xref: /openbmc/linux/drivers/base/node.c (revision cb1aaebe)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Basic Node interface support
4  */
5 
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/mm.h>
9 #include <linux/memory.h>
10 #include <linux/vmstat.h>
11 #include <linux/notifier.h>
12 #include <linux/node.h>
13 #include <linux/hugetlb.h>
14 #include <linux/compaction.h>
15 #include <linux/cpumask.h>
16 #include <linux/topology.h>
17 #include <linux/nodemask.h>
18 #include <linux/cpu.h>
19 #include <linux/device.h>
20 #include <linux/pm_runtime.h>
21 #include <linux/swap.h>
22 #include <linux/slab.h>
23 
24 static struct bus_type node_subsys = {
25 	.name = "node",
26 	.dev_name = "node",
27 };
28 
29 
30 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
31 {
32 	ssize_t n;
33 	cpumask_var_t mask;
34 	struct node *node_dev = to_node(dev);
35 
36 	/* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
37 	BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
38 
39 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
40 		return 0;
41 
42 	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
43 	n = cpumap_print_to_pagebuf(list, buf, mask);
44 	free_cpumask_var(mask);
45 
46 	return n;
47 }
48 
49 static inline ssize_t node_read_cpumask(struct device *dev,
50 				struct device_attribute *attr, char *buf)
51 {
52 	return node_read_cpumap(dev, false, buf);
53 }
54 static inline ssize_t node_read_cpulist(struct device *dev,
55 				struct device_attribute *attr, char *buf)
56 {
57 	return node_read_cpumap(dev, true, buf);
58 }
59 
60 static DEVICE_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
61 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
62 
63 /**
64  * struct node_access_nodes - Access class device to hold user visible
65  * 			      relationships to other nodes.
66  * @dev:	Device for this memory access class
67  * @list_node:	List element in the node's access list
68  * @access:	The access class rank
69  */
70 struct node_access_nodes {
71 	struct device		dev;
72 	struct list_head	list_node;
73 	unsigned		access;
74 #ifdef CONFIG_HMEM_REPORTING
75 	struct node_hmem_attrs	hmem_attrs;
76 #endif
77 };
78 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
79 
80 static struct attribute *node_init_access_node_attrs[] = {
81 	NULL,
82 };
83 
84 static struct attribute *node_targ_access_node_attrs[] = {
85 	NULL,
86 };
87 
88 static const struct attribute_group initiators = {
89 	.name	= "initiators",
90 	.attrs	= node_init_access_node_attrs,
91 };
92 
93 static const struct attribute_group targets = {
94 	.name	= "targets",
95 	.attrs	= node_targ_access_node_attrs,
96 };
97 
98 static const struct attribute_group *node_access_node_groups[] = {
99 	&initiators,
100 	&targets,
101 	NULL,
102 };
103 
104 static void node_remove_accesses(struct node *node)
105 {
106 	struct node_access_nodes *c, *cnext;
107 
108 	list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
109 		list_del(&c->list_node);
110 		device_unregister(&c->dev);
111 	}
112 }
113 
114 static void node_access_release(struct device *dev)
115 {
116 	kfree(to_access_nodes(dev));
117 }
118 
119 static struct node_access_nodes *node_init_node_access(struct node *node,
120 						       unsigned access)
121 {
122 	struct node_access_nodes *access_node;
123 	struct device *dev;
124 
125 	list_for_each_entry(access_node, &node->access_list, list_node)
126 		if (access_node->access == access)
127 			return access_node;
128 
129 	access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
130 	if (!access_node)
131 		return NULL;
132 
133 	access_node->access = access;
134 	dev = &access_node->dev;
135 	dev->parent = &node->dev;
136 	dev->release = node_access_release;
137 	dev->groups = node_access_node_groups;
138 	if (dev_set_name(dev, "access%u", access))
139 		goto free;
140 
141 	if (device_register(dev))
142 		goto free_name;
143 
144 	pm_runtime_no_callbacks(dev);
145 	list_add_tail(&access_node->list_node, &node->access_list);
146 	return access_node;
147 free_name:
148 	kfree_const(dev->kobj.name);
149 free:
