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
24static const struct bus_type node_subsys = {
25 .name = "node",
26 .dev_name = "node",
27};
28
29static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
30 struct bin_attribute *attr, char *buf,
31 loff_t off, size_t count)
32{
33 struct device *dev = kobj_to_dev(kobj);
34 struct node *node_dev = to_node(dev);
35 cpumask_var_t mask;
36 ssize_t n;
37
38 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
39 return 0;
40
41 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
42 n = cpumap_print_bitmask_to_buf(buf, mask, off, count);
43 free_cpumask_var(mask);
44
45 return n;
46}
47
48static BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
49
50static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
51 struct bin_attribute *attr, char *buf,
52 loff_t off, size_t count)
53{
54 struct device *dev = kobj_to_dev(kobj);
55 struct node *node_dev = to_node(dev);
56 cpumask_var_t mask;
57 ssize_t n;
58
59 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
60 return 0;
61
62 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
63 n = cpumap_print_list_to_buf(buf, mask, off, count);
64 free_cpumask_var(mask);
65
66 return n;
67}
68
69static BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
70
71/**
72 * struct node_access_nodes - Access class device to hold user visible
73 * relationships to other nodes.
74 * @dev: Device for this memory access class
75 * @list_node: List element in the node's access list
76 * @access: The access class rank
77 * @coord: Heterogeneous memory performance coordinates
78 */
79struct node_access_nodes {
80 struct device dev;
81 struct list_head list_node;
82 unsigned int access;
83#ifdef CONFIG_HMEM_REPORTING
84 struct access_coordinate coord;
85#endif
86};
87#define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
88
89static struct attribute *node_init_access_node_attrs[] = {
90 NULL,
91};
92
93static struct attribute *node_targ_access_node_attrs[] = {
94 NULL,
95};
96
97static const struct attribute_group initiators = {
98 .name = "initiators",
99 .attrs = node_init_access_node_attrs,
100};
101
102static const struct attribute_group targets = {
103 .name = "targets",
104 .attrs = node_targ_access_node_attrs,
105};
106
107static const struct attribute_group *node_access_node_groups[] = {
108 &initiators,
109 &targets,
110 NULL,
111};
112
113static void node_remove_accesses(struct node *node)
114{
115 struct node_access_nodes *c, *cnext;
116
117 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
118 list_del(&c->list_node);
119 device_unregister(&c->dev);
120 }
121}
122
123static void node_access_release(struct device *dev)
124{
125 kfree(to_access_nodes(dev));
126}
127
128static struct node_access_nodes *node_init_node_access(struct node *node,
129 unsigned int access)
130{
131 struct node_access_nodes *access_node;
132 struct device *dev;
133
134 list_for_each_entry(access_node, &node->access_list, list_node)
135 if (access_node->access == access)
136 return access_node;
137
138 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
139 if (!access_node)
140 return NULL;
141
142 access_node->access = access;
143 dev = &access_node->dev;
144 dev->parent = &node->dev;
145 dev->release = node_access_release;
146 dev->groups = node_access_node_groups;
147 if (dev_set_name(dev, "access%u", access))
148 goto free;
149
150 if (device_register(dev))
151 goto free_name;
152
153 pm_runtime_no_callbacks(dev);
154 list_add_tail(&access_node->list_node, &node->access_list);
155 return access_node;
156free_name:
157 kfree_const(dev->kobj.name);
158free:
159 kfree(access_node);
160 return NULL;
161}
162
163#ifdef CONFIG_HMEM_REPORTING
164#define ACCESS_ATTR(property) \
165static ssize_t property##_show(struct device *dev, \
166 struct device_attribute *attr, \
167 char *buf) \
168{ \
169 return sysfs_emit(buf, "%u\n", \
170 to_access_nodes(dev)->coord.property); \
171} \
172static DEVICE_ATTR_RO(property)
173
174ACCESS_ATTR(read_bandwidth);
175ACCESS_ATTR(read_latency);
176ACCESS_ATTR(write_bandwidth);
177ACCESS_ATTR(write_latency);
178
179static struct attribute *access_attrs[] = {
180 &dev_attr_read_bandwidth.attr,
181 &dev_attr_read_latency.attr,
182 &dev_attr_write_bandwidth.attr,
183 &dev_attr_write_latency.attr,
184 NULL,
185};
186
187/**
188 * node_set_perf_attrs - Set the performance values for given access class
189 * @nid: Node identifier to be set
190 * @coord: Heterogeneous memory performance coordinates
191 * @access: The access class the for the given attributes
192 */
193void node_set_perf_attrs(unsigned int nid, struct access_coordinate *coord,
194 unsigned int access)
195{
196 struct node_access_nodes *c;
197 struct node *node;
198 int i;
199
200 if (WARN_ON_ONCE(!node_online(nid)))
201 return;
202
203 node = node_devices[nid];
204 c = node_init_node_access(node, access);
205 if (!c)
206 return;
207
208 c->coord = *coord;
209 for (i = 0; access_attrs[i] != NULL; i++) {
210 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
211 "initiators")) {
212 pr_info("failed to add performance attribute to node %d\n",
213 nid);
214 break;
215 }
216 }
217}
218
219/**
220 * struct node_cache_info - Internal tracking for memory node caches
221 * @dev: Device represeting the cache level
222 * @node: List element for tracking in the node
223 * @cache_attrs:Attributes for this cache level
224 */
225struct node_cache_info {
226 struct device dev;
227 struct list_head node;
228 struct node_cache_attrs cache_attrs;
229};
230#define to_cache_info(device) container_of(device, struct node_cache_info, dev)
