1 // SPDX-License-Identifier: GPL-2.0
3 * Basic Node interface support
6 #include <linux/module.h>
7 #include <linux/init.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>
23 static struct bus_type node_subsys = {
29 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
33 struct node *node_dev = to_node(dev);
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));
38 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
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);
48 static inline ssize_t node_read_cpumask(struct device *dev,
49 struct device_attribute *attr, char *buf)
51 return node_read_cpumap(dev, false, buf);
53 static inline ssize_t node_read_cpulist(struct device *dev,
54 struct device_attribute *attr, char *buf)
56 return node_read_cpumap(dev, true, buf);
59 static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
60 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
62 #define K(x) ((x) << (PAGE_SHIFT - 10))
63 static ssize_t node_read_meminfo(struct device *dev,
64 struct device_attribute *attr, char *buf)
68 struct pglist_data *pgdat = NODE_DATA(nid);
71 si_meminfo_node(&i, nid);
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",
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)));
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",
106 nid, K(i.totalram - i.totalhigh),
107 nid, K(i.freeram - i.freehigh));
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"
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)),
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) *
148 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
150 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
153 nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)));
155 n += hugetlb_report_node_meminfo(nid, buf + n);
160 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
162 static ssize_t node_read_numastat(struct device *dev,
163 struct device_attribute *attr, char *buf)
169 "interleave_hit %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));
179 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
181 static ssize_t node_read_vmstat(struct device *dev,
182 struct device_attribute *attr, char *buf)
185 struct pglist_data *pgdat = NODE_DATA(nid);
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));
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));
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')
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));
213 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
215 static ssize_t node_read_distance(struct device *dev,
216 struct device_attribute *attr, char *buf)
223 * buf is currently PAGE_SIZE in length and each node needs 4 chars
224 * at the most (distance + space or newline).
226 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
228 for_each_online_node(i)
229 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
231 len += sprintf(buf + len, "\n");
234 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
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,
245 ATTRIBUTE_GROUPS(node_dev);
247 #ifdef CONFIG_HUGETLBFS
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.
257 static node_registration_func_t __hugetlb_register_node;
258 static node_registration_func_t __hugetlb_unregister_node;
260 static inline bool hugetlb_register_node(struct node *node)
262 if (__hugetlb_register_node &&
263 node_state(node->dev.id, N_MEMORY)) {
264 __hugetlb_register_node(node);
270 static inline void hugetlb_unregister_node(struct node *node)
272 if (__hugetlb_unregister_node)
273 __hugetlb_unregister_node(node);
276 void register_hugetlbfs_with_node(node_registration_func_t doregister,
277 node_registration_func_t unregister)
279 __hugetlb_register_node = doregister;
280 __hugetlb_unregister_node = unregister;
283 static inline void hugetlb_register_node(struct node *node) {}
285 static inline void hugetlb_unregister_node(struct node *node) {}
288 static void node_device_release(struct device *dev)
290 struct node *node = to_node(dev);
292 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
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.
299 * The work is using node->node_work, so we should
300 * flush work before freeing the memory.
302 flush_work(&node->node_work);
308 * register_node - Setup a sysfs device for a node.
309 * @num - Node number to use when creating the device.
311 * Initialize and register the node device.
313 static int register_node(struct node *node, int 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);
324 put_device(&node->dev);
326 hugetlb_register_node(node);
328 compaction_register_node(node);
334 * unregister_node - unregister a node device
335 * @node: node going away
337 * Unregisters a node device @node. All the devices on the node must be
338 * unregistered before calling this function.
340 void unregister_node(struct node *node)
342 hugetlb_unregister_node(node); /* no-op, if memoryless node */
344 device_unregister(&node->dev);
347 struct node *node_devices[MAX_NUMNODES];
350 * register cpu under node
352 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
357 if (!node_online(nid))
360 obj = get_cpu_device(cpu);
364 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
366 kobject_name(&obj->kobj));
370 return sysfs_create_link(&obj->kobj,
371 &node_devices[nid]->dev.kobj,
372 kobject_name(&node_devices[nid]->dev.kobj));
375 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
379 if (!node_online(nid))
382 obj = get_cpu_device(cpu);
386 sysfs_remove_link(&node_devices[nid]->dev.kobj,
387 kobject_name(&obj->kobj));
388 sysfs_remove_link(&obj->kobj,
389 kobject_name(&node_devices[nid]->dev.kobj));
394 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
395 static int __ref get_nid_for_pfn(unsigned long pfn)
397 if (!pfn_valid_within(pfn))
399 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
400 if (system_state < SYSTEM_RUNNING)
401 return early_pfn_to_nid(pfn);
403 return pfn_to_nid(pfn);
406 static int do_register_memory_block_under_node(int nid,
407 struct memory_block *mem_blk)
412 * If this memory block spans multiple nodes, we only indicate
413 * the last processed node.
417 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
419 kobject_name(&mem_blk->dev.kobj));
423 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
424 &node_devices[nid]->dev.kobj,
425 kobject_name(&node_devices[nid]->dev.kobj));
428 /* register memory section under specified node if it spans that node */
429 int register_mem_block_under_node_early(struct memory_block *mem_blk, void *arg)
431 int nid = *(int *)arg;
432 unsigned long pfn, sect_start_pfn, sect_end_pfn;
434 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
435 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
436 sect_end_pfn += PAGES_PER_SECTION - 1;
437 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
441 * memory block could have several absent sections from start.
