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, int type, char *buf)
29{
30 struct node *node_dev = to_node(dev);
31 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
32 int len;
33
34 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
35 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
36
37 len = type?
38 cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
39 cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
40 buf[len++] = '\n';
41 buf[len] = '\0';
42 return len;
43}
44
45static inline ssize_t node_read_cpumask(struct device *dev,
46 struct device_attribute *attr, char *buf)
47{
48 return node_read_cpumap(dev, 0, buf);
49}
50static inline ssize_t node_read_cpulist(struct device *dev,
51 struct device_attribute *attr, char *buf)
52{
53 return node_read_cpumap(dev, 1, buf);
54}
55
56static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
57static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
58
59#define K(x) ((x) << (PAGE_SHIFT - 10))
60static ssize_t node_read_meminfo(struct device *dev,
61 struct device_attribute *attr, char *buf)
62{
63 int n;
64 int nid = dev->id;
65 struct sysinfo i;
66
67 si_meminfo_node(&i, nid);
68 n = sprintf(buf,
69 "Node %d MemTotal: %8lu kB\n"
70 "Node %d MemFree: %8lu kB\n"
71 "Node %d MemUsed: %8lu kB\n"
72 "Node %d Active: %8lu kB\n"
73 "Node %d Inactive: %8lu kB\n"
74 "Node %d Active(anon): %8lu kB\n"
75 "Node %d Inactive(anon): %8lu kB\n"
76 "Node %d Active(file): %8lu kB\n"
77 "Node %d Inactive(file): %8lu kB\n"
78 "Node %d Unevictable: %8lu kB\n"
79 "Node %d Mlocked: %8lu kB\n",
80 nid, K(i.totalram),
81 nid, K(i.freeram),
82 nid, K(i.totalram - i.freeram),
83 nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
84 node_page_state(nid, NR_ACTIVE_FILE)),
85 nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
86 node_page_state(nid, NR_INACTIVE_FILE)),
87 nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
88 nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
89 nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
90 nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
91 nid, K(node_page_state(nid, NR_UNEVICTABLE)),
92 nid, K(node_page_state(nid, NR_MLOCK)));
93
94#ifdef CONFIG_HIGHMEM
95 n += sprintf(buf + n,
96 "Node %d HighTotal: %8lu kB\n"
97 "Node %d HighFree: %8lu kB\n"
98 "Node %d LowTotal: %8lu kB\n"
99 "Node %d LowFree: %8lu kB\n",
100 nid, K(i.totalhigh),
101 nid, K(i.freehigh),
102 nid, K(i.totalram - i.totalhigh),
103 nid, K(i.freeram - i.freehigh));
104#endif
105 n += sprintf(buf + n,
106 "Node %d Dirty: %8lu kB\n"
107 "Node %d Writeback: %8lu kB\n"
108 "Node %d FilePages: %8lu kB\n"
109 "Node %d Mapped: %8lu kB\n"
110 "Node %d AnonPages: %8lu kB\n"
111 "Node %d Shmem: %8lu kB\n"
112 "Node %d KernelStack: %8lu kB\n"
113 "Node %d PageTables: %8lu kB\n"
114 "Node %d NFS_Unstable: %8lu kB\n"
115 "Node %d Bounce: %8lu kB\n"
116 "Node %d WritebackTmp: %8lu kB\n"
117 "Node %d Slab: %8lu kB\n"
118 "Node %d SReclaimable: %8lu kB\n"
119 "Node %d SUnreclaim: %8lu kB\n"
120#ifdef CONFIG_TRANSPARENT_HUGEPAGE
121 "Node %d AnonHugePages: %8lu kB\n"
122#endif
123 ,
124 nid, K(node_page_state(nid, NR_FILE_DIRTY)),
125 nid, K(node_page_state(nid, NR_WRITEBACK)),
126 nid, K(node_page_state(nid, NR_FILE_PAGES)),
127 nid, K(node_page_state(nid, NR_FILE_MAPPED)),
128 nid, K(node_page_state(nid, NR_ANON_PAGES)),
129 nid, K(node_page_state(nid, NR_SHMEM)),
130 nid, node_page_state(nid, NR_KERNEL_STACK) *
131 THREAD_SIZE / 1024,
132 nid, K(node_page_state(nid, NR_PAGETABLE)),
133 nid, K(node_page_state(nid, NR_UNSTABLE_NFS)),
134 nid, K(node_page_state(nid, NR_BOUNCE)),
135 nid, K(node_page_state(nid, NR_WRITEBACK_TEMP)),
136 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE) +
137 node_page_state(nid, NR_SLAB_UNRECLAIMABLE)),
138 nid, K(node_page_state(nid, NR_SLAB_RECLAIMABLE)),
139#ifdef CONFIG_TRANSPARENT_HUGEPAGE
140 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE))
141 , nid,
142 K(node_page_state(nid, NR_ANON_TRANSPARENT_HUGEPAGES) *
143 HPAGE_PMD_NR));
144#else
145 nid, K(node_page_state(nid, NR_SLAB_UNRECLAIMABLE)));
146#endif
147 n += hugetlb_report_node_meminfo(nid, buf + n);
148 return n;
149}
150
151#undef K
152static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
153
154static ssize_t node_read_numastat(struct device *dev,
155 struct device_attribute *attr, char *buf)
156{
157 return sprintf(buf,
158 "numa_hit %lu\n"
159 "numa_miss %lu\n"
160 "numa_foreign %lu\n"
161 "interleave_hit %lu\n"
162 "local_node %lu\n"
163 "other_node %lu\n",
164 node_page_state(dev->id, NUMA_HIT),
165 node_page_state(dev->id, NUMA_MISS),
166 node_page_state(dev->id, NUMA_FOREIGN),
167 node_page_state(dev->id, NUMA_INTERLEAVE_HIT),
168 node_page_state(dev->id, NUMA_LOCAL),
169 node_page_state(dev->id, NUMA_OTHER));
170}
171static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
172
173static ssize_t node_read_vmstat(struct device *dev,
174 struct device_attribute *attr, char *buf)
175{
176 int nid = dev->id;
177 int i;
178 int n = 0;
179
180 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
181 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
182 node_page_state(nid, i));
183
184 return n;
185}
186static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
187
188static ssize_t node_read_distance(struct device *dev,
189 struct device_attribute *attr, char * buf)
190{
191 int nid = dev->id;
192 int len = 0;
193 int i;
194
195 /*
196 * buf is currently PAGE_SIZE in length and each node needs 4 chars
197 * at the most (distance + space or newline).
