xref: /openbmc/linux/drivers/base/node.c (revision 287dcc2b)
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 #include <linux/hugetlb.h>
24 
25 static struct bus_type node_subsys = {
26 	.name = "node",
27 	.dev_name = "node",
28 };
29 
30 static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
31 				  struct bin_attribute *attr, char *buf,
32 				  loff_t off, size_t count)
33 {
34 	struct device *dev = kobj_to_dev(kobj);
35 	struct node *node_dev = to_node(dev);
36 	cpumask_var_t mask;
37 	ssize_t n;
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_bitmask_to_buf(buf, mask, off, count);
44 	free_cpumask_var(mask);
45 
46 	return n;
47 }
48 
49 static BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
50 
51 static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
52 				   struct bin_attribute *attr, char *buf,
53 				   loff_t off, size_t count)
54 {
55 	struct device *dev = kobj_to_dev(kobj);
56 	struct node *node_dev = to_node(dev);
57 	cpumask_var_t mask;
58 	ssize_t n;
59 
60 	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
61 		return 0;
62 
63 	cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
64 	n = cpumap_print_list_to_buf(buf, mask, off, count);
65 	free_cpumask_var(mask);
66 
67 	return n;
68 }
69 
70 static BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
71 
72 /**
73  * struct node_access_nodes - Access class device to hold user visible
74  * 			      relationships to other nodes.
75  * @dev:	Device for this memory access class
76  * @list_node:	List element in the node's access list
77  * @access:	The access class rank
78  * @hmem_attrs: Heterogeneous memory performance attributes
79  */
80 struct node_access_nodes {
81 	struct device		dev;
82 	struct list_head	list_node;
83 	unsigned int		access;
84 #ifdef CONFIG_HMEM_REPORTING
85 	struct node_hmem_attrs	hmem_attrs;
86 #endif
87 };
88 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
89 
90 static struct attribute *node_init_access_node_attrs[] = {
91 	NULL,
92 };
93 
94 static struct attribute *node_targ_access_node_attrs[] = {
95 	NULL,
96 };
97 
98 static const struct attribute_group initiators = {
99 	.name	= "initiators",
100 	.attrs	= node_init_access_node_attrs,
101 };
102 
103 static const struct attribute_group targets = {
104 	.name	= "targets",
105 	.attrs	= node_targ_access_node_attrs,
106 };
107 
108 static const struct attribute_group *node_access_node_groups[] = {
109 	&initiators,
110 	&targets,
111 	NULL,
112 };
113 
114 static void node_remove_accesses(struct node *node)
115 {
116 	struct node_access_nodes *c, *cnext;
117 
118 	list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
119 		list_del(&c->list_node);
120 		device_unregister(&c->dev);
121 	}
122 }
123 
124 static void node_access_release(struct device *dev)
125 {
126 	kfree(to_access_nodes(dev));
127 }
128 
129 static struct node_access_nodes *node_init_node_access(struct node *node,
130 						       unsigned int access)
131 {
132 	struct node_access_nodes *access_node;
133 	struct device *dev;
134 
135 	list_for_each_entry(access_node, &node->access_list, list_node)
136 		if (access_node->access == access)
137 			return access_node;
138 
139 	access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
140 	if (!access_node)
141 		return NULL;
142 
143 	access_node->access = access;
144 	dev = &access_node->dev;
145 	dev->parent = &node->dev;
146 	dev->release = node_access_release;
147 	dev->groups = node_access_node_groups;
148 	if (dev_set_name(dev, "access%u", access))
149 		goto free;
150 
151 	if (device_register(dev))
152 		goto free_name;
153 
154 	pm_runtime_no_callbacks(dev);
155 	list_add_tail(&access_node->list_node, &node->access_list);
156 	return access_node;
157 free_name:
158 	kfree_const(dev->kobj.name);
159 free:
160 	kfree(access_node);
161 	return NULL;
162 }
163 
164 #ifdef CONFIG_HMEM_REPORTING
165 #define ACCESS_ATTR(name)						\
166 static ssize_t name##_show(struct device *dev,				\
167 			   struct device_attribute *attr,		\
168 			   char *buf)					\
169 {									\
170 	return sysfs_emit(buf, "%u\n",					\
171 			  to_access_nodes(dev)->hmem_attrs.name);	\
172 }									\
173 static DEVICE_ATTR_RO(name)
