xref: /linux/drivers/base/node.c (revision 23c48a124b469cee2eb0c75e6d22d366d1caa118)
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 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  * @hmem_attrs: Heterogeneous memory performance attributes
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 node_hmem_attrs	hmem_attrs;
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 						       unsigned int 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(name)						\
165 static ssize_t name##_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)->hmem_attrs.name);	\
171 }									\
172 static DEVICE_ATTR_RO(name)
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  * @hmem_attrs: Heterogeneous memory performance attributes
191  * @access: The access class the for the given attributes
192  */
193 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
194 			 unsigned int 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->hmem_attrs = *hmem_attrs;
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 
219 /**
220  * struct node_cache_info - Internal tracking for memory node caches
221  * @dev:	Device represeting the cache level
222  * @node:	List element for tracking in the node
223  * @cache_attrs:Attributes for this cache level
224  */
225 struct node_cache_info {
226 	struct device dev;
227 	struct list_head node;
228 	struct node_cache_attrs cache_attrs;
229 };
230 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
231 
232 #define CACHE_ATTR(name, fmt) 						\
233 static ssize_t name##_show(struct device *dev,				\
234 			   struct device_attribute *attr,		\
235 			   char *buf)					\
236 {									\
237 	return sysfs_emit(buf, fmt "\n",				\
238 			  to_cache_info(dev)->cache_attrs.name);	\
239 }									\
240 static DEVICE_ATTR_RO(name);
241 
242 CACHE_ATTR(size, "%llu")
243 CACHE_ATTR(line_size, "%u")
244 CACHE_ATTR(indexing, "%u")
245 CACHE_ATTR(write_policy, "%u")
246 
247 static struct attribute *cache_attrs[] = {
248 	&dev_attr_indexing.attr,
249 	&dev_attr_size.attr,
250 	&dev_attr_line_size.attr,
251 	&dev_attr_write_policy.attr,
252 	NULL,
253 };
254 ATTRIBUTE_GROUPS(cache);
255 
256 static void node_cache_release(struct device *dev)
257 {
258 	kfree(dev);
259 }
260 
261 static void node_cacheinfo_release(struct device *dev)
262 {
263 	struct node_cache_info *info = to_cache_info(dev);
264 	kfree(info);
265 }
266 
267 static void node_init_cache_dev(struct node *node)
268 {
269 	struct device *dev;
270 
271 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
272 	if (!dev)
273 		return;
274 
275 	device_initialize(dev);
276 	dev->parent = &node->dev;
277 	dev->release = node_cache_release;
278 	if (dev_set_name(dev, "memory_side_cache"))
279 		goto put_device;
280 
281 	if (device_add(dev))
282 		goto put_device;
283 
284 	pm_runtime_no_callbacks(dev);
285 	node->cache_dev = dev;
286 	return;
287 put_device:
288 	put_device(dev);
289 }
290 
291 /**
292  * node_add_cache() - add cache attribute to a memory node
293  * @nid: Node identifier that has new cache attributes
294  * @cache_attrs: Attributes for the cache being added
295  */
296 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
297 {
298 	struct node_cache_info *info;
299 	struct device *dev;
300 	struct node *node;
301 
302 	if (!node_online(nid) || !node_devices[nid])
303 		return;
304 
305 	node = node_devices[nid];
306 	list_for_each_entry(info, &node->cache_attrs, node) {
307 		if (info->cache_attrs.level == cache_attrs->level) {
308 			dev_warn(&node->dev,
309 				"attempt to add duplicate cache level:%d\n",
310 				cache_attrs->level);
311 			return;
312 		}
313 	}
314 
315 	if (!node->cache_dev)
316 		node_init_cache_dev(node);
317 	if (!node->cache_dev)
318 		return;
319 
320 	info = kzalloc(sizeof(*info), GFP_KERNEL);
321 	if (!info)
322 		return;
323 
324 	dev = &info->dev;
325 	device_initialize(dev);
326 	dev->parent = node->cache_dev;
327 	dev->release = node_cacheinfo_release;
328 	dev->groups = cache_groups;
329 	if (dev_set_name(dev, "index%d", cache_attrs->level))
330 		goto put_device;