198 */
199 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
200
201 for_each_online_node(i)
202 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
203
204 len += sprintf(buf + len, "\n");
205 return len;
206}
207static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
208
209#ifdef CONFIG_HUGETLBFS
210/*
211 * hugetlbfs per node attributes registration interface:
212 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
213 * it will register its per node attributes for all online nodes with
214 * memory. It will also call register_hugetlbfs_with_node(), below, to
215 * register its attribute registration functions with this node driver.
216 * Once these hooks have been initialized, the node driver will call into
217 * the hugetlb module to [un]register attributes for hot-plugged nodes.
218 */
219static node_registration_func_t __hugetlb_register_node;
220static node_registration_func_t __hugetlb_unregister_node;
221
222static inline bool hugetlb_register_node(struct node *node)
223{
224 if (__hugetlb_register_node &&
225 node_state(node->dev.id, N_MEMORY)) {
226 __hugetlb_register_node(node);
227 return true;
228 }
229 return false;
230}
231
232static inline void hugetlb_unregister_node(struct node *node)
233{
234 if (__hugetlb_unregister_node)
235 __hugetlb_unregister_node(node);
236}
237
238void register_hugetlbfs_with_node(node_registration_func_t doregister,
239 node_registration_func_t unregister)
240{
241 __hugetlb_register_node = doregister;
242 __hugetlb_unregister_node = unregister;
243}
244#else
245static inline void hugetlb_register_node(struct node *node) {}
246
247static inline void hugetlb_unregister_node(struct node *node) {}
248#endif
249
250static void node_device_release(struct device *dev)
251{
252 struct node *node = to_node(dev);
253
254#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
255 /*
256 * We schedule the work only when a memory section is
257 * onlined/offlined on this node. When we come here,
258 * all the memory on this node has been offlined,
259 * so we won't enqueue new work to this work.
260 *
261 * The work is using node->node_work, so we should
262 * flush work before freeing the memory.
263 */
264 flush_work(&node->node_work);
265#endif
266 kfree(node);
267}
268
269/*
270 * register_node - Setup a sysfs device for a node.
271 * @num - Node number to use when creating the device.
272 *
273 * Initialize and register the node device.
274 */
275static int register_node(struct node *node, int num, struct node *parent)
276{
277 int error;
278
279 node->dev.id = num;
280 node->dev.bus = &node_subsys;
281 node->dev.release = node_device_release;
282 error = device_register(&node->dev);
283
284 if (!error){
285 device_create_file(&node->dev, &dev_attr_cpumap);
286 device_create_file(&node->dev, &dev_attr_cpulist);
287 device_create_file(&node->dev, &dev_attr_meminfo);
288 device_create_file(&node->dev, &dev_attr_numastat);
289 device_create_file(&node->dev, &dev_attr_distance);
290 device_create_file(&node->dev, &dev_attr_vmstat);
291
292 scan_unevictable_register_node(node);
293
294 hugetlb_register_node(node);
295
296 compaction_register_node(node);
297 }
298 return error;
299}
300
301/**
302 * unregister_node - unregister a node device
303 * @node: node going away
304 *
305 * Unregisters a node device @node. All the devices on the node must be
306 * unregistered before calling this function.
307 */
308void unregister_node(struct node *node)
309{
310 device_remove_file(&node->dev, &dev_attr_cpumap);
311 device_remove_file(&node->dev, &dev_attr_cpulist);
312 device_remove_file(&node->dev, &dev_attr_meminfo);
313 device_remove_file(&node->dev, &dev_attr_numastat);
314 device_remove_file(&node->dev, &dev_attr_distance);
315 device_remove_file(&node->dev, &dev_attr_vmstat);
316
317 scan_unevictable_unregister_node(node);
318 hugetlb_unregister_node(node); /* no-op, if memoryless node */
319
320 device_unregister(&node->dev);
321}
322
323struct node *node_devices[MAX_NUMNODES];
324
325/*
326 * register cpu under node
327 */
328int register_cpu_under_node(unsigned int cpu, unsigned int nid)
329{
330 int ret;
331 struct device *obj;
332
333 if (!node_online(nid))
334 return 0;
335
336 obj = get_cpu_device(cpu);
337 if (!obj)
338 return 0;
339
340 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
341 &obj->kobj,
342 kobject_name(&obj->kobj));
343 if (ret)
344 return ret;
345
346 return sysfs_create_link(&obj->kobj,
347 &node_devices[nid]->dev.kobj,
348 kobject_name(&node_devices[nid]->dev.kobj));
349}
350
351int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
352{
353 struct device *obj;
354
355 if (!node_online(nid))
356 return 0;
357
358 obj = get_cpu_device(cpu);
359 if (!obj)
360 return 0;
361
362 sysfs_remove_link(&node_devices[nid]->dev.kobj,
363 kobject_name(&obj->kobj));
364 sysfs_remove_link(&obj->kobj,
365 kobject_name(&node_devices[nid]->dev.kobj));
366
367 return 0;
368}
369
370#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
371#define page_initialized(page) (page->lru.next)
372
373static int get_nid_for_pfn(unsigned long pfn)
374{
375 struct page *page;
376
377 if (!pfn_valid_within(pfn))
378 return -1;
379 page = pfn_to_page(pfn);
380 if (!page_initialized(page))
381 return -1;
382 return pfn_to_nid(pfn);
383}
384
385/* register memory section under specified node if it spans that node */
386int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
387{
388 int ret;
389 unsigned long pfn, sect_start_pfn, sect_end_pfn;
390
391 if (!mem_blk)
392 return -EFAULT;
393 if (!node_online(nid))
394 return 0;
395
396 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
397 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
398 sect_end_pfn += PAGES_PER_SECTION - 1;
399 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
400 int page_nid;
401
402 page_nid = get_nid_for_pfn(pfn);
403 if (page_nid < 0)
404 continue;
405 if (page_nid != nid)
