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/mempolicy.h>
11 #include <linux/vmstat.h>
12 #include <linux/notifier.h>
13 #include <linux/node.h>
14 #include <linux/hugetlb.h>
15 #include <linux/compaction.h>
16 #include <linux/cpumask.h>
17 #include <linux/topology.h>
18 #include <linux/nodemask.h>
19 #include <linux/cpu.h>
20 #include <linux/device.h>
21 #include <linux/pm_runtime.h>
22 #include <linux/swap.h>
23 #include <linux/slab.h>
24 #include <linux/memblock.h>
25
26 static const struct bus_type node_subsys = {
27 .name = "node",
28 .dev_name = "node",
29 };
30
cpumap_read(struct file * file,struct kobject * kobj,const struct bin_attribute * attr,char * buf,loff_t off,size_t count)31 static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
32 const struct bin_attribute *attr, char *buf,
33 loff_t off, size_t count)
34 {
35 struct device *dev = kobj_to_dev(kobj);
36 struct node *node_dev = to_node(dev);
37 cpumask_var_t mask;
38 ssize_t n;
39
40 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
41 return 0;
42
43 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
44 n = cpumap_print_bitmask_to_buf(buf, mask, off, count);
45 free_cpumask_var(mask);
46
47 return n;
48 }
49
50 static const BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
51
cpulist_read(struct file * file,struct kobject * kobj,const struct bin_attribute * attr,char * buf,loff_t off,size_t count)52 static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
53 const struct bin_attribute *attr, char *buf,
54 loff_t off, size_t count)
55 {
56 struct device *dev = kobj_to_dev(kobj);
57 struct node *node_dev = to_node(dev);
58 cpumask_var_t mask;
59 ssize_t n;
60
61 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
62 return 0;
63
64 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
65 n = cpumap_print_list_to_buf(buf, mask, off, count);
66 free_cpumask_var(mask);
67
68 return n;
69 }
70
71 static const BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
72
73 /**
74 * struct node_access_nodes - Access class device to hold user visible
75 * relationships to other nodes.
76 * @dev: Device for this memory access class
77 * @list_node: List element in the node's access list
78 * @access: The access class rank
79 * @coord: Heterogeneous memory performance coordinates
80 */
81 struct node_access_nodes {
82 struct device dev;
83 struct list_head list_node;
84 unsigned int access;
85 #ifdef CONFIG_HMEM_REPORTING
86 struct access_coordinate coord;
87 #endif
88 };
89 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
90
91 static struct attribute *node_init_access_node_attrs[] = {
92 NULL,
93 };
94
95 static struct attribute *node_targ_access_node_attrs[] = {
96 NULL,
97 };
98
99 static const struct attribute_group initiators = {
100 .name = "initiators",
101 .attrs = node_init_access_node_attrs,
102 };
103
104 static const struct attribute_group targets = {
105 .name = "targets",
106 .attrs = node_targ_access_node_attrs,
107 };
108
109 static const struct attribute_group *node_access_node_groups[] = {
110 &initiators,
111 &targets,
112 NULL,
113 };
114
115 #ifdef CONFIG_MEMORY_HOTPLUG
116 static BLOCKING_NOTIFIER_HEAD(node_chain);
117
register_node_notifier(struct notifier_block * nb)118 int register_node_notifier(struct notifier_block *nb)
119 {
120 return blocking_notifier_chain_register(&node_chain, nb);
121 }
122 EXPORT_SYMBOL(register_node_notifier);
123
unregister_node_notifier(struct notifier_block * nb)124 void unregister_node_notifier(struct notifier_block *nb)
125 {
126 blocking_notifier_chain_unregister(&node_chain, nb);
127 }
128 EXPORT_SYMBOL(unregister_node_notifier);
