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/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);
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 pglist_data *pgdat = NODE_DATA(nid);
60 struct sysinfo i;
61
62 si_meminfo_node(&i, nid);
63 n = sprintf(buf,
64 "Node %d MemTotal: %8lu kB\n"
65 "Node %d MemFree: %8lu kB\n"
66 "Node %d MemUsed: %8lu kB\n"
67 "Node %d Active: %8lu kB\n"
68 "Node %d Inactive: %8lu kB\n"
69 "Node %d Active(anon): %8lu kB\n"
70 "Node %d Inactive(anon): %8lu kB\n"
71 "Node %d Active(file): %8lu kB\n"
72 "Node %d Inactive(file): %8lu kB\n"
73 "Node %d Unevictable: %8lu kB\n"
74 "Node %d Mlocked: %8lu kB\n",
75 nid, K(i.totalram),
76 nid, K(i.freeram),
77 nid, K(i.totalram - i.freeram),
78 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
79 node_page_state(pgdat, NR_ACTIVE_FILE)),
80 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
81 node_page_state(pgdat, NR_INACTIVE_FILE)),
82 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
83 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
84 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
85 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
86 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
87 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
88
89#ifdef CONFIG_HIGHMEM
90 n += sprintf(buf + n,
91 "Node %d HighTotal: %8lu kB\n"
92 "Node %d HighFree: %8lu kB\n"
93 "Node %d LowTotal: %8lu kB\n"
94 "Node %d LowFree: %8lu kB\n",
95 nid, K(i.totalhigh),
96 nid, K(i.freehigh),
97 nid, K(i.totalram - i.totalhigh),
98 nid, K(i.freeram - i.freehigh));
99#endif
100 n += sprintf(buf + n,
101 "Node %d Dirty: %8lu kB\n"
102 "Node %d Writeback: %8lu kB\n"
103 "Node %d FilePages: %8lu kB\n"
104 "Node %d Mapped: %8lu kB\n"
105 "Node %d AnonPages: %8lu kB\n"
106 "Node %d Shmem: %8lu kB\n"
107 "Node %d KernelStack: %8lu kB\n"
108 "Node %d PageTables: %8lu kB\n"
109 "Node %d NFS_Unstable: %8lu kB\n"
110 "Node %d Bounce: %8lu kB\n"
111 "Node %d WritebackTmp: %8lu kB\n"
112 "Node %d Slab: %8lu kB\n"
113 "Node %d SReclaimable: %8lu kB\n"
114 "Node %d SUnreclaim: %8lu kB\n"
115#ifdef CONFIG_TRANSPARENT_HUGEPAGE
116 "Node %d AnonHugePages: %8lu kB\n"
117 "Node %d ShmemHugePages: %8lu kB\n"
118 "Node %d ShmemPmdMapped: %8lu kB\n"
119#endif
120 ,
121 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
122 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
123 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
124 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
125 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
126 nid, K(i.sharedram),
127 nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
128 nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
129 nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
130 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
131 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
132 nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE) +
133 sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
134 nid, K(sum_zone_node_page_state(nid, NR_SLAB_RECLAIMABLE)),
135#ifdef CONFIG_TRANSPARENT_HUGEPAGE
136 nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
137 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
138 HPAGE_PMD_NR),
139 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
140 HPAGE_PMD_NR),
141 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
142 HPAGE_PMD_NR));
143#else
144 nid, K(sum_zone_node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
145#endif
146 n += hugetlb_report_node_meminfo(nid, buf + n);
147 return n;
148}
149
150#undef K
151static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
152
153static ssize_t node_read_numastat(struct device *dev,
154 struct device_attribute *attr, char *buf)
155{
156 return sprintf(buf,
157 "numa_hit %lu\n"
158 "numa_miss %lu\n"
159 "numa_foreign %lu\n"
160 "interleave_hit %lu\n"
161 "local_node %lu\n"
162 "other_node %lu\n",
163 sum_zone_node_page_state(dev->id, NUMA_HIT),
164 sum_zone_node_page_state(dev->id, NUMA_MISS),
165 sum_zone_node_page_state(dev->id, NUMA_FOREIGN),
166 sum_zone_node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
167 sum_zone_node_page_state(dev->id, NUMA_LOCAL),
168 sum_zone_node_page_state(dev->id, NUMA_OTHER));
169}
170static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
171
172static ssize_t node_read_vmstat(struct device *dev,
173 struct device_attribute *attr, char *buf)
174{
175 int nid = dev->id;
176 struct pglist_data *pgdat = NODE_DATA(nid);
177 int i;
178 int n = 0;
179
180 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
181 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
182 sum_zone_node_page_state(nid, i));
183
184 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
185 n += sprintf(buf+n, "%s %lu\n",
186 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
187 node_page_state(pgdat, 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
214static struct attribute *node_dev_attrs[] = {
215 &dev_attr_cpumap.attr,
216 &dev_attr_cpulist.attr,
217 &dev_attr_meminfo.attr,
218 &dev_attr_numastat.attr,
219 &dev_attr_distance.attr,
220 &dev_attr_vmstat.attr,
221 NULL
222};
223ATTRIBUTE_GROUPS(node_dev);
224
225#ifdef CONFIG_HUGETLBFS
226/*
227 * hugetlbfs per node attributes registration interface:
228 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
229 * it will register its per node attributes for all online nodes with
230 * memory. It will also call register_hugetlbfs_with_node(), below, to
231 * register its attribute registration functions with this node driver.
