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/pm_runtime.h>
21 #include <linux/swap.h>
22 #include <linux/slab.h>
24 static struct bus_type node_subsys = {
30 static ssize_t node_read_cpumap(struct device *dev, bool list, char *buf)
34 struct node *node_dev = to_node(dev);
36 /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
37 BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
39 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
42 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
43 n = cpumap_print_to_pagebuf(list, buf, mask);
44 free_cpumask_var(mask);
49 static inline ssize_t cpumap_show(struct device *dev,
50 struct device_attribute *attr,
53 return node_read_cpumap(dev, false, buf);
56 static DEVICE_ATTR_RO(cpumap);
58 static inline ssize_t cpulist_show(struct device *dev,
59 struct device_attribute *attr,
62 return node_read_cpumap(dev, true, buf);
65 static DEVICE_ATTR_RO(cpulist);
68 * struct node_access_nodes - Access class device to hold user visible
69 * relationships to other nodes.
70 * @dev: Device for this memory access class
71 * @list_node: List element in the node's access list
72 * @access: The access class rank
73 * @hmem_attrs: Heterogeneous memory performance attributes
75 struct node_access_nodes {
77 struct list_head list_node;
79 #ifdef CONFIG_HMEM_REPORTING
80 struct node_hmem_attrs hmem_attrs;
83 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
85 static struct attribute *node_init_access_node_attrs[] = {
89 static struct attribute *node_targ_access_node_attrs[] = {
93 static const struct attribute_group initiators = {
95 .attrs = node_init_access_node_attrs,
98 static const struct attribute_group targets = {
100 .attrs = node_targ_access_node_attrs,
103 static const struct attribute_group *node_access_node_groups[] = {
109 static void node_remove_accesses(struct node *node)
111 struct node_access_nodes *c, *cnext;
113 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
114 list_del(&c->list_node);
115 device_unregister(&c->dev);
119 static void node_access_release(struct device *dev)
121 kfree(to_access_nodes(dev));
124 static struct node_access_nodes *node_init_node_access(struct node *node,
127 struct node_access_nodes *access_node;
130 list_for_each_entry(access_node, &node->access_list, list_node)
131 if (access_node->access == access)
134 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
138 access_node->access = access;
139 dev = &access_node->dev;
140 dev->parent = &node->dev;
141 dev->release = node_access_release;
142 dev->groups = node_access_node_groups;
143 if (dev_set_name(dev, "access%u", access))
146 if (device_register(dev))
149 pm_runtime_no_callbacks(dev);
150 list_add_tail(&access_node->list_node, &node->access_list);
153 kfree_const(dev->kobj.name);
159 #ifdef CONFIG_HMEM_REPORTING
160 #define ACCESS_ATTR(name) \
161 static ssize_t name##_show(struct device *dev, \
162 struct device_attribute *attr, \
165 return sysfs_emit(buf, "%u\n", \
166 to_access_nodes(dev)->hmem_attrs.name); \
168 static DEVICE_ATTR_RO(name)
170 ACCESS_ATTR(read_bandwidth);
171 ACCESS_ATTR(read_latency);
172 ACCESS_ATTR(write_bandwidth);
173 ACCESS_ATTR(write_latency);
175 static struct attribute *access_attrs[] = {
176 &dev_attr_read_bandwidth.attr,
177 &dev_attr_read_latency.attr,
178 &dev_attr_write_bandwidth.attr,
179 &dev_attr_write_latency.attr,
184 * node_set_perf_attrs - Set the performance values for given access class
185 * @nid: Node identifier to be set
186 * @hmem_attrs: Heterogeneous memory performance attributes
187 * @access: The access class the for the given attributes
189 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
192 struct node_access_nodes *c;
196 if (WARN_ON_ONCE(!node_online(nid)))
199 node = node_devices[nid];
200 c = node_init_node_access(node, access);
204 c->hmem_attrs = *hmem_attrs;
205 for (i = 0; access_attrs[i] != NULL; i++) {
206 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
208 pr_info("failed to add performance attribute to node %d\n",
