Linux Audio

Check our new training course

Loading...
v4.10.11
 
  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);
v4.17
  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
 23static struct bus_type node_subsys = {
 24	.name = "node",
 25	.dev_name = "node",
 26};
 27
 28
 29static 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
 48static 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}
 53static 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
 59static DEVICE_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
 60static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
 61
 62#define K(x) ((x) << (PAGE_SHIFT - 10))
 63static 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
160static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
161
162static 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}
179static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
180
181static 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		n += sprintf(buf+n, "%s %lu\n",
202			     vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
203			     NR_VM_NUMA_STAT_ITEMS],
204			     node_page_state(pgdat, i));
205
206	return n;
207}
208static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
209
210static ssize_t node_read_distance(struct device *dev,
211			struct device_attribute *attr, char *buf)
212{
213	int nid = dev->id;
214	int len = 0;
215	int i;
216
217	/*
218	 * buf is currently PAGE_SIZE in length and each node needs 4 chars
219	 * at the most (distance + space or newline).
220	 */
221	BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
222
223	for_each_online_node(i)
224		len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
225
226	len += sprintf(buf + len, "\n");
227	return len;
228}
229static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
230
231static struct attribute *node_dev_attrs[] = {
232	&dev_attr_cpumap.attr,
233	&dev_attr_cpulist.attr,
234	&dev_attr_meminfo.attr,
235	&dev_attr_numastat.attr,
236	&dev_attr_distance.attr,
237	&dev_attr_vmstat.attr,
238	NULL
239};
240ATTRIBUTE_GROUPS(node_dev);
241
242#ifdef CONFIG_HUGETLBFS
243/*
244 * hugetlbfs per node attributes registration interface:
245 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
246 * it will register its per node attributes for all online nodes with
247 * memory.  It will also call register_hugetlbfs_with_node(), below, to
248 * register its attribute registration functions with this node driver.
249 * Once these hooks have been initialized, the node driver will call into
250 * the hugetlb module to [un]register attributes for hot-plugged nodes.
251 */
252static node_registration_func_t __hugetlb_register_node;
253static node_registration_func_t __hugetlb_unregister_node;
254
255static inline bool hugetlb_register_node(struct node *node)
256{
257	if (__hugetlb_register_node &&
258			node_state(node->dev.id, N_MEMORY)) {
259		__hugetlb_register_node(node);
260		return true;
261	}
262	return false;
263}
264
265static inline void hugetlb_unregister_node(struct node *node)
266{
267	if (__hugetlb_unregister_node)
268		__hugetlb_unregister_node(node);
269}
270
271void register_hugetlbfs_with_node(node_registration_func_t doregister,
272				  node_registration_func_t unregister)
273{
274	__hugetlb_register_node   = doregister;
275	__hugetlb_unregister_node = unregister;
276}
277#else
278static inline void hugetlb_register_node(struct node *node) {}
279
280static inline void hugetlb_unregister_node(struct node *node) {}
281#endif
282
283static void node_device_release(struct device *dev)
284{
285	struct node *node = to_node(dev);
286
287#if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
288	/*
289	 * We schedule the work only when a memory section is
290	 * onlined/offlined on this node. When we come here,
291	 * all the memory on this node has been offlined,
292	 * so we won't enqueue new work to this work.
293	 *
294	 * The work is using node->node_work, so we should
295	 * flush work before freeing the memory.
296	 */
297	flush_work(&node->node_work);
298#endif
299	kfree(node);
300}
301
302/*
303 * register_node - Setup a sysfs device for a node.
304 * @num - Node number to use when creating the device.
305 *
306 * Initialize and register the node device.
307 */
308static int register_node(struct node *node, int num)
309{
310	int error;
311
312	node->dev.id = num;
313	node->dev.bus = &node_subsys;
314	node->dev.release = node_device_release;
315	node->dev.groups = node_dev_groups;
316	error = device_register(&node->dev);
317
318	if (error)
319		put_device(&node->dev);
320	else {
321		hugetlb_register_node(node);
322
323		compaction_register_node(node);
324	}
325	return error;
326}
327
328/**
329 * unregister_node - unregister a node device
330 * @node: node going away
331 *
332 * Unregisters a node device @node.  All the devices on the node must be
333 * unregistered before calling this function.
334 */
335void unregister_node(struct node *node)
336{
337	hugetlb_unregister_node(node);		/* no-op, if memoryless node */
338
339	device_unregister(&node->dev);
340}
341
342struct node *node_devices[MAX_NUMNODES];
343
344/*
345 * register cpu under node
346 */
347int register_cpu_under_node(unsigned int cpu, unsigned int nid)
348{
349	int ret;
350	struct device *obj;
351
352	if (!node_online(nid))
353		return 0;
354
355	obj = get_cpu_device(cpu);
356	if (!obj)
357		return 0;
358
359	ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
360				&obj->kobj,
361				kobject_name(&obj->kobj));
362	if (ret)
363		return ret;
364
365	return sysfs_create_link(&obj->kobj,
366				 &node_devices[nid]->dev.kobj,
367				 kobject_name(&node_devices[nid]->dev.kobj));
368}
369
370int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
371{
372	struct device *obj;
373
374	if (!node_online(nid))
375		return 0;
376
377	obj = get_cpu_device(cpu);
378	if (!obj)
379		return 0;
380
381	sysfs_remove_link(&node_devices[nid]->dev.kobj,
382			  kobject_name(&obj->kobj));
383	sysfs_remove_link(&obj->kobj,
384			  kobject_name(&node_devices[nid]->dev.kobj));
385
386	return 0;
387}
388
389#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 
 
