Linux 4.16.11
[linux/fpc-iii.git] / drivers / base / node.c
blobee090ab9171c269dfaeade5f1c1c4e5584566e70
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Basic Node interface support
4 */
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>
23 static struct bus_type node_subsys = {
24 .name = "node",
25 .dev_name = "node",
29 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
31 ssize_t n;
32 cpumask_var_t mask;
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))
39 return 0;
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);
45 return n;
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)
66 int n;
67 int nid = dev->id;
68 struct pglist_data *pgdat = NODE_DATA(nid);
69 struct sysinfo i;
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)));
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
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;
159 #undef K
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)
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));
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)
184 int nid = dev->id;
185 struct pglist_data *pgdat = NODE_DATA(nid);
186 int i;
187 int n = 0;
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));
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
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));
206 return n;
208 static DEVICE_ATTR(vmstat, S_IRUGO, node_read_vmstat, NULL);
210 static ssize_t node_read_distance(struct device *dev,
211 struct device_attribute *attr, char *buf)
213 int nid = dev->id;
214 int len = 0;
215 int i;
218 * buf is currently PAGE_SIZE in length and each node needs 4 chars
219 * at the most (distance + space or newline).
221 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
223 for_each_online_node(i)
224 len += sprintf(buf + len, "%s%d", i ? " " : "", node_distance(nid, i));
226 len += sprintf(buf + len, "\n");
227 return len;
229 static DEVICE_ATTR(distance, S_IRUGO, node_read_distance, NULL);
231 static 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
240 ATTRIBUTE_GROUPS(node_dev);
242 #ifdef CONFIG_HUGETLBFS
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.
252 static node_registration_func_t __hugetlb_register_node;
253 static node_registration_func_t __hugetlb_unregister_node;
255 static inline bool hugetlb_register_node(struct node *node)
257 if (__hugetlb_register_node &&
258 node_state(node->dev.id, N_MEMORY)) {
259 __hugetlb_register_node(node);
260 return true;
262 return false;
265 static inline void hugetlb_unregister_node(struct node *node)
267 if (__hugetlb_unregister_node)
268 __hugetlb_unregister_node(node);
271 void register_hugetlbfs_with_node(node_registration_func_t doregister,
272 node_registration_func_t unregister)
274 __hugetlb_register_node = doregister;
275 __hugetlb_unregister_node = unregister;
277 #else
278 static inline void hugetlb_register_node(struct node *node) {}
280 static inline void hugetlb_unregister_node(struct node *node) {}
281 #endif
283 static void node_device_release(struct device *dev)
285 struct node *node = to_node(dev);
287 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
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.
294 * The work is using node->node_work, so we should
295 * flush work before freeing the memory.
297 flush_work(&node->node_work);
298 #endif
299 kfree(node);
303 * register_node - Setup a sysfs device for a node.
304 * @num - Node number to use when creating the device.
306 * Initialize and register the node device.
308 static int register_node(struct node *node, int num)
310 int error;
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);
318 if (!error){
319 hugetlb_register_node(node);
321 compaction_register_node(node);
323 return error;
327 * unregister_node - unregister a node device
328 * @node: node going away
330 * Unregisters a node device @node. All the devices on the node must be
331 * unregistered before calling this function.
333 void unregister_node(struct node *node)
335 hugetlb_unregister_node(node); /* no-op, if memoryless node */
337 device_unregister(&node->dev);
340 struct node *node_devices[MAX_NUMNODES];
343 * register cpu under node
345 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
347 int ret;
348 struct device *obj;
350 if (!node_online(nid))
351 return 0;
353 obj = get_cpu_device(cpu);
354 if (!obj)
355 return 0;
357 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
358 &obj->kobj,
359 kobject_name(&obj->kobj));
360 if (ret)
361 return ret;
363 return sysfs_create_link(&obj->kobj,
364 &node_devices[nid]->dev.kobj,
365 kobject_name(&node_devices[nid]->dev.kobj));
368 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
370 struct device *obj;
372 if (!node_online(nid))
373 return 0;
375 obj = get_cpu_device(cpu);
376 if (!obj)
377 return 0;
379 sysfs_remove_link(&node_devices[nid]->dev.kobj,
380 kobject_name(&obj->kobj));
381 sysfs_remove_link(&obj->kobj,
382 kobject_name(&node_devices[nid]->dev.kobj));
384 return 0;
387 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
388 static int __ref get_nid_for_pfn(unsigned long pfn)
390 if (!pfn_valid_within(pfn))
391 return -1;
392 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
393 if (system_state < SYSTEM_RUNNING)
394 return early_pfn_to_nid(pfn);
395 #endif
396 return pfn_to_nid(pfn);
399 /* register memory section under specified node if it spans that node */
400 int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
402 int ret;
403 unsigned long pfn, sect_start_pfn, sect_end_pfn;
405 if (!mem_blk)
406 return -EFAULT;
407 if (!node_online(nid))
408 return 0;
410 sect_start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
411 sect_end_pfn = section_nr_to_pfn(mem_blk->end_section_nr);
412 sect_end_pfn += PAGES_PER_SECTION - 1;
413 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
414 int page_nid;
417 * memory block could have several absent sections from start.