216 * struct node_cache_info - Internal tracking for memory node caches
217 * @dev: Device represeting the cache level
218 * @node: List element for tracking in the node
219 * @cache_attrs:Attributes for this cache level
221 struct node_cache_info {
223 struct list_head node;
224 struct node_cache_attrs cache_attrs;
226 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
228 #define CACHE_ATTR(name, fmt) \
229 static ssize_t name##_show(struct device *dev, \
230 struct device_attribute *attr, \
233 return sysfs_emit(buf, fmt "\n", \
234 to_cache_info(dev)->cache_attrs.name); \
236 DEVICE_ATTR_RO(name);
238 CACHE_ATTR(size, "%llu")
239 CACHE_ATTR(line_size, "%u")
240 CACHE_ATTR(indexing, "%u")
241 CACHE_ATTR(write_policy, "%u")
243 static struct attribute *cache_attrs[] = {
244 &dev_attr_indexing.attr,
246 &dev_attr_line_size.attr,
247 &dev_attr_write_policy.attr,
250 ATTRIBUTE_GROUPS(cache);
252 static void node_cache_release(struct device *dev)
257 static void node_cacheinfo_release(struct device *dev)
259 struct node_cache_info *info = to_cache_info(dev);
263 static void node_init_cache_dev(struct node *node)
267 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
271 device_initialize(dev);
272 dev->parent = &node->dev;
273 dev->release = node_cache_release;
274 if (dev_set_name(dev, "memory_side_cache"))
280 pm_runtime_no_callbacks(dev);
281 node->cache_dev = dev;
288 * node_add_cache() - add cache attribute to a memory node
289 * @nid: Node identifier that has new cache attributes
290 * @cache_attrs: Attributes for the cache being added
292 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
294 struct node_cache_info *info;
298 if (!node_online(nid) || !node_devices[nid])
301 node = node_devices[nid];
302 list_for_each_entry(info, &node->cache_attrs, node) {
303 if (info->cache_attrs.level == cache_attrs->level) {
305 "attempt to add duplicate cache level:%d\n",
311 if (!node->cache_dev)
312 node_init_cache_dev(node);
313 if (!node->cache_dev)
316 info = kzalloc(sizeof(*info), GFP_KERNEL);
321 device_initialize(dev);
322 dev->parent = node->cache_dev;
323 dev->release = node_cacheinfo_release;
324 dev->groups = cache_groups;
325 if (dev_set_name(dev, "index%d", cache_attrs->level))
328 info->cache_attrs = *cache_attrs;
329 if (device_add(dev)) {
330 dev_warn(&node->dev, "failed to add cache level:%d\n",
334 pm_runtime_no_callbacks(dev);
335 list_add_tail(&info->node, &node->cache_attrs);
341 static void node_remove_caches(struct node *node)
343 struct node_cache_info *info, *next;
345 if (!node->cache_dev)
348 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
349 list_del(&info->node);
350 device_unregister(&info->dev);
352 device_unregister(node->cache_dev);
355 static void node_init_caches(unsigned int nid)
357 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
360 static void node_init_caches(unsigned int nid) { }
361 static void node_remove_caches(struct node *node) { }
364 #define K(x) ((x) << (PAGE_SHIFT - 10))
365 static ssize_t node_read_meminfo(struct device *dev,
366 struct device_attribute *attr, char *buf)
370 struct pglist_data *pgdat = NODE_DATA(nid);
372 unsigned long sreclaimable, sunreclaimable;
374 si_meminfo_node(&i, nid);
375 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
376 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
377 len = sysfs_emit_at(buf, len,
378 "Node %d MemTotal: %8lu kB\n"
379 "Node %d MemFree: %8lu kB\n"
380 "Node %d MemUsed: %8lu kB\n"
381 "Node %d Active: %8lu kB\n"
382 "Node %d Inactive: %8lu kB\n"
383 "Node %d Active(anon): %8lu kB\n"
384 "Node %d Inactive(anon): %8lu kB\n"
385 "Node %d Active(file): %8lu kB\n"
386 "Node %d Inactive(file): %8lu kB\n"
387 "Node %d Unevictable: %8lu kB\n"
388 "Node %d Mlocked: %8lu kB\n",
391 nid, K(i.totalram - i.freeram),
392 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
393 node_page_state(pgdat, NR_ACTIVE_FILE)),
394 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
395 node_page_state(pgdat, NR_INACTIVE_FILE)),
396 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
