xref: /linux/drivers/base/node.c (revision 6c363eafc4d637ac4bd83d4a7dd06dd3cfbe7c5f)
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 cpumap_show(struct device *dev,
50 				  struct device_attribute *attr,
51 				  char *buf)
52 {
53 	return node_read_cpumap(dev, false, buf);
54 }
55 
56 static DEVICE_ATTR_RO(cpumap);
57 
58 static inline ssize_t cpulist_show(struct device *dev,
59 				   struct device_attribute *attr,
60 				   char *buf)
61 {
62 	return node_read_cpumap(dev, true, buf);
63 }
64 
65 static DEVICE_ATTR_RO(cpulist);
66 
67 /**
68  * struct node_access_nodes - Access class device to hold user visible
69  * 			      relationships to other nodes.
70  * @dev:	Device for this memory access class
71  * @list_node:	List element in the node's access list
72  * @access:	The access class rank
73  * @hmem_attrs: Heterogeneous memory performance attributes
74  */
75 struct node_access_nodes {
76 	struct device		dev;
77 	struct list_head	list_node;
78 	unsigned		access;
79 #ifdef CONFIG_HMEM_REPORTING
80 	struct node_hmem_attrs	hmem_attrs;
81 #endif
82 };
83 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
84 
85 static struct attribute *node_init_access_node_attrs[] = {
86 	NULL,
87 };
88 
89 static struct attribute *node_targ_access_node_attrs[] = {
90 	NULL,
91 };
92 
93 static const struct attribute_group initiators = {
94 	.name	= "initiators",
95 	.attrs	= node_init_access_node_attrs,
96 };
97 
98 static const struct attribute_group targets = {
99 	.name	= "targets",
100 	.attrs	= node_targ_access_node_attrs,
101 };
102 
103 static const struct attribute_group *node_access_node_groups[] = {
104 	&initiators,
105 	&targets,
106 	NULL,
107 };
108 
109 static void node_remove_accesses(struct node *node)
110 {
111 	struct node_access_nodes *c, *cnext;
112 
113 	list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
114 		list_del(&c->list_node);
115 		device_unregister(&c->dev);
116 	}
117 }
118 
119 static void node_access_release(struct device *dev)
120 {
121 	kfree(to_access_nodes(dev));
122 }
123 
124 static struct node_access_nodes *node_init_node_access(struct node *node,
125 						       unsigned access)
126 {
127 	struct node_access_nodes *access_node;
128 	struct device *dev;
129 
130 	list_for_each_entry(access_node, &node->access_list, list_node)
131 		if (access_node->access == access)
132 			return access_node;
133 
134 	access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
135 	if (!access_node)
136 		return NULL;
137 
138 	access_node->access = access;
139 	dev = &access_node->dev;
140 	dev->parent = &node->dev;
141 	dev->release = node_access_release;
142 	dev->groups = node_access_node_groups;
143 	if (dev_set_name(dev, "access%u", access))
144 		goto free;
145 
146 	if (device_register(dev))
147 		goto free_name;
148 
149 	pm_runtime_no_callbacks(dev);
150 	list_add_tail(&access_node->list_node, &node->access_list);
151 	return access_node;
152 free_name:
153 	kfree_const(dev->kobj.name);
154 free:
155 	kfree(access_node);
156 	return NULL;
157 }
158 
159 #ifdef CONFIG_HMEM_REPORTING
160 #define ACCESS_ATTR(name)						\
161 static ssize_t name##_show(struct device *dev,				\
162 			   struct device_attribute *attr,		\
163 			   char *buf)					\
164 {									\
165 	return sysfs_emit(buf, "%u\n",					\
166 			  to_access_nodes(dev)->hmem_attrs.name);	\
167 }									\
168 static DEVICE_ATTR_RO(name)
169 
170 ACCESS_ATTR(read_bandwidth);
171 ACCESS_ATTR(read_latency);
172 ACCESS_ATTR(write_bandwidth);
173 ACCESS_ATTR(write_latency);
174 
175 static struct attribute *access_attrs[] = {
176 	&dev_attr_read_bandwidth.attr,
177 	&dev_attr_read_latency.attr,
178 	&dev_attr_write_bandwidth.attr,
179 	&dev_attr_write_latency.attr,
180 	NULL,
181 };
182 
183 /**
184  * node_set_perf_attrs - Set the performance values for given access class
185  * @nid: Node identifier to be set
186  * @hmem_attrs: Heterogeneous memory performance attributes
