xref: /linux/drivers/base/node.c (revision 1553a1c48281243359a9529a10ddb551f3b967ab)
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 const struct bus_type node_subsys = {
25 	.name = "node",
26 	.dev_name = "node",
27 };
28 
29 static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
30 				  struct bin_attribute *attr, char *buf,
31 				  loff_t off, size_t count)
32 {
33 	struct device *dev = kobj_to_dev(kobj);
34 	struct node *node_dev = to_node(dev);
35 	cpumask_var_t mask;
36 	ssize_t n;
37 
38 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
39 		return 0;
40 
41 	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
42 	n = cpumap_print_bitmask_to_buf(buf, mask, off, count);
43 	free_cpumask_var(mask);
44 
45 	return n;
46 }
47 
48 static BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
49 
50 static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
51 				   struct bin_attribute *attr, char *buf,
52 				   loff_t off, size_t count)
53 {
54 	struct device *dev = kobj_to_dev(kobj);
55 	struct node *node_dev = to_node(dev);
56 	cpumask_var_t mask;
57 	ssize_t n;
58 
59 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
60 		return 0;
61 
62 	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
63 	n = cpumap_print_list_to_buf(buf, mask, off, count);
64 	free_cpumask_var(mask);
65 
66 	return n;
67 }
68 
69 static BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
70 
71 /**
72  * struct node_access_nodes - Access class device to hold user visible
73  * 			      relationships to other nodes.
74  * @dev:	Device for this memory access class
75  * @list_node:	List element in the node's access list
76  * @access:	The access class rank
77  * @coord:	Heterogeneous memory performance coordinates
78  */
79 struct node_access_nodes {
80 	struct device		dev;
81 	struct list_head	list_node;
82 	unsigned int		access;
83 #ifdef CONFIG_HMEM_REPORTING
84 	struct access_coordinate	coord;
85 #endif
86 };
87 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
88 
89 static struct attribute *node_init_access_node_attrs[] = {
90 	NULL,
91 };
92 
93 static struct attribute *node_targ_access_node_attrs[] = {
94 	NULL,
95 };
96 
97 static const struct attribute_group initiators = {
98 	.name	= "initiators",
99 	.attrs	= node_init_access_node_attrs,
100 };
101 
102 static const struct attribute_group targets = {
103 	.name	= "targets",
104 	.attrs	= node_targ_access_node_attrs,
105 };
106 
107 static const struct attribute_group *node_access_node_groups[] = {
108 	&initiators,
109 	&targets,
110 	NULL,
111 };
112 
113 static void node_remove_accesses(struct node *node)
114 {
115 	struct node_access_nodes *c, *cnext;
116 
117 	list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
118 		list_del(&c->list_node);
119 		device_unregister(&c->dev);
120 	}
121 }
122 
123 static void node_access_release(struct device *dev)
124 {
125 	kfree(to_access_nodes(dev));
126 }
127 
128 static struct node_access_nodes *node_init_node_access(struct node *node,
129 						       enum access_coordinate_class access)
130 {
131 	struct node_access_nodes *access_node;
132 	struct device *dev;
133 
134 	list_for_each_entry(access_node, &node->access_list, list_node)
135 		if (access_node->access == access)
136 			return access_node;
137 
138 	access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
139 	if (!access_node)
140 		return NULL;
141 
142 	access_node->access = access;
143 	dev = &access_node->dev;
144 	dev->parent = &node->dev;
145 	dev->release = node_access_release;
146 	dev->groups = node_access_node_groups;
147 	if (dev_set_name(dev, "access%u", access))
148 		goto free;
149 
150 	if (device_register(dev))
151 		goto free_name;
152 
153 	pm_runtime_no_callbacks(dev);
154 	list_add_tail(&access_node->list_node, &node->access_list);
155 	return access_node;
156 free_name:
157 	kfree_const(dev->kobj.name);
158 free:
159 	kfree(access_node);
160 	return NULL;
161 }
162 
163 #ifdef CONFIG_HMEM_REPORTING
164 #define ACCESS_ATTR(property)						\
165 static ssize_t property##_show(struct device *dev,			\
166 			   struct device_attribute *attr,		\
167 			   char *buf)					\
168 {									\
169 	return sysfs_emit(buf, "%u\n",					\
170 			  to_access_nodes(dev)->coord.property);	\
171 }									\
