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 = {
29 static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
30 struct bin_attribute *attr, char *buf,
31 loff_t off, size_t count)
33 struct device *dev = kobj_to_dev(kobj);
34 struct node *node_dev = to_node(dev);
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_bitmask_to_buf(buf, mask, off, count);
43 free_cpumask_var(mask);
48 static BIN_ATTR_RO(cpumap, CPUMAP_FILE_MAX_BYTES);
50 static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
51 struct bin_attribute *attr, char *buf,
52 loff_t off, size_t count)
54 struct device *dev = kobj_to_dev(kobj);
55 struct node *node_dev = to_node(dev);
59 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
62 cpumask_and(mask, cpumask_of_node(node_dev->dev.id), cpu_online_mask);
63 n = cpumap_print_list_to_buf(buf, mask, off, count);
64 free_cpumask_var(mask);
69 static BIN_ATTR_RO(cpulist, CPULIST_FILE_MAX_BYTES);
72 * struct node_access_nodes - Access class device to hold user visible
73 * relationships to other nodes.
74 * @dev: Device for this memory access class
75 * @list_node: List element in the node's access list
76 * @access: The access class rank
77 * @hmem_attrs: Heterogeneous memory performance attributes
79 struct node_access_nodes {
81 struct list_head list_node;
83 #ifdef CONFIG_HMEM_REPORTING
84 struct node_hmem_attrs hmem_attrs;
87 #define to_access_nodes(dev) container_of(dev, struct node_access_nodes, dev)
89 static struct attribute *node_init_access_node_attrs[] = {
93 static struct attribute *node_targ_access_node_attrs[] = {
97 static const struct attribute_group initiators = {
99 .attrs = node_init_access_node_attrs,
102 static const struct attribute_group targets = {
104 .attrs = node_targ_access_node_attrs,
107 static const struct attribute_group *node_access_node_groups[] = {
113 static void node_remove_accesses(struct node *node)
115 struct node_access_nodes *c, *cnext;
117 list_for_each_entry_safe(c, cnext, &node->access_list, list_node) {
118 list_del(&c->list_node);
119 device_unregister(&c->dev);
123 static void node_access_release(struct device *dev)
125 kfree(to_access_nodes(dev));
128 static struct node_access_nodes *node_init_node_access(struct node *node,
131 struct node_access_nodes *access_node;
134 list_for_each_entry(access_node, &node->access_list, list_node)
135 if (access_node->access == access)
138 access_node = kzalloc(sizeof(*access_node), GFP_KERNEL);
142 access_node->access = access;
143 dev = &access_node->dev;
144 dev->parent = &node->dev;
145 dev->release = node_access_release;
146 dev->groups = node_access_node_groups;
147 if (dev_set_name(dev, "access%u", access))
150 if (device_register(dev))
153 pm_runtime_no_callbacks(dev);
154 list_add_tail(&access_node->list_node, &node->access_list);
157 kfree_const(dev->kobj.name);
163 #ifdef CONFIG_HMEM_REPORTING
164 #define ACCESS_ATTR(name) \
165 static ssize_t name##_show(struct device *dev, \
166 struct device_attribute *attr, \
169 return sysfs_emit(buf, "%u\n", \
170 to_access_nodes(dev)->hmem_attrs.name); \
172 static DEVICE_ATTR_RO(name)
174 ACCESS_ATTR(read_bandwidth);
175 ACCESS_ATTR(read_latency);
176 ACCESS_ATTR(write_bandwidth);