187  * @access: The access class the for the given attributes
188  */
189 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
190 			 unsigned access)
191 {
192 	struct node_access_nodes *c;
193 	struct node *node;
194 	int i;
195 
196 	if (WARN_ON_ONCE(!node_online(nid)))
197 		return;
198 
199 	node = node_devices[nid];
200 	c = node_init_node_access(node, access);
201 	if (!c)
202 		return;
203 
204 	c->hmem_attrs = *hmem_attrs;
205 	for (i = 0; access_attrs[i] != NULL; i++) {
206 		if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
207 					    "initiators")) {
208 			pr_info("failed to add performance attribute to node %d\n",
209 				nid);
210 			break;
211 		}
212 	}
213 }
214 
215 /**
216  * struct node_cache_info - Internal tracking for memory node caches
217  * @dev:	Device represeting the cache level
218  * @node:	List element for tracking in the node
219  * @cache_attrs:Attributes for this cache level
220  */
221 struct node_cache_info {
222 	struct device dev;
223 	struct list_head node;
224 	struct node_cache_attrs cache_attrs;
225 };
226 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
227 
228 #define CACHE_ATTR(name, fmt) 						\
229 static ssize_t name##_show(struct device *dev,				\
230 			   struct device_attribute *attr,		\
231 			   char *buf)					\
232 {									\
233 	return sysfs_emit(buf, fmt "\n",				\
234 			  to_cache_info(dev)->cache_attrs.name);	\
235 }									\
236 DEVICE_ATTR_RO(name);
237 
238 CACHE_ATTR(size, "%llu")
239 CACHE_ATTR(line_size, "%u")
240 CACHE_ATTR(indexing, "%u")
241 CACHE_ATTR(write_policy, "%u")
242 
243 static struct attribute *cache_attrs[] = {
244 	&dev_attr_indexing.attr,
245 	&dev_attr_size.attr,
246 	&dev_attr_line_size.attr,
247 	&dev_attr_write_policy.attr,
248 	NULL,
249 };
250 ATTRIBUTE_GROUPS(cache);
251 
252 static void node_cache_release(struct device *dev)
253 {
254 	kfree(dev);
255 }
256 
257 static void node_cacheinfo_release(struct device *dev)
258 {
259 	struct node_cache_info *info = to_cache_info(dev);
260 	kfree(info);
261 }
262 
263 static void node_init_cache_dev(struct node *node)
264 {
265 	struct device *dev;
266 
267 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
268 	if (!dev)
269 		return;
270 
271 	dev->parent = &node->dev;
272 	dev->release = node_cache_release;
273 	if (dev_set_name(dev, "memory_side_cache"))
274 		goto free_dev;
275 
276 	if (device_register(dev))
277 		goto free_name;
278 
279 	pm_runtime_no_callbacks(dev);
280 	node->cache_dev = dev;
281 	return;
282 free_name:
283 	kfree_const(dev->kobj.name);
284 free_dev:
285 	kfree(dev);
286 }
287 
288 /**
289  * node_add_cache() - add cache attribute to a memory node
290  * @nid: Node identifier that has new cache attributes
291  * @cache_attrs: Attributes for the cache being added
292  */
293 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
294 {
295 	struct node_cache_info *info;
296 	struct device *dev;
297 	struct node *node;
298 
299 	if (!node_online(nid) || !node_devices[nid])
300 		return;
301 
302 	node = node_devices[nid];
303 	list_for_each_entry(info, &node->cache_attrs, node) {
304 		if (info->cache_attrs.level == cache_attrs->level) {
305 			dev_warn(&node->dev,
306 				"attempt to add duplicate cache level:%d\n",
307 				cache_attrs->level);
308 			return;
309 		}
310 	}
311 
312 	if (!node->cache_dev)
313 		node_init_cache_dev(node);
314 	if (!node->cache_dev)
315 		return;
316 
317 	info = kzalloc(sizeof(*info), GFP_KERNEL);
318 	if (!info)
319 		return;
320 
321 	dev = &info->dev;
322 	dev->parent = node->cache_dev;
323 	dev->release = node_cacheinfo_release;
324 	dev->groups = cache_groups;
325 	if (dev_set_name(dev, "index%d", cache_attrs->level))
326 		goto free_cache;
327 
328 	info->cache_attrs = *cache_attrs;
329 	if (device_register(dev)) {
330 		dev_warn(&node->dev, "failed to add cache level:%d\n",
331 			 cache_attrs->level);
332 		goto free_name;
333 	}