150 	kfree(access_node);
151 	return NULL;
152 }
153 
154 #ifdef CONFIG_HMEM_REPORTING
155 #define ACCESS_ATTR(name) 						   \
156 static ssize_t name##_show(struct device *dev,				   \
157 			   struct device_attribute *attr,		   \
158 			   char *buf)					   \
159 {									   \
160 	return sprintf(buf, "%u\n", to_access_nodes(dev)->hmem_attrs.name); \
161 }									   \
162 static DEVICE_ATTR_RO(name);
163 
164 ACCESS_ATTR(read_bandwidth)
165 ACCESS_ATTR(read_latency)
166 ACCESS_ATTR(write_bandwidth)
167 ACCESS_ATTR(write_latency)
168 
169 static struct attribute *access_attrs[] = {
170 	&dev_attr_read_bandwidth.attr,
171 	&dev_attr_read_latency.attr,
172 	&dev_attr_write_bandwidth.attr,
173 	&dev_attr_write_latency.attr,
174 	NULL,
175 };
176 
177 /**
178  * node_set_perf_attrs - Set the performance values for given access class
179  * @nid: Node identifier to be set
180  * @hmem_attrs: Heterogeneous memory performance attributes
181  * @access: The access class the for the given attributes
182  */
183 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
184 			 unsigned access)
185 {
186 	struct node_access_nodes *c;
187 	struct node *node;
188 	int i;
189 
190 	if (WARN_ON_ONCE(!node_online(nid)))
191 		return;
192 
193 	node = node_devices[nid];
194 	c = node_init_node_access(node, access);
195 	if (!c)
196 		return;
197 
198 	c->hmem_attrs = *hmem_attrs;
199 	for (i = 0; access_attrs[i] != NULL; i++) {
200 		if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
201 					    "initiators")) {
202 			pr_info("failed to add performance attribute to node %d\n",
203 				nid);
204 			break;
205 		}
206 	}
207 }
208 
209 /**
210  * struct node_cache_info - Internal tracking for memory node caches
211  * @dev:	Device represeting the cache level
212  * @node:	List element for tracking in the node
213  * @cache_attrs:Attributes for this cache level
214  */
215 struct node_cache_info {
216 	struct device dev;
217 	struct list_head node;
218 	struct node_cache_attrs cache_attrs;
219 };
220 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
221 
222 #define CACHE_ATTR(name, fmt) 						\
223 static ssize_t name##_show(struct device *dev,				\
224 			   struct device_attribute *attr,		\
225 			   char *buf)					\
226 {									\
227 	return sprintf(buf, fmt "\n", to_cache_info(dev)->cache_attrs.name);\
228 }									\
229 DEVICE_ATTR_RO(name);
230 
231 CACHE_ATTR(size, "%llu")
232 CACHE_ATTR(line_size, "%u")
233 CACHE_ATTR(indexing, "%u")
234 CACHE_ATTR(write_policy, "%u")
235 
236 static struct attribute *cache_attrs[] = {
237 	&dev_attr_indexing.attr,
238 	&dev_attr_size.attr,
239 	&dev_attr_line_size.attr,
240 	&dev_attr_write_policy.attr,
241 	NULL,
242 };
243 ATTRIBUTE_GROUPS(cache);
244 
245 static void node_cache_release(struct device *dev)
246 {
247 	kfree(dev);
248 }
249 
250 static void node_cacheinfo_release(struct device *dev)
251 {
252 	struct node_cache_info *info = to_cache_info(dev);
253 	kfree(info);
254 }
255 
256 static void node_init_cache_dev(struct node *node)
257 {
258 	struct device *dev;
259 
260 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
261 	if (!dev)
262 		return;
263 
264 	dev->parent = &node->dev;
265 	dev->release = node_cache_release;
266 	if (dev_set_name(dev, "memory_side_cache"))
267 		goto free_dev;
268 
269 	if (device_register(dev))
270 		goto free_name;
271 
272 	pm_runtime_no_callbacks(dev);
273 	node->cache_dev = dev;
274 	return;
275 free_name:
276 	kfree_const(dev->kobj.name);
277 free_dev:
278 	kfree(dev);
279 }
280 
281 /**
282  * node_add_cache() - add cache attribute to a memory node
283  * @nid: Node identifier that has new cache attributes
284  * @cache_attrs: Attributes for the cache being added
285  */
286 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
287 {
288 	struct node_cache_info *info;
289 	struct device *dev;
290 	struct node *node;
291 
292 	if (!node_online(nid) || !node_devices[nid])
293 		return;
294 
295 	node = node_devices[nid];
296 	list_for_each_entry(info, &node->cache_attrs, node) {
297 		if (info->cache_attrs.level == cache_attrs->level) {