172 static DEVICE_ATTR_RO(property)
173 
174 ACCESS_ATTR(read_bandwidth);
175 ACCESS_ATTR(read_latency);
176 ACCESS_ATTR(write_bandwidth);
177 ACCESS_ATTR(write_latency);
178 
179 static struct attribute *access_attrs[] = {
180 	&dev_attr_read_bandwidth.attr,
181 	&dev_attr_read_latency.attr,
182 	&dev_attr_write_bandwidth.attr,
183 	&dev_attr_write_latency.attr,
184 	NULL,
185 };
186 
187 /**
188  * node_set_perf_attrs - Set the performance values for given access class
189  * @nid: Node identifier to be set
190  * @coord: Heterogeneous memory performance coordinates
191  * @access: The access class the for the given attributes
192  */
193 void node_set_perf_attrs(unsigned int nid, struct access_coordinate *coord,
194 			 enum access_coordinate_class access)
195 {
196 	struct node_access_nodes *c;
197 	struct node *node;
198 	int i;
199 
200 	if (WARN_ON_ONCE(!node_online(nid)))
201 		return;
202 
203 	node = node_devices[nid];
204 	c = node_init_node_access(node, access);
205 	if (!c)
206 		return;
207 
208 	c->coord = *coord;
209 	for (i = 0; access_attrs[i] != NULL; i++) {
210 		if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
211 					    "initiators")) {
212 			pr_info("failed to add performance attribute to node %d\n",
213 				nid);
214 			break;
215 		}
216 	}
217 }
218 EXPORT_SYMBOL_GPL(node_set_perf_attrs);
219 
220 /**
221  * struct node_cache_info - Internal tracking for memory node caches
222  * @dev:	Device represeting the cache level
223  * @node:	List element for tracking in the node
224  * @cache_attrs:Attributes for this cache level
225  */
226 struct node_cache_info {
227 	struct device dev;
228 	struct list_head node;
229 	struct node_cache_attrs cache_attrs;
230 };
231 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
232 
233 #define CACHE_ATTR(name, fmt) 						\
234 static ssize_t name##_show(struct device *dev,				\
235 			   struct device_attribute *attr,		\
236 			   char *buf)					\
237 {									\
238 	return sysfs_emit(buf, fmt "\n",				\
239 			  to_cache_info(dev)->cache_attrs.name);	\
240 }									\
241 static DEVICE_ATTR_RO(name);
242 
243 CACHE_ATTR(size, "%llu")
244 CACHE_ATTR(line_size, "%u")
245 CACHE_ATTR(indexing, "%u")
246 CACHE_ATTR(write_policy, "%u")
247 
248 static struct attribute *cache_attrs[] = {
249 	&dev_attr_indexing.attr,
250 	&dev_attr_size.attr,
251 	&dev_attr_line_size.attr,
252 	&dev_attr_write_policy.attr,
253 	NULL,
254 };
255 ATTRIBUTE_GROUPS(cache);
256 
257 static void node_cache_release(struct device *dev)
258 {
259 	kfree(dev);
260 }
261 
262 static void node_cacheinfo_release(struct device *dev)
263 {
264 	struct node_cache_info *info = to_cache_info(dev);
265 	kfree(info);
266 }
267 
268 static void node_init_cache_dev(struct node *node)
269 {
270 	struct device *dev;
271 
272 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
273 	if (!dev)
274 		return;
275 
276 	device_initialize(dev);
277 	dev->parent = &node->dev;
278 	dev->release = node_cache_release;
279 	if (dev_set_name(dev, "memory_side_cache"))
280 		goto put_device;
281 
282 	if (device_add(dev))
283 		goto put_device;
284 
285 	pm_runtime_no_callbacks(dev);
286 	node->cache_dev = dev;
287 	return;
288 put_device:
289 	put_device(dev);
290 }
291 
292 /**
293  * node_add_cache() - add cache attribute to a memory node
294  * @nid: Node identifier that has new cache attributes
295  * @cache_attrs: Attributes for the cache being added
296  */
297 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
298 {
299 	struct node_cache_info *info;
300 	struct device *dev;
301 	struct node *node;
302 
303 	if (!node_online(nid) || !node_devices[nid])
304 		return;
305 
306 	node = node_devices[nid];
307 	list_for_each_entry(info, &node->cache_attrs, node) {
308 		if (info->cache_attrs.level == cache_attrs->level) {
309 			dev_warn(&node->dev,
310 				"attempt to add duplicate cache level:%d\n",
311 				cache_attrs->level);
312 			return;
313 		}
314 	}
315 
316 	if (!node->cache_dev)
317 		node_init_cache_dev(node);
318 	if (!node->cache_dev)
319 		return;
320 
321 	info = kzalloc(sizeof(*info), GFP_KERNEL);
322 	if (!info)
323 		return;
324 
325 	dev = &info->dev;
326 	device_initialize(dev);
327 	dev->parent = node->cache_dev;
328 	dev->release = node_cacheinfo_release;
