Loading...
1/*
2 * Basic Node interface support
3 */
4
5#include <linux/module.h>
6#include <linux/init.h>
7#include <linux/mm.h>
8#include <linux/memory.h>
9#include <linux/vmstat.h>
10#include <linux/notifier.h>
11#include <linux/node.h>
12#include <linux/hugetlb.h>
13#include <linux/compaction.h>
14#include <linux/cpumask.h>
15#include <linux/topology.h>
16#include <linux/nodemask.h>
17#include <linux/cpu.h>
18#include <linux/device.h>
19#include <linux/swap.h>
20#include <linux/slab.h>
21
22static struct bus_type node_subsys = {
23 .name = "node",
24 .dev_name = "node",
25};
26
27
28static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
29{
30 struct node *node_dev = to_node(dev);
31 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
32
33 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
34 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
35
36 return cpumap_print_to_pagebuf(list, buf, mask);
37}
38
39static inline ssize_t node_read_cpumask(struct device *dev,
40 struct device_attribute *attr, char *buf)
41{
42 return node_read_cpumap(dev, false, buf);
43}
44static inline ssize_t node_read_cpulist(struct device *dev,
45 struct device_attribute *attr, char *buf)
46{
47 return node_read_cpumap(dev, true, buf);
48}
49
50static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
51static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
52
53#define K(x) ((x) << (PAGE_SHIFT - 10))
54static ssize_t node_read_meminfo(struct device *dev,
55 struct device_attribute *attr, char *buf)
56{
57 int n;
58 int nid = dev->id;
59 struct 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);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Basic Node interface support
4 */
5
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/mm.h>
9#include <linux/memory.h>
10#include <linux/vmstat.h>
11#include <linux/notifier.h>
12#include <linux/node.h>
13#include <linux/hugetlb.h>
14#include <linux/compaction.h>
15#include <linux/cpumask.h>
16#include <linux/topology.h>
17#include <linux/nodemask.h>
18#include <linux/cpu.h>
19#include <linux/device.h>
20#include <linux/pm_runtime.h>
21#include <linux/swap.h>
22#include <linux/slab.h>
23
24static struct bus_type node_subsys = {
25 .name = "node",
26 .dev_name = "node",
27};
28
29
30static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
31{
32 ssize_t n;
33 cpumask_var_t mask;
34 struct node *node_dev = to_node(dev);
35
36 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
37 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
38
39 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
40 return 0;
41
42 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
43 n = cpumap_print_to_pagebuf(list, buf, mask);
44 free_cpumask_var(mask);
45
46 return n;
47}
48
49static inline ssize_t node_read_cpumask(struct device *dev,
50 struct device_attribute *attr, char *buf)
51{
52 return node_read_cpumap(dev, false, buf);
53}
54static inline ssize_t node_read_cpulist(struct device *dev,
55 struct device_attribute *attr, char *buf)
56{
57 return node_read_cpumap(dev, true, buf);
58}
59
60static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
61static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
62
63/**
64 * struct node_access_nodes - Access class device to hold user visible
65 * relationships to other nodes.