129
node_notify(unsigned long val,void * v)130 int node_notify(unsigned long val, void *v)
131 {
132 return blocking_notifier_call_chain(&node_chain, val, v);
133 }
134 #endif
135
node_remove_accesses(struct node * node)136 static void node_remove_accesses(struct node *node)
137 {
138 struct node_access_nodes *c, *cnext;
139
140 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
141 list_del(&c->list_node);
142 device_unregister(&c->dev);
143 }
144 }
145
node_access_release(struct device * dev)146 static void node_access_release(struct device *dev)
147 {
148 kfree(to_access_nodes(dev));
149 }
150
node_init_node_access(struct node * node,enum access_coordinate_class access)151 static struct node_access_nodes *node_init_node_access(struct node *node,
152 enum access_coordinate_class access)
153 {
154 struct node_access_nodes *access_node;
155 struct device *dev;
156
157 list_for_each_entry(access_node, &node->access_list, list_node)
158 if (access_node->access == access)
159 return access_node;
160
161 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
162 if (!access_node)
163 return NULL;
164
165 access_node->access = access;
166 dev = &access_node->dev;
167 dev->parent = &node->dev;
168 dev->release = node_access_release;
169 dev->groups = node_access_node_groups;
170 if (dev_set_name(dev, "access%u", access))
171 goto free;
172
173 if (device_register(dev))
174 goto free_name;
175
176 pm_runtime_no_callbacks(dev);
177 list_add_tail(&access_node->list_node, &node->access_list);
178 return access_node;
179 free_name:
180 kfree_const(dev->kobj.name);
181 free:
182 kfree(access_node);
183 return NULL;
184 }
185
186 #ifdef CONFIG_HMEM_REPORTING
187 #define ACCESS_ATTR(property) \
188 static ssize_t property##_show(struct device *dev, \
189 struct device_attribute *attr, \
190 char *buf) \
191 { \
192 return sysfs_emit(buf, "%u\n", \
193 to_access_nodes(dev)->coord.property); \
194 } \
195 static DEVICE_ATTR_RO(property)
196
197 ACCESS_ATTR(read_bandwidth);
198 ACCESS_ATTR(read_latency);
199 ACCESS_ATTR(write_bandwidth);
200 ACCESS_ATTR(write_latency);
201
202 static struct attribute *access_attrs[] = {
203 &dev_attr_read_bandwidth.attr,
204 &dev_attr_read_latency.attr,
205 &dev_attr_write_bandwidth.attr,
206 &dev_attr_write_latency.attr,
207 NULL,
208 };
209
210 /**
211 * node_set_perf_attrs - Set the performance values for given access class
212 * @nid: Node identifier to be set
213 * @coord: Heterogeneous memory performance coordinates
214 * @access: The access class the for the given attributes
215 */
node_set_perf_attrs(unsigned int nid,struct access_coordinate * coord,enum access_coordinate_class access)216 void node_set_perf_attrs(unsigned int nid, struct access_coordinate *coord,
217 enum access_coordinate_class access)
218 {
219 struct node_access_nodes *c;
220 struct node *node;
221 int i;
222
223 if (WARN_ON_ONCE(!node_online(nid)))
224 return;
225
226 node = node_devices[nid];
227 c = node_init_node_access(node, access);
228 if (!c)
229 return;
230
231 c->coord = *coord;
232 for (i = 0; access_attrs[i] != NULL; i++) {
233 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
234 "initiators")) {
235 pr_info("failed to add performance attribute to node %d\n",
236 nid);
237 break;
238 }
239 }
240
241 /* When setting CPU access coordinates, update mempolicy */
242 if (access == ACCESS_COORDINATE_CPU) {
243 if (mempolicy_set_node_perf(nid, coord)) {
244 pr_info("failed to set mempolicy attrs for node %d\n",
245 nid);
246 }
247 }
248 }
249 EXPORT_SYMBOL_GPL(node_set_perf_attrs);
250
251 /**
252 * struct node_cache_info - Internal tracking for memory node caches
