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