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// 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 struct bus_type node_subsys = {
25 .name = "node",
26 .dev_name = "node",
27};
28
29
30static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
31{
32 ssize_t n;
33 cpumask_var_t mask;
34 struct node *node_dev = to_node(dev);
35
36 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
37 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
38
39 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
40 return 0;
41
42 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
43 n = cpumap_print_to_pagebuf(list, buf, mask);
44 free_cpumask_var(mask);
45
46 return n;
47}
48
49static inline ssize_t cpumap_show(struct device *dev,
50 struct device_attribute *attr,
51 char *buf)
52{
53 return node_read_cpumap(dev, false, buf);
54}
55
56static DEVICE_ATTR_RO(cpumap);
57
58static inline ssize_t cpulist_show(struct device *dev,
59 struct device_attribute *attr,
60 char *buf)
61{
62 return node_read_cpumap(dev, true, buf);
63}
64
65static DEVICE_ATTR_RO(cpulist);
66
67/**
68 * struct node_access_nodes - Access class device to hold user visible
69 * relationships to other nodes.
70 * @dev: Device for this memory access class
71 * @list_node: List element in the node's access list
72 * @access: The access class rank
73 * @hmem_attrs: Heterogeneous memory performance attributes
74 */
75struct node_access_nodes {
76 struct device dev;
77 struct list_head list_node;
78 unsigned access;
79#ifdef CONFIG_HMEM_REPORTING
80 struct node_hmem_attrs hmem_attrs;
81#endif
82};
83#define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
84
85static struct attribute *node_init_access_node_attrs[] = {
86 NULL,
87};
88
89static struct attribute *node_targ_access_node_attrs[] = {
90 NULL,
91};
92
93static const struct attribute_group initiators = {
94 .name = "initiators",
95 .attrs = node_init_access_node_attrs,
96};
97
98static const struct attribute_group targets = {
99 .name = "targets",
100 .attrs = node_targ_access_node_attrs,
101};
102
103static const struct attribute_group *node_access_node_groups[] = {
104 &initiators,
105 &targets,
106 NULL,
107};
108
109static void node_remove_accesses(struct node *node)
110{
111 struct node_access_nodes *c, *cnext;
112
113 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
114 list_del(&c->list_node);
115 device_unregister(&c->dev);
116 }
117}
118
119static void node_access_release(struct device *dev)
120{
121 kfree(to_access_nodes(dev));
122}
123
124static struct node_access_nodes *node_init_node_access(struct node *node,
125 unsigned access)
126{
127 struct node_access_nodes *access_node;
128 struct device *dev;
129
130 list_for_each_entry(access_node, &node->access_list, list_node)
131 if (access_node->access == access)
132 return access_node;
133
134 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
135 if (!access_node)
136 return NULL;
137
138 access_node->access = access;
139 dev = &access_node->dev;
140 dev->parent = &node->dev;
141 dev->release = node_access_release;
142 dev->groups = node_access_node_groups;
143 if (dev_set_name(dev, "access%u", access))
144 goto free;
145
146 if (device_register(dev))
147 goto free_name;
148
149 pm_runtime_no_callbacks(dev);
150 list_add_tail(&access_node->list_node, &node->access_list);
151 return access_node;
152free_name:
153 kfree_const(dev->kobj.name);
154free:
155 kfree(access_node);
156 return NULL;
157}
158
159#ifdef CONFIG_HMEM_REPORTING
160#define ACCESS_ATTR(name) \
161static ssize_t name##_show(struct device *dev, \
162 struct device_attribute *attr, \
163 char *buf) \
164{ \
165 return sysfs_emit(buf, "%u\n", \
166 to_access_nodes(dev)->hmem_attrs.name); \
167} \
168static DEVICE_ATTR_RO(name)
169
170ACCESS_ATTR(read_bandwidth);
171ACCESS_ATTR(read_latency);
172ACCESS_ATTR(write_bandwidth);
173ACCESS_ATTR(write_latency);
174
175static struct attribute *access_attrs[] = {
176 &dev_attr_read_bandwidth.attr,
177 &dev_attr_read_latency.attr,