253 * @dev: Device represeting the cache level
254 * @node: List element for tracking in the node
255 * @cache_attrs:Attributes for this cache level
256 */
257 struct node_cache_info {
258 struct device dev;
259 struct list_head node;
260 struct node_cache_attrs cache_attrs;
261 };
262 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
263
264 #define CACHE_ATTR(name, fmt) \
265 static ssize_t name##_show(struct device *dev, \
266 struct device_attribute *attr, \
267 char *buf) \
268 { \
269 return sysfs_emit(buf, fmt "\n", \
270 to_cache_info(dev)->cache_attrs.name); \
271 } \
272 static DEVICE_ATTR_RO(name);
273
274 CACHE_ATTR(size, "%llu")
275 CACHE_ATTR(line_size, "%u")
276 CACHE_ATTR(indexing, "%u")
277 CACHE_ATTR(write_policy, "%u")
278 CACHE_ATTR(address_mode, "%#x")
279
280 static struct attribute *cache_attrs[] = {
281 &dev_attr_indexing.attr,
282 &dev_attr_size.attr,
283 &dev_attr_line_size.attr,
284 &dev_attr_write_policy.attr,
285 &dev_attr_address_mode.attr,
286 NULL,
287 };
288 ATTRIBUTE_GROUPS(cache);
289
node_cache_release(struct device * dev)290 static void node_cache_release(struct device *dev)
291 {
292 kfree(dev);
293 }
294
node_cacheinfo_release(struct device * dev)295 static void node_cacheinfo_release(struct device *dev)
296 {
297 struct node_cache_info *info = to_cache_info(dev);
298 kfree(info);
299 }
300
node_init_cache_dev(struct node * node)301 static void node_init_cache_dev(struct node *node)
302 {
303 struct device *dev;
304
305 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
306 if (!dev)
307 return;
308
309 device_initialize(dev);
310 dev->parent = &node->dev;
311 dev->release = node_cache_release;
312 if (dev_set_name(dev, "memory_side_cache"))
313 goto put_device;
314
315 if (device_add(dev))
316 goto put_device;
317
318 pm_runtime_no_callbacks(dev);
319 node->cache_dev = dev;
320 return;
321 put_device:
322 put_device(dev);
323 }
324
325 /**
326 * node_add_cache() - add cache attribute to a memory node
327 * @nid: Node identifier that has new cache attributes
328 * @cache_attrs: Attributes for the cache being added
329 */
node_add_cache(unsigned int nid,struct node_cache_attrs * cache_attrs)330 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
331 {
332 struct node_cache_info *info;
333 struct device *dev;
334 struct node *node;
335
336 if (!node_online(nid) || !node_devices[nid])
337 return;
338
339 node = node_devices[nid];
340 list_for_each_entry(info, &node->cache_attrs, node) {
341 if (info->cache_attrs.level == cache_attrs->level) {
342 dev_warn(&node->dev,
343 "attempt to add duplicate cache level:%d\n",
344 cache_attrs->level);
345 return;
346 }
347 }
348
349 if (!node->cache_dev)
350 node_init_cache_dev(node);
351 if (!node->cache_dev)
352 return;
353
354 info = kzalloc(sizeof(*info), GFP_KERNEL);
355 if (!info)
356 return;
357
358 dev = &info->dev;
359 device_initialize(dev);
360 dev->parent = node->cache_dev;
361 dev->release = node_cacheinfo_release;
362 dev->groups = cache_groups;
363 if (dev_set_name(dev, "index%d", cache_attrs->level))
364 goto put_device;
365
366 info->cache_attrs = *cache_attrs;
367 if (device_add(dev)) {
368 dev_warn(&node->dev, "failed to add cache level:%d\n",
369 cache_attrs->level);
370 goto put_device;
371 }
372 pm_runtime_no_callbacks(dev);
373 list_add_tail(&info->node, &node->cache_attrs);
374 return;
375 put_device:
376 put_device(dev);
377 }
378
node_remove_caches(struct node * node)379 static void node_remove_caches(struct node *node)
380 {
381 struct node_cache_info *info, *next;
382
383 if (!node->cache_dev)
384 return;
385
386 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