406 continue;
407 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
408 &mem_blk->dev.kobj,
409 kobject_name(&mem_blk->dev.kobj));
410 if (ret)
411 return ret;
412
413 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
414 &node_devices[nid]->dev.kobj,
415 kobject_name(&node_devices[nid]->dev.kobj));
416 }
417 /* mem section does not span the specified node */
418 return 0;
419}
420
421/* unregister memory section under all nodes that it spans */
422int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
423 unsigned long phys_index)
424{
425 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
426 unsigned long pfn, sect_start_pfn, sect_end_pfn;
427
428 if (!mem_blk) {
429 NODEMASK_FREE(unlinked_nodes);
430 return -EFAULT;
431 }
432 if (!unlinked_nodes)
433 return -ENOMEM;
434 nodes_clear(*unlinked_nodes);
435
436 sect_start_pfn = section_nr_to_pfn(phys_index);
437 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
438 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
439 int nid;
440
441 nid = get_nid_for_pfn(pfn);
442 if (nid < 0)
443 continue;
444 if (!node_online(nid))
445 continue;
446 if (node_test_and_set(nid, *unlinked_nodes))
447 continue;
448 sysfs_remove_link(&node_devices[nid]->dev.kobj,
449 kobject_name(&mem_blk->dev.kobj));
450 sysfs_remove_link(&mem_blk->dev.kobj,
451 kobject_name(&node_devices[nid]->dev.kobj));
452 }
453 NODEMASK_FREE(unlinked_nodes);
454 return 0;
455}
456
457static int link_mem_sections(int nid)
458{
459 unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
460 unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
461 unsigned long pfn;
462 struct memory_block *mem_blk = NULL;
463 int err = 0;
464
465 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
466 unsigned long section_nr = pfn_to_section_nr(pfn);
467 struct mem_section *mem_sect;
468 int ret;
469
470 if (!present_section_nr(section_nr))
471 continue;
472 mem_sect = __nr_to_section(section_nr);
473
474 /* same memblock ? */
475 if (mem_blk)
476 if ((section_nr >= mem_blk->start_section_nr) &&
477 (section_nr <= mem_blk->end_section_nr))
478 continue;
479
480 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
481
482 ret = register_mem_sect_under_node(mem_blk, nid);
483 if (!err)
484 err = ret;
485
486 /* discard ref obtained in find_memory_block() */
487 }
488
489 if (mem_blk)
490 kobject_put(&mem_blk->dev.kobj);
491 return err;
492}
493
494#ifdef CONFIG_HUGETLBFS
495/*
496 * Handle per node hstate attribute [un]registration on transistions
497 * to/from memoryless state.
498 */
499static void node_hugetlb_work(struct work_struct *work)
500{
501 struct node *node = container_of(work, struct node, node_work);
502
503 /*
504 * We only get here when a node transitions to/from memoryless state.
505 * We can detect which transition occurred by examining whether the
506 * node has memory now. hugetlb_register_node() already check this
507 * so we try to register the attributes. If that fails, then the
508 * node has transitioned to memoryless, try to unregister the
509 * attributes.
510 */
511 if (!hugetlb_register_node(node))
512 hugetlb_unregister_node(node);
513}
514
515static void init_node_hugetlb_work(int nid)
516{
517 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
518}
519
520static int node_memory_callback(struct notifier_block *self,
521 unsigned long action, void *arg)
522{
523 struct memory_notify *mnb = arg;
524 int nid = mnb->status_change_nid;
525
526 switch (action) {
527 case MEM_ONLINE:
528 case MEM_OFFLINE:
529 /*
530 * offload per node hstate [un]registration to a work thread
531 * when transitioning to/from memoryless state.
532 */
533 if (nid != NUMA_NO_NODE)
534 schedule_work(&node_devices[nid]->node_work);
535 break;
536
537 case MEM_GOING_ONLINE:
538 case MEM_GOING_OFFLINE:
539 case MEM_CANCEL_ONLINE:
540 case MEM_CANCEL_OFFLINE:
541 default:
542 break;
543 }
544
545 return NOTIFY_OK;
546}
547#endif /* CONFIG_HUGETLBFS */
548#else /* !CONFIG_MEMORY_HOTPLUG_SPARSE */
549
550static int link_mem_sections(int nid) { return 0; }
551#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
552
553#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
554 !defined(CONFIG_HUGETLBFS)
555static inline int node_memory_callback(struct notifier_block *self,
556 unsigned long action, void *arg)
557{
558 return NOTIFY_OK;
559}
560
561static void init_node_hugetlb_work(int nid) { }
562
563#endif
564
565int register_one_node(int nid)
566{
567 int error = 0;
568 int cpu;
569
570 if (node_online(nid)) {
571 int p_node = parent_node(nid);
572 struct node *parent = NULL;
573
574 if (p_node != nid)
575 parent = node_devices[p_node];
576
577 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
578 if (!node_devices[nid])
579 return -ENOMEM;
580
581 error = register_node(node_devices[nid], nid, parent);
582
583 /* link cpu under this node */
584 for_each_present_cpu(cpu) {
585 if (cpu_to_node(cpu) == nid)
586 register_cpu_under_node(cpu, nid);
587 }
588
589 /* link memory sections under this node */
590 error = link_mem_sections(nid);
591
592 /* initialize work queue for memory hot plug */
593 init_node_hugetlb_work(nid);
594 }
595
596 return error;
597
598}
599
600void unregister_one_node(int nid)
601{
602 if (!node_devices[nid])
603 return;
604
605 unregister_node(node_devices[nid]);
606 kfree(node_devices[nid]);
607 node_devices[nid] = NULL;
608}
609
610/*
611 * node states attributes
612 */
613
614static ssize_t print_nodes_state(enum node_states state, char *buf)
615{
616 int n;
617
618 n = nodelist_scnprintf(buf, PAGE_SIZE-2, node_states[state]);
619 buf[n++] = '\n';
620 buf[n] = '\0';
621 return n;
622}
623
624struct node_attr {
625 struct device_attribute attr;
626 enum node_states state;
627};
628
629static ssize_t show_node_state(struct device *dev,
630 struct device_attribute *attr, char *buf)
631{
632 struct node_attr *na = container_of(attr, struct node_attr, attr);
