Loading...
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, bool list, char *buf)
29{
30 struct node *node_dev = to_node(dev);
31 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
32
33 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
34 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
35
36 return cpumap_print_to_pagebuf(list, buf, mask);
37}
38
39static inline ssize_t node_read_cpumask(struct device *dev,
40 struct device_attribute *attr, char *buf)
41{
42 return node_read_cpumap(dev, false, buf);
43}
44static inline ssize_t node_read_cpulist(struct device *dev,
45 struct device_attribute *attr, char *buf)
46{
47 return node_read_cpumap(dev, true, buf);
48}
49
50static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
51static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
52
53#define K(x) ((x) << (PAGE_SHIFT - 10))
54static ssize_t node_read_meminfo(struct device *dev,
55 struct device_attribute *attr, char *buf)
56{
57 int n;
58 int nid = dev->id;
59 struct sysinfo i;
60
61 si_meminfo_node(&i, nid);
62 n = sprintf(buf,
63 "Node %d MemTotal: %8lu kB\n"
64 "Node %d MemFree: %8lu kB\n"
65 "Node %d MemUsed: %8lu kB\n"
66 "Node %d Active: %8lu kB\n"
67 "Node %d Inactive: %8lu kB\n"
68 "Node %d Active(anon): %8lu kB\n"
69 "Node %d Inactive(anon): %8lu kB\n"
70 "Node %d Active(file): %8lu kB\n"
71 "Node %d Inactive(file): %8lu kB\n"
72 "Node %d Unevictable: %8lu kB\n"
73 "Node %d Mlocked: %8lu kB\n",
74 nid, K(i.totalram),
75 nid, K(i.freeram),
76 nid, K(i.totalram - i.freeram),
77 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
78 node_page_state(nid, NR_ACTIVE_FILE)),
79 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
80 node_page_state(nid, NR_INACTIVE_FILE)),
81 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
82 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
83 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
84 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
85 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
86 nid, K(node_page_state(nid, NR_MLOCK)));
87
88#ifdef CONFIG_HIGHMEM
89 n += sprintf(buf + n,
90 "Node %d HighTotal: %8lu kB\n"
91 "Node %d HighFree: %8lu kB\n"
92 "Node %d LowTotal: %8lu kB\n"
93 "Node %d LowFree: %8lu kB\n",
94 nid, K(i.totalhigh),
95 nid, K(i.freehigh),
96 nid, K(i.totalram - i.totalhigh),
97 nid, K(i.freeram - i.freehigh));
98#endif
99 n += sprintf(buf + n,
100 "Node %d Dirty: %8lu kB\n"
101 "Node %d Writeback: %8lu kB\n"
102 "Node %d FilePages: %8lu kB\n"
103 "Node %d Mapped: %8lu kB\n"
104 "Node %d AnonPages: %8lu kB\n"
105 "Node %d Shmem: %8lu kB\n"
106 "Node %d KernelStack: %8lu kB\n"
107 "Node %d PageTables: %8lu kB\n"
108 "Node %d NFS_Unstable: %8lu kB\n"
109 "Node %d Bounce: %8lu kB\n"
110 "Node %d WritebackTmp: %8lu kB\n"
111 "Node %d Slab: %8lu kB\n"
112 "Node %d SReclaimable: %8lu kB\n"
113 "Node %d SUnreclaim: %8lu kB\n"
114#ifdef CONFIG_TRANSPARENT_HUGEPAGE
115 "Node %d AnonHugePages: %8lu kB\n"
116#endif
117 ,
118 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
119 nid, K(node_page_state(nid, NR_WRITEBACK)),
120 nid, K(node_page_state(nid, NR_FILE_PAGES)),
121 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
122 nid, K(node_page_state(nid, NR_ANON_PAGES)),
123 nid, K(i.sharedram),
124 nid, node_page_state(nid, NR_KERNEL_STACK) *
125 THREAD_SIZE / 1024,
126 nid, K(node_page_state(nid, NR_PAGETABLE)),
127 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
128 nid, K(node_page_state(nid, NR_BOUNCE)),
129 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
130 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
131 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
132 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
133#ifdef CONFIG_TRANSPARENT_HUGEPAGE
134 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
135 , nid,
136 K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
137 HPAGE_PMD_NR));
138#else
139 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
140#endif
141 n += hugetlb_report_node_meminfo(nid, buf + n);
142 return n;
143}
144
145#undef K
146static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
147
148static ssize_t node_read_numastat(struct device *dev,
149 struct device_attribute *attr, char *buf)
150{
151 return sprintf(buf,
152 "numa_hit %lu\n"
153 "numa_miss %lu\n"
154 "numa_foreign %lu\n"
155 "interleave_hit %lu\n"
156 "local_node %lu\n"
157 "other_node %lu\n",
158 node_page_state(dev->id, NUMA_HIT),
159 node_page_state(dev->id, NUMA_MISS),
160 node_page_state(dev->id, NUMA_FOREIGN),
161 node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
162 node_page_state(dev->id, NUMA_LOCAL),
163 node_page_state(dev->id, NUMA_OTHER));
164}
165static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
166
167static ssize_t node_read_vmstat(struct device *dev,
168 struct device_attribute *attr, char *buf)
169{
170 int nid = dev->id;
171 int i;
172 int n = 0;
173
174 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
175 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
176 node_page_state(nid, i));
177
178 return n;
179}
180static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
181
182static ssize_t node_read_distance(struct device *dev,
183 struct device_attribute *attr, char *buf)
184{
185 int nid = dev->id;
186 int len = 0;
187 int i;
188
189 /*
190 * buf is currently PAGE_SIZE in length and each node needs 4 chars
191 * at the most (distance + space or newline).