334 	pm_runtime_no_callbacks(dev);
335 	list_add_tail(&info->node, &node->cache_attrs);
336 	return;
337 free_name:
338 	kfree_const(dev->kobj.name);
339 free_cache:
340 	kfree(info);
341 }
342 
343 static void node_remove_caches(struct node *node)
344 {
345 	struct node_cache_info *info, *next;
346 
347 	if (!node->cache_dev)
348 		return;
349 
350 	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
351 		list_del(&info->node);
352 		device_unregister(&info->dev);
353 	}
354 	device_unregister(node->cache_dev);
355 }
356 
357 static void node_init_caches(unsigned int nid)
358 {
359 	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
360 }
361 #else
362 static void node_init_caches(unsigned int nid) { }
363 static void node_remove_caches(struct node *node) { }
364 #endif
365 
366 #define K(x) ((x) << (PAGE_SHIFT - 10))
367 static ssize_t node_read_meminfo(struct device *dev,
368 			struct device_attribute *attr, char *buf)
369 {
370 	int len = 0;
371 	int nid = dev->id;
372 	struct pglist_data *pgdat = NODE_DATA(nid);
373 	struct sysinfo i;
374 	unsigned long sreclaimable, sunreclaimable;
375 	unsigned long swapcached = 0;
376 
377 	si_meminfo_node(&i, nid);
378 	sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
379 	sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
380 #ifdef CONFIG_SWAP
381 	swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
382 #endif
383 	len = sysfs_emit_at(buf, len,
384 			    "Node %d MemTotal:       %8lu kB\n"
385 			    "Node %d MemFree:        %8lu kB\n"
386 			    "Node %d MemUsed:        %8lu kB\n"
387 			    "Node %d SwapCached:     %8lu kB\n"
388 			    "Node %d Active:         %8lu kB\n"
389 			    "Node %d Inactive:       %8lu kB\n"
390 			    "Node %d Active(anon):   %8lu kB\n"
391 			    "Node %d Inactive(anon): %8lu kB\n"
392 			    "Node %d Active(file):   %8lu kB\n"
393 			    "Node %d Inactive(file): %8lu kB\n"
394 			    "Node %d Unevictable:    %8lu kB\n"
395 			    "Node %d Mlocked:        %8lu kB\n",
396 			    nid, K(i.totalram),
397 			    nid, K(i.freeram),
398 			    nid, K(i.totalram - i.freeram),
399 			    nid, K(swapcached),
400 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
401 				   node_page_state(pgdat, NR_ACTIVE_FILE)),
402 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
403 				   node_page_state(pgdat, NR_INACTIVE_FILE)),
404 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
405 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
406 			    nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
407 			    nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
408 			    nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
409 			    nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
410 
411 #ifdef CONFIG_HIGHMEM
412 	len += sysfs_emit_at(buf, len,
413 			     "Node %d HighTotal:      %8lu kB\n"
414 			     "Node %d HighFree:       %8lu kB\n"
415 			     "Node %d LowTotal:       %8lu kB\n"
416 			     "Node %d LowFree:        %8lu kB\n",
417 			     nid, K(i.totalhigh),
418 			     nid, K(i.freehigh),
419 			     nid, K(i.totalram - i.totalhigh),
420 			     nid, K(i.freeram - i.freehigh));
421 #endif
422 	len += sysfs_emit_at(buf, len,
423 			     "Node %d Dirty:          %8lu kB\n"
424 			     "Node %d Writeback:      %8lu kB\n"
425 			     "Node %d FilePages:      %8lu kB\n"
426 			     "Node %d Mapped:         %8lu kB\n"
427 			     "Node %d AnonPages:      %8lu kB\n"
428 			     "Node %d Shmem:          %8lu kB\n"
429 			     "Node %d KernelStack:    %8lu kB\n"
430 #ifdef CONFIG_SHADOW_CALL_STACK
431 			     "Node %d ShadowCallStack:%8lu kB\n"
432 #endif
433 			     "Node %d PageTables:     %8lu kB\n"
434 			     "Node %d NFS_Unstable:   %8lu kB\n"
435 			     "Node %d Bounce:         %8lu kB\n"
436 			     "Node %d WritebackTmp:   %8lu kB\n"
437 			     "Node %d KReclaimable:   %8lu kB\n"
438 			     "Node %d Slab:           %8lu kB\n"
439 			     "Node %d SReclaimable:   %8lu kB\n"