633 return print_nodes_state(na->state, buf);
634}
635
636#define _NODE_ATTR(name, state) \
637 { __ATTR(name, 0444, show_node_state, NULL), state }
638
639static struct node_attr node_state_attr[] = {
640 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
641 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
642 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
643#ifdef CONFIG_HIGHMEM
644 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
645#endif
646#ifdef CONFIG_MOVABLE_NODE
647 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
648#endif
649 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
650};
651
652static struct attribute *node_state_attrs[] = {
653 &node_state_attr[N_POSSIBLE].attr.attr,
654 &node_state_attr[N_ONLINE].attr.attr,
655 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
656#ifdef CONFIG_HIGHMEM
657 &node_state_attr[N_HIGH_MEMORY].attr.attr,
658#endif
659#ifdef CONFIG_MOVABLE_NODE
660 &node_state_attr[N_MEMORY].attr.attr,
661#endif
662 &node_state_attr[N_CPU].attr.attr,
663 NULL
664};
665
666static struct attribute_group memory_root_attr_group = {
667 .attrs = node_state_attrs,
668};
669
670static const struct attribute_group *cpu_root_attr_groups[] = {
671 &memory_root_attr_group,
672 NULL,
673};
674
675#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
676static int __init register_node_type(void)
677{
678 int ret;
679
680 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
681 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
682
683 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
684 if (!ret) {
685 static struct notifier_block node_memory_callback_nb = {
686 .notifier_call = node_memory_callback,
687 .priority = NODE_CALLBACK_PRI,
688 };
689 register_hotmemory_notifier(&node_memory_callback_nb);
690 }
691
692 /*
693 * Note: we're not going to unregister the node class if we fail
694 * to register the node state class attribute files.
695 */
696 return ret;
697}
698postcore_initcall(register_node_type);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Basic Node interface support
4 */
5
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/mm.h>
9#include <linux/memory.h>
10#include <linux/vmstat.h>
11#include <linux/notifier.h>
12#include <linux/node.h>
13#include <linux/hugetlb.h>
14#include <linux/compaction.h>
15#include <linux/cpumask.h>
16#include <linux/topology.h>
17#include <linux/nodemask.h>
18#include <linux/cpu.h>
19#include <linux/device.h>
20#include <linux/pm_runtime.h>
21#include <linux/swap.h>
22#include <linux/slab.h>
23#include <linux/hugetlb.h>
24
25static struct bus_type node_subsys = {
26 .name = "node",
27 .dev_name = "node",
28};
29
30static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
31 struct bin_attribute *attr, char *buf,
32 loff_t off, size_t count)
33{
34 struct device *dev = kobj_to_dev(kobj);
35 struct node *node_dev = to_node(dev);
36 cpumask_var_t mask;
37 ssize_t n;
38
39 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
40 return 0;
41
42 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
43 n = cpumap_print_bitmask_to_buf(buf, mask, off, count);
44 free_cpumask_var(mask);
45
46 return n;
47}
48
49static BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
50
51static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
52 struct bin_attribute *attr, char *buf,
53 loff_t off, size_t count)
54{
55 struct device *dev = kobj_to_dev(kobj);
56 struct node *node_dev = to_node(dev);
57 cpumask_var_t mask;
58 ssize_t n;
59
60 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
61 return 0;
62
63 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
64 n = cpumap_print_list_to_buf(buf, mask, off, count);
65 free_cpumask_var(mask);
66
67 return n;
68}
69
70static BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
71
72/**
73 * struct node_access_nodes - Access class device to hold user visible
74 * relationships to other nodes.
75 * @dev: Device for this memory access class
76 * @list_node: List element in the node's access list
77 * @access: The access class rank
78 * @hmem_attrs: Heterogeneous memory performance attributes
79 */
80struct node_access_nodes {
81 struct device dev;
82 struct list_head list_node;
83 unsigned int access;
84#ifdef CONFIG_HMEM_REPORTING
85 struct node_hmem_attrs hmem_attrs;
86#endif
87};
88#define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
89
90static struct attribute *node_init_access_node_attrs[] = {
91 NULL,
92};
93
94static struct attribute *node_targ_access_node_attrs[] = {
95 NULL,
96};
97
98static const struct attribute_group initiators = {
99 .name = "initiators",
100 .attrs = node_init_access_node_attrs,
101};
102
103static const struct attribute_group targets = {
104 .name = "targets",
105 .attrs = node_targ_access_node_attrs,
106};
107
108static const struct attribute_group *node_access_node_groups[] = {
109 &initiators,
110 &targets,
111 NULL,
112};
113
114static void node_remove_accesses(struct node *node)
115{
116 struct node_access_nodes *c, *cnext;
117
118 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
119 list_del(&c->list_node);
120 device_unregister(&c->dev);
121 }
122}
123
124static void node_access_release(struct device *dev)
125{
126 kfree(to_access_nodes(dev));
127}
128
129static struct node_access_nodes *node_init_node_access(struct node *node,
130 unsigned int access)
131{
132 struct node_access_nodes *access_node;
133 struct device *dev;
134
135 list_for_each_entry(access_node, &node->access_list, list_node)
136 if (access_node->access == access)
137 return access_node;
138
139 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
140 if (!access_node)
141 return NULL;
142
143 access_node->access = access;
144 dev = &access_node->dev;
145 dev->parent = &node->dev;
146 dev->release = node_access_release;
147 dev->groups = node_access_node_groups;
148 if (dev_set_name(dev, "access%u", access))
149 goto free;
150
151 if (device_register(dev))