397 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
398 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
399 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
400 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
401 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
403 #ifdef CONFIG_HIGHMEM
404 len += sysfs_emit_at(buf, len,
405 "Node %d HighTotal: %8lu kB\n"
406 "Node %d HighFree: %8lu kB\n"
407 "Node %d LowTotal: %8lu kB\n"
408 "Node %d LowFree: %8lu kB\n",
411 nid, K(i.totalram - i.totalhigh),
412 nid, K(i.freeram - i.freehigh));
414 len += sysfs_emit_at(buf, len,
415 "Node %d Dirty: %8lu kB\n"
416 "Node %d Writeback: %8lu kB\n"
417 "Node %d FilePages: %8lu kB\n"
418 "Node %d Mapped: %8lu kB\n"
419 "Node %d AnonPages: %8lu kB\n"
420 "Node %d Shmem: %8lu kB\n"
421 "Node %d KernelStack: %8lu kB\n"
422 #ifdef CONFIG_SHADOW_CALL_STACK
423 "Node %d ShadowCallStack:%8lu kB\n"
425 "Node %d PageTables: %8lu kB\n"
426 "Node %d NFS_Unstable: %8lu kB\n"
427 "Node %d Bounce: %8lu kB\n"
428 "Node %d WritebackTmp: %8lu kB\n"
429 "Node %d KReclaimable: %8lu kB\n"
430 "Node %d Slab: %8lu kB\n"
431 "Node %d SReclaimable: %8lu kB\n"
432 "Node %d SUnreclaim: %8lu kB\n"
433 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
434 "Node %d AnonHugePages: %8lu kB\n"
435 "Node %d ShmemHugePages: %8lu kB\n"
436 "Node %d ShmemPmdMapped: %8lu kB\n"
437 "Node %d FileHugePages: %8lu kB\n"
438 "Node %d FilePmdMapped: %8lu kB\n"
441 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
442 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
443 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
444 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
445 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
447 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
448 #ifdef CONFIG_SHADOW_CALL_STACK
449 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
451 nid, K(sum_zone_node_page_state(nid, NR_PAGETABLE)),
453 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
454 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
455 nid, K(sreclaimable +
456 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
457 nid, K(sreclaimable + sunreclaimable),
458 nid, K(sreclaimable),
459 nid, K(sunreclaimable)
460 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
462 nid, K(node_page_state(pgdat, NR_ANON_THPS) *
464 nid, K(node_page_state(pgdat, NR_SHMEM_THPS) *
466 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
468 nid, K(node_page_state(pgdat, NR_FILE_THPS) *
470 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED) *
474 len += hugetlb_report_node_meminfo(buf, len, nid);
479 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
481 static ssize_t node_read_numastat(struct device *dev,
482 struct device_attribute *attr, char *buf)
484 return sysfs_emit(buf,
488 "interleave_hit %lu\n"
491 sum_zone_numa_state(dev->id, NUMA_HIT),
492 sum_zone_numa_state(dev->id, NUMA_MISS),
493 sum_zone_numa_state(dev->id, NUMA_FOREIGN),
494 sum_zone_numa_state(dev->id, NUMA_INTERLEAVE_HIT),
495 sum_zone_numa_state(dev->id, NUMA_LOCAL),
496 sum_zone_numa_state(dev->id, NUMA_OTHER));
498 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
500 static ssize_t node_read_vmstat(struct device *dev,
501 struct device_attribute *attr, char *buf)
504 struct pglist_data *pgdat = NODE_DATA(nid);
508 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
509 len += sysfs_emit_at(buf, len, "%s %lu\n",
511 sum_zone_node_page_state(nid, i));
514 for (i = 0; i < NR_VM_NUMA_STAT_ITEMS; i++)
515 len += sysfs_emit_at(buf, len, "%s %lu\n",
517 sum_zone_numa_state(nid, i));
520 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
521 len += sysfs_emit_at(buf, len, "%s %lu\n",
523 node_page_state_pages(pgdat, i));
527 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
529 static ssize_t node_read_distance(struct device *dev,
530 struct device_attribute *attr, char *buf)
537 * buf is currently PAGE_SIZE in length and each node needs 4 chars
538 * at the most (distance + space or newline).