178 &dev_attr_write_bandwidth.attr,
179 &dev_attr_write_latency.attr,
180 NULL,
181};
182
183/**
184 * node_set_perf_attrs - Set the performance values for given access class
185 * @nid: Node identifier to be set
186 * @hmem_attrs: Heterogeneous memory performance attributes
187 * @access: The access class the for the given attributes
188 */
189void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
190 unsigned access)
191{
192 struct node_access_nodes *c;
193 struct node *node;
194 int i;
195
196 if (WARN_ON_ONCE(!node_online(nid)))
197 return;
198
199 node = node_devices[nid];
200 c = node_init_node_access(node, access);
201 if (!c)
202 return;
203
204 c->hmem_attrs = *hmem_attrs;
205 for (i = 0; access_attrs[i] != NULL; i++) {
206 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
207 "initiators")) {
208 pr_info("failed to add performance attribute to node %d\n",
209 nid);
210 break;
211 }
212 }
213}
214
215/**
216 * struct node_cache_info - Internal tracking for memory node caches
217 * @dev: Device represeting the cache level
218 * @node: List element for tracking in the node
219 * @cache_attrs:Attributes for this cache level
220 */
221struct node_cache_info {
222 struct device dev;
223 struct list_head node;
224 struct node_cache_attrs cache_attrs;
225};
226#define to_cache_info(device) container_of(device, struct node_cache_info, dev)
227
228#define CACHE_ATTR(name, fmt) \
229static ssize_t name##_show(struct device *dev, \
230 struct device_attribute *attr, \
231 char *buf) \
232{ \
233 return sysfs_emit(buf, fmt "\n", \
234 to_cache_info(dev)->cache_attrs.name); \
235} \
236static DEVICE_ATTR_RO(name);
237
238CACHE_ATTR(size, "%llu")
239CACHE_ATTR(line_size, "%u")
240CACHE_ATTR(indexing, "%u")
241CACHE_ATTR(write_policy, "%u")
242
243static struct attribute *cache_attrs[] = {
244 &dev_attr_indexing.attr,
245 &dev_attr_size.attr,
246 &dev_attr_line_size.attr,
247 &dev_attr_write_policy.attr,
248 NULL,
249};
250ATTRIBUTE_GROUPS(cache);
251
252static void node_cache_release(struct device *dev)
253{
254 kfree(dev);
255}
256
257static void node_cacheinfo_release(struct device *dev)
258{
259 struct node_cache_info *info = to_cache_info(dev);
260 kfree(info);
261}
262
263static void node_init_cache_dev(struct node *node)
264{
265 struct device *dev;
266
267 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
268 if (!dev)
269 return;
270
271 device_initialize(dev);
272 dev->parent = &node->dev;
273 dev->release = node_cache_release;
274 if (dev_set_name(dev, "memory_side_cache"))
275 goto put_device;
276
277 if (device_add(dev))
278 goto put_device;
279
280 pm_runtime_no_callbacks(dev);
281 node->cache_dev = dev;
282 return;
283put_device:
284 put_device(dev);
285}
286
287/**
288 * node_add_cache() - add cache attribute to a memory node
289 * @nid: Node identifier that has new cache attributes
290 * @cache_attrs: Attributes for the cache being added
291 */
292void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
293{
294 struct node_cache_info *info;
295 struct device *dev;
296 struct node *node;
297
298 if (!node_online(nid) || !node_devices[nid])
299 return;
300
301 node = node_devices[nid];
302 list_for_each_entry(info, &node->cache_attrs, node) {
303 if (info->cache_attrs.level == cache_attrs->level) {
304 dev_warn(&node->dev,
305 "attempt to add duplicate cache level:%d\n",
306 cache_attrs->level);
307 return;
308 }
309 }
310
311 if (!node->cache_dev)
312 node_init_cache_dev(node);
313 if (!node->cache_dev)
314 return;
315
316 info = kzalloc(sizeof(*info), GFP_KERNEL);
317 if (!info)
318 return;
319
320 dev = &info->dev;
321 device_initialize(dev);
322 dev->parent = node->cache_dev;
323 dev->release = node_cacheinfo_release;
324 dev->groups = cache_groups;
325 if (dev_set_name(dev, "index%d", cache_attrs->level))
326 goto put_device;
327
328 info->cache_attrs = *cache_attrs;
329 if (device_add(dev)) {
330 dev_warn(&node->dev, "failed to add cache level:%d\n",
331 cache_attrs->level);
332 goto put_device;
333 }
334 pm_runtime_no_callbacks(dev);
