cpumap.attr,
542 	&dev_attr_cpulist.attr,
543 	&dev_attr_meminfo.attr,
544 	&dev_attr_numastat.attr,
545 	&dev_attr_distance.attr,
546 	&dev_attr_vmstat.attr,
547 	NULL
548 };
549 ATTRIBUTE_GROUPS(node_dev);
550 
551 #ifdef CONFIG_HUGETLBFS
552 /*
553  * hugetlbfs per node attributes registration interface:
554  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
555  * it will register its per node attributes for all online nodes with
556  * memory.  It will also call register_hugetlbfs_with_node(), below, to
557  * register its attribute registration functions with this node driver.
558  * Once these hooks have been initialized, the node driver will call into
559  * the hugetlb module to [un]register attributes for hot-plugged nodes.
560  */
561 static node_registration_func_t __hugetlb_register_node;
562 static node_registration_func_t __hugetlb_unregister_node;
563 
564 static inline bool hugetlb_register_node(struct node *node)
565 {
566 	if (__hugetlb_register_node &&
567 			node_state(node->dev.id, N_MEMORY)) {
568 		__hugetlb_register_node(node);
569 		return true;
570 	}
571 	return false;
572 }
573 
574 static inline void hugetlb_unregister_node(struct node *node)
575 {
576 	if (__hugetlb_unregister_node)
577 		__hugetlb_unregister_node(node);
578 }
579 
580 void register_hugetlbfs_with_node(node_registration_func_t doregister,
581 				  node_registration_func_t unregister)
582 {
583 	__hugetlb_register_node   = doregister;
584 	__hugetlb_unregister_node = unregister;
585 }
586 #else
587 static inline void hugetlb_register_node(struct node *node) {}
588 
589 static inline void hugetlb_unregister_node(struct node *node) {}
590 #endif
591 
592 static void node_device_release(struct device *dev)
593 {
594 	struct node *node = to_node(dev);
595 
596 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
597 	/*
598 	 * We schedule the work only when a memory section is
599 	 * onlined/offlined on this node. When we come here,
600 	 * all the memory on this node has been offlined,
601 	 * so we won't enqueue new work to this work.
602 	 *
603 	 * The work is using node->node_work, so we should
604 	 * flush work before freeing the memory.
605 	 */
606 	flush_work(&node->node_work);
607 #endif
608 	kfree(node);
609 }
610 
611 /*
612  * register_node - Setup a sysfs device for a node.
613  * @num - Node number to use when creating the device.
614  *
615  * Initialize and register the node device.
616  */
617 static int register_node(struct node *node, int num)
618 {
619 	int error;
620 
621 	node->dev.id = num;
622 	node->dev.bus = &node_subsys;
623 	node->dev.release = node_device_release;
624 	node->dev.groups = node_dev_groups;
625 	error = device_register(&node->dev);
626 
627 	if (error)
628 		put_device(&node->dev);
629 	else {
630 		hugetlb_register_node(node);
631 
632 		compaction_register_node(node);
633 	}
634 	return error;
635 }
636 
637 /**
638  * unregister_node - unregister a node device
639  * @node: node going away
640  *
641  * Unregisters a node device @node.  All the devices on the node must be
642  * unregistered before calling this function.
643  */
644 void unregister_node(struct node *node)
645 {
646 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
647 	node_remove_accesses(node);
648 	node_remove_caches(node);
649 	device_unregister(&node->dev);
650 }
651 
652 struct node *node_devices[MAX_NUMNODES];
653 
654 /*
655  * register cpu under node
656  */
657 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
658 {
659 	int ret;
660 	struct device *obj;
661 
662 	if (!node_online(nid))
663 		return 0;
664 
665 	obj = get_cpu_device(cpu);
666 	if (!obj)
667 		return 0;
668 
669 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
670 				&obj->kobj,
671 				kobject_name(&obj->kobj));
672 	if (ret)
673 		return ret;
674 
675 	return sysfs_create_link(&obj->kobj,
676 				 &node_devices[nid]->dev.kobj,
677 				 kobject_name(&node_devices[nid]->dev.kobj));
678 }
679 
680 /**
681  * register_memory_node_under_compute_node - link memory node to its compute
682  *					     node for a given access class.
683  * @mem_nid:	Memory node number
684  * @cpu_nid:	Cpu  node number
685  * @access:	Access class to register
686  *
687  * Description:
688  * 	For use with platforms that may have separate memory and compute nodes.
689  * 	This function will export node relationships linking which memory
690  * 	initiator nodes can access memory targets at a given ranked access
691  * 	class.
