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 const struct bus_type node_subsys = {
25 .name = "node",
26 .dev_name = "node",
27 };
28
cpumap_read(struct file * file,struct kobject * kobj,const struct bin_attribute * attr,char * buf,loff_t off,size_t count)29 static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
30 const 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 const BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
49
cpulist_read(struct file * file,struct kobject * kobj,const struct bin_attribute * attr,char * buf,loff_t off,size_t count)50 static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
51 const 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 const 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 * @coord: Heterogeneous memory performance coordinates
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 access_coordinate coord;
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
node_remove_accesses(struct node * node)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
node_access_release(struct device * dev)123 static void node_access_release(struct device *dev)
124 {
125 kfree(to_access_nodes(dev));
126 }
127
node_init_node_access(struct node * node,enum access_coordinate_class access)128 static struct node_access_nodes *node_init_node_access(struct node *node,
129 enum access_coordinate_class 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(property) \
165 static ssize_t property##_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)->coord.property); \
171 } \
172 static DEVICE_ATTR_RO(property)
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 * @coord: Heterogeneous memory performance coordinates
191 * @access: The access class the for the given attributes
192 */
node_set_perf_attrs(unsigned int nid,struct access_coordinate * coord,enum access_coordinate_class access)193 void node_set_perf_attrs(unsigned int nid, struct access_coordinate *coord,
194 enum access_coordinate_class 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->coord = *coord;
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 EXPORT_SYMBOL_GPL(node_set_perf_attrs);
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 CACHE_ATTR(address_mode, "%#x")
248
249 static struct attribute *cache_attrs[] = {
250 &dev_attr_indexing.attr,
251 &dev_attr_size.attr,
252 &dev_attr_line_size.attr,
253 &dev_attr_write_policy.attr,
254 &dev_attr_address_mode.attr,
255 NULL,
256 };
257 ATTRIBUTE_GROUPS(cache);
258
node_cache_release(struct device * dev)259 static void node_cache_release(struct device *dev)
260 {
261 kfree(dev);
262 }
263
node_cacheinfo_release(struct device * dev)264 static void node_cacheinfo_release(struct device *dev)
265 {
266 struct node_cache_info *info = to_cache_info(dev);
267 kfree(info);
268 }
269
node_init_cache_dev(struct node * node)270 static void node_init_cache_dev(struct node *node)
271 {
272 struct device *dev;
273
274 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
275 if (!dev)
276 return;
277
278 device_initialize(dev);
279 dev->parent = &node->dev;
280 dev->release = node_cache_release;
281 if (dev_set_name(dev, "memory_side_cache"))
282 goto put_device;
283
284 if (device_add(dev))
285 goto put_device;
286
287 pm_runtime_no_callbacks(dev);
288 node->cache_dev = dev;
289 return;
290 put_device:
291 put_device(dev);
292 }
293
294 /**
295 * node_add_cache() - add cache attribute to a memory node
296 * @nid: Node identifier that has new cache attributes
297 * @cache_attrs: Attributes for the cache being added
298 */
node_add_cache(unsigned int nid,struct node_cache_attrs * cache_attrs)299 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
300 {
301 struct node_cache_info *info;
302 struct device *dev;
303 struct node *node;
304
305 if (!node_online(nid) || !node_devices[nid])
306 return;
307
308 node = node_devices[nid];
309 list_for_each_entry(info, &node->cache_attrs, node) {
310 if (info->cache_attrs.level == cache_attrs->level) {
311 dev_warn(&node->dev,
312 "attempt to add duplicate cache level:%d\n",
313 cache_attrs->level);
314 return;
315 }
316 }
317
318 if (!node->cache_dev)
319 node_init_cache_dev(node);