232 * Once these hooks have been initialized, the node driver will call into
233 * the hugetlb module to [un]register attributes for hot-plugged nodes.
234 */
235static node_registration_func_t __hugetlb_register_node;
236static node_registration_func_t __hugetlb_unregister_node;
237
238static inline bool hugetlb_register_node(struct node *node)
239{
240 if (__hugetlb_register_node &&
241 node_state(node->dev.id, N_MEMORY)) {
242 __hugetlb_register_node(node);
243 return true;
244 }
245 return false;
246}
247
248static inline void hugetlb_unregister_node(struct node *node)
249{
250 if (__hugetlb_unregister_node)
251 __hugetlb_unregister_node(node);
252}
253
254void register_hugetlbfs_with_node(node_registration_func_t doregister,
255 node_registration_func_t unregister)
256{
257 __hugetlb_register_node = doregister;
258 __hugetlb_unregister_node = unregister;
259}
260#else
261static inline void hugetlb_register_node(struct node *node) {}
262
263static inline void hugetlb_unregister_node(struct node *node) {}
264#endif
265
266static void node_device_release(struct device *dev)
267{
268 struct node *node = to_node(dev);
269
270#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
271 /*
272 * We schedule the work only when a memory section is
273 * onlined/offlined on this node. When we come here,
274 * all the memory on this node has been offlined,
275 * so we won't enqueue new work to this work.
276 *
277 * The work is using node->node_work, so we should
278 * flush work before freeing the memory.
279 */
280 flush_work(&node->node_work);
281#endif
282 kfree(node);
283}
284
285/*
286 * register_node - Setup a sysfs device for a node.
287 * @num - Node number to use when creating the device.
288 *
289 * Initialize and register the node device.
290 */
291static int register_node(struct node *node, int num, struct node *parent)
292{
293 int error;
294
295 node->dev.id = num;
296 node->dev.bus = &node_subsys;
297 node->dev.release = node_device_release;
298 node->dev.groups = node_dev_groups;
299 error = device_register(&node->dev);
300
301 if (!error){
302 hugetlb_register_node(node);
303
304 compaction_register_node(node);
305 }
306 return error;
307}
308
309/**
310 * unregister_node - unregister a node device
311 * @node: node going away
312 *
313 * Unregisters a node device @node. All the devices on the node must be
314 * unregistered before calling this function.
315 */
316void unregister_node(struct node *node)
317{
318 hugetlb_unregister_node(node); /* no-op, if memoryless node */
319
320 device_unregister(&node->dev);
321}
322
323struct node *node_devices[MAX_NUMNODES];
324
325/*
326 * register cpu under node
327 */
328int register_cpu_under_node(unsigned int cpu, unsigned int nid)
329{
330 int ret;
331 struct device *obj;
332
333 if (!node_online(nid))
334 return 0;
335
336 obj = get_cpu_device(cpu);
337 if (!obj)
338 return 0;
339
340 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
341 &obj->kobj,
342 kobject_name(&obj->kobj));
343 if (ret)
344 return ret;
345
346 return sysfs_create_link(&obj->kobj,
347 &node_devices[nid]->dev.kobj,
348 kobject_name(&node_devices[nid]->dev.kobj));
349}
350
351int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
352{
353 struct device *obj;
354
355 if (!node_online(nid))
356 return 0;
357
358 obj = get_cpu_device(cpu);
359 if (!obj)
360 return 0;
361
362 sysfs_remove_link(&node_devices[nid]->dev.kobj,
363 kobject_name(&obj->kobj));
364 sysfs_remove_link(&obj->kobj,
365 kobject_name(&node_devices[nid]->dev.kobj));
366
367 return 0;
368}
369
370#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
371#define page_initialized(page) (page->lru.next)
372
373static int __ref get_nid_for_pfn(unsigned long pfn)
374{
375 struct page *page;
376
377 if (!pfn_valid_within(pfn))
378 return -1;
379#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
380 if (system_state == SYSTEM_BOOTING)
381 return early_pfn_to_nid(pfn);
382#endif
383 page = pfn_to_page(pfn);
384 if (!page_initialized(page))
385 return -1;
386 return pfn_to_nid(pfn);
387}
388
389/* register memory section under specified node if it spans that node */
390int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
391{
392 int ret;
393 unsigned long pfn, sect_start_pfn, sect_end_pfn;
394
395 if (!mem_blk)
396 return -EFAULT;
397 if (!node_online(nid))
398 return 0;
399
400 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
401 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
402 sect_end_pfn += PAGES_PER_SECTION - 1;
403 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
404 int page_nid;
405
406 /*
407 * memory block could have several absent sections from start.