692  */
693 int register_memory_node_under_compute_node(unsigned int mem_nid,
694 					    unsigned int cpu_nid,
695 					    unsigned access)
696 {
697 	struct node *init_node, *targ_node;
698 	struct node_access_nodes *initiator, *target;
699 	int ret;
700 
701 	if (!node_online(cpu_nid) || !node_online(mem_nid))
702 		return -ENODEV;
703 
704 	init_node = node_devices[cpu_nid];
705 	targ_node = node_devices[mem_nid];
706 	initiator = node_init_node_access(init_node, access);
707 	target = node_init_node_access(targ_node, access);
708 	if (!initiator || !target)
709 		return -ENOMEM;
710 
711 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
712 				      &targ_node->dev.kobj,
713 				      dev_name(&targ_node->dev));
714 	if (ret)
715 		return ret;
716 
717 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
718 				      &init_node->dev.kobj,
719 				      dev_name(&init_node->dev));
720 	if (ret)
721 		goto err;
722 
723 	return 0;
724  err:
725 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
726 				     dev_name(&targ_node->dev));
727 	return ret;
728 }
729 
730 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
731 {
732 	struct device *obj;
733 
734 	if (!node_online(nid))
735 		return 0;
736 
737 	obj = get_cpu_device(cpu);
738 	if (!obj)
739 		return 0;
740 
741 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
742 			  kobject_name(&obj->kobj));
743 	sysfs_remove_link(&obj->kobj,
744 			  kobject_name(&node_devices[nid]->dev.kobj));
745 
746 	return 0;
747 }
748 
749 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
750 static int __ref get_nid_for_pfn(unsigned long pfn)
751 {
752 	if (!pfn_valid_within(pfn))
753 		return -1;
754 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
755 	if (system_state < SYSTEM_RUNNING)
756 		return early_pfn_to_nid(pfn);
757 #endif
758 	return pfn_to_nid(pfn);
759 }
760 
761 /* register memory section under specified node if it spans that node */
762 static int register_mem_sect_under_node(struct memory_block *mem_blk,
763 					 void *arg)
764 {
765 	unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
766 	unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
767 	unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
768 	int ret, nid = *(int *)arg;
769 	unsigned long pfn;
770 
771 	for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
772 		int page_nid;
773 
774 		/*
775 		 * memory block could have several absent sections from start.
776 		 * skip pfn range from absent section
777 		 */
778 		if (!pfn_present(pfn)) {
779 			pfn = round_down(pfn + PAGES_PER_SECTION,
780 					 PAGES_PER_SECTION) - 1;
781 			continue;
782 		}
783 
784 		/*
785 		 * We need to check if page belongs to nid only for the boot
786 		 * case, during hotplug we know that all pages in the memory
787 		 * block belong to the same node.
788 		 */
789 		if (system_state == SYSTEM_BOOTING) {
790 			page_nid = get_nid_for_pfn(pfn);
791 			if (page_nid < 0)
792 				continue;
793 			if (page_nid != nid)
794 				continue;
795 		}
796 
797 		/*
798 		 * If this memory block spans multiple nodes, we only indicate
799 		 * the last processed node.
800 		 */
801 		mem_blk->nid = nid;
802 
803 		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
804 					&mem_blk->dev.kobj,
805 					kobject_name(&mem_blk->dev.kobj));
806 		if (ret)
807 			return ret;
808 
809 		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
810 				&node_devices[nid]->dev.kobj,
811 				kobject_name(&node_devices[nid]->dev.kobj));
812 	}
813 	/* mem section does not span the specified node */
814 	return 0;
815 }
816 
817 /*
818  * Unregister a memory block device under the node it spans. Memory blocks
819  * with multiple nodes cannot be offlined and therefore also never be removed.
820  */
821 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
822 {
823 	if (mem_blk->nid == NUMA_NO_NODE)
824 		return;
825 
826 	sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
827 			  kobject_name(&mem_blk->dev.kobj));
828 	sysfs_remove_link(&mem_blk->dev.kobj,
829 			  kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
830 }
831 
832 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn)
833 {
834 	return walk_memory_blocks(PFN_PHYS(start_pfn),
835 				  PFN_PHYS(end_pfn - start_pfn), (void *)&nid,
836 				  register_mem_sect_under_node);
837 }
838 
839 #ifdef CONFIG_HUGETLBFS
840 /*
841  * Handle per node hstate attribute [un]registration on transistions
842  * to/from memoryless state.
843  */
844 static void node_hugetlb_work(struct work_struct *work)
845 {
846 	struct node *node = container_of(work, struct node, node_work);
847 
848 	/*
849 	 * We only get here when a node transitions to/from memoryless state.
850 	 * We can detect which transition occurred by examining whether the
851 	 * node has memory now.  hugetlb_register_node() already check this
852 	 * so we try to register the attributes.  If that fails, then the
853 	 * node has transitioned to memoryless, try to unregister the
854 	 * attributes.
