xref: /linux/drivers/base/node.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
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 	device_initialize(dev);
272 	dev->parent = &node->dev;
273 	dev->release = node_cache_release;
274 	if (dev_set_name(dev, "memory_side_cache"))
275 		goto put_device;
276 
277 	if (device_add(dev))
278 		goto put_device;
279 
280 	pm_runtime_no_callbacks(dev);
281 	node->cache_dev = dev;
282 	return;
283 put_device:
284 	put_device(dev);
285 }
286 
287 /**
288  * node_add_cache() - add cache attribute to a memory node
289  * @nid: Node identifier that has new cache attributes
290  * @cache_attrs: Attributes for the cache being added
291  */
292 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
293 {
294 	struct node_cache_info *info;
295 	struct device *dev;
296 	struct node *node;
297 
298 	if (!node_online(nid) || !node_devices[nid])
299 		return;
300 
301 	node = node_devices[nid];
302 	list_for_each_entry(info, &node->cache_attrs, node) {
303 		if (info->cache_attrs.level == cache_attrs->level) {
304 			dev_warn(&node->dev,
305 				"attempt to add duplicate cache level:%d\n",
306 				cache_attrs->level);
307 			return;
308 		}
309 	}
310 
311 	if (!node->cache_dev)
312 		node_init_cache_dev(node);
313 	if (!node->cache_dev)
314 		return;
315 
316 	info = kzalloc(sizeof(*info), GFP_KERNEL);
317 	if (!info)
318 		return;
319 
320 	dev = &info->dev;
321 	device_initialize(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 put_device;
327 
328 	info->cache_attrs = *cache_attrs;
329 	if (device_add(dev)) {
330 		dev_warn(&node->dev, "failed to add cache level:%d\n",
331 			 cache_attrs->level);
332 		goto put_device;
333 	}
334 	pm_runtime_no_callbacks(dev);
335 	list_add_tail(&info->node, &node->cache_attrs);
336 	return;
337 put_device:
338 	put_device(dev);
339 }
340 
341 static void node_remove_caches(struct node *node)
342 {
343 	struct node_cache_info *info, *next;
344 
345 	if (!node->cache_dev)
346 		return;
347 
348 	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
349 		list_del(&info->node);
350 		device_unregister(&info->dev);
351 	}
352 	device_unregister(node->cache_dev);
353 }
354 
355 static void node_init_caches(unsigned int nid)
356 {
357 	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
358 }
359 #else
360 static void node_init_caches(unsigned int nid) { }
361 static void node_remove_caches(struct node *node) { }
362 #endif
363 
364 #define K(x) ((x) << (PAGE_SHIFT - 10))
365 static ssize_t node_read_meminfo(struct device *dev,
366 			struct device_attribute *attr, char *buf)
367 {
368 	int len = 0;
369 	int nid = dev->id;
370 	struct pglist_data *pgdat = NODE_DATA(nid);
371 	struct sysinfo i;
372 	unsigned long sreclaimable, sunreclaimable;
373 	unsigned long swapcached = 0;
374 
375 	si_meminfo_node(&i, nid);
376 	sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
377 	sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
378 #ifdef CONFIG_SWAP
379 	swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
380 #endif
381 	len = sysfs_emit_at(buf, len,
382 			    "Node %d MemTotal:       %8lu kB\n"
383 			    "Node %d MemFree:        %8lu kB\n"
384 			    "Node %d MemUsed:        %8lu kB\n"
385 			    "Node %d SwapCached:     %8lu kB\n"
386 			    "Node %d Active:         %8lu kB\n"
387 			    "Node %d Inactive:       %8lu kB\n"
388 			    "Node %d Active(anon):   %8lu kB\n"
389 			    "Node %d Inactive(anon): %8lu kB\n"
390 			    "Node %d Active(file):   %8lu kB\n"
391 			    "Node %d Inactive(file): %8lu kB\n"
392 			    "Node %d Unevictable:    %8lu kB\n"
393 			    "Node %d Mlocked:        %8lu kB\n",
394 			    nid, K(i.totalram),
395 			    nid, K(i.freeram),
396 			    nid, K(i.totalram - i.freeram),
397 			    nid, K(swapcached),
398 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
399 				   node_page_state(pgdat, NR_ACTIVE_FILE)),
400 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
401 				   node_page_state(pgdat, NR_INACTIVE_FILE)),
402 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
403 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
404 			    nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
405 			    nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
406 			    nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
407 			    nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
408 
409 #ifdef CONFIG_HIGHMEM
410 	len += sysfs_emit_at(buf, len,
411 			     "Node %d HighTotal:      %8lu kB\n"
412 			     "Node %d HighFree:       %8lu kB\n"
413 			     "Node %d LowTotal:       %8lu kB\n"
414 			     "Node %d LowFree:        %8lu kB\n",
415 			     nid, K(i.totalhigh),
416 			     nid, K(i.freehigh),
417 			     nid, K(i.totalram - i.totalhigh),