66 * @dev: Device for this memory access class
67 * @list_node: List element in the node's access list
68 * @access: The access class rank
69 * @hmem_attrs: Heterogeneous memory performance attributes
70 */
71struct node_access_nodes {
72 struct device dev;
73 struct list_head list_node;
74 unsigned access;
75#ifdef CONFIG_HMEM_REPORTING
76 struct node_hmem_attrs hmem_attrs;
77#endif
78};
79#define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
80
81static struct attribute *node_init_access_node_attrs[] = {
82 NULL,
83};
84
85static struct attribute *node_targ_access_node_attrs[] = {
86 NULL,
87};
88
89static const struct attribute_group initiators = {
90 .name = "initiators",
91 .attrs = node_init_access_node_attrs,
92};
93
94static const struct attribute_group targets = {
95 .name = "targets",
96 .attrs = node_targ_access_node_attrs,
97};
98
99static const struct attribute_group *node_access_node_groups[] = {
100 &initiators,
101 &targets,
102 NULL,
103};
104
105static void node_remove_accesses(struct node *node)
106{
107 struct node_access_nodes *c, *cnext;
108
109 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
110 list_del(&c->list_node);
111 device_unregister(&c->dev);
112 }
113}
114
115static void node_access_release(struct device *dev)
116{
117 kfree(to_access_nodes(dev));
118}
119
120static struct node_access_nodes *node_init_node_access(struct node *node,
121 unsigned access)
122{
123 struct node_access_nodes *access_node;
124 struct device *dev;
125
126 list_for_each_entry(access_node, &node->access_list, list_node)
127 if (access_node->access == access)
128 return access_node;
129
130 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
131 if (!access_node)
132 return NULL;
133
134 access_node->access = access;
135 dev = &access_node->dev;
136 dev->parent = &node->dev;
137 dev->release = node_access_release;
138 dev->groups = node_access_node_groups;
139 if (dev_set_name(dev, "access%u", access))
140 goto free;
141
142 if (device_register(dev))
143 goto free_name;
144
145 pm_runtime_no_callbacks(dev);
146 list_add_tail(&access_node->list_node, &node->access_list);
147 return access_node;
148free_name:
149 kfree_const(dev->kobj.name);
150free:
151 kfree(access_node);
152 return NULL;
153}
154
155#ifdef CONFIG_HMEM_REPORTING
156#define ACCESS_ATTR(name) \
157static ssize_t name##_show(struct device *dev, \
158 struct device_attribute *attr, \
159 char *buf) \
160{ \
161 return sprintf(buf, "%u\n", to_access_nodes(dev)->hmem_attrs.name); \
162} \
163static DEVICE_ATTR_RO(name);
164
165ACCESS_ATTR(read_bandwidth)
166ACCESS_ATTR(read_latency)
167ACCESS_ATTR(write_bandwidth)
168ACCESS_ATTR(write_latency)
169
170static struct attribute *access_attrs[] = {
171 &dev_attr_read_bandwidth.attr,
172 &dev_attr_read_latency.attr,
173 &dev_attr_write_bandwidth.attr,
174 &dev_attr_write_latency.attr,
175 NULL,
176};
177
178/**
179 * node_set_perf_attrs - Set the performance values for given access class
180 * @nid: Node identifier to be set
181 * @hmem_attrs: Heterogeneous memory performance attributes
182 * @access: The access class the for the given attributes
183 */
184void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
185 unsigned access)
186{
187 struct node_access_nodes *c;
188 struct node *node;
189 int i;
190
191 if (WARN_ON_ONCE(!node_online(nid)))
192 return;
193
194 node = node_devices[nid];
195 c = node_init_node_access(node, access);
196 if (!c)
197 return;
198
199 c->hmem_attrs = *hmem_attrs;
200 for (i = 0; access_attrs[i] != NULL; i++) {
201 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
202 "initiators")) {
203 pr_info("failed to add performance attribute to node %d\n",
204 nid);
205 break;
206 }
207 }
208}
209
210/**
211 * struct node_cache_info - Internal tracking for memory node caches
212 * @dev: Device represeting the cache level
213 * @node: List element for tracking in the node
214 * @cache_attrs:Attributes for this cache level