408 * skip pfn range from absent section
409 */
410 if (!pfn_present(pfn)) {
411 pfn = round_down(pfn + PAGES_PER_SECTION,
412 PAGES_PER_SECTION) - 1;
413 continue;
414 }
415
416 page_nid = get_nid_for_pfn(pfn);
417 if (page_nid < 0)
418 continue;
419 if (page_nid != nid)
420 continue;
421 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
422 &mem_blk->dev.kobj,
423 kobject_name(&mem_blk->dev.kobj));
424 if (ret)
425 return ret;
426
427 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
428 &node_devices[nid]->dev.kobj,
429 kobject_name(&node_devices[nid]->dev.kobj));
430 }
431 /* mem section does not span the specified node */
432 return 0;
433}
434
435/* unregister memory section under all nodes that it spans */
436int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
437 unsigned long phys_index)
438{
439 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
440 unsigned long pfn, sect_start_pfn, sect_end_pfn;
441
442 if (!mem_blk) {
443 NODEMASK_FREE(unlinked_nodes);
444 return -EFAULT;
445 }
446 if (!unlinked_nodes)
447 return -ENOMEM;
448 nodes_clear(*unlinked_nodes);
449
450 sect_start_pfn = section_nr_to_pfn(phys_index);
451 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
452 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
453 int nid;
454
455 nid = get_nid_for_pfn(pfn);
456 if (nid < 0)
457 continue;
458 if (!node_online(nid))
459 continue;
460 if (node_test_and_set(nid, *unlinked_nodes))
461 continue;
462 sysfs_remove_link(&node_devices[nid]->dev.kobj,
463 kobject_name(&mem_blk->dev.kobj));
464 sysfs_remove_link(&mem_blk->dev.kobj,
465 kobject_name(&node_devices[nid]->dev.kobj));
466 }
467 NODEMASK_FREE(unlinked_nodes);
468 return 0;
469}
470
471static int link_mem_sections(int nid)
472{
473 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
474 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
475 unsigned long pfn;
476 struct memory_block *mem_blk = NULL;
477 int err = 0;
478
479 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
480 unsigned long section_nr = pfn_to_section_nr(pfn);
481 struct mem_section *mem_sect;
482 int ret;
483
484 if (!present_section_nr(section_nr))
485 continue;
486 mem_sect = __nr_to_section(section_nr);
487
488 /* same memblock ? */
489 if (mem_blk)
490 if ((section_nr >= mem_blk->start_section_nr) &&
491 (section_nr <= mem_blk->end_section_nr))
492 continue;
493
494 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
495
496 ret = register_mem_sect_under_node(mem_blk, nid);
497 if (!err)
498 err = ret;
499
500 /* discard ref obtained in find_memory_block() */
501 }
502
503 if (mem_blk)
504 kobject_put(&mem_blk->dev.kobj);
505 return err;
506}
507
508#ifdef CONFIG_HUGETLBFS
509/*
510 * Handle per node hstate attribute [un]registration on transistions
511 * to/from memoryless state.
512 */
513static void node_hugetlb_work(struct work_struct *work)
514{
515 struct node *node = container_of(work, struct node, node_work);
516
517 /*
518 * We only get here when a node transitions to/from memoryless state.
519 * We can detect which transition occurred by examining whether the
520 * node has memory now. hugetlb_register_node() already check this
521 * so we try to register the attributes. If that fails, then the
522 * node has transitioned to memoryless, try to unregister the
523 * attributes.