540 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
542 for_each_online_node(i) {
543 len += sysfs_emit_at(buf, len, "%s%d",
544 i ? " " : "", node_distance(nid, i));
547 len += sysfs_emit_at(buf, len, "\n");
550 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
552 static struct attribute *node_dev_attrs[] = {
553 &dev_attr_cpumap.attr,
554 &dev_attr_cpulist.attr,
555 &dev_attr_meminfo.attr,
556 &dev_attr_numastat.attr,
557 &dev_attr_distance.attr,
558 &dev_attr_vmstat.attr,
561 ATTRIBUTE_GROUPS(node_dev);
563 #ifdef CONFIG_HUGETLBFS
565 * hugetlbfs per node attributes registration interface:
566 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
567 * it will register its per node attributes for all online nodes with
568 * memory. It will also call register_hugetlbfs_with_node(), below, to
569 * register its attribute registration functions with this node driver.
570 * Once these hooks have been initialized, the node driver will call into
571 * the hugetlb module to [un]register attributes for hot-plugged nodes.
573 static node_registration_func_t __hugetlb_register_node;
574 static node_registration_func_t __hugetlb_unregister_node;
576 static inline bool hugetlb_register_node(struct node *node)
578 if (__hugetlb_register_node &&
579 node_state(node->dev.id, N_MEMORY)) {
580 __hugetlb_register_node(node);
586 static inline void hugetlb_unregister_node(struct node *node)
588 if (__hugetlb_unregister_node)
589 __hugetlb_unregister_node(node);
592 void register_hugetlbfs_with_node(node_registration_func_t doregister,
593 node_registration_func_t unregister)
595 __hugetlb_register_node = doregister;
596 __hugetlb_unregister_node = unregister;
599 static inline void hugetlb_register_node(struct node *node) {}
601 static inline void hugetlb_unregister_node(struct node *node) {}
604 static void node_device_release(struct device *dev)
606 struct node *node = to_node(dev);
608 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
610 * We schedule the work only when a memory section is
611 * onlined/offlined on this node. When we come here,
612 * all the memory on this node has been offlined,
613 * so we won't enqueue new work to this work.
615 * The work is using node->node_work, so we should
616 * flush work before freeing the memory.
618 flush_work(&node->node_work);
624 * register_node - Setup a sysfs device for a node.
625 * @num - Node number to use when creating the device.
627 * Initialize and register the node device.
629 static int register_node(struct node *node, int num)
634 node->dev.bus = &node_subsys;
635 node->dev.release = node_device_release;
636 node->dev.groups = node_dev_groups;
637 error = device_register(&node->dev);
640 put_device(&node->dev);
642 hugetlb_register_node(node);
644 compaction_register_node(node);
650 * unregister_node - unregister a node device
651 * @node: node going away
653 * Unregisters a node device @node. All the devices on the node must be
654 * unregistered before calling this function.
656 void unregister_node(struct node *node)
658 compaction_unregister_node(node);
659 hugetlb_unregister_node(node); /* no-op, if memoryless node */
660 node_remove_accesses(node);
661 node_remove_caches(node);
662 device_unregister(&node->dev);
665 struct node *node_devices[MAX_NUMNODES];
668 * register cpu under node
670 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
675 if (!node_online(nid))
678 obj = get_cpu_device(cpu);
682 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
684 kobject_name(&obj->kobj));
688 return sysfs_create_link(&obj->kobj,
689 &node_devices[nid]->dev.kobj,
690 kobject_name(&node_devices[nid]->dev.kobj));
694 * register_memory_node_under_compute_node - link memory node to its compute
695 * node for a given access class.
