Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Basic Node interface support
4 */
5
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/mm.h>
9#include <linux/memory.h>
10#include <linux/vmstat.h>
11#include <linux/notifier.h>
12#include <linux/node.h>
13#include <linux/hugetlb.h>
14#include <linux/compaction.h>
15#include <linux/cpumask.h>
16#include <linux/topology.h>
17#include <linux/nodemask.h>
18#include <linux/cpu.h>
19#include <linux/device.h>
20#include <linux/pm_runtime.h>
21#include <linux/swap.h>
22#include <linux/slab.h>
23#include <linux/hugetlb.h>
24
25static struct bus_type node_subsys = {
26 .name = "node",
27 .dev_name = "node",
28};
29
30static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
31 struct bin_attribute *attr, char *buf,
32 loff_t off, size_t count)
33{
34 struct device *dev = kobj_to_dev(kobj);
35 struct node *node_dev = to_node(dev);
36 cpumask_var_t mask;
37 ssize_t n;
38
39 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
40 return 0;
41
42 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
43 n = cpumap_print_bitmask_to_buf(buf, mask, off, count);
44 free_cpumask_var(mask);
45
46 return n;
47}
48
49static BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
50
51static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
52 struct bin_attribute *attr, char *buf,
53 loff_t off, size_t count)
54{
55 struct device *dev = kobj_to_dev(kobj);
56 struct node *node_dev = to_node(dev);
57 cpumask_var_t mask;
58 ssize_t n;
59
60 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
61 return 0;
62
63 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
64 n = cpumap_print_list_to_buf(buf, mask, off, count);
65 free_cpumask_var(mask);
66
67 return n;
68}
69
70static BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
71
72/**
73 * struct node_access_nodes - Access class device to hold user visible
74 * relationships to other nodes.
75 * @dev: Device for this memory access class
76 * @list_node: List element in the node's access list
77 * @access: The access class rank
78 * @hmem_attrs: Heterogeneous memory performance attributes
79 */
80struct node_access_nodes {
81 struct device dev;
82 struct list_head list_node;
83 unsigned int access;
84#ifdef CONFIG_HMEM_REPORTING
85 struct node_hmem_attrs hmem_attrs;
86#endif
87};
88#define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
89
90static struct attribute *node_init_access_node_attrs[] = {
91 NULL,
92};
93
94static struct attribute *node_targ_access_node_attrs[] = {
95 NULL,
96};
97
98static const struct attribute_group initiators = {
99 .name = "initiators",
100 .attrs = node_init_access_node_attrs,
101};
102
103static const struct attribute_group targets = {
104 .name = "targets",
105 .attrs = node_targ_access_node_attrs,
106};
107
108static const struct attribute_group *node_access_node_groups[] = {
109 &initiators,
110 &targets,
111 NULL,
112};
113
114static void node_remove_accesses(struct node *node)
115{
116 struct node_access_nodes *c, *cnext;
117
118 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
119 list_del(&c->list_node);
120 device_unregister(&c->dev);
121 }
122}
123
124static void node_access_release(struct device *dev)
125{
126 kfree(to_access_nodes(dev));
127}
128
129static struct node_access_nodes *node_init_node_access(struct node *node,
130 unsigned int access)
131{
132 struct node_access_nodes *access_node;
133 struct device *dev;
134
135 list_for_each_entry(access_node, &node->access_list, list_node)
136 if (access_node->access == access)
137 return access_node;
138
139 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
140 if (!access_node)
141 return NULL;
142
143 access_node->access = access;
144 dev = &access_node->dev;
145 dev->parent = &node->dev;
146 dev->release = node_access_release;
147 dev->groups = node_access_node_groups;
148 if (dev_set_name(dev, "access%u", access))
149 goto free;
150
151 if (device_register(dev))
152 goto free_name;
153
154 pm_runtime_no_callbacks(dev);
155 list_add_tail(&access_node->list_node, &node->access_list);
156 return access_node;
157free_name:
158 kfree_const(dev->kobj.name);
159free:
160 kfree(access_node);
161 return NULL;
162}
163
164#ifdef CONFIG_HMEM_REPORTING
165#define ACCESS_ATTR(name) \
166static ssize_t name##_show(struct device *dev, \
167 struct device_attribute *attr, \
168 char *buf) \
169{ \
170 return sysfs_emit(buf, "%u\n", \
171 to_access_nodes(dev)->hmem_attrs.name); \
172} \
173static DEVICE_ATTR_RO(name)
174
175ACCESS_ATTR(read_bandwidth);
176ACCESS_ATTR(read_latency);
177ACCESS_ATTR(write_bandwidth);
178ACCESS_ATTR(write_latency);
179
180static struct attribute *access_attrs[] = {
181 &dev_attr_read_bandwidth.attr,
182 &dev_attr_read_latency.attr,
183 &dev_attr_write_bandwidth.attr,
184 &dev_attr_write_latency.attr,
185 NULL,
186};
187
188/**
189 * node_set_perf_attrs - Set the performance values for given access class
190 * @nid: Node identifier to be set
191 * @hmem_attrs: Heterogeneous memory performance attributes
192 * @access: The access class the for the given attributes
193 */
194void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
195 unsigned int access)
196{
197 struct node_access_nodes *c;
198 struct node *node;
199 int i;
200
201 if (WARN_ON_ONCE(!node_online(nid)))
202 return;
203
204 node = node_devices[nid];
205 c = node_init_node_access(node, access);
206 if (!c)
207 return;
208
209 c->hmem_attrs = *hmem_attrs;
210 for (i = 0; access_attrs[i] != NULL; i++) {
211 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
212 "initiators")) {