440 			     "Node %d SUnreclaim:     %8lu kB\n"
441 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
442 			     "Node %d AnonHugePages:  %8lu kB\n"
443 			     "Node %d ShmemHugePages: %8lu kB\n"
444 			     "Node %d ShmemPmdMapped: %8lu kB\n"
445 			     "Node %d FileHugePages: %8lu kB\n"
446 			     "Node %d FilePmdMapped: %8lu kB\n"
447 #endif
448 			     ,
449 			     nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
450 			     nid, K(node_page_state(pgdat, NR_WRITEBACK)),
451 			     nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
452 			     nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
453 			     nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
454 			     nid, K(i.sharedram),
455 			     nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
456 #ifdef CONFIG_SHADOW_CALL_STACK
457 			     nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
458 #endif
459 			     nid, K(node_page_state(pgdat, NR_PAGETABLE)),
460 			     nid, 0UL,
461 			     nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
462 			     nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
463 			     nid, K(sreclaimable +
464 				    node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
465 			     nid, K(sreclaimable + sunreclaimable),
466 			     nid, K(sreclaimable),
467 			     nid, K(sunreclaimable)
468 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
469 			     ,
470 			     nid, K(node_page_state(pgdat, NR_ANON_THPS)),
471 			     nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
472 			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
473 			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
474 			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
475 #endif
476 			    );
477 	len += hugetlb_report_node_meminfo(buf, len, nid);
478 	return len;
479 }
480 
481 #undef K
482 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
483 
484 static ssize_t node_read_numastat(struct device *dev,
485 				  struct device_attribute *attr, char *buf)
486 {
487 	return sysfs_emit(buf,
488 			  "numa_hit %lu\n"
489 			  "numa_miss %lu\n"
490 			  "numa_foreign %lu\n"
491 			  "interleave_hit %lu\n"
492 			  "local_node %lu\n"
493 			  "other_node %lu\n",
494 			  sum_zone_numa_state(dev->id, NUMA_HIT),
495 			  sum_zone_numa_state(dev->id, NUMA_MISS),
496 			  sum_zone_numa_state(dev->id, NUMA_FOREIGN),
497 			  sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
498 			  sum_zone_numa_state(dev->id, NUMA_LOCAL),
499 			  sum_zone_numa_state(dev->id, NUMA_OTHER));
500 }
501 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
502 
503 static ssize_t node_read_vmstat(struct device *dev,
504 				struct device_attribute *attr, char *buf)
505 {
506 	int nid = dev->id;
507 	struct pglist_data *pgdat = NODE_DATA(nid);
508 	int i;
509 	int len = 0;
510 
511 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
512 		len += sysfs_emit_at(buf, len, "%s %lu\n",
513 				     zone_stat_name(i),
514 				     sum_zone_node_page_state(nid, i));
515 
516 #ifdef CONFIG_NUMA
517 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
518 		len += sysfs_emit_at(buf, len, "%s %lu\n",
519 				     numa_stat_name(i),
520 				     sum_zone_numa_state(nid, i));
521 
522 #endif
523 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
524 		unsigned long pages = node_page_state_pages(pgdat, i);
525 
526 		if (vmstat_item_print_in_thp(i))
527 			pages /= HPAGE_PMD_NR;
528 		len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
529 				     pages);
530 	}
531 
532 	return len;
533 }
534 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
535 
536 static ssize_t node_read_distance(struct device *dev,
537 				  struct device_attribute *attr, char *buf)
538 {
539 	int nid = dev->id;
540 	int len = 0;
541 	int i;
542 
543 	/*
544 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
545 	 * at the most (distance + space or newline).