320 if (!node->cache_dev)
321 return;
322
323 info = kzalloc(sizeof(*info), GFP_KERNEL);
324 if (!info)
325 return;
326
327 dev = &info->dev;
328 device_initialize(dev);
329 dev->parent = node->cache_dev;
330 dev->release = node_cacheinfo_release;
331 dev->groups = cache_groups;
332 if (dev_set_name(dev, "index%d", cache_attrs->level))
333 goto put_device;
334
335 info->cache_attrs = *cache_attrs;
336 if (device_add(dev)) {
337 dev_warn(&node->dev, "failed to add cache level:%d\n",
338 cache_attrs->level);
339 goto put_device;
340 }
341 pm_runtime_no_callbacks(dev);
342 list_add_tail(&info->node, &node->cache_attrs);
343 return;
344 put_device:
345 put_device(dev);
346 }
347
node_remove_caches(struct node * node)348 static void node_remove_caches(struct node *node)
349 {
350 struct node_cache_info *info, *next;
351
352 if (!node->cache_dev)
353 return;
354
355 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
356 list_del(&info->node);
357 device_unregister(&info->dev);
358 }
359 device_unregister(node->cache_dev);
360 }
361
node_init_caches(unsigned int nid)362 static void node_init_caches(unsigned int nid)
363 {
364 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
365 }
366 #else
node_init_caches(unsigned int nid)367 static void node_init_caches(unsigned int nid) { }
node_remove_caches(struct node * node)368 static void node_remove_caches(struct node *node) { }
369 #endif
370
371 #define K(x) ((x) << (PAGE_SHIFT - 10))
node_read_meminfo(struct device * dev,struct device_attribute * attr,char * buf)372 static ssize_t node_read_meminfo(struct device *dev,
373 struct device_attribute *attr, char *buf)
374 {
375 int len = 0;
376 int nid = dev->id;
377 struct pglist_data *pgdat = NODE_DATA(nid);
378 struct sysinfo i;
379 unsigned long sreclaimable, sunreclaimable;
380 unsigned long swapcached = 0;
381
382 si_meminfo_node(&i, nid);
383 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
384 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
385 #ifdef CONFIG_SWAP
386 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
387 #endif
388 len = sysfs_emit_at(buf, len,
389 "Node %d MemTotal: %8lu kB\n"
390 "Node %d MemFree: %8lu kB\n"
391 "Node %d MemUsed: %8lu kB\n"
392 "Node %d SwapCached: %8lu kB\n"
393 "Node %d Active: %8lu kB\n"
394 "Node %d Inactive: %8lu kB\n"
395 "Node %d Active(anon): %8lu kB\n"
396 "Node %d Inactive(anon): %8lu kB\n"
397 "Node %d Active(file): %8lu kB\n"
398 "Node %d Inactive(file): %8lu kB\n"
399 "Node %d Unevictable: %8lu kB\n"
400 "Node %d Mlocked: %8lu kB\n",
401 nid, K(i.totalram),
402 nid, K(i.freeram),
403 nid, K(i.totalram - i.freeram),
404 nid, K(swapcached),
405 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
406 node_page_state(pgdat, NR_ACTIVE_FILE)),
407 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
408 node_page_state(pgdat, NR_INACTIVE_FILE)),
409 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
410 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
411 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
412 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
413 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
414 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
415
416 #ifdef CONFIG_HIGHMEM
417 len += sysfs_emit_at(buf, len,
418 "Node %d HighTotal: %8lu kB\n"
419 "Node %d HighFree: %8lu kB\n"
420 "Node %d LowTotal: %8lu kB\n"
421 "Node %d LowFree: %8lu kB\n",
422 nid, K(i.totalhigh),
423 nid, K(i.freehigh),
424 nid, K(i.totalram - i.totalhigh),
425 nid, K(i.freeram - i.freehigh));
426 #endif
427 len += sysfs_emit_at(buf, len,
428 "Node %d Dirty: %8lu kB\n"
429 "Node %d Writeback: %8lu kB\n"
430 "Node %d FilePages: %8lu kB\n"
431 "Node %d Mapped: %8lu kB\n"
432 "Node %d AnonPages: %8lu kB\n"
433 "Node %d Shmem: %8lu kB\n"
434 "Node %d KernelStack: %8lu kB\n"
435 #ifdef CONFIG_SHADOW_CALL_STACK
436 "Node %d ShadowCallStack:%8lu kB\n"
437 #endif
438 "Node %d PageTables: %8lu kB\n"
439 "Node %d SecPageTables: %8lu kB\n"
440 "Node %d NFS_Unstable: %8lu kB\n"
441 "Node %d Bounce: %8lu kB\n"
442 "Node %d WritebackTmp: %8lu kB\n"
443 "Node %d KReclaimable: %8lu kB\n"
444 "Node %d Slab: %8lu kB\n"