213 pr_info("failed to add performance attribute to node %d\n",
214 nid);
215 break;
216 }
217 }
218}
219
220/**
221 * struct node_cache_info - Internal tracking for memory node caches
222 * @dev: Device represeting the cache level
223 * @node: List element for tracking in the node
224 * @cache_attrs:Attributes for this cache level
225 */
226struct 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) \
234static 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} \
241static DEVICE_ATTR_RO(name);
242
243CACHE_ATTR(size, "%llu")
244CACHE_ATTR(line_size, "%u")
245CACHE_ATTR(indexing, "%u")
246CACHE_ATTR(write_policy, "%u")
247
248static struct attribute *cache_attrs[] = {
249 &dev_attr_indexing.attr,
250 &dev_attr_size.attr,
251 &dev_attr_line_size.attr,
252 &dev_attr_write_policy.attr,
253 NULL,
254};
255ATTRIBUTE_GROUPS(cache);
256
257static void node_cache_release(struct device *dev)
258{
259 kfree(dev);
260}
261
262static void node_cacheinfo_release(struct device *dev)
263{
264 struct node_cache_info *info = to_cache_info(dev);
265 kfree(info);
266}
267
268static void node_init_cache_dev(struct node *node)
269{
270 struct device *dev;
271
272 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
273 if (!dev)
274 return;
275
276 device_initialize(dev);
277 dev->parent = &node->dev;
278 dev->release = node_cache_release;
279 if (dev_set_name(dev, "memory_side_cache"))
280 goto put_device;
281
282 if (device_add(dev))
283 goto put_device;
284
285 pm_runtime_no_callbacks(dev);
286 node->cache_dev = dev;
287 return;
288put_device:
289 put_device(dev);
290}
291
292/**
293 * node_add_cache() - add cache attribute to a memory node
294 * @nid: Node identifier that has new cache attributes
295 * @cache_attrs: Attributes for the cache being added
296 */
297void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
298{
299 struct node_cache_info *info;
300 struct device *dev;
301 struct node *node;
302
303 if (!node_online(nid) || !node_devices[nid])
304 return;
305
306 node = node_devices[nid];
307 list_for_each_entry(info, &node->cache_attrs, node) {
308 if (info->cache_attrs.level == cache_attrs->level) {
309 dev_warn(&node->dev,
310 "attempt to add duplicate cache level:%d\n",
311 cache_attrs->level);
312 return;
313 }
314 }
315
316 if (!node->cache_dev)
317 node_init_cache_dev(node);
318 if (!node->cache_dev)
319 return;
320
321 info = kzalloc(sizeof(*info), GFP_KERNEL);
322 if (!info)
323 return;
324
325 dev = &info->dev;
326 device_initialize(dev);
327 dev->parent = node->cache_dev;
328 dev->release = node_cacheinfo_release;
329 dev->groups = cache_groups;
330 if (dev_set_name(dev, "index%d", cache_attrs->level))
331 goto put_device;
332
333 info->cache_attrs = *cache_attrs;
334 if (device_add(dev)) {
335 dev_warn(&node->dev, "failed to add cache level:%d\n",
336 cache_attrs->level);
337 goto put_device;
338 }
339 pm_runtime_no_callbacks(dev);
340 list_add_tail(&info->node, &node->cache_attrs);
341 return;
342put_device:
343 put_device(dev);
344}
345
346static void node_remove_caches(struct node *node)
347{
348 struct node_cache_info *info, *next;
349
350 if (!node->cache_dev)
351 return;
352
353 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
354 list_del(&info->node);
355 device_unregister(&info->dev);
356 }
357 device_unregister(node->cache_dev);
358}
359
360static void node_init_caches(unsigned int nid)
361{
362 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
363}
364#else
365static void node_init_caches(unsigned int nid) { }
366static void node_remove_caches(struct node *node) { }
367#endif
368
369#define K(x) ((x) << (PAGE_SHIFT - 10))
370static ssize_t node_read_meminfo(struct device *dev,
371 struct device_attribute *attr, char *buf)
372{
373 int len = 0;
374 int nid = dev->id;
375 struct pglist_data *pgdat = NODE_DATA(nid);
376 struct sysinfo i;
377 unsigned long sreclaimable, sunreclaimable;
378 unsigned long swapcached = 0;
379
380 si_meminfo_node(&i, nid);
381 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
382 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
383#ifdef CONFIG_SWAP
384 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
385#endif
386 len = sysfs_emit_at(buf, len,
387 "Node %d MemTotal: %8lu kB\n"
388 "Node %d MemFree: %8lu kB\n"
389 "Node %d MemUsed: %8lu kB\n"
390 "Node %d SwapCached: %8lu kB\n"
391 "Node %d Active: %8lu kB\n"
392 "Node %d Inactive: %8lu kB\n"
393 "Node %d Active(anon): %8lu kB\n"
394 "Node %d Inactive(anon): %8lu kB\n"
395 "Node %d Active(file): %8lu kB\n"
396 "Node %d Inactive(file): %8lu kB\n"
397 "Node %d Unevictable: %8lu kB\n"
398 "Node %d Mlocked: %8lu kB\n",
399 nid, K(i.totalram),
400 nid, K(i.freeram),
401 nid, K(i.totalram - i.freeram),
402 nid, K(swapcached),
403 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
404 node_page_state(pgdat, NR_ACTIVE_FILE)),
405 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
406 node_page_state(pgdat, NR_INACTIVE_FILE)),
407 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
408 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
409 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
410 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
411 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
412 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
413
414#ifdef CONFIG_HIGHMEM
415 len += sysfs_emit_at(buf, len,
416 "Node %d HighTotal: %8lu kB\n"
417 "Node %d HighFree: %8lu kB\n"
418 "Node %d LowTotal: %8lu kB\n"
419 "Node %d LowFree: %8lu kB\n",
420 nid, K(i.totalhigh),
421 nid, K(i.freehigh),
422 nid, K(i.totalram - i.totalhigh),