696 * @mem_nid: Memory node number
697 * @cpu_nid: Cpu node number
698 * @access: Access class to register
701 * For use with platforms that may have separate memory and compute nodes.
702 * This function will export node relationships linking which memory
703 * initiator nodes can access memory targets at a given ranked access
706 int register_memory_node_under_compute_node(unsigned int mem_nid,
707 unsigned int cpu_nid,
710 struct node *init_node, *targ_node;
711 struct node_access_nodes *initiator, *target;
714 if (!node_online(cpu_nid) || !node_online(mem_nid))
717 init_node = node_devices[cpu_nid];
718 targ_node = node_devices[mem_nid];
719 initiator = node_init_node_access(init_node, access);
720 target = node_init_node_access(targ_node, access);
721 if (!initiator || !target)
724 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
725 &targ_node->dev.kobj,
726 dev_name(&targ_node->dev));
730 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
731 &init_node->dev.kobj,
732 dev_name(&init_node->dev));
738 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
739 dev_name(&targ_node->dev));
743 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
747 if (!node_online(nid))
750 obj = get_cpu_device(cpu);
754 sysfs_remove_link(&node_devices[nid]->dev.kobj,
755 kobject_name(&obj->kobj));
756 sysfs_remove_link(&obj->kobj,
757 kobject_name(&node_devices[nid]->dev.kobj));
762 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
763 static int __ref get_nid_for_pfn(unsigned long pfn)
765 if (!pfn_valid_within(pfn))
767 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
768 if (system_state < SYSTEM_RUNNING)
769 return early_pfn_to_nid(pfn);
771 return pfn_to_nid(pfn);
774 static void do_register_memory_block_under_node(int nid,
775 struct memory_block *mem_blk)
780 * If this memory block spans multiple nodes, we only indicate
781 * the last processed node.
785 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
787 kobject_name(&mem_blk->dev.kobj));
788 if (ret && ret != -EEXIST)
789 dev_err_ratelimited(&node_devices[nid]->dev,
790 "can't create link to %s in sysfs (%d)\n",
791 kobject_name(&mem_blk->dev.kobj), ret);
793 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
794 &node_devices[nid]->dev.kobj,
795 kobject_name(&node_devices[nid]->dev.kobj));
796 if (ret && ret != -EEXIST)
797 dev_err_ratelimited(&mem_blk->dev,
798 "can't create link to %s in sysfs (%d)\n",
799 kobject_name(&node_devices[nid]->dev.kobj),
803 /* register memory section under specified node if it spans that node */
804 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
807 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
808 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
809 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
810 int nid = *(int *)arg;
813 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
817 * memory block could have several absent sections from start.
818 * skip pfn range from absent section
820 if (!pfn_in_present_section(pfn)) {
821 pfn = round_down(pfn + PAGES_PER_SECTION,
822 PAGES_PER_SECTION) - 1;
827 * We need to check if page belongs to nid only at the boot
828 * case because node's ranges can be interleaved.
830 page_nid = get_nid_for_pfn(pfn);
836 do_register_memory_block_under_node(nid, mem_blk);
839 /* mem section does not span the specified node */
844 * During hotplug we know that all pages in the memory block belong to the same
847 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
850 int nid = *(int *)arg;
852 do_register_memory_block_under_node(nid, mem_blk);
857 * Unregister a memory block device under the node it spans. Memory blocks
858 * with multiple nodes cannot be offlined and therefore also never be removed.
860 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
862 if (mem_blk->nid == NUMA_NO_NODE)
865 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
866 kobject_name(&mem_blk->dev.kobj));
867 sysfs_remove_link(&mem_blk->dev.kobj,
868 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
871 void link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn,
872 enum meminit_context context)
874 walk_memory_blocks_func_t func;
876 if (context == MEMINIT_HOTPLUG)
877 func = register_mem_block_under_node_hotplug;
879 func = register_mem_block_under_node_early;
881 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
886 #ifdef CONFIG_HUGETLBFS
888 * Handle per node hstate attribute [un]registration on transistions
889 * to/from memoryless state.