329 	dev->groups = cache_groups;
330 	if (dev_set_name(dev, "index%d", cache_attrs->level))
331 		goto put_device;
332 
333 	info->cache_attrs = *cache_attrs;
334 	if (device_add(dev)) {
335 		dev_warn(&node->dev, "failed to add cache level:%d\n",
336 			 cache_attrs->level);
337 		goto put_device;
338 	}
339 	pm_runtime_no_callbacks(dev);
340 	list_add_tail(&info->node, &node->cache_attrs);
341 	return;
342 put_device:
343 	put_device(dev);
344 }
345 
346 static void node_remove_caches(struct node *node)
347 {
348 	struct node_cache_info *info, *next;
349 
350 	if (!node->cache_dev)
351 		return;
352 
353 	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
354 		list_del(&info->node);
355 		device_unregister(&info->dev);
356 	}
357 	device_unregister(node->cache_dev);
358 }
359 
360 static void node_init_caches(unsigned int nid)
361 {
362 	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
363 }
364 #else
365 static void node_init_caches(unsigned int nid) { }
366 static void node_remove_caches(struct node *node) { }
367 #endif
368 
369 #define K(x) ((x) << (PAGE_SHIFT - 10))
370 static ssize_t node_read_meminfo(struct device *dev,
371 			struct device_attribute *attr, char *buf)
372 {
373 	int len = 0;
374 	int nid = dev->id;
375 	struct pglist_data *pgdat = NODE_DATA(nid);
376 	struct sysinfo i;
377 	unsigned long sreclaimable, sunreclaimable;
378 	unsigned long swapcached = 0;
379 
380 	si_meminfo_node(&i, nid);
381 	sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
382 	sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
383 #ifdef CONFIG_SWAP
384 	swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
385 #endif
386 	len = sysfs_emit_at(buf, len,
387 			    "Node %d MemTotal:       %8lu kB\n"
388 			    "Node %d MemFree:        %8lu kB\n"
389 			    "Node %d MemUsed:        %8lu kB\n"
390 			    "Node %d SwapCached:     %8lu kB\n"
391 			    "Node %d Active:         %8lu kB\n"
392 			    "Node %d Inactive:       %8lu kB\n"
393 			    "Node %d Active(anon):   %8lu kB\n"
394 			    "Node %d Inactive(anon): %8lu kB\n"
395 			    "Node %d Active(file):   %8lu kB\n"
396 			    "Node %d Inactive(file): %8lu kB\n"
397 			    "Node %d Unevictable:    %8lu kB\n"
398 			    "Node %d Mlocked:        %8lu kB\n",
399 			    nid, K(i.totalram),
400 			    nid, K(i.freeram),
401 			    nid, K(i.totalram - i.freeram),
402 			    nid, K(swapcached),
403 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
404 				   node_page_state(pgdat, NR_ACTIVE_FILE)),
405 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
406 				   node_page_state(pgdat, NR_INACTIVE_FILE)),
407 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
408 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
409 			    nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
410 			    nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
411 			    nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
412 			    nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
413 
414 #ifdef CONFIG_HIGHMEM
415 	len += sysfs_emit_at(buf, len,
416 			     "Node %d HighTotal:      %8lu kB\n"
417 			     "Node %d HighFree:       %8lu kB\n"
418 			     "Node %d LowTotal:       %8lu kB\n"
419 			     "Node %d LowFree:        %8lu kB\n",
420 			     nid, K(i.totalhigh),
421 			     nid, K(i.freehigh),
422 			     nid, K(i.totalram - i.totalhigh),
423 			     nid, K(i.freeram - i.freehigh));
424 #endif
425 	len += sysfs_emit_at(buf, len,
426 			     "Node %d Dirty:          %8lu kB\n"
427 			     "Node %d Writeback:      %8lu kB\n"
428 			     "Node %d FilePages:      %8lu kB\n"
429 			     "Node %d Mapped:         %8lu kB\n"
430 			     "Node %d AnonPages:      %8lu kB\n"
431 			     "Node %d Shmem:          %8lu kB\n"
432 			     "Node %d KernelStack:    %8lu kB\n"
433 #ifdef CONFIG_SHADOW_CALL_STACK
434 			     "Node %d ShadowCallStack:%8lu kB\n"
435 #endif
436 			     "Node %d PageTables:     %8lu kB\n"
437 			     "Node %d SecPageTables:  %8lu kB\n"
438 			     "Node %d NFS_Unstable:   %8lu kB\n"
439 			     "Node %d Bounce:         %8lu kB\n"
440 			     "Node %d WritebackTmp:   %8lu kB\n"
441 			     "Node %d KReclaimable:   %8lu kB\n"
442 			     "Node %d Slab:           %8lu kB\n"
443 			     "Node %d SReclaimable:   %8lu kB\n"
444 			     "Node %d SUnreclaim:     %8lu kB\n"