387 list_del(&info->node);
388 device_unregister(&info->dev);
389 }
390 device_unregister(node->cache_dev);
391 }
392
node_init_caches(unsigned int nid)393 static void node_init_caches(unsigned int nid)
394 {
395 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
396 }
397 #else
node_init_caches(unsigned int nid)398 static void node_init_caches(unsigned int nid) { }
node_remove_caches(struct node * node)399 static void node_remove_caches(struct node *node) { }
400 #endif
401
402 #define K(x) ((x) << (PAGE_SHIFT - 10))
node_read_meminfo(struct device * dev,struct device_attribute * attr,char * buf)403 static ssize_t node_read_meminfo(struct device *dev,
404 struct device_attribute *attr, char *buf)
405 {
406 int len = 0;
407 int nid = dev->id;
408 struct pglist_data *pgdat = NODE_DATA(nid);
409 struct sysinfo i;
410 unsigned long sreclaimable, sunreclaimable;
411 unsigned long swapcached = 0;
412
413 si_meminfo_node(&i, nid);
414 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
415 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
416 #ifdef CONFIG_SWAP
417 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
418 #endif
419 len = sysfs_emit_at(buf, len,
420 "Node %d MemTotal: %8lu kB\n"
421 "Node %d MemFree: %8lu kB\n"
422 "Node %d MemUsed: %8lu kB\n"
423 "Node %d SwapCached: %8lu kB\n"
424 "Node %d Active: %8lu kB\n"
425 "Node %d Inactive: %8lu kB\n"
426 "Node %d Active(anon): %8lu kB\n"
427 "Node %d Inactive(anon): %8lu kB\n"
428 "Node %d Active(file): %8lu kB\n"
429 "Node %d Inactive(file): %8lu kB\n"
430 "Node %d Unevictable: %8lu kB\n"
431 "Node %d Mlocked: %8lu kB\n",
432 nid, K(i.totalram),
433 nid, K(i.freeram),
434 nid, K(i.totalram - i.freeram),
435 nid, K(swapcached),
436 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
437 node_page_state(pgdat, NR_ACTIVE_FILE)),
438 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
439 node_page_state(pgdat, NR_INACTIVE_FILE)),
440 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
441 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
442 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
443 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
444 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
445 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
446
447 #ifdef CONFIG_HIGHMEM
448 len += sysfs_emit_at(buf, len,
449 "Node %d HighTotal: %8lu kB\n"
450 "Node %d HighFree: %8lu kB\n"
451 "Node %d LowTotal: %8lu kB\n"
452 "Node %d LowFree: %8lu kB\n",
453 nid, K(i.totalhigh),
454 nid, K(i.freehigh),
455 nid, K(i.totalram - i.totalhigh),
456 nid, K(i.freeram - i.freehigh));
457 #endif
458 len += sysfs_emit_at(buf, len,
459 "Node %d Dirty: %8lu kB\n"
460 "Node %d Writeback: %8lu kB\n"
461 "Node %d FilePages: %8lu kB\n"
462 "Node %d Mapped: %8lu kB\n"
463 "Node %d AnonPages: %8lu kB\n"
464 "Node %d Shmem: %8lu kB\n"
465 "Node %d KernelStack: %8lu kB\n"
466 #ifdef CONFIG_SHADOW_CALL_STACK
467 "Node %d ShadowCallStack:%8lu kB\n"
468 #endif
469 "Node %d PageTables: %8lu kB\n"
470 "Node %d SecPageTables: %8lu kB\n"
471 "Node %d NFS_Unstable: %8lu kB\n"
472 "Node %d Bounce: %8lu kB\n"
473 "Node %d WritebackTmp: %8lu kB\n"
474 "Node %d KReclaimable: %8lu kB\n"
475 "Node %d Slab: %8lu kB\n"
476 "Node %d SReclaimable: %8lu kB\n"
477 "Node %d SUnreclaim: %8lu kB\n"
478 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
479 "Node %d AnonHugePages: %8lu kB\n"
480 "Node %d ShmemHugePages: %8lu kB\n"
481 "Node %d ShmemPmdMapped: %8lu kB\n"
482 "Node %d FileHugePages: %8lu kB\n"
483 "Node %d FilePmdMapped: %8lu kB\n"
484 #endif
485 #ifdef CONFIG_UNACCEPTED_MEMORY
486 "Node %d Unaccepted: %8lu kB\n"
487 #endif
