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
node_read_cpumap(struct device * dev,bool list,char * buf)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
node_read_cpumask(struct device * dev,struct device_attribute * attr,char * buf)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 }
node_read_cpulist(struct device * dev,struct device_attribute * attr,char * buf)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 * @hmem_attrs: Heterogeneous memory performance attributes
70 */
71 struct node_access_nodes {
72 struct device dev;
73 struct list_head list_node;
74 unsigned access;
75 #ifdef CONFIG_HMEM_REPORTING
76 struct node_hmem_attrs hmem_attrs;
77 #endif
78 };
79 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
80
81 static struct attribute *node_init_access_node_attrs[] = {
82 NULL,
83 };
84
85 static struct attribute *node_targ_access_node_attrs[] = {
86 NULL,
87 };
88
89 static const struct attribute_group initiators = {
90 .name = "initiators",
91 .attrs = node_init_access_node_attrs,
92 };
93
94 static const struct attribute_group targets = {
95 .name = "targets",
96 .attrs = node_targ_access_node_attrs,
97 };
98
99 static const struct attribute_group *node_access_node_groups[] = {
100 &initiators,
101 &targets,
102 NULL,
103 };
104
node_remove_accesses(struct node * node)105 static void node_remove_accesses(struct node *node)
106 {
107 struct node_access_nodes *c, *cnext;
108
109 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
110 list_del(&c->list_node);
111 device_unregister(&c->dev);
112 }
113 }
114
node_access_release(struct device * dev)115 static void node_access_release(struct device *dev)
116 {
117 kfree(to_access_nodes(dev));
118 }
119
node_init_node_access(struct node * node,unsigned access)120 static struct node_access_nodes *node_init_node_access(struct node *node,
121 unsigned access)
122 {
123 struct node_access_nodes *access_node;
124 struct device *dev;
125
126 list_for_each_entry(access_node, &node->access_list, list_node)
127 if (access_node->access == access)
128 return access_node;
129
130 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
131 if (!access_node)
132 return NULL;
133
134 access_node->access = access;
135 dev = &access_node->dev;
136 dev->parent = &node->dev;
137 dev->release = node_access_release;
138 dev->groups = node_access_node_groups;
139 if (dev_set_name(dev, "access%u", access))
140 goto free;
141
142 if (device_register(dev))
143 goto free_name;
144
145 pm_runtime_no_callbacks(dev);
146 list_add_tail(&access_node->list_node, &node->access_list);
147 return access_node;
148 free_name:
149 kfree_const(dev->kobj.name);
150 free:
151 kfree(access_node);
152 return NULL;
153 }
154
155 #ifdef CONFIG_HMEM_REPORTING
156 #define ACCESS_ATTR(name) \
157 static ssize_t name##_show(struct device *dev, \
158 struct device_attribute *attr, \
159 char *buf) \
160 { \
161 return sprintf(buf, "%u\n", to_access_nodes(dev)->hmem_attrs.name); \
162 } \
163 static DEVICE_ATTR_RO(name);
164
165 ACCESS_ATTR(read_bandwidth)
166 ACCESS_ATTR(read_latency)
167 ACCESS_ATTR(write_bandwidth)
168 ACCESS_ATTR(write_latency)
169
170 static struct attribute *access_attrs[] = {
171 &dev_attr_read_bandwidth.attr,
172 &dev_attr_read_latency.attr,
173 &dev_attr_write_bandwidth.attr,
174 &dev_attr_write_latency.attr,
175 NULL,
176 };
177
178 /**
179 * node_set_perf_attrs - Set the performance values for given access class
180 * @nid: Node identifier to be set
181 * @hmem_attrs: Heterogeneous memory performance attributes
182 * @access: The access class the for the given attributes
183 */
node_set_perf_attrs(unsigned int nid,struct node_hmem_attrs * hmem_attrs,unsigned access)184 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
185 unsigned access)
186 {
187 struct node_access_nodes *c;
188 struct node *node;
189 int i;
190
191 if (WARN_ON_ONCE(!node_online(nid)))
192 return;
193
194 node = node_devices[nid];
195 c = node_init_node_access(node, access);
196 if (!c)
197 return;
198
199 c->hmem_attrs = *hmem_attrs;
200 for (i = 0; access_attrs[i] != NULL; i++) {
201 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
202 "initiators")) {
