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