418 * skip pfn range from absent section
420 if (!pfn_present(pfn)) {
421 pfn = round_down(pfn + PAGES_PER_SECTION,
422 PAGES_PER_SECTION) - 1;
423 continue;
426 page_nid = get_nid_for_pfn(pfn);
427 if (page_nid < 0)
428 continue;
429 if (page_nid != nid)
430 continue;
431 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
432 &mem_blk->dev.kobj,
433 kobject_name(&mem_blk->dev.kobj));
434 if (ret)
435 return ret;
437 return sysfs_create_link_nowarn(&mem_blk->dev.kobj,
438 &node_devices[nid]->dev.kobj,
439 kobject_name(&node_devices[nid]->dev.kobj));
441 /* mem section does not span the specified node */
442 return 0;
445 /* unregister memory section under all nodes that it spans */
446 int unregister_mem_sect_under_nodes(struct memory_block *mem_blk,
447 unsigned long phys_index)
449 NODEMASK_ALLOC(nodemask_t, unlinked_nodes, GFP_KERNEL);
450 unsigned long pfn, sect_start_pfn, sect_end_pfn;
452 if (!mem_blk) {
453 NODEMASK_FREE(unlinked_nodes);
454 return -EFAULT;
456 if (!unlinked_nodes)
457 return -ENOMEM;
458 nodes_clear(*unlinked_nodes);
460 sect_start_pfn = section_nr_to_pfn(phys_index);
461 sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
462 for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
463 int nid;
465 nid = get_nid_for_pfn(pfn);
466 if (nid < 0)
467 continue;
468 if (!node_online(nid))
469 continue;
470 if (node_test_and_set(nid, *unlinked_nodes))
471 continue;
472 sysfs_remove_link(&node_devices[nid]->dev.kobj,
473 kobject_name(&mem_blk->dev.kobj));
474 sysfs_remove_link(&mem_blk->dev.kobj,
475 kobject_name(&node_devices[nid]->dev.kobj));
477 NODEMASK_FREE(unlinked_nodes);
478 return 0;
481 int link_mem_sections(int nid, unsigned long start_pfn, unsigned long nr_pages)
483 unsigned long end_pfn = start_pfn + nr_pages;
484 unsigned long pfn;
485 struct memory_block *mem_blk = NULL;
486 int err = 0;
488 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
489 unsigned long section_nr = pfn_to_section_nr(pfn);
490 struct mem_section *mem_sect;
491 int ret;
493 if (!present_section_nr(section_nr))
494 continue;
495 mem_sect = __nr_to_section(section_nr);
497 /* same memblock ? */
498 if (mem_blk)
499 if ((section_nr >= mem_blk->start_section_nr) &&
500 (section_nr <= mem_blk->end_section_nr))
501 continue;
503 mem_blk = find_memory_block_hinted(mem_sect, mem_blk);
505 ret = register_mem_sect_under_node(mem_blk, nid);
506 if (!err)
507 err = ret;
509 /* discard ref obtained in find_memory_block() */
512 if (mem_blk)
513 kobject_put(&mem_blk->dev.kobj);
514 return err;
517 #ifdef CONFIG_HUGETLBFS
519 * Handle per node hstate attribute [un]registration on transistions
520 * to/from memoryless state.
522 static void node_hugetlb_work(struct work_struct *work)
524 struct node *node = container_of(work, struct node, node_work);
527 * We only get here when a node transitions to/from memoryless state.
