2 * Basic Node interface support
5 #include <linux/module.h>
6 #include <linux/init.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>
22 static struct bus_type node_subsys = {
28 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
30 struct node *node_dev = to_node(dev);
31 const struct cpumask *mask = cpumask_of_node(node_dev->dev.id);
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));
36 return cpumap_print_to_pagebuf(list, buf, mask);
39 static inline ssize_t node_read_cpumask(struct device *dev,
40 struct device_attribute *attr, char *buf)
42 return node_read_cpumap(dev, false, buf);
44 static inline ssize_t node_read_cpulist(struct device *dev,
45 struct device_attribute *attr, char *buf)
47 return node_read_cpumap(dev, true, buf);
50 static DEVICE_ATTR(cpumap, S_IRUGO, node_read_cpumask, NULL);
51 static DEVICE_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
53 #define K(x) ((x) << (PAGE_SHIFT - 10))
54 static ssize_t node_read_meminfo(struct device *dev,
55 struct device_attribute *attr, char *buf)
59 struct pglist_data *pgdat = NODE_DATA(nid);
62 si_meminfo_node(&i, nid);
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",
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)));
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",
97 nid, K(i.totalram - i.totalhigh),
98 nid, K(i.freeram - i.freehigh));
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"
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)),
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(node_page_state(pgdat, NR_SLAB_RECLAIMABLE) +
133 node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
134 nid, K(node_page_state(pgdat, NR_SLAB_RECLAIMABLE)),
135 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
136 nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)),
137 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
139 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
141 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
144 nid, K(node_page_state(pgdat, NR_SLAB_UNRECLAIMABLE)));
146 n += hugetlb_report_node_meminfo(nid, buf + n);
151 static DEVICE_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
153 static ssize_t node_read_numastat(struct device *dev,
154 struct device_attribute *attr, char *buf)
160 "interleave_hit %lu\n"
163 sum_zone_numa_state(dev->id, NUMA_HIT),
164 sum_zone_numa_state(dev->id, NUMA_MISS),
165 sum_zone_numa_state(dev->id, NUMA_FOREIGN),
166 sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
167 sum_zone_numa_state(dev->id, NUMA_LOCAL),
168 sum_zone_numa_state(dev->id, NUMA_OTHER));
170 static DEVICE_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
172 static ssize_t node_read_vmstat(struct device *dev,
173 struct device_attribute *attr, char *buf)
176 struct pglist_data *pgdat = NODE_DATA(nid);
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));
185 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
186 n += sprintf(buf+n, "%s %lu\n",
187 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS],
188 sum_zone_numa_state(nid, i));
191 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
192 n += sprintf(buf+n, "%s %lu\n",
193 vmstat_text[i + NR_VM_ZONE_STAT_ITEMS +
194 NR_VM_NUMA_STAT_ITEMS],
195 node_page_state(pgdat, i));
199 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
201 static ssize_t node_read_distance(struct device *dev,
202 struct device_attribute *attr, char *buf)
209 * buf is currently PAGE_SIZE in length and each node needs 4 chars
210 * at the most (distance + space or newline).
212 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
214 for_each_online_node(i)
215 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
217 len += sprintf(buf + len, "\n");
220 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
222 static struct attribute *node_dev_attrs[] = {
223 &dev_attr_cpumap.attr,
224 &dev_attr_cpulist.attr,
225 &dev_attr_meminfo.attr,
226 &dev_attr_numastat.attr,
227 &dev_attr_distance.attr,
228 &dev_attr_vmstat.attr,
231 ATTRIBUTE_GROUPS(node_dev);
233 #ifdef CONFIG_HUGETLBFS
235 * hugetlbfs per node attributes registration interface:
236 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
237 * it will register its per node attributes for all online nodes with
238 * memory. It will also call register_hugetlbfs_with_node(), below, to
239 * register its attribute registration functions with this node driver.
240 * Once these hooks have been initialized, the node driver will call into
241 * the hugetlb module to [un]register attributes for hot-plugged nodes.