418 			     nid, K(i.freeram - i.freehigh));
419 #endif
420 	len += sysfs_emit_at(buf, len,
421 			     "Node %d Dirty:          %8lu kB\n"
422 			     "Node %d Writeback:      %8lu kB\n"
423 			     "Node %d FilePages:      %8lu kB\n"
424 			     "Node %d Mapped:         %8lu kB\n"
425 			     "Node %d AnonPages:      %8lu kB\n"
426 			     "Node %d Shmem:          %8lu kB\n"
427 			     "Node %d KernelStack:    %8lu kB\n"
428 #ifdef CONFIG_SHADOW_CALL_STACK
429 			     "Node %d ShadowCallStack:%8lu kB\n"
430 #endif
431 			     "Node %d PageTables:     %8lu kB\n"
432 			     "Node %d NFS_Unstable:   %8lu kB\n"
433 			     "Node %d Bounce:         %8lu kB\n"
434 			     "Node %d WritebackTmp:   %8lu kB\n"
435 			     "Node %d KReclaimable:   %8lu kB\n"
436 			     "Node %d Slab:           %8lu kB\n"
437 			     "Node %d SReclaimable:   %8lu kB\n"
438 			     "Node %d SUnreclaim:     %8lu kB\n"
439 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
440 			     "Node %d AnonHugePages:  %8lu kB\n"
441 			     "Node %d ShmemHugePages: %8lu kB\n"
442 			     "Node %d ShmemPmdMapped: %8lu kB\n"
443 			     "Node %d FileHugePages: %8lu kB\n"
444 			     "Node %d FilePmdMapped: %8lu kB\n"
445 #endif
446 			     ,
447 			     nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
448 			     nid, K(node_page_state(pgdat, NR_WRITEBACK)),
449 			     nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
450 			     nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
451 			     nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
452 			     nid, K(i.sharedram),
453 			     nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
454 #ifdef CONFIG_SHADOW_CALL_STACK
455 			     nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
456 #endif
457 			     nid, K(node_page_state(pgdat, NR_PAGETABLE)),
458 			     nid, 0UL,
459 			     nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
460 			     nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
461 			     nid, K(sreclaimable +
462 				    node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
463 			     nid, K(sreclaimable + sunreclaimable),
464 			     nid, K(sreclaimable),
465 			     nid, K(sunreclaimable)
466 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
467 			     ,
468 			     nid, K(node_page_state(pgdat, NR_ANON_THPS)),
469 			     nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
470 			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
471 			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
472 			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
473 #endif
474 			    );
475 	len += hugetlb_report_node_meminfo(buf, len, nid);
476 	return len;
477 }
478 
479 #undef K
480 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
481 
482 static ssize_t node_read_numastat(struct device *dev,
483 				  struct device_attribute *attr, char *buf)
484 {
485 	return sysfs_emit(buf,
486 			  "numa_hit %lu\n"
487 			  "numa_miss %lu\n"
488 			  "numa_foreign %lu\n"
489 			  "interleave_hit %lu\n"
490 			  "local_node %lu\n"
491 			  "other_node %lu\n",
492 			  sum_zone_numa_state(dev->id, NUMA_HIT),
493 			  sum_zone_numa_state(dev->id, NUMA_MISS),
494 			  sum_zone_numa_state(dev->id, NUMA_FOREIGN),
495 			  sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
496 			  sum_zone_numa_state(dev->id, NUMA_LOCAL),
497 			  sum_zone_numa_state(dev->id, NUMA_OTHER));
498 }
499 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
500 
501 static ssize_t node_read_vmstat(struct device *dev,
502 				struct device_attribute *attr, char *buf)
503 {
504 	int nid = dev->id;
505 	struct pglist_data *pgdat = NODE_DATA(nid);
506 	int i;
507 	int len = 0;
508 
509 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
510 		len += sysfs_emit_at(buf, len, "%s %lu\n",
511 				     zone_stat_name(i),
512 				     sum_zone_node_page_state(nid, i));
513 
514 #ifdef CONFIG_NUMA
515 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
516 		len += sysfs_emit_at(buf, len, "%s %lu\n",
517 				     numa_stat_name(i),
518 				     sum_zone_numa_state(nid, i));
519 
520 #endif
521 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
522 		unsigned long pages = node_page_state_pages(pgdat, i);
523 
524 		if (vmstat_item_print_in_thp(i))
525 			pages /= HPAGE_PMD_NR;
526 		len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
527 				     pages);
528 	}
529 
530 	return len;
531 }
532 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
533 
534 static ssize_t node_read_distance(struct device *dev,
535 				  struct device_attribute *attr, char *buf)
536 {
537 	int nid = dev->id;
538 	int len = 0;
539 	int i;
540 
541 	/*
542 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
543 	 * at the most (distance + space or newline).