298 			dev_warn(&node->dev,
299 				"attempt to add duplicate cache level:%d\n",
300 				cache_attrs->level);
301 			return;
302 		}
303 	}
304 
305 	if (!node->cache_dev)
306 		node_init_cache_dev(node);
307 	if (!node->cache_dev)
308 		return;
309 
310 	info = kzalloc(sizeof(*info), GFP_KERNEL);
311 	if (!info)
312 		return;
313 
314 	dev = &info->dev;
315 	dev->parent = node->cache_dev;
316 	dev->release = node_cacheinfo_release;
317 	dev->groups = cache_groups;
318 	if (dev_set_name(dev, "index%d", cache_attrs->level))
319 		goto free_cache;
320 
321 	info->cache_attrs = *cache_attrs;
322 	if (device_register(dev)) {
323 		dev_warn(&node->dev, "failed to add cache level:%d\n",
324 			 cache_attrs->level);
325 		goto free_name;
326 	}
327 	pm_runtime_no_callbacks(dev);
328 	list_add_tail(&info->node, &node->cache_attrs);
329 	return;
330 free_name:
331 	kfree_const(dev->kobj.name);
332 free_cache:
333 	kfree(info);
334 }
335 
336 static void node_remove_caches(struct node *node)
337 {
338 	struct node_cache_info *info, *next;
339 
340 	if (!node->cache_dev)
341 		return;
342 
343 	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
344 		list_del(&info->node);
345 		device_unregister(&info->dev);
346 	}
347 	device_unregister(node->cache_dev);
348 }
349 
350 static void node_init_caches(unsigned int nid)
351 {
352 	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
353 }
354 #else
355 static void node_init_caches(unsigned int nid) { }
356 static void node_remove_caches(struct node *node) { }
357 #endif
358 
359 #define K(x) ((x) << (PAGE_SHIFT - 10))
360 static ssize_t node_read_meminfo(struct device *dev,
361 			struct device_attribute *attr, char *buf)
362 {
363 	int n;
364 	int nid = dev->id;
365 	struct pglist_data *pgdat = NODE_DATA(nid);
366 	struct sysinfo i;
367 	unsigned long sreclaimable, sunreclaimable;
368 
369 	si_meminfo_node(&i, nid);
370 	sreclaimable = node_page_state(pgdat, NR_SLAB_RECLAIMABLE);
371 	sunreclaimable = node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE);
372 	n = sprintf(buf,
373 		       "Node %d MemTotal:       %8lu kB\n"
374 		       "Node %d MemFree:        %8lu kB\n"
375 		       "Node %d MemUsed:        %8lu kB\n"
376 		       "Node %d Active:         %8lu kB\n"
377 		       "Node %d Inactive:       %8lu kB\n"
378 		       "Node %d Active(anon):   %8lu kB\n"
379 		       "Node %d Inactive(anon): %8lu kB\n"
380 		       "Node %d Active(file):   %8lu kB\n"
381 		       "Node %d Inactive(file): %8lu kB\n"
382 		       "Node %d Unevictable:    %8lu kB\n"
383 		       "Node %d Mlocked:        %8lu kB\n",
384 		       nid, K(i.totalram),
385 		       nid, K(i.freeram),
386 		       nid, K(i.totalram - i.freeram),
387 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
388 				node_page_state(pgdat, NR_ACTIVE_FILE)),
389 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
390 				node_page_state(pgdat, NR_INACTIVE_FILE)),
391 		       nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
392 		       nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
393 		       nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
394 		       nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
395 		       nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
396 		       nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
397 
398 #ifdef CONFIG_HIGHMEM
399 	n += sprintf(buf + n,
400 		       "Node %d HighTotal:      %8lu kB\n"
401 		       "Node %d HighFree:       %8lu kB\n"
402 		       "Node %d LowTotal:       %8lu kB\n"
403 		       "Node %d LowFree:        %8lu kB\n",
404 		       nid, K(i.totalhigh),
405 		       nid, K(i.freehigh),
406 		       nid, K(i.totalram - i.totalhigh),
407 		       nid, K(i.freeram - i.freehigh));
408 #endif
409 	n += sprintf(buf + n,
410 		       "Node %d Dirty:          %8lu kB\n"
411 		       "Node %d Writeback:      %8lu kB\n"
412 		       "Node %d FilePages:      %8lu kB\n"
413 		       "Node %d Mapped:         %8lu kB\n"
414 		       "Node %d AnonPages:      %8lu kB\n"
415 		       "Node %d Shmem:          %8lu kB\n"
416 		       "Node %d KernelStack:    %8lu kB\n"