152 goto free_name;
153
154 pm_runtime_no_callbacks(dev);
155 list_add_tail(&access_node->list_node, &node->access_list);
156 return access_node;
157free_name:
158 kfree_const(dev->kobj.name);
159free:
160 kfree(access_node);
161 return NULL;
162}
163
164#ifdef CONFIG_HMEM_REPORTING
165#define ACCESS_ATTR(name) \
166static ssize_t name##_show(struct device *dev, \
167 struct device_attribute *attr, \
168 char *buf) \
169{ \
170 return sysfs_emit(buf, "%u\n", \
171 to_access_nodes(dev)->hmem_attrs.name); \
172} \
173static DEVICE_ATTR_RO(name)
174
175ACCESS_ATTR(read_bandwidth);
176ACCESS_ATTR(read_latency);
177ACCESS_ATTR(write_bandwidth);
178ACCESS_ATTR(write_latency);
179
180static struct attribute *access_attrs[] = {
181 &dev_attr_read_bandwidth.attr,
182 &dev_attr_read_latency.attr,
183 &dev_attr_write_bandwidth.attr,
184 &dev_attr_write_latency.attr,
185 NULL,
186};
187
188/**
189 * node_set_perf_attrs - Set the performance values for given access class
190 * @nid: Node identifier to be set
191 * @hmem_attrs: Heterogeneous memory performance attributes
192 * @access: The access class the for the given attributes
193 */
194void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
195 unsigned int access)
196{
197 struct node_access_nodes *c;
198 struct node *node;
199 int i;
200
201 if (WARN_ON_ONCE(!node_online(nid)))
202 return;
203
204 node = node_devices[nid];
205 c = node_init_node_access(node, access);
206 if (!c)
207 return;
208
209 c->hmem_attrs = *hmem_attrs;
210 for (i = 0; access_attrs[i] != NULL; i++) {
211 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
212 "initiators")) {
213 pr_info("failed to add performance attribute to node %d\n",
214 nid);
215 break;
216 }
217 }
218}
219
220/**
221 * struct node_cache_info - Internal tracking for memory node caches
222 * @dev: Device represeting the cache level
223 * @node: List element for tracking in the node
224 * @cache_attrs:Attributes for this cache level
225 */
226struct node_cache_info {
227 struct device dev;
228 struct list_head node;
229 struct node_cache_attrs cache_attrs;
230};
231#define to_cache_info(device) container_of(device, struct node_cache_info, dev)
232
233#define CACHE_ATTR(name, fmt) \
234static ssize_t name##_show(struct device *dev, \
235 struct device_attribute *attr, \
236 char *buf) \
237{ \
238 return sysfs_emit(buf, fmt "\n", \
239 to_cache_info(dev)->cache_attrs.name); \
240} \
241static DEVICE_ATTR_RO(name);
242
243CACHE_ATTR(size, "%llu")
244CACHE_ATTR(line_size, "%u")
245CACHE_ATTR(indexing, "%u")
246CACHE_ATTR(write_policy, "%u")
247
248static struct attribute *cache_attrs[] = {
249 &dev_attr_indexing.attr,
250 &dev_attr_size.attr,
251 &dev_attr_line_size.attr,
252 &dev_attr_write_policy.attr,
253 NULL,
254};
255ATTRIBUTE_GROUPS(cache);
256
257static void node_cache_release(struct device *dev)
258{
259 kfree(dev);
260}
261
262static void node_cacheinfo_release(struct device *dev)
263{
264 struct node_cache_info *info = to_cache_info(dev);
265 kfree(info);
266}
267
268static void node_init_cache_dev(struct node *node)
269{
270 struct device *dev;
271
272 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
273 if (!dev)
274 return;
275
276 device_initialize(dev);
277 dev->parent = &node->dev;
278 dev->release = node_cache_release;
279 if (dev_set_name(dev, "memory_side_cache"))
280 goto put_device;
281
282 if (device_add(dev))
283 goto put_device;
284
285 pm_runtime_no_callbacks(dev);
286 node->cache_dev = dev;
287 return;
288put_device:
289 put_device(dev);
290}
291
292/**
293 * node_add_cache() - add cache attribute to a memory node
294 * @nid: Node identifier that has new cache attributes
295 * @cache_attrs: Attributes for the cache being added
296 */
297void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
298{
299 struct node_cache_info *info;
300 struct device *dev;
301 struct node *node;
302
303 if (!node_online(nid) || !node_devices[nid])
304 return;
305
306 node = node_devices[nid];
307 list_for_each_entry(info, &node->cache_attrs, node) {
308 if (info->cache_attrs.level == cache_attrs->level) {
309 dev_warn(&node->dev,
310 "attempt to add duplicate cache level:%d\n",
311 cache_attrs->level);
312 return;
313 }
314 }
315
316 if (!node->cache_dev)
317 node_init_cache_dev(node);
318 if (!node->cache_dev)
319 return;
320
321 info = kzalloc(sizeof(*info), GFP_KERNEL);
322 if (!info)
323 return;
324
325 dev = &info->dev;
326 device_initialize(dev);
327 dev->parent = node->cache_dev;
328 dev->release = node_cacheinfo_release;
329 dev->groups = cache_groups;
330 if (dev_set_name(dev, "index%d", cache_attrs->level))
331 goto put_device;
332
333 info->cache_attrs = *cache_attrs;
334 if (device_add(dev)) {
335 dev_warn(&node->dev, "failed to add cache level:%d\n",
336 cache_attrs->level);
337 goto put_device;
338 }
339 pm_runtime_no_callbacks(dev);
340 list_add_tail(&info->node, &node->cache_attrs);
341 return;
342put_device:
343 put_device(dev);
344}
345
346static void node_remove_caches(struct node *node)
347{
348 struct node_cache_info *info, *next;
349
350 if (!node->cache_dev)
351 return;
352
353 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
354 list_del(&info->node);
355 device_unregister(&info->dev);
356 }
357 device_unregister(node->cache_dev);
358}
359
360static void node_init_caches(unsigned int nid)
361{
362 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
363}
364#else
365static void node_init_caches(unsigned int nid) { }
366static void node_remove_caches(struct node *node) { }
367#endif
368
369#define K(x) ((x) << (PAGE_SHIFT - 10))
370static ssize_t node_read_meminfo(struct device *dev,
371 struct device_attribute *attr, char *buf)
372{
373 int len = 0;
374 int nid = dev->id;
375 struct pglist_data *pgdat = NODE_DATA(nid);
376 struct sysinfo i;
377 unsigned long sreclaimable, sunreclaimable;
378 unsigned long swapcached = 0;
379
380 si_meminfo_node(&i, nid);