488 ,
489 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
490 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
491 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
492 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
493 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
494 nid, K(i.sharedram),
495 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
496 #ifdef CONFIG_SHADOW_CALL_STACK
497 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
498 #endif
499 nid, K(node_page_state(pgdat, NR_PAGETABLE)),
500 nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
501 nid, 0UL,
502 nid, 0UL,
503 nid, 0UL,
504 nid, K(sreclaimable +
505 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
506 nid, K(sreclaimable + sunreclaimable),
507 nid, K(sreclaimable),
508 nid, K(sunreclaimable)
509 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
510 ,
511 nid, K(node_page_state(pgdat, NR_ANON_THPS)),
512 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
513 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
514 nid, K(node_page_state(pgdat, NR_FILE_THPS)),
515 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
516 #endif
517 #ifdef CONFIG_UNACCEPTED_MEMORY
518 ,
519 nid, K(sum_zone_node_page_state(nid, NR_UNACCEPTED))
520 #endif
521 );
522 len += hugetlb_report_node_meminfo(buf, len, nid);
523 return len;
524 }
525
526 #undef K
527 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
528
node_read_numastat(struct device * dev,struct device_attribute * attr,char * buf)529 static ssize_t node_read_numastat(struct device *dev,
530 struct device_attribute *attr, char *buf)
531 {
532 fold_vm_numa_events();
533 return sysfs_emit(buf,
534 "numa_hit %lu\n"
535 "numa_miss %lu\n"
536 "numa_foreign %lu\n"
537 "interleave_hit %lu\n"
538 "local_node %lu\n"
539 "other_node %lu\n",
540 sum_zone_numa_event_state(dev->id, NUMA_HIT),
541 sum_zone_numa_event_state(dev->id, NUMA_MISS),
542 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
543 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
544 sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
545 sum_zone_numa_event_state(dev->id, NUMA_OTHER));
546 }
547 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
548
node_read_vmstat(struct device * dev,struct device_attribute * attr,char * buf)549 static ssize_t node_read_vmstat(struct device *dev,
550 struct device_attribute *attr, char *buf)
551 {
552 int nid = dev->id;
553 struct pglist_data *pgdat = NODE_DATA(nid);
554 int i;
555 int len = 0;
556
557 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
558 len += sysfs_emit_at(buf, len, "%s %lu\n",
559 zone_stat_name(i),
560 sum_zone_node_page_state(nid, i));
561
562 #ifdef CONFIG_NUMA
563 fold_vm_numa_events();
564 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
565 len += sysfs_emit_at(buf, len, "%s %lu\n",
566 numa_stat_name(i),
567 sum_zone_numa_event_state(nid, i));
568
569 #endif
570 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
571 unsigned long pages = node_page_state_pages(pgdat, i);
572
573 if (vmstat_item_print_in_thp(i))
574 pages /= HPAGE_PMD_NR;
575 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
576 pages);
577 }
578
579 return len;
580 }
581 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
582
node_read_distance(struct device * dev,struct device_attribute * attr,char * buf)583 static ssize_t node_read_distance(struct device *dev,
584 struct device_attribute *attr, char *buf)
585 {
586 int nid = dev->id;
587 int len = 0;
588 int i;
589
590 /*
591 * buf is currently PAGE_SIZE in length and each node needs 4 chars
592 * at the most (distance + space or newline).
