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 dev->parent = &node->dev;
266 dev->release = node_cache_release;
267 if (dev_set_name(dev, "memory_side_cache"))
268 goto free_dev;
269
270 if (device_register(dev))
271 goto free_name;
272
273 pm_runtime_no_callbacks(dev);
274 node->cache_dev = dev;
275 return;
276 free_name:
277 kfree_const(dev->kobj.name);
278 free_dev:
279 kfree(dev);
280 }
281
282 /**
283 * node_add_cache() - add cache attribute to a memory node
284 * @nid: Node identifier that has new cache attributes
285 * @cache_attrs: Attributes for the cache being added
286 */
node_add_cache(unsigned int nid,struct node_cache_attrs * cache_attrs)287 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
288 {
289 struct node_cache_info *info;
290 struct device *dev;
291 struct node *node;
292
293 if (!node_online(nid) || !node_devices[nid])
294 return;
295
296 node = node_devices[nid];
297 list_for_each_entry(info, &node->cache_attrs, node) {
298 if (info->cache_attrs.level == cache_attrs->level) {
299 dev_warn(&node->dev,
300 "attempt to add duplicate cache level:%d\n",
301 cache_attrs->level);
302 return;
303 }
304 }
305
306 if (!node->cache_dev)
307 node_init_cache_dev(node);
308 if (!node->cache_dev)
309 return;
310
311 info = kzalloc(sizeof(*info), GFP_KERNEL);
312 if (!info)
313 return;
314
315 dev = &info->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 free_cache;
321
322 info->cache_attrs = *cache_attrs;
323 if (device_register(dev)) {
324 dev_warn(&node->dev, "failed to add cache level:%d\n",
325 cache_attrs->level);
326 goto free_name;
327 }
328 pm_runtime_no_callbacks(dev);
329 list_add_tail(&info->node, &node->cache_attrs);
330 return;
331 free_name:
332 kfree_const(dev->kobj.name);
333 free_cache:
334 kfree(info);
335 }
336
node_remove_caches(struct node * node)337 static void node_remove_caches(struct node *node)
338 {
339 struct node_cache_info *info, *next;
340
341 if (!node->cache_dev)
342 return;
343
344 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
345 list_del(&info->node);
346 device_unregister(&info->dev);
347 }
348 device_unregister(node->cache_dev);
349 }
350
node_init_caches(unsigned int nid)351 static void node_init_caches(unsigned int nid)
352 {
353 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
354 }
355 #else
node_init_caches(unsigned int nid)356 static void node_init_caches(unsigned int nid) { }
node_remove_caches(struct node * node)357 static void node_remove_caches(struct node *node) { }
358 #endif
359
360 #define K(x) ((x) << (PAGE_SHIFT - 10))
node_read_meminfo(struct device * dev,struct device_attribute * attr,char * buf)361 static ssize_t node_read_meminfo(struct device *dev,
362 struct device_attribute *attr, char *buf)
363 {
364 int n;
365 int nid = dev->id;
366 struct pglist_data *pgdat = NODE_DATA(nid);
367 struct sysinfo i;
368 unsigned long sreclaimable, sunreclaimable;
369
370 si_meminfo_node(&i, nid);
371 sreclaimable = node_page_state(pgdat, NR_SLAB_RECLAIMABLE);
372 sunreclaimable = node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE);
373 n = sprintf(buf,
374 "Node %d MemTotal: %8lu kB\n"
375 "Node %d MemFree: %8lu kB\n"
376 "Node %d MemUsed: %8lu kB\n"
377 "Node %d Active: %8lu kB\n"
378 "Node %d Inactive: %8lu kB\n"
379 "Node %d Active(anon): %8lu kB\n"
380 "Node %d Inactive(anon): %8lu kB\n"
381 "Node %d Active(file): %8lu kB\n"
382 "Node %d Inactive(file): %8lu kB\n"
383 "Node %d Unevictable: %8lu kB\n"
384 "Node %d Mlocked: %8lu kB\n",
385 nid, K(i.totalram),
386 nid, K(i.freeram),
387 nid, K(i.totalram - i.freeram),
388 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
389 node_page_state(pgdat, NR_ACTIVE_FILE)),
390 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
391 node_page_state(pgdat, NR_INACTIVE_FILE)),
392 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
393 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
394 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
395 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
396 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
397 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
398
399 #ifdef CONFIG_HIGHMEM
400 n += sprintf(buf + n,
401 "Node %d HighTotal: %8lu kB\n"
402 "Node %d HighFree: %8lu kB\n"
403 "Node %d LowTotal: %8lu kB\n"
404 "Node %d LowFree: %8lu kB\n",
405 nid, K(i.totalhigh),
406 nid, K(i.freehigh),
407 nid, K(i.totalram - i.totalhigh),
408 nid, K(i.freeram - i.freehigh));
409 #endif