192 */
193 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
194
195 for_each_online_node(i)
196 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
197
198 len += sprintf(buf + len, "\n");
199 return len;
200}
201static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
202
203static struct attribute *node_dev_attrs[] = {
204 &dev_attr_cpumap.attr,
205 &dev_attr_cpulist.attr,
206 &dev_attr_meminfo.attr,
207 &dev_attr_numastat.attr,
208 &dev_attr_distance.attr,
209 &dev_attr_vmstat.attr,
210 NULL
211};
212ATTRIBUTE_GROUPS(node_dev);
213
214#ifdef CONFIG_HUGETLBFS
215/*
216 * hugetlbfs per node attributes registration interface:
217 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
218 * it will register its per node attributes for all online nodes with
219 * memory. It will also call register_hugetlbfs_with_node(), below, to
220 * register its attribute registration functions with this node driver.
221 * Once these hooks have been initialized, the node driver will call into
222 * the hugetlb module to [un]register attributes for hot-plugged nodes.
223 */
224static node_registration_func_t __hugetlb_register_node;
225static node_registration_func_t __hugetlb_unregister_node;
226
227static inline bool hugetlb_register_node(struct node *node)
228{
229 if (__hugetlb_register_node &&
230 node_state(node->dev.id, N_MEMORY)) {
231 __hugetlb_register_node(node);
232 return true;
233 }
234 return false;
235}
236
237static inline void hugetlb_unregister_node(struct node *node)
238{
239 if (__hugetlb_unregister_node)
240 __hugetlb_unregister_node(node);
241}
242
243void register_hugetlbfs_with_node(node_registration_func_t doregister,
244 node_registration_func_t unregister)
245{
246 __hugetlb_register_node = doregister;
247 __hugetlb_unregister_node = unregister;
248}
249#else
250static inline void hugetlb_register_node(struct node *node) {}
251
252static inline void hugetlb_unregister_node(struct node *node) {}
253#endif
254
255static void node_device_release(struct device *dev)
256{
257 struct node *node = to_node(dev);
258
259#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
260 /*
261 * We schedule the work only when a memory section is
262 * onlined/offlined on this node. When we come here,
263 * all the memory on this node has been offlined,
264 * so we won't enqueue new work to this work.
265 *
266 * The work is using node->node_work, so we should
267 * flush work before freeing the memory.
268 */
269 flush_work(&node->node_work);
270#endif
271 kfree(node);
272}
273
274/*
275 * register_node - Setup a sysfs device for a node.
276 * @num - Node number to use when creating the device.
277 *
278 * Initialize and register the node device.
279 */
280static int register_node(struct node *node, int num, struct node *parent)
281{
282 int error;
283
284 node->dev.id = num;
285 node->dev.bus = &node_subsys;
286 node->dev.release = node_device_release;
287 node->dev.groups = node_dev_groups;
288 error = device_register(&node->dev);
289
290 if (!error){
291 hugetlb_register_node(node);
292
293 compaction_register_node(node);
294 }
295 return error;
296}
297
298/**
299 * unregister_node - unregister a node device
300 * @node: node going away
301 *
302 * Unregisters a node device @node. All the devices on the node must be
303 * unregistered before calling this function.