524 */
525 if (!hugetlb_register_node(node))
526 hugetlb_unregister_node(node);
527}
528
529static void init_node_hugetlb_work(int nid)
530{
531 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
532}
533
534static int node_memory_callback(struct notifier_block *self,
535 unsigned long action, void *arg)
536{
537 struct memory_notify *mnb = arg;
538 int nid = mnb->status_change_nid;
539
540 switch (action) {
541 case MEM_ONLINE:
542 case MEM_OFFLINE:
543 /*
544 * offload per node hstate [un]registration to a work thread
545 * when transitioning to/from memoryless state.
546 */
547 if (nid != NUMA_NO_NODE)
548 schedule_work(&node_devices[nid]->node_work);
549 break;
550
551 case MEM_GOING_ONLINE:
552 case MEM_GOING_OFFLINE:
553 case MEM_CANCEL_ONLINE:
554 case MEM_CANCEL_OFFLINE:
555 default:
556 break;
557 }
558
559 return NOTIFY_OK;
560}
561#endif /* CONFIG_HUGETLBFS */
562#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
563
564static int link_mem_sections(int nid) { return 0; }
565#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
566
567#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
568 !defined(CONFIG_HUGETLBFS)
569static inline int node_memory_callback(struct notifier_block *self,
570 unsigned long action, void *arg)
571{
572 return NOTIFY_OK;
573}
574
575static void init_node_hugetlb_work(int nid) { }
576
577#endif
578
579int register_one_node(int nid)
580{
581 int error = 0;
582 int cpu;
583
584 if (node_online(nid)) {
585 int p_node = parent_node(nid);
586 struct node *parent = NULL;
587
588 if (p_node != nid)
589 parent = node_devices[p_node];
590
591 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
592 if (!node_devices[nid])
593 return -ENOMEM;
594
595 error = register_node(node_devices[nid], nid, parent);
596
597 /* link cpu under this node */
598 for_each_present_cpu(cpu) {
599 if (cpu_to_node(cpu) == nid)
600 register_cpu_under_node(cpu, nid);
601 }
602
603 /* link memory sections under this node */
604 error = link_mem_sections(nid);
605
606 /* initialize work queue for memory hot plug */
607 init_node_hugetlb_work(nid);
608 }
609
610 return error;
611
612}
613
614void unregister_one_node(int nid)
615{
616 if (!node_devices[nid])
617 return;
618
619 unregister_node(node_devices[nid]);
620 node_devices[nid] = NULL;
621}
622
623/*
624 * node states attributes
625 */
626
627static ssize_t print_nodes_state(enum node_states state, char *buf)
628{
629 int n;
630
631 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
632 nodemask_pr_args(&node_states[state]));
633 buf[n++] = '\n';
634 buf[n] = '\0';
635 return n;
636}
637
638struct node_attr {
639 struct device_attribute attr;
640 enum node_states state;
641};
642
643static ssize_t show_node_state(struct device *dev,
644 struct device_attribute *attr, char *buf)
645{
646 struct node_attr *na = container_of(attr, struct node_attr, attr);
647 return print_nodes_state(na->state, buf);
648}
649
650#define _NODE_ATTR(name, state) \
651 { __ATTR(name, 0444, show_node_state, NULL), state }
652
653static struct node_attr node_state_attr[] = {
654 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
655 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
656 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
657#ifdef CONFIG_HIGHMEM
658 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
659#endif
660#ifdef CONFIG_MOVABLE_NODE
661 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
662#endif
663 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
664};
665
666static struct attribute *node_state_attrs[] = {
667 &node_state_attr[N_POSSIBLE].attr.attr,
668 &node_state_attr[N_ONLINE].attr.attr,
669 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
670#ifdef CONFIG_HIGHMEM
671 &node_state_attr[N_HIGH_MEMORY].attr.attr,
672#endif
673#ifdef CONFIG_MOVABLE_NODE
674 &node_state_attr[N_MEMORY].attr.attr,
675#endif
676 &node_state_attr[N_CPU].attr.attr,
677 NULL
678};
679
680static struct attribute_group memory_root_attr_group = {
681 .attrs = node_state_attrs,
682};
683
684static const struct attribute_group *cpu_root_attr_groups[] = {
685 &memory_root_attr_group,
686 NULL,
687};
688
689#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
690static int __init register_node_type(void)
691{
692 int ret;
693
694 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
695 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
696
697 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
698 if (!ret) {
699 static struct notifier_block node_memory_callback_nb = {
700 .notifier_call = node_memory_callback,
701 .priority = NODE_CALLBACK_PRI,
702 };
703 register_hotmemory_notifier(&node_memory_callback_nb);
704 }
705
706 /*
707 * Note: we're not going to unregister the node class if we fail
708 * to register the node state class attribute files.
709 */
710 return ret;
711}
712postcore_initcall(register_node_type);