410 n += sprintf(buf + n,
411 "Node %d Dirty: %8lu kB\n"
412 "Node %d Writeback: %8lu kB\n"
413 "Node %d FilePages: %8lu kB\n"
414 "Node %d Mapped: %8lu kB\n"
415 "Node %d AnonPages: %8lu kB\n"
416 "Node %d Shmem: %8lu kB\n"
417 "Node %d KernelStack: %8lu kB\n"
418 #ifdef CONFIG_SHADOW_CALL_STACK
419 "Node %d ShadowCallStack:%8lu kB\n"
420 #endif
421 "Node %d PageTables: %8lu kB\n"
422 "Node %d NFS_Unstable: %8lu kB\n"
423 "Node %d Bounce: %8lu kB\n"
424 "Node %d WritebackTmp: %8lu kB\n"
425 "Node %d KReclaimable: %8lu kB\n"
426 "Node %d Slab: %8lu kB\n"
427 "Node %d SReclaimable: %8lu kB\n"
428 "Node %d SUnreclaim: %8lu kB\n"
429 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
430 "Node %d AnonHugePages: %8lu kB\n"
431 "Node %d ShmemHugePages: %8lu kB\n"
432 "Node %d ShmemPmdMapped: %8lu kB\n"
433 "Node %d FileHugePages: %8lu kB\n"
434 "Node %d FilePmdMapped: %8lu kB\n"
435 #endif
436 ,
437 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
438 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
439 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
440 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
441 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
442 nid, K(i.sharedram),
443 nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
444 #ifdef CONFIG_SHADOW_CALL_STACK
445 nid, sum_zone_node_page_state(nid, NR_KERNEL_SCS_BYTES) / 1024,
446 #endif
447 nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
448 nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
449 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
450 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
451 nid, K(sreclaimable +
452 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
453 nid, K(sreclaimable + sunreclaimable),
454 nid, K(sreclaimable),
455 nid, K(sunreclaimable)
456 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
457 ,
458 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
459 HPAGE_PMD_NR),
460 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
461 HPAGE_PMD_NR),
462 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
463 HPAGE_PMD_NR),
464 nid, K(node_page_state(pgdat, NR_FILE_THPS) *
465 HPAGE_PMD_NR),
466 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED) *
467 HPAGE_PMD_NR)
468 #endif
469 );
470 n += hugetlb_report_node_meminfo(nid, buf + n);
471 return n;
472 }
473
474 #undef K
475 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
476
node_read_numastat(struct device * dev,struct device_attribute * attr,char * buf)477 static ssize_t node_read_numastat(struct device *dev,
478 struct device_attribute *attr, char *buf)
479 {
480 return sprintf(buf,
481 "numa_hit %lu\n"
482 "numa_miss %lu\n"
483 "numa_foreign %lu\n"
484 "interleave_hit %lu\n"
485 "local_node %lu\n"
486 "other_node %lu\n",
487 sum_zone_numa_state(dev->id, NUMA_HIT),
488 sum_zone_numa_state(dev->id, NUMA_MISS),
489 sum_zone_numa_state(dev->id, NUMA_FOREIGN),
490 sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
491 sum_zone_numa_state(dev->id, NUMA_LOCAL),
492 sum_zone_numa_state(dev->id, NUMA_OTHER));
493 }
494 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
495
node_read_vmstat(struct device * dev,struct device_attribute * attr,char * buf)496 static ssize_t node_read_vmstat(struct device *dev,
497 struct device_attribute *attr, char *buf)
498 {
499 int nid = dev->id;
500 struct pglist_data *pgdat = NODE_DATA(nid);
501 int i;
502 int n = 0;
503
504 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
505 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
506 sum_zone_node_page_state(nid, i));
507
508 #ifdef CONFIG_NUMA
509 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
510 n += sprintf(buf+n, "%s %lu\n",
511 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
512 sum_zone_numa_state(nid, i));
513 #endif
514
515 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
516 n += sprintf(buf+n, "%s %lu\n",
517 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
518 NR_VM_NUMA_STAT_ITEMS],
519 node_page_state(pgdat, i));
520
521 return n;
522 }
523 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
524
node_read_distance(struct device * dev,struct device_attribute * attr,char * buf)525 static ssize_t node_read_distance(struct device *dev,
526 struct device_attribute *attr, char *buf)
527 {
528 int nid = dev->id;
529 int len = 0;
530 int i;
531
532 /*
533 * buf is currently PAGE_SIZE in length and each node needs 4 chars
534 * at the most (distance + space or newline).