231
232#define CACHE_ATTR(name, fmt) \
233static ssize_t name##_show(struct device *dev, \
234 struct device_attribute *attr, \
235 char *buf) \
236{ \
237 return sysfs_emit(buf, fmt "\n", \
238 to_cache_info(dev)->cache_attrs.name); \
239} \
240static DEVICE_ATTR_RO(name);
241
242CACHE_ATTR(size, "%llu")
243CACHE_ATTR(line_size, "%u")
244CACHE_ATTR(indexing, "%u")
245CACHE_ATTR(write_policy, "%u")
246
247static struct attribute *cache_attrs[] = {
248 &dev_attr_indexing.attr,
249 &dev_attr_size.attr,
250 &dev_attr_line_size.attr,
251 &dev_attr_write_policy.attr,
252 NULL,
253};
254ATTRIBUTE_GROUPS(cache);
255
256static void node_cache_release(struct device *dev)
257{
258 kfree(dev);
259}
260
261static void node_cacheinfo_release(struct device *dev)
262{
263 struct node_cache_info *info = to_cache_info(dev);
264 kfree(info);
265}
266
267static void node_init_cache_dev(struct node *node)
268{
269 struct device *dev;
270
271 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
272 if (!dev)
273 return;
274
275 device_initialize(dev);
276 dev->parent = &node->dev;
277 dev->release = node_cache_release;
278 if (dev_set_name(dev, "memory_side_cache"))
279 goto put_device;
280
281 if (device_add(dev))
282 goto put_device;
283
284 pm_runtime_no_callbacks(dev);
285 node->cache_dev = dev;
286 return;
287put_device:
288 put_device(dev);
289}
290
291/**
292 * node_add_cache() - add cache attribute to a memory node
293 * @nid: Node identifier that has new cache attributes
294 * @cache_attrs: Attributes for the cache being added
295 */
296void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
297{
298 struct node_cache_info *info;
299 struct device *dev;
300 struct node *node;
301
302 if (!node_online(nid) || !node_devices[nid])
303 return;
304
305 node = node_devices[nid];
306 list_for_each_entry(info, &node->cache_attrs, node) {
307 if (info->cache_attrs.level == cache_attrs->level) {
308 dev_warn(&node->dev,
309 "attempt to add duplicate cache level:%d\n",
310 cache_attrs->level);
311 return;
312 }
313 }
314
315 if (!node->cache_dev)
316 node_init_cache_dev(node);
317 if (!node->cache_dev)
318 return;
319
320 info = kzalloc(sizeof(*info), GFP_KERNEL);
321 if (!info)
322 return;
323
324 dev = &info->dev;
325 device_initialize(dev);
326 dev->parent = node->cache_dev;
327 dev->release = node_cacheinfo_release;
328 dev->groups = cache_groups;
329 if (dev_set_name(dev, "index%d", cache_attrs->level))
330 goto put_device;
331
332 info->cache_attrs = *cache_attrs;
333 if (device_add(dev)) {
334 dev_warn(&node->dev, "failed to add cache level:%d\n",
335 cache_attrs->level);
336 goto put_device;
337 }
338 pm_runtime_no_callbacks(dev);
339 list_add_tail(&info->node, &node->cache_attrs);
340 return;
341put_device:
342 put_device(dev);
343}
344
345static void node_remove_caches(struct node *node)
346{
347 struct node_cache_info *info, *next;
348
349 if (!node->cache_dev)
350 return;
351
352 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
353 list_del(&info->node);
354 device_unregister(&info->dev);
355 }
356 device_unregister(node->cache_dev);
357}
358
359static void node_init_caches(unsigned int nid)
360{
361 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
362}
363#else
364static void node_init_caches(unsigned int nid) { }
365static void node_remove_caches(struct node *node) { }
366#endif
367
368#define K(x) ((x) << (PAGE_SHIFT - 10))
369static ssize_t node_read_meminfo(struct device *dev,
370 struct device_attribute *attr, char *buf)
371{
372 int len = 0;
373 int nid = dev->id;
374 struct pglist_data *pgdat = NODE_DATA(nid);
375 struct sysinfo i;
376 unsigned long sreclaimable, sunreclaimable;
377 unsigned long swapcached = 0;
378
379 si_meminfo_node(&i, nid);
380 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
381 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
382#ifdef CONFIG_SWAP
383 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
384#endif
385 len = sysfs_emit_at(buf, len,
386 "Node %d MemTotal: %8lu kB\n"
387 "Node %d MemFree: %8lu kB\n"
388 "Node %d MemUsed: %8lu kB\n"
389 "Node %d SwapCached: %8lu kB\n"
390 "Node %d Active: %8lu kB\n"
391 "Node %d Inactive: %8lu kB\n"
392 "Node %d Active(anon): %8lu kB\n"
393 "Node %d Inactive(anon): %8lu kB\n"
394 "Node %d Active(file): %8lu kB\n"
395 "Node %d Inactive(file): %8lu kB\n"
396 "Node %d Unevictable: %8lu kB\n"
397 "Node %d Mlocked: %8lu kB\n",
398 nid, K(i.totalram),
399 nid, K(i.freeram),
400 nid, K(i.totalram - i.freeram),
401 nid, K(swapcached),
402 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
403 node_page_state(pgdat, NR_ACTIVE_FILE)),
404 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
405 node_page_state(pgdat, NR_INACTIVE_FILE)),
406 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
407 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
408 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
409 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
410 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
411 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
412
413#ifdef CONFIG_HIGHMEM
414 len += sysfs_emit_at(buf, len,
415 "Node %d HighTotal: %8lu kB\n"
416 "Node %d HighFree: %8lu kB\n"
417 "Node %d LowTotal: %8lu kB\n"
418 "Node %d LowFree: %8lu kB\n",
419 nid, K(i.totalhigh),
420 nid, K(i.freehigh),
421 nid, K(i.totalram - i.totalhigh),
422 nid, K(i.freeram - i.freehigh));
423#endif
424 len += sysfs_emit_at(buf, len,
425 "Node %d Dirty: %8lu kB\n"
426 "Node %d Writeback: %8lu kB\n"
427 "Node %d FilePages: %8lu kB\n"
428 "Node %d Mapped: %8lu kB\n"