203 pr_info("failed to add performance attribute to node %d\n",
204 nid);
205 break;
206 }
207 }
208 }
209
210 /**
211 * struct node_cache_info - Internal tracking for memory node caches
212 * @dev: Device represeting the cache level
213 * @node: List element for tracking in the node
214 * @cache_attrs:Attributes for this cache level
215 */
216 struct node_cache_info {
217 struct device dev;
218 struct list_head node;
219 struct node_cache_attrs cache_attrs;
220 };
221 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
222
223 #define CACHE_ATTR(name, fmt) \
224 static ssize_t name##_show(struct device *dev, \
225 struct device_attribute *attr, \
226 char *buf) \
227 { \
228 return sprintf(buf, fmt "\n", to_cache_info(dev)->cache_attrs.name);\
229 } \
230 DEVICE_ATTR_RO(name);
231
232 CACHE_ATTR(size, "%llu")
233 CACHE_ATTR(line_size, "%u")
234 CACHE_ATTR(indexing, "%u")
235 CACHE_ATTR(write_policy, "%u")
236
237 static struct attribute *cache_attrs[] = {
238 &dev_attr_indexing.attr,
239 &dev_attr_size.attr,
240 &dev_attr_line_size.attr,
241 &dev_attr_write_policy.attr,
242 NULL,
243 };
244 ATTRIBUTE_GROUPS(cache);
245
node_cache_release(struct device * dev)246 static void node_cache_release(struct device *dev)
247 {
248 kfree(dev);
249 }
250
node_cacheinfo_release(struct device * dev)251 static void node_cacheinfo_release(struct device *dev)
252 {
253 struct node_cache_info *info = to_cache_info(dev);
254 kfree(info);
255 }
256
node_init_cache_dev(struct node * node)257 static void node_init_cache_dev(struct node *node)
258 {
259 struct device *dev;
260
261 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
262 if (!dev)
263 return;
264
265 device_initialize(dev);
266 dev->parent = &node->dev;
267 dev->release = node_cache_release;
268 if (dev_set_name(dev, "memory_side_cache"))
269 goto put_device;
270
271 if (device_add(dev))
272 goto put_device;
273
274 pm_runtime_no_callbacks(dev);
275 node->cache_dev = dev;
276 return;
277 put_device:
278 put_device(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 */
node_add_cache(unsigned int nid,struct node_cache_attrs * cache_attrs)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 device_initialize(dev);
316 dev->parent = node->cache_dev;
317 dev->release = node_cacheinfo_release;
318 dev->groups = cache_groups;
319 if (dev_set_name(dev, "index%d", cache_attrs->level))
320 goto put_device;
321
322 info->cache_attrs = *cache_attrs;
323 if (device_add(dev)) {
324 dev_warn(&node->dev, "failed to add cache level:%d\n",
325 cache_attrs->level);
326 goto put_device;
327 }
328 pm_runtime_no_callbacks(dev);
329 list_add_tail(&info->node, &node->cache_attrs);
330 return;
331 put_device:
332 put_device(dev);
333 }
334
node_remove_caches(struct node * node)335 static void node_remove_caches(struct node *node)
336 {
337 struct node_cache_info *info, *next;
338
339 if (!node->cache_dev)
340 return;
341
342 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
343 list_del(&info->node);
344 device_unregister(&info->dev);
345 }
346 device_unregister(node->cache_dev);
347 }
348
node_init_caches(unsigned int nid)349 static void node_init_caches(unsigned int nid)
350 {
351 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
352 }
353 #else
node_init_caches(unsigned int nid)354 static void node_init_caches(unsigned int nid) { }
node_remove_caches(struct node * node)355 static void node_remove_caches(struct node *node) { }
356 #endif
357
358 #define K(x) ((x) << (PAGE_SHIFT - 10))
node_read_meminfo(struct device * dev,struct device_attribute * attr,char * buf)359 static ssize_t node_read_meminfo(struct device *dev,
360 struct device_attribute *attr, char *buf)
361 {
362 int n;
363 int nid = dev->id;
364 struct pglist_data *pgdat = NODE_DATA(nid);
365 struct sysinfo i;
366 unsigned long sreclaimable, sunreclaimable;
367
368 si_meminfo_node(&i, nid);
369 sreclaimable = node_page_state(pgdat, NR_SLAB_RECLAIMABLE);
370 sunreclaimable = node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE);
371 n = sysfs_emit(buf,
372 "Node %d MemTotal: %8lu kB\n"
373 "Node %d MemFree: %8lu kB\n"
374 "Node %d MemUsed: %8lu kB\n"
375 "Node %d Active: %8lu kB\n"
376 "Node %d Inactive: %8lu kB\n"