335 list_add_tail(&info->node, &node->cache_attrs);
336 return;
337put_device:
338 put_device(dev);
339}
340
341static void node_remove_caches(struct node *node)
342{
343 struct node_cache_info *info, *next;
344
345 if (!node->cache_dev)
346 return;
347
348 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
349 list_del(&info->node);
350 device_unregister(&info->dev);
351 }
352 device_unregister(node->cache_dev);
353}
354
355static void node_init_caches(unsigned int nid)
356{
357 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
358}
359#else
360static void node_init_caches(unsigned int nid) { }
361static void node_remove_caches(struct node *node) { }
362#endif
363
364#define K(x) ((x) << (PAGE_SHIFT - 10))
365static ssize_t node_read_meminfo(struct device *dev,
366 struct device_attribute *attr, char *buf)
367{
368 int len = 0;
369 int nid = dev->id;
370 struct pglist_data *pgdat = NODE_DATA(nid);
371 struct sysinfo i;
372 unsigned long sreclaimable, sunreclaimable;
373 unsigned long swapcached = 0;
374
375 si_meminfo_node(&i, nid);
376 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
377 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
378#ifdef CONFIG_SWAP
379 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
380#endif
381 len = sysfs_emit_at(buf, len,
382 "Node %d MemTotal: %8lu kB\n"
383 "Node %d MemFree: %8lu kB\n"
384 "Node %d MemUsed: %8lu kB\n"
385 "Node %d SwapCached: %8lu kB\n"
386 "Node %d Active: %8lu kB\n"
387 "Node %d Inactive: %8lu kB\n"
388 "Node %d Active(anon): %8lu kB\n"
389 "Node %d Inactive(anon): %8lu kB\n"
390 "Node %d Active(file): %8lu kB\n"
391 "Node %d Inactive(file): %8lu kB\n"
392 "Node %d Unevictable: %8lu kB\n"
393 "Node %d Mlocked: %8lu kB\n",
394 nid, K(i.totalram),
395 nid, K(i.freeram),
396 nid, K(i.totalram - i.freeram),
397 nid, K(swapcached),
398 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
399 node_page_state(pgdat, NR_ACTIVE_FILE)),
400 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
401 node_page_state(pgdat, NR_INACTIVE_FILE)),
402 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
403 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
404 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
405 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
406 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
407 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
408
409#ifdef CONFIG_HIGHMEM
410 len += sysfs_emit_at(buf, len,
411 "Node %d HighTotal: %8lu kB\n"
412 "Node %d HighFree: %8lu kB\n"
413 "Node %d LowTotal: %8lu kB\n"
414 "Node %d LowFree: %8lu kB\n",
415 nid, K(i.totalhigh),
416 nid, K(i.freehigh),
417 nid, K(i.totalram - i.totalhigh),
418 nid, K(i.freeram - i.freehigh));
419#endif
420 len += sysfs_emit_at(buf, len,
421 "Node %d Dirty: %8lu kB\n"
422 "Node %d Writeback: %8lu kB\n"
423 "Node %d FilePages: %8lu kB\n"
424 "Node %d Mapped: %8lu kB\n"
425 "Node %d AnonPages: %8lu kB\n"
426 "Node %d Shmem: %8lu kB\n"
427 "Node %d KernelStack: %8lu kB\n"
428#ifdef CONFIG_SHADOW_CALL_STACK
429 "Node %d ShadowCallStack:%8lu kB\n"
430#endif
431 "Node %d PageTables: %8lu kB\n"
432 "Node %d NFS_Unstable: %8lu kB\n"
433 "Node %d Bounce: %8lu kB\n"
434 "Node %d WritebackTmp: %8lu kB\n"
435 "Node %d KReclaimable: %8lu kB\n"
436 "Node %d Slab: %8lu kB\n"
437 "Node %d SReclaimable: %8lu kB\n"
438 "Node %d SUnreclaim: %8lu kB\n"
439#ifdef CONFIG_TRANSPARENT_HUGEPAGE
440 "Node %d AnonHugePages: %8lu kB\n"
441 "Node %d ShmemHugePages: %8lu kB\n"
442 "Node %d ShmemPmdMapped: %8lu kB\n"
443 "Node %d FileHugePages: %8lu kB\n"
444 "Node %d FilePmdMapped: %8lu kB\n"
445#endif
446 ,
447 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
448 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
449 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
450 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
451 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
452 nid, K(i.sharedram),
453 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