381 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
382 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
383#ifdef CONFIG_SWAP
384 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
385#endif
386 len = sysfs_emit_at(buf, len,
387 "Node %d MemTotal: %8lu kB\n"
388 "Node %d MemFree: %8lu kB\n"
389 "Node %d MemUsed: %8lu kB\n"
390 "Node %d SwapCached: %8lu kB\n"
391 "Node %d Active: %8lu kB\n"
392 "Node %d Inactive: %8lu kB\n"
393 "Node %d Active(anon): %8lu kB\n"
394 "Node %d Inactive(anon): %8lu kB\n"
395 "Node %d Active(file): %8lu kB\n"
396 "Node %d Inactive(file): %8lu kB\n"
397 "Node %d Unevictable: %8lu kB\n"
398 "Node %d Mlocked: %8lu kB\n",
399 nid, K(i.totalram),
400 nid, K(i.freeram),
401 nid, K(i.totalram - i.freeram),
402 nid, K(swapcached),
403 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
404 node_page_state(pgdat, NR_ACTIVE_FILE)),
405 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
406 node_page_state(pgdat, NR_INACTIVE_FILE)),
407 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
408 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
409 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
410 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
411 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
412 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
413
414#ifdef CONFIG_HIGHMEM
415 len += sysfs_emit_at(buf, len,
416 "Node %d HighTotal: %8lu kB\n"
417 "Node %d HighFree: %8lu kB\n"
418 "Node %d LowTotal: %8lu kB\n"
419 "Node %d LowFree: %8lu kB\n",
420 nid, K(i.totalhigh),
421 nid, K(i.freehigh),
422 nid, K(i.totalram - i.totalhigh),
423 nid, K(i.freeram - i.freehigh));
424#endif
425 len += sysfs_emit_at(buf, len,
426 "Node %d Dirty: %8lu kB\n"
427 "Node %d Writeback: %8lu kB\n"
428 "Node %d FilePages: %8lu kB\n"
429 "Node %d Mapped: %8lu kB\n"
430 "Node %d AnonPages: %8lu kB\n"
431 "Node %d Shmem: %8lu kB\n"
432 "Node %d KernelStack: %8lu kB\n"
433#ifdef CONFIG_SHADOW_CALL_STACK
434 "Node %d ShadowCallStack:%8lu kB\n"
435#endif
436 "Node %d PageTables: %8lu kB\n"
437 "Node %d SecPageTables: %8lu kB\n"
438 "Node %d NFS_Unstable: %8lu kB\n"
439 "Node %d Bounce: %8lu kB\n"
440 "Node %d WritebackTmp: %8lu kB\n"
441 "Node %d KReclaimable: %8lu kB\n"
442 "Node %d Slab: %8lu kB\n"
443 "Node %d SReclaimable: %8lu kB\n"
444 "Node %d SUnreclaim: %8lu kB\n"
445#ifdef CONFIG_TRANSPARENT_HUGEPAGE
446 "Node %d AnonHugePages: %8lu kB\n"
447 "Node %d ShmemHugePages: %8lu kB\n"
448 "Node %d ShmemPmdMapped: %8lu kB\n"
449 "Node %d FileHugePages: %8lu kB\n"
450 "Node %d FilePmdMapped: %8lu kB\n"
451#endif
452 ,
453 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
454 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
455 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
456 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
457 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
458 nid, K(i.sharedram),
459 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
460#ifdef CONFIG_SHADOW_CALL_STACK
461 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
462#endif
463 nid, K(node_page_state(pgdat, NR_PAGETABLE)),
464 nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
465 nid, 0UL,
466 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
467 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
468 nid, K(sreclaimable +
469 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
470 nid, K(sreclaimable + sunreclaimable),
471 nid, K(sreclaimable),
472 nid, K(sunreclaimable)
473#ifdef CONFIG_TRANSPARENT_HUGEPAGE
474 ,
475 nid, K(node_page_state(pgdat, NR_ANON_THPS)),
476 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
477 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
478 nid, K(node_page_state(pgdat, NR_FILE_THPS)),
479 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
480#endif
481 );
482 len += hugetlb_report_node_meminfo(buf, len, nid);
483 return len;
484}
485
486#undef K
487static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
488
489static ssize_t node_read_numastat(struct device *dev,
490 struct device_attribute *attr, char *buf)
491{
492 fold_vm_numa_events();
493 return sysfs_emit(buf,
494 "numa_hit %lu\n"
495 "numa_miss %lu\n"
496 "numa_foreign %lu\n"
497 "interleave_hit %lu\n"
498 "local_node %lu\n"
499 "other_node %lu\n",
500 sum_zone_numa_event_state(dev->id, NUMA_HIT),
501 sum_zone_numa_event_state(dev->id, NUMA_MISS),
502 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
503 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
504 sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
505 sum_zone_numa_event_state(dev->id, NUMA_OTHER));
506}
507static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
508
509static ssize_t node_read_vmstat(struct device *dev,
510 struct device_attribute *attr, char *buf)
511{
512 int nid = dev->id;
513 struct pglist_data *pgdat = NODE_DATA(nid);
514 int i;
515 int len = 0;
516
517 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
518 len += sysfs_emit_at(buf, len, "%s %lu\n",
519 zone_stat_name(i),
520 sum_zone_node_page_state(nid, i));
521
522#ifdef CONFIG_NUMA
523 fold_vm_numa_events();
524 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
525 len += sysfs_emit_at(buf, len, "%s %lu\n",
526 numa_stat_name(i),
527 sum_zone_numa_event_state(nid, i));
528
529#endif
530 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
531 unsigned long pages = node_page_state_pages(pgdat, i);
532
533 if (vmstat_item_print_in_thp(i))
534 pages /= HPAGE_PMD_NR;
535 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
536 pages);
537 }
538
539 return len;
540}
541static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
542
543static ssize_t node_read_distance(struct device *dev,
544 struct device_attribute *attr, char *buf)
545{
546 int nid = dev->id;
547 int len = 0;
548 int i;
549
550 /*
551 * buf is currently PAGE_SIZE in length and each node needs 4 chars
552 * at the most (distance + space or newline).