442 * skip pfn range from absent section
444 if (!pfn_present(pfn)) {
445 pfn = round_down(pfn + PAGES_PER_SECTION,
446 PAGES_PER_SECTION) - 1;
451 * We need to check if page belongs to nid only at the boot
452 * case because node's ranges can be interleaved.
454 page_nid = get_nid_for_pfn(pfn);
460 return do_register_memory_block_under_node(nid, mem_blk);
462 /* mem section does not span the specified node */
467 * During hotplug we know that all pages in the memory block belong to the same
470 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
473 int nid = *(int *)arg;
475 return do_register_memory_block_under_node(nid, mem_blk);
479 * Unregister a memory block device under the node it spans. Memory blocks
480 * with multiple nodes cannot be offlined and therefore also never be removed.
482 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
484 if (mem_blk->nid == NUMA_NO_NODE)
487 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
488 kobject_name(&mem_blk->dev.kobj));
489 sysfs_remove_link(&mem_blk->dev.kobj,
490 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
493 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn,
494 enum meminit_context context)
496 walk_memory_blocks_func_t func;
498 if (context == MEMINIT_HOTPLUG)
499 func = register_mem_block_under_node_hotplug;
501 func = register_mem_block_under_node_early;
503 return walk_memory_range(start_pfn, end_pfn, (void *)&nid, func);
506 #ifdef CONFIG_HUGETLBFS
508 * Handle per node hstate attribute [un]registration on transistions
509 * to/from memoryless state.
511 static void node_hugetlb_work(struct work_struct *work)
513 struct node *node = container_of(work, struct node, node_work);
516 * We only get here when a node transitions to/from memoryless state.
517 * We can detect which transition occurred by examining whether the
518 * node has memory now. hugetlb_register_node() already check this
519 * so we try to register the attributes. If that fails, then the
520 * node has transitioned to memoryless, try to unregister the
523 if (!hugetlb_register_node(node))
524 hugetlb_unregister_node(node);
527 static void init_node_hugetlb_work(int nid)
529 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
532 static int node_memory_callback(struct notifier_block *self,
533 unsigned long action, void *arg)
535 struct memory_notify *mnb = arg;
536 int nid = mnb->status_change_nid;
542 * offload per node hstate [un]registration to a work thread
543 * when transitioning to/from memoryless state.
545 if (nid != NUMA_NO_NODE)
546 schedule_work(&node_devices[nid]->node_work);
549 case MEM_GOING_ONLINE:
550 case MEM_GOING_OFFLINE:
551 case MEM_CANCEL_ONLINE:
552 case MEM_CANCEL_OFFLINE:
559 #endif /* CONFIG_HUGETLBFS */
560 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
562 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
563 !defined(CONFIG_HUGETLBFS)
564 static inline int node_memory_callback(struct notifier_block *self,
565 unsigned long action, void *arg)
570 static void init_node_hugetlb_work(int nid) { }
574 int __register_one_node(int nid)
579 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
580 if (!node_devices[nid])
583 error = register_node(node_devices[nid], nid);
585 /* link cpu under this node */
586 for_each_present_cpu(cpu) {
587 if (cpu_to_node(cpu) == nid)
588 register_cpu_under_node(cpu, nid);
591 /* initialize work queue for memory hot plug */
592 init_node_hugetlb_work(nid);
597 void unregister_one_node(int nid)
599 if (!node_devices[nid])
602 unregister_node(node_devices[nid]);
603 node_devices[nid] = NULL;
607 * node states attributes
610 static ssize_t print_nodes_state(enum node_states state, char *buf)
614 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
615 nodemask_pr_args(&node_states[state]));
622 struct device_attribute attr;
623 enum node_states state;
626 static ssize_t show_node_state(struct device *dev,
627 struct device_attribute *attr, char *buf)
629 struct node_attr *na = container_of(attr, struct node_attr, attr);
630 return print_nodes_state(na->state, buf);
633 #define _NODE_ATTR(name, state) \
634 { __ATTR(name, 0444, show_node_state, NULL), state }
636 static struct node_attr node_state_attr[] = {
637 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
638 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
639 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
640 #ifdef CONFIG_HIGHMEM
641 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
643 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
644 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
647 static struct attribute *node_state_attrs[] = {
648 &node_state_attr[N_POSSIBLE].attr.attr,
649 &node_state_attr[N_ONLINE].attr.attr,
650 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
651 #ifdef CONFIG_HIGHMEM
652 &node_state_attr[N_HIGH_MEMORY].attr.attr,
654 &node_state_attr[N_MEMORY].attr.attr,
655 &node_state_attr[N_CPU].attr.attr,
659 static struct attribute_group memory_root_attr_group = {
660 .attrs = node_state_attrs,
663 static const struct attribute_group *cpu_root_attr_groups[] = {
664 &memory_root_attr_group,
668 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
669 static int __init register_node_type(void)
673 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
674 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
676 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
678 static struct notifier_block node_memory_callback_nb = {
679 .notifier_call = node_memory_callback,
680 .priority = NODE_CALLBACK_PRI,
682 register_hotmemory_notifier(&node_memory_callback_nb);
686 * Note: we're not going to unregister the node class if we fail
687 * to register the node state class attribute files.
691 postcore_initcall(register_node_type);