215 */
216struct node_cache_info {
217 struct device dev;
218 struct list_head node;
219 struct node_cache_attrs cache_attrs;
220};
221#define to_cache_info(device) container_of(device, struct node_cache_info, dev)
222
223#define CACHE_ATTR(name, fmt) \
224static ssize_t name##_show(struct device *dev, \
225 struct device_attribute *attr, \
226 char *buf) \
227{ \
228 return sprintf(buf, fmt "\n", to_cache_info(dev)->cache_attrs.name);\
229} \
230DEVICE_ATTR_RO(name);
231
232CACHE_ATTR(size, "%llu")
233CACHE_ATTR(line_size, "%u")
234CACHE_ATTR(indexing, "%u")
235CACHE_ATTR(write_policy, "%u")
236
237static struct attribute *cache_attrs[] = {
238 &dev_attr_indexing.attr,
239 &dev_attr_size.attr,
240 &dev_attr_line_size.attr,
241 &dev_attr_write_policy.attr,
242 NULL,
243};
244ATTRIBUTE_GROUPS(cache);
245
246static void node_cache_release(struct device *dev)
247{
248 kfree(dev);
249}
250
251static void node_cacheinfo_release(struct device *dev)
252{
253 struct node_cache_info *info = to_cache_info(dev);
254 kfree(info);
255}
256
257static void node_init_cache_dev(struct node *node)
258{
259 struct device *dev;
260
261 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
262 if (!dev)
263 return;
264
265 dev->parent = &node->dev;
266 dev->release = node_cache_release;
267 if (dev_set_name(dev, "memory_side_cache"))
268 goto free_dev;
269
270 if (device_register(dev))
271 goto free_name;
272
273 pm_runtime_no_callbacks(dev);
274 node->cache_dev = dev;
275 return;
276free_name:
277 kfree_const(dev->kobj.name);
278free_dev:
279 kfree(dev);
280}
281
282/**
283 * node_add_cache() - add cache attribute to a memory node
284 * @nid: Node identifier that has new cache attributes
285 * @cache_attrs: Attributes for the cache being added
286 */
287void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
288{
289 struct node_cache_info *info;
290 struct device *dev;
291 struct node *node;
292
293 if (!node_online(nid) || !node_devices[nid])
294 return;
295
296 node = node_devices[nid];
297 list_for_each_entry(info, &node->cache_attrs, node) {
298 if (info->cache_attrs.level == cache_attrs->level) {
299 dev_warn(&node->dev,
300 "attempt to add duplicate cache level:%d\n",
301 cache_attrs->level);
302 return;
303 }
304 }
305
306 if (!node->cache_dev)
307 node_init_cache_dev(node);
308 if (!node->cache_dev)
309 return;
310
311 info = kzalloc(sizeof(*info), GFP_KERNEL);
312 if (!info)
313 return;
314
315 dev = &info->dev;
316 dev->parent = node->cache_dev;
317 dev->release = node_cacheinfo_release;
318 dev->groups = cache_groups;
319 if (dev_set_name(dev, "index%d", cache_attrs->level))
320 goto free_cache;
321
322 info->cache_attrs = *cache_attrs;
323 if (device_register(dev)) {
324 dev_warn(&node->dev, "failed to add cache level:%d\n",
325 cache_attrs->level);
326 goto free_name;
327 }
328 pm_runtime_no_callbacks(dev);
329 list_add_tail(&info->node, &node->cache_attrs);
330 return;
331free_name:
332 kfree_const(dev->kobj.name);
333free_cache:
334 kfree(info);
335}
336
337static void node_remove_caches(struct node *node)
338{
339 struct node_cache_info *info, *next;
340
341 if (!node->cache_dev)
342 return;
343
344 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
345 list_del(&info->node);
346 device_unregister(&info->dev);
347 }
348 device_unregister(node->cache_dev);
349}
350
351static void node_init_caches(unsigned int nid)
352{
353 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
354}
355#else
356static void node_init_caches(unsigned int nid) { }
357static void node_remove_caches(struct node *node) { }
358#endif
359
360#define K(x) ((x) << (PAGE_SHIFT - 10))
361static ssize_t node_read_meminfo(struct device *dev,
362 struct device_attribute *attr, char *buf)
363{
364 int n;
365 int nid = dev->id;
366 struct pglist_data *pgdat = NODE_DATA(nid);
367 struct sysinfo i;
368 unsigned long sreclaimable, sunreclaimable;
369
370 si_meminfo_node(&i, nid);