417 		       "Node %d PageTables:     %8lu kB\n"
418 		       "Node %d NFS_Unstable:   %8lu kB\n"
419 		       "Node %d Bounce:         %8lu kB\n"
420 		       "Node %d WritebackTmp:   %8lu kB\n"
421 		       "Node %d KReclaimable:   %8lu kB\n"
422 		       "Node %d Slab:           %8lu kB\n"
423 		       "Node %d SReclaimable:   %8lu kB\n"
424 		       "Node %d SUnreclaim:     %8lu kB\n"
425 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
426 		       "Node %d AnonHugePages:  %8lu kB\n"
427 		       "Node %d ShmemHugePages: %8lu kB\n"
428 		       "Node %d ShmemPmdMapped: %8lu kB\n"
429 #endif
430 			,
431 		       nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
432 		       nid, K(node_page_state(pgdat, NR_WRITEBACK)),
433 		       nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
434 		       nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
435 		       nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
436 		       nid, K(i.sharedram),
437 		       nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
438 		       nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
439 		       nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
440 		       nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
441 		       nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
442 		       nid, K(sreclaimable +
443 			      node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
444 		       nid, K(sreclaimable + sunreclaimable),
445 		       nid, K(sreclaimable),
446 		       nid, K(sunreclaimable)
447 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
448 		       ,
449 		       nid, K(node_page_state(pgdat, NR_ANON_THPS) *
450 				       HPAGE_PMD_NR),
451 		       nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
452 				       HPAGE_PMD_NR),
453 		       nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
454 				       HPAGE_PMD_NR)
455 #endif
456 		       );
457 	n += hugetlb_report_node_meminfo(nid, buf + n);
458 	return n;
459 }
460 
461 #undef K
462 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
463 
464 static ssize_t node_read_numastat(struct device *dev,
465 				struct device_attribute *attr, char *buf)
466 {
467 	return sprintf(buf,
468 		       "numa_hit %lu\n"
469 		       "numa_miss %lu\n"
470 		       "numa_foreign %lu\n"
471 		       "interleave_hit %lu\n"
472 		       "local_node %lu\n"
473 		       "other_node %lu\n",
474 		       sum_zone_numa_state(dev->id, NUMA_HIT),
475 		       sum_zone_numa_state(dev->id, NUMA_MISS),
476 		       sum_zone_numa_state(dev->id, NUMA_FOREIGN),
477 		       sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
478 		       sum_zone_numa_state(dev->id, NUMA_LOCAL),
479 		       sum_zone_numa_state(dev->id, NUMA_OTHER));
480 }
481 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
482 
483 static ssize_t node_read_vmstat(struct device *dev,
484 				struct device_attribute *attr, char *buf)
485 {
486 	int nid = dev->id;
487 	struct pglist_data *pgdat = NODE_DATA(nid);
488 	int i;
489 	int n = 0;
490 
491 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
492 		n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
493 			     sum_zone_node_page_state(nid, i));
494 
495 #ifdef CONFIG_NUMA
496 	for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
497 		n += sprintf(buf+n, "%s %lu\n",
498 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
499 			     sum_zone_numa_state(nid, i));
500 #endif
501 
502 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
503 		n += sprintf(buf+n, "%s %lu\n",
504 			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
505 			     NR_VM_NUMA_STAT_ITEMS],
506 			     node_page_state(pgdat, i));
507 
508 	return n;
509 }
510 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
511 
512 static ssize_t node_read_distance(struct device *dev,
513 			struct device_attribute *attr, char *buf)
514 {
515 	int nid = dev->id;
516 	int len = 0;
517 	int i;
518 
519 	/*
520 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
521 	 * at the most (distance + space or newline).