553 */
554 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
555
556 for_each_online_node(i) {
557 len += sysfs_emit_at(buf, len, "%s%d",
558 i ? " " : "", node_distance(nid, i));
559 }
560
561 len += sysfs_emit_at(buf, len, "\n");
562 return len;
563}
564static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
565
566static struct attribute *node_dev_attrs[] = {
567 &dev_attr_meminfo.attr,
568 &dev_attr_numastat.attr,
569 &dev_attr_distance.attr,
570 &dev_attr_vmstat.attr,
571 NULL
572};
573
574static struct bin_attribute *node_dev_bin_attrs[] = {
575 &bin_attr_cpumap,
576 &bin_attr_cpulist,
577 NULL
578};
579
580static const struct attribute_group node_dev_group = {
581 .attrs = node_dev_attrs,
582 .bin_attrs = node_dev_bin_attrs
583};
584
585static const struct attribute_group *node_dev_groups[] = {
586 &node_dev_group,
587#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
588 &arch_node_dev_group,
589#endif
590 NULL
591};
592
593static void node_device_release(struct device *dev)
594{
595 kfree(to_node(dev));
596}
597
598/*
599 * register_node - Setup a sysfs device for a node.
600 * @num - Node number to use when creating the device.
601 *
602 * Initialize and register the node device.
603 */
604static int register_node(struct node *node, int num)
605{
606 int error;
607
608 node->dev.id = num;
609 node->dev.bus = &node_subsys;
610 node->dev.release = node_device_release;
611 node->dev.groups = node_dev_groups;
612 error = device_register(&node->dev);
613
614 if (error) {
615 put_device(&node->dev);
616 } else {
617 hugetlb_register_node(node);
618 compaction_register_node(node);
619 }
620
621 return error;
622}
623
624/**
625 * unregister_node - unregister a node device
626 * @node: node going away
627 *
628 * Unregisters a node device @node. All the devices on the node must be
629 * unregistered before calling this function.
630 */
631void unregister_node(struct node *node)
632{
633 hugetlb_unregister_node(node);
634 compaction_unregister_node(node);
635 node_remove_accesses(node);
636 node_remove_caches(node);
637 device_unregister(&node->dev);
638}
639
640struct node *node_devices[MAX_NUMNODES];
641
642/*
643 * register cpu under node
644 */
645int register_cpu_under_node(unsigned int cpu, unsigned int nid)
646{
647 int ret;
648 struct device *obj;
649
650 if (!node_online(nid))
651 return 0;
652
653 obj = get_cpu_device(cpu);
654 if (!obj)
655 return 0;
656
657 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
658 &obj->kobj,
659 kobject_name(&obj->kobj));
660 if (ret)
661 return ret;
662
663 return sysfs_create_link(&obj->kobj,
664 &node_devices[nid]->dev.kobj,
665 kobject_name(&node_devices[nid]->dev.kobj));
666}
667
668/**
669 * register_memory_node_under_compute_node - link memory node to its compute
670 * node for a given access class.
671 * @mem_nid: Memory node number
672 * @cpu_nid: Cpu node number
673 * @access: Access class to register
674 *
675 * Description:
676 * For use with platforms that may have separate memory and compute nodes.
677 * This function will export node relationships linking which memory
678 * initiator nodes can access memory targets at a given ranked access
679 * class.
680 */
681int register_memory_node_under_compute_node(unsigned int mem_nid,
682 unsigned int cpu_nid,
683 unsigned int access)
684{
685 struct node *init_node, *targ_node;
686 struct node_access_nodes *initiator, *target;
687 int ret;
688
689 if (!node_online(cpu_nid) || !node_online(mem_nid))
690 return -ENODEV;
691
692 init_node = node_devices[cpu_nid];
693 targ_node = node_devices[mem_nid];
694 initiator = node_init_node_access(init_node, access);
695 target = node_init_node_access(targ_node, access);
696 if (!initiator || !target)
697 return -ENOMEM;
698
699 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
700 &targ_node->dev.kobj,
701 dev_name(&targ_node->dev));
702 if (ret)
703 return ret;
704
705 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
706 &init_node->dev.kobj,
707 dev_name(&init_node->dev));
708 if (ret)
709 goto err;
710
711 return 0;
712 err:
713 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
714 dev_name(&targ_node->dev));
715 return ret;
716}
717
718int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
719{
720 struct device *obj;
721
722 if (!node_online(nid))
723 return 0;
724
725 obj = get_cpu_device(cpu);
726 if (!obj)
727 return 0;
728
729 sysfs_remove_link(&node_devices[nid]->dev.kobj,
730 kobject_name(&obj->kobj));
731 sysfs_remove_link(&obj->kobj,
732 kobject_name(&node_devices[nid]->dev.kobj));
733
734 return 0;
735}
736
737#ifdef CONFIG_MEMORY_HOTPLUG
738static int __ref get_nid_for_pfn(unsigned long pfn)
739{
740#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
741 if (system_state < SYSTEM_RUNNING)
742 return early_pfn_to_nid(pfn);
743#endif
744 return pfn_to_nid(pfn);
745}
746
747static void do_register_memory_block_under_node(int nid,
748 struct memory_block *mem_blk,
749 enum meminit_context context)
750{
751 int ret;
752
753 memory_block_add_nid(mem_blk, nid, context);
754
755 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
756 &mem_blk->dev.kobj,
757 kobject_name(&mem_blk->dev.kobj));
758 if (ret && ret != -EEXIST)
759 dev_err_ratelimited(&node_devices[nid]->dev,
760 "can't create link to %s in sysfs (%d)\n",
761 kobject_name(&mem_blk->dev.kobj), ret);
762
763 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
764 &node_devices[nid]->dev.kobj,
765 kobject_name(&node_devices[nid]->dev.kobj));
766 if (ret && ret != -EEXIST)
767 dev_err_ratelimited(&mem_blk->dev,
768 "can't create link to %s in sysfs (%d)\n",
769 kobject_name(&node_devices[nid]->dev.kobj),
770 ret);
771}
772
773/* register memory section under specified node if it spans that node */
774static int register_mem_block_under_node_early(struct memory_block *mem_blk,
775 void *arg)
776{
777 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
778 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
779 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
780 int nid = *(int *)arg;
781 unsigned long pfn;
782
783 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
784 int page_nid;
785
786 /*
787 * memory block could have several absent sections from start.