304 */
305void unregister_node(struct node *node)
306{
307 hugetlb_unregister_node(node); /* no-op, if memoryless node */
308
309 device_unregister(&node->dev);
310}
311
312struct node *node_devices[MAX_NUMNODES];
313
314/*
315 * register cpu under node
316 */
317int register_cpu_under_node(unsigned int cpu, unsigned int nid)
318{
319 int ret;
320 struct device *obj;
321
322 if (!node_online(nid))
323 return 0;
324
325 obj = get_cpu_device(cpu);
326 if (!obj)
327 return 0;
328
329 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
330 &obj->kobj,
331 kobject_name(&obj->kobj));
332 if (ret)
333 return ret;
334
335 return sysfs_create_link(&obj->kobj,
336 &node_devices[nid]->dev.kobj,
337 kobject_name(&node_devices[nid]->dev.kobj));
338}
339
340int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
341{
342 struct device *obj;
343
344 if (!node_online(nid))
345 return 0;
346
347 obj = get_cpu_device(cpu);
348 if (!obj)
349 return 0;
350
351 sysfs_remove_link(&node_devices[nid]->dev.kobj,
352 kobject_name(&obj->kobj));
353 sysfs_remove_link(&obj->kobj,
354 kobject_name(&node_devices[nid]->dev.kobj));
355
356 return 0;
357}
358
359#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
360#define page_initialized(page) (page->lru.next)
361
362static int __init_refok get_nid_for_pfn(unsigned long pfn)
363{
364 struct page *page;
365
366 if (!pfn_valid_within(pfn))
367 return -1;
368#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
369 if (system_state == SYSTEM_BOOTING)
370 return early_pfn_to_nid(pfn);
371#endif
372 page = pfn_to_page(pfn);
373 if (!page_initialized(page))
374 return -1;
375 return pfn_to_nid(pfn);
376}
377
378/* register memory section under specified node if it spans that node */
379int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
380{
381 int ret;
382 unsigned long pfn, sect_start_pfn, sect_end_pfn;
383
384 if (!mem_blk)
385 return -EFAULT;
386 if (!node_online(nid))
387 return 0;
388
389 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
390 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
391 sect_end_pfn += PAGES_PER_SECTION - 1;
392 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
393 int page_nid;
394
395 /*
396 * memory block could have several absent sections from start.
397 * skip pfn range from absent section
398 */
399 if (!pfn_present(pfn)) {
400 pfn = round_down(pfn + PAGES_PER_SECTION,
401 PAGES_PER_SECTION) - 1;
402 continue;
403 }
404
405 page_nid = get_nid_for_pfn(pfn);
406 if (page_nid < 0)
407 continue;
408 if (page_nid != nid)
409 continue;
410 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
411 &mem_blk->dev.kobj,
412 kobject_name(&mem_blk->dev.kobj));
413 if (ret)
414 return ret;
415
416 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
417 &node_devices[nid]->dev.kobj,
418 kobject_name(&node_devices[nid]->dev.kobj));
419 }
420 /* mem section does not span the specified node */
421 return 0;
422}
423
424/* unregister memory section under all nodes that it spans */
425int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
426 unsigned long phys_index)
427{
428 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
429 unsigned long pfn, sect_start_pfn, sect_end_pfn;
430
431 if (!mem_blk) {
432 NODEMASK_FREE(unlinked_nodes);
433 return -EFAULT;
434 }
435 if (!unlinked_nodes)
436 return -ENOMEM;
437 nodes_clear(*unlinked_nodes);
438
439 sect_start_pfn = section_nr_to_pfn(phys_index);
440 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
441 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
442 int nid;
443
444 nid = get_nid_for_pfn(pfn);
445 if (nid < 0)
446 continue;
447 if (!node_online(nid))
448 continue;
449 if (node_test_and_set(nid, *unlinked_nodes))
450 continue;
451 sysfs_remove_link(&node_devices[nid]->dev.kobj,
452 kobject_name(&mem_blk->dev.kobj));
453 sysfs_remove_link(&mem_blk->dev.kobj,
454 kobject_name(&node_devices[nid]->dev.kobj));
455 }
456 NODEMASK_FREE(unlinked_nodes);
457 return 0;
458}
459
460static int link_mem_sections(int nid)
461{
462 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
463 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
464 unsigned long pfn;
465 struct memory_block *mem_blk = NULL;
466 int err = 0;
467
468 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
469 unsigned long section_nr = pfn_to_section_nr(pfn);
470 struct mem_section *mem_sect;
471 int ret;
472
473 if (!present_section_nr(section_nr))
474 continue;
475 mem_sect = __nr_to_section(section_nr);
476
477 /* same memblock ? */
478 if (mem_blk)
479 if ((section_nr >= mem_blk->start_section_nr) &&
480 (section_nr <= mem_blk->end_section_nr))
481 continue;
482
483 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
484
485 ret = register_mem_sect_under_node(mem_blk, nid);
486 if (!err)
487 err = ret;
488
489 /* discard ref obtained in find_memory_block() */
490 }
491
492 if (mem_blk)
493 kobject_put(&mem_blk->dev.kobj);
494 return err;
495}
496
497#ifdef CONFIG_HUGETLBFS
498/*
499 * Handle per node hstate attribute [un]registration on transistions
500 * to/from memoryless state.
501 */
502static void node_hugetlb_work(struct work_struct *work)
503{
504 struct node *node = container_of(work, struct node, node_work);
505
506 /*
507 * We only get here when a node transitions to/from memoryless state.
508 * We can detect which transition occurred by examining whether the
509 * node has memory now. hugetlb_register_node() already check this
510 * so we try to register the attributes. If that fails, then the
511 * node has transitioned to memoryless, try to unregister the
512 * attributes.