331 
332 	info->cache_attrs = *cache_attrs;
333 	if (device_add(dev)) {
334 		dev_warn(&node->dev, "failed to add cache level:%d\n",
335 			 cache_attrs->level);
336 		goto put_device;
337 	}
338 	pm_runtime_no_callbacks(dev);
339 	list_add_tail(&info->node, &node->cache_attrs);
340 	return;
341 put_device:
342 	put_device(dev);
343 }
344 
345 static void node_remove_caches(struct node *node)
346 {
347 	struct node_cache_info *info, *next;
348 
349 	if (!node->cache_dev)
350 		return;
351 
352 	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
353 		list_del(&info->node);
354 		device_unregister(&info->dev);
355 	}
356 	device_unregister(node->cache_dev);
357 }
358 
359 static void node_init_caches(unsigned int nid)
360 {
361 	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
362 }
363 #else
364 static void node_init_caches(unsigned int nid) { }
365 static void node_remove_caches(struct node *node) { }
366 #endif
367 
368 #define K(x) ((x) << (PAGE_SHIFT - 10))
369 static ssize_t node_read_meminfo(struct device *dev,
370 			struct device_attribute *attr, char *buf)
371 {
372 	int len = 0;
373 	int nid = dev->id;
374 	struct pglist_data *pgdat = NODE_DATA(nid);
375 	struct sysinfo i;
376 	unsigned long sreclaimable, sunreclaimable;
377 	unsigned long swapcached = 0;
378 
379 	si_meminfo_node(&i, nid);
380 	sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
381 	sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
382 #ifdef CONFIG_SWAP
383 	swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
384 #endif
385 	len = sysfs_emit_at(buf, len,
386 			    "Node %d MemTotal:       %8lu kB\n"
387 			    "Node %d MemFree:        %8lu kB\n"
388 			    "Node %d MemUsed:        %8lu kB\n"
389 			    "Node %d SwapCached:     %8lu kB\n"
390 			    "Node %d Active:         %8lu kB\n"
391 			    "Node %d Inactive:       %8lu kB\n"
392 			    "Node %d Active(anon):   %8lu kB\n"
393 			    "Node %d Inactive(anon): %8lu kB\n"
394 			    "Node %d Active(file):   %8lu kB\n"
395 			    "Node %d Inactive(file): %8lu kB\n"
396 			    "Node %d Unevictable:    %8lu kB\n"
397 			    "Node %d Mlocked:        %8lu kB\n",
398 			    nid, K(i.totalram),
399 			    nid, K(i.freeram),
400 			    nid, K(i.totalram - i.freeram),
401 			    nid, K(swapcached),
402 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
403 				   node_page_state(pgdat, NR_ACTIVE_FILE)),
404 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
405 				   node_page_state(pgdat, NR_INACTIVE_FILE)),
406 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
407 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
408 			    nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
409 			    nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
410 			    nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
411 			    nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
412 
413 #ifdef CONFIG_HIGHMEM
414 	len += sysfs_emit_at(buf, len,
415 			     "Node %d HighTotal:      %8lu kB\n"
416 			     "Node %d HighFree:       %8lu kB\n"
417 			     "Node %d LowTotal:       %8lu kB\n"
418 			     "Node %d LowFree:        %8lu kB\n",
419 			     nid, K(i.totalhigh),
420 			     nid, K(i.freehigh),
421 			     nid, K(i.totalram - i.totalhigh),
422 			     nid, K(i.freeram - i.freehigh));
423 #endif
424 	len += sysfs_emit_at(buf, len,
425 			     "Node %d Dirty:          %8lu kB\n"
426 			     "Node %d Writeback:      %8lu kB\n"
427 			     "Node %d FilePages:      %8lu kB\n"
428 			     "Node %d Mapped:         %8lu kB\n"
429 			     "Node %d AnonPages:      %8lu kB\n"
430 			     "Node %d Shmem:          %8lu kB\n"
431 			     "Node %d KernelStack:    %8lu kB\n"
432 #ifdef CONFIG_SHADOW_CALL_STACK
433 			     "Node %d ShadowCallStack:%8lu kB\n"
434 #endif
435 			     "Node %d PageTables:     %8lu kB\n"
436 			     "Node %d NFS_Unstable:   %8lu kB\n"
437 			     "Node %d Bounce:         %8lu kB\n"
438 			     "Node %d WritebackTmp:   %8lu kB\n"
439 			     "Node %d KReclaimable:   %8lu kB\n"
440 			     "Node %d Slab:           %8lu kB\n"