528 * We can detect which transition occurred by examining whether the
529 * node has memory now. hugetlb_register_node() already check this
530 * so we try to register the attributes. If that fails, then the
531 * node has transitioned to memoryless, try to unregister the
532 * attributes.
534 if (!hugetlb_register_node(node))
535 hugetlb_unregister_node(node);
538 static void init_node_hugetlb_work(int nid)
540 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
543 static int node_memory_callback(struct notifier_block *self,
544 unsigned long action, void *arg)
546 struct memory_notify *mnb = arg;
547 int nid = mnb->status_change_nid;
549 switch (action) {
550 case MEM_ONLINE:
551 case MEM_OFFLINE:
553 * offload per node hstate [un]registration to a work thread
554 * when transitioning to/from memoryless state.
556 if (nid != NUMA_NO_NODE)
557 schedule_work(&node_devices[nid]->node_work);
558 break;
560 case MEM_GOING_ONLINE:
561 case MEM_GOING_OFFLINE:
562 case MEM_CANCEL_ONLINE:
563 case MEM_CANCEL_OFFLINE:
564 default:
565 break;
568 return NOTIFY_OK;
570 #endif /* CONFIG_HUGETLBFS */
571 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
573 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
574 !defined(CONFIG_HUGETLBFS)
575 static inline int node_memory_callback(struct notifier_block *self,
576 unsigned long action, void *arg)
578 return NOTIFY_OK;
581 static void init_node_hugetlb_work(int nid) { }
583 #endif
585 int __register_one_node(int nid)
587 int error;
588 int cpu;
590 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
591 if (!node_devices[nid])
592 return -ENOMEM;
594 error = register_node(node_devices[nid], nid);
596 /* link cpu under this node */
597 for_each_present_cpu(cpu) {
598 if (cpu_to_node(cpu) == nid)
599 register_cpu_under_node(cpu, nid);
602 /* initialize work queue for memory hot plug */
603 init_node_hugetlb_work(nid);
605 return error;
608 void unregister_one_node(int nid)
610 if (!node_devices[nid])
611 return;
613 unregister_node(node_devices[nid]);
614 node_devices[nid] = NULL;
618 * node states attributes
621 static ssize_t print_nodes_state(enum node_states state, char *buf)
623 int n;
625 n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
626 nodemask_pr_args(&node_states[state]));
627 buf[n++] = '\n';
628 buf[n] = '\0';
629 return n;
632 struct node_attr {
633 struct device_attribute attr;
634 enum node_states state;
637 static ssize_t show_node_state(struct device *dev,
638 struct device_attribute *attr, char *buf)
640 struct node_attr *na = container_of(attr, struct node_attr, attr);
641 return print_nodes_state(na->state, buf);
644 #define _NODE_ATTR(name, state) \
645 { __ATTR(name, 0444, show_node_state, NULL), state }
647 static struct node_attr node_state_attr[] = {
648 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
649 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
650 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
651 #ifdef CONFIG_HIGHMEM
652 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
653 #endif
654 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
655 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
658 static struct attribute *node_state_attrs[] = {
659 &node_state_attr[N_POSSIBLE].attr.attr,
660 &node_state_attr[N_ONLINE].attr.attr,
661 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
662 #ifdef CONFIG_HIGHMEM
663 &node_state_attr[N_HIGH_MEMORY].attr.attr,
664 #endif
665 &node_state_attr[N_MEMORY].attr.attr,
666 &node_state_attr[N_CPU].attr.attr,
667 NULL
670 static struct attribute_group memory_root_attr_group = {
671 .attrs = node_state_attrs,
674 static const struct attribute_group *cpu_root_attr_groups[] = {
675 &memory_root_attr_group,
676 NULL,
679 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
680 static int __init register_node_type(void)
682 int ret;
684 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
685 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
687 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
688 if (!ret) {
689 static struct notifier_block node_memory_callback_nb = {
690 .notifier_call = node_memory_callback,
691 .priority = NODE_CALLBACK_PRI,
693 register_hotmemory_notifier(&node_memory_callback_nb);
697 * Note: we're not going to unregister the node class if we fail
698 * to register the node state class attribute files.
700 return ret;
702 postcore_initcall(register_node_type);