423 nid, K(i.freeram - i.freehigh));
424#endif
425 len += sysfs_emit_at(buf, len,
426 "Node %d Dirty: %8lu kB\n"
427 "Node %d Writeback: %8lu kB\n"
428 "Node %d FilePages: %8lu kB\n"
429 "Node %d Mapped: %8lu kB\n"
430 "Node %d AnonPages: %8lu kB\n"
431 "Node %d Shmem: %8lu kB\n"
432 "Node %d KernelStack: %8lu kB\n"
433#ifdef CONFIG_SHADOW_CALL_STACK
434 "Node %d ShadowCallStack:%8lu kB\n"
435#endif
436 "Node %d PageTables: %8lu kB\n"
437 "Node %d SecPageTables: %8lu kB\n"
438 "Node %d NFS_Unstable: %8lu kB\n"
439 "Node %d Bounce: %8lu kB\n"
440 "Node %d WritebackTmp: %8lu kB\n"
441 "Node %d KReclaimable: %8lu kB\n"
442 "Node %d Slab: %8lu kB\n"
443 "Node %d SReclaimable: %8lu kB\n"
444 "Node %d SUnreclaim: %8lu kB\n"
445#ifdef CONFIG_TRANSPARENT_HUGEPAGE
446 "Node %d AnonHugePages: %8lu kB\n"
447 "Node %d ShmemHugePages: %8lu kB\n"
448 "Node %d ShmemPmdMapped: %8lu kB\n"
449 "Node %d FileHugePages: %8lu kB\n"
450 "Node %d FilePmdMapped: %8lu kB\n"
451#endif
452 ,
453 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
454 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
455 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
456 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
457 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
458 nid, K(i.sharedram),
459 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
460#ifdef CONFIG_SHADOW_CALL_STACK
461 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
462#endif
463 nid, K(node_page_state(pgdat, NR_PAGETABLE)),
464 nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
465 nid, 0UL,
466 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
467 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
468 nid, K(sreclaimable +
469 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
470 nid, K(sreclaimable + sunreclaimable),
471 nid, K(sreclaimable),
472 nid, K(sunreclaimable)
473#ifdef CONFIG_TRANSPARENT_HUGEPAGE
474 ,
475 nid, K(node_page_state(pgdat, NR_ANON_THPS)),
476 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
477 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
478 nid, K(node_page_state(pgdat, NR_FILE_THPS)),
479 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
480#endif
481 );
482 len += hugetlb_report_node_meminfo(buf, len, nid);
483 return len;
484}
485
486#undef K
487static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
488
489static ssize_t node_read_numastat(struct device *dev,
490 struct device_attribute *attr, char *buf)
491{
492 fold_vm_numa_events();
493 return sysfs_emit(buf,
494 "numa_hit %lu\n"
495 "numa_miss %lu\n"
496 "numa_foreign %lu\n"
497 "interleave_hit %lu\n"
498 "local_node %lu\n"
499 "other_node %lu\n",
500 sum_zone_numa_event_state(dev->id, NUMA_HIT),
501 sum_zone_numa_event_state(dev->id, NUMA_MISS),
502 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
503 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
504 sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
505 sum_zone_numa_event_state(dev->id, NUMA_OTHER));
506}
507static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
508
509static ssize_t node_read_vmstat(struct device *dev,
510 struct device_attribute *attr, char *buf)
511{
512 int nid = dev->id;
513 struct pglist_data *pgdat = NODE_DATA(nid);
514 int i;
515 int len = 0;
516
517 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
518 len += sysfs_emit_at(buf, len, "%s %lu\n",
519 zone_stat_name(i),
520 sum_zone_node_page_state(nid, i));
521
522#ifdef CONFIG_NUMA
523 fold_vm_numa_events();
524 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
525 len += sysfs_emit_at(buf, len, "%s %lu\n",
526 numa_stat_name(i),
527 sum_zone_numa_event_state(nid, i));
528
529#endif
530 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
531 unsigned long pages = node_page_state_pages(pgdat, i);
532
533 if (vmstat_item_print_in_thp(i))
534 pages /= HPAGE_PMD_NR;
535 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
536 pages);
537 }
538
539 return len;
540}
541static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
542
543static ssize_t node_read_distance(struct device *dev,
544 struct device_attribute *attr, char *buf)
545{
546 int nid = dev->id;
547 int len = 0;
548 int i;
549
550 /*
551 * buf is currently PAGE_SIZE in length and each node needs 4 chars
552 * at the most (distance + space or newline).
553 */
554 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
555
556 for_each_online_node(i) {
557 len += sysfs_emit_at(buf, len, "%s%d",
558 i ? " " : "", node_distance(nid, i));
559 }
560
561 len += sysfs_emit_at(buf, len, "\n");
562 return len;
563}
564static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
565
566static struct attribute *node_dev_attrs[] = {
567 &dev_attr_meminfo.attr,
568 &dev_attr_numastat.attr,
569 &dev_attr_distance.attr,
570 &dev_attr_vmstat.attr,
571 NULL
572};
573
574static struct bin_attribute *node_dev_bin_attrs[] = {
575 &bin_attr_cpumap,
576 &bin_attr_cpulist,
577 NULL
578};
579
580static const struct attribute_group node_dev_group = {
581 .attrs = node_dev_attrs,
582 .bin_attrs = node_dev_bin_attrs
583};
584
585static const struct attribute_group *node_dev_groups[] = {
586 &node_dev_group,
587#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
588 &arch_node_dev_group,
589#endif
590 NULL
591};
592
593static void node_device_release(struct device *dev)
594{
595 kfree(to_node(dev));
596}
597
598/*
599 * register_node - Setup a sysfs device for a node.
600 * @num - Node number to use when creating the device.
601 *
602 * Initialize and register the node device.