593 */
594 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
595
596 for_each_online_node(i) {
597 len += sysfs_emit_at(buf, len, "%s%d",
598 i ? " " : "", node_distance(nid, i));
599 }
600
601 len += sysfs_emit_at(buf, len, "\n");
602 return len;
603 }
604 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
605
606 static struct attribute *node_dev_attrs[] = {
607 &dev_attr_meminfo.attr,
608 &dev_attr_numastat.attr,
609 &dev_attr_distance.attr,
610 &dev_attr_vmstat.attr,
611 NULL
612 };
613
614 static const struct bin_attribute *node_dev_bin_attrs[] = {
615 &bin_attr_cpumap,
616 &bin_attr_cpulist,
617 NULL
618 };
619
620 static const struct attribute_group node_dev_group = {
621 .attrs = node_dev_attrs,
622 .bin_attrs = node_dev_bin_attrs,
623 };
624
625 static const struct attribute_group *node_dev_groups[] = {
626 &node_dev_group,
627 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
628 &arch_node_dev_group,
629 #endif
630 #ifdef CONFIG_MEMORY_FAILURE
631 &memory_failure_attr_group,
632 #endif
633 NULL
634 };
635
node_device_release(struct device * dev)636 static void node_device_release(struct device *dev)
637 {
638 kfree(to_node(dev));
639 }
640
641 /*
642 * register_node - Setup a sysfs device for a node.
643 * @num - Node number to use when creating the device.
644 *
645 * Initialize and register the node device.
646 */
register_node(struct node * node,int num)647 static int register_node(struct node *node, int num)
648 {
649 int error;
650
651 node->dev.id = num;
652 node->dev.bus = &node_subsys;
653 node->dev.release = node_device_release;
654 node->dev.groups = node_dev_groups;
655 error = device_register(&node->dev);
656
657 if (error) {
658 put_device(&node->dev);
659 } else {
660 hugetlb_register_node(node);
661 compaction_register_node(node);
662 reclaim_register_node(node);
663 }
664
665 return error;
666 }
667
668 /**
669 * unregister_node - unregister a node device
670 * @node: node going away
671 *
672 * Unregisters a node device @node. All the devices on the node must be
673 * unregistered before calling this function.
674 */
unregister_node(struct node * node)675 void unregister_node(struct node *node)
676 {
677 hugetlb_unregister_node(node);
678 compaction_unregister_node(node);
679 reclaim_unregister_node(node);
680 node_remove_accesses(node);
681 node_remove_caches(node);
682 device_unregister(&node->dev);
683 }
684
685 struct node *node_devices[MAX_NUMNODES];
686
687 /*
688 * register cpu under node
689 */
register_cpu_under_node(unsigned int cpu,unsigned int nid)690 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
691 {
692 int ret;
693 struct device *obj;
694
695 if (!node_online(nid))
696 return 0;
697
698 obj = get_cpu_device(cpu);
699 if (!obj)
700 return 0;
701
702 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
703 &obj->kobj,
704 kobject_name(&obj->kobj));
705 if (ret)
706 return ret;
707
708 return sysfs_create_link(&obj->kobj,
709 &node_devices[nid]->dev.kobj,
710 kobject_name(&node_devices[nid]->dev.kobj));
711 }
712
713 /**
714 * register_memory_node_under_compute_node - link memory node to its compute
715 * node for a given access class.
716 * @mem_nid: Memory node number
717 * @cpu_nid: Cpu node number
718 * @access: Access class to register
719 *
720 * Description:
721 * For use with platforms that may have separate memory and compute nodes.
722 * This function will export node relationships linking which memory
723 * initiator nodes can access memory targets at a given ranked access
724 * class.