174 
175 ACCESS_ATTR(read_bandwidth);
176 ACCESS_ATTR(read_latency);
177 ACCESS_ATTR(write_bandwidth);
178 ACCESS_ATTR(write_latency);
179 
180 static struct attribute *access_attrs[] = {
181 	&dev_attr_read_bandwidth.attr,
182 	&dev_attr_read_latency.attr,
183 	&dev_attr_write_bandwidth.attr,
184 	&dev_attr_write_latency.attr,
185 	NULL,
186 };
187 
188 /**
189  * node_set_perf_attrs - Set the performance values for given access class
190  * @nid: Node identifier to be set
191  * @hmem_attrs: Heterogeneous memory performance attributes
192  * @access: The access class the for the given attributes
193  */
194 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
195 			 unsigned int access)
196 {
197 	struct node_access_nodes *c;
198 	struct node *node;
199 	int i;
200 
201 	if (WARN_ON_ONCE(!node_online(nid)))
202 		return;
203 
204 	node = node_devices[nid];
205 	c = node_init_node_access(node, access);
206 	if (!c)
207 		return;
208 
209 	c->hmem_attrs = *hmem_attrs;
210 	for (i = 0; access_attrs[i] != NULL; i++) {
211 		if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
212 					    "initiators")) {
213 			pr_info("failed to add performance attribute to node %d\n",
214 				nid);
215 			break;
216 		}
217 	}
218 }
219 
220 /**
221  * struct node_cache_info - Internal tracking for memory node caches
222  * @dev:	Device represeting the cache level
223  * @node:	List element for tracking in the node
224  * @cache_attrs:Attributes for this cache level
225  */
226 struct node_cache_info {
227 	struct device dev;
228 	struct list_head node;
229 	struct node_cache_attrs cache_attrs;
230 };
231 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
232 
233 #define CACHE_ATTR(name, fmt) 						\
234 static ssize_t name##_show(struct device *dev,				\
235 			   struct device_attribute *attr,		\
236 			   char *buf)					\
237 {									\
238 	return sysfs_emit(buf, fmt "\n",				\
239 			  to_cache_info(dev)->cache_attrs.name);	\
240 }									\
241 static DEVICE_ATTR_RO(name);
242 
243 CACHE_ATTR(size, "%llu")
244 CACHE_ATTR(line_size, "%u")
245 CACHE_ATTR(indexing, "%u")
246 CACHE_ATTR(write_policy, "%u")
247 
248 static struct attribute *cache_attrs[] = {
249 	&dev_attr_indexing.attr,
250 	&dev_attr_size.attr,
251 	&dev_attr_line_size.attr,
252 	&dev_attr_write_policy.attr,
253 	NULL,
254 };
255 ATTRIBUTE_GROUPS(cache);
256 
257 static void node_cache_release(struct device *dev)
258 {
259 	kfree(dev);
260 }
261 
262 static void node_cacheinfo_release(struct device *dev)
263 {
264 	struct node_cache_info *info = to_cache_info(dev);
265 	kfree(info);
266 }
267 
268 static void node_init_cache_dev(struct node *node)
269 {
270 	struct device *dev;
271 
272 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
273 	if (!dev)
274 		return;
275 
276 	device_initialize(dev);
277 	dev->parent = &node->dev;
278 	dev->release = node_cache_release;
279 	if (dev_set_name(dev, "memory_side_cache"))
280 		goto put_device;
281 
282 	if (device_add(dev))
283 		goto put_device;
284 
285 	pm_runtime_no_callbacks(dev);
286 	node->cache_dev = dev;
287 	return;
288 put_device:
289 	put_device(dev);
290 }
291 
292 /**
293  * node_add_cache() - add cache attribute to a memory node
294  * @nid: Node identifier that has new cache attributes
295  * @cache_attrs: Attributes for the cache being added
296  */
297 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
298 {
299 	struct node_cache_info *info;
300 	struct device *dev;
301 	struct node *node;
302 
303 	if (!node_online(nid) || !node_devices[nid])
304 		return;
305 
306 	node = node_devices[nid];
307 	list_for_each_entry(info, &node->cache_attrs, node) {
308 		if (info->cache_attrs.level == cache_attrs->level) {
309 			dev_warn(&node->dev,
310 				"attempt to add duplicate cache level:%d\n",
311 				cache_attrs->level);
312 			return;
313 		}
314 	}
315 
316 	if (!node->cache_dev)
317 		node_init_cache_dev(node);
318 	if (!node->cache_dev)
319 		return;
320 
321 	info = kzalloc(sizeof(*info), GFP_KERNEL);
322 	if (!info)
323 		return;
324 
325 	dev = &info->dev;
326 	device_initialize(dev);
327 	dev->parent = node->cache_dev;
328 	dev->release = node_cacheinfo_release;