429 "Node %d AnonPages: %8lu kB\n"
430 "Node %d Shmem: %8lu kB\n"
431 "Node %d KernelStack: %8lu kB\n"
432#ifdef CONFIG_SHADOW_CALL_STACK
433 "Node %d ShadowCallStack:%8lu kB\n"
434#endif
435 "Node %d PageTables: %8lu kB\n"
436 "Node %d SecPageTables: %8lu kB\n"
437 "Node %d NFS_Unstable: %8lu kB\n"
438 "Node %d Bounce: %8lu kB\n"
439 "Node %d WritebackTmp: %8lu kB\n"
440 "Node %d KReclaimable: %8lu kB\n"
441 "Node %d Slab: %8lu kB\n"
442 "Node %d SReclaimable: %8lu kB\n"
443 "Node %d SUnreclaim: %8lu kB\n"
444#ifdef CONFIG_TRANSPARENT_HUGEPAGE
445 "Node %d AnonHugePages: %8lu kB\n"
446 "Node %d ShmemHugePages: %8lu kB\n"
447 "Node %d ShmemPmdMapped: %8lu kB\n"
448 "Node %d FileHugePages: %8lu kB\n"
449 "Node %d FilePmdMapped: %8lu kB\n"
450#endif
451#ifdef CONFIG_UNACCEPTED_MEMORY
452 "Node %d Unaccepted: %8lu kB\n"
453#endif
454 ,
455 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
456 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
457 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
458 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
459 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
460 nid, K(i.sharedram),
461 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
462#ifdef CONFIG_SHADOW_CALL_STACK
463 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
464#endif
465 nid, K(node_page_state(pgdat, NR_PAGETABLE)),
466 nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
467 nid, 0UL,
468 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
469 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
470 nid, K(sreclaimable +
471 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
472 nid, K(sreclaimable + sunreclaimable),
473 nid, K(sreclaimable),
474 nid, K(sunreclaimable)
475#ifdef CONFIG_TRANSPARENT_HUGEPAGE
476 ,
477 nid, K(node_page_state(pgdat, NR_ANON_THPS)),
478 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
479 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
480 nid, K(node_page_state(pgdat, NR_FILE_THPS)),
481 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
482#endif
483#ifdef CONFIG_UNACCEPTED_MEMORY
484 ,
485 nid, K(sum_zone_node_page_state(nid, NR_UNACCEPTED))
486#endif
487 );
488 len += hugetlb_report_node_meminfo(buf, len, nid);
489 return len;
490}
491
492#undef K
493static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
494
495static ssize_t node_read_numastat(struct device *dev,
496 struct device_attribute *attr, char *buf)
497{
498 fold_vm_numa_events();
499 return sysfs_emit(buf,
500 "numa_hit %lu\n"
501 "numa_miss %lu\n"
502 "numa_foreign %lu\n"
503 "interleave_hit %lu\n"
504 "local_node %lu\n"
505 "other_node %lu\n",
506 sum_zone_numa_event_state(dev->id, NUMA_HIT),
507 sum_zone_numa_event_state(dev->id, NUMA_MISS),
508 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
509 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
510 sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
511 sum_zone_numa_event_state(dev->id, NUMA_OTHER));
512}
513static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
514
515static ssize_t node_read_vmstat(struct device *dev,
516 struct device_attribute *attr, char *buf)
517{
518 int nid = dev->id;
519 struct pglist_data *pgdat = NODE_DATA(nid);
520 int i;
521 int len = 0;
522
523 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
524 len += sysfs_emit_at(buf, len, "%s %lu\n",
525 zone_stat_name(i),
526 sum_zone_node_page_state(nid, i));
527
528#ifdef CONFIG_NUMA
529 fold_vm_numa_events();
530 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
531 len += sysfs_emit_at(buf, len, "%s %lu\n",
532 numa_stat_name(i),
533 sum_zone_numa_event_state(nid, i));
534
535#endif
536 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
537 unsigned long pages = node_page_state_pages(pgdat, i);
538
539 if (vmstat_item_print_in_thp(i))
540 pages /= HPAGE_PMD_NR;
541 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
542 pages);
543 }
544
545 return len;
546}
547static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
548
549static ssize_t node_read_distance(struct device *dev,
550 struct device_attribute *attr, char *buf)
551{
552 int nid = dev->id;
553 int len = 0;
554 int i;
555
556 /*
557 * buf is currently PAGE_SIZE in length and each node needs 4 chars
558 * at the most (distance + space or newline).
559 */
560 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
561
562 for_each_online_node(i) {
563 len += sysfs_emit_at(buf, len, "%s%d",
564 i ? " " : "", node_distance(nid, i));
565 }
566
567 len += sysfs_emit_at(buf, len, "\n");
568 return len;
569}
570static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
571
572static struct attribute *node_dev_attrs[] = {
573 &dev_attr_meminfo.attr,
574 &dev_attr_numastat.attr,
575 &dev_attr_distance.attr,
576 &dev_attr_vmstat.attr,
577 NULL
578};
579
580static struct bin_attribute *node_dev_bin_attrs[] = {
581 &bin_attr_cpumap,
582 &bin_attr_cpulist,
583 NULL
584};
585
586static const struct attribute_group node_dev_group = {
587 .attrs = node_dev_attrs,
588 .bin_attrs = node_dev_bin_attrs
589};
590
591static const struct attribute_group *node_dev_groups[] = {
592 &node_dev_group,
593#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
594 &arch_node_dev_group,
595#endif
596#ifdef CONFIG_MEMORY_FAILURE
597 &memory_failure_attr_group,
598#endif
599 NULL
600};
601
602static void node_device_release(struct device *dev)
603{
604 kfree(to_node(dev));
605}
606
607/*
608 * register_node - Setup a sysfs device for a node.
609 * @num - Node number to use when creating the device.
610 *
611 * Initialize and register the node device.