177 ACCESS_ATTR(write_latency);
179 static struct attribute *access_attrs[] = {
180 &dev_attr_read_bandwidth.attr,
181 &dev_attr_read_latency.attr,
182 &dev_attr_write_bandwidth.attr,
183 &dev_attr_write_latency.attr,
188 * node_set_perf_attrs - Set the performance values for given access class
189 * @nid: Node identifier to be set
190 * @hmem_attrs: Heterogeneous memory performance attributes
191 * @access: The access class the for the given attributes
193 void node_set_perf_attrs(unsigned int nid, struct node_hmem_attrs *hmem_attrs,
196 struct node_access_nodes *c;
200 if (WARN_ON_ONCE(!node_online(nid)))
203 node = node_devices[nid];
204 c = node_init_node_access(node, access);
208 c->hmem_attrs = *hmem_attrs;
209 for (i = 0; access_attrs[i] != NULL; i++) {
210 if (sysfs_add_file_to_group(&c->dev.kobj, access_attrs[i],
212 pr_info("failed to add performance attribute to node %d\n",
220 * struct node_cache_info - Internal tracking for memory node caches
221 * @dev: Device represeting the cache level
222 * @node: List element for tracking in the node
223 * @cache_attrs:Attributes for this cache level
225 struct node_cache_info {
227 struct list_head node;
228 struct node_cache_attrs cache_attrs;
230 #define to_cache_info(device) container_of(device, struct node_cache_info, dev)
232 #define CACHE_ATTR(name, fmt) \
233 static ssize_t name##_show(struct device *dev, \
234 struct device_attribute *attr, \
237 return sysfs_emit(buf, fmt "\n", \
238 to_cache_info(dev)->cache_attrs.name); \
240 static DEVICE_ATTR_RO(name);
242 CACHE_ATTR(size, "%llu")
243 CACHE_ATTR(line_size, "%u")
244 CACHE_ATTR(indexing, "%u")
245 CACHE_ATTR(write_policy, "%u")
247 static struct attribute *cache_attrs[] = {
248 &dev_attr_indexing.attr,
250 &dev_attr_line_size.attr,
251 &dev_attr_write_policy.attr,
254 ATTRIBUTE_GROUPS(cache);
256 static void node_cache_release(struct device *dev)
261 static void node_cacheinfo_release(struct device *dev)
263 struct node_cache_info *info = to_cache_info(dev);
267 static void node_init_cache_dev(struct node *node)
271 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
275 device_initialize(dev);
276 dev->parent = &node->dev;
277 dev->release = node_cache_release;
278 if (dev_set_name(dev, "memory_side_cache"))
284 pm_runtime_no_callbacks(dev);
285 node->cache_dev = dev;
292 * node_add_cache() - add cache attribute to a memory node
293 * @nid: Node identifier that has new cache attributes
294 * @cache_attrs: Attributes for the cache being added
296 void node_add_cache(unsigned int nid, struct node_cache_attrs *cache_attrs)
298 struct node_cache_info *info;
302 if (!node_online(nid) || !node_devices[nid])
305 node = node_devices[nid];
306 list_for_each_entry(info, &node->cache_attrs, node) {
307 if (info->cache_attrs.level == cache_attrs->level) {
309 "attempt to add duplicate cache level:%d\n",
315 if (!node->cache_dev)
316 node_init_cache_dev(node);
317 if (!node->cache_dev)
320 info = kzalloc(sizeof(*info), GFP_KERNEL);
325 device_initialize(dev);
326 dev->parent = node->cache_dev;
327 dev->release = node_cacheinfo_release;
328 dev->groups = cache_groups;
329 if (dev_set_name(dev, "index%d", cache_attrs->level))