535 */
536 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
537
538 for_each_online_node(i)
539 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
540
541 len += sprintf(buf + len, "\n");
542 return len;
543 }
544 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
545
546 static struct attribute *node_dev_attrs[] = {
547 &dev_attr_cpumap.attr,
548 &dev_attr_cpulist.attr,
549 &dev_attr_meminfo.attr,
550 &dev_attr_numastat.attr,
551 &dev_attr_distance.attr,
552 &dev_attr_vmstat.attr,
553 NULL
554 };
555 ATTRIBUTE_GROUPS(node_dev);
556
557 #ifdef CONFIG_HUGETLBFS
558 /*
559 * hugetlbfs per node attributes registration interface:
560 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
561 * it will register its per node attributes for all online nodes with
562 * memory. It will also call register_hugetlbfs_with_node(), below, to
563 * register its attribute registration functions with this node driver.
564 * Once these hooks have been initialized, the node driver will call into
565 * the hugetlb module to [un]register attributes for hot-plugged nodes.
566 */
567 static node_registration_func_t __hugetlb_register_node;
568 static node_registration_func_t __hugetlb_unregister_node;
569
hugetlb_register_node(struct node * node)570 static inline bool hugetlb_register_node(struct node *node)
571 {
572 if (__hugetlb_register_node &&
573 node_state(node->dev.id, N_MEMORY)) {
574 __hugetlb_register_node(node);
575 return true;
576 }
577 return false;
578 }
579
hugetlb_unregister_node(struct node * node)580 static inline void hugetlb_unregister_node(struct node *node)
581 {
582 if (__hugetlb_unregister_node)
583 __hugetlb_unregister_node(node);
584 }
585
register_hugetlbfs_with_node(node_registration_func_t doregister,node_registration_func_t unregister)586 void register_hugetlbfs_with_node(node_registration_func_t doregister,
587 node_registration_func_t unregister)
588 {
589 __hugetlb_register_node = doregister;
590 __hugetlb_unregister_node = unregister;
591 }
592 #else
hugetlb_register_node(struct node * node)593 static inline void hugetlb_register_node(struct node *node) {}
594
hugetlb_unregister_node(struct node * node)595 static inline void hugetlb_unregister_node(struct node *node) {}
596 #endif
597
node_device_release(struct device * dev)598 static void node_device_release(struct device *dev)
599 {
600 struct node *node = to_node(dev);
601
602 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
603 /*
604 * We schedule the work only when a memory section is
605 * onlined/offlined on this node. When we come here,
606 * all the memory on this node has been offlined,
607 * so we won't enqueue new work to this work.
608 *
609 * The work is using node->node_work, so we should
610 * flush work before freeing the memory.
611 */
612 flush_work(&node->node_work);
613 #endif
614 kfree(node);
615 }
616
617 /*
618 * register_node - Setup a sysfs device for a node.
619 * @num - Node number to use when creating the device.