513 */
514 if (!hugetlb_register_node(node))
515 hugetlb_unregister_node(node);
516}
517
518static void init_node_hugetlb_work(int nid)
519{
520 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
521}
522
523static int node_memory_callback(struct notifier_block *self,
524 unsigned long action, void *arg)
525{
526 struct memory_notify *mnb = arg;
527 int nid = mnb->status_change_nid;
528
529 switch (action) {
530 case MEM_ONLINE:
531 case MEM_OFFLINE:
532 /*
533 * offload per node hstate [un]registration to a work thread
534 * when transitioning to/from memoryless state.
535 */
536 if (nid != NUMA_NO_NODE)
537 schedule_work(&node_devices[nid]->node_work);
538 break;
539
540 case MEM_GOING_ONLINE:
541 case MEM_GOING_OFFLINE:
542 case MEM_CANCEL_ONLINE:
543 case MEM_CANCEL_OFFLINE:
544 default:
545 break;
546 }
547
548 return NOTIFY_OK;
549}
550#endif /* CONFIG_HUGETLBFS */
551#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
552
553static int link_mem_sections(int nid) { return 0; }
554#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
555
556#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
557 !defined(CONFIG_HUGETLBFS)
558static inline int node_memory_callback(struct notifier_block *self,
559 unsigned long action, void *arg)
560{
561 return NOTIFY_OK;
562}
563
564static void init_node_hugetlb_work(int nid) { }
565
566#endif
567
568int register_one_node(int nid)
569{
570 int error = 0;
571 int cpu;
572
573 if (node_online(nid)) {
574 int p_node = parent_node(nid);
575 struct node *parent = NULL;
576
577 if (p_node != nid)
578 parent = node_devices[p_node];
579
580 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
581 if (!node_devices[nid])
582 return -ENOMEM;
583
584 error = register_node(node_devices[nid], nid, parent);
585
586 /* link cpu under this node */
587 for_each_present_cpu(cpu) {
588 if (cpu_to_node(cpu) == nid)
589 register_cpu_under_node(cpu, nid);
590 }
591
592 /* link memory sections under this node */
593 error = link_mem_sections(nid);
594
595 /* initialize work queue for memory hot plug */
596 init_node_hugetlb_work(nid);
597 }
598
599 return error;
600
601}
602
603void unregister_one_node(int nid)
604{
605 if (!node_devices[nid])
606 return;
607
608 unregister_node(node_devices[nid]);
609 node_devices[nid] = NULL;
610}
611
612/*
613 * node states attributes
614 */
615
616static ssize_t print_nodes_state(enum node_states state, char *buf)
617{
618 int n;
619
620 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
621 nodemask_pr_args(&node_states[state]));
622 buf[n++] = '\n';
623 buf[n] = '\0';
624 return n;
625}
626
627struct node_attr {
628 struct device_attribute attr;
629 enum node_states state;
630};
631
632static ssize_t show_node_state(struct device *dev,
633 struct device_attribute *attr, char *buf)
634{
635 struct node_attr *na = container_of(attr, struct node_attr, attr);
636 return print_nodes_state(na->state, buf);
637}
638
639#define _NODE_ATTR(name, state) \
640 { __ATTR(name, 0444, show_node_state, NULL), state }
641
642static struct node_attr node_state_attr[] = {
643 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
644 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
645 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
646#ifdef CONFIG_HIGHMEM
647 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
648#endif
649#ifdef CONFIG_MOVABLE_NODE
650 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
651#endif
652 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
653};
654
655static struct attribute *node_state_attrs[] = {
656 &node_state_attr[N_POSSIBLE].attr.attr,
657 &node_state_attr[N_ONLINE].attr.attr,
658 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
659#ifdef CONFIG_HIGHMEM
660 &node_state_attr[N_HIGH_MEMORY].attr.attr,
661#endif
662#ifdef CONFIG_MOVABLE_NODE
663 &node_state_attr[N_MEMORY].attr.attr,
664#endif
665 &node_state_attr[N_CPU].attr.attr,
666 NULL
667};
668
669static struct attribute_group memory_root_attr_group = {
670 .attrs = node_state_attrs,
671};
672
673static const struct attribute_group *cpu_root_attr_groups[] = {
674 &memory_root_attr_group,
675 NULL,
676};
677
678#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
679static int __init register_node_type(void)
680{
681 int ret;
682
683 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
684 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
685
686 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
687 if (!ret) {
688 static struct notifier_block node_memory_callback_nb = {
689 .notifier_call = node_memory_callback,
690 .priority = NODE_CALLBACK_PRI,
691 };
692 register_hotmemory_notifier(&node_memory_callback_nb);
693 }
694
695 /*
696 * Note: we're not going to unregister the node class if we fail
697 * to register the node state class attribute files.