546 	 */
547 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
548 
549 	for_each_online_node(i) {
550 		len += sysfs_emit_at(buf, len, "%s%d",
551 				     i ? " " : "", node_distance(nid, i));
552 	}
553 
554 	len += sysfs_emit_at(buf, len, "\n");
555 	return len;
556 }
557 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
558 
559 static struct attribute *node_dev_attrs[] = {
560 	&dev_attr_cpumap.attr,
561 	&dev_attr_cpulist.attr,
562 	&dev_attr_meminfo.attr,
563 	&dev_attr_numastat.attr,
564 	&dev_attr_distance.attr,
565 	&dev_attr_vmstat.attr,
566 	NULL
567 };
568 ATTRIBUTE_GROUPS(node_dev);
569 
570 #ifdef CONFIG_HUGETLBFS
571 /*
572  * hugetlbfs per node attributes registration interface:
573  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
574  * it will register its per node attributes for all online nodes with
575  * memory.  It will also call register_hugetlbfs_with_node(), below, to
576  * register its attribute registration functions with this node driver.
577  * Once these hooks have been initialized, the node driver will call into
578  * the hugetlb module to [un]register attributes for hot-plugged nodes.
579  */
580 static node_registration_func_t __hugetlb_register_node;
581 static node_registration_func_t __hugetlb_unregister_node;
582 
583 static inline bool hugetlb_register_node(struct node *node)
584 {
585 	if (__hugetlb_register_node &&
586 			node_state(node->dev.id, N_MEMORY)) {
587 		__hugetlb_register_node(node);
588 		return true;
589 	}
590 	return false;
591 }
592 
593 static inline void hugetlb_unregister_node(struct node *node)
594 {
595 	if (__hugetlb_unregister_node)
596 		__hugetlb_unregister_node(node);
597 }
598 
599 void register_hugetlbfs_with_node(node_registration_func_t doregister,
600 				  node_registration_func_t unregister)
601 {
602 	__hugetlb_register_node   = doregister;
603 	__hugetlb_unregister_node = unregister;
604 }
605 #else
606 static inline void hugetlb_register_node(struct node *node) {}
607 
608 static inline void hugetlb_unregister_node(struct node *node) {}
609 #endif
610 
611 static void node_device_release(struct device *dev)
612 {
613 	struct node *node = to_node(dev);
614 
615 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
616 	/*
617 	 * We schedule the work only when a memory section is
618 	 * onlined/offlined on this node. When we come here,
619 	 * all the memory on this node has been offlined,
620 	 * so we won't enqueue new work to this work.
621 	 *
622 	 * The work is using node->node_work, so we should
623 	 * flush work before freeing the memory.
624 	 */
625 	flush_work(&node->node_work);
626 #endif
627 	kfree(node);
628 }
629 
630 /*
631  * register_node - Setup a sysfs device for a node.
632  * @num - Node number to use when creating the device.
633  *
634  * Initialize and register the node device.
635  */
636 static int register_node(struct node *node, int num)
637 {
638 	int error;
639 
640 	node->dev.id = num;
641 	node->dev.bus = &node_subsys;
642 	node->dev.release = node_device_release;
643 	node->dev.groups = node_dev_groups;
644 	error = device_register(&node->dev);
645 
646 	if (error)
647 		put_device(&node->dev);
648 	else {
649 		hugetlb_register_node(node);
650 
651 		compaction_register_node(node);
652 	}
653 	return error;
654 }
655 
656 /**
657  * unregister_node - unregister a node device
658  * @node: node going away
659  *
660  * Unregisters a node device @node.  All the devices on the node must be
661  * unregistered before calling this function.