620 *
621 * Initialize and register the node device.
622 */
register_node(struct node * node,int num)623 static int register_node(struct node *node, int num)
624 {
625 int error;
626
627 node->dev.id = num;
628 node->dev.bus = &node_subsys;
629 node->dev.release = node_device_release;
630 node->dev.groups = node_dev_groups;
631 error = device_register(&node->dev);
632
633 if (error)
634 put_device(&node->dev);
635 else {
636 hugetlb_register_node(node);
637
638 compaction_register_node(node);
639 }
640 return error;
641 }
642
643 /**
644 * unregister_node - unregister a node device
645 * @node: node going away
646 *
647 * Unregisters a node device @node. All the devices on the node must be
648 * unregistered before calling this function.
649 */
unregister_node(struct node * node)650 void unregister_node(struct node *node)
651 {
652 hugetlb_unregister_node(node); /* no-op, if memoryless node */
653 node_remove_accesses(node);
654 node_remove_caches(node);
655 device_unregister(&node->dev);
656 }
657
658 struct node *node_devices[MAX_NUMNODES];
659
660 /*
661 * register cpu under node
662 */
register_cpu_under_node(unsigned int cpu,unsigned int nid)663 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
664 {
665 int ret;
666 struct device *obj;
667
668 if (!node_online(nid))
669 return 0;
670
671 obj = get_cpu_device(cpu);
672 if (!obj)
673 return 0;
674
675 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
676 &obj->kobj,
677 kobject_name(&obj->kobj));
678 if (ret)
679 return ret;
680
681 return sysfs_create_link(&obj->kobj,
682 &node_devices[nid]->dev.kobj,
683 kobject_name(&node_devices[nid]->dev.kobj));
684 }
685
686 /**
687 * register_memory_node_under_compute_node - link memory node to its compute
688 * node for a given access class.
689 * @mem_nid: Memory node number
690 * @cpu_nid: Cpu node number
691 * @access: Access class to register
692 *
693 * Description:
694 * For use with platforms that may have separate memory and compute nodes.
695 * This function will export node relationships linking which memory
696 * initiator nodes can access memory targets at a given ranked access
697 * class.
698 */
register_memory_node_under_compute_node(unsigned int mem_nid,unsigned int cpu_nid,unsigned access)699 int register_memory_node_under_compute_node(unsigned int mem_nid,
700 unsigned int cpu_nid,
701 unsigned access)
702 {
703 struct node *init_node, *targ_node;
704 struct node_access_nodes *initiator, *target;
705 int ret;
706
707 if (!node_online(cpu_nid) || !node_online(mem_nid))
708 return -ENODEV;
709
710 init_node = node_devices[cpu_nid];
711 targ_node = node_devices[mem_nid];
712 initiator = node_init_node_access(init_node, access);
713 target = node_init_node_access(targ_node, access);
714 if (!initiator || !target)
715 return -ENOMEM;
716
717 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
718 &targ_node->dev.kobj,
719 dev_name(&targ_node->dev));
720 if (ret)
721 return ret;
722
723 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
724 &init_node->dev.kobj,
725 dev_name(&init_node->dev));
726 if (ret)
727 goto err;
728
729 return 0;
730 err:
731 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
732 dev_name(&targ_node->dev));
733 return ret;
734 }
735
unregister_cpu_under_node(unsigned int cpu,unsigned int nid)736 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
737 {
738 struct device *obj;
739
740 if (!node_online(nid))
741 return 0;
742
743 obj = get_cpu_device(cpu);
744 if (!obj)
745 return 0;
746
747 sysfs_remove_link(&node_devices[nid]->dev.kobj,
748 kobject_name(&obj->kobj));
749 sysfs_remove_link(&obj->kobj,
750 kobject_name(&node_devices[nid]->dev.kobj));
751
752 return 0;
753 }
754
755 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
get_nid_for_pfn(unsigned long pfn)756 static int __ref get_nid_for_pfn(unsigned long pfn)
757 {
758 if (!pfn_valid_within(pfn))
759 return -1;
760 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
761 if (system_state < SYSTEM_RUNNING)
762 return early_pfn_to_nid(pfn);
763 #endif
764 return pfn_to_nid(pfn);
765 }
766
767 /* register memory section under specified node if it spans that node */
register_mem_sect_under_node(struct memory_block * mem_blk,void * arg)768 static int register_mem_sect_under_node(struct memory_block *mem_blk,
769 void *arg)
770 {
771 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
772 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
773 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
774 int ret, nid = *(int *)arg;
775 unsigned long pfn;
776
777 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
778 int page_nid;
779
780 /*
781 * memory block could have several absent sections from start.