544 	 */
545 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
546 
547 	for_each_online_node(i) {
548 		len += sysfs_emit_at(buf, len, "%s%d",
549 				     i ? " " : "", node_distance(nid, i));
550 	}
551 
552 	len += sysfs_emit_at(buf, len, "\n");
553 	return len;
554 }
555 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
556 
557 static struct attribute *node_dev_attrs[] = {
558 	&dev_attr_cpumap.attr,
559 	&dev_attr_cpulist.attr,
560 	&dev_attr_meminfo.attr,
561 	&dev_attr_numastat.attr,
562 	&dev_attr_distance.attr,
563 	&dev_attr_vmstat.attr,
564 	NULL
565 };
566 ATTRIBUTE_GROUPS(node_dev);
567 
568 #ifdef CONFIG_HUGETLBFS
569 /*
570  * hugetlbfs per node attributes registration interface:
571  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
572  * it will register its per node attributes for all online nodes with
573  * memory.  It will also call register_hugetlbfs_with_node(), below, to
574  * register its attribute registration functions with this node driver.
575  * Once these hooks have been initialized, the node driver will call into
576  * the hugetlb module to [un]register attributes for hot-plugged nodes.
577  */
578 static node_registration_func_t __hugetlb_register_node;
579 static node_registration_func_t __hugetlb_unregister_node;
580 
581 static inline bool hugetlb_register_node(struct node *node)
582 {
583 	if (__hugetlb_register_node &&
584 			node_state(node->dev.id, N_MEMORY)) {
585 		__hugetlb_register_node(node);
586 		return true;
587 	}
588 	return false;
589 }
590 
591 static inline void hugetlb_unregister_node(struct node *node)
592 {
593 	if (__hugetlb_unregister_node)
594 		__hugetlb_unregister_node(node);
595 }
596 
597 void register_hugetlbfs_with_node(node_registration_func_t doregister,
598 				  node_registration_func_t unregister)
599 {
600 	__hugetlb_register_node   = doregister;
601 	__hugetlb_unregister_node = unregister;
602 }
603 #else
604 static inline void hugetlb_register_node(struct node *node) {}
605 
606 static inline void hugetlb_unregister_node(struct node *node) {}
607 #endif
608 
609 static void node_device_release(struct device *dev)
610 {
611 	struct node *node = to_node(dev);
612 
613 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
614 	/*
615 	 * We schedule the work only when a memory section is
616 	 * onlined/offlined on this node. When we come here,
617 	 * all the memory on this node has been offlined,
618 	 * so we won't enqueue new work to this work.
619 	 *
620 	 * The work is using node->node_work, so we should
621 	 * flush work before freeing the memory.
622 	 */
623 	flush_work(&node->node_work);
624 #endif
625 	kfree(node);
626 }
627 
628 /*
629  * register_node - Setup a sysfs device for a node.
630  * @num - Node number to use when creating the device.
631  *
632  * Initialize and register the node device.