371 sreclaimable = node_page_state(pgdat, NR_SLAB_RECLAIMABLE);
372 sunreclaimable = node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE);
373 n = sprintf(buf,
374 "Node %d MemTotal: %8lu kB\n"
375 "Node %d MemFree: %8lu kB\n"
376 "Node %d MemUsed: %8lu kB\n"
377 "Node %d Active: %8lu kB\n"
378 "Node %d Inactive: %8lu kB\n"
379 "Node %d Active(anon): %8lu kB\n"
380 "Node %d Inactive(anon): %8lu kB\n"
381 "Node %d Active(file): %8lu kB\n"
382 "Node %d Inactive(file): %8lu kB\n"
383 "Node %d Unevictable: %8lu kB\n"
384 "Node %d Mlocked: %8lu kB\n",
385 nid, K(i.totalram),
386 nid, K(i.freeram),
387 nid, K(i.totalram - i.freeram),
388 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
389 node_page_state(pgdat, NR_ACTIVE_FILE)),
390 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
391 node_page_state(pgdat, NR_INACTIVE_FILE)),
392 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
393 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
394 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
395 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
396 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
397 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
398
399#ifdef CONFIG_HIGHMEM
400 n += sprintf(buf + n,
401 "Node %d HighTotal: %8lu kB\n"
402 "Node %d HighFree: %8lu kB\n"
403 "Node %d LowTotal: %8lu kB\n"
404 "Node %d LowFree: %8lu kB\n",
405 nid, K(i.totalhigh),
406 nid, K(i.freehigh),
407 nid, K(i.totalram - i.totalhigh),
408 nid, K(i.freeram - i.freehigh));
409#endif
410 n += sprintf(buf + n,
411 "Node %d Dirty: %8lu kB\n"
412 "Node %d Writeback: %8lu kB\n"
413 "Node %d FilePages: %8lu kB\n"
414 "Node %d Mapped: %8lu kB\n"
415 "Node %d AnonPages: %8lu kB\n"
416 "Node %d Shmem: %8lu kB\n"
417 "Node %d KernelStack: %8lu kB\n"
418 "Node %d PageTables: %8lu kB\n"
419 "Node %d NFS_Unstable: %8lu kB\n"
420 "Node %d Bounce: %8lu kB\n"
421 "Node %d WritebackTmp: %8lu kB\n"
422 "Node %d KReclaimable: %8lu kB\n"
423 "Node %d Slab: %8lu kB\n"
424 "Node %d SReclaimable: %8lu kB\n"
425 "Node %d SUnreclaim: %8lu kB\n"
426#ifdef CONFIG_TRANSPARENT_HUGEPAGE
427 "Node %d AnonHugePages: %8lu kB\n"
428 "Node %d ShmemHugePages: %8lu kB\n"
429 "Node %d ShmemPmdMapped: %8lu kB\n"
430 "Node %d FileHugePages: %8lu kB\n"
431 "Node %d FilePmdMapped: %8lu kB\n"
432#endif
433 ,
434 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
435 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
436 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
437 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
438 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
439 nid, K(i.sharedram),
440 nid, sum_zone_node_page_state(nid, NR_KERNEL_STACK_KB),
441 nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
442 nid, K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
443 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
444 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
445 nid, K(sreclaimable +
446 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
447 nid, K(sreclaimable + sunreclaimable),
448 nid, K(sreclaimable),
449 nid, K(sunreclaimable)
450#ifdef CONFIG_TRANSPARENT_HUGEPAGE
451 ,
452 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
453 HPAGE_PMD_NR),
454 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
455 HPAGE_PMD_NR),
456 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
457 HPAGE_PMD_NR),
458 nid, K(node_page_state(pgdat, NR_FILE_THPS) *
459 HPAGE_PMD_NR),
460 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED) *
461 HPAGE_PMD_NR)
462#endif
463 );
464 n += hugetlb_report_node_meminfo(nid, buf + n);
465 return n;
466}
467
468#undef K
469static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
470
471static ssize_t node_read_numastat(struct device *dev,
472 struct device_attribute *attr, char *buf)
473{
474 return sprintf(buf,
475 "numa_hit %lu\n"
476 "numa_miss %lu\n"
477 "numa_foreign %lu\n"
478 "interleave_hit %lu\n"