855 	 */
856 	if (!hugetlb_register_node(node))
857 		hugetlb_unregister_node(node);
858 }
859 
860 static void init_node_hugetlb_work(int nid)
861 {
862 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
863 }
864 
865 static int node_memory_callback(struct notifier_block *self,
866 				unsigned long action, void *arg)
867 {
868 	struct memory_notify *mnb = arg;
869 	int nid = mnb->status_change_nid;
870 
871 	switch (action) {
872 	case MEM_ONLINE:
873 	case MEM_OFFLINE:
874 		/*
875 		 * offload per node hstate [un]registration to a work thread
876 		 * when transitioning to/from memoryless state.
877 		 */
878 		if (nid != NUMA_NO_NODE)
879 			schedule_work(&node_devices[nid]->node_work);
880 		break;
881 
882 	case MEM_GOING_ONLINE:
883 	case MEM_GOING_OFFLINE:
884 	case MEM_CANCEL_ONLINE:
885 	case MEM_CANCEL_OFFLINE:
886 	default:
887 		break;
888 	}
889 
890 	return NOTIFY_OK;
891 }
892 #endif	/* CONFIG_HUGETLBFS */
893 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
894 
895 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
896     !defined(CONFIG_HUGETLBFS)
897 static inline int node_memory_callback(struct notifier_block *self,
898 				unsigned long action, void *arg)
899 {
900 	return NOTIFY_OK;
901 }
902 
903 static void init_node_hugetlb_work(int nid) { }
904 
905 #endif
906 
907 int __register_one_node(int nid)
908 {
909 	int error;
910 	int cpu;
911 
912 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
913 	if (!node_devices[nid])
914 		return -ENOMEM;
915 
916 	error = register_node(node_devices[nid], nid);
917 
918 	/* link cpu under this node */
919 	for_each_present_cpu(cpu) {
920 		if (cpu_to_node(cpu) == nid)
921 			register_cpu_under_node(cpu, nid);
922 	}
923 
924 	INIT_LIST_HEAD(&node_devices[nid]->access_list);
925 	/* initialize work queue for memory hot plug */
926 	init_node_hugetlb_work(nid);
927 	node_init_caches(nid);
928 
929 	return error;
930 }
931 
932 void unregister_one_node(int nid)
933 {
934 	if (!node_devices[nid])
935 		return;
936 
937 	unregister_node(node_devices[nid]);
938 	node_devices[nid] = NULL;
939 }
940 
941 /*
942  * node states attributes
943  */
944 
945 static ssize_t print_nodes_state(enum node_states state, char *buf)
946 {
947 	int n;
948 
949 	n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
950 		      nodemask_pr_args(&node_states[state]));
951 	buf[n++] = '\n';
952 	buf[n] = '\0';
953 	return n;
954 }
955 
956 struct node_attr {
957 	struct device_attribute attr;
958 	enum node_states state;
959 };
960 
961 static ssize_t show_node_state(struct device *dev,
962 			       struct device_attribute *attr, char *buf)
963 {
964 	struct node_attr *na = container_of(attr, struct node_attr, attr);
965 	return print_nodes_state(na->state, buf);
966 }
967 
968 #define _NODE_ATTR(name, state) \
969 	{ __ATTR(name, 0444, show_node_state, NULL), state }
970 
971 static struct node_attr node_state_attr[] = {
972 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
973 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
974 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
975 #ifdef CONFIG_HIGHMEM
976 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
977 #endif
978 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
979 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
980 };
981 
982 static struct attribute *node_state_attrs[] = {
983 	&node_state_attr[N_POSSIBLE].attr.attr,
984 	&node_state_attr[N_ONLINE].attr.attr,
985 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
986 #ifdef CONFIG_HIGHMEM
987 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
988 #endif
989 	&node_state_attr[N_MEMORY].attr.attr,
990 	&node_state_attr[N_CPU].attr.attr,
991 	NULL
992 };
993 
994 static struct attribute_group memory_root_attr_group = {
995 	.attrs = node_state_attrs,
996 };
997 
998 static const struct attribute_group *cpu_root_attr_groups[] = {
999 	&memory_root_attr_group,
1000 	NULL,
1001 };
1002 
1003 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
1004 static int __init register_node_type(void)
1005 {
1006 	int ret;
1007 
1008  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1009  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1010 
1011 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1012 	if (!ret) {
1013 		static struct notifier_block node_memory_callback_nb = {
1014 			.notifier_call = node_memory_callback,
1015 			.priority = NODE_CALLBACK_PRI,
1016 		};
1017 		register_hotmemory_notifier(&node_memory_callback_nb);
1018 	}
1019 
1020 	/*
1021 	 * Note:  we're not going to unregister the node class if we fail
1022 	 * to register the node state class attribute files.
1023 	 */
1024 	return ret;
1025 }
1026 postcore_initcall(register_node_type);
1027