698 */
699 return ret;
700}
701postcore_initcall(register_node_type);
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/node.h>
11#include <linux/hugetlb.h>
12#include <linux/compaction.h>
13#include <linux/cpumask.h>
14#include <linux/topology.h>
15#include <linux/nodemask.h>
16#include <linux/cpu.h>
17#include <linux/device.h>
18#include <linux/swap.h>
19#include <linux/slab.h>
20
21static struct bus_type node_subsys = {
22 .name = "node",
23 .dev_name = "node",
24};
25
26
27static ssize_t node_read_cpumap(struct device *dev, int type, char *buf)
28{
29 struct node *node_dev = to_node(dev);
30 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
31 int len;
32
33 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
34 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
35
36 len = type?
37 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
38 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
39 buf[len++] = '\n';
40 buf[len] = '\0';
41 return len;
42}
43
44static inline ssize_t node_read_cpumask(struct device *dev,
45 struct device_attribute *attr, char *buf)
46{
47 return node_read_cpumap(dev, 0, buf);
48}
49static inline ssize_t node_read_cpulist(struct device *dev,
50 struct device_attribute *attr, char *buf)
51{
52 return node_read_cpumap(dev, 1, buf);
53}
54
55static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
56static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
57
58#define K(x) ((x) << (PAGE_SHIFT - 10))
59static ssize_t node_read_meminfo(struct device *dev,
60 struct device_attribute *attr, char *buf)
61{
62 int n;
63 int nid = dev->id;
64 struct sysinfo i;
65
66 si_meminfo_node(&i, nid);
67 n = sprintf(buf,
68 "Node %d MemTotal: %8lu kB\n"
69 "Node %d MemFree: %8lu kB\n"
70 "Node %d MemUsed: %8lu kB\n"
71 "Node %d Active: %8lu kB\n"
72 "Node %d Inactive: %8lu kB\n"
73 "Node %d Active(anon): %8lu kB\n"
74 "Node %d Inactive(anon): %8lu kB\n"
75 "Node %d Active(file): %8lu kB\n"
76 "Node %d Inactive(file): %8lu kB\n"
77 "Node %d Unevictable: %8lu kB\n"
78 "Node %d Mlocked: %8lu kB\n",
79 nid, K(i.totalram),
80 nid, K(i.freeram),
81 nid, K(i.totalram - i.freeram),
82 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
83 node_page_state(nid, NR_ACTIVE_FILE)),
84 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
85 node_page_state(nid, NR_INACTIVE_FILE)),
86 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
87 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
88 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
89 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
90 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
91 nid, K(node_page_state(nid, NR_MLOCK)));
92
93#ifdef CONFIG_HIGHMEM
94 n += sprintf(buf + n,
95 "Node %d HighTotal: %8lu kB\n"
96 "Node %d HighFree: %8lu kB\n"
97 "Node %d LowTotal: %8lu kB\n"
98 "Node %d LowFree: %8lu kB\n",
99 nid, K(i.totalhigh),
100 nid, K(i.freehigh),
101 nid, K(i.totalram - i.totalhigh),
102 nid, K(i.freeram - i.freehigh));
103#endif
104 n += sprintf(buf + n,
105 "Node %d Dirty: %8lu kB\n"
106 "Node %d Writeback: %8lu kB\n"
107 "Node %d FilePages: %8lu kB\n"
108 "Node %d Mapped: %8lu kB\n"
109 "Node %d AnonPages: %8lu kB\n"
110 "Node %d Shmem: %8lu kB\n"
111 "Node %d KernelStack: %8lu kB\n"
112 "Node %d PageTables: %8lu kB\n"
113 "Node %d NFS_Unstable: %8lu kB\n"
114 "Node %d Bounce: %8lu kB\n"
115 "Node %d WritebackTmp: %8lu kB\n"
116 "Node %d Slab: %8lu kB\n"
117 "Node %d SReclaimable: %8lu kB\n"
118 "Node %d SUnreclaim: %8lu kB\n"
119#ifdef CONFIG_TRANSPARENT_HUGEPAGE
120 "Node %d AnonHugePages: %8lu kB\n"
121#endif
122 ,
123 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
124 nid, K(node_page_state(nid, NR_WRITEBACK)),
125 nid, K(node_page_state(nid, NR_FILE_PAGES)),
126 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
127#ifdef CONFIG_TRANSPARENT_HUGEPAGE
128 nid, K(node_page_state(nid, NR_ANON_PAGES)
129 + node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
130 HPAGE_PMD_NR),
131#else
132 nid, K(node_page_state(nid, NR_ANON_PAGES)),
133#endif
134 nid, K(node_page_state(nid, NR_SHMEM)),
135 nid, node_page_state(nid, NR_KERNEL_STACK) *
136 THREAD_SIZE / 1024,
137 nid, K(node_page_state(nid, NR_PAGETABLE)),
138 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
139 nid, K(node_page_state(nid, NR_BOUNCE)),
140 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
141 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
142 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
143 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