633  */
634 static int register_node(struct node *node, int num)
635 {
636 	int error;
637 
638 	node->dev.id = num;
639 	node->dev.bus = &node_subsys;
640 	node->dev.release = node_device_release;
641 	node->dev.groups = node_dev_groups;
642 	error = device_register(&node->dev);
643 
644 	if (error)
645 		put_device(&node->dev);
646 	else {
647 		hugetlb_register_node(node);
648 
649 		compaction_register_node(node);
650 	}
651 	return error;
652 }
653 
654 /**
655  * unregister_node - unregister a node device
656  * @node: node going away
657  *
658  * Unregisters a node device @node.  All the devices on the node must be
659  * unregistered before calling this function.
660  */
661 void unregister_node(struct node *node)
662 {
663 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
664 	node_remove_accesses(node);
665 	node_remove_caches(node);
666 	device_unregister(&node->dev);
667 }
668 
669 struct node *node_devices[MAX_NUMNODES];
670 
671 /*
672  * register cpu under node
673  */
674 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
675 {
676 	int ret;
677 	struct device *obj;
678 
679 	if (!node_online(nid))
680 		return 0;
681 
682 	obj = get_cpu_device(cpu);
683 	if (!obj)
684 		return 0;
685 
686 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
687 				&obj->kobj,
688 				kobject_name(&obj->kobj));
689 	if (ret)
690 		return ret;
691 
692 	return sysfs_create_link(&obj->kobj,
693 				 &node_devices[nid]->dev.kobj,
694 				 kobject_name(&node_devices[nid]->dev.kobj));
695 }
696 
697 /**
698  * register_memory_node_under_compute_node - link memory node to its compute
699  *					     node for a given access class.
700  * @mem_nid:	Memory node number
701  * @cpu_nid:	Cpu  node number
702  * @access:	Access class to register
703  *
704  * Description:
705  * 	For use with platforms that may have separate memory and compute nodes.
706  * 	This function will export node relationships linking which memory
707  * 	initiator nodes can access memory targets at a given ranked access
708  * 	class.
709  */
710 int register_memory_node_under_compute_node(unsigned int mem_nid,
711 					    unsigned int cpu_nid,
712 					    unsigned access)
713 {
714 	struct node *init_node, *targ_node;
715 	struct node_access_nodes *initiator, *target;
716 	int ret;
717 
718 	if (!node_online(cpu_nid) || !node_online(mem_nid))
719 		return -ENODEV;
720 
721 	init_node = node_devices[cpu_nid];
722 	targ_node = node_devices[mem_nid];
723 	initiator = node_init_node_access(init_node, access);
724 	target = node_init_node_access(targ_node, access);
725 	if (!initiator || !target)
726 		return -ENOMEM;
727 
728 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
729 				      &targ_node->dev.kobj,
730 				      dev_name(&targ_node->dev));
731 	if (ret)
732 		return ret;
733 
734 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
735 				      &init_node->dev.kobj,
736 				      dev_name(&init_node->dev));
737 	if (ret)
738 		goto err;
739 
740 	return 0;
741  err:
742 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
743 				     dev_name(&targ_node->dev));
744 	return ret;
745 }
746 
747 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
748 {
749 	struct device *obj;
750 
751 	if (!node_online(nid))
752 		return 0;
753 
754 	obj = get_cpu_device(cpu);
755 	if (!obj)
756 		return 0;
757 
758 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
759 			  kobject_name(&obj->kobj));
760 	sysfs_remove_link(&obj->kobj,
761 			  kobject_name(&node_devices[nid]->dev.kobj));
762 
763 	return 0;
764 }
765 
766 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
767 static int __ref get_nid_for_pfn(unsigned long pfn)
768 {
769 	if (!pfn_valid_within(pfn))
770 		return -1;
771 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
772 	if (system_state < SYSTEM_RUNNING)
773 		return early_pfn_to_nid(pfn);
774 #endif
775 	return pfn_to_nid(pfn);
776 }
777 
778 static void do_register_memory_block_under_node(int nid,
779 						struct memory_block *mem_blk)
780 {
781 	int ret;
782 
783 	/*
784 	 * If this memory block spans multiple nodes, we only indicate
785 	 * the last processed node.