522 	 */
523 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
524 
525 	for_each_online_node(i)
526 		len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
527 
528 	len += sprintf(buf + len, "\n");
529 	return len;
530 }
531 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
532 
533 static struct attribute *node_dev_attrs[] = {
534 	&dev_attr_cpumap.attr,
535 	&dev_attr_cpulist.attr,
536 	&dev_attr_meminfo.attr,
537 	&dev_attr_numastat.attr,
538 	&dev_attr_distance.attr,
539 	&dev_attr_vmstat.attr,
540 	NULL
541 };
542 ATTRIBUTE_GROUPS(node_dev);
543 
544 #ifdef CONFIG_HUGETLBFS
545 /*
546  * hugetlbfs per node attributes registration interface:
547  * When/if hugetlb[fs] subsystem initializes [sometime after this module],
548  * it will register its per node attributes for all online nodes with
549  * memory.  It will also call register_hugetlbfs_with_node(), below, to
550  * register its attribute registration functions with this node driver.
551  * Once these hooks have been initialized, the node driver will call into
552  * the hugetlb module to [un]register attributes for hot-plugged nodes.
553  */
554 static node_registration_func_t __hugetlb_register_node;
555 static node_registration_func_t __hugetlb_unregister_node;
556 
557 static inline bool hugetlb_register_node(struct node *node)
558 {
559 	if (__hugetlb_register_node &&
560 			node_state(node->dev.id, N_MEMORY)) {
561 		__hugetlb_register_node(node);
562 		return true;
563 	}
564 	return false;
565 }
566 
567 static inline void hugetlb_unregister_node(struct node *node)
568 {
569 	if (__hugetlb_unregister_node)
570 		__hugetlb_unregister_node(node);
571 }
572 
573 void register_hugetlbfs_with_node(node_registration_func_t doregister,
574 				  node_registration_func_t unregister)
575 {
576 	__hugetlb_register_node   = doregister;
577 	__hugetlb_unregister_node = unregister;
578 }
579 #else
580 static inline void hugetlb_register_node(struct node *node) {}
581 
582 static inline void hugetlb_unregister_node(struct node *node) {}
583 #endif
584 
585 static void node_device_release(struct device *dev)
586 {
587 	struct node *node = to_node(dev);
588 
589 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
590 	/*
591 	 * We schedule the work only when a memory section is
592 	 * onlined/offlined on this node. When we come here,
593 	 * all the memory on this node has been offlined,
594 	 * so we won't enqueue new work to this work.
595 	 *
596 	 * The work is using node->node_work, so we should
597 	 * flush work before freeing the memory.
598 	 */
599 	flush_work(&node->node_work);
600 #endif
601 	kfree(node);
602 }
603 
604 /*
605  * register_node - Setup a sysfs device for a node.
606  * @num - Node number to use when creating the device.
607  *
608  * Initialize and register the node device.