612 */
613static int register_node(struct node *node, int num)
614{
615 int error;
616
617 node->dev.id = num;
618 node->dev.bus = &node_subsys;
619 node->dev.release = node_device_release;
620 node->dev.groups = node_dev_groups;
621 error = device_register(&node->dev);
622
623 if (error) {
624 put_device(&node->dev);
625 } else {
626 hugetlb_register_node(node);
627 compaction_register_node(node);
628 }
629
630 return error;
631}
632
633/**
634 * unregister_node - unregister a node device
635 * @node: node going away
636 *
637 * Unregisters a node device @node. All the devices on the node must be
638 * unregistered before calling this function.
639 */
640void unregister_node(struct node *node)
641{
642 hugetlb_unregister_node(node);
643 compaction_unregister_node(node);
644 node_remove_accesses(node);
645 node_remove_caches(node);
646 device_unregister(&node->dev);
647}
648
649struct node *node_devices[MAX_NUMNODES];
650
651/*
652 * register cpu under node
653 */
654int register_cpu_under_node(unsigned int cpu, unsigned int nid)
655{
656 int ret;
657 struct device *obj;
658
659 if (!node_online(nid))
660 return 0;
661
662 obj = get_cpu_device(cpu);
663 if (!obj)
664 return 0;
665
666 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
667 &obj->kobj,
668 kobject_name(&obj->kobj));
669 if (ret)
670 return ret;
671
672 return sysfs_create_link(&obj->kobj,
673 &node_devices[nid]->dev.kobj,
674 kobject_name(&node_devices[nid]->dev.kobj));
675}
676
677/**
678 * register_memory_node_under_compute_node - link memory node to its compute
679 * node for a given access class.
680 * @mem_nid: Memory node number
681 * @cpu_nid: Cpu node number
682 * @access: Access class to register
683 *
684 * Description:
685 * For use with platforms that may have separate memory and compute nodes.
686 * This function will export node relationships linking which memory
687 * initiator nodes can access memory targets at a given ranked access
688 * class.
689 */
690int register_memory_node_under_compute_node(unsigned int mem_nid,
691 unsigned int cpu_nid,
692 unsigned int access)
693{
694 struct node *init_node, *targ_node;
695 struct node_access_nodes *initiator, *target;
696 int ret;
697
698 if (!node_online(cpu_nid) || !node_online(mem_nid))
699 return -ENODEV;
700
701 init_node = node_devices[cpu_nid];
702 targ_node = node_devices[mem_nid];
703 initiator = node_init_node_access(init_node, access);
704 target = node_init_node_access(targ_node, access);
705 if (!initiator || !target)
706 return -ENOMEM;
707
708 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
709 &targ_node->dev.kobj,
710 dev_name(&targ_node->dev));
711 if (ret)
712 return ret;
713
714 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
715 &init_node->dev.kobj,
716 dev_name(&init_node->dev));
717 if (ret)
718 goto err;
719
720 return 0;
721 err:
722 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
723 dev_name(&targ_node->dev));
724 return ret;
725}
726
727int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
728{
729 struct device *obj;
730
731 if (!node_online(nid))
732 return 0;
733
734 obj = get_cpu_device(cpu);
735 if (!obj)
736 return 0;
737
738 sysfs_remove_link(&node_devices[nid]->dev.kobj,
739 kobject_name(&obj->kobj));
740 sysfs_remove_link(&obj->kobj,
741 kobject_name(&node_devices[nid]->dev.kobj));
742
743 return 0;
744}
745
746#ifdef CONFIG_MEMORY_HOTPLUG
747static int __ref get_nid_for_pfn(unsigned long pfn)
748{
749#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
750 if (system_state < SYSTEM_RUNNING)
751 return early_pfn_to_nid(pfn);
752#endif
753 return pfn_to_nid(pfn);
754}
755
756static void do_register_memory_block_under_node(int nid,
757 struct memory_block *mem_blk,
758 enum meminit_context context)
759{
760 int ret;
761
762 memory_block_add_nid(mem_blk, nid, context);
763
764 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
765 &mem_blk->dev.kobj,
766 kobject_name(&mem_blk->dev.kobj));
767 if (ret && ret != -EEXIST)
768 dev_err_ratelimited(&node_devices[nid]->dev,
769 "can't create link to %s in sysfs (%d)\n",
770 kobject_name(&mem_blk->dev.kobj), ret);
771
772 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
773 &node_devices[nid]->dev.kobj,
774 kobject_name(&node_devices[nid]->dev.kobj));
775 if (ret && ret != -EEXIST)
776 dev_err_ratelimited(&mem_blk->dev,
777 "can't create link to %s in sysfs (%d)\n",
778 kobject_name(&node_devices[nid]->dev.kobj),
779 ret);
780}
781
782/* register memory section under specified node if it spans that node */
783static int register_mem_block_under_node_early(struct memory_block *mem_blk,
784 void *arg)
785{
786 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
787 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
788 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
789 int nid = *(int *)arg;
790 unsigned long pfn;
791
792 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
793 int page_nid;
794
795 /*
796 * memory block could have several absent sections from start.
797 * skip pfn range from absent section
798 */
799 if (!pfn_in_present_section(pfn)) {
800 pfn = round_down(pfn + PAGES_PER_SECTION,
801 PAGES_PER_SECTION) - 1;
802 continue;
803 }
804
805 /*
806 * We need to check if page belongs to nid only at the boot
807 * case because node's ranges can be interleaved.
808 */
809 page_nid = get_nid_for_pfn(pfn);
810 if (page_nid < 0)
811 continue;
812 if (page_nid != nid)
813 continue;
814
815 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
816 return 0;
817 }
818 /* mem section does not span the specified node */
819 return 0;
820}
821
822/*
823 * During hotplug we know that all pages in the memory block belong to the same
824 * node.
825 */
826static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
827 void *arg)
828{
829 int nid = *(int *)arg;
830
831 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
832 return 0;
833}
834
835/*
836 * Unregister a memory block device under the node it spans. Memory blocks
837 * with multiple nodes cannot be offlined and therefore also never be removed.