454#ifdef CONFIG_SHADOW_CALL_STACK
455 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
456#endif
457 nid, K(node_page_state(pgdat, NR_PAGETABLE)),
458 nid, 0UL,
459 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
460 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
461 nid, K(sreclaimable +
462 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
463 nid, K(sreclaimable + sunreclaimable),
464 nid, K(sreclaimable),
465 nid, K(sunreclaimable)
466#ifdef CONFIG_TRANSPARENT_HUGEPAGE
467 ,
468 nid, K(node_page_state(pgdat, NR_ANON_THPS)),
469 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
470 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
471 nid, K(node_page_state(pgdat, NR_FILE_THPS)),
472 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
473#endif
474 );
475 len += hugetlb_report_node_meminfo(buf, len, nid);
476 return len;
477}
478
479#undef K
480static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
481
482static ssize_t node_read_numastat(struct device *dev,
483 struct device_attribute *attr, char *buf)
484{
485 fold_vm_numa_events();
486 return sysfs_emit(buf,
487 "numa_hit %lu\n"
488 "numa_miss %lu\n"
489 "numa_foreign %lu\n"
490 "interleave_hit %lu\n"
491 "local_node %lu\n"
492 "other_node %lu\n",
493 sum_zone_numa_event_state(dev->id, NUMA_HIT),
494 sum_zone_numa_event_state(dev->id, NUMA_MISS),
495 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
496 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
497 sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
498 sum_zone_numa_event_state(dev->id, NUMA_OTHER));
499}
500static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
501
502static ssize_t node_read_vmstat(struct device *dev,
503 struct device_attribute *attr, char *buf)
504{
505 int nid = dev->id;
506 struct pglist_data *pgdat = NODE_DATA(nid);
507 int i;
508 int len = 0;
509
510 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
511 len += sysfs_emit_at(buf, len, "%s %lu\n",
512 zone_stat_name(i),
513 sum_zone_node_page_state(nid, i));
514
515#ifdef CONFIG_NUMA
516 fold_vm_numa_events();
517 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
518 len += sysfs_emit_at(buf, len, "%s %lu\n",
519 numa_stat_name(i),
520 sum_zone_numa_event_state(nid, i));
521
522#endif
523 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
524 unsigned long pages = node_page_state_pages(pgdat, i);
525
526 if (vmstat_item_print_in_thp(i))
527 pages /= HPAGE_PMD_NR;
528 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
529 pages);
530 }
531
532 return len;
533}
534static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
535
536static ssize_t node_read_distance(struct device *dev,
537 struct device_attribute *attr, char *buf)
538{
539 int nid = dev->id;
540 int len = 0;
541 int i;
542
543 /*
544 * buf is currently PAGE_SIZE in length and each node needs 4 chars
545 * at the most (distance + space or newline).
546 */
547 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
548
549 for_each_online_node(i) {
550 len += sysfs_emit_at(buf, len, "%s%d",
551 i ? " " : "", node_distance(nid, i));
552 }
553
554 len += sysfs_emit_at(buf, len, "\n");
555 return len;
556}
557static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
558
559static struct attribute *node_dev_attrs[] = {
560 &dev_attr_cpumap.attr,
561 &dev_attr_cpulist.attr,
562 &dev_attr_meminfo.attr,
563 &dev_attr_numastat.attr,
564 &dev_attr_distance.attr,
565 &dev_attr_vmstat.attr,
566 NULL
567};
568ATTRIBUTE_GROUPS(node_dev);
569
570#ifdef CONFIG_HUGETLBFS
571/*
572 * hugetlbfs per node attributes registration interface:
573 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
574 * it will register its per node attributes for all online nodes with
575 * memory. It will also call register_hugetlbfs_with_node(), below, to
576 * register its attribute registration functions with this node driver.
577 * Once these hooks have been initialized, the node driver will call into
578 * the hugetlb module to [un]register attributes for hot-plugged nodes.