390static int __ref get_nid_for_pfn(unsigned long pfn)
391{
 
 
392	if (!pfn_valid_within(pfn))
393		return -1;
394#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
395	if (system_state < SYSTEM_RUNNING)
396		return early_pfn_to_nid(pfn);
397#endif
 
 
 
398	return pfn_to_nid(pfn);
399}
400
401/* register memory section under specified node if it spans that node */
402int register_mem_sect_under_node(struct memory_block *mem_blk, int nid,
403				 bool check_nid)
404{
405	int ret;
406	unsigned long pfn, sect_start_pfn, sect_end_pfn;
407
408	if (!mem_blk)
409		return -EFAULT;
410
411	mem_blk->nid = nid;
412	if (!node_online(nid))
413		return 0;
414
415	sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
416	sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
417	sect_end_pfn += PAGES_PER_SECTION - 1;
418	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
419		int page_nid;
420
421		/*
422		 * memory block could have several absent sections from start.
423		 * skip pfn range from absent section
424		 */
425		if (!pfn_present(pfn)) {
426			pfn = round_down(pfn + PAGES_PER_SECTION,
427					 PAGES_PER_SECTION) - 1;
428			continue;
429		}
430
431		/*
432		 * We need to check if page belongs to nid only for the boot
433		 * case, during hotplug we know that all pages in the memory
434		 * block belong to the same node.
435		 */
436		if (check_nid) {
437			page_nid = get_nid_for_pfn(pfn);
438			if (page_nid < 0)
439				continue;
440			if (page_nid != nid)
441				continue;
442		}
443		ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
444					&mem_blk->dev.kobj,
445					kobject_name(&mem_blk->dev.kobj));
446		if (ret)
447			return ret;
448
449		return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
450				&node_devices[nid]->dev.kobj,
451				kobject_name(&node_devices[nid]->dev.kobj));
452	}
453	/* mem section does not span the specified node */
454	return 0;
455}
456
457/* unregister memory section under all nodes that it spans */
458int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
459				    unsigned long phys_index)
460{
461	NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
462	unsigned long pfn, sect_start_pfn, sect_end_pfn;
463
464	if (!mem_blk) {
465		NODEMASK_FREE(unlinked_nodes);
466		return -EFAULT;
467	}
468	if (!unlinked_nodes)
469		return -ENOMEM;
470	nodes_clear(*unlinked_nodes);
471
472	sect_start_pfn = section_nr_to_pfn(phys_index);
473	sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
474	for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
475		int nid;
476
477		nid = get_nid_for_pfn(pfn);
478		if (nid < 0)
479			continue;
480		if (!node_online(nid))
481			continue;
482		if (node_test_and_set(nid, *unlinked_nodes))
483			continue;
484		sysfs_remove_link(&node_devices[nid]->dev.kobj,
485			 kobject_name(&mem_blk->dev.kobj));
486		sysfs_remove_link(&mem_blk->dev.kobj,
487			 kobject_name(&node_devices[nid]->dev.kobj));
488	}
489	NODEMASK_FREE(unlinked_nodes);
490	return 0;
491}
492
493int link_mem_sections(int nid, unsigned long start_pfn, unsigned long nr_pages,
494		      bool check_nid)
495{
496	unsigned long end_pfn = start_pfn + nr_pages;
 