609  */
610 static int register_node(struct node *node, int num)
611 {
612 	int error;
613 
614 	node->dev.id = num;
615 	node->dev.bus = &node_subsys;
616 	node->dev.release = node_device_release;
617 	node->dev.groups = node_dev_groups;
618 	error = device_register(&node->dev);
619 
620 	if (error)
621 		put_device(&node->dev);
622 	else {
623 		hugetlb_register_node(node);
624 
625 		compaction_register_node(node);
626 	}
627 	return error;
628 }
629 
630 /**
631  * unregister_node - unregister a node device
632  * @node: node going away
633  *
634  * Unregisters a node device @node.  All the devices on the node must be
635  * unregistered before calling this function.
636  */
637 void unregister_node(struct node *node)
638 {
639 	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
640 	node_remove_accesses(node);
641 	node_remove_caches(node);
642 	device_unregister(&node->dev);
643 }
644 
645 struct node *node_devices[MAX_NUMNODES];
646 
647 /*
648  * register cpu under node
649  */
650 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
651 {
652 	int ret;
653 	struct device *obj;
654 
655 	if (!node_online(nid))
656 		return 0;
657 
658 	obj = get_cpu_device(cpu);
659 	if (!obj)
660 		return 0;
661 
662 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
663 				&obj->kobj,
664 				kobject_name(&obj->kobj));
665 	if (ret)
666 		return ret;
667 
668 	return sysfs_create_link(&obj->kobj,
669 				 &node_devices[nid]->dev.kobj,
670 				 kobject_name(&node_devices[nid]->dev.kobj));
671 }
672 
673 /**
674  * register_memory_node_under_compute_node - link memory node to its compute
675  *					     node for a given access class.
676  * @mem_node:	Memory node number
677  * @cpu_node:	Cpu  node number
678  * @access:	Access class to register
679  *
680  * Description:
681  * 	For use with platforms that may have separate memory and compute nodes.
682  * 	This function will export node relationships linking which memory
683  * 	initiator nodes can access memory targets at a given ranked access
684  * 	class.
685  */
686 int register_memory_node_under_compute_node(unsigned int mem_nid,
687 					    unsigned int cpu_nid,
688 					    unsigned access)
689 {
690 	struct node *init_node, *targ_node;
691 	struct node_access_nodes *initiator, *target;
692 	int ret;
693 
694 	if (!node_online(cpu_nid) || !node_online(mem_nid))
695 		return -ENODEV;
696 
697 	init_node = node_devices[cpu_nid];
698 	targ_node = node_devices[mem_nid];
699 	initiator = node_init_node_access(init_node, access);
700 	target = node_init_node_access(targ_node, access);
701 	if (!initiator || !target)
702 		return -ENOMEM;
703 
704 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
705 				      &targ_node->dev.kobj,
706 				      dev_name(&targ_node->dev));
707 	if (ret)
708 		return ret;
709 
710 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
711 				      &init_node->dev.kobj,
712 				      dev_name(&init_node->dev));
713 	if (ret)
714 		goto err;
715 
716 	return 0;
717  err:
718 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
719 				     dev_name(&targ_node->dev));
720 	return ret;
721 }
722 
723 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
724 {
725 	struct device *obj;
726 
727 	if (!node_online(nid))
728 		return 0;
729 
730 	obj = get_cpu_device(cpu);
731 	if (!obj)
732 		return 0;
733 
734 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
735 			  kobject_name(&obj->kobj));
736 	sysfs_remove_link(&obj->kobj,
737 			  kobject_name(&node_devices[nid]->dev.kobj));
738 
739 	return 0;
740 }
741 
742 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
743 static int __ref get_nid_for_pfn(unsigned long pfn)
744 {
745 	if (!pfn_valid_within(pfn))
746 		return -1;
747 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
748 	if (system_state < SYSTEM_RUNNING)
749 		return early_pfn_to_nid(pfn);
750 #endif
751 	return pfn_to_nid(pfn);
752 }
753 
754 /* register memory section under specified node if it spans that node */
755 int register_mem_sect_under_node(struct memory_block *mem_blk, void *arg)
756 {
757 	int ret, nid = *(int *)arg;
758 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
759 
760 	mem_blk->nid = nid;
761 
762 	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
763 	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
764 	sect_end_pfn += PAGES_PER_SECTION - 1;
765 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
766 		int page_nid;
767 
768 		/*
769 		 * memory block could have several absent sections from start.