377 "Node %d Active(anon): %8lu kB\n"
378 "Node %d Inactive(anon): %8lu kB\n"
379 "Node %d Active(file): %8lu kB\n"
380 "Node %d Inactive(file): %8lu kB\n"
381 "Node %d Unevictable: %8lu kB\n"
382 "Node %d Mlocked: %8lu kB\n",
383 nid, K(i.totalram),
384 nid, K(i.freeram),
385 nid, K(i.totalram - i.freeram),
386 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
387 node_page_state(pgdat, NR_ACTIVE_FILE)),
388 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
389 node_page_state(pgdat, NR_INACTIVE_FILE)),
390 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
391 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
392 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
393 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
394 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
395 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
396
397 #ifdef CONFIG_HIGHMEM
398 n += sprintf(buf + n,
399 "Node %d HighTotal: %8lu kB\n"
400 "Node %d HighFree: %8lu kB\n"
401 "Node %d LowTotal: %8lu kB\n"
402 "Node %d LowFree: %8lu kB\n",
403 nid, K(i.totalhigh),
404 nid, K(i.freehigh),
405 nid, K(i.totalram - i.totalhigh),
406 nid, K(i.freeram - i.freehigh));
407 #endif
408 n += sprintf(buf + n,
409 "Node %d Dirty: %8lu kB\n"
410 "Node %d Writeback: %8lu kB\n"
411 "Node %d FilePages: %8lu kB\n"
412 "Node %d Mapped: %8lu kB\n"
413 "Node %d AnonPages: %8lu kB\n"
414 "Node %d Shmem: %8lu kB\n"
415 "Node %d KernelStack: %8lu kB\n"
416 #ifdef CONFIG_SHADOW_CALL_STACK
417 "Node %d ShadowCallStack:%8lu kB\n"
418 #endif
419 "Node %d PageTables: %8lu kB\n"
420 "Node %d NFS_Unstable: %8lu kB\n"
421 "Node %d Bounce: %8lu kB\n"
422 "Node %d WritebackTmp: %8lu kB\n"
423 "Node %d KReclaimable: %8lu kB\n"
424 "Node %d Slab: %8lu kB\n"
425 "Node %d SReclaimable: %8lu kB\n"
426 "Node %d SUnreclaim: %8lu kB\n"
427 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
428 "Node %d AnonHugePages: %8lu kB\n"
429 "Node %d ShmemHugePages: %8lu kB\n"
430 "Node %d ShmemPmdMapped: %8lu kB\n"
431 "Node %d FileHugePages: %8lu kB\n"
432 "Node %d FilePmdMapped: %8lu kB\n"
433 #endif
434 ,
435 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
436 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
437 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
438 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
439 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
440 nid, K(i.sharedram),
441 nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
442 #ifdef CONFIG_SHADOW_CALL_STACK
443 nid, sum_zone_node_page_state(nid, NR_KERNEL_SCS_BYTES) / 1024,
444 #endif
445 nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
446 nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
447 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
448 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
449 nid, K(sreclaimable +
450 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
451 nid, K(sreclaimable + sunreclaimable),
452 nid, K(sreclaimable),
453 nid, K(sunreclaimable)
454 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
455 ,
456 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
457 HPAGE_PMD_NR),
458 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
459 HPAGE_PMD_NR),
460 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
461 HPAGE_PMD_NR),
462 nid, K(node_page_state(pgdat, NR_FILE_THPS) *
463 HPAGE_PMD_NR),
464 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED) *
465 HPAGE_PMD_NR)
466 #endif
467 );
468 n += hugetlb_report_node_meminfo(nid, buf + n);
469 return n;
470 }
471
472 #undef K
473 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
474
node_read_numastat(struct device * dev,struct device_attribute * attr,char * buf)475 static ssize_t node_read_numastat(struct device *dev,
476 struct device_attribute *attr, char *buf)
477 {
478 return sysfs_emit(buf,
479 "numa_hit %lu\n"
480 "numa_miss %lu\n"
481 "numa_foreign %lu\n"
482 "interleave_hit %lu\n"
483 "local_node %lu\n"
484 "other_node %lu\n",
485 sum_zone_numa_state(dev->id, NUMA_HIT),
486 sum_zone_numa_state(dev->id, NUMA_MISS),
487 sum_zone_numa_state(dev->id, NUMA_FOREIGN),
488 sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