838 */
839void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
840{
841 if (mem_blk->nid == NUMA_NO_NODE)
842 return;
843
844 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
845 kobject_name(&mem_blk->dev.kobj));
846 sysfs_remove_link(&mem_blk->dev.kobj,
847 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
848}
849
850void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
851 unsigned long end_pfn,
852 enum meminit_context context)
853{
854 walk_memory_blocks_func_t func;
855
856 if (context == MEMINIT_HOTPLUG)
857 func = register_mem_block_under_node_hotplug;
858 else
859 func = register_mem_block_under_node_early;
860
861 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
862 (void *)&nid, func);
863 return;
864}
865#endif /* CONFIG_MEMORY_HOTPLUG */
866
867int __register_one_node(int nid)
868{
869 int error;
870 int cpu;
871 struct node *node;
872
873 node = kzalloc(sizeof(struct node), GFP_KERNEL);
874 if (!node)
875 return -ENOMEM;
876
877 INIT_LIST_HEAD(&node->access_list);
878 node_devices[nid] = node;
879
880 error = register_node(node_devices[nid], nid);
881
882 /* link cpu under this node */
883 for_each_present_cpu(cpu) {
884 if (cpu_to_node(cpu) == nid)
885 register_cpu_under_node(cpu, nid);
886 }
887
888 node_init_caches(nid);
889
890 return error;
891}
892
893void unregister_one_node(int nid)
894{
895 if (!node_devices[nid])
896 return;
897
898 unregister_node(node_devices[nid]);
899 node_devices[nid] = NULL;
900}
901
902/*
903 * node states attributes
904 */
905
906struct node_attr {
907 struct device_attribute attr;
908 enum node_states state;
909};
910
911static ssize_t show_node_state(struct device *dev,
912 struct device_attribute *attr, char *buf)
913{
914 struct node_attr *na = container_of(attr, struct node_attr, attr);
915
916 return sysfs_emit(buf, "%*pbl\n",
917 nodemask_pr_args(&node_states[na->state]));
918}
919
920#define _NODE_ATTR(name, state) \
921 { __ATTR(name, 0444, show_node_state, NULL), state }
922
923static struct node_attr node_state_attr[] = {
924 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
925 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
926 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
927#ifdef CONFIG_HIGHMEM
928 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
929#endif
930 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
931 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
932 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
933 N_GENERIC_INITIATOR),
934};
935
936static struct attribute *node_state_attrs[] = {
937 &node_state_attr[N_POSSIBLE].attr.attr,
938 &node_state_attr[N_ONLINE].attr.attr,
939 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
940#ifdef CONFIG_HIGHMEM
941 &node_state_attr[N_HIGH_MEMORY].attr.attr,
942#endif
943 &node_state_attr[N_MEMORY].attr.attr,
944 &node_state_attr[N_CPU].attr.attr,
945 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
946 NULL
947};
948
949static const struct attribute_group memory_root_attr_group = {
950 .attrs = node_state_attrs,
951};
952
953static const struct attribute_group *cpu_root_attr_groups[] = {
954 &memory_root_attr_group,
955 NULL,
956};
957
958void __init node_dev_init(void)
959{
960 int ret, i;
961
962 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
963 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
964
965 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
966 if (ret)
967 panic("%s() failed to register subsystem: %d\n", __func__, ret);
968
969 /*
970 * Create all node devices, which will properly link the node
971 * to applicable memory block devices and already created cpu devices.
972 */
973 for_each_online_node(i) {
974 ret = register_one_node(i);
975 if (ret)
976 panic("%s() failed to add node: %d\n", __func__, ret);
977 }
978}
1/*
2 * Basic Node interface support
3 */
4
5#include <linux/module.h>
6#include <linux/init.h>
7#include <linux/mm.h>
8#include <linux/memory.h>
9#include <linux/vmstat.h>
10#include <linux/notifier.h>
11#include <linux/node.h>
12#include <linux/hugetlb.h>
13#include <linux/compaction.h>
14#include <linux/cpumask.h>
15#include <linux/topology.h>
16#include <linux/nodemask.h>
17#include <linux/cpu.h>
18#include <linux/device.h>
19#include <linux/swap.h>
20#include <linux/slab.h>
21
22static struct bus_type node_subsys = {
23 .name = "node",
24 .dev_name = "node",
25};
26
27
28static ssize_t node_read_cpumap(struct device *dev, int type, char *buf)
29{
30 struct node *node_dev = to_node(dev);
31 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
32 int len;
33
34 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
35 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
36
37 len = type?