144#ifdef CONFIG_TRANSPARENT_HUGEPAGE
145 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
146 , nid,
147 K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
148 HPAGE_PMD_NR));
149#else
150 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
151#endif
152 n += hugetlb_report_node_meminfo(nid, buf + n);
153 return n;
154}
155
156#undef K
157static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
158
159static ssize_t node_read_numastat(struct device *dev,
160 struct device_attribute *attr, char *buf)
161{
162 return sprintf(buf,
163 "numa_hit %lu\n"
164 "numa_miss %lu\n"
165 "numa_foreign %lu\n"
166 "interleave_hit %lu\n"
167 "local_node %lu\n"
168 "other_node %lu\n",
169 node_page_state(dev->id, NUMA_HIT),
170 node_page_state(dev->id, NUMA_MISS),
171 node_page_state(dev->id, NUMA_FOREIGN),
172 node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
173 node_page_state(dev->id, NUMA_LOCAL),
174 node_page_state(dev->id, NUMA_OTHER));
175}
176static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
177
178static ssize_t node_read_vmstat(struct device *dev,
179 struct device_attribute *attr, char *buf)
180{
181 int nid = dev->id;
182 int i;
183 int n = 0;
184
185 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
186 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
187 node_page_state(nid, i));
188
189 return n;
190}
191static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
192
193static ssize_t node_read_distance(struct device *dev,
194 struct device_attribute *attr, char * buf)
195{
196 int nid = dev->id;
197 int len = 0;
198 int i;
199
200 /*
201 * buf is currently PAGE_SIZE in length and each node needs 4 chars
202 * at the most (distance + space or newline).
203 */
204 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
205
206 for_each_online_node(i)
207 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
208
209 len += sprintf(buf + len, "\n");
210 return len;
211}
212static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
213
214#ifdef CONFIG_HUGETLBFS
215/*
216 * hugetlbfs per node attributes registration interface:
217 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
218 * it will register its per node attributes for all online nodes with
219 * memory. It will also call register_hugetlbfs_with_node(), below, to
220 * register its attribute registration functions with this node driver.
221 * Once these hooks have been initialized, the node driver will call into
222 * the hugetlb module to [un]register attributes for hot-plugged nodes.
223 */
224static node_registration_func_t __hugetlb_register_node;
225static node_registration_func_t __hugetlb_unregister_node;
226
227static inline bool hugetlb_register_node(struct node *node)
228{
229 if (__hugetlb_register_node &&
230 node_state(node->dev.id, N_HIGH_MEMORY)) {
231 __hugetlb_register_node(node);
232 return true;
233 }
234 return false;
235}
236
237static inline void hugetlb_unregister_node(struct node *node)
238{
239 if (__hugetlb_unregister_node)
240 __hugetlb_unregister_node(node);
241}
242
243void register_hugetlbfs_with_node(node_registration_func_t doregister,
244 node_registration_func_t unregister)
245{
246 __hugetlb_register_node = doregister;
247 __hugetlb_unregister_node = unregister;
248}
249#else
250static inline void hugetlb_register_node(struct node *node) {}
251
252static inline void hugetlb_unregister_node(struct node *node) {}
253#endif
254
255
256/*
257 * register_node - Setup a sysfs device for a node.
258 * @num - Node number to use when creating the device.
259 *
260 * Initialize and register the node device.
261 */
262int register_node(struct node *node, int num, struct node *parent)
263{
264 int error;
265
266 node->dev.id = num;
267 node->dev.bus = &node_subsys;
268 error = device_register(&node->dev);
269
270 if (!error){
271 device_create_file(&node->dev, &dev_attr_cpumap);
272 device_create_file(&node->dev, &dev_attr_cpulist);
273 device_create_file(&node->dev, &dev_attr_meminfo);
274 device_create_file(&node->dev, &dev_attr_numastat);
275 device_create_file(&node->dev, &dev_attr_distance);
276 device_create_file(&node->dev, &dev_attr_vmstat);
277
278 scan_unevictable_register_node(node);
279
280 hugetlb_register_node(node);
281
282 compaction_register_node(node);
283 }
284 return error;
285}
286
287/**
288 * unregister_node - unregister a node device
289 * @node: node going away
290 *
291 * Unregisters a node device @node. All the devices on the node must be
292 * unregistered before calling this function.