489 sum_zone_numa_state(dev->id, NUMA_LOCAL),
490 sum_zone_numa_state(dev->id, NUMA_OTHER));
491 }
492 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
493
node_read_vmstat(struct device * dev,struct device_attribute * attr,char * buf)494 static ssize_t node_read_vmstat(struct device *dev,
495 struct device_attribute *attr, char *buf)
496 {
497 int nid = dev->id;
498 struct pglist_data *pgdat = NODE_DATA(nid);
499 int i;
500 int n = 0;
501
502 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
503 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
504 sum_zone_node_page_state(nid, i));
505
506 #ifdef CONFIG_NUMA
507 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
508 n += sprintf(buf+n, "%s %lu\n",
509 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
510 sum_zone_numa_state(nid, i));
511 #endif
512
513 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
514 n += sprintf(buf+n, "%s %lu\n",
515 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
516 NR_VM_NUMA_STAT_ITEMS],
517 node_page_state(pgdat, i));
518
519 return n;
520 }
521 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
522
node_read_distance(struct device * dev,struct device_attribute * attr,char * buf)523 static ssize_t node_read_distance(struct device *dev,
524 struct device_attribute *attr, char *buf)
525 {
526 int nid = dev->id;
527 int len = 0;
528 int i;
529
530 /*
531 * buf is currently PAGE_SIZE in length and each node needs 4 chars
532 * at the most (distance + space or newline).
533 */
534 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
535
536 for_each_online_node(i)
537 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
538
539 len += sprintf(buf + len, "\n");
540 return len;
541 }
542 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
543
544 static struct attribute *node_dev_attrs[] = {
545 &dev_attr_cpumap.attr,
546 &dev_attr_cpulist.attr,
547 &dev_attr_meminfo.attr,
548 &dev_attr_numastat.attr,
549 &dev_attr_distance.attr,
550 &dev_attr_vmstat.attr,
551 NULL
552 };
553 ATTRIBUTE_GROUPS(node_dev);
554
555 #ifdef CONFIG_HUGETLBFS
556 /*
557 * hugetlbfs per node attributes registration interface:
558 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
559 * it will register its per node attributes for all online nodes with
560 * memory. It will also call register_hugetlbfs_with_node(), below, to
561 * register its attribute registration functions with this node driver.
562 * Once these hooks have been initialized, the node driver will call into
563 * the hugetlb module to [un]register attributes for hot-plugged nodes.
564 */
565 static node_registration_func_t __hugetlb_register_node;
566 static node_registration_func_t __hugetlb_unregister_node;
567
hugetlb_register_node(struct node * node)568 static inline bool hugetlb_register_node(struct node *node)
569 {
570 if (__hugetlb_register_node &&
571 node_state(node->dev.id, N_MEMORY)) {
572 __hugetlb_register_node(node);
573 return true;
574 }
575 return false;
576 }
577
hugetlb_unregister_node(struct node * node)578 static inline void hugetlb_unregister_node(struct node *node)
579 {
580 if (__hugetlb_unregister_node)
581 __hugetlb_unregister_node(node);
582 }
583
register_hugetlbfs_with_node(node_registration_func_t doregister,node_registration_func_t unregister)584 void register_hugetlbfs_with_node(node_registration_func_t doregister,
585 node_registration_func_t unregister)
586 {
587 __hugetlb_register_node = doregister;
588 __hugetlb_unregister_node = unregister;
589 }
590 #else
hugetlb_register_node(struct node * node)591 static inline void hugetlb_register_node(struct node *node) {}
592
hugetlb_unregister_node(struct node * node)593 static inline void hugetlb_unregister_node(struct node *node) {}
594 #endif
595
node_device_release(struct device * dev)596 static void node_device_release(struct device *dev)
597 {
598 struct node *node = to_node(dev);
599
600 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
601 /*
602 * We schedule the work only when a memory section is
603 * onlined/offlined on this node. When we come here,
604 * all the memory on this node has been offlined,
605 * so we won't enqueue new work to this work.