441 			     "Node %d SReclaimable:   %8lu kB\n"
442 			     "Node %d SUnreclaim:     %8lu kB\n"
443 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
444 			     "Node %d AnonHugePages:  %8lu kB\n"
445 			     "Node %d ShmemHugePages: %8lu kB\n"
446 			     "Node %d ShmemPmdMapped: %8lu kB\n"
447 			     "Node %d FileHugePages: %8lu kB\n"
448 			     "Node %d FilePmdMapped: %8lu kB\n"
449 #endif
450 			     ,
451 			     nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
452 			     nid, K(node_page_state(pgdat, NR_WRITEBACK)),
453 			     nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
454 			     nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
455 			     nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
456 			     nid, K(i.sharedram),
457 			     nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
458 #ifdef CONFIG_SHADOW_CALL_STACK
459 			     nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
460 #endif
461 			     nid, K(node_page_state(pgdat, NR_PAGETABLE)),
462 			     nid, 0UL,
463 			     nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
464 			     nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
465 			     nid, K(sreclaimable +
466 				    node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
467 			     nid, K(sreclaimable + sunreclaimable),
468 			     nid, K(sreclaimable),
469 			     nid, K(sunreclaimable)
470 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
471 			     ,
472 			     nid, K(node_page_state(pgdat, NR_ANON_THPS)),
473 			     nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
474 			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
475 			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
476 			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
477 #endif
478 			    );
479 	len += hugetlb_report_node_meminfo(buf, len, nid);
480 	return len;
481 }
482 
483 #undef K
484 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
485 
486 static ssize_t node_read_numastat(struct device *dev,
487 				  struct device_attribute *attr, char *buf)
488 {
489 	fold_vm_numa_events();
490 	return sysfs_emit(buf,
491 			  "numa_hit %lu\n"
492 			  "numa_miss %lu\n"
493 			  "numa_foreign %lu\n"
494 			  "interleave_hit %lu\n"
495 			  "local_node %lu\n"
496 			  "other_node %lu\n",
497 			  sum_zone_numa_event_state(dev->id, NUMA_HIT),
498 			  sum_zone_numa_event_state(dev->id, NUMA_MISS),
499 			  sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
500 			  sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
501 			  sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
502 			  sum_zone_numa_event_state(dev->id, NUMA_OTHER));
503 }
504 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
505 
506 static ssize_t node_read_vmstat(struct device *dev,
507 				struct device_attribute *attr, char *buf)
508 {
509 	int nid = dev->id;
510 	struct pglist_data *pgdat = NODE_DATA(nid);
511 	int i;
512 	int len = 0;
513 
514 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
515 		len += sysfs_emit_at(buf, len, "%s %lu\n",
516 				     zone_stat_name(i),
517 				     sum_zone_node_page_state(nid, i));
518 
519 #ifdef CONFIG_NUMA
520 	fold_vm_numa_events();
521 	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
522 		len += sysfs_emit_at(buf, len, "%s %lu\n",
523 				     numa_stat_name(i),
524 				     sum_zone_numa_event_state(nid, i));
525 
526 #endif
527 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
528 		unsigned long pages = node_page_state_pages(pgdat, i);
529 
530 		if (vmstat_item_print_in_thp(i))
531 			pages /= HPAGE_PMD_NR;
532 		len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
533 				     pages);
534 	}
535 
536 	return len;
537 }
538 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
539 
540 static ssize_t node_read_distance(struct device *dev,
541 				  struct device_attribute *attr, char *buf)
542 {
543 	int nid = dev->id;
544 	int len = 0;
545 	int i;
546 
547 	/*
548 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
549 	 * at the most (distance + space or newline).