38 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
39 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
40 buf[len++] = '\n';
41 buf[len] = '\0';
42 return len;
43}
44
45static inline ssize_t node_read_cpumask(struct device *dev,
46 struct device_attribute *attr, char *buf)
47{
48 return node_read_cpumap(dev, 0, buf);
49}
50static inline ssize_t node_read_cpulist(struct device *dev,
51 struct device_attribute *attr, char *buf)
52{
53 return node_read_cpumap(dev, 1, buf);
54}
55
56static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
57static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
58
59#define K(x) ((x) << (PAGE_SHIFT - 10))
60static ssize_t node_read_meminfo(struct device *dev,
61 struct device_attribute *attr, char *buf)
62{
63 int n;
64 int nid = dev->id;
65 struct sysinfo i;
66
67 si_meminfo_node(&i, nid);
68 n = sprintf(buf,
69 "Node %d MemTotal: %8lu kB\n"
70 "Node %d MemFree: %8lu kB\n"
71 "Node %d MemUsed: %8lu kB\n"
72 "Node %d Active: %8lu kB\n"
73 "Node %d Inactive: %8lu kB\n"
74 "Node %d Active(anon): %8lu kB\n"
75 "Node %d Inactive(anon): %8lu kB\n"
76 "Node %d Active(file): %8lu kB\n"
77 "Node %d Inactive(file): %8lu kB\n"
78 "Node %d Unevictable: %8lu kB\n"
79 "Node %d Mlocked: %8lu kB\n",
80 nid, K(i.totalram),
81 nid, K(i.freeram),
82 nid, K(i.totalram - i.freeram),
83 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
84 node_page_state(nid, NR_ACTIVE_FILE)),
85 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
86 node_page_state(nid, NR_INACTIVE_FILE)),
87 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
88 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
89 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
90 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
91 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
92 nid, K(node_page_state(nid, NR_MLOCK)));
93
94#ifdef CONFIG_HIGHMEM
95 n += sprintf(buf + n,
96 "Node %d HighTotal: %8lu kB\n"
97 "Node %d HighFree: %8lu kB\n"
98 "Node %d LowTotal: %8lu kB\n"
99 "Node %d LowFree: %8lu kB\n",
100 nid, K(i.totalhigh),
101 nid, K(i.freehigh),
102 nid, K(i.totalram - i.totalhigh),
103 nid, K(i.freeram - i.freehigh));
104#endif
105 n += sprintf(buf + n,
106 "Node %d Dirty: %8lu kB\n"
107 "Node %d Writeback: %8lu kB\n"
108 "Node %d FilePages: %8lu kB\n"
109 "Node %d Mapped: %8lu kB\n"
110 "Node %d AnonPages: %8lu kB\n"
111 "Node %d Shmem: %8lu kB\n"
112 "Node %d KernelStack: %8lu kB\n"
113 "Node %d PageTables: %8lu kB\n"
114 "Node %d NFS_Unstable: %8lu kB\n"
115 "Node %d Bounce: %8lu kB\n"
116 "Node %d WritebackTmp: %8lu kB\n"
117 "Node %d Slab: %8lu kB\n"
118 "Node %d SReclaimable: %8lu kB\n"
119 "Node %d SUnreclaim: %8lu kB\n"
120#ifdef CONFIG_TRANSPARENT_HUGEPAGE
121 "Node %d AnonHugePages: %8lu kB\n"
122#endif
123 ,
124 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
125 nid, K(node_page_state(nid, NR_WRITEBACK)),
126 nid, K(node_page_state(nid, NR_FILE_PAGES)),
127 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
128 nid, K(node_page_state(nid, NR_ANON_PAGES)),
129 nid, K(node_page_state(nid, NR_SHMEM)),
130 nid, node_page_state(nid, NR_KERNEL_STACK) *
131 THREAD_SIZE / 1024,
132 nid, K(node_page_state(nid, NR_PAGETABLE)),
133 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
134 nid, K(node_page_state(nid, NR_BOUNCE)),
135 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
136 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
137 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
138 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
139#ifdef CONFIG_TRANSPARENT_HUGEPAGE
140 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
141 , nid,
142 K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
143 HPAGE_PMD_NR));
144#else
145 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
146#endif
147 n += hugetlb_report_node_meminfo(nid, buf + n);
148 return n;
149}
150
151#undef K
152static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
153
154static ssize_t node_read_numastat(struct device *dev,
155 struct device_attribute *attr, char *buf)
156{
157 return sprintf(buf,
158 "numa_hit %lu\n"
159 "numa_miss %lu\n"
160 "numa_foreign %lu\n"
161 "interleave_hit %lu\n"
162 "local_node %lu\n"
163 "other_node %lu\n",
164 node_page_state(dev->id, NUMA_HIT),
165 node_page_state(dev->id, NUMA_MISS),
166 node_page_state(dev->id, NUMA_FOREIGN),
167 node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
168 node_page_state(dev->id, NUMA_LOCAL),
169 node_page_state(dev->id, NUMA_OTHER));
170}
171static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
172
173static ssize_t node_read_vmstat(struct device *dev,
174 struct device_attribute *attr, char *buf)
175{
176 int nid = dev->id;
177 int i;
178 int n = 0;
179
180 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
181 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
182 node_page_state(nid, i));
183
184 return n;
185}
186static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
187
188static ssize_t node_read_distance(struct device *dev,
189 struct device_attribute *attr, char * buf)
190{
191 int nid = dev->id;
192 int len = 0;
193 int i;
194
195 /*
196 * buf is currently PAGE_SIZE in length and each node needs 4 chars
197 * at the most (distance + space or newline).
198 */
199 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
200
201 for_each_online_node(i)
202 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
203
204 len += sprintf(buf + len, "\n");
205 return len;
206}
207static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
208
209#ifdef CONFIG_HUGETLBFS
210/*
211 * hugetlbfs per node attributes registration interface:
212 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
213 * it will register its per node attributes for all online nodes with
214 * memory. It will also call register_hugetlbfs_with_node(), below, to
215 * register its attribute registration functions with this node driver.
216 * Once these hooks have been initialized, the node driver will call into
217 * the hugetlb module to [un]register attributes for hot-plugged nodes.
218 */
219static node_registration_func_t __hugetlb_register_node;
220static node_registration_func_t __hugetlb_unregister_node;
221
222static inline bool hugetlb_register_node(struct node *node)
223{
224 if (__hugetlb_register_node &&
225 node_state(node->dev.id, N_MEMORY)) {
226 __hugetlb_register_node(node);
227 return true;
228 }
229 return false;
230}
231
232static inline void hugetlb_unregister_node(struct node *node)
233{
234 if (__hugetlb_unregister_node)
235 __hugetlb_unregister_node(node);
236}
237
238void register_hugetlbfs_with_node(node_registration_func_t doregister,
239 node_registration_func_t unregister)
240{
241 __hugetlb_register_node = doregister;
242 __hugetlb_unregister_node = unregister;
243}
244#else
245static inline void hugetlb_register_node(struct node *node) {}
246
247static inline void hugetlb_unregister_node(struct node *node) {}
248#endif
249
250static void node_device_release(struct device *dev)
251{
252 struct node *node = to_node(dev);
253
254#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
255 /*
256 * We schedule the work only when a memory section is
257 * onlined/offlined on this node. When we come here,
258 * all the memory on this node has been offlined,
259 * so we won't enqueue new work to this work.
260 *
261 * The work is using node->node_work, so we should
262 * flush work before freeing the memory.