329 	dev->groups = cache_groups;
330 	if (dev_set_name(dev, "index%d", cache_attrs->level))
331 		goto put_device;
332 
333 	info->cache_attrs = *cache_attrs;
334 	if (device_add(dev)) {
335 		dev_warn(&node->dev, "failed to add cache level:%d\n",
336 			 cache_attrs->level);
337 		goto put_device;
338 	}
339 	pm_runtime_no_callbacks(dev);
340 	list_add_tail(&info->node, &node->cache_attrs);
341 	return;
342 put_device:
343 	put_device(dev);
344 }
345 
346 static void node_remove_caches(struct node *node)
347 {
348 	struct node_cache_info *info, *next;
349 
350 	if (!node->cache_dev)
351 		return;
352 
353 	list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
354 		list_del(&info->node);
355 		device_unregister(&info->dev);
356 	}
357 	device_unregister(node->cache_dev);
358 }
359 
360 static void node_init_caches(unsigned int nid)
361 {
362 	INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
363 }
364 #else
365 static void node_init_caches(unsigned int nid) { }
366 static void node_remove_caches(struct node *node) { }
367 #endif
368 
369 #define K(x) ((x) << (PAGE_SHIFT - 10))
370 static ssize_t node_read_meminfo(struct device *dev,
371 			struct device_attribute *attr, char *buf)
372 {
373 	int len = 0;
374 	int nid = dev->id;
375 	struct pglist_data *pgdat = NODE_DATA(nid);
376 	struct sysinfo i;
377 	unsigned long sreclaimable, sunreclaimable;
378 	unsigned long swapcached = 0;
379 
380 	si_meminfo_node(&i, nid);
381 	sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
382 	sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
383 #ifdef CONFIG_SWAP
384 	swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
385 #endif
386 	len = sysfs_emit_at(buf, len,
387 			    "Node %d MemTotal:       %8lu kB\n"
388 			    "Node %d MemFree:        %8lu kB\n"
389 			    "Node %d MemUsed:        %8lu kB\n"
390 			    "Node %d SwapCached:     %8lu kB\n"
391 			    "Node %d Active:         %8lu kB\n"
392 			    "Node %d Inactive:       %8lu kB\n"
393 			    "Node %d Active(anon):   %8lu kB\n"
394 			    "Node %d Inactive(anon): %8lu kB\n"
395 			    "Node %d Active(file):   %8lu kB\n"
396 			    "Node %d Inactive(file): %8lu kB\n"
397 			    "Node %d Unevictable:    %8lu kB\n"
398 			    "Node %d Mlocked:        %8lu kB\n",
399 			    nid, K(i.totalram),
400 			    nid, K(i.freeram),
401 			    nid, K(i.totalram - i.freeram),
402 			    nid, K(swapcached),
403 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
404 				   node_page_state(pgdat, NR_ACTIVE_FILE)),
405 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
406 				   node_page_state(pgdat, NR_INACTIVE_FILE)),
407 			    nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
408 			    nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
409 			    nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
410 			    nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
411 			    nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
412 			    nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
413 
414 #ifdef CONFIG_HIGHMEM
415 	len += sysfs_emit_at(buf, len,
416 			     "Node %d HighTotal:      %8lu kB\n"
417 			     "Node %d HighFree:       %8lu kB\n"
418 			     "Node %d LowTotal:       %8lu kB\n"
419 			     "Node %d LowFree:        %8lu kB\n",
420 			     nid, K(i.totalhigh),
421 			     nid, K(i.freehigh),
422 			     nid, K(i.totalram - i.totalhigh),
423 			     nid, K(i.freeram - i.freehigh));
424 #endif
425 	len += sysfs_emit_at(buf, len,
426 			     "Node %d Dirty:          %8lu kB\n"
427 			     "Node %d Writeback:      %8lu kB\n"
428 			     "Node %d FilePages:      %8lu kB\n"
429 			     "Node %d Mapped:         %8lu kB\n"
430 			     "Node %d AnonPages:      %8lu kB\n"
431 			     "Node %d Shmem:          %8lu kB\n"
432 			     "Node %d KernelStack:    %8lu kB\n"
433 #ifdef CONFIG_SHADOW_CALL_STACK
434 			     "Node %d ShadowCallStack:%8lu kB\n"
435 #endif
436 			     "Node %d PageTables:     %8lu kB\n"
437 			     "Node %d SecPageTables:  %8lu kB\n"
438 			     "Node %d NFS_Unstable:   %8lu kB\n"
439 			     "Node %d Bounce:         %8lu kB\n"
440 			     "Node %d WritebackTmp:   %8lu kB\n"
441 			     "Node %d KReclaimable:   %8lu kB\n"