770 		 * skip pfn range from absent section
771 		 */
772 		if (!pfn_present(pfn)) {
773 			pfn = round_down(pfn + PAGES_PER_SECTION,
774 					 PAGES_PER_SECTION) - 1;
775 			continue;
776 		}
777 
778 		/*
779 		 * We need to check if page belongs to nid only for the boot
780 		 * case, during hotplug we know that all pages in the memory
781 		 * block belong to the same node.
782 		 */
783 		if (system_state == SYSTEM_BOOTING) {
784 			page_nid = get_nid_for_pfn(pfn);
785 			if (page_nid < 0)
786 				continue;
787 			if (page_nid != nid)
788 				continue;
789 		}
790 		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
791 					&mem_blk->dev.kobj,
792 					kobject_name(&mem_blk->dev.kobj));
793 		if (ret)
794 			return ret;
795 
796 		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
797 				&node_devices[nid]->dev.kobj,
798 				kobject_name(&node_devices[nid]->dev.kobj));
799 	}
800 	/* mem section does not span the specified node */
801 	return 0;
802 }
803 
804 /* unregister memory section under all nodes that it spans */
805 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
806 				    unsigned long phys_index)
807 {
808 	NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
809 	unsigned long pfn, sect_start_pfn, sect_end_pfn;
810 
811 	if (!mem_blk) {
812 		NODEMASK_FREE(unlinked_nodes);
813 		return -EFAULT;
814 	}
815 	if (!unlinked_nodes)
816 		return -ENOMEM;
817 	nodes_clear(*unlinked_nodes);
818 
819 	sect_start_pfn = section_nr_to_pfn(phys_index);
820 	sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
821 	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
822 		int nid;
823 
824 		nid = get_nid_for_pfn(pfn);
825 		if (nid < 0)
826 			continue;
827 		if (!node_online(nid))
828 			continue;
829 		if (node_test_and_set(nid, *unlinked_nodes))
830 			continue;
831 		sysfs_remove_link(&node_devices[nid]->dev.kobj,
832 			 kobject_name(&mem_blk->dev.kobj));
833 		sysfs_remove_link(&mem_blk->dev.kobj,
834 			 kobject_name(&node_devices[nid]->dev.kobj));
835 	}
836 	NODEMASK_FREE(unlinked_nodes);
837 	return 0;
838 }
839 
840 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn)
841 {
842 	return walk_memory_range(start_pfn, end_pfn, (void *)&nid,
843 					register_mem_sect_under_node);
844 }
845 
846 #ifdef CONFIG_HUGETLBFS
847 /*
848  * Handle per node hstate attribute [un]registration on transistions
849  * to/from memoryless state.
850  */
851 static void node_hugetlb_work(struct work_struct *work)
852 {
853 	struct node *node = container_of(work, struct node, node_work);
854 
855 	/*
856 	 * We only get here when a node transitions to/from memoryless state.
857 	 * We can detect which transition occurred by examining whether the
858 	 * node has memory now.  hugetlb_register_node() already check this
859 	 * so we try to register the attributes.  If that fails, then the
860 	 * node has transitioned to memoryless, try to unregister the
861 	 * attributes.
862 	 */
863 	if (!hugetlb_register_node(node))
864 		hugetlb_unregister_node(node);
865 }
866 
867 static void init_node_hugetlb_work(int nid)
868 {
869 	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
870 }
871 
872 static int node_memory_callback(struct notifier_block *self,
873 				unsigned long action, void *arg)
874 {
875 	struct memory_notify *mnb = arg;
876 	int nid = mnb->status_change_nid;
877 
878 	switch (action) {
879 	case MEM_ONLINE:
880 	case MEM_OFFLINE:
881 		/*
882 		 * offload per node hstate [un]registration to a work thread
883 		 * when transitioning to/from memoryless state.