445 "Node %d SReclaimable: %8lu kB\n"
446 "Node %d SUnreclaim: %8lu kB\n"
447 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
448 "Node %d AnonHugePages: %8lu kB\n"
449 "Node %d ShmemHugePages: %8lu kB\n"
450 "Node %d ShmemPmdMapped: %8lu kB\n"
451 "Node %d FileHugePages: %8lu kB\n"
452 "Node %d FilePmdMapped: %8lu kB\n"
453 #endif
454 #ifdef CONFIG_UNACCEPTED_MEMORY
455 "Node %d Unaccepted: %8lu kB\n"
456 #endif
457 ,
458 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
459 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
460 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
461 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
462 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
463 nid, K(i.sharedram),
464 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
465 #ifdef CONFIG_SHADOW_CALL_STACK
466 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
467 #endif
468 nid, K(node_page_state(pgdat, NR_PAGETABLE)),
469 nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
470 nid, 0UL,
471 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
472 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
473 nid, K(sreclaimable +
474 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
475 nid, K(sreclaimable + sunreclaimable),
476 nid, K(sreclaimable),
477 nid, K(sunreclaimable)
478 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
479 ,
480 nid, K(node_page_state(pgdat, NR_ANON_THPS)),
481 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
482 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
483 nid, K(node_page_state(pgdat, NR_FILE_THPS)),
484 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
485 #endif
486 #ifdef CONFIG_UNACCEPTED_MEMORY
487 ,
488 nid, K(sum_zone_node_page_state(nid, NR_UNACCEPTED))
489 #endif
490 );
491 len += hugetlb_report_node_meminfo(buf, len, nid);
492 return len;
493 }
494
495 #undef K
496 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
497
node_read_numastat(struct device * dev,struct device_attribute * attr,char * buf)498 static ssize_t node_read_numastat(struct device *dev,
499 struct device_attribute *attr, char *buf)
500 {
501 fold_vm_numa_events();
502 return sysfs_emit(buf,
503 "numa_hit %lu\n"
504 "numa_miss %lu\n"
505 "numa_foreign %lu\n"
506 "interleave_hit %lu\n"
507 "local_node %lu\n"
508 "other_node %lu\n",
509 sum_zone_numa_event_state(dev->id, NUMA_HIT),
510 sum_zone_numa_event_state(dev->id, NUMA_MISS),
511 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
512 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
513 sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
514 sum_zone_numa_event_state(dev->id, NUMA_OTHER));
515 }
516 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
517
node_read_vmstat(struct device * dev,struct device_attribute * attr,char * buf)518 static ssize_t node_read_vmstat(struct device *dev,
519 struct device_attribute *attr, char *buf)
520 {
521 int nid = dev->id;
522 struct pglist_data *pgdat = NODE_DATA(nid);
523 int i;
524 int len = 0;
525
526 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
527 len += sysfs_emit_at(buf, len, "%s %lu\n",
528 zone_stat_name(i),
529 sum_zone_node_page_state(nid, i));
530
531 #ifdef CONFIG_NUMA
532 fold_vm_numa_events();
533 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
534 len += sysfs_emit_at(buf, len, "%s %lu\n",
535 numa_stat_name(i),
536 sum_zone_numa_event_state(nid, i));
537
538 #endif
539 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
540 unsigned long pages = node_page_state_pages(pgdat, i);
541
542 if (vmstat_item_print_in_thp(i))
543 pages /= HPAGE_PMD_NR;
544 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
545 pages);
546 }
547
548 return len;
549 }
550 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
551
node_read_distance(struct device * dev,struct device_attribute * attr,char * buf)552 static ssize_t node_read_distance(struct device *dev,
553 struct device_attribute *attr, char *buf)
554 {
555 int nid = dev->id;
556 int len = 0;
557 int i;
558
559 /*
560 * buf is currently PAGE_SIZE in length and each node needs 4 chars
561 * at the most (distance + space or newline).