550 	 */
551 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
552 
553 	for_each_online_node(i) {
554 		len += sysfs_emit_at(buf, len, "%s%d",
555 				     i ? " " : "", node_distance(nid, i));
556 	}
557 
558 	len += sysfs_emit_at(buf, len, "\n");
559 	return len;
560 }
561 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
562 
563 static struct attribute *node_dev_attrs[] = {
564 	&dev_attr_meminfo.attr,
565 	&dev_attr_numastat.attr,
566 	&dev_attr_distance.attr,
567 	&dev_attr_vmstat.attr,
568 	NULL
569 };
570 
571 static struct bin_attribute *node_dev_bin_attrs[] = {
572 	&bin_attr_cpumap,
573 	&bin_attr_cpulist,
574 	NULL
575 };
576 
577 static const struct attribute_group node_dev_group = {
578 	.attrs = node_dev_attrs,
579 	.bin_attrs = node_dev_bin_attrs
580 };
581 
582 static const struct attribute_group *node_dev_groups[] = {
583 	&node_dev_group,
584 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
585 	&arch_node_dev_group,
586 #endif
587 	NULL
588 };
589 
590 #ifdef CONFIG_HUGETLBFS
591 /*
592  * hugetlbfs per node attributes registration interface:
593  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
594  * it will register its per node attributes for all online nodes with
595  * memory.  It will also call register_hugetlbfs_with_node(), below, to
596  * register its attribute registration functions with this node driver.
597  * Once these hooks have been initialized, the node driver will call into
598  * the hugetlb module to [un]register attributes for hot-plugged nodes.
599  */
600 static node_registration_func_t __hugetlb_register_node;
601 static node_registration_func_t __hugetlb_unregister_node;
602 
603 static inline bool hugetlb_register_node(struct node *node)
604 {
605 	if (__hugetlb_register_node &&
606 			node_state(node->dev.id, N_MEMORY)) {
607 		__hugetlb_register_node(node);
608 		return true;
609 	}
610 	return false;
611 }
612 
613 static inline void hugetlb_unregister_node(struct node *node)
614 {
615 	if (__hugetlb_unregister_node)
616 		__hugetlb_unregister_node(node);
617 }
618 
619 void register_hugetlbfs_with_node(node_registration_func_t doregister,
620 				  node_registration_func_t unregister)
621 {
622 	__hugetlb_register_node   = doregister;
623 	__hugetlb_unregister_node = unregister;
624 }
625 #else
626 static inline void hugetlb_register_node(struct node *node) {}
627 
628 static inline void hugetlb_unregister_node(struct node *node) {}
629 #endif
630 
631 static void node_device_release(struct device *dev)
632 {
633 	struct node *node = to_node(dev);
634 
635 #if defined(CONFIG_MEMORY_HOTPLUG) && defined(CONFIG_HUGETLBFS)
636 	/*
637 	 * We schedule the work only when a memory section is
638 	 * onlined/offlined on this node. When we come here,
639 	 * all the memory on this node has been offlined,
640 	 * so we won't enqueue new work to this work.
641 	 *
642 	 * The work is using node->node_work, so we should
643 	 * flush work before freeing the memory.
644 	 */
645 	flush_work(&node->node_work);
646 #endif
647 	kfree(node);
648 }
649 
650 /*
651  * register_node - Setup a sysfs device for a node.
652  * @num - Node number to use when creating the device.
653  *
654  * Initialize and register the node device.
655  */
656 static int register_node(struct node *node, int num)
657 {
658 	int error;
659 
660 	node->dev.id = num;
661 	node->dev.bus = &node_subsys;
662 	node->dev.release = node_device_release;
663 	node->dev.groups = node_dev_groups;
664 	error = device_register(&node->dev);
665 
666 	if (error)
667 		put_device(&node->dev);
668 	else {
669 		hugetlb_register_node(node);
670 
671 		compaction_register_node(node);
672 	}
673 	return error;
674 }
675 
676 /**
677  * unregister_node - unregister a node device
678  * @node: node going away
679  *
680  * Unregisters a node device @node.  All the devices on the node must be
681  * unregistered before calling this function.