662  */
663 void unregister_node(struct node *node)
664 {
665 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
666 	node_remove_accesses(node);
667 	node_remove_caches(node);
668 	device_unregister(&node->dev);
669 }
670 
671 struct node *node_devices[MAX_NUMNODES];
672 
673 /*
674  * register cpu under node
675  */
676 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
677 {
678 	int ret;
679 	struct device *obj;
680 
681 	if (!node_online(nid))
682 		return 0;
683 
684 	obj = get_cpu_device(cpu);
685 	if (!obj)
686 		return 0;
687 
688 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
689 				&obj->kobj,
690 				kobject_name(&obj->kobj));
691 	if (ret)
692 		return ret;
693 
694 	return sysfs_create_link(&obj->kobj,
695 				 &node_devices[nid]->dev.kobj,
696 				 kobject_name(&node_devices[nid]->dev.kobj));
697 }
698 
699 /**
700  * register_memory_node_under_compute_node - link memory node to its compute
701  *					     node for a given access class.
702  * @mem_nid:	Memory node number
703  * @cpu_nid:	Cpu  node number
704  * @access:	Access class to register
705  *
706  * Description:
707  * 	For use with platforms that may have separate memory and compute nodes.
708  * 	This function will export node relationships linking which memory
709  * 	initiator nodes can access memory targets at a given ranked access
710  * 	class.
711  */
712 int register_memory_node_under_compute_node(unsigned int mem_nid,
713 					    unsigned int cpu_nid,
714 					    unsigned access)
715 {
716 	struct node *init_node, *targ_node;
717 	struct node_access_nodes *initiator, *target;
718 	int ret;
719 
720 	if (!node_online(cpu_nid) || !node_online(mem_nid))
721 		return -ENODEV;
722 
723 	init_node = node_devices[cpu_nid];
724 	targ_node = node_devices[mem_nid];
725 	initiator = node_init_node_access(init_node, access);
726 	target = node_init_node_access(targ_node, access);
727 	if (!initiator || !target)
728 		return -ENOMEM;
729 
730 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
731 				      &targ_node->dev.kobj,
732 				      dev_name(&targ_node->dev));
733 	if (ret)
734 		return ret;
735 
736 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
737 				      &init_node->dev.kobj,
738 				      dev_name(&init_node->dev));
739 	if (ret)
740 		goto err;
741 
742 	return 0;
743  err:
744 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
745 				     dev_name(&targ_node->dev));
746 	return ret;
747 }
748 
749 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
750 {
751 	struct device *obj;
752 
753 	if (!node_online(nid))
754 		return 0;
755 
756 	obj = get_cpu_device(cpu);
757 	if (!obj)
758 		return 0;
759 
760 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
761 			  kobject_name(&obj->kobj));
762 	sysfs_remove_link(&obj->kobj,
763 			  kobject_name(&node_devices[nid]->dev.kobj));
764 
765 	return 0;
766 }
767 
768 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
769 static int __ref get_nid_for_pfn(unsigned long pfn)
770 {
771 	if (!pfn_valid_within(pfn))
772 		return -1;
773 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
774 	if (system_state < SYSTEM_RUNNING)
775 		return early_pfn_to_nid(pfn);
776 #endif
777 	return pfn_to_nid(pfn);
778 }
779 
780 static void do_register_memory_block_under_node(int nid,
781 						struct memory_block *mem_blk)
782 {
783 	int ret;
784 
785 	/*
786 	 * If this memory block spans multiple nodes, we only indicate
787 	 * the last processed node.