891 static void node_hugetlb_work(struct work_struct *work)
893 struct node *node = container_of(work, struct node, node_work);
896 * We only get here when a node transitions to/from memoryless state.
897 * We can detect which transition occurred by examining whether the
898 * node has memory now. hugetlb_register_node() already check this
899 * so we try to register the attributes. If that fails, then the
900 * node has transitioned to memoryless, try to unregister the
903 if (!hugetlb_register_node(node))
904 hugetlb_unregister_node(node);
907 static void init_node_hugetlb_work(int nid)
909 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
912 static int node_memory_callback(struct notifier_block *self,
913 unsigned long action, void *arg)
915 struct memory_notify *mnb = arg;
916 int nid = mnb->status_change_nid;
922 * offload per node hstate [un]registration to a work thread
923 * when transitioning to/from memoryless state.
925 if (nid != NUMA_NO_NODE)
926 schedule_work(&node_devices[nid]->node_work);
929 case MEM_GOING_ONLINE:
930 case MEM_GOING_OFFLINE:
931 case MEM_CANCEL_ONLINE:
932 case MEM_CANCEL_OFFLINE:
939 #endif /* CONFIG_HUGETLBFS */
940 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
942 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
943 !defined(CONFIG_HUGETLBFS)
944 static inline int node_memory_callback(struct notifier_block *self,
945 unsigned long action, void *arg)
950 static void init_node_hugetlb_work(int nid) { }
954 int __register_one_node(int nid)
959 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
960 if (!node_devices[nid])
963 error = register_node(node_devices[nid], nid);
965 /* link cpu under this node */
966 for_each_present_cpu(cpu) {
967 if (cpu_to_node(cpu) == nid)
968 register_cpu_under_node(cpu, nid);
971 INIT_LIST_HEAD(&node_devices[nid]->access_list);
972 /* initialize work queue for memory hot plug */
973 init_node_hugetlb_work(nid);
974 node_init_caches(nid);
979 void unregister_one_node(int nid)
981 if (!node_devices[nid])
984 unregister_node(node_devices[nid]);
985 node_devices[nid] = NULL;
989 * node states attributes
993 struct device_attribute attr;
994 enum node_states state;
997 static ssize_t show_node_state(struct device *dev,
998 struct device_attribute *attr, char *buf)
1000 struct node_attr *na = container_of(attr, struct node_attr, attr);
1002 return sysfs_emit(buf, "%*pbl\n",
1003 nodemask_pr_args(&node_states[na->state]));
1006 #define _NODE_ATTR(name, state) \
1007 { __ATTR(name, 0444, show_node_state, NULL), state }
1009 static struct node_attr node_state_attr[] = {
1010 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1011 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1012 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1013 #ifdef CONFIG_HIGHMEM
1014 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1016 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1017 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1018 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
1019 N_GENERIC_INITIATOR),
1022 static struct attribute *node_state_attrs[] = {
1023 &node_state_attr[N_POSSIBLE].attr.attr,
1024 &node_state_attr[N_ONLINE].attr.attr,
1025 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
1026 #ifdef CONFIG_HIGHMEM
1027 &node_state_attr[N_HIGH_MEMORY].attr.attr,
1029 &node_state_attr[N_MEMORY].attr.attr,
1030 &node_state_attr[N_CPU].attr.attr,
1031 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
1035 static struct attribute_group memory_root_attr_group = {
1036 .attrs = node_state_attrs,
1039 static const struct attribute_group *cpu_root_attr_groups[] = {
1040 &memory_root_attr_group,
1044 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
1045 static int __init register_node_type(void)
1049 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1050 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1052 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1054 static struct notifier_block node_memory_callback_nb = {
1055 .notifier_call = node_memory_callback,
1056 .priority = NODE_CALLBACK_PRI,
1058 register_hotmemory_notifier(&node_memory_callback_nb);
1062 * Note: we're not going to unregister the node class if we fail
1063 * to register the node state class attribute files.
1067 postcore_initcall(register_node_type);