445 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
446 			     "Node %d AnonHugePages:  %8lu kB\n"
447 			     "Node %d ShmemHugePages: %8lu kB\n"
448 			     "Node %d ShmemPmdMapped: %8lu kB\n"
449 			     "Node %d FileHugePages:  %8lu kB\n"
450 			     "Node %d FilePmdMapped:  %8lu kB\n"
451 #endif
452 #ifdef CONFIG_UNACCEPTED_MEMORY
453 			     "Node %d Unaccepted:     %8lu kB\n"
454 #endif
455 			     ,
456 			     nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
457 			     nid, K(node_page_state(pgdat, NR_WRITEBACK)),
458 			     nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
459 			     nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
460 			     nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
461 			     nid, K(i.sharedram),
462 			     nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
463 #ifdef CONFIG_SHADOW_CALL_STACK
464 			     nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
465 #endif
466 			     nid, K(node_page_state(pgdat, NR_PAGETABLE)),
467 			     nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
468 			     nid, 0UL,
469 			     nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
470 			     nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
471 			     nid, K(sreclaimable +
472 				    node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
473 			     nid, K(sreclaimable + sunreclaimable),
474 			     nid, K(sreclaimable),
475 			     nid, K(sunreclaimable)
476 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
477 			     ,
478 			     nid, K(node_page_state(pgdat, NR_ANON_THPS)),
479 			     nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
480 			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
481 			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
482 			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
483 #endif
484 #ifdef CONFIG_UNACCEPTED_MEMORY
485 			     ,
486 			     nid, K(sum_zone_node_page_state(nid, NR_UNACCEPTED))
487 #endif
488 			    );
489 	len += hugetlb_report_node_meminfo(buf, len, nid);
490 	return len;
491 }
492 
493 #undef K
494 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
495 
496 static ssize_t node_read_numastat(struct device *dev,
497 				  struct device_attribute *attr, char *buf)
498 {
499 	fold_vm_numa_events();
500 	return sysfs_emit(buf,
501 			  "numa_hit %lu\n"
502 			  "numa_miss %lu\n"
503 			  "numa_foreign %lu\n"
504 			  "interleave_hit %lu\n"
505 			  "local_node %lu\n"
506 			  "other_node %lu\n",
507 			  sum_zone_numa_event_state(dev->id, NUMA_HIT),
508 			  sum_zone_numa_event_state(dev->id, NUMA_MISS),
509 			  sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
510 			  sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
511 			  sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
512 			  sum_zone_numa_event_state(dev->id, NUMA_OTHER));
513 }
514 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
515 
516 static ssize_t node_read_vmstat(struct device *dev,
517 				struct device_attribute *attr, char *buf)
518 {
519 	int nid = dev->id;
520 	struct pglist_data *pgdat = NODE_DATA(nid);
521 	int i;
522 	int len = 0;
523 
524 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
525 		len += sysfs_emit_at(buf, len, "%s %lu\n",
526 				     zone_stat_name(i),
527 				     sum_zone_node_page_state(nid, i));
528 
529 #ifdef CONFIG_NUMA
530 	fold_vm_numa_events();
531 	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
532 		len += sysfs_emit_at(buf, len, "%s %lu\n",
533 				     numa_stat_name(i),
534 				     sum_zone_numa_event_state(nid, i));
535 
536 #endif
537 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
538 		unsigned long pages = node_page_state_pages(pgdat, i);
539 
540 		if (vmstat_item_print_in_thp(i))
541 			pages /= HPAGE_PMD_NR;
542 		len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
543 				     pages);
544 	}
545 
546 	return len;
547 }
548 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
549 
550 static ssize_t node_read_distance(struct device *dev,
551 				  struct device_attribute *attr, char *buf)
552 {
553 	int nid = dev->id;
554 	int len = 0;
555 	int i;
556 
557 	/*
558 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
559 	 * at the most (distance + space or newline).