263 */
264 flush_work(&node->node_work);
265#endif
266 kfree(node);
267}
268
269/*
270 * register_node - Setup a sysfs device for a node.
271 * @num - Node number to use when creating the device.
272 *
273 * Initialize and register the node device.
274 */
275static int register_node(struct node *node, int num, struct node *parent)
276{
277 int error;
278
279 node->dev.id = num;
280 node->dev.bus = &node_subsys;
281 node->dev.release = node_device_release;
282 error = device_register(&node->dev);
283
284 if (!error){
285 device_create_file(&node->dev, &dev_attr_cpumap);
286 device_create_file(&node->dev, &dev_attr_cpulist);
287 device_create_file(&node->dev, &dev_attr_meminfo);
288 device_create_file(&node->dev, &dev_attr_numastat);
289 device_create_file(&node->dev, &dev_attr_distance);
290 device_create_file(&node->dev, &dev_attr_vmstat);
291
292 scan_unevictable_register_node(node);
293
294 hugetlb_register_node(node);
295
296 compaction_register_node(node);
297 }
298 return error;
299}
300
301/**
302 * unregister_node - unregister a node device
303 * @node: node going away
304 *
305 * Unregisters a node device @node. All the devices on the node must be
306 * unregistered before calling this function.
307 */
308void unregister_node(struct node *node)
309{
310 device_remove_file(&node->dev, &dev_attr_cpumap);
311 device_remove_file(&node->dev, &dev_attr_cpulist);
312 device_remove_file(&node->dev, &dev_attr_meminfo);
313 device_remove_file(&node->dev, &dev_attr_numastat);
314 device_remove_file(&node->dev, &dev_attr_distance);
315 device_remove_file(&node->dev, &dev_attr_vmstat);
316
317 scan_unevictable_unregister_node(node);
318 hugetlb_unregister_node(node); /* no-op, if memoryless node */
319
320 device_unregister(&node->dev);
321}
322
323struct node *node_devices[MAX_NUMNODES];
324
325/*
326 * register cpu under node
327 */
328int register_cpu_under_node(unsigned int cpu, unsigned int nid)
329{
330 int ret;
331 struct device *obj;
332
333 if (!node_online(nid))
334 return 0;
335
336 obj = get_cpu_device(cpu);
337 if (!obj)
338 return 0;
339
340 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
341 &obj->kobj,
342 kobject_name(&obj->kobj));
343 if (ret)
344 return ret;
345
346 return sysfs_create_link(&obj->kobj,
347 &node_devices[nid]->dev.kobj,
348 kobject_name(&node_devices[nid]->dev.kobj));
349}
350
351int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
352{
353 struct device *obj;
354
355 if (!node_online(nid))
356 return 0;
357
358 obj = get_cpu_device(cpu);
359 if (!obj)
360 return 0;
361
362 sysfs_remove_link(&node_devices[nid]->dev.kobj,
363 kobject_name(&obj->kobj));
364 sysfs_remove_link(&obj->kobj,
365 kobject_name(&node_devices[nid]->dev.kobj));
366
367 return 0;
368}
369
370#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
371#define page_initialized(page) (page->lru.next)
372
373static int get_nid_for_pfn(unsigned long pfn)
374{
375 struct page *page;
376
377 if (!pfn_valid_within(pfn))
378 return -1;
379 page = pfn_to_page(pfn);
380 if (!page_initialized(page))
381 return -1;
382 return pfn_to_nid(pfn);
383}
384
385/* register memory section under specified node if it spans that node */
386int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
387{
388 int ret;
389 unsigned long pfn, sect_start_pfn, sect_end_pfn;
390
391 if (!mem_blk)
392 return -EFAULT;
393 if (!node_online(nid))
394 return 0;
395
396 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
397 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
398 sect_end_pfn += PAGES_PER_SECTION - 1;
399 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
400 int page_nid;
401
402 page_nid = get_nid_for_pfn(pfn);
403 if (page_nid < 0)
404 continue;
405 if (page_nid != nid)
406 continue;
407 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
408 &mem_blk->dev.kobj,
409 kobject_name(&mem_blk->dev.kobj));
410 if (ret)
411 return ret;
412
413 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
414 &node_devices[nid]->dev.kobj,
415 kobject_name(&node_devices[nid]->dev.kobj));
416 }
417 /* mem section does not span the specified node */
418 return 0;
419}
420
421/* unregister memory section under all nodes that it spans */
422int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
423 unsigned long phys_index)
424{
425 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
426 unsigned long pfn, sect_start_pfn, sect_end_pfn;
427
428 if (!mem_blk) {
429 NODEMASK_FREE(unlinked_nodes);
430 return -EFAULT;
431 }
432 if (!unlinked_nodes)
433 return -ENOMEM;
434 nodes_clear(*unlinked_nodes);
435
436 sect_start_pfn = section_nr_to_pfn(phys_index);
437 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
438 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
439 int nid;
440
441 nid = get_nid_for_pfn(pfn);
442 if (nid < 0)
443 continue;
444 if (!node_online(nid))
445 continue;
446 if (node_test_and_set(nid, *unlinked_nodes))
447 continue;
448 sysfs_remove_link(&node_devices[nid]->dev.kobj,
449 kobject_name(&mem_blk->dev.kobj));
450 sysfs_remove_link(&mem_blk->dev.kobj,
451 kobject_name(&node_devices[nid]->dev.kobj));
452 }
453 NODEMASK_FREE(unlinked_nodes);
454 return 0;
455}
456
457static int link_mem_sections(int nid)
458{
459 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
460 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
461 unsigned long pfn;
462 struct memory_block *mem_blk = NULL;
463 int err = 0;
464
465 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
466 unsigned long section_nr = pfn_to_section_nr(pfn);
467 struct mem_section *mem_sect;
468 int ret;
469
470 if (!present_section_nr(section_nr))
471 continue;
472 mem_sect = __nr_to_section(section_nr);
473
474 /* same memblock ? */
475 if (mem_blk)
476 if ((section_nr >= mem_blk->start_section_nr) &&
477 (section_nr <= mem_blk->end_section_nr))
478 continue;
479
480 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
481
482 ret = register_mem_sect_under_node(mem_blk, nid);
483 if (!err)
484 err = ret;
485
486 /* discard ref obtained in find_memory_block() */
487 }
488
489 if (mem_blk)
490 kobject_put(&mem_blk->dev.kobj);
491 return err;
492}
493
494#ifdef CONFIG_HUGETLBFS
495/*
496 * Handle per node hstate attribute [un]registration on transistions
497 * to/from memoryless state.