788 	 */
789 	mem_blk->nid = nid;
790 
791 	ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
792 				       &mem_blk->dev.kobj,
793 				       kobject_name(&mem_blk->dev.kobj));
794 	if (ret && ret != -EEXIST)
795 		dev_err_ratelimited(&node_devices[nid]->dev,
796 				    "can't create link to %s in sysfs (%d)\n",
797 				    kobject_name(&mem_blk->dev.kobj), ret);
798 
799 	ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
800 				&node_devices[nid]->dev.kobj,
801 				kobject_name(&node_devices[nid]->dev.kobj));
802 	if (ret && ret != -EEXIST)
803 		dev_err_ratelimited(&mem_blk->dev,
804 				    "can't create link to %s in sysfs (%d)\n",
805 				    kobject_name(&node_devices[nid]->dev.kobj),
806 				    ret);
807 }
808 
809 /* register memory section under specified node if it spans that node */
810 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
811 					       void *arg)
812 {
813 	unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
814 	unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
815 	unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
816 	int nid = *(int *)arg;
817 	unsigned long pfn;
818 
819 	for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
820 		int page_nid;
821 
822 		/*
823 		 * memory block could have several absent sections from start.
824 		 * skip pfn range from absent section
825 		 */
826 		if (!pfn_in_present_section(pfn)) {
827 			pfn = round_down(pfn + PAGES_PER_SECTION,
828 					 PAGES_PER_SECTION) - 1;
829 			continue;
830 		}
831 
832 		/*
833 		 * We need to check if page belongs to nid only at the boot
834 		 * case because node's ranges can be interleaved.
835 		 */
836 		page_nid = get_nid_for_pfn(pfn);
837 		if (page_nid < 0)
838 			continue;
839 		if (page_nid != nid)
840 			continue;
841 
842 		do_register_memory_block_under_node(nid, mem_blk);
843 		return 0;
844 	}
845 	/* mem section does not span the specified node */
846 	return 0;
847 }
848 
849 /*
850  * During hotplug we know that all pages in the memory block belong to the same
851  * node.
852  */
853 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
854 						 void *arg)
855 {
856 	int nid = *(int *)arg;
857 
858 	do_register_memory_block_under_node(nid, mem_blk);
859 	return 0;
860 }
861 
862 /*
863  * Unregister a memory block device under the node it spans. Memory blocks
864  * with multiple nodes cannot be offlined and therefore also never be removed.
865  */
866 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
867 {
868 	if (mem_blk->nid == NUMA_NO_NODE)
869 		return;
870 
871 	sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
872 			  kobject_name(&mem_blk->dev.kobj));
873 	sysfs_remove_link(&mem_blk->dev.kobj,
874 			  kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
875 }
876 
877 void link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn,
878 		       enum meminit_context context)
879 {
880 	walk_memory_blocks_func_t func;
881 
882 	if (context == MEMINIT_HOTPLUG)
883 		func = register_mem_block_under_node_hotplug;
884 	else
885 		func = register_mem_block_under_node_early;
886 
887 	walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
888 			   (void *)&nid, func);
889 	return;
890 }
891 
892 #ifdef CONFIG_HUGETLBFS
893 /*
894  * Handle per node hstate attribute [un]registration on transistions
895  * to/from memoryless state.
896  */
897 static void node_hugetlb_work(struct work_struct *work)
898 {
899 	struct node *node = container_of(work, struct node, node_work);
900 
901 	/*
902 	 * We only get here when a node transitions to/from memoryless state.
903 	 * We can detect which transition occurred by examining whether the
904 	 * node has memory now.  hugetlb_register_node() already check this
905 	 * so we try to register the attributes.  If that fails, then the
906 	 * node has transitioned to memoryless, try to unregister the
907 	 * attributes.
908 	 */
909 	if (!hugetlb_register_node(node))
910 		hugetlb_unregister_node(node);
911 }
912 
913 static void init_node_hugetlb_work(int nid)
914 {
915 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
916 }
917 
918 static int node_memory_callback(struct notifier_block *self,
919 				unsigned long action, void *arg)
920 {
921 	struct memory_notify *mnb = arg;
922 	int nid = mnb->status_change_nid;
923 
924 	switch (action) {
925 	case MEM_ONLINE:
926 	case MEM_OFFLINE:
927 		/*
928 		 * offload per node hstate [un]registration to a work thread
929 		 * when transitioning to/from memoryless state.