560 	 */
561 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
562 
563 	for_each_online_node(i) {
564 		len += sysfs_emit_at(buf, len, "%s%d",
565 				     i ? " " : "", node_distance(nid, i));
566 	}
567 
568 	len += sysfs_emit_at(buf, len, "\n");
569 	return len;
570 }
571 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
572 
573 static struct attribute *node_dev_attrs[] = {
574 	&dev_attr_meminfo.attr,
575 	&dev_attr_numastat.attr,
576 	&dev_attr_distance.attr,
577 	&dev_attr_vmstat.attr,
578 	NULL
579 };
580 
581 static struct bin_attribute *node_dev_bin_attrs[] = {
582 	&bin_attr_cpumap,
583 	&bin_attr_cpulist,
584 	NULL
585 };
586 
587 static const struct attribute_group node_dev_group = {
588 	.attrs = node_dev_attrs,
589 	.bin_attrs = node_dev_bin_attrs
590 };
591 
592 static const struct attribute_group *node_dev_groups[] = {
593 	&node_dev_group,
594 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
595 	&arch_node_dev_group,
596 #endif
597 #ifdef CONFIG_MEMORY_FAILURE
598 	&memory_failure_attr_group,
599 #endif
600 	NULL
601 };
602 
603 static void node_device_release(struct device *dev)
604 {
605 	kfree(to_node(dev));
606 }
607 
608 /*
609  * register_node - Setup a sysfs device for a node.
610  * @num - Node number to use when creating the device.
611  *
612  * Initialize and register the node device.
613  */
614 static int register_node(struct node *node, int num)
615 {
616 	int error;
617 
618 	node->dev.id = num;
619 	node->dev.bus = &node_subsys;
620 	node->dev.release = node_device_release;
621 	node->dev.groups = node_dev_groups;
622 	error = device_register(&node->dev);
623 
624 	if (error) {
625 		put_device(&node->dev);
626 	} else {
627 		hugetlb_register_node(node);
628 		compaction_register_node(node);
629 	}
630 
631 	return error;
632 }
633 
634 /**
635  * unregister_node - unregister a node device
636  * @node: node going away
637  *
638  * Unregisters a node device @node.  All the devices on the node must be
639  * unregistered before calling this function.
640  */
641 void unregister_node(struct node *node)
642 {
643 	hugetlb_unregister_node(node);
644 	compaction_unregister_node(node);
645 	node_remove_accesses(node);
646 	node_remove_caches(node);
647 	device_unregister(&node->dev);
648 }
649 
650 struct node *node_devices[MAX_NUMNODES];
651 
652 /*
653  * register cpu under node
654  */
655 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
656 {
657 	int ret;
658 	struct device *obj;
659 
660 	if (!node_online(nid))
661 		return 0;
662 
663 	obj = get_cpu_device(cpu);
664 	if (!obj)
665 		return 0;
666 
667 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
668 				&obj->kobj,
669 				kobject_name(&obj->kobj));
670 	if (ret)
671 		return ret;
672 
673 	return sysfs_create_link(&obj->kobj,
674 				 &node_devices[nid]->dev.kobj,
675 				 kobject_name(&node_devices[nid]->dev.kobj));
676 }
677 
678 /**
679  * register_memory_node_under_compute_node - link memory node to its compute
680  *					     node for a given access class.
681  * @mem_nid:	Memory node number
682  * @cpu_nid:	Cpu  node number
683  * @access:	Access class to register
684  *
685  * Description:
686  * 	For use with platforms that may have separate memory and compute nodes.