293 */
294void unregister_node(struct node *node)
295{
296 device_remove_file(&node->dev, &dev_attr_cpumap);
297 device_remove_file(&node->dev, &dev_attr_cpulist);
298 device_remove_file(&node->dev, &dev_attr_meminfo);
299 device_remove_file(&node->dev, &dev_attr_numastat);
300 device_remove_file(&node->dev, &dev_attr_distance);
301 device_remove_file(&node->dev, &dev_attr_vmstat);
302
303 scan_unevictable_unregister_node(node);
304 hugetlb_unregister_node(node); /* no-op, if memoryless node */
305
306 device_unregister(&node->dev);
307}
308
309struct node node_devices[MAX_NUMNODES];
310
311/*
312 * register cpu under node
313 */
314int register_cpu_under_node(unsigned int cpu, unsigned int nid)
315{
316 int ret;
317 struct device *obj;
318
319 if (!node_online(nid))
320 return 0;
321
322 obj = get_cpu_device(cpu);
323 if (!obj)
324 return 0;
325
326 ret = sysfs_create_link(&node_devices[nid].dev.kobj,
327 &obj->kobj,
328 kobject_name(&obj->kobj));
329 if (ret)
330 return ret;
331
332 return sysfs_create_link(&obj->kobj,
333 &node_devices[nid].dev.kobj,
334 kobject_name(&node_devices[nid].dev.kobj));
335}
336
337int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
338{
339 struct device *obj;
340
341 if (!node_online(nid))
342 return 0;
343
344 obj = get_cpu_device(cpu);
345 if (!obj)
346 return 0;
347
348 sysfs_remove_link(&node_devices[nid].dev.kobj,
349 kobject_name(&obj->kobj));
350 sysfs_remove_link(&obj->kobj,
351 kobject_name(&node_devices[nid].dev.kobj));
352
353 return 0;
354}
355
356#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
357#define page_initialized(page) (page->lru.next)
358
359static int get_nid_for_pfn(unsigned long pfn)
360{
361 struct page *page;
362
363 if (!pfn_valid_within(pfn))
364 return -1;
365 page = pfn_to_page(pfn);
366 if (!page_initialized(page))
367 return -1;
368 return pfn_to_nid(pfn);
369}
370
371/* register memory section under specified node if it spans that node */
372int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
373{
374 int ret;
375 unsigned long pfn, sect_start_pfn, sect_end_pfn;
376
377 if (!mem_blk)
378 return -EFAULT;
379 if (!node_online(nid))
380 return 0;
381
382 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
383 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
384 sect_end_pfn += PAGES_PER_SECTION - 1;
385 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
386 int page_nid;
387
388 page_nid = get_nid_for_pfn(pfn);
389 if (page_nid < 0)
390 continue;
391 if (page_nid != nid)
392 continue;
393 ret = sysfs_create_link_nowarn(&node_devices[nid].dev.kobj,
394 &mem_blk->dev.kobj,
395 kobject_name(&mem_blk->dev.kobj));
396 if (ret)
397 return ret;
398
399 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
400 &node_devices[nid].dev.kobj,
401 kobject_name(&node_devices[nid].dev.kobj));
402 }
403 /* mem section does not span the specified node */
404 return 0;
405}
406
407/* unregister memory section under all nodes that it spans */
408int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
409 unsigned long phys_index)
410{
411 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
412 unsigned long pfn, sect_start_pfn, sect_end_pfn;
413
414 if (!mem_blk) {
415 NODEMASK_FREE(unlinked_nodes);
416 return -EFAULT;
417 }
418 if (!unlinked_nodes)
419 return -ENOMEM;
420 nodes_clear(*unlinked_nodes);
421
422 sect_start_pfn = section_nr_to_pfn(phys_index);
423 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
424 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
425 int nid;
426
427 nid = get_nid_for_pfn(pfn);
428 if (nid < 0)
429 continue;
430 if (!node_online(nid))
431 continue;
432 if (node_test_and_set(nid, *unlinked_nodes))
433 continue;
434 sysfs_remove_link(&node_devices[nid].dev.kobj,
435 kobject_name(&mem_blk->dev.kobj));
436 sysfs_remove_link(&mem_blk->dev.kobj,
437 kobject_name(&node_devices[nid].dev.kobj));
438 }
439 NODEMASK_FREE(unlinked_nodes);
440 return 0;
441}
442
443static int link_mem_sections(int nid)
444{
445 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
446 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
447 unsigned long pfn;
448 struct memory_block *mem_blk = NULL;
449 int err = 0;
450
451 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
452 unsigned long section_nr = pfn_to_section_nr(pfn);
453 struct mem_section *mem_sect;
454 int ret;
455
456 if (!present_section_nr(section_nr))
457 continue;
458 mem_sect = __nr_to_section(section_nr);
459
460 /* same memblock ? */
461 if (mem_blk)
462 if ((section_nr >= mem_blk->start_section_nr) &&
463 (section_nr <= mem_blk->end_section_nr))
464 continue;
465
466 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
467
468 ret = register_mem_sect_under_node(mem_blk, nid);
469 if (!err)
470 err = ret;
471
472 /* discard ref obtained in find_memory_block() */
473 }
474
475 if (mem_blk)
476 kobject_put(&mem_blk->dev.kobj);
477 return err;
478}
479
480#ifdef CONFIG_HUGETLBFS
481/*
482 * Handle per node hstate attribute [un]registration on transistions
483 * to/from memoryless state.
484 */
485static void node_hugetlb_work(struct work_struct *work)
486{
487 struct node *node = container_of(work, struct node, node_work);
488
489 /*
490 * We only get here when a node transitions to/from memoryless state.