782 * skip pfn range from absent section
783 */
784 if (!pfn_present(pfn)) {
785 pfn = round_down(pfn + PAGES_PER_SECTION,
786 PAGES_PER_SECTION) - 1;
787 continue;
788 }
789
790 /*
791 * We need to check if page belongs to nid only for the boot
792 * case, during hotplug we know that all pages in the memory
793 * block belong to the same node.
794 */
795 if (system_state == SYSTEM_BOOTING) {
796 page_nid = get_nid_for_pfn(pfn);
797 if (page_nid < 0)
798 continue;
799 if (page_nid != nid)
800 continue;
801 }
802
803 /*
804 * If this memory block spans multiple nodes, we only indicate
805 * the last processed node.
806 */
807 mem_blk->nid = nid;
808
809 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
810 &mem_blk->dev.kobj,
811 kobject_name(&mem_blk->dev.kobj));
812 if (ret)
813 return ret;
814
815 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
816 &node_devices[nid]->dev.kobj,
817 kobject_name(&node_devices[nid]->dev.kobj));
818 }
819 /* mem section does not span the specified node */
820 return 0;
821 }
822
823 /*
824 * Unregister a memory block device under the node it spans. Memory blocks
825 * with multiple nodes cannot be offlined and therefore also never be removed.
826 */
unregister_memory_block_under_nodes(struct memory_block * mem_blk)827 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
828 {
829 if (mem_blk->nid == NUMA_NO_NODE)
830 return;
831
832 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
833 kobject_name(&mem_blk->dev.kobj));
834 sysfs_remove_link(&mem_blk->dev.kobj,
835 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
836 }
837
link_mem_sections(int nid,unsigned long start_pfn,unsigned long end_pfn)838 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn)
839 {
840 return walk_memory_blocks(PFN_PHYS(start_pfn),
841 PFN_PHYS(end_pfn - start_pfn), (void *)&nid,
842 register_mem_sect_under_node);
843 }
844
845 #ifdef CONFIG_HUGETLBFS
846 /*
847 * Handle per node hstate attribute [un]registration on transistions
848 * to/from memoryless state.
849 */
node_hugetlb_work(struct work_struct * work)850 static void node_hugetlb_work(struct work_struct *work)
851 {
852 struct node *node = container_of(work, struct node, node_work);
853
854 /*
855 * We only get here when a node transitions to/from memoryless state.
856 * We can detect which transition occurred by examining whether the
857 * node has memory now. hugetlb_register_node() already check this
858 * so we try to register the attributes. If that fails, then the
859 * node has transitioned to memoryless, try to unregister the
860 * attributes.
861 */
862 if (!hugetlb_register_node(node))
863 hugetlb_unregister_node(node);
864 }
865
init_node_hugetlb_work(int nid)866 static void init_node_hugetlb_work(int nid)
867 {
868 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
869 }
870
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)871 static int node_memory_callback(struct notifier_block *self,
872 unsigned long action, void *arg)
873 {
874 struct memory_notify *mnb = arg;
875 int nid = mnb->status_change_nid;
876
877 switch (action) {
878 case MEM_ONLINE:
879 case MEM_OFFLINE:
880 /*
881 * offload per node hstate [un]registration to a work thread
882 * when transitioning to/from memoryless state.