606 *
607 * The work is using node->node_work, so we should
608 * flush work before freeing the memory.
609 */
610 flush_work(&node->node_work);
611 #endif
612 kfree(node);
613 }
614
615 /*
616 * register_node - Setup a sysfs device for a node.
617 * @num - Node number to use when creating the device.
618 *
619 * Initialize and register the node device.
620 */
register_node(struct node * node,int num)621 static int register_node(struct node *node, int num)
622 {
623 int error;
624
625 node->dev.id = num;
626 node->dev.bus = &node_subsys;
627 node->dev.release = node_device_release;
628 node->dev.groups = node_dev_groups;
629 error = device_register(&node->dev);
630
631 if (error)
632 put_device(&node->dev);
633 else {
634 hugetlb_register_node(node);
635
636 compaction_register_node(node);
637 }
638 return error;
639 }
640
641 /**
642 * unregister_node - unregister a node device
643 * @node: node going away
644 *
645 * Unregisters a node device @node. All the devices on the node must be
646 * unregistered before calling this function.
647 */
unregister_node(struct node * node)648 void unregister_node(struct node *node)
649 {
650 compaction_unregister_node(node);
651 hugetlb_unregister_node(node); /* no-op, if memoryless node */
652 node_remove_accesses(node);
653 node_remove_caches(node);
654 device_unregister(&node->dev);
655 }
656
657 struct node *node_devices[MAX_NUMNODES];
658
659 /*
660 * register cpu under node
661 */
register_cpu_under_node(unsigned int cpu,unsigned int nid)662 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
663 {
664 int ret;
665 struct device *obj;
666
667 if (!node_online(nid))
668 return 0;
669
670 obj = get_cpu_device(cpu);
671 if (!obj)
672 return 0;
673
674 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
675 &obj->kobj,
676 kobject_name(&obj->kobj));
677 if (ret)
678 return ret;
679
680 return sysfs_create_link(&obj->kobj,
681 &node_devices[nid]->dev.kobj,
682 kobject_name(&node_devices[nid]->dev.kobj));
683 }
684
685 /**
686 * register_memory_node_under_compute_node - link memory node to its compute
687 * node for a given access class.
688 * @mem_nid: Memory node number
689 * @cpu_nid: Cpu node number
690 * @access: Access class to register
691 *
692 * Description:
693 * For use with platforms that may have separate memory and compute nodes.
694 * This function will export node relationships linking which memory
695 * initiator nodes can access memory targets at a given ranked access
696 * class.
697 */
register_memory_node_under_compute_node(unsigned int mem_nid,unsigned int cpu_nid,unsigned access)698 int register_memory_node_under_compute_node(unsigned int mem_nid,
699 unsigned int cpu_nid,
700 unsigned access)
701 {
702 struct node *init_node, *targ_node;
703 struct node_access_nodes *initiator, *target;
704 int ret;
705
706 if (!node_online(cpu_nid) || !node_online(mem_nid))
707 return -ENODEV;
708
709 init_node = node_devices[cpu_nid];
710 targ_node = node_devices[mem_nid];
711 initiator = node_init_node_access(init_node, access);
712 target = node_init_node_access(targ_node, access);
713 if (!initiator || !target)
714 return -ENOMEM;
715
716 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
717 &targ_node->dev.kobj,
718 dev_name(&targ_node->dev));
719 if (ret)
720 return ret;
721
722 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
723 &init_node->dev.kobj,
724 dev_name(&init_node->dev));
725 if (ret)
726 goto err;
727
728 return 0;
729 err:
730 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
731 dev_name(&targ_node->dev));
732 return ret;
733 }
734
unregister_cpu_under_node(unsigned int cpu,unsigned int nid)735 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
736 {
737 struct device *obj;
738
739 if (!node_online(nid))
740 return 0;
741
742 obj = get_cpu_device(cpu);
743 if (!obj)
744 return 0;
745
746 sysfs_remove_link(&node_devices[nid]->dev.kobj,
747 kobject_name(&obj->kobj));
748 sysfs_remove_link(&obj->kobj,
749 kobject_name(&node_devices[nid]->dev.kobj));
750
751 return 0;
752 }
753
754 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
get_nid_for_pfn(unsigned long pfn)755 static int __ref get_nid_for_pfn(unsigned long pfn)
756 {
757 if (!pfn_valid_within(pfn))
758 return -1;
759 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
760 if (system_state < SYSTEM_RUNNING)
761 return early_pfn_to_nid(pfn);
762 #endif
763 return pfn_to_nid(pfn);
764 }
765
do_register_memory_block_under_node(int nid,struct memory_block * mem_blk)766 static int do_register_memory_block_under_node(int nid,
767 struct memory_block *mem_blk)
768 {
769 int ret;
770
771 /*
772 * If this memory block spans multiple nodes, we only indicate
773 * the last processed node.