884 		 */
885 		if (nid != NUMA_NO_NODE)
886 			schedule_work(&node_devices[nid]->node_work);
887 		break;
888 
889 	case MEM_GOING_ONLINE:
890 	case MEM_GOING_OFFLINE:
891 	case MEM_CANCEL_ONLINE:
892 	case MEM_CANCEL_OFFLINE:
893 	default:
894 		break;
895 	}
896 
897 	return NOTIFY_OK;
898 }
899 #endif	/* CONFIG_HUGETLBFS */
900 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
901 
902 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
903     !defined(CONFIG_HUGETLBFS)
904 static inline int node_memory_callback(struct notifier_block *self,
905 				unsigned long action, void *arg)
906 {
907 	return NOTIFY_OK;
908 }
909 
910 static void init_node_hugetlb_work(int nid) { }
911 
912 #endif
913 
914 int __register_one_node(int nid)
915 {
916 	int error;
917 	int cpu;
918 
919 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
920 	if (!node_devices[nid])
921 		return -ENOMEM;
922 
923 	error = register_node(node_devices[nid], nid);
924 
925 	/* link cpu under this node */
926 	for_each_present_cpu(cpu) {
927 		if (cpu_to_node(cpu) == nid)
928 			register_cpu_under_node(cpu, nid);
929 	}
930 
931 	INIT_LIST_HEAD(&node_devices[nid]->access_list);
932 	/* initialize work queue for memory hot plug */
933 	init_node_hugetlb_work(nid);
934 	node_init_caches(nid);
935 
936 	return error;
937 }
938 
939 void unregister_one_node(int nid)
940 {
941 	if (!node_devices[nid])
942 		return;
943 
944 	unregister_node(node_devices[nid]);
945 	node_devices[nid] = NULL;
946 }
947 
948 /*
949  * node states attributes
950  */
951 
952 static ssize_t print_nodes_state(enum node_states state, char *buf)
953 {
954 	int n;
955 
956 	n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
957 		      nodemask_pr_args(&node_states[state]));
958 	buf[n++] = '\n';
959 	buf[n] = '\0';
960 	return n;
961 }
962 
963 struct node_attr {
964 	struct device_attribute attr;
965 	enum node_states state;
966 };
967 
968 static ssize_t show_node_state(struct device *dev,
969 			       struct device_attribute *attr, char *buf)
970 {
971 	struct node_attr *na = container_of(attr, struct node_attr, attr);
972 	return print_nodes_state(na->state, buf);
973 }
974 
975 #define _NODE_ATTR(name, state) \
976 	{ __ATTR(name, 0444, show_node_state, NULL), state }
977 
978 static struct node_attr node_state_attr[] = {
979 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
980 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
981 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
982 #ifdef CONFIG_HIGHMEM
983 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
984 #endif
985 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
986 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
987 };
988 
989 static struct attribute *node_state_attrs[] = {
990 	&node_state_attr[N_POSSIBLE].attr.attr,
991 	&node_state_attr[N_ONLINE].attr.attr,
992 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
993 #ifdef CONFIG_HIGHMEM
994 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
995 #endif
996 	&node_state_attr[N_MEMORY].attr.attr,
997 	&node_state_attr[N_CPU].attr.attr,
998 	NULL
999 };
1000 
1001 static struct attribute_group memory_root_attr_group = {
1002 	.attrs = node_state_attrs,
1003 };
1004 
1005 static const struct attribute_group *cpu_root_attr_groups[] = {
1006 	&memory_root_attr_group,
1007 	NULL,
1008 };
1009 
1010 #define NODE_CALLBACK_PRI	2	/* lower than SLAB */
1011 static int __init register_node_type(void)
1012 {
1013 	int ret;
1014 
1015  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1016  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1017 
1018 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1019 	if (!ret) {
1020 		static struct notifier_block node_memory_callback_nb = {
1021 			.notifier_call = node_memory_callback,
1022 			.priority = NODE_CALLBACK_PRI,
1023 		};
1024 		register_hotmemory_notifier(&node_memory_callback_nb);
1025 	}
1026 
1027 	/*
1028 	 * Note:  we're not going to unregister the node class if we fail
1029 	 * to register the node state class attribute files.
1030 	 */
1031 	return ret;
1032 }
1033 postcore_initcall(register_node_type);
1034