497	unsigned long pfn;
498	struct memory_block *mem_blk = NULL;
499	int err = 0;
500
501	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
502		unsigned long section_nr = pfn_to_section_nr(pfn);
503		struct mem_section *mem_sect;
504		int ret;
505
506		if (!present_section_nr(section_nr))
507			continue;
508		mem_sect = __nr_to_section(section_nr);
509
510		/* same memblock ? */
511		if (mem_blk)
512			if ((section_nr >= mem_blk->start_section_nr) &&
513			    (section_nr <= mem_blk->end_section_nr))
514				continue;
515
516		mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
517
518		ret = register_mem_sect_under_node(mem_blk, nid, check_nid);
519		if (!err)
520			err = ret;
521
522		/* discard ref obtained in find_memory_block() */
523	}
524
525	if (mem_blk)
526		kobject_put(&mem_blk->dev.kobj);
527	return err;
528}
529
530#ifdef CONFIG_HUGETLBFS
531/*
532 * Handle per node hstate attribute [un]registration on transistions
533 * to/from memoryless state.
534 */
535static void node_hugetlb_work(struct work_struct *work)
536{
537	struct node *node = container_of(work, struct node, node_work);
538
539	/*
540	 * We only get here when a node transitions to/from memoryless state.
541	 * We can detect which transition occurred by examining whether the
542	 * node has memory now.  hugetlb_register_node() already check this
543	 * so we try to register the attributes.  If that fails, then the
544	 * node has transitioned to memoryless, try to unregister the
545	 * attributes.
546	 */
547	if (!hugetlb_register_node(node))
548		hugetlb_unregister_node(node);
549}
550
551static void init_node_hugetlb_work(int nid)
552{
553	INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
554}
555
556static int node_memory_callback(struct notifier_block *self,
557				unsigned long action, void *arg)
558{
559	struct memory_notify *mnb = arg;
560	int nid = mnb->status_change_nid;
561
562	switch (action) {
563	case MEM_ONLINE:
564	case MEM_OFFLINE:
565		/*
566		 * offload per node hstate [un]registration to a work thread
567		 * when transitioning to/from memoryless state.
568		 */
569		if (nid != NUMA_NO_NODE)
570			schedule_work(&node_devices[nid]->node_work);
571		break;
572
573	case MEM_GOING_ONLINE:
574	case MEM_GOING_OFFLINE:
575	case MEM_CANCEL_ONLINE:
576	case MEM_CANCEL_OFFLINE:
577	default:
578		break;
579	}
580
581	return NOTIFY_OK;
582}
583#endif	/* CONFIG_HUGETLBFS */
584#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 
 
 
585
586#if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
587    !defined(CONFIG_HUGETLBFS)
588static inline int node_memory_callback(struct notifier_block *self,
589				unsigned long action, void *arg)
590{
591	return NOTIFY_OK;
592}
593
594static void init_node_hugetlb_work(int nid) { }
595
596#endif
597
598int __register_one_node(int nid)
599{
600	int error;
601	int cpu;
602
603	node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
604	if (!node_devices[nid])
605		return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
606
607	error = register_node(node_devices[nid], nid);
 
608
609	/* link cpu under this node */
610	for_each_present_cpu(cpu) {
611		if (cpu_to_node(cpu) == nid)
612			register_cpu_under_node(cpu, nid);
613	}
614
615	/* initialize work queue for memory hot plug */
616	init_node_hugetlb_work(nid);
617
618	return error;
619}
620
621void unregister_one_node(int nid)
622{
623	if (!node_devices[nid])
624		return;
625
626	unregister_node(node_devices[nid]);
627	node_devices[nid] = NULL;
628}
629
630/*
631 * node states attributes
632 */
633
634static ssize_t print_nodes_state(enum node_states state, char *buf)
635{
636	int n;
637
638	n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
639		      nodemask_pr_args(&node_states[state]));
640	buf[n++] = '\n';
641	buf[n] = '\0';
642	return n;
643}
644
645struct node_attr {
646	struct device_attribute attr;
647	enum node_states state;
648};
649
650static ssize_t show_node_state(struct device *dev,
651			       struct device_attribute *attr, char *buf)
652{
653	struct node_attr *na = container_of(attr, struct node_attr, attr);
654	return print_nodes_state(na->state, buf);
655}
656
657#define _NODE_ATTR(name, state) \
658	{ __ATTR(name, 0444, show_node_state, NULL), state }
659
660static struct node_attr node_state_attr[] = {
661	[N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
662	[N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
663	[N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
664#ifdef CONFIG_HIGHMEM
665	[N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
666#endif
 
667	[N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
 
668	[N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
669};
670
671static struct attribute *node_state_attrs[] = {
672	&node_state_attr[N_POSSIBLE].attr.attr,
673	&node_state_attr[N_ONLINE].attr.attr,
674	&node_state_attr[N_NORMAL_MEMORY].attr.attr,
675#ifdef CONFIG_HIGHMEM
676	&node_state_attr[N_HIGH_MEMORY].attr.attr,
677#endif
 
678	&node_state_attr[N_MEMORY].attr.attr,
 
679	&node_state_attr[N_CPU].attr.attr,
680	NULL
681};
682
683static struct attribute_group memory_root_attr_group = {
684	.attrs = node_state_attrs,
685};
686
687static const struct attribute_group *cpu_root_attr_groups[] = {
688	&memory_root_attr_group,
689	NULL,
690};
691
692#define NODE_CALLBACK_PRI	2	/* lower than SLAB */
693static int __init register_node_type(void)
694{
695	int ret;
696
697 	BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
698 	BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
699
700	ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
701	if (!ret) {
702		static struct notifier_block node_memory_callback_nb = {
703			.notifier_call = node_memory_callback,
704			.priority = NODE_CALLBACK_PRI,
705		};
706		register_hotmemory_notifier(&node_memory_callback_nb);
707	}
708
709	/*
710	 * Note:  we're not going to unregister the node class if we fail
711	 * to register the node state class attribute files.
712	 */
713	return ret;
714}
715postcore_initcall(register_node_type);