562 */
563 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
564
565 for_each_online_node(i) {
566 len += sysfs_emit_at(buf, len, "%s%d",
567 i ? " " : "", node_distance(nid, i));
568 }
569
570 len += sysfs_emit_at(buf, len, "\n");
571 return len;
572 }
573 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
574
575 static struct attribute *node_dev_attrs[] = {
576 &dev_attr_meminfo.attr,
577 &dev_attr_numastat.attr,
578 &dev_attr_distance.attr,
579 &dev_attr_vmstat.attr,
580 NULL
581 };
582
583 static const struct bin_attribute *node_dev_bin_attrs[] = {
584 &bin_attr_cpumap,
585 &bin_attr_cpulist,
586 NULL
587 };
588
589 static const struct attribute_group node_dev_group = {
590 .attrs = node_dev_attrs,
591 .bin_attrs_new = node_dev_bin_attrs,
592 };
593
594 static const struct attribute_group *node_dev_groups[] = {
595 &node_dev_group,
596 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
597 &arch_node_dev_group,
598 #endif
599 #ifdef CONFIG_MEMORY_FAILURE
600 &memory_failure_attr_group,
601 #endif
602 NULL
603 };
604
node_device_release(struct device * dev)605 static void node_device_release(struct device *dev)
606 {
607 kfree(to_node(dev));
608 }
609
610 /*
611 * register_node - Setup a sysfs device for a node.
612 * @num - Node number to use when creating the device.
613 *
614 * Initialize and register the node device.
615 */
register_node(struct node * node,int num)616 static int register_node(struct node *node, int num)
617 {
618 int error;
619
620 node->dev.id = num;
621 node->dev.bus = &node_subsys;
622 node->dev.release = node_device_release;
623 node->dev.groups = node_dev_groups;
624 error = device_register(&node->dev);
625
626 if (error) {
627 put_device(&node->dev);
628 } else {
629 hugetlb_register_node(node);
630 compaction_register_node(node);
631 }
632
633 return error;
634 }
635
636 /**
637 * unregister_node - unregister a node device
638 * @node: node going away
639 *
640 * Unregisters a node device @node. All the devices on the node must be
641 * unregistered before calling this function.
642 */
unregister_node(struct node * node)643 void unregister_node(struct node *node)
644 {
645 hugetlb_unregister_node(node);
646 compaction_unregister_node(node);
647 node_remove_accesses(node);
648 node_remove_caches(node);
649 device_unregister(&node->dev);
650 }
651
652 struct node *node_devices[MAX_NUMNODES];
653
654 /*
655 * register cpu under node
656 */
register_cpu_under_node(unsigned int cpu,unsigned int nid)657 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
658 {
659 int ret;
660 struct device *obj;
661
662 if (!node_online(nid))
663 return 0;
664
665 obj = get_cpu_device(cpu);
666 if (!obj)
667 return 0;
668
669 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
670 &obj->kobj,
671 kobject_name(&obj->kobj));
672 if (ret)
673 return ret;
674
675 return sysfs_create_link(&obj->kobj,
676 &node_devices[nid]->dev.kobj,
677 kobject_name(&node_devices[nid]->dev.kobj));
678 }
679
680 /**
681 * register_memory_node_under_compute_node - link memory node to its compute
682 * node for a given access class.
683 * @mem_nid: Memory node number
684 * @cpu_nid: Cpu node number
685 * @access: Access class to register
686 *
687 * Description:
688 * For use with platforms that may have separate memory and compute nodes.
689 * This function will export node relationships linking which memory
690 * initiator nodes can access memory targets at a given ranked access
691 * class.