725 */
register_memory_node_under_compute_node(unsigned int mem_nid,unsigned int cpu_nid,enum access_coordinate_class access)726 int register_memory_node_under_compute_node(unsigned int mem_nid,
727 unsigned int cpu_nid,
728 enum access_coordinate_class access)
729 {
730 struct node *init_node, *targ_node;
731 struct node_access_nodes *initiator, *target;
732 int ret;
733
734 if (!node_online(cpu_nid) || !node_online(mem_nid))
735 return -ENODEV;
736
737 init_node = node_devices[cpu_nid];
738 targ_node = node_devices[mem_nid];
739 initiator = node_init_node_access(init_node, access);
740 target = node_init_node_access(targ_node, access);
741 if (!initiator || !target)
742 return -ENOMEM;
743
744 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
745 &targ_node->dev.kobj,
746 dev_name(&targ_node->dev));
747 if (ret)
748 return ret;
749
750 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
751 &init_node->dev.kobj,
752 dev_name(&init_node->dev));
753 if (ret)
754 goto err;
755
756 return 0;
757 err:
758 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
759 dev_name(&targ_node->dev));
760 return ret;
761 }
762
unregister_cpu_under_node(unsigned int cpu,unsigned int nid)763 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
764 {
765 struct device *obj;
766
767 if (!node_online(nid))
768 return 0;
769
770 obj = get_cpu_device(cpu);
771 if (!obj)
772 return 0;
773
774 sysfs_remove_link(&node_devices[nid]->dev.kobj,
775 kobject_name(&obj->kobj));
776 sysfs_remove_link(&obj->kobj,
777 kobject_name(&node_devices[nid]->dev.kobj));
778
779 return 0;
780 }
781
782 #ifdef CONFIG_MEMORY_HOTPLUG
do_register_memory_block_under_node(int nid,struct memory_block * mem_blk,enum meminit_context context)783 static void do_register_memory_block_under_node(int nid,
784 struct memory_block *mem_blk,
785 enum meminit_context context)
786 {
787 int ret;
788
789 memory_block_add_nid(mem_blk, nid, context);
790
791 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
792 &mem_blk->dev.kobj,
793 kobject_name(&mem_blk->dev.kobj));
794 if (ret && ret != -EEXIST)
795 dev_err_ratelimited(&node_devices[nid]->dev,
796 "can't create link to %s in sysfs (%d)\n",
797 kobject_name(&mem_blk->dev.kobj), ret);
798
799 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
800 &node_devices[nid]->dev.kobj,
801 kobject_name(&node_devices[nid]->dev.kobj));
802 if (ret && ret != -EEXIST)
803 dev_err_ratelimited(&mem_blk->dev,
804 "can't create link to %s in sysfs (%d)\n",
805 kobject_name(&node_devices[nid]->dev.kobj),
806 ret);
807 }
808
809 /*
810 * During hotplug we know that all pages in the memory block belong to the same
811 * node.
812 */
register_mem_block_under_node_hotplug(struct memory_block * mem_blk,void * arg)813 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
814 void *arg)
815 {
816 int nid = *(int *)arg;
817
818 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
819 return 0;
820 }
821
822 /*
823 * Unregister a memory block device under the node it spans. Memory blocks
824 * with multiple nodes cannot be offlined and therefore also never be removed.