491 * We can detect which transition occurred by examining whether the
492 * node has memory now. hugetlb_register_node() already check this
493 * so we try to register the attributes. If that fails, then the
494 * node has transitioned to memoryless, try to unregister the
495 * attributes.
496 */
497 if (!hugetlb_register_node(node))
498 hugetlb_unregister_node(node);
499}
500
501static void init_node_hugetlb_work(int nid)
502{
503 INIT_WORK(&node_devices[nid].node_work, node_hugetlb_work);
504}
505
506static int node_memory_callback(struct notifier_block *self,
507 unsigned long action, void *arg)
508{
509 struct memory_notify *mnb = arg;
510 int nid = mnb->status_change_nid;
511
512 switch (action) {
513 case MEM_ONLINE:
514 case MEM_OFFLINE:
515 /*
516 * offload per node hstate [un]registration to a work thread
517 * when transitioning to/from memoryless state.
518 */
519 if (nid != NUMA_NO_NODE)
520 schedule_work(&node_devices[nid].node_work);
521 break;
522
523 case MEM_GOING_ONLINE:
524 case MEM_GOING_OFFLINE:
525 case MEM_CANCEL_ONLINE:
526 case MEM_CANCEL_OFFLINE:
527 default:
528 break;
529 }
530
531 return NOTIFY_OK;
532}
533#endif /* CONFIG_HUGETLBFS */
534#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
535
536static int link_mem_sections(int nid) { return 0; }
537#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
538
539#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
540 !defined(CONFIG_HUGETLBFS)
541static inline int node_memory_callback(struct notifier_block *self,
542 unsigned long action, void *arg)
543{
544 return NOTIFY_OK;
545}
546
547static void init_node_hugetlb_work(int nid) { }
548
549#endif
550
551int register_one_node(int nid)
552{
553 int error = 0;
554 int cpu;
555
556 if (node_online(nid)) {
557 int p_node = parent_node(nid);
558 struct node *parent = NULL;
559
560 if (p_node != nid)
561 parent = &node_devices[p_node];
562
563 error = register_node(&node_devices[nid], nid, parent);
564
565 /* link cpu under this node */
566 for_each_present_cpu(cpu) {
567 if (cpu_to_node(cpu) == nid)
568 register_cpu_under_node(cpu, nid);
569 }
570
571 /* link memory sections under this node */
572 error = link_mem_sections(nid);
573
574 /* initialize work queue for memory hot plug */
575 init_node_hugetlb_work(nid);
576 }
577
578 return error;
579
580}
581
582void unregister_one_node(int nid)
583{
584 unregister_node(&node_devices[nid]);
585}
586
587/*
588 * node states attributes
589 */
590
591static ssize_t print_nodes_state(enum node_states state, char *buf)
592{
593 int n;
594
595 n = nodelist_scnprintf(buf, PAGE_SIZE-2, node_states[state]);
596 buf[n++] = '\n';
597 buf[n] = '\0';
598 return n;
599}
600
601struct node_attr {
602 struct device_attribute attr;
603 enum node_states state;
604};
605
606static ssize_t show_node_state(struct device *dev,
607 struct device_attribute *attr, char *buf)
608{
609 struct node_attr *na = container_of(attr, struct node_attr, attr);
610 return print_nodes_state(na->state, buf);
611}
612
613#define _NODE_ATTR(name, state) \
614 { __ATTR(name, 0444, show_node_state, NULL), state }
615
616static struct node_attr node_state_attr[] = {
617 _NODE_ATTR(possible, N_POSSIBLE),
618 _NODE_ATTR(online, N_ONLINE),
619 _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
620 _NODE_ATTR(has_cpu, N_CPU),
621#ifdef CONFIG_HIGHMEM
622 _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
623#endif
624};
625
626static struct attribute *node_state_attrs[] = {
627 &node_state_attr[0].attr.attr,
628 &node_state_attr[1].attr.attr,
629 &node_state_attr[2].attr.attr,
630 &node_state_attr[3].attr.attr,
631#ifdef CONFIG_HIGHMEM
632 &node_state_attr[4].attr.attr,
633#endif
634 NULL
635};
636
637static struct attribute_group memory_root_attr_group = {
638 .attrs = node_state_attrs,
639};
640
641static const struct attribute_group *cpu_root_attr_groups[] = {
642 &memory_root_attr_group,
643 NULL,
644};
645
646#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
647static int __init register_node_type(void)
648{
649 int ret;
650
651 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
652 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
653
654 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
655 if (!ret) {
656 hotplug_memory_notifier(node_memory_callback,
657 NODE_CALLBACK_PRI);
658 }
659
660 /*
661 * Note: we're not going to unregister the node class if we fail
662 * to register the node state class attribute files.
663 */
664 return ret;
665}
666postcore_initcall(register_node_type);