603 */
604static int register_node(struct node *node, int num)
605{
606 int error;
607
608 node->dev.id = num;
609 node->dev.bus = &node_subsys;
610 node->dev.release = node_device_release;
611 node->dev.groups = node_dev_groups;
612 error = device_register(&node->dev);
613
614 if (error) {
615 put_device(&node->dev);
616 } else {
617 hugetlb_register_node(node);
618 compaction_register_node(node);
619 }
620
621 return error;
622}
623
624/**
625 * unregister_node - unregister a node device
626 * @node: node going away
627 *
628 * Unregisters a node device @node. All the devices on the node must be
629 * unregistered before calling this function.
630 */
631void unregister_node(struct node *node)
632{
633 hugetlb_unregister_node(node);
634 compaction_unregister_node(node);
635 node_remove_accesses(node);
636 node_remove_caches(node);
637 device_unregister(&node->dev);
638}
639
640struct node *node_devices[MAX_NUMNODES];
641
642/*
643 * register cpu under node
644 */
645int register_cpu_under_node(unsigned int cpu, unsigned int nid)
646{
647 int ret;
648 struct device *obj;
649
650 if (!node_online(nid))
651 return 0;
652
653 obj = get_cpu_device(cpu);
654 if (!obj)
655 return 0;
656
657 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
658 &obj->kobj,
659 kobject_name(&obj->kobj));
660 if (ret)
661 return ret;
662
663 return sysfs_create_link(&obj->kobj,
664 &node_devices[nid]->dev.kobj,
665 kobject_name(&node_devices[nid]->dev.kobj));
666}
667
668/**
669 * register_memory_node_under_compute_node - link memory node to its compute
670 * node for a given access class.
671 * @mem_nid: Memory node number
672 * @cpu_nid: Cpu node number
673 * @access: Access class to register
674 *
675 * Description:
676 * For use with platforms that may have separate memory and compute nodes.
677 * This function will export node relationships linking which memory
678 * initiator nodes can access memory targets at a given ranked access
679 * class.
680 */
681int register_memory_node_under_compute_node(unsigned int mem_nid,
682 unsigned int cpu_nid,
683 unsigned int access)
684{
685 struct node *init_node, *targ_node;
686 struct node_access_nodes *initiator, *target;
687 int ret;
688
689 if (!node_online(cpu_nid) || !node_online(mem_nid))
690 return -ENODEV;
691
692 init_node = node_devices[cpu_nid];
693 targ_node = node_devices[mem_nid];
694 initiator = node_init_node_access(init_node, access);
695 target = node_init_node_access(targ_node, access);
696 if (!initiator || !target)
697 return -ENOMEM;
698
699 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
700 &targ_node->dev.kobj,
701 dev_name(&targ_node->dev));
702 if (ret)
703 return ret;
704
705 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
706 &init_node->dev.kobj,
707 dev_name(&init_node->dev));
708 if (ret)
709 goto err;
710
711 return 0;
712 err:
713 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
714 dev_name(&targ_node->dev));
715 return ret;
716}
717
718int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
719{
720 struct device *obj;
721
722 if (!node_online(nid))
723 return 0;
724
725 obj = get_cpu_device(cpu);
726 if (!obj)
727 return 0;
728
729 sysfs_remove_link(&node_devices[nid]->dev.kobj,
730 kobject_name(&obj->kobj));
731 sysfs_remove_link(&obj->kobj,
732 kobject_name(&node_devices[nid]->dev.kobj));
733
734 return 0;
735}
736
737#ifdef CONFIG_MEMORY_HOTPLUG
738static int __ref get_nid_for_pfn(unsigned long pfn)
739{
740#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
741 if (system_state < SYSTEM_RUNNING)
742 return early_pfn_to_nid(pfn);
743#endif
744 return pfn_to_nid(pfn);
745}
746
747static void do_register_memory_block_under_node(int nid,
748 struct memory_block *mem_blk,
749 enum meminit_context context)
750{
751 int ret;
752
753 memory_block_add_nid(mem_blk, nid, context);
754
755 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
756 &mem_blk->dev.kobj,
757 kobject_name(&mem_blk->dev.kobj));
758 if (ret && ret != -EEXIST)
759 dev_err_ratelimited(&node_devices[nid]->dev,
760 "can't create link to %s in sysfs (%d)\n",
761 kobject_name(&mem_blk->dev.kobj), ret);
762
763 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
764 &node_devices[nid]->dev.kobj,
765 kobject_name(&node_devices[nid]->dev.kobj));
766 if (ret && ret != -EEXIST)
767 dev_err_ratelimited(&mem_blk->dev,
768 "can't create link to %s in sysfs (%d)\n",
769 kobject_name(&node_devices[nid]->dev.kobj),
770 ret);
771}
772
773/* register memory section under specified node if it spans that node */
774static int register_mem_block_under_node_early(struct memory_block *mem_blk,
775 void *arg)
776{
777 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
778 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
779 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
780 int nid = *(int *)arg;
781 unsigned long pfn;
782
783 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
784 int page_nid;
785
786 /*
787 * memory block could have several absent sections from start.
788 * skip pfn range from absent section
789 */
790 if (!pfn_in_present_section(pfn)) {
791 pfn = round_down(pfn + PAGES_PER_SECTION,
792 PAGES_PER_SECTION) - 1;
793 continue;
794 }
795
796 /*
797 * We need to check if page belongs to nid only at the boot
798 * case because node's ranges can be interleaved.
799 */
800 page_nid = get_nid_for_pfn(pfn);
801 if (page_nid < 0)
802 continue;
803 if (page_nid != nid)
804 continue;
805
806 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
807 return 0;
808 }
809 /* mem section does not span the specified node */
810 return 0;
811}
812
813/*
814 * During hotplug we know that all pages in the memory block belong to the same
815 * node.
816 */
817static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
818 void *arg)
819{
820 int nid = *(int *)arg;
821
822 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
823 return 0;
824}
825
826/*
827 * Unregister a memory block device under the node it spans. Memory blocks
828 * with multiple nodes cannot be offlined and therefore also never be removed.