930 		 */
931 		if (nid != NUMA_NO_NODE)
932 			schedule_work(&node_devices[nid]->node_work);
933 		break;
934 
935 	case MEM_GOING_ONLINE:
936 	case MEM_GOING_OFFLINE:
937 	case MEM_CANCEL_ONLINE:
938 	case MEM_CANCEL_OFFLINE:
939 	default:
940 		break;
941 	}
942 
943 	return NOTIFY_OK;
944 }
945 #endif	/* CONFIG_HUGETLBFS */
946 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
947 
948 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
949     !defined(CONFIG_HUGETLBFS)
950 static inline int node_memory_callback(struct notifier_block *self,
951 				unsigned long action, void *arg)
952 {
953 	return NOTIFY_OK;
954 }
955 
956 static void init_node_hugetlb_work(int nid) { }
957 
958 #endif
959 
960 int __register_one_node(int nid)
961 {
962 	int error;
963 	int cpu;
964 
965 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
966 	if (!node_devices[nid])
967 		return -ENOMEM;
968 
969 	error = register_node(node_devices[nid], nid);
970 
971 	/* link cpu under this node */
972 	for_each_present_cpu(cpu) {
973 		if (cpu_to_node(cpu) == nid)
974 			register_cpu_under_node(cpu, nid);
975 	}
976 
977 	INIT_LIST_HEAD(&node_devices[nid]->access_list);
978 	/* initialize work queue for memory hot plug */
979 	init_node_hugetlb_work(nid);
980 	node_init_caches(nid);
981 
982 	return error;
983 }
984 
985 void unregister_one_node(int nid)
986 {
987 	if (!node_devices[nid])
988 		return;
989 
990 	unregister_node(node_devices[nid]);
991 	node_devices[nid] = NULL;
992 }
993 
994 /*
995  * node states attributes
996  */
997 
998 struct node_attr {
999 	struct device_attribute attr;
1000 	enum node_states state;
1001 };
1002 
1003 static ssize_t show_node_state(struct device *dev,
1004 			       struct device_attribute *attr, char *buf)
1005 {
1006 	struct node_attr *na = container_of(attr, struct node_attr, attr);
1007 
1008 	return sysfs_emit(buf, "%*pbl\n",
1009 			  nodemask_pr_args(&node_states[na->state]));
1010 }
1011 
1012 #define _NODE_ATTR(name, state) \
1013 	{ __ATTR(name, 0444, show_node_state, NULL), state }
1014 
1015 static struct node_attr node_state_attr[] = {
1016 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1017 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1018 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1019 #ifdef CONFIG_HIGHMEM
1020 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1021 #endif
1022 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1023 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1024 	[N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
1025 					   N_GENERIC_INITIATOR),
1026 };
1027 
1028 static struct attribute *node_state_attrs[] = {
1029 	&node_state_attr[N_POSSIBLE].attr.attr,
1030 	&node_state_attr[N_ONLINE].attr.attr,
1031 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
1032 #ifdef CONFIG_HIGHMEM
1033 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
1034 #endif
1035 	&node_state_attr[N_MEMORY].attr.attr,
1036 	&node_state_attr[N_CPU].attr.attr,
1037 	&node_state_attr[N_GENERIC_INITIATOR].attr.attr,
1038 	NULL
1039 };
1040 
1041 static struct attribute_group memory_root_attr_group = {
1042 	.attrs = node_state_attrs,
1043 };
1044 
1045 static const struct attribute_group *cpu_root_attr_groups[] = {
1046 	&memory_root_attr_group,
1047 	NULL,
1048 };
1049 
1050 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
1051 static int __init register_node_type(void)
1052 {
1053 	int ret;
1054 
1055  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1056  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1057 
1058 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1059 	if (!ret) {
1060 		static struct notifier_block node_memory_callback_nb = {
1061 			.notifier_call = node_memory_callback,
1062 			.priority = NODE_CALLBACK_PRI,
1063 		};
1064 		register_hotmemory_notifier(&node_memory_callback_nb);
1065 	}
1066 
1067 	/*
1068 	 * Note:  we're not going to unregister the node class if we fail
1069 	 * to register the node state class attribute files.
1070 	 */
1071 	return ret;
1072 }
1073 postcore_initcall(register_node_type);
1074