479 "local_node %lu\n"
480 "other_node %lu\n",
481 sum_zone_numa_state(dev->id, NUMA_HIT),
482 sum_zone_numa_state(dev->id, NUMA_MISS),
483 sum_zone_numa_state(dev->id, NUMA_FOREIGN),
484 sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
485 sum_zone_numa_state(dev->id, NUMA_LOCAL),
486 sum_zone_numa_state(dev->id, NUMA_OTHER));
487}
488static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
489
490static ssize_t node_read_vmstat(struct device *dev,
491 struct device_attribute *attr, char *buf)
492{
493 int nid = dev->id;
494 struct pglist_data *pgdat = NODE_DATA(nid);
495 int i;
496 int n = 0;
497
498 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
499 n += sprintf(buf+n, "%s %lu\n", vmstat_text[i],
500 sum_zone_node_page_state(nid, i));
501
502#ifdef CONFIG_NUMA
503 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
504 n += sprintf(buf+n, "%s %lu\n",
505 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
506 sum_zone_numa_state(nid, i));
507#endif
508
509 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
510 n += sprintf(buf+n, "%s %lu\n",
511 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
512 NR_VM_NUMA_STAT_ITEMS],
513 node_page_state(pgdat, i));
514
515 return n;
516}
517static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
518
519static ssize_t node_read_distance(struct device *dev,
520 struct device_attribute *attr, char *buf)
521{
522 int nid = dev->id;
523 int len = 0;
524 int i;
525
526 /*
527 * buf is currently PAGE_SIZE in length and each node needs 4 chars
528 * at the most (distance + space or newline).
529 */
530 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
531
532 for_each_online_node(i)
533 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
534
535 len += sprintf(buf + len, "\n");
536 return len;
537}
538static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
539
540static struct attribute *node_dev_attrs[] = {
541 &dev_attr_cpumap.attr,
542 &dev_attr_cpulist.attr,
543 &dev_attr_meminfo.attr,
544 &dev_attr_numastat.attr,
545 &dev_attr_distance.attr,
546 &dev_attr_vmstat.attr,
547 NULL
548};
549ATTRIBUTE_GROUPS(node_dev);
550
551#ifdef CONFIG_HUGETLBFS
552/*
553 * hugetlbfs per node attributes registration interface:
554 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
555 * it will register its per node attributes for all online nodes with
556 * memory. It will also call register_hugetlbfs_with_node(), below, to
557 * register its attribute registration functions with this node driver.
558 * Once these hooks have been initialized, the node driver will call into
559 * the hugetlb module to [un]register attributes for hot-plugged nodes.
560 */
561static node_registration_func_t __hugetlb_register_node;
562static node_registration_func_t __hugetlb_unregister_node;
563
564static inline bool hugetlb_register_node(struct node *node)
565{
566 if (__hugetlb_register_node &&
567 node_state(node->dev.id, N_MEMORY)) {
568 __hugetlb_register_node(node);
569 return true;
570 }
571 return false;
572}
573
574static inline void hugetlb_unregister_node(struct node *node)
575{
576 if (__hugetlb_unregister_node)
577 __hugetlb_unregister_node(node);
578}
579
580void register_hugetlbfs_with_node(node_registration_func_t doregister,
581 node_registration_func_t unregister)
582{
583 __hugetlb_register_node = doregister;
584 __hugetlb_unregister_node = unregister;
585}
586#else
587static inline void hugetlb_register_node(struct node *node) {}
588
589static inline void hugetlb_unregister_node(struct node *node) {}
590#endif
591
592static void node_device_release(struct device *dev)
593{
594 struct node *node = to_node(dev);
595
596#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
597 /*
598 * We schedule the work only when a memory section is
599 * onlined/offlined on this node. When we come here,
600 * all the memory on this node has been offlined,
601 * so we won't enqueue new work to this work.