829 */
830void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
831{
832 if (mem_blk->nid == NUMA_NO_NODE)
833 return;
834
835 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
836 kobject_name(&mem_blk->dev.kobj));
837 sysfs_remove_link(&mem_blk->dev.kobj,
838 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
839}
840
841void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
842 unsigned long end_pfn,
843 enum meminit_context context)
844{
845 walk_memory_blocks_func_t func;
846
847 if (context == MEMINIT_HOTPLUG)
848 func = register_mem_block_under_node_hotplug;
849 else
850 func = register_mem_block_under_node_early;
851
852 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
853 (void *)&nid, func);
854 return;
855}
856#endif /* CONFIG_MEMORY_HOTPLUG */
857
858int __register_one_node(int nid)
859{
860 int error;
861 int cpu;
862
863 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
864 if (!node_devices[nid])
865 return -ENOMEM;
866
867 error = register_node(node_devices[nid], nid);
868
869 /* link cpu under this node */
870 for_each_present_cpu(cpu) {
871 if (cpu_to_node(cpu) == nid)
872 register_cpu_under_node(cpu, nid);
873 }
874
875 INIT_LIST_HEAD(&node_devices[nid]->access_list);
876 node_init_caches(nid);
877
878 return error;
879}
880
881void unregister_one_node(int nid)
882{
883 if (!node_devices[nid])
884 return;
885
886 unregister_node(node_devices[nid]);
887 node_devices[nid] = NULL;
888}
889
890/*
891 * node states attributes
892 */
893
894struct node_attr {
895 struct device_attribute attr;
896 enum node_states state;
897};
898
899static ssize_t show_node_state(struct device *dev,
900 struct device_attribute *attr, char *buf)
901{
902 struct node_attr *na = container_of(attr, struct node_attr, attr);
903
904 return sysfs_emit(buf, "%*pbl\n",
905 nodemask_pr_args(&node_states[na->state]));
906}
907
908#define _NODE_ATTR(name, state) \
909 { __ATTR(name, 0444, show_node_state, NULL), state }
910
911static struct node_attr node_state_attr[] = {
912 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
913 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
914 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
915#ifdef CONFIG_HIGHMEM
916 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
917#endif
918 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
919 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
920 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
921 N_GENERIC_INITIATOR),
922};
923
924static struct attribute *node_state_attrs[] = {
925 &node_state_attr[N_POSSIBLE].attr.attr,
926 &node_state_attr[N_ONLINE].attr.attr,
927 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
928#ifdef CONFIG_HIGHMEM
929 &node_state_attr[N_HIGH_MEMORY].attr.attr,
930#endif
931 &node_state_attr[N_MEMORY].attr.attr,
932 &node_state_attr[N_CPU].attr.attr,
933 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
934 NULL
935};
936
937static const struct attribute_group memory_root_attr_group = {
938 .attrs = node_state_attrs,
939};
940
941static const struct attribute_group *cpu_root_attr_groups[] = {
942 &memory_root_attr_group,
943 NULL,
944};
945
946void __init node_dev_init(void)
947{
948 int ret, i;
949
950 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
951 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
952
953 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
954 if (ret)
955 panic("%s() failed to register subsystem: %d\n", __func__, ret);
956
957 /*
958 * Create all node devices, which will properly link the node
959 * to applicable memory block devices and already created cpu devices.
960 */
961 for_each_online_node(i) {
962 ret = register_one_node(i);
963 if (ret)
964 panic("%s() failed to add node: %d\n", __func__, ret);
965 }
966}
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/swap.h>
21#include <linux/slab.h>
22
23static struct bus_type node_subsys = {
24 .name = "node",
25 .dev_name = "node",
26};
27
28
29static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
30{
31 ssize_t n;
32 cpumask_var_t mask;
33 struct node *node_dev = to_node(dev);
34
35 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
36 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
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_to_pagebuf(list, buf, mask);
43 free_cpumask_var(mask);
44
45 return n;
46}
47
48static inline ssize_t node_read_cpumask(struct device *dev,
49 struct device_attribute *attr, char *buf)
50{
51 return node_read_cpumap(dev, false, buf);
52}
53static inline ssize_t node_read_cpulist(struct device *dev,
54 struct device_attribute *attr, char *buf)
55{
56 return node_read_cpumap(dev, true, buf);
57}
58
59static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
60static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
61
62#define K(x) ((x) << (PAGE_SHIFT - 10))
63static ssize_t node_read_meminfo(struct device *dev,
64 struct device_attribute *attr, char *buf)
65{
66 int n;
67 int nid = dev->id;
68 struct pglist_data *pgdat = NODE_DATA(nid);
69 struct sysinfo i;
70
71 si_meminfo_node(&i, nid);
72 n = sprintf(buf,
73 "Node %d MemTotal: %8lu kB\n"
74 "Node %d MemFree: %8lu kB\n"
75 "Node %d MemUsed: %8lu kB\n"
76 "Node %d Active: %8lu kB\n"
77 "Node %d Inactive: %8lu kB\n"
78 "Node %d Active(anon): %8lu kB\n"
79 "Node %d Inactive(anon): %8lu kB\n"
80 "Node %d Active(file): %8lu kB\n"
81 "Node %d Inactive(file): %8lu kB\n"
82 "Node %d Unevictable: %8lu kB\n"
83 "Node %d Mlocked: %8lu kB\n",
84 nid, K(i.totalram),
85 nid, K(i.freeram),
86 nid, K(i.totalram - i.freeram),
87 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
88 node_page_state(pgdat, NR_ACTIVE_FILE)),
89 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
90 node_page_state(pgdat, NR_INACTIVE_FILE)),
91 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
92 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
93 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
94 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
95 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
96 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
97
98#ifdef CONFIG_HIGHMEM
99 n += sprintf(buf + n,
100 "Node %d HighTotal: %8lu kB\n"
101 "Node %d HighFree: %8lu kB\n"
102 "Node %d LowTotal: %8lu kB\n"
103 "Node %d LowFree: %8lu kB\n",
104 nid, K(i.totalhigh),