579 */
580static node_registration_func_t __hugetlb_register_node;
581static node_registration_func_t __hugetlb_unregister_node;
582
583static inline bool hugetlb_register_node(struct node *node)
584{
585 if (__hugetlb_register_node &&
586 node_state(node->dev.id, N_MEMORY)) {
587 __hugetlb_register_node(node);
588 return true;
589 }
590 return false;
591}
592
593static inline void hugetlb_unregister_node(struct node *node)
594{
595 if (__hugetlb_unregister_node)
596 __hugetlb_unregister_node(node);
597}
598
599void register_hugetlbfs_with_node(node_registration_func_t doregister,
600 node_registration_func_t unregister)
601{
602 __hugetlb_register_node = doregister;
603 __hugetlb_unregister_node = unregister;
604}
605#else
606static inline void hugetlb_register_node(struct node *node) {}
607
608static inline void hugetlb_unregister_node(struct node *node) {}
609#endif
610
611static void node_device_release(struct device *dev)
612{
613 struct node *node = to_node(dev);
614
615#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
616 /*
617 * We schedule the work only when a memory section is
618 * onlined/offlined on this node. When we come here,
619 * all the memory on this node has been offlined,
620 * so we won't enqueue new work to this work.
621 *
622 * The work is using node->node_work, so we should
623 * flush work before freeing the memory.
624 */
625 flush_work(&node->node_work);
626#endif
627 kfree(node);
628}
629
630/*
631 * register_node - Setup a sysfs device for a node.
632 * @num - Node number to use when creating the device.
633 *
634 * Initialize and register the node device.
635 */
636static int register_node(struct node *node, int num)
637{
638 int error;
639
640 node->dev.id = num;
641 node->dev.bus = &node_subsys;
642 node->dev.release = node_device_release;
643 node->dev.groups = node_dev_groups;
644 error = device_register(&node->dev);
645
646 if (error)
647 put_device(&node->dev);
648 else {
649 hugetlb_register_node(node);
650
651 compaction_register_node(node);
652 }
653 return error;
654}
655
656/**
657 * unregister_node - unregister a node device
658 * @node: node going away
659 *
660 * Unregisters a node device @node. All the devices on the node must be
661 * unregistered before calling this function.
662 */
663void unregister_node(struct node *node)
664{
665 hugetlb_unregister_node(node); /* no-op, if memoryless node */
666 node_remove_accesses(node);
667 node_remove_caches(node);
668 device_unregister(&node->dev);
669}
670
671struct node *node_devices[MAX_NUMNODES];
672
673/*
674 * register cpu under node
675 */
676int register_cpu_under_node(unsigned int cpu, unsigned int nid)
677{
678 int ret;
679 struct device *obj;
680
681 if (!node_online(nid))
682 return 0;
683
684 obj = get_cpu_device(cpu);
685 if (!obj)
686 return 0;
687
688 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
689 &obj->kobj,
690 kobject_name(&obj->kobj));
691 if (ret)
692 return ret;
693
694 return sysfs_create_link(&obj->kobj,
695 &node_devices[nid]->dev.kobj,
696 kobject_name(&node_devices[nid]->dev.kobj));
697}
698
699/**
700 * register_memory_node_under_compute_node - link memory node to its compute
701 * node for a given access class.
702 * @mem_nid: Memory node number
703 * @cpu_nid: Cpu node number
704 * @access: Access class to register
705 *
706 * Description:
707 * For use with platforms that may have separate memory and compute nodes.
708 * This function will export node relationships linking which memory
709 * initiator nodes can access memory targets at a given ranked access
710 * class.
711 */
712int register_memory_node_under_compute_node(unsigned int mem_nid,
713 unsigned int cpu_nid,
714 unsigned access)
715{
716 struct node *init_node, *targ_node;
717 struct node_access_nodes *initiator, *target;
718 int ret;
719
720 if (!node_online(cpu_nid) || !node_online(mem_nid))
721 return -ENODEV;
722
723 init_node = node_devices[cpu_nid];
724 targ_node = node_devices[mem_nid];
725 initiator = node_init_node_access(init_node, access);
726 target = node_init_node_access(targ_node, access);
727 if (!initiator || !target)
728 return -ENOMEM;
729
730 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
731 &targ_node->dev.kobj,
732 dev_name(&targ_node->dev));
733 if (ret)
734 return ret;
735
736 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
737 &init_node->dev.kobj,
738 dev_name(&init_node->dev));
739 if (ret)
740 goto err;
741
742 return 0;
743 err:
744 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
745 dev_name(&targ_node->dev));
746 return ret;
747}
748
749int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
750{
751 struct device *obj;
752
753 if (!node_online(nid))
754 return 0;
755
756 obj = get_cpu_device(cpu);
757 if (!obj)
758 return 0;
759
760 sysfs_remove_link(&node_devices[nid]->dev.kobj,
761 kobject_name(&obj->kobj));
762 sysfs_remove_link(&obj->kobj,
763 kobject_name(&node_devices[nid]->dev.kobj));
764
765 return 0;
766}
767
768#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
769static int __ref get_nid_for_pfn(unsigned long pfn)
770{
771 if (!pfn_valid_within(pfn))
772 return -1;
773#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
774 if (system_state < SYSTEM_RUNNING)
775 return early_pfn_to_nid(pfn);
776#endif
777 return pfn_to_nid(pfn);
778}
779
780static void do_register_memory_block_under_node(int nid,
781 struct memory_block *mem_blk)
782{
783 int ret;
784
785 /*
786 * If this memory block spans multiple nodes, we only indicate
787 * the last processed node.