774 */
775 mem_blk->nid = nid;
776
777 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
778 &mem_blk->dev.kobj,
779 kobject_name(&mem_blk->dev.kobj));
780 if (ret)
781 return ret;
782
783 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
784 &node_devices[nid]->dev.kobj,
785 kobject_name(&node_devices[nid]->dev.kobj));
786 }
787
788 /* register memory section under specified node if it spans that node */
register_mem_block_under_node_early(struct memory_block * mem_blk,void * arg)789 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
790 void *arg)
791 {
792 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
793 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
794 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
795 int nid = *(int *)arg;
796 unsigned long pfn;
797
798 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
799 int page_nid;
800
801 /*
802 * memory block could have several absent sections from start.
803 * skip pfn range from absent section
804 */
805 if (!pfn_present(pfn)) {
806 pfn = round_down(pfn + PAGES_PER_SECTION,
807 PAGES_PER_SECTION) - 1;
808 continue;
809 }
810
811 /*
812 * We need to check if page belongs to nid only at the boot
813 * case because node's ranges can be interleaved.
814 */
815 page_nid = get_nid_for_pfn(pfn);
816 if (page_nid < 0)
817 continue;
818 if (page_nid != nid)
819 continue;
820
821 return do_register_memory_block_under_node(nid, mem_blk);
822 }
823 /* mem section does not span the specified node */
824 return 0;
825 }
826
827 /*
828 * During hotplug we know that all pages in the memory block belong to the same
829 * node.
830 */
register_mem_block_under_node_hotplug(struct memory_block * mem_blk,void * arg)831 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
832 void *arg)
833 {
834 int nid = *(int *)arg;
835
836 return do_register_memory_block_under_node(nid, mem_blk);
837 }
838
839 /*
840 * Unregister a memory block device under the node it spans. Memory blocks
841 * with multiple nodes cannot be offlined and therefore also never be removed.
842 */
unregister_memory_block_under_nodes(struct memory_block * mem_blk)843 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
844 {
845 if (mem_blk->nid == NUMA_NO_NODE)
846 return;
847
848 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
849 kobject_name(&mem_blk->dev.kobj));
850 sysfs_remove_link(&mem_blk->dev.kobj,
851 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
852 }
853
link_mem_sections(int nid,unsigned long start_pfn,unsigned long end_pfn,enum meminit_context context)854 int link_mem_sections(int nid, unsigned long start_pfn, 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 return walk_memory_blocks(PFN_PHYS(start_pfn),
865 PFN_PHYS(end_pfn - start_pfn), (void *)&nid,
866 func);
867 }
868
869 #ifdef CONFIG_HUGETLBFS
870 /*
871 * Handle per node hstate attribute [un]registration on transistions
872 * to/from memoryless state.
873 */
node_hugetlb_work(struct work_struct * work)874 static void node_hugetlb_work(struct work_struct *work)
875 {
876 struct node *node = container_of(work, struct node, node_work);
877
878 /*
879 * We only get here when a node transitions to/from memoryless state.
880 * We can detect which transition occurred by examining whether the
881 * node has memory now. hugetlb_register_node() already check this
882 * so we try to register the attributes. If that fails, then the
883 * node has transitioned to memoryless, try to unregister the
884 * attributes.
885 */
886 if (!hugetlb_register_node(node))
887 hugetlb_unregister_node(node);
888 }
889
init_node_hugetlb_work(int nid)890 static void init_node_hugetlb_work(int nid)
891 {
892 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
893 }
894
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)895 static int node_memory_callback(struct notifier_block *self,
896 unsigned long action, void *arg)
897 {
898 struct memory_notify *mnb = arg;
899 int nid = mnb->status_change_nid;
900
901 switch (action) {
902 case MEM_ONLINE:
903 case MEM_OFFLINE:
904 /*
905 * offload per node hstate [un]registration to a work thread
906 * when transitioning to/from memoryless state.