105 nid, K(i.freehigh),
106 nid, K(i.totalram - i.totalhigh),
107 nid, K(i.freeram - i.freehigh));
108#endif
109 n += sprintf(buf + n,
110 "Node %d Dirty: %8lu kB\n"
111 "Node %d Writeback: %8lu kB\n"
112 "Node %d FilePages: %8lu kB\n"
113 "Node %d Mapped: %8lu kB\n"
114 "Node %d AnonPages: %8lu kB\n"
115 "Node %d Shmem: %8lu kB\n"
116 "Node %d KernelStack: %8lu kB\n"
117 "Node %d PageTables: %8lu kB\n"
118 "Node %d NFS_Unstable: %8lu kB\n"
119 "Node %d Bounce: %8lu kB\n"
120 "Node %d WritebackTmp: %8lu kB\n"
121 "Node %d Slab: %8lu kB\n"
122 "Node %d SReclaimable: %8lu kB\n"
123 "Node %d SUnreclaim: %8lu kB\n"
124#ifdef CONFIG_TRANSPARENT_HUGEPAGE
125 "Node %d AnonHugePages: %8lu kB\n"
126 "Node %d ShmemHugePages: %8lu kB\n"
127 "Node %d ShmemPmdMapped: %8lu kB\n"
128#endif
129 ,
130 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
131 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
132 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
133 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
134 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
135 nid, K(i.sharedram),
136 nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
137 nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
138 nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
139 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
140 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
141 nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE) +
142 node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
143 nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE)),
144#ifdef CONFIG_TRANSPARENT_HUGEPAGE
145 nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
146 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
147 HPAGE_PMD_NR),
148 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
149 HPAGE_PMD_NR),
150 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
151 HPAGE_PMD_NR));
152#else
153 nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)));
154#endif
155 n += hugetlb_report_node_meminfo(nid, buf + n);
156 return n;
157}
158
159#undef K
160static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
161
162static ssize_t node_read_numastat(struct device *dev,
163 struct device_attribute *attr, char *buf)
164{
165 return sprintf(buf,
166 "numa_hit %lu\n"
167 "numa_miss %lu\n"
168 "numa_foreign %lu\n"
169 "interleave_hit %lu\n"
170 "local_node %lu\n"
171 "other_node %lu\n",
172 sum_zone_numa_state(dev->id, NUMA_HIT),
173 sum_zone_numa_state(dev->id, NUMA_MISS),
174 sum_zone_numa_state(dev->id, NUMA_FOREIGN),
175 sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
176 sum_zone_numa_state(dev->id, NUMA_LOCAL),
177 sum_zone_numa_state(dev->id, NUMA_OTHER));
178}
179static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
180
181static ssize_t node_read_vmstat(struct device *dev,
182 struct device_attribute *attr, char *buf)
183{
184 int nid = dev->id;
185 struct pglist_data *pgdat = NODE_DATA(nid);
186 int i;
187 int n = 0;
188
189 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
190 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
191 sum_zone_node_page_state(nid, i));
192
193#ifdef CONFIG_NUMA
194 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
195 n += sprintf(buf+n, "%s %lu\n",
196 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
197 sum_zone_numa_state(nid, i));
198#endif
199
200 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
201 n += sprintf(buf+n, "%s %lu\n",
202 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
203 NR_VM_NUMA_STAT_ITEMS],
204 node_page_state(pgdat, i));
205
206 return n;
207}
208static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
209
210static ssize_t node_read_distance(struct device *dev,
211 struct device_attribute *attr, char *buf)
212{
213 int nid = dev->id;
214 int len = 0;
215 int i;
216
217 /*
218 * buf is currently PAGE_SIZE in length and each node needs 4 chars
219 * at the most (distance + space or newline).
220 */
221 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
222
223 for_each_online_node(i)
224 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
225
226 len += sprintf(buf + len, "\n");
227 return len;
228}
229static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
230
231static struct attribute *node_dev_attrs[] = {
232 &dev_attr_cpumap.attr,
233 &dev_attr_cpulist.attr,
234 &dev_attr_meminfo.attr,
235 &dev_attr_numastat.attr,
236 &dev_attr_distance.attr,
237 &dev_attr_vmstat.attr,
238 NULL
239};
240ATTRIBUTE_GROUPS(node_dev);
241
242#ifdef CONFIG_HUGETLBFS
243/*
244 * hugetlbfs per node attributes registration interface:
245 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
246 * it will register its per node attributes for all online nodes with
247 * memory. It will also call register_hugetlbfs_with_node(), below, to
248 * register its attribute registration functions with this node driver.
249 * Once these hooks have been initialized, the node driver will call into
250 * the hugetlb module to [un]register attributes for hot-plugged nodes.
251 */
252static node_registration_func_t __hugetlb_register_node;
253static node_registration_func_t __hugetlb_unregister_node;
254
255static inline bool hugetlb_register_node(struct node *node)
256{
257 if (__hugetlb_register_node &&
258 node_state(node->dev.id, N_MEMORY)) {
259 __hugetlb_register_node(node);
260 return true;
261 }
262 return false;
263}
264
265static inline void hugetlb_unregister_node(struct node *node)
266{
267 if (__hugetlb_unregister_node)
268 __hugetlb_unregister_node(node);
269}
270
271void register_hugetlbfs_with_node(node_registration_func_t doregister,
272 node_registration_func_t unregister)
273{
274 __hugetlb_register_node = doregister;
275 __hugetlb_unregister_node = unregister;
276}
277#else
278static inline void hugetlb_register_node(struct node *node) {}
279
280static inline void hugetlb_unregister_node(struct node *node) {}
281#endif
282
283static void node_device_release(struct device *dev)
284{
285 struct node *node = to_node(dev);
286
287#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
288 /*
289 * We schedule the work only when a memory section is
290 * onlined/offlined on this node. When we come here,
291 * all the memory on this node has been offlined,
292 * so we won't enqueue new work to this work.