332 info->cache_attrs = *cache_attrs;
333 if (device_add(dev)) {
334 dev_warn(&node->dev, "failed to add cache level:%d\n",
338 pm_runtime_no_callbacks(dev);
339 list_add_tail(&info->node, &node->cache_attrs);
345 static void node_remove_caches(struct node *node)
347 struct node_cache_info *info, *next;
349 if (!node->cache_dev)
352 list_for_each_entry_safe(info, next, &node->cache_attrs, node) {
353 list_del(&info->node);
354 device_unregister(&info->dev);
356 device_unregister(node->cache_dev);
359 static void node_init_caches(unsigned int nid)
361 INIT_LIST_HEAD(&node_devices[nid]->cache_attrs);
364 static void node_init_caches(unsigned int nid) { }
365 static void node_remove_caches(struct node *node) { }
368 #define K(x) ((x) << (PAGE_SHIFT - 10))
369 static ssize_t node_read_meminfo(struct device *dev,
370 struct device_attribute *attr, char *buf)
374 struct pglist_data *pgdat = NODE_DATA(nid);
376 unsigned long sreclaimable, sunreclaimable;
377 unsigned long swapcached = 0;
379 si_meminfo_node(&i, nid);
380 sreclaimable = node_page_state_pages(pgdat, NR_SLAB_RECLAIMABLE_B);
381 sunreclaimable = node_page_state_pages(pgdat, NR_SLAB_UNRECLAIMABLE_B);
383 swapcached = node_page_state_pages(pgdat, NR_SWAPCACHE);
385 len = sysfs_emit_at(buf, len,
386 "Node %d MemTotal: %8lu kB\n"
387 "Node %d MemFree: %8lu kB\n"
388 "Node %d MemUsed: %8lu kB\n"
389 "Node %d SwapCached: %8lu kB\n"
390 "Node %d Active: %8lu kB\n"
391 "Node %d Inactive: %8lu kB\n"
392 "Node %d Active(anon): %8lu kB\n"
393 "Node %d Inactive(anon): %8lu kB\n"
394 "Node %d Active(file): %8lu kB\n"
395 "Node %d Inactive(file): %8lu kB\n"
396 "Node %d Unevictable: %8lu kB\n"
397 "Node %d Mlocked: %8lu kB\n",
400 nid, K(i.totalram - i.freeram),
402 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON) +
403 node_page_state(pgdat, NR_ACTIVE_FILE)),
404 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON) +
405 node_page_state(pgdat, NR_INACTIVE_FILE)),
406 nid, K(node_page_state(pgdat, NR_ACTIVE_ANON)),
407 nid, K(node_page_state(pgdat, NR_INACTIVE_ANON)),
408 nid, K(node_page_state(pgdat, NR_ACTIVE_FILE)),
409 nid, K(node_page_state(pgdat, NR_INACTIVE_FILE)),
410 nid, K(node_page_state(pgdat, NR_UNEVICTABLE)),
411 nid, K(sum_zone_node_page_state(nid, NR_MLOCK)));
413 #ifdef CONFIG_HIGHMEM
414 len += sysfs_emit_at(buf, len,
415 "Node %d HighTotal: %8lu kB\n"
416 "Node %d HighFree: %8lu kB\n"
417 "Node %d LowTotal: %8lu kB\n"
418 "Node %d LowFree: %8lu kB\n",
421 nid, K(i.totalram - i.totalhigh),
422 nid, K(i.freeram - i.freehigh));
424 len += sysfs_emit_at(buf, len,
425 "Node %d Dirty: %8lu kB\n"
426 "Node %d Writeback: %8lu kB\n"
427 "Node %d FilePages: %8lu kB\n"
428 "Node %d Mapped: %8lu kB\n"
429 "Node %d AnonPages: %8lu kB\n"
430 "Node %d Shmem: %8lu kB\n"
431 "Node %d KernelStack: %8lu kB\n"
432 #ifdef CONFIG_SHADOW_CALL_STACK
433 "Node %d ShadowCallStack:%8lu kB\n"
435 "Node %d PageTables: %8lu kB\n"
436 "Node %d NFS_Unstable: %8lu kB\n"
437 "Node %d Bounce: %8lu kB\n"
438 "Node %d WritebackTmp: %8lu kB\n"
439 "Node %d KReclaimable: %8lu kB\n"
440 "Node %d Slab: %8lu kB\n"
441 "Node %d SReclaimable: %8lu kB\n"
442 "Node %d SUnreclaim: %8lu kB\n"