692 */
register_memory_node_under_compute_node(unsigned int mem_nid,unsigned int cpu_nid,enum access_coordinate_class access)693 int register_memory_node_under_compute_node(unsigned int mem_nid,
694 unsigned int cpu_nid,
695 enum access_coordinate_class access)
696 {
697 struct node *init_node, *targ_node;
698 struct node_access_nodes *initiator, *target;
699 int ret;
700
701 if (!node_online(cpu_nid) || !node_online(mem_nid))
702 return -ENODEV;
703
704 init_node = node_devices[cpu_nid];
705 targ_node = node_devices[mem_nid];
706 initiator = node_init_node_access(init_node, access);
707 target = node_init_node_access(targ_node, access);
708 if (!initiator || !target)
709 return -ENOMEM;
710
711 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
712 &targ_node->dev.kobj,
713 dev_name(&targ_node->dev));
714 if (ret)
715 return ret;
716
717 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
718 &init_node->dev.kobj,
719 dev_name(&init_node->dev));
720 if (ret)
721 goto err;
722
723 return 0;
724 err:
725 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
726 dev_name(&targ_node->dev));
727 return ret;
728 }
729
unregister_cpu_under_node(unsigned int cpu,unsigned int nid)730 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
731 {
732 struct device *obj;
733
734 if (!node_online(nid))
735 return 0;
736
737 obj = get_cpu_device(cpu);
738 if (!obj)
739 return 0;
740
741 sysfs_remove_link(&node_devices[nid]->dev.kobj,
742 kobject_name(&obj->kobj));
743 sysfs_remove_link(&obj->kobj,
744 kobject_name(&node_devices[nid]->dev.kobj));
745
746 return 0;
747 }
748
749 #ifdef CONFIG_MEMORY_HOTPLUG
get_nid_for_pfn(unsigned long pfn)750 static int __ref get_nid_for_pfn(unsigned long pfn)
751 {
752 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
753 if (system_state < SYSTEM_RUNNING)
754 return early_pfn_to_nid(pfn);
755 #endif
756 return pfn_to_nid(pfn);
757 }
758
do_register_memory_block_under_node(int nid,struct memory_block * mem_blk,enum meminit_context context)759 static void do_register_memory_block_under_node(int nid,
760 struct memory_block *mem_blk,
761 enum meminit_context context)
762 {
763 int ret;
764
765 memory_block_add_nid(mem_blk, nid, context);
766
767 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
768 &mem_blk->dev.kobj,
769 kobject_name(&mem_blk->dev.kobj));
770 if (ret && ret != -EEXIST)
771 dev_err_ratelimited(&node_devices[nid]->dev,
772 "can't create link to %s in sysfs (%d)\n",
773 kobject_name(&mem_blk->dev.kobj), ret);
774
775 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
776 &node_devices[nid]->dev.kobj,
777 kobject_name(&node_devices[nid]->dev.kobj));
778 if (ret && ret != -EEXIST)
779 dev_err_ratelimited(&mem_blk->dev,
780 "can't create link to %s in sysfs (%d)\n",
781 kobject_name(&node_devices[nid]->dev.kobj),
782 ret);
783 }
784
785 /* register memory section under specified node if it spans that node */
register_mem_block_under_node_early(struct memory_block * mem_blk,void * arg)786 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
787 void *arg)
788 {
789 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
790 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
791 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
792 int nid = *(int *)arg;
793 unsigned long pfn;
794
795 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
796 int page_nid;
797
798 /*
799 * memory block could have several absent sections from start.
800 * skip pfn range from absent section
801 */
802 if (!pfn_in_present_section(pfn)) {
803 pfn = round_down(pfn + PAGES_PER_SECTION,
804 PAGES_PER_SECTION) - 1;
805 continue;
806 }
807
808 /*
809 * We need to check if page belongs to nid only at the boot
810 * case because node's ranges can be interleaved.
811 */
812 page_nid = get_nid_for_pfn(pfn);
813 if (page_nid < 0)
814 continue;
815 if (page_nid != nid)
816 continue;
817
818 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
819 return 0;
820 }
821 /* mem section does not span the specified node */
822 return 0;
823 }
824
825 /*
826 * During hotplug we know that all pages in the memory block belong to the same
827 * node.
828 */
register_mem_block_under_node_hotplug(struct memory_block * mem_blk,void * arg)829 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
830 void *arg)
831 {
832 int nid = *(int *)arg;
833
834 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
835 return 0;
836 }
837
838 /*
839 * Unregister a memory block device under the node it spans. Memory blocks
840 * with multiple nodes cannot be offlined and therefore also never be removed.