602 *
603 * The work is using node->node_work, so we should
604 * flush work before freeing the memory.
605 */
606 flush_work(&node->node_work);
607#endif
608 kfree(node);
609}
610
611/*
612 * register_node - Setup a sysfs device for a node.
613 * @num - Node number to use when creating the device.
614 *
615 * Initialize and register the node device.
616 */
617static int register_node(struct node *node, int num)
618{
619 int error;
620
621 node->dev.id = num;
622 node->dev.bus = &node_subsys;
623 node->dev.release = node_device_release;
624 node->dev.groups = node_dev_groups;
625 error = device_register(&node->dev);
626
627 if (error)
628 put_device(&node->dev);
629 else {
630 hugetlb_register_node(node);
631
632 compaction_register_node(node);
633 }
634 return error;
635}
636
637/**
638 * unregister_node - unregister a node device
639 * @node: node going away
640 *
641 * Unregisters a node device @node. All the devices on the node must be
642 * unregistered before calling this function.
643 */
644void unregister_node(struct node *node)
645{
646 hugetlb_unregister_node(node); /* no-op, if memoryless node */
647 node_remove_accesses(node);
648 node_remove_caches(node);
649 device_unregister(&node->dev);
650}
651
652struct node *node_devices[MAX_NUMNODES];
653
654/*
655 * register cpu under node
656 */
657int register_cpu_under_node(unsigned int cpu, unsigned int nid)
658{
659 int ret;
660 struct device *obj;
661
662 if (!node_online(nid))
663 return 0;
664
665 obj = get_cpu_device(cpu);
666 if (!obj)
667 return 0;
668
669 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
670 &obj->kobj,
671 kobject_name(&obj->kobj));
672 if (ret)
673 return ret;
674
675 return sysfs_create_link(&obj->kobj,
676 &node_devices[nid]->dev.kobj,
677 kobject_name(&node_devices[nid]->dev.kobj));
678}
679
680/**
681 * register_memory_node_under_compute_node - link memory node to its compute
682 * node for a given access class.
683 * @mem_nid: Memory node number
684 * @cpu_nid: Cpu node number
685 * @access: Access class to register
686 *
687 * Description:
688 * For use with platforms that may have separate memory and compute nodes.
689 * This function will export node relationships linking which memory
690 * initiator nodes can access memory targets at a given ranked access
691 * class.
692 */
693int register_memory_node_under_compute_node(unsigned int mem_nid,
694 unsigned int cpu_nid,
695 unsigned access)
696{
697 struct node *init_node, *targ_node;
698 struct node_access_nodes *initiator, *target;
699 int ret;
700
701 if (!node_online(cpu_nid) || !node_online(mem_nid))
702 return -ENODEV;
703
704 init_node = node_devices[cpu_nid];
705 targ_node = node_devices[mem_nid];
706 initiator = node_init_node_access(init_node, access);
707 target = node_init_node_access(targ_node, access);
708 if (!initiator || !target)
709 return -ENOMEM;
710
711 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
712 &targ_node->dev.kobj,
713 dev_name(&targ_node->dev));
714 if (ret)
715 return ret;
716
717 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
718 &init_node->dev.kobj,
719 dev_name(&init_node->dev));
720 if (ret)
721 goto err;
722
723 return 0;
724 err:
725 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
726 dev_name(&targ_node->dev));
727 return ret;
728}
729
730int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
731{
732 struct device *obj;
733
734 if (!node_online(nid))
735 return 0;
736
737 obj = get_cpu_device(cpu);
738 if (!obj)
739 return 0;
740
741 sysfs_remove_link(&node_devices[nid]->dev.kobj,
742 kobject_name(&obj->kobj));
743 sysfs_remove_link(&obj->kobj,
744 kobject_name(&node_devices[nid]->dev.kobj));
745
746 return 0;
747}
748
749#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
750static int __ref get_nid_for_pfn(unsigned long pfn)
751{
752 if (!pfn_valid_within(pfn))
753 return -1;
754#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
755 if (system_state < SYSTEM_RUNNING)
756 return early_pfn_to_nid(pfn);
757#endif
758 return pfn_to_nid(pfn);
759}
760
761/* register memory section under specified node if it spans that node */
762static int register_mem_sect_under_node(struct memory_block *mem_blk,
763 void *arg)
764{
765 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
766 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
767 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
768 int ret, nid = *(int *)arg;
769 unsigned long pfn;
770
771 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
772 int page_nid;
773
774 /*
775 * memory block could have several absent sections from start.