786 	 */
787 	mem_blk->nid = nid;
788 
789 	ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
790 				       &mem_blk->dev.kobj,
791 				       kobject_name(&mem_blk->dev.kobj));
792 	if (ret && ret != -EEXIST)
793 		dev_err_ratelimited(&node_devices[nid]->dev,
794 				    "can't create link to %s in sysfs (%d)\n",
795 				    kobject_name(&mem_blk->dev.kobj), ret);
796 
797 	ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
798 				&node_devices[nid]->dev.kobj,
799 				kobject_name(&node_devices[nid]->dev.kobj));
800 	if (ret && ret != -EEXIST)
801 		dev_err_ratelimited(&mem_blk->dev,
802 				    "can't create link to %s in sysfs (%d)\n",
803 				    kobject_name(&node_devices[nid]->dev.kobj),
804 				    ret);
805 }
806 
807 /* register memory section under specified node if it spans that node */
808 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
809 					       void *arg)
810 {
811 	unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
812 	unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
813 	unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
814 	int nid = *(int *)arg;
815 	unsigned long pfn;
816 
817 	for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
818 		int page_nid;
819 
820 		/*
821 		 * memory block could have several absent sections from start.
822 		 * skip pfn range from absent section
823 		 */
824 		if (!pfn_in_present_section(pfn)) {
825 			pfn = round_down(pfn + PAGES_PER_SECTION,
826 					 PAGES_PER_SECTION) - 1;
827 			continue;
828 		}
829 
830 		/*
831 		 * We need to check if page belongs to nid only at the boot
832 		 * case because node's ranges can be interleaved.
833 		 */
834 		page_nid = get_nid_for_pfn(pfn);
835 		if (page_nid < 0)
836 			continue;
837 		if (page_nid != nid)
838 			continue;
839 
840 		do_register_memory_block_under_node(nid, mem_blk);
841 		return 0;
842 	}
843 	/* mem section does not span the specified node */
844 	return 0;
845 }
846 
847 /*
848  * During hotplug we know that all pages in the memory block belong to the same
849  * node.
850  */
851 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
852 						 void *arg)
853 {
854 	int nid = *(int *)arg;
855 
856 	do_register_memory_block_under_node(nid, mem_blk);
857 	return 0;
858 }
859 
860 /*
861  * Unregister a memory block device under the node it spans. Memory blocks
862  * with multiple nodes cannot be offlined and therefore also never be removed.
863  */
864 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
865 {
866 	if (mem_blk->nid == NUMA_NO_NODE)
867 		return;
868 
869 	sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
870 			  kobject_name(&mem_blk->dev.kobj));
871 	sysfs_remove_link(&mem_blk->dev.kobj,
872 			  kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
873 }
874 
875 void link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn,
876 		       enum meminit_context context)
877 {
878 	walk_memory_blocks_func_t func;
879 
880 	if (context == MEMINIT_HOTPLUG)
881 		func = register_mem_block_under_node_hotplug;
882 	else
883 		func = register_mem_block_under_node_early;
884 
885 	walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
886 			   (void *)&nid, func);
887 	return;
888 }
889 
890 #ifdef CONFIG_HUGETLBFS
891 /*
892  * Handle per node hstate attribute [un]registration on transistions
893  * to/from memoryless state.
894  */
895 static void node_hugetlb_work(struct work_struct *work)
896 {
897 	struct node *node = container_of(work, struct node, node_work);
898 
899 	/*
900 	 * We only get here when a node transitions to/from memoryless state.
901 	 * We can detect which transition occurred by examining whether the
902 	 * node has memory now.  hugetlb_register_node() already check this
903 	 * so we try to register the attributes.  If that fails, then the
904 	 * node has transitioned to memoryless, try to unregister the
905 	 * attributes.
906 	 */
907 	if (!hugetlb_register_node(node))
908 		hugetlb_unregister_node(node);
909 }
910 
911 static void init_node_hugetlb_work(int nid)
912 {
913 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
914 }
915 
916 static int node_memory_callback(struct notifier_block *self,
917 				unsigned long action, void *arg)
918 {
919 	struct memory_notify *mnb = arg;
920 	int nid = mnb->status_change_nid;
921 
922 	switch (action) {
923 	case MEM_ONLINE:
924 	case MEM_OFFLINE:
925 		/*
926 		 * offload per node hstate [un]registration to a work thread
927 		 * when transitioning to/from memoryless state.