442 			     "Node %d Slab:           %8lu kB\n"
443 			     "Node %d SReclaimable:   %8lu kB\n"
444 			     "Node %d SUnreclaim:     %8lu kB\n"
445 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
446 			     "Node %d AnonHugePages:  %8lu kB\n"
447 			     "Node %d ShmemHugePages: %8lu kB\n"
448 			     "Node %d ShmemPmdMapped: %8lu kB\n"
449 			     "Node %d FileHugePages: %8lu kB\n"
450 			     "Node %d FilePmdMapped: %8lu kB\n"
451 #endif
452 			     ,
453 			     nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
454 			     nid, K(node_page_state(pgdat, NR_WRITEBACK)),
455 			     nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
456 			     nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
457 			     nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
458 			     nid, K(i.sharedram),
459 			     nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
460 #ifdef CONFIG_SHADOW_CALL_STACK
461 			     nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
462 #endif
463 			     nid, K(node_page_state(pgdat, NR_PAGETABLE)),
464 			     nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
465 			     nid, 0UL,
466 			     nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
467 			     nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
468 			     nid, K(sreclaimable +
469 				    node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
470 			     nid, K(sreclaimable + sunreclaimable),
471 			     nid, K(sreclaimable),
472 			     nid, K(sunreclaimable)
473 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
474 			     ,
475 			     nid, K(node_page_state(pgdat, NR_ANON_THPS)),
476 			     nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
477 			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
478 			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
479 			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
480 #endif
481 			    );
482 	len += hugetlb_report_node_meminfo(buf, len, nid);
483 	return len;
484 }
485 
486 #undef K
487 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
488 
489 static ssize_t node_read_numastat(struct device *dev,
490 				  struct device_attribute *attr, char *buf)
491 {
492 	fold_vm_numa_events();
493 	return sysfs_emit(buf,
494 			  "numa_hit %lu\n"
495 			  "numa_miss %lu\n"
496 			  "numa_foreign %lu\n"
497 			  "interleave_hit %lu\n"
498 			  "local_node %lu\n"
499 			  "other_node %lu\n",
500 			  sum_zone_numa_event_state(dev->id, NUMA_HIT),
501 			  sum_zone_numa_event_state(dev->id, NUMA_MISS),
502 			  sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
503 			  sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
504 			  sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
505 			  sum_zone_numa_event_state(dev->id, NUMA_OTHER));
506 }
507 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
508 
509 static ssize_t node_read_vmstat(struct device *dev,
510 				struct device_attribute *attr, char *buf)
511 {
512 	int nid = dev->id;
513 	struct pglist_data *pgdat = NODE_DATA(nid);
514 	int i;
515 	int len = 0;
516 
517 	for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
518 		len += sysfs_emit_at(buf, len, "%s %lu\n",
519 				     zone_stat_name(i),
520 				     sum_zone_node_page_state(nid, i));
521 
522 #ifdef CONFIG_NUMA
523 	fold_vm_numa_events();
524 	for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
525 		len += sysfs_emit_at(buf, len, "%s %lu\n",
526 				     numa_stat_name(i),
527 				     sum_zone_numa_event_state(nid, i));
528 
529 #endif
530 	for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
531 		unsigned long pages = node_page_state_pages(pgdat, i);
532 
533 		if (vmstat_item_print_in_thp(i))
534 			pages /= HPAGE_PMD_NR;
535 		len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
536 				     pages);
537 	}
538 
539 	return len;
540 }
541 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
542 
543 static ssize_t node_read_distance(struct device *dev,
544 				  struct device_attribute *attr, char *buf)
545 {
546 	int nid = dev->id;
547 	int len = 0;
548 	int i;
549 
550 	/*
551 	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
552 	 * at the most (distance + space or newline).