293 *
294 * The work is using node->node_work, so we should
295 * flush work before freeing the memory.
296 */
297 flush_work(&node->node_work);
298#endif
299 kfree(node);
300}
301
302/*
303 * register_node - Setup a sysfs device for a node.
304 * @num - Node number to use when creating the device.
305 *
306 * Initialize and register the node device.
307 */
308static int register_node(struct node *node, int num)
309{
310 int error;
311
312 node->dev.id = num;
313 node->dev.bus = &node_subsys;
314 node->dev.release = node_device_release;
315 node->dev.groups = node_dev_groups;
316 error = device_register(&node->dev);
317
318 if (error)
319 put_device(&node->dev);
320 else {
321 hugetlb_register_node(node);
322
323 compaction_register_node(node);
324 }
325 return error;
326}
327
328/**
329 * unregister_node - unregister a node device
330 * @node: node going away
331 *
332 * Unregisters a node device @node. All the devices on the node must be
333 * unregistered before calling this function.
334 */
335void unregister_node(struct node *node)
336{
337 hugetlb_unregister_node(node); /* no-op, if memoryless node */
338
339 device_unregister(&node->dev);
340}
341
342struct node *node_devices[MAX_NUMNODES];
343
344/*
345 * register cpu under node
346 */
347int register_cpu_under_node(unsigned int cpu, unsigned int nid)
348{
349 int ret;
350 struct device *obj;
351
352 if (!node_online(nid))
353 return 0;
354
355 obj = get_cpu_device(cpu);
356 if (!obj)
357 return 0;
358
359 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
360 &obj->kobj,
361 kobject_name(&obj->kobj));
362 if (ret)
363 return ret;
364
365 return sysfs_create_link(&obj->kobj,
366 &node_devices[nid]->dev.kobj,
367 kobject_name(&node_devices[nid]->dev.kobj));
368}
369
370int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
371{
372 struct device *obj;
373
374 if (!node_online(nid))
375 return 0;
376
377 obj = get_cpu_device(cpu);
378 if (!obj)
379 return 0;
380
381 sysfs_remove_link(&node_devices[nid]->dev.kobj,
382 kobject_name(&obj->kobj));
383 sysfs_remove_link(&obj->kobj,
384 kobject_name(&node_devices[nid]->dev.kobj));
385
386 return 0;
387}
388
389#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
390static int __ref get_nid_for_pfn(unsigned long pfn)
391{
392 if (!pfn_valid_within(pfn))
393 return -1;
394#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
395 if (system_state < SYSTEM_RUNNING)
396 return early_pfn_to_nid(pfn);
397#endif
398 return pfn_to_nid(pfn);
399}
400
401/* register memory section under specified node if it spans that node */
402int register_mem_sect_under_node(struct memory_block *mem_blk, int nid,
403 bool check_nid)
404{
405 int ret;
406 unsigned long pfn, sect_start_pfn, sect_end_pfn;
407
408 if (!mem_blk)
409 return -EFAULT;
410
411 mem_blk->nid = nid;
412 if (!node_online(nid))
413 return 0;
414
415 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
416 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
417 sect_end_pfn += PAGES_PER_SECTION - 1;
418 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
419 int page_nid;
420
421 /*
422 * memory block could have several absent sections from start.
423 * skip pfn range from absent section
424 */
425 if (!pfn_present(pfn)) {
426 pfn = round_down(pfn + PAGES_PER_SECTION,
427 PAGES_PER_SECTION) - 1;
428 continue;
429 }
430
431 /*
432 * We need to check if page belongs to nid only for the boot
433 * case, during hotplug we know that all pages in the memory
434 * block belong to the same node.
435 */
436 if (check_nid) {
437 page_nid = get_nid_for_pfn(pfn);
438 if (page_nid < 0)
439 continue;
440 if (page_nid != nid)
441 continue;
442 }
443 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
444 &mem_blk->dev.kobj,
445 kobject_name(&mem_blk->dev.kobj));
446 if (ret)
447 return ret;
448
449 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
450 &node_devices[nid]->dev.kobj,
451 kobject_name(&node_devices[nid]->dev.kobj));
452 }
453 /* mem section does not span the specified node */
454 return 0;
455}
456
457/* unregister memory section under all nodes that it spans */
458int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
459 unsigned long phys_index)
460{
461 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
462 unsigned long pfn, sect_start_pfn, sect_end_pfn;
463
464 if (!mem_blk) {
465 NODEMASK_FREE(unlinked_nodes);
466 return -EFAULT;
467 }
468 if (!unlinked_nodes)
469 return -ENOMEM;
470 nodes_clear(*unlinked_nodes);
471
472 sect_start_pfn = section_nr_to_pfn(phys_index);
473 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
474 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
475 int nid;
476
477 nid = get_nid_for_pfn(pfn);
478 if (nid < 0)
479 continue;
480 if (!node_online(nid))
481 continue;
482 if (node_test_and_set(nid, *unlinked_nodes))
483 continue;
484 sysfs_remove_link(&node_devices[nid]->dev.kobj,
485 kobject_name(&mem_blk->dev.kobj));
486 sysfs_remove_link(&mem_blk->dev.kobj,
487 kobject_name(&node_devices[nid]->dev.kobj));
488 }
489 NODEMASK_FREE(unlinked_nodes);
490 return 0;
491}
492
493int link_mem_sections(int nid, unsigned long start_pfn, unsigned long nr_pages,
494 bool check_nid)
495{
496 unsigned long end_pfn = start_pfn + nr_pages;
497 unsigned long pfn;
498 struct memory_block *mem_blk = NULL;
499 int err = 0;
500
501 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
502 unsigned long section_nr = pfn_to_section_nr(pfn);
503 struct mem_section *mem_sect;
504 int ret;
505
506 if (!present_section_nr(section_nr))
507 continue;
508 mem_sect = __nr_to_section(section_nr);
509
510 /* same memblock ? */
511 if (mem_blk)
512 if ((section_nr >= mem_blk->start_section_nr) &&
513 (section_nr <= mem_blk->end_section_nr))
514 continue;
515
516 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
517
518 ret = register_mem_sect_under_node(mem_blk, nid, check_nid);
519 if (!err)
520 err = ret;
521
522 /* discard ref obtained in find_memory_block() */
523 }
524
525 if (mem_blk)
526 kobject_put(&mem_blk->dev.kobj);
527 return err;
528}
529
530#ifdef CONFIG_HUGETLBFS
531/*
532 * Handle per node hstate attribute [un]registration on transistions
533 * to/from memoryless state.