788 */
789 mem_blk->nid = nid;
790
791 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
792 &mem_blk->dev.kobj,
793 kobject_name(&mem_blk->dev.kobj));
794 if (ret && ret != -EEXIST)
795 dev_err_ratelimited(&node_devices[nid]->dev,
796 "can't create link to %s in sysfs (%d)\n",
797 kobject_name(&mem_blk->dev.kobj), ret);
798
799 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
800 &node_devices[nid]->dev.kobj,
801 kobject_name(&node_devices[nid]->dev.kobj));
802 if (ret && ret != -EEXIST)
803 dev_err_ratelimited(&mem_blk->dev,
804 "can't create link to %s in sysfs (%d)\n",
805 kobject_name(&node_devices[nid]->dev.kobj),
806 ret);
807}
808
809/* register memory section under specified node if it spans that node */
810static int register_mem_block_under_node_early(struct memory_block *mem_blk,
811 void *arg)
812{
813 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
814 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
815 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
816 int nid = *(int *)arg;
817 unsigned long pfn;
818
819 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
820 int page_nid;
821
822 /*
823 * memory block could have several absent sections from start.
824 * skip pfn range from absent section
825 */
826 if (!pfn_in_present_section(pfn)) {
827 pfn = round_down(pfn + PAGES_PER_SECTION,
828 PAGES_PER_SECTION) - 1;
829 continue;
830 }
831
832 /*
833 * We need to check if page belongs to nid only at the boot
834 * case because node's ranges can be interleaved.
835 */
836 page_nid = get_nid_for_pfn(pfn);
837 if (page_nid < 0)
838 continue;
839 if (page_nid != nid)
840 continue;
841
842 do_register_memory_block_under_node(nid, mem_blk);
843 return 0;
844 }
845 /* mem section does not span the specified node */
846 return 0;
847}
848
849/*
850 * During hotplug we know that all pages in the memory block belong to the same
851 * node.
852 */
853static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
854 void *arg)
855{
856 int nid = *(int *)arg;
857
858 do_register_memory_block_under_node(nid, mem_blk);
859 return 0;
860}
861
862/*
863 * Unregister a memory block device under the node it spans. Memory blocks
864 * with multiple nodes cannot be offlined and therefore also never be removed.
865 */
866void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
867{
868 if (mem_blk->nid == NUMA_NO_NODE)
869 return;
870
871 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
872 kobject_name(&mem_blk->dev.kobj));
873 sysfs_remove_link(&mem_blk->dev.kobj,
874 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
875}
876
877void link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn,
878 enum meminit_context context)
879{
880 walk_memory_blocks_func_t func;
881
882 if (context == MEMINIT_HOTPLUG)
883 func = register_mem_block_under_node_hotplug;
884 else
885 func = register_mem_block_under_node_early;
886
887 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
888 (void *)&nid, func);
889 return;
890}
891
892#ifdef CONFIG_HUGETLBFS
893/*
894 * Handle per node hstate attribute [un]registration on transistions
895 * to/from memoryless state.
896 */
897static void node_hugetlb_work(struct work_struct *work)
898{
899 struct node *node = container_of(work, struct node, node_work);
900
901 /*
902 * We only get here when a node transitions to/from memoryless state.
903 * We can detect which transition occurred by examining whether the
904 * node has memory now. hugetlb_register_node() already check this
905 * so we try to register the attributes. If that fails, then the
906 * node has transitioned to memoryless, try to unregister the
907 * attributes.