498 */
499static void node_hugetlb_work(struct work_struct *work)
500{
501 struct node *node = container_of(work, struct node, node_work);
502
503 /*
504 * We only get here when a node transitions to/from memoryless state.
505 * We can detect which transition occurred by examining whether the
506 * node has memory now. hugetlb_register_node() already check this
507 * so we try to register the attributes. If that fails, then the
508 * node has transitioned to memoryless, try to unregister the
509 * attributes.
510 */
511 if (!hugetlb_register_node(node))
512 hugetlb_unregister_node(node);
513}
514
515static void init_node_hugetlb_work(int nid)
516{
517 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
518}
519
520static int node_memory_callback(struct notifier_block *self,
521 unsigned long action, void *arg)
522{
523 struct memory_notify *mnb = arg;
524 int nid = mnb->status_change_nid;
525
526 switch (action) {
527 case MEM_ONLINE:
528 case MEM_OFFLINE:
529 /*
530 * offload per node hstate [un]registration to a work thread
531 * when transitioning to/from memoryless state.
532 */
533 if (nid != NUMA_NO_NODE)
534 schedule_work(&node_devices[nid]->node_work);
535 break;
536
537 case MEM_GOING_ONLINE:
538 case MEM_GOING_OFFLINE:
539 case MEM_CANCEL_ONLINE:
540 case MEM_CANCEL_OFFLINE:
541 default:
542 break;
543 }
544
545 return NOTIFY_OK;
546}
547#endif /* CONFIG_HUGETLBFS */
548#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
549
550static int link_mem_sections(int nid) { return 0; }
551#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
552
553#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
554 !defined(CONFIG_HUGETLBFS)
555static inline int node_memory_callback(struct notifier_block *self,
556 unsigned long action, void *arg)
557{
558 return NOTIFY_OK;
559}
560
561static void init_node_hugetlb_work(int nid) { }
562
563#endif
564
565int register_one_node(int nid)
566{
567 int error = 0;
568 int cpu;
569
570 if (node_online(nid)) {
571 int p_node = parent_node(nid);
572 struct node *parent = NULL;
573
574 if (p_node != nid)
575 parent = node_devices[p_node];
576
577 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
578 if (!node_devices[nid])
579 return -ENOMEM;
580
581 error = register_node(node_devices[nid], nid, parent);
582
583 /* link cpu under this node */
584 for_each_present_cpu(cpu) {
585 if (cpu_to_node(cpu) == nid)
586 register_cpu_under_node(cpu, nid);
587 }
588
589 /* link memory sections under this node */
590 error = link_mem_sections(nid);
591
592 /* initialize work queue for memory hot plug */
593 init_node_hugetlb_work(nid);
594 }
595
596 return error;
597
598}
599
600void unregister_one_node(int nid)
601{
602 if (!node_devices[nid])
603 return;
604
605 unregister_node(node_devices[nid]);
606 kfree(node_devices[nid]);
607 node_devices[nid] = NULL;
608}
609
610/*
611 * node states attributes
612 */
613
614static ssize_t print_nodes_state(enum node_states state, char *buf)
615{
616 int n;
617
618 n = nodelist_scnprintf(buf, PAGE_SIZE-2, node_states[state]);
619 buf[n++] = '\n';
620 buf[n] = '\0';
621 return n;
622}
623
624struct node_attr {
625 struct device_attribute attr;
626 enum node_states state;
627};
628
629static ssize_t show_node_state(struct device *dev,
630 struct device_attribute *attr, char *buf)
631{
632 struct node_attr *na = container_of(attr, struct node_attr, attr);
633 return print_nodes_state(na->state, buf);
634}
635
636#define _NODE_ATTR(name, state) \
637 { __ATTR(name, 0444, show_node_state, NULL), state }
638
639static struct node_attr node_state_attr[] = {
640 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
641 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
642 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
643#ifdef CONFIG_HIGHMEM
644 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
645#endif
646#ifdef CONFIG_MOVABLE_NODE
647 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
648#endif
649 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
650};
651
652static struct attribute *node_state_attrs[] = {
653 &node_state_attr[N_POSSIBLE].attr.attr,
654 &node_state_attr[N_ONLINE].attr.attr,
655 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
656#ifdef CONFIG_HIGHMEM
657 &node_state_attr[N_HIGH_MEMORY].attr.attr,
658#endif
659#ifdef CONFIG_MOVABLE_NODE
660 &node_state_attr[N_MEMORY].attr.attr,
661#endif
662 &node_state_attr[N_CPU].attr.attr,
663 NULL
664};
665
666static struct attribute_group memory_root_attr_group = {
667 .attrs = node_state_attrs,
668};
669
670static const struct attribute_group *cpu_root_attr_groups[] = {
671 &memory_root_attr_group,
672 NULL,
673};
674
675#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
676static int __init register_node_type(void)
677{
678 int ret;
679
680 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
681 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
682
683 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
684 if (!ret) {
685 static struct notifier_block node_memory_callback_nb = {
686 .notifier_call = node_memory_callback,
687 .priority = NODE_CALLBACK_PRI,
688 };
689 register_hotmemory_notifier(&node_memory_callback_nb);
690 }
691
692 /*
693 * Note: we're not going to unregister the node class if we fail
694 * to register the node state class attribute files.
695 */
696 return ret;
697}
698postcore_initcall(register_node_type);