243 static node_registration_func_t __hugetlb_register_node;
244 static node_registration_func_t __hugetlb_unregister_node;
246 static inline bool hugetlb_register_node(struct node *node)
248 if (__hugetlb_register_node &&
249 node_state(node->dev.id, N_MEMORY)) {
250 __hugetlb_register_node(node);
256 static inline void hugetlb_unregister_node(struct node *node)
258 if (__hugetlb_unregister_node)
259 __hugetlb_unregister_node(node);
262 void register_hugetlbfs_with_node(node_registration_func_t doregister,
263 node_registration_func_t unregister)
265 __hugetlb_register_node = doregister;
266 __hugetlb_unregister_node = unregister;
269 static inline void hugetlb_register_node(struct node *node) {}
271 static inline void hugetlb_unregister_node(struct node *node) {}
274 static void node_device_release(struct device *dev)
276 struct node *node = to_node(dev);
278 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
280 * We schedule the work only when a memory section is
281 * onlined/offlined on this node. When we come here,
282 * all the memory on this node has been offlined,
283 * so we won't enqueue new work to this work.
285 * The work is using node->node_work, so we should
286 * flush work before freeing the memory.
288 flush_work(&node->node_work);
294 * register_node - Setup a sysfs device for a node.
295 * @num - Node number to use when creating the device.
297 * Initialize and register the node device.
299 static int register_node(struct node *node, int num)
304 node->dev.bus = &node_subsys;
305 node->dev.release = node_device_release;
306 node->dev.groups = node_dev_groups;
307 error = device_register(&node->dev);
310 hugetlb_register_node(node);
312 compaction_register_node(node);
318 * unregister_node - unregister a node device
319 * @node: node going away
321 * Unregisters a node device @node. All the devices on the node must be
322 * unregistered before calling this function.
324 void unregister_node(struct node *node)
326 hugetlb_unregister_node(node); /* no-op, if memoryless node */
328 device_unregister(&node->dev);
331 struct node *node_devices[MAX_NUMNODES];
334 * register cpu under node
336 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
341 if (!node_online(nid))
344 obj = get_cpu_device(cpu);
348 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
350 kobject_name(&obj->kobj));
354 return sysfs_create_link(&obj->kobj,
355 &node_devices[nid]->dev.kobj,
356 kobject_name(&node_devices[nid]->dev.kobj));
359 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
363 if (!node_online(nid))
366 obj = get_cpu_device(cpu);
370 sysfs_remove_link(&node_devices[nid]->dev.kobj,
371 kobject_name(&obj->kobj));
372 sysfs_remove_link(&obj->kobj,
373 kobject_name(&node_devices[nid]->dev.kobj));
378 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
379 static int __ref get_nid_for_pfn(unsigned long pfn)
381 if (!pfn_valid_within(pfn))
383 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
384 if (system_state < SYSTEM_RUNNING)
385 return early_pfn_to_nid(pfn);
387 return pfn_to_nid(pfn);
390 /* register memory section under specified node if it spans that node */
391 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
394 unsigned long pfn, sect_start_pfn, sect_end_pfn;
398 if (!node_online(nid))
401 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
402 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
403 sect_end_pfn += PAGES_PER_SECTION - 1;
404 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
408 * memory block could have several absent sections from start.
409 * skip pfn range from absent section
411 if (!pfn_present(pfn)) {
412 pfn = round_down(pfn + PAGES_PER_SECTION,
413 PAGES_PER_SECTION) - 1;
417 page_nid = get_nid_for_pfn(pfn);
422 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
424 kobject_name(&mem_blk->dev.kobj));
428 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
429 &node_devices[nid]->dev.kobj,
430 kobject_name(&node_devices[nid]->dev.kobj));
432 /* mem section does not span the specified node */
436 /* unregister memory section under all nodes that it spans */
437 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
438 unsigned long phys_index)
440 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
441 unsigned long pfn, sect_start_pfn, sect_end_pfn;
444 NODEMASK_FREE(unlinked_nodes);
449 nodes_clear(*unlinked_nodes);
451 sect_start_pfn = section_nr_to_pfn(phys_index);
452 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
453 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
456 nid = get_nid_for_pfn(pfn);
459 if (!node_online(nid))
461 if (node_test_and_set(nid, *unlinked_nodes))
463 sysfs_remove_link(&node_devices[nid]->dev.kobj,
464 kobject_name(&mem_blk->dev.kobj));
465 sysfs_remove_link(&mem_blk->dev.kobj,
466 kobject_name(&node_devices[nid]->dev.kobj));
468 NODEMASK_FREE(unlinked_nodes);
472 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long nr_pages)
474 unsigned long end_pfn = start_pfn + nr_pages;
476 struct memory_block *mem_blk = NULL;
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;
484 if (!present_section_nr(section_nr))
486 mem_sect = __nr_to_section(section_nr);
488 /* same memblock ? */
490 if ((section_nr >= mem_blk->start_section_nr) &&
491 (section_nr <= mem_blk->end_section_nr))
494 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
496 ret = register_mem_sect_under_node(mem_blk, nid);
500 /* discard ref obtained in find_memory_block() */
504 kobject_put(&mem_blk->dev.kobj);
508 #ifdef CONFIG_HUGETLBFS
510 * Handle per node hstate attribute [un]registration on transistions
511 * to/from memoryless state.