825 */
unregister_memory_block_under_nodes(struct memory_block * mem_blk)826 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
827 {
828 if (mem_blk->nid == NUMA_NO_NODE)
829 return;
830
831 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
832 kobject_name(&mem_blk->dev.kobj));
833 sysfs_remove_link(&mem_blk->dev.kobj,
834 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
835 }
836
837 /* register all memory blocks under the corresponding nodes */
register_memory_blocks_under_nodes(void)838 static void register_memory_blocks_under_nodes(void)
839 {
840 struct memblock_region *r;
841
842 for_each_mem_region(r) {
843 const unsigned long start_block_id = phys_to_block_id(r->base);
844 const unsigned long end_block_id = phys_to_block_id(r->base + r->size - 1);
845 const int nid = memblock_get_region_node(r);
846 unsigned long block_id;
847
848 if (!node_online(nid))
849 continue;
850
851 for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
852 struct memory_block *mem;
853
854 mem = find_memory_block_by_id(block_id);
855 if (!mem)
856 continue;
857
858 do_register_memory_block_under_node(nid, mem, MEMINIT_EARLY);
859 put_device(&mem->dev);
860 }
861
862 }
863 }
864
register_memory_blocks_under_node_hotplug(int nid,unsigned long start_pfn,unsigned long end_pfn)865 void register_memory_blocks_under_node_hotplug(int nid, unsigned long start_pfn,
866 unsigned long end_pfn)
867 {
868 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
869 (void *)&nid, register_mem_block_under_node_hotplug);
870 return;
871 }
872 #endif /* CONFIG_MEMORY_HOTPLUG */
873
register_one_node(int nid)874 int register_one_node(int nid)
875 {
876 int error;
877 int cpu;
878 struct node *node;
879
880 node = kzalloc(sizeof(struct node), GFP_KERNEL);
881 if (!node)
882 return -ENOMEM;
883
884 INIT_LIST_HEAD(&node->access_list);
885 node_devices[nid] = node;
886
887 error = register_node(node_devices[nid], nid);
888
889 /* link cpu under this node */
890 for_each_present_cpu(cpu) {
891 if (cpu_to_node(cpu) == nid)
892 register_cpu_under_node(cpu, nid);
893 }
894
895 node_init_caches(nid);
896
897 return error;
898 }
899
unregister_one_node(int nid)900 void unregister_one_node(int nid)
901 {
902 if (!node_devices[nid])
903 return;
904
905 unregister_node(node_devices[nid]);
906 node_devices[nid] = NULL;
907 }
908
909 /*
910 * node states attributes
911 */
912
913 struct node_attr {
914 struct device_attribute attr;
915 enum node_states state;
916 };
917
show_node_state(struct device * dev,struct device_attribute * attr,char * buf)918 static ssize_t show_node_state(struct device *dev,
919 struct device_attribute *attr, char *buf)
920 {
921 struct node_attr *na = container_of(attr, struct node_attr, attr);
922
923 return sysfs_emit(buf, "%*pbl\n",
924 nodemask_pr_args(&node_states[na->state]));
925 }
926
927 #define _NODE_ATTR(name, state) \
928 { __ATTR(name, 0444, show_node_state, NULL), state }
929
930 static struct node_attr node_state_attr[] = {
931 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
932 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
933 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
934 #ifdef CONFIG_HIGHMEM
935 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
936 #endif
937 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
938 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
939 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
940 N_GENERIC_INITIATOR),
941 };
942
943 static struct attribute *node_state_attrs[] = {
944 &node_state_attr[N_POSSIBLE].attr.attr,
945 &node_state_attr[N_ONLINE].attr.attr,
946 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
947 #ifdef CONFIG_HIGHMEM
948 &node_state_attr[N_HIGH_MEMORY].attr.attr,
949 #endif
950 &node_state_attr[N_MEMORY].attr.attr,
951 &node_state_attr[N_CPU].attr.attr,
952 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
953 NULL
954 };
955
956 static const struct attribute_group memory_root_attr_group = {
957 .attrs = node_state_attrs,
958 };
959
960 static const struct attribute_group *cpu_root_attr_groups[] = {
961 &memory_root_attr_group,
962 NULL,
963 };
964
node_dev_init(void)965 void __init node_dev_init(void)
966 {
967 int ret, i;
968
969 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
970 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
971
972 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
973 if (ret)
974 panic("%s() failed to register subsystem: %d\n", __func__, ret);
975
976 /*
977 * Create all node devices, which will properly link the node
978 * to already created cpu devices.
979 */
980 for_each_online_node(i) {
981 ret = register_one_node(i);
982 if (ret)
983 panic("%s() failed to add node: %d\n", __func__, ret);
984 }
985
986 register_memory_blocks_under_nodes();
987 }
988