553 	 */
554 	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
555 
556 	for_each_online_node(i) {
557 		len += sysfs_emit_at(buf, len, "%s%d",
558 				     i ? " " : "", node_distance(nid, i));
559 	}
560 
561 	len += sysfs_emit_at(buf, len, "\n");
562 	return len;
563 }
564 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
565 
566 static struct attribute *node_dev_attrs[] = {
567 	&dev_attr_meminfo.attr,
568 	&dev_attr_numastat.attr,
569 	&dev_attr_distance.attr,
570 	&dev_attr_vmstat.attr,
571 	NULL
572 };
573 
574 static struct bin_attribute *node_dev_bin_attrs[] = {
575 	&bin_attr_cpumap,
576 	&bin_attr_cpulist,
577 	NULL
578 };
579 
580 static const struct attribute_group node_dev_group = {
581 	.attrs = node_dev_attrs,
582 	.bin_attrs = node_dev_bin_attrs
583 };
584 
585 static const struct attribute_group *node_dev_groups[] = {
586 	&node_dev_group,
587 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
588 	&arch_node_dev_group,
589 #endif
590 #ifdef CONFIG_MEMORY_FAILURE
591 	&memory_failure_attr_group,
592 #endif
593 	NULL
594 };
595 
596 static void node_device_release(struct device *dev)
597 {
598 	kfree(to_node(dev));
599 }
600 
601 /*
602  * register_node - Setup a sysfs device for a node.
603  * @num - Node number to use when creating the device.
604  *
605  * Initialize and register the node device.
606  */
607 static int register_node(struct node *node, int num)
608 {
609 	int error;
610 
611 	node->dev.id = num;
612 	node->dev.bus = &node_subsys;
613 	node->dev.release = node_device_release;
614 	node->dev.groups = node_dev_groups;
615 	error = device_register(&node->dev);
616 
617 	if (error) {
618 		put_device(&node->dev);
619 	} else {
620 		hugetlb_register_node(node);
621 		compaction_register_node(node);
622 	}
623 
624 	return error;
625 }
626 
627 /**
628  * unregister_node - unregister a node device
629  * @node: node going away
630  *
631  * Unregisters a node device @node.  All the devices on the node must be
632  * unregistered before calling this function.
633  */
634 void unregister_node(struct node *node)
635 {
636 	hugetlb_unregister_node(node);
637 	compaction_unregister_node(node);
638 	node_remove_accesses(node);
639 	node_remove_caches(node);
640 	device_unregister(&node->dev);
641 }
642 
643 struct node *node_devices[MAX_NUMNODES];
644 
645 /*
646  * register cpu under node
647  */
648 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
649 {
650 	int ret;
651 	struct device *obj;
652 
653 	if (!node_online(nid))
654 		return 0;
655 
656 	obj = get_cpu_device(cpu);
657 	if (!obj)
658 		return 0;
659 
660 	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
661 				&obj->kobj,
662 				kobject_name(&obj->kobj));
663 	if (ret)
664 		return ret;
665 
666 	return sysfs_create_link(&obj->kobj,
667 				 &node_devices[nid]->dev.kobj,
668 				 kobject_name(&node_devices[nid]->dev.kobj));
669 }
670 
671 /**
672  * register_memory_node_under_compute_node - link memory node to its compute
673  *					     node for a given access class.
674  * @mem_nid:	Memory node number
675  * @cpu_nid:	Cpu  node number
676  * @access:	Access class to register
677  *
678  * Description:
679  * 	For use with platforms that may have separate memory and compute nodes.
680  * 	This function will export node relationships linking which memory
681  * 	initiator nodes can access memory targets at a given ranked access
682  * 	class.