907 */
908 if (nid != NUMA_NO_NODE)
909 schedule_work(&node_devices[nid]->node_work);
910 break;
911
912 case MEM_GOING_ONLINE:
913 case MEM_GOING_OFFLINE:
914 case MEM_CANCEL_ONLINE:
915 case MEM_CANCEL_OFFLINE:
916 default:
917 break;
918 }
919
920 return NOTIFY_OK;
921 }
922 #endif /* CONFIG_HUGETLBFS */
923 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
924
925 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
926 !defined(CONFIG_HUGETLBFS)
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)927 static inline int node_memory_callback(struct notifier_block *self,
928 unsigned long action, void *arg)
929 {
930 return NOTIFY_OK;
931 }
932
init_node_hugetlb_work(int nid)933 static void init_node_hugetlb_work(int nid) { }
934
935 #endif
936
__register_one_node(int nid)937 int __register_one_node(int nid)
938 {
939 int error;
940 int cpu;
941
942 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
943 if (!node_devices[nid])
944 return -ENOMEM;
945
946 error = register_node(node_devices[nid], nid);
947
948 /* link cpu under this node */
949 for_each_present_cpu(cpu) {
950 if (cpu_to_node(cpu) == nid)
951 register_cpu_under_node(cpu, nid);
952 }
953
954 INIT_LIST_HEAD(&node_devices[nid]->access_list);
955 /* initialize work queue for memory hot plug */
956 init_node_hugetlb_work(nid);
957 node_init_caches(nid);
958
959 return error;
960 }
961
unregister_one_node(int nid)962 void unregister_one_node(int nid)
963 {
964 if (!node_devices[nid])
965 return;
966
967 unregister_node(node_devices[nid]);
968 node_devices[nid] = NULL;
969 }
970
971 /*
972 * node states attributes
973 */
974
print_nodes_state(enum node_states state,char * buf)975 static ssize_t print_nodes_state(enum node_states state, char *buf)
976 {
977 int n;
978
979 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
980 nodemask_pr_args(&node_states[state]));
981 buf[n++] = '\n';
982 buf[n] = '\0';
983 return n;
984 }
985
986 struct node_attr {
987 struct device_attribute attr;
988 enum node_states state;
989 };
990
show_node_state(struct device * dev,struct device_attribute * attr,char * buf)991 static ssize_t show_node_state(struct device *dev,
992 struct device_attribute *attr, char *buf)
993 {
994 struct node_attr *na = container_of(attr, struct node_attr, attr);
995 return print_nodes_state(na->state, buf);
996 }
997
998 #define _NODE_ATTR(name, state) \
999 { __ATTR(name, 0444, show_node_state, NULL), state }
1000
1001 static struct node_attr node_state_attr[] = {
1002 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1003 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1004 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1005 #ifdef CONFIG_HIGHMEM
1006 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1007 #endif
1008 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1009 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1010 };
1011
1012 static struct attribute *node_state_attrs[] = {
1013 &node_state_attr[N_POSSIBLE].attr.attr,
1014 &node_state_attr[N_ONLINE].attr.attr,
1015 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
1016 #ifdef CONFIG_HIGHMEM
1017 &node_state_attr[N_HIGH_MEMORY].attr.attr,
1018 #endif
1019 &node_state_attr[N_MEMORY].attr.attr,
1020 &node_state_attr[N_CPU].attr.attr,
1021 NULL
1022 };
1023
1024 static struct attribute_group memory_root_attr_group = {
1025 .attrs = node_state_attrs,
1026 };
1027
1028 static const struct attribute_group *cpu_root_attr_groups[] = {
1029 &memory_root_attr_group,
1030 NULL,
1031 };
1032
1033 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
register_node_type(void)1034 static int __init register_node_type(void)
1035 {
1036 int ret;
1037
1038 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1039 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1040
1041 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1042 if (!ret) {
1043 static struct notifier_block node_memory_callback_nb = {
1044 .notifier_call = node_memory_callback,
1045 .priority = NODE_CALLBACK_PRI,
1046 };
1047 register_hotmemory_notifier(&node_memory_callback_nb);
1048 }
1049
1050 /*
1051 * Note: we're not going to unregister the node class if we fail
1052 * to register the node state class attribute files.
1053 */
1054 return ret;
1055 }
1056 postcore_initcall(register_node_type);
1057