513 static void node_hugetlb_work(struct work_struct *work)
515 struct node *node = container_of(work, struct node, node_work);
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
525 if (!hugetlb_register_node(node))
526 hugetlb_unregister_node(node);
529 static void init_node_hugetlb_work(int nid)
531 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
534 static int node_memory_callback(struct notifier_block *self,
535 unsigned long action, void *arg)
537 struct memory_notify *mnb = arg;
538 int nid = mnb->status_change_nid;
544 * offload per node hstate [un]registration to a work thread
545 * when transitioning to/from memoryless state.
547 if (nid != NUMA_NO_NODE)
548 schedule_work(&node_devices[nid]->node_work);
551 case MEM_GOING_ONLINE:
552 case MEM_GOING_OFFLINE:
553 case MEM_CANCEL_ONLINE:
554 case MEM_CANCEL_OFFLINE:
561 #endif /* CONFIG_HUGETLBFS */
562 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
564 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
565 !defined(CONFIG_HUGETLBFS)
566 static inline int node_memory_callback(struct notifier_block *self,
567 unsigned long action, void *arg)
572 static void init_node_hugetlb_work(int nid) { }
576 int __register_one_node(int nid)
581 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
582 if (!node_devices[nid])
585 error = register_node(node_devices[nid], nid);
587 /* link cpu under this node */
588 for_each_present_cpu(cpu) {
589 if (cpu_to_node(cpu) == nid)
590 register_cpu_under_node(cpu, nid);
593 /* initialize work queue for memory hot plug */
594 init_node_hugetlb_work(nid);
599 void unregister_one_node(int nid)
601 if (!node_devices[nid])
604 unregister_node(node_devices[nid]);
605 node_devices[nid] = NULL;
609 * node states attributes
612 static ssize_t print_nodes_state(enum node_states state, char *buf)
616 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
617 nodemask_pr_args(&node_states[state]));
624 struct device_attribute attr;
625 enum node_states state;
628 static ssize_t show_node_state(struct device *dev,
629 struct device_attribute *attr, char *buf)
631 struct node_attr *na = container_of(attr, struct node_attr, attr);
632 return print_nodes_state(na->state, buf);
635 #define _NODE_ATTR(name, state) \
636 { __ATTR(name, 0444, show_node_state, NULL), state }
638 static struct node_attr node_state_attr[] = {
639 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
640 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
641 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
642 #ifdef CONFIG_HIGHMEM
643 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
645 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
646 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
649 static struct attribute *node_state_attrs[] = {
650 &node_state_attr[N_POSSIBLE].attr.attr,
651 &node_state_attr[N_ONLINE].attr.attr,
652 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
653 #ifdef CONFIG_HIGHMEM
654 &node_state_attr[N_HIGH_MEMORY].attr.attr,
656 &node_state_attr[N_MEMORY].attr.attr,
657 &node_state_attr[N_CPU].attr.attr,
661 static struct attribute_group memory_root_attr_group = {
662 .attrs = node_state_attrs,
665 static const struct attribute_group *cpu_root_attr_groups[] = {
666 &memory_root_attr_group,
670 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
671 static int __init register_node_type(void)
675 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
676 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
678 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
680 static struct notifier_block node_memory_callback_nb = {
681 .notifier_call = node_memory_callback,
682 .priority = NODE_CALLBACK_PRI,
684 register_hotmemory_notifier(&node_memory_callback_nb);
688 * Note: we're not going to unregister the node class if we fail
689 * to register the node state class attribute files.
693 postcore_initcall(register_node_type);