683  */
684 int register_memory_node_under_compute_node(unsigned int mem_nid,
685 					    unsigned int cpu_nid,
686 					    unsigned int access)
687 {
688 	struct node *init_node, *targ_node;
689 	struct node_access_nodes *initiator, *target;
690 	int ret;
691 
692 	if (!node_online(cpu_nid) || !node_online(mem_nid))
693 		return -ENODEV;
694 
695 	init_node = node_devices[cpu_nid];
696 	targ_node = node_devices[mem_nid];
697 	initiator = node_init_node_access(init_node, access);
698 	target = node_init_node_access(targ_node, access);
699 	if (!initiator || !target)
700 		return -ENOMEM;
701 
702 	ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
703 				      &targ_node->dev.kobj,
704 				      dev_name(&targ_node->dev));
705 	if (ret)
706 		return ret;
707 
708 	ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
709 				      &init_node->dev.kobj,
710 				      dev_name(&init_node->dev));
711 	if (ret)
712 		goto err;
713 
714 	return 0;
715  err:
716 	sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
717 				     dev_name(&targ_node->dev));
718 	return ret;
719 }
720 
721 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
722 {
723 	struct device *obj;
724 
725 	if (!node_online(nid))
726 		return 0;
727 
728 	obj = get_cpu_device(cpu);
729 	if (!obj)
730 		return 0;
731 
732 	sysfs_remove_link(&node_devices[nid]->dev.kobj,
733 			  kobject_name(&obj->kobj));
734 	sysfs_remove_link(&obj->kobj,
735 			  kobject_name(&node_devices[nid]->dev.kobj));
736 
737 	return 0;
738 }
739 
740 #ifdef CONFIG_MEMORY_HOTPLUG
741 static int __ref get_nid_for_pfn(unsigned long pfn)
742 {
743 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
744 	if (system_state < SYSTEM_RUNNING)
745 		return early_pfn_to_nid(pfn);
746 #endif
747 	return pfn_to_nid(pfn);
748 }
749 
750 static void do_register_memory_block_under_node(int nid,
751 						struct memory_block *mem_blk,
752 						enum meminit_context context)
753 {
754 	int ret;
755 
756 	memory_block_add_nid(mem_blk, nid, context);
757 
758 	ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
759 				       &mem_blk->dev.kobj,
760 				       kobject_name(&mem_blk->dev.kobj));
761 	if (ret && ret != -EEXIST)
762 		dev_err_ratelimited(&node_devices[nid]->dev,
763 				    "can't create link to %s in sysfs (%d)\n",
764 				    kobject_name(&mem_blk->dev.kobj), ret);
765 
766 	ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
767 				&node_devices[nid]->dev.kobj,
768 				kobject_name(&node_devices[nid]->dev.kobj));
769 	if (ret && ret != -EEXIST)
770 		dev_err_ratelimited(&mem_blk->dev,
771 				    "can't create link to %s in sysfs (%d)\n",
772 				    kobject_name(&node_devices[nid]->dev.kobj),
773 				    ret);
774 }
775 
776 /* register memory section under specified node if it spans that node */
777 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
778 					       void *arg)
779 {
780 	unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
781 	unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
782 	unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
783 	int nid = *(int *)arg;
784 	unsigned long pfn;
785 
786 	for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
787 		int page_nid;
788 
789 		/*
790 		 * memory block could have several absent sections from start.
791 		 * skip pfn range from absent section
792 		 */
793 		if (!pfn_in_present_section(pfn)) {
794 			pfn = round_down(pfn + PAGES_PER_SECTION,
795 					 PAGES_PER_SECTION) - 1;
796 			continue;
797 		}
798 
799 		/*
800 		 * We need to check if page belongs to nid only at the boot
801 		 * case because node's ranges can be interleaved.
802 		 */
803 		page_nid = get_nid_for_pfn(pfn);
804 		if (page_nid < 0)
805 			continue;
806 		if (page_nid != nid)
807 			continue;
808 
809 		do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
810 		return 0;
811 	}
812 	/* mem section does not span the specified node */
813 	return 0;
814 }
815 
816 /*
817  * During hotplug we know that all pages in the memory block belong to the same
818  * node.