682  */
683 void unregister_node(struct node *node)
684 {
685 	compaction_unregister_node(node);
686 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
687 	node_remove_accesses(node);
688 	node_remove_caches(node);
689 	device_unregister(&node->dev);
690 }
691 
692 struct node *node_devices[MAX_NUMNODES];
693 
694 /*
695  * register cpu under node
696  */
697 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
698 {
699 	int ret;
700 	struct device *obj;
701 
702 	if (!node_online(nid))
703 		return 0;
704 
705 	obj = get_cpu_device(cpu);
706 	if (!obj)
707 		return 0;
708 
709 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
710 				&obj->kobj,
711 				kobject_name(&obj->kobj));
712 	if (ret)
713 		return ret;
714 
715 	return sysfs_create_link(&obj->kobj,
716 				 &node_devices[nid]->dev.kobj,
717 				 kobject_name(&node_devices[nid]->dev.kobj));
718 }
719 
720 /**
721  * register_memory_node_under_compute_node - link memory node to its compute
722  *					     node for a given access class.
723  * @mem_nid:	Memory node number
724  * @cpu_nid:	Cpu  node number
725  * @access:	Access class to register
726  *
727  * Description:
728  * 	For use with platforms that may have separate memory and compute nodes.
729  * 	This function will export node relationships linking which memory
730  * 	initiator nodes can access memory targets at a given ranked access
731  * 	class.
732  */
733 int register_memory_node_under_compute_node(unsigned int mem_nid,
734 					    unsigned int cpu_nid,
735 					    unsigned int access)
736 {
737 	struct node *init_node, *targ_node;
738 	struct node_access_nodes *initiator, *target;
739 	int ret;
740 
741 	if (!node_online(cpu_nid) || !node_online(mem_nid))
742 		return -ENODEV;
743 
744 	init_node = node_devices[cpu_nid];
745 	targ_node = node_devices[mem_nid];
746 	initiator = node_init_node_access(init_node, access);
747 	target = node_init_node_access(targ_node, access);
748 	if (!initiator || !target)
749 		return -ENOMEM;
750 
751 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
752 				      &targ_node->dev.kobj,
753 				      dev_name(&targ_node->dev));
754 	if (ret)
755 		return ret;
756 
757 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
758 				      &init_node->dev.kobj,
759 				      dev_name(&init_node->dev));
760 	if (ret)
761 		goto err;
762 
763 	return 0;
764  err:
765 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
766 				     dev_name(&targ_node->dev));
767 	return ret;
768 }
769 
770 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
771 {
772 	struct device *obj;
773 
774 	if (!node_online(nid))
775 		return 0;
776 
777 	obj = get_cpu_device(cpu);
778 	if (!obj)
779 		return 0;
780 
781 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
782 			  kobject_name(&obj->kobj));
783 	sysfs_remove_link(&obj->kobj,
784 			  kobject_name(&node_devices[nid]->dev.kobj));
785 
786 	return 0;
787 }
788 
789 #ifdef CONFIG_MEMORY_HOTPLUG
790 static int __ref get_nid_for_pfn(unsigned long pfn)
791 {
792 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
793 	if (system_state < SYSTEM_RUNNING)
794 		return early_pfn_to_nid(pfn);
795 #endif
796 	return pfn_to_nid(pfn);
797 }
798 
799 static void do_register_memory_block_under_node(int nid,
800 						struct memory_block *mem_blk,
801 						enum meminit_context context)
802 {
803 	int ret;
804 
805 	memory_block_add_nid(mem_blk, nid, context);
806 
807 	ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
808 				       &mem_blk->dev.kobj,
809 				       kobject_name(&mem_blk->dev.kobj));
810 	if (ret && ret != -EEXIST)
811 		dev_err_ratelimited(&node_devices[nid]->dev,
812 				    "can't create link to %s in sysfs (%d)\n",
813 				    kobject_name(&mem_blk->dev.kobj), ret);
814 
815 	ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
816 				&node_devices[nid]->dev.kobj,
817 				kobject_name(&node_devices[nid]->dev.kobj));
818 	if (ret && ret != -EEXIST)
819 		dev_err_ratelimited(&mem_blk->dev,
820 				    "can't create link to %s in sysfs (%d)\n",
821 				    kobject_name(&node_devices[nid]->dev.kobj),
822 				    ret);
823 }
824 
825 /* register memory section under specified node if it spans that node */
826 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
827 					       void *arg)
828 {
829 	unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
830 	unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
831 	unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
832 	int nid = *(int *)arg;
833 	unsigned long pfn;
834 
835 	for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
836 		int page_nid;
837 
838 		/*
839 		 * memory block could have several absent sections from start.