841 */
unregister_memory_block_under_nodes(struct memory_block * mem_blk)842 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
843 {
844 if (mem_blk->nid == NUMA_NO_NODE)
845 return;
846
847 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
848 kobject_name(&mem_blk->dev.kobj));
849 sysfs_remove_link(&mem_blk->dev.kobj,
850 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
851 }
852
register_memory_blocks_under_node(int nid,unsigned long start_pfn,unsigned long end_pfn,enum meminit_context context)853 void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
854 unsigned long end_pfn,
855 enum meminit_context context)
856 {
857 walk_memory_blocks_func_t func;
858
859 if (context == MEMINIT_HOTPLUG)
860 func = register_mem_block_under_node_hotplug;
861 else
862 func = register_mem_block_under_node_early;
863
864 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
865 (void *)&nid, func);
866 return;
867 }
868 #endif /* CONFIG_MEMORY_HOTPLUG */
869
__register_one_node(int nid)870 int __register_one_node(int nid)
871 {
872 int error;
873 int cpu;
874 struct node *node;
875
876 node = kzalloc(sizeof(struct node), GFP_KERNEL);
877 if (!node)
878 return -ENOMEM;
879
880 INIT_LIST_HEAD(&node->access_list);
881 node_devices[nid] = node;
882
883 error = register_node(node_devices[nid], nid);
884
885 /* link cpu under this node */
886 for_each_present_cpu(cpu) {
887 if (cpu_to_node(cpu) == nid)
888 register_cpu_under_node(cpu, nid);
889 }
890
891 node_init_caches(nid);
892
893 return error;
894 }
895
unregister_one_node(int nid)896 void unregister_one_node(int nid)
897 {
898 if (!node_devices[nid])
899 return;
900
901 unregister_node(node_devices[nid]);
902 node_devices[nid] = NULL;
903 }
904
905 /*
906 * node states attributes
907 */
908
909 struct node_attr {
910 struct device_attribute attr;
911 enum node_states state;
912 };
913
show_node_state(struct device * dev,struct device_attribute * attr,char * buf)914 static ssize_t show_node_state(struct device *dev,
915 struct device_attribute *attr, char *buf)
916 {
917 struct node_attr *na = container_of(attr, struct node_attr, attr);
918
919 return sysfs_emit(buf, "%*pbl\n",
920 nodemask_pr_args(&node_states[na->state]));
921 }
922
923 #define _NODE_ATTR(name, state) \
924 { __ATTR(name, 0444, show_node_state, NULL), state }
925
926 static struct node_attr node_state_attr[] = {
927 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
928 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
929 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
930 #ifdef CONFIG_HIGHMEM
931 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
932 #endif
933 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
934 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
935 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
936 N_GENERIC_INITIATOR),
937 };
938
939 static struct attribute *node_state_attrs[] = {
940 &node_state_attr[N_POSSIBLE].attr.attr,
941 &node_state_attr[N_ONLINE].attr.attr,
942 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
943 #ifdef CONFIG_HIGHMEM
944 &node_state_attr[N_HIGH_MEMORY].attr.attr,
945 #endif
946 &node_state_attr[N_MEMORY].attr.attr,
947 &node_state_attr[N_CPU].attr.attr,
948 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
949 NULL
950 };
951
952 static const struct attribute_group memory_root_attr_group = {
953 .attrs = node_state_attrs,
954 };
955
956 static const struct attribute_group *cpu_root_attr_groups[] = {
957 &memory_root_attr_group,
958 NULL,
959 };
960
node_dev_init(void)961 void __init node_dev_init(void)
962 {
963 int ret, i;
964
965 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
966 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
967
968 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
969 if (ret)
970 panic("%s() failed to register subsystem: %d\n", __func__, ret);
971
972 /*
973 * Create all node devices, which will properly link the node
974 * to applicable memory block devices and already created cpu devices.
975 */
976 for_each_online_node(i) {
977 ret = register_one_node(i);
978 if (ret)
979 panic("%s() failed to add node: %d\n", __func__, ret);
980 }
981 }
982