819  */
820 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
821 						 void *arg)
822 {
823 	int nid = *(int *)arg;
824 
825 	do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
826 	return 0;
827 }
828 
829 /*
830  * Unregister a memory block device under the node it spans. Memory blocks
831  * with multiple nodes cannot be offlined and therefore also never be removed.
832  */
833 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
834 {
835 	if (mem_blk->nid == NUMA_NO_NODE)
836 		return;
837 
838 	sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
839 			  kobject_name(&mem_blk->dev.kobj));
840 	sysfs_remove_link(&mem_blk->dev.kobj,
841 			  kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
842 }
843 
844 void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
845 				       unsigned long end_pfn,
846 				       enum meminit_context context)
847 {
848 	walk_memory_blocks_func_t func;
849 
850 	if (context == MEMINIT_HOTPLUG)
851 		func = register_mem_block_under_node_hotplug;
852 	else
853 		func = register_mem_block_under_node_early;
854 
855 	walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
856 			   (void *)&nid, func);
857 	return;
858 }
859 #endif /* CONFIG_MEMORY_HOTPLUG */
860 
861 int __register_one_node(int nid)
862 {
863 	int error;
864 	int cpu;
865 
866 	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
867 	if (!node_devices[nid])
868 		return -ENOMEM;
869 
870 	error = register_node(node_devices[nid], nid);
871 
872 	/* link cpu under this node */
873 	for_each_present_cpu(cpu) {
874 		if (cpu_to_node(cpu) == nid)
875 			register_cpu_under_node(cpu, nid);
876 	}
877 
878 	INIT_LIST_HEAD(&node_devices[nid]->access_list);
879 	node_init_caches(nid);
880 
881 	return error;
882 }
883 
884 void unregister_one_node(int nid)
885 {
886 	if (!node_devices[nid])
887 		return;
888 
889 	unregister_node(node_devices[nid]);
890 	node_devices[nid] = NULL;
891 }
892 
893 /*
894  * node states attributes
895  */
896 
897 struct node_attr {
898 	struct device_attribute attr;
899 	enum node_states state;
900 };
901 
902 static ssize_t show_node_state(struct device *dev,
903 			       struct device_attribute *attr, char *buf)
904 {
905 	struct node_attr *na = container_of(attr, struct node_attr, attr);
906 
907 	return sysfs_emit(buf, "%*pbl\n",
908 			  nodemask_pr_args(&node_states[na->state]));
909 }
910 
911 #define _NODE_ATTR(name, state) \
912 	{ __ATTR(name, 0444, show_node_state, NULL), state }
913 
914 static struct node_attr node_state_attr[] = {
915 	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
916 	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
917 	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
918 #ifdef CONFIG_HIGHMEM
919 	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
920 #endif
921 	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
922 	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
923 	[N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
924 					   N_GENERIC_INITIATOR),
925 };
926 
927 static struct attribute *node_state_attrs[] = {
928 	&node_state_attr[N_POSSIBLE].attr.attr,
929 	&node_state_attr[N_ONLINE].attr.attr,
930 	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
931 #ifdef CONFIG_HIGHMEM
932 	&node_state_attr[N_HIGH_MEMORY].attr.attr,
933 #endif
934 	&node_state_attr[N_MEMORY].attr.attr,
935 	&node_state_attr[N_CPU].attr.attr,
936 	&node_state_attr[N_GENERIC_INITIATOR].attr.attr,
937 	NULL
938 };
939 
940 static const struct attribute_group memory_root_attr_group = {
941 	.attrs = node_state_attrs,
942 };
943 
944 static const struct attribute_group *cpu_root_attr_groups[] = {
945 	&memory_root_attr_group,
946 	NULL,
947 };
948 
949 void __init node_dev_init(void)
950 {
951 	int ret, i;
952 
953  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
954  	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
955 
956 	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
957 	if (ret)
958 		panic("%s() failed to register subsystem: %d\n", __func__, ret);
959 
960 	/*
961 	 * Create all node devices, which will properly link the node
962 	 * to applicable memory block devices and already created cpu devices.
963 	 */
964 	for_each_online_node(i) {
965 		ret = register_one_node(i);
966 		if (ret)
967 			panic("%s() failed to add node: %d\n", __func__, ret);
968 	}
969 }
970