840 		 * skip pfn range from absent section
841 		 */
842 		if (!pfn_in_present_section(pfn)) {
843 			pfn = round_down(pfn + PAGES_PER_SECTION,
844 					 PAGES_PER_SECTION) - 1;
845 			continue;
846 		}
847 
848 		/*
849 		 * We need to check if page belongs to nid only at the boot
850 		 * case because node's ranges can be interleaved.
851 		 */
852 		page_nid = get_nid_for_pfn(pfn);
853 		if (page_nid < 0)
854 			continue;
855 		if (page_nid != nid)
856 			continue;
857 
858 		do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
859 		return 0;
860 	}
861 	/* mem section does not span the specified node */
862 	return 0;
863 }
864 
865 /*
866  * During hotplug we know that all pages in the memory block belong to the same
867  * node.
868  */
869 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
870 						 void *arg)
871 {
872 	int nid = *(int *)arg;
873 
874 	do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
875 	return 0;
876 }
877 
878 /*
879  * Unregister a memory block device under the node it spans. Memory blocks
880  * with multiple nodes cannot be offlined and therefore also never be removed.
881  */
882 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
883 {
884 	if (mem_blk->nid == NUMA_NO_NODE)
885 		return;
886 
887 	sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
888 			  kobject_name(&mem_blk->dev.kobj));
889 	sysfs_remove_link(&mem_blk->dev.kobj,
890 			  kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
891 }
892 
893 void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
894 				       unsigned long end_pfn,
895 				       enum meminit_context context)
896 {
897 	walk_memory_blocks_func_t func;
898 
899 	if (context == MEMINIT_HOTPLUG)
900 		func = register_mem_block_under_node_hotplug;
901 	else
902 		func = register_mem_block_under_node_early;
903 
904 	walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
905 			   (void *)&nid, func);
906 	return;
907 }
908 
909 #ifdef CONFIG_HUGETLBFS
910 /*
911  * Handle per node hstate attribute [un]registration on transistions
912  * to/from memoryless state.
913  */
914 static void node_hugetlb_work(struct work_struct *work)
915 {
916 	struct node *node = container_of(work, struct node, node_work);
917 
918 	/*
919 	 * We only get here when a node transitions to/from memoryless state.
920 	 * We can detect which transition occurred by examining whether the
921 	 * node has memory now.  hugetlb_register_node() already check this
922 	 * so we try to register the attributes.  If that fails, then the
923 	 * node has transitioned to memoryless, try to unregister the
924 	 * attributes.
925 	 */
926 	if (!hugetlb_register_node(node))
927 		hugetlb_unregister_node(node);
928 }
929 
930 static void init_node_hugetlb_work(int nid)
931 {
932 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
933 }
934 
935 static int node_memory_callback(struct notifier_block *self,
936 				unsigned long action, void *arg)
937 {
938 	struct memory_notify *mnb = arg;
939 	int nid = mnb->status_change_nid;
940 
941 	switch (action) {
942 	case MEM_ONLINE:
943 	case MEM_OFFLINE:
944 		/*
945 		 * offload per node hstate [un]registration to a work thread
946 		 * when transitioning to/from memoryless state.