776 * skip pfn range from absent section
777 */
778 if (!pfn_present(pfn)) {
779 pfn = round_down(pfn + PAGES_PER_SECTION,
780 PAGES_PER_SECTION) - 1;
781 continue;
782 }
783
784 /*
785 * We need to check if page belongs to nid only for the boot
786 * case, during hotplug we know that all pages in the memory
787 * block belong to the same node.
788 */
789 if (system_state == SYSTEM_BOOTING) {
790 page_nid = get_nid_for_pfn(pfn);
791 if (page_nid < 0)
792 continue;
793 if (page_nid != nid)
794 continue;
795 }
796
797 /*
798 * If this memory block spans multiple nodes, we only indicate
799 * the last processed node.
800 */
801 mem_blk->nid = nid;
802
803 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
804 &mem_blk->dev.kobj,
805 kobject_name(&mem_blk->dev.kobj));
806 if (ret)
807 return ret;
808
809 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
810 &node_devices[nid]->dev.kobj,
811 kobject_name(&node_devices[nid]->dev.kobj));
812 }
813 /* mem section does not span the specified node */
814 return 0;
815}
816
817/*
818 * Unregister a memory block device under the node it spans. Memory blocks
819 * with multiple nodes cannot be offlined and therefore also never be removed.
820 */
821void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
822{
823 if (mem_blk->nid == NUMA_NO_NODE)
824 return;
825
826 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
827 kobject_name(&mem_blk->dev.kobj));
828 sysfs_remove_link(&mem_blk->dev.kobj,
829 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
830}
831
832int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn)
833{
834 return walk_memory_blocks(PFN_PHYS(start_pfn),
835 PFN_PHYS(end_pfn - start_pfn), (void *)&nid,
836 register_mem_sect_under_node);
837}
838
839#ifdef CONFIG_HUGETLBFS
840/*
841 * Handle per node hstate attribute [un]registration on transistions
842 * to/from memoryless state.
843 */
844static void node_hugetlb_work(struct work_struct *work)
845{
846 struct node *node = container_of(work, struct node, node_work);
847
848 /*
849 * We only get here when a node transitions to/from memoryless state.
850 * We can detect which transition occurred by examining whether the
851 * node has memory now. hugetlb_register_node() already check this
852 * so we try to register the attributes. If that fails, then the
853 * node has transitioned to memoryless, try to unregister the
854 * attributes.
855 */
856 if (!hugetlb_register_node(node))
857 hugetlb_unregister_node(node);
858}
859
860static void init_node_hugetlb_work(int nid)
861{
862 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
863}
864
865static int node_memory_callback(struct notifier_block *self,
866 unsigned long action, void *arg)
867{
868 struct memory_notify *mnb = arg;
869 int nid = mnb->status_change_nid;
870
871 switch (action) {
872 case MEM_ONLINE:
873 case MEM_OFFLINE:
874 /*
875 * offload per node hstate [un]registration to a work thread
876 * when transitioning to/from memoryless state.