687  * 	This function will export node relationships linking which memory
688  * 	initiator nodes can access memory targets at a given ranked access
689  * 	class.
690  */
691 int register_memory_node_under_compute_node(unsigned int mem_nid,
692 					    unsigned int cpu_nid,
693 					    enum access_coordinate_class access)
694 {
695 	struct node *init_node, *targ_node;
696 	struct node_access_nodes *initiator, *target;
697 	int ret;
698 
699 	if (!node_online(cpu_nid) || !node_online(mem_nid))
700 		return -ENODEV;
701 
702 	init_node = node_devices[cpu_nid];
703 	targ_node = node_devices[mem_nid];
704 	initiator = node_init_node_access(init_node, access);
705 	target = node_init_node_access(targ_node, access);
706 	if (!initiator || !target)
707 		return -ENOMEM;
708 
709 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
710 				      &targ_node->dev.kobj,
711 				      dev_name(&targ_node->dev));
712 	if (ret)
713 		return ret;
714 
715 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
716 				      &init_node->dev.kobj,
717 				      dev_name(&init_node->dev));
718 	if (ret)
719 		goto err;
720 
721 	return 0;
722  err:
723 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
724 				     dev_name(&targ_node->dev));
725 	return ret;
726 }
727 
728 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
729 {
730 	struct device *obj;
731 
732 	if (!node_online(nid))
733 		return 0;
734 
735 	obj = get_cpu_device(cpu);
736 	if (!obj)
737 		return 0;
738 
739 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
740 			  kobject_name(&obj->kobj));
741 	sysfs_remove_link(&obj->kobj,
742 			  kobject_name(&node_devices[nid]->dev.kobj));
743 
744 	return 0;
745 }
746 
747 #ifdef CONFIG_MEMORY_HOTPLUG
748 static int __ref get_nid_for_pfn(unsigned long pfn)
749 {
750 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
751 	if (system_state < SYSTEM_RUNNING)
752 		return early_pfn_to_nid(pfn);
753 #endif
754 	return pfn_to_nid(pfn);
755 }
756 
757 static void do_register_memory_block_under_node(int nid,
758 						struct memory_block *mem_blk,
759 						enum meminit_context context)
760 {
761 	int ret;
762 
763 	memory_block_add_nid(mem_blk, nid, context);
764 
765 	ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
766 				       &mem_blk->dev.kobj,
767 				       kobject_name(&mem_blk->dev.kobj));
768 	if (ret && ret != -EEXIST)
769 		dev_err_ratelimited(&node_devices[nid]->dev,
770 				    "can't create link to %s in sysfs (%d)\n",
771 				    kobject_name(&mem_blk->dev.kobj), ret);
772 
773 	ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
774 				&node_devices[nid]->dev.kobj,
775 				kobject_name(&node_devices[nid]->dev.kobj));
776 	if (ret && ret != -EEXIST)
777 		dev_err_ratelimited(&mem_blk->dev,
778 				    "can't create link to %s in sysfs (%d)\n",
779 				    kobject_name(&node_devices[nid]->dev.kobj),
780 				    ret);
781 }
782 
783 /* register memory section under specified node if it spans that node */
784 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
785 					       void *arg)
786 {
787 	unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
788 	unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
789 	unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
790 	int nid = *(int *)arg;
791 	unsigned long pfn;
792 
793 	for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
794 		int page_nid;
795 
796 		/*
797 		 * memory block could have several absent sections from start.
798 		 * skip pfn range from absent section
799 		 */
800 		if (!pfn_in_present_section(pfn)) {
801 			pfn = round_down(pfn + PAGES_PER_SECTION,
802 					 PAGES_PER_SECTION) - 1;
803 			continue;
804 		}
805 
806 		/*
807 		 * We need to check if page belongs to nid only at the boot
808 		 * case because node's ranges can be interleaved.
809 		 */
810 		page_nid = get_nid_for_pfn(pfn);
811 		if (page_nid < 0)
812 			continue;
813 		if (page_nid != nid)
814 			continue;
815 
816 		do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
817 		return 0;
818 	}
819 	/* mem section does not span the specified node */
820 	return 0;
821 }
822 
823 /*
824  * During hotplug we know that all pages in the memory block belong to the same
825  * node.