788 * skip pfn range from absent section
789 */
790 if (!pfn_in_present_section(pfn)) {
791 pfn = round_down(pfn + PAGES_PER_SECTION,
792 PAGES_PER_SECTION) - 1;
793 continue;
794 }
795
796 /*
797 * We need to check if page belongs to nid only at the boot
798 * case because node's ranges can be interleaved.
799 */
800 page_nid = get_nid_for_pfn(pfn);
801 if (page_nid < 0)
802 continue;
803 if (page_nid != nid)
804 continue;
805
806 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_EARLY);
807 return 0;
808 }
809 /* mem section does not span the specified node */
810 return 0;
811}
812
813/*
814 * During hotplug we know that all pages in the memory block belong to the same
815 * node.
816 */
817static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
818 void *arg)
819{
820 int nid = *(int *)arg;
821
822 do_register_memory_block_under_node(nid, mem_blk, MEMINIT_HOTPLUG);
823 return 0;
824}
825
826/*
827 * Unregister a memory block device under the node it spans. Memory blocks
828 * with multiple nodes cannot be offlined and therefore also never be removed.
829 */
830void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
831{
832 if (mem_blk->nid == NUMA_NO_NODE)
833 return;
834
835 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
836 kobject_name(&mem_blk->dev.kobj));
837 sysfs_remove_link(&mem_blk->dev.kobj,
838 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
839}
840
841void register_memory_blocks_under_node(int nid, unsigned long start_pfn,
842 unsigned long end_pfn,
843 enum meminit_context context)
844{
845 walk_memory_blocks_func_t func;
846
847 if (context == MEMINIT_HOTPLUG)
848 func = register_mem_block_under_node_hotplug;
849 else
850 func = register_mem_block_under_node_early;
851
852 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
853 (void *)&nid, func);
854 return;
855}
856#endif /* CONFIG_MEMORY_HOTPLUG */
857
858int __register_one_node(int nid)
859{
860 int error;
861 int cpu;
862
863 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
864 if (!node_devices[nid])
865 return -ENOMEM;
866
867 error = register_node(node_devices[nid], nid);
868
869 /* link cpu under this node */
870 for_each_present_cpu(cpu) {
871 if (cpu_to_node(cpu) == nid)
872 register_cpu_under_node(cpu, nid);
873 }
874
875 INIT_LIST_HEAD(&node_devices[nid]->access_list);
876 node_init_caches(nid);
877
878 return error;
879}
880
881void unregister_one_node(int nid)
882{
883 if (!node_devices[nid])
884 return;
885
886 unregister_node(node_devices[nid]);
887 node_devices[nid] = NULL;
888}
889
890/*
891 * node states attributes
892 */
893
894struct node_attr {
895 struct device_attribute attr;
896 enum node_states state;
897};
898
899static ssize_t show_node_state(struct device *dev,
900 struct device_attribute *attr, char *buf)
901{
902 struct node_attr *na = container_of(attr, struct node_attr, attr);
903
904 return sysfs_emit(buf, "%*pbl\n",
905 nodemask_pr_args(&node_states[na->state]));
906}
907
908#define _NODE_ATTR(name, state) \
909 { __ATTR(name, 0444, show_node_state, NULL), state }
910
911static struct node_attr node_state_attr[] = {
912 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
913 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
914 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
915#ifdef CONFIG_HIGHMEM
916 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
917#endif
918 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
919 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
920 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
921 N_GENERIC_INITIATOR),
922};
923
924static struct attribute *node_state_attrs[] = {
925 &node_state_attr[N_POSSIBLE].attr.attr,
926 &node_state_attr[N_ONLINE].attr.attr,
927 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
928#ifdef CONFIG_HIGHMEM
929 &node_state_attr[N_HIGH_MEMORY].attr.attr,
930#endif
931 &node_state_attr[N_MEMORY].attr.attr,
932 &node_state_attr[N_CPU].attr.attr,
933 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
934 NULL
935};
936
937static const struct attribute_group memory_root_attr_group = {
938 .attrs = node_state_attrs,
939};
940
941static const struct attribute_group *cpu_root_attr_groups[] = {
942 &memory_root_attr_group,
943 NULL,
944};
945
946void __init node_dev_init(void)
947{
948 int ret, i;
949
950 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
951 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
952
953 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
954 if (ret)
955 panic("%s() failed to register subsystem: %d\n", __func__, ret);
956
957 /*
958 * Create all node devices, which will properly link the node
959 * to applicable memory block devices and already created cpu devices.
960 */
961 for_each_online_node(i) {
962 ret = register_one_node(i);
963 if (ret)
964 panic("%s() failed to add node: %d\n", __func__, ret);
965 }
966}