883 */
884 if (nid != NUMA_NO_NODE)
885 schedule_work(&node_devices[nid]->node_work);
886 break;
887
888 case MEM_GOING_ONLINE:
889 case MEM_GOING_OFFLINE:
890 case MEM_CANCEL_ONLINE:
891 case MEM_CANCEL_OFFLINE:
892 default:
893 break;
894 }
895
896 return NOTIFY_OK;
897 }
898 #endif /* CONFIG_HUGETLBFS */
899 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
900
901 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
902 !defined(CONFIG_HUGETLBFS)
node_memory_callback(struct notifier_block * self,unsigned long action,void * arg)903 static inline int node_memory_callback(struct notifier_block *self,
904 unsigned long action, void *arg)
905 {
906 return NOTIFY_OK;
907 }
908
init_node_hugetlb_work(int nid)909 static void init_node_hugetlb_work(int nid) { }
910
911 #endif
912
__register_one_node(int nid)913 int __register_one_node(int nid)
914 {
915 int error;
916 int cpu;
917
918 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
919 if (!node_devices[nid])
920 return -ENOMEM;
921
922 error = register_node(node_devices[nid], nid);
923
924 /* link cpu under this node */
925 for_each_present_cpu(cpu) {
926 if (cpu_to_node(cpu) == nid)
927 register_cpu_under_node(cpu, nid);
928 }
929
930 INIT_LIST_HEAD(&node_devices[nid]->access_list);
931 /* initialize work queue for memory hot plug */
932 init_node_hugetlb_work(nid);
933 node_init_caches(nid);
934
935 return error;
936 }
937
unregister_one_node(int nid)938 void unregister_one_node(int nid)
939 {
940 if (!node_devices[nid])
941 return;
942
943 unregister_node(node_devices[nid]);
944 node_devices[nid] = NULL;
945 }
946
947 /*
948 * node states attributes
949 */
950
print_nodes_state(enum node_states state,char * buf)951 static ssize_t print_nodes_state(enum node_states state, char *buf)
952 {
953 int n;
954
955 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
956 nodemask_pr_args(&node_states[state]));
957 buf[n++] = '\n';
958 buf[n] = '\0';
959 return n;
960 }
961
962 struct node_attr {
963 struct device_attribute attr;
964 enum node_states state;
965 };
966
show_node_state(struct device * dev,struct device_attribute * attr,char * buf)967 static ssize_t show_node_state(struct device *dev,
968 struct device_attribute *attr, char *buf)
969 {
970 struct node_attr *na = container_of(attr, struct node_attr, attr);
971 return print_nodes_state(na->state, buf);
972 }
973
974 #define _NODE_ATTR(name, state) \
975 { __ATTR(name, 0444, show_node_state, NULL), state }
976
977 static struct node_attr node_state_attr[] = {
978 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
979 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
980 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
981 #ifdef CONFIG_HIGHMEM
982 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
983 #endif
984 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
985 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
986 };
987
988 static struct attribute *node_state_attrs[] = {
989 &node_state_attr[N_POSSIBLE].attr.attr,
990 &node_state_attr[N_ONLINE].attr.attr,
991 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
992 #ifdef CONFIG_HIGHMEM
993 &node_state_attr[N_HIGH_MEMORY].attr.attr,
994 #endif
995 &node_state_attr[N_MEMORY].attr.attr,
996 &node_state_attr[N_CPU].attr.attr,
997 NULL
998 };
999
1000 static struct attribute_group memory_root_attr_group = {
1001 .attrs = node_state_attrs,
1002 };
1003
1004 static const struct attribute_group *cpu_root_attr_groups[] = {
1005 &memory_root_attr_group,
1006 NULL,
1007 };
1008
1009 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
register_node_type(void)1010 static int __init register_node_type(void)
1011 {
1012 int ret;
1013
1014 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1015 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1016
1017 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1018 if (!ret) {
1019 static struct notifier_block node_memory_callback_nb = {
1020 .notifier_call = node_memory_callback,
1021 .priority = NODE_CALLBACK_PRI,
1022 };
1023 register_hotmemory_notifier(&node_memory_callback_nb);
1024 }
1025
1026 /*
1027 * Note: we're not going to unregister the node class if we fail
1028 * to register the node state class attribute files.
1029 */
1030 return ret;
1031 }
1032 postcore_initcall(register_node_type);
1033