947 		 */
948 		if (nid != NUMA_NO_NODE)
949 			schedule_work(&node_devices[nid]->node_work);
950 		break;
951 
952 	case MEM_GOING_ONLINE:
953 	case MEM_GOING_OFFLINE:
954 	case MEM_CANCEL_ONLINE:
955 	case MEM_CANCEL_OFFLINE:
956 	default:
957 		break;
958 	}
959 
960 	return NOTIFY_OK;
961 }
962 #endif	/* CONFIG_HUGETLBFS */
963 #endif /* CONFIG_MEMORY_HOTPLUG */
964 
965 #if !defined(CONFIG_MEMORY_HOTPLUG) || !defined(CONFIG_HUGETLBFS)
966 static inline int node_memory_callback(struct notifier_block *self,
967 				unsigned long action, void *arg)
968 {
969 	return NOTIFY_OK;
970 }
971 
972 static void init_node_hugetlb_work(int nid) { }
973 
974 #endif
975 
976 int __register_one_node(int nid)
977 {
978 	int error;
979 	int cpu;
980 
981 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
982 	if (!node_devices[nid])
983 		return -ENOMEM;
984 
985 	error = register_node(node_devices[nid], nid);
986 
987 	/* link cpu under this node */
988 	for_each_present_cpu(cpu) {
989 		if (cpu_to_node(cpu) == nid)
990 			register_cpu_under_node(cpu, nid);
991 	}
992 
993 	INIT_LIST_HEAD(&node_devices[nid]->access_list);
994 	/* initialize work queue for memory hot plug */
995 	init_node_hugetlb_work(nid);
996 	node_init_caches(nid);
997 
998 	return error;
999 }
1000 
1001 void unregister_one_node(int nid)
1002 {
1003 	if (!node_devices[nid])
1004 		return;
1005 
1006 	unregister_node(node_devices[nid]);
1007 	node_devices[nid] = NULL;
1008 }
1009 
1010 /*
1011  * node states attributes
1012  */
1013 
1014 struct node_attr {
1015 	struct device_attribute attr;
1016 	enum node_states state;
1017 };
1018 
1019 static ssize_t show_node_state(struct device *dev,
1020 			       struct device_attribute *attr, char *buf)
1021 {
1022 	struct node_attr *na = container_of(attr, struct node_attr, attr);
1023 
1024 	return sysfs_emit(buf, "%*pbl\n",
1025 			  nodemask_pr_args(&node_states[na->state]));
1026 }
1027 
1028 #define _NODE_ATTR(name, state) \
1029 	{ __ATTR(name, 0444, show_node_state, NULL), state }
1030 
1031 static struct node_attr node_state_attr[] = {
1032 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1033 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1034 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1035 #ifdef CONFIG_HIGHMEM
1036 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1037 #endif
1038 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1039 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1040 	[N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
1041 					   N_GENERIC_INITIATOR),
1042 };
1043 
1044 static struct attribute *node_state_attrs[] = {
1045 	&node_state_attr[N_POSSIBLE].attr.attr,
1046 	&node_state_attr[N_ONLINE].attr.attr,
1047 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
1048 #ifdef CONFIG_HIGHMEM
1049 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
1050 #endif
1051 	&node_state_attr[N_MEMORY].attr.attr,
1052 	&node_state_attr[N_CPU].attr.attr,
1053 	&node_state_attr[N_GENERIC_INITIATOR].attr.attr,
1054 	NULL
1055 };
1056 
1057 static const struct attribute_group memory_root_attr_group = {
1058 	.attrs = node_state_attrs,
1059 };
1060 
1061 static const struct attribute_group *cpu_root_attr_groups[] = {
1062 	&memory_root_attr_group,
1063 	NULL,
1064 };
1065 
1066 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
1067 void __init node_dev_init(void)
1068 {
1069 	static struct notifier_block node_memory_callback_nb = {
1070 		.notifier_call = node_memory_callback,
1071 		.priority = NODE_CALLBACK_PRI,
1072 	};
1073 	int ret, i;
1074 
1075  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1076  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1077 
1078 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1079 	if (ret)
1080 		panic("%s() failed to register subsystem: %d\n", __func__, ret);
1081 
1082 	register_hotmemory_notifier(&node_memory_callback_nb);
1083 
1084 	/*
1085 	 * Create all node devices, which will properly link the node
1086 	 * to applicable memory block devices and already created cpu devices.
1087 	 */
1088 	for_each_online_node(i) {
1089 		ret = register_one_node(i);
1090 		if (ret)
1091 			panic("%s() failed to add node: %d\n", __func__, ret);
1092 	}
1093 }
1094