908 */
909 if (!hugetlb_register_node(node))
910 hugetlb_unregister_node(node);
911}
912
913static void init_node_hugetlb_work(int nid)
914{
915 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
916}
917
918static int node_memory_callback(struct notifier_block *self,
919 unsigned long action, void *arg)
920{
921 struct memory_notify *mnb = arg;
922 int nid = mnb->status_change_nid;
923
924 switch (action) {
925 case MEM_ONLINE:
926 case MEM_OFFLINE:
927 /*
928 * offload per node hstate [un]registration to a work thread
929 * when transitioning to/from memoryless state.
930 */
931 if (nid != NUMA_NO_NODE)
932 schedule_work(&node_devices[nid]->node_work);
933 break;
934
935 case MEM_GOING_ONLINE:
936 case MEM_GOING_OFFLINE:
937 case MEM_CANCEL_ONLINE:
938 case MEM_CANCEL_OFFLINE:
939 default:
940 break;
941 }
942
943 return NOTIFY_OK;
944}
945#endif /* CONFIG_HUGETLBFS */
946#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
947
948#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
949 !defined(CONFIG_HUGETLBFS)
950static inline int node_memory_callback(struct notifier_block *self,
951 unsigned long action, void *arg)
952{
953 return NOTIFY_OK;
954}
955
956static void init_node_hugetlb_work(int nid) { }
957
958#endif
959
960int __register_one_node(int nid)
961{
962 int error;
963 int cpu;
964
965 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
966 if (!node_devices[nid])
967 return -ENOMEM;
968
969 error = register_node(node_devices[nid], nid);
970
971 /* link cpu under this node */
972 for_each_present_cpu(cpu) {
973 if (cpu_to_node(cpu) == nid)
974 register_cpu_under_node(cpu, nid);
975 }
976
977 INIT_LIST_HEAD(&node_devices[nid]->access_list);
978 /* initialize work queue for memory hot plug */
979 init_node_hugetlb_work(nid);
980 node_init_caches(nid);
981
982 return error;
983}
984
985void unregister_one_node(int nid)
986{
987 if (!node_devices[nid])
988 return;
989
990 unregister_node(node_devices[nid]);
991 node_devices[nid] = NULL;
992}
993
994/*
995 * node states attributes
996 */
997
998struct node_attr {
999 struct device_attribute attr;
1000 enum node_states state;
1001};
1002
1003static ssize_t show_node_state(struct device *dev,
1004 struct device_attribute *attr, char *buf)
1005{
1006 struct node_attr *na = container_of(attr, struct node_attr, attr);
1007
1008 return sysfs_emit(buf, "%*pbl\n",
1009 nodemask_pr_args(&node_states[na->state]));
1010}
1011
1012#define _NODE_ATTR(name, state) \
1013 { __ATTR(name, 0444, show_node_state, NULL), state }
1014
1015static struct node_attr node_state_attr[] = {
1016 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1017 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1018 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1019#ifdef CONFIG_HIGHMEM
1020 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1021#endif
1022 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1023 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1024 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
1025 N_GENERIC_INITIATOR),
1026};
1027
1028static struct attribute *node_state_attrs[] = {
1029 &node_state_attr[N_POSSIBLE].attr.attr,
1030 &node_state_attr[N_ONLINE].attr.attr,
1031 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
1032#ifdef CONFIG_HIGHMEM
1033 &node_state_attr[N_HIGH_MEMORY].attr.attr,
1034#endif
1035 &node_state_attr[N_MEMORY].attr.attr,
1036 &node_state_attr[N_CPU].attr.attr,
1037 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
1038 NULL
1039};
1040
1041static const struct attribute_group memory_root_attr_group = {
1042 .attrs = node_state_attrs,
1043};
1044
1045static const struct attribute_group *cpu_root_attr_groups[] = {
1046 &memory_root_attr_group,
1047 NULL,
1048};
1049
1050#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
1051static int __init register_node_type(void)
1052{
1053 int ret;
1054
1055 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1056 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1057
1058 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1059 if (!ret) {
1060 static struct notifier_block node_memory_callback_nb = {
1061 .notifier_call = node_memory_callback,
1062 .priority = NODE_CALLBACK_PRI,
1063 };
1064 register_hotmemory_notifier(&node_memory_callback_nb);
1065 }
1066
1067 /*
1068 * Note: we're not going to unregister the node class if we fail
1069 * to register the node state class attribute files.
1070 */
1071 return ret;
1072}
1073postcore_initcall(register_node_type);