534 */
535static void node_hugetlb_work(struct work_struct *work)
536{
537 struct node *node = container_of(work, struct node, node_work);
538
539 /*
540 * We only get here when a node transitions to/from memoryless state.
541 * We can detect which transition occurred by examining whether the
542 * node has memory now. hugetlb_register_node() already check this
543 * so we try to register the attributes. If that fails, then the
544 * node has transitioned to memoryless, try to unregister the
545 * attributes.
546 */
547 if (!hugetlb_register_node(node))
548 hugetlb_unregister_node(node);
549}
550
551static void init_node_hugetlb_work(int nid)
552{
553 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
554}
555
556static int node_memory_callback(struct notifier_block *self,
557 unsigned long action, void *arg)
558{
559 struct memory_notify *mnb = arg;
560 int nid = mnb->status_change_nid;
561
562 switch (action) {
563 case MEM_ONLINE:
564 case MEM_OFFLINE:
565 /*
566 * offload per node hstate [un]registration to a work thread
567 * when transitioning to/from memoryless state.
568 */
569 if (nid != NUMA_NO_NODE)
570 schedule_work(&node_devices[nid]->node_work);
571 break;
572
573 case MEM_GOING_ONLINE:
574 case MEM_GOING_OFFLINE:
575 case MEM_CANCEL_ONLINE:
576 case MEM_CANCEL_OFFLINE:
577 default:
578 break;
579 }
580
581 return NOTIFY_OK;
582}
583#endif /* CONFIG_HUGETLBFS */
584#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
585
586#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
587 !defined(CONFIG_HUGETLBFS)
588static inline int node_memory_callback(struct notifier_block *self,
589 unsigned long action, void *arg)
590{
591 return NOTIFY_OK;
592}
593
594static void init_node_hugetlb_work(int nid) { }
595
596#endif
597
598int __register_one_node(int nid)
599{
600 int error;
601 int cpu;
602
603 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
604 if (!node_devices[nid])
605 return -ENOMEM;
606
607 error = register_node(node_devices[nid], nid);
608
609 /* link cpu under this node */
610 for_each_present_cpu(cpu) {
611 if (cpu_to_node(cpu) == nid)
612 register_cpu_under_node(cpu, nid);
613 }
614
615 /* initialize work queue for memory hot plug */
616 init_node_hugetlb_work(nid);
617
618 return error;
619}
620
621void unregister_one_node(int nid)
622{
623 if (!node_devices[nid])
624 return;
625
626 unregister_node(node_devices[nid]);
627 node_devices[nid] = NULL;
628}
629
630/*
631 * node states attributes
632 */
633
634static ssize_t print_nodes_state(enum node_states state, char *buf)
635{
636 int n;
637
638 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
639 nodemask_pr_args(&node_states[state]));
640 buf[n++] = '\n';
641 buf[n] = '\0';
642 return n;
643}
644
645struct node_attr {
646 struct device_attribute attr;
647 enum node_states state;
648};
649
650static ssize_t show_node_state(struct device *dev,
651 struct device_attribute *attr, char *buf)
652{
653 struct node_attr *na = container_of(attr, struct node_attr, attr);
654 return print_nodes_state(na->state, buf);
655}
656
657#define _NODE_ATTR(name, state) \
658 { __ATTR(name, 0444, show_node_state, NULL), state }
659
660static struct node_attr node_state_attr[] = {
661 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
662 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
663 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
664#ifdef CONFIG_HIGHMEM
665 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
666#endif
667 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
668 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
669};
670
671static struct attribute *node_state_attrs[] = {
672 &node_state_attr[N_POSSIBLE].attr.attr,
673 &node_state_attr[N_ONLINE].attr.attr,
674 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
675#ifdef CONFIG_HIGHMEM
676 &node_state_attr[N_HIGH_MEMORY].attr.attr,
677#endif
678 &node_state_attr[N_MEMORY].attr.attr,
679 &node_state_attr[N_CPU].attr.attr,
680 NULL
681};
682
683static struct attribute_group memory_root_attr_group = {
684 .attrs = node_state_attrs,
685};
686
687static const struct attribute_group *cpu_root_attr_groups[] = {
688 &memory_root_attr_group,
689 NULL,
690};
691
692#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
693static int __init register_node_type(void)
694{
695 int ret;
696
697 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
698 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
699
700 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
701 if (!ret) {
702 static struct notifier_block node_memory_callback_nb = {
703 .notifier_call = node_memory_callback,
704 .priority = NODE_CALLBACK_PRI,
705 };
706 register_hotmemory_notifier(&node_memory_callback_nb);
707 }
708
709 /*
710 * Note: we're not going to unregister the node class if we fail
711 * to register the node state class attribute files.
712 */
713 return ret;
714}
715postcore_initcall(register_node_type);