928 		 */
929 		if (nid != NUMA_NO_NODE)
930 			schedule_work(&node_devices[nid]->node_work);
931 		break;
932 
933 	case MEM_GOING_ONLINE:
934 	case MEM_GOING_OFFLINE:
935 	case MEM_CANCEL_ONLINE:
936 	case MEM_CANCEL_OFFLINE:
937 	default:
938 		break;
939 	}
940 
941 	return NOTIFY_OK;
942 }
943 #endif	/* CONFIG_HUGETLBFS */
944 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
945 
946 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
947     !defined(CONFIG_HUGETLBFS)
948 static inline int node_memory_callback(struct notifier_block *self,
949 				unsigned long action, void *arg)
950 {
951 	return NOTIFY_OK;
952 }
953 
954 static void init_node_hugetlb_work(int nid) { }
955 
956 #endif
957 
958 int __register_one_node(int nid)
959 {
960 	int error;
961 	int cpu;
962 
963 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
964 	if (!node_devices[nid])
965 		return -ENOMEM;
966 
967 	error = register_node(node_devices[nid], nid);
968 
969 	/* link cpu under this node */
970 	for_each_present_cpu(cpu) {
971 		if (cpu_to_node(cpu) == nid)
972 			register_cpu_under_node(cpu, nid);
973 	}
974 
975 	INIT_LIST_HEAD(&node_devices[nid]->access_list);
976 	/* initialize work queue for memory hot plug */
977 	init_node_hugetlb_work(nid);
978 	node_init_caches(nid);
979 
980 	return error;
981 }
982 
983 void unregister_one_node(int nid)
984 {
985 	if (!node_devices[nid])
986 		return;
987 
988 	unregister_node(node_devices[nid]);
989 	node_devices[nid] = NULL;
990 }
991 
992 /*
993  * node states attributes
994  */
995 
996 struct node_attr {
997 	struct device_attribute attr;
998 	enum node_states state;
999 };
1000 
1001 static ssize_t show_node_state(struct device *dev,
1002 			       struct device_attribute *attr, char *buf)
1003 {
1004 	struct node_attr *na = container_of(attr, struct node_attr, attr);
1005 
1006 	return sysfs_emit(buf, "%*pbl\n",
1007 			  nodemask_pr_args(&node_states[na->state]));
1008 }
1009 
1010 #define _NODE_ATTR(name, state) \
1011 	{ __ATTR(name, 0444, show_node_state, NULL), state }
1012 
1013 static struct node_attr node_state_attr[] = {
1014 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1015 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1016 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1017 #ifdef CONFIG_HIGHMEM
1018 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1019 #endif
1020 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1021 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1022 	[N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
1023 					   N_GENERIC_INITIATOR),
1024 };
1025 
1026 static struct attribute *node_state_attrs[] = {
1027 	&node_state_attr[N_POSSIBLE].attr.attr,
1028 	&node_state_attr[N_ONLINE].attr.attr,
1029 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
1030 #ifdef CONFIG_HIGHMEM
1031 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
1032 #endif
1033 	&node_state_attr[N_MEMORY].attr.attr,
1034 	&node_state_attr[N_CPU].attr.attr,
1035 	&node_state_attr[N_GENERIC_INITIATOR].attr.attr,
1036 	NULL
1037 };
1038 
1039 static struct attribute_group memory_root_attr_group = {
1040 	.attrs = node_state_attrs,
1041 };
1042 
1043 static const struct attribute_group *cpu_root_attr_groups[] = {
1044 	&memory_root_attr_group,
1045 	NULL,
1046 };
1047 
1048 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
1049 static int __init register_node_type(void)
1050 {
1051 	int ret;
1052 
1053  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1054  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1055 
1056 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1057 	if (!ret) {
1058 		static struct notifier_block node_memory_callback_nb = {
1059 			.notifier_call = node_memory_callback,
1060 			.priority = NODE_CALLBACK_PRI,
1061 		};
1062 		register_hotmemory_notifier(&node_memory_callback_nb);
1063 	}
1064 
1065 	/*
1066 	 * Note:  we're not going to unregister the node class if we fail
1067 	 * to register the node state class attribute files.
1068 	 */
1069 	return ret;
1070 }
1071 postcore_initcall(register_node_type);
1072