826  */
827 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
828 						 void *arg)
829 {
830 	int nid = *(int *)arg;
831 
832 	do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
833 	return 0;
834 }
835 
836 /*
837  * Unregister a memory block device under the node it spans. Memory blocks
838  * with multiple nodes cannot be offlined and therefore also never be removed.
839  */
840 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
841 {
842 	if (mem_blk->nid == NUMA_NO_NODE)
843 		return;
844 
845 	sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
846 			  kobject_name(&mem_blk->dev.kobj));
847 	sysfs_remove_link(&mem_blk->dev.kobj,
848 			  kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
849 }
850 
851 void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
852 				       unsigned long end_pfn,
853 				       enum meminit_context context)
854 {
855 	walk_memory_blocks_func_t func;
856 
857 	if (context == MEMINIT_HOTPLUG)
858 		func = register_mem_block_under_node_hotplug;
859 	else
860 		func = register_mem_block_under_node_early;
861 
862 	walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
863 			   (void *)&nid, func);
864 	return;
865 }
866 #endif /* CONFIG_MEMORY_HOTPLUG */
867 
868 int __register_one_node(int nid)
869 {
870 	int error;
871 	int cpu;
872 	struct node *node;
873 
874 	node = kzalloc(sizeof(struct node), GFP_KERNEL);
875 	if (!node)
876 		return -ENOMEM;
877 
878 	INIT_LIST_HEAD(&node->access_list);
879 	node_devices[nid] = node;
880 
881 	error = register_node(node_devices[nid], nid);
882 
883 	/* link cpu under this node */
884 	for_each_present_cpu(cpu) {
885 		if (cpu_to_node(cpu) == nid)
886 			register_cpu_under_node(cpu, nid);
887 	}
888 
889 	node_init_caches(nid);
890 
891 	return error;
892 }
893 
894 void unregister_one_node(int nid)
895 {
896 	if (!node_devices[nid])
897 		return;
898 
899 	unregister_node(node_devices[nid]);
900 	node_devices[nid] = NULL;
901 }
902 
903 /*
904  * node states attributes
905  */
906 
907 struct node_attr {
908 	struct device_attribute attr;
909 	enum node_states state;
910 };
911 
912 static ssize_t show_node_state(struct device *dev,
913 			       struct device_attribute *attr, char *buf)
914 {
915 	struct node_attr *na = container_of(attr, struct node_attr, attr);
916 
917 	return sysfs_emit(buf, "%*pbl\n",
918 			  nodemask_pr_args(&node_states[na->state]));
919 }
920 
921 #define _NODE_ATTR(name, state) \
922 	{ __ATTR(name, 0444, show_node_state, NULL), state }
923 
924 static struct node_attr node_state_attr[] = {
925 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
926 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
927 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
928 #ifdef CONFIG_HIGHMEM
929 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
930 #endif
931 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
932 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
933 	[N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
934 					   N_GENERIC_INITIATOR),
935 };
936 
937 static struct attribute *node_state_attrs[] = {
938 	&node_state_attr[N_POSSIBLE].attr.attr,
939 	&node_state_attr[N_ONLINE].attr.attr,
940 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
941 #ifdef CONFIG_HIGHMEM
942 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
943 #endif
944 	&node_state_attr[N_MEMORY].attr.attr,
945 	&node_state_attr[N_CPU].attr.attr,
946 	&node_state_attr[N_GENERIC_INITIATOR].attr.attr,
947 	NULL
948 };
949 
950 static const struct attribute_group memory_root_attr_group = {
951 	.attrs = node_state_attrs,
952 };
953 
954 static const struct attribute_group *cpu_root_attr_groups[] = {
955 	&memory_root_attr_group,
956 	NULL,
957 };
958 
959 void __init node_dev_init(void)
960 {
961 	int ret, i;
962 
963  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
964  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
965 
966 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
967 	if (ret)
968 		panic("%s() failed to register subsystem: %d\n", __func__, ret);
969 
970 	/*
971 	 * Create all node devices, which will properly link the node
972 	 * to applicable memory block devices and already created cpu devices.
973 	 */
974 	for_each_online_node(i) {
975 		ret = register_one_node(i);
976 		if (ret)
977 			panic("%s() failed to add node: %d\n", __func__, ret);
978 	}
979 }
980