877 */
878 if (nid != NUMA_NO_NODE)
879 schedule_work(&node_devices[nid]->node_work);
880 break;
881
882 case MEM_GOING_ONLINE:
883 case MEM_GOING_OFFLINE:
884 case MEM_CANCEL_ONLINE:
885 case MEM_CANCEL_OFFLINE:
886 default:
887 break;
888 }
889
890 return NOTIFY_OK;
891}
892#endif /* CONFIG_HUGETLBFS */
893#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
894
895#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
896 !defined(CONFIG_HUGETLBFS)
897static inline int node_memory_callback(struct notifier_block *self,
898 unsigned long action, void *arg)
899{
900 return NOTIFY_OK;
901}
902
903static void init_node_hugetlb_work(int nid) { }
904
905#endif
906
907int __register_one_node(int nid)
908{
909 int error;
910 int cpu;
911
912 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
913 if (!node_devices[nid])
914 return -ENOMEM;
915
916 error = register_node(node_devices[nid], nid);
917
918 /* link cpu under this node */
919 for_each_present_cpu(cpu) {
920 if (cpu_to_node(cpu) == nid)
921 register_cpu_under_node(cpu, nid);
922 }
923
924 INIT_LIST_HEAD(&node_devices[nid]->access_list);
925 /* initialize work queue for memory hot plug */
926 init_node_hugetlb_work(nid);
927 node_init_caches(nid);
928
929 return error;
930}
931
932void unregister_one_node(int nid)
933{
934 if (!node_devices[nid])
935 return;
936
937 unregister_node(node_devices[nid]);
938 node_devices[nid] = NULL;
939}
940
941/*
942 * node states attributes
943 */
944
945static ssize_t print_nodes_state(enum node_states state, char *buf)
946{
947 int n;
948
949 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
950 nodemask_pr_args(&node_states[state]));
951 buf[n++] = '\n';
952 buf[n] = '\0';
953 return n;
954}
955
956struct node_attr {
957 struct device_attribute attr;
958 enum node_states state;
959};
960
961static ssize_t show_node_state(struct device *dev,
962 struct device_attribute *attr, char *buf)
963{
964 struct node_attr *na = container_of(attr, struct node_attr, attr);
965 return print_nodes_state(na->state, buf);
966}
967
968#define _NODE_ATTR(name, state) \
969 { __ATTR(name, 0444, show_node_state, NULL), state }
970
971static struct node_attr node_state_attr[] = {
972 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
973 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
974 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
975#ifdef CONFIG_HIGHMEM
976 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
977#endif
978 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
979 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
980};
981
982static struct attribute *node_state_attrs[] = {
983 &node_state_attr[N_POSSIBLE].attr.attr,
984 &node_state_attr[N_ONLINE].attr.attr,
985 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
986#ifdef CONFIG_HIGHMEM
987 &node_state_attr[N_HIGH_MEMORY].attr.attr,
988#endif
989 &node_state_attr[N_MEMORY].attr.attr,
990 &node_state_attr[N_CPU].attr.attr,
991 NULL
992};
993
994static struct attribute_group memory_root_attr_group = {
995 .attrs = node_state_attrs,
996};
997
998static const struct attribute_group *cpu_root_attr_groups[] = {
999 &memory_root_attr_group,
1000 NULL,
1001};
1002
1003#define NODE_CALLBACK_PRI 2 /* lower than SLAB */
1004static int __init register_node_type(void)
1005{
1006 int ret;
1007
1008 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1009 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1010
1011 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1012 if (!ret) {
1013 static struct notifier_block node_memory_callback_nb = {
1014 .notifier_call = node_memory_callback,
1015 .priority = NODE_CALLBACK_PRI,
1016 };
1017 register_hotmemory_notifier(&node_memory_callback_nb);
1018 }
1019
1020 /*
1021 * Note: we're not going to unregister the node class if we fail
1022 * to register the node state class attribute files.
1023 */
1024 return ret;
1025}
1026postcore_initcall(register_node_type);