443 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
444 "Node %d AnonHugePages: %8lu kB\n"
445 "Node %d ShmemHugePages: %8lu kB\n"
446 "Node %d ShmemPmdMapped: %8lu kB\n"
447 "Node %d FileHugePages: %8lu kB\n"
448 "Node %d FilePmdMapped: %8lu kB\n"
451 nid, K(node_page_state(pgdat, NR_FILE_DIRTY)),
452 nid, K(node_page_state(pgdat, NR_WRITEBACK)),
453 nid, K(node_page_state(pgdat, NR_FILE_PAGES)),
454 nid, K(node_page_state(pgdat, NR_FILE_MAPPED)),
455 nid, K(node_page_state(pgdat, NR_ANON_MAPPED)),
457 nid, node_page_state(pgdat, NR_KERNEL_STACK_KB),
458 #ifdef CONFIG_SHADOW_CALL_STACK
459 nid, node_page_state(pgdat, NR_KERNEL_SCS_KB),
461 nid, K(node_page_state(pgdat, NR_PAGETABLE)),
463 nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
464 nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
465 nid, K(sreclaimable +
466 node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
467 nid, K(sreclaimable + sunreclaimable),
468 nid, K(sreclaimable),
469 nid, K(sunreclaimable)
470 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 nid, K(node_page_state(pgdat, NR_ANON_THPS)),
473 nid, K(node_page_state(pgdat, NR_SHMEM_THPS)),
474 nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
475 nid, K(node_page_state(pgdat, NR_FILE_THPS)),
476 nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED))
479 len += hugetlb_report_node_meminfo(buf, len, nid);
484 static DEVICE_ATTR(meminfo, 0444, node_read_meminfo, NULL);
486 static ssize_t node_read_numastat(struct device *dev,
487 struct device_attribute *attr, char *buf)
489 fold_vm_numa_events();
490 return sysfs_emit(buf,
494 "interleave_hit %lu\n"
497 sum_zone_numa_event_state(dev->id, NUMA_HIT),
498 sum_zone_numa_event_state(dev->id, NUMA_MISS),
499 sum_zone_numa_event_state(dev->id, NUMA_FOREIGN),
500 sum_zone_numa_event_state(dev->id, NUMA_INTERLEAVE_HIT),
501 sum_zone_numa_event_state(dev->id, NUMA_LOCAL),
502 sum_zone_numa_event_state(dev->id, NUMA_OTHER));
504 static DEVICE_ATTR(numastat, 0444, node_read_numastat, NULL);
506 static ssize_t node_read_vmstat(struct device *dev,
507 struct device_attribute *attr, char *buf)
510 struct pglist_data *pgdat = NODE_DATA(nid);
514 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
515 len += sysfs_emit_at(buf, len, "%s %lu\n",
517 sum_zone_node_page_state(nid, i));
520 fold_vm_numa_events();
521 for (i = 0; i < NR_VM_NUMA_EVENT_ITEMS; i++)
522 len += sysfs_emit_at(buf, len, "%s %lu\n",
524 sum_zone_numa_event_state(nid, i));
527 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
528 unsigned long pages = node_page_state_pages(pgdat, i);
530 if (vmstat_item_print_in_thp(i))
531 pages /= HPAGE_PMD_NR;
532 len += sysfs_emit_at(buf, len, "%s %lu\n", node_stat_name(i),
538 static DEVICE_ATTR(vmstat, 0444, node_read_vmstat, NULL);
540 static ssize_t node_read_distance(struct device *dev,
541 struct device_attribute *attr, char *buf)
548 * buf is currently PAGE_SIZE in length and each node needs 4 chars
549 * at the most (distance + space or newline).
551 BUILD_BUG_ON(MAX_NUMNODES * 4 > PAGE_SIZE);
553 for_each_online_node(i) {
554 len += sysfs_emit_at(buf, len, "%s%d",
555 i ? " " : "", node_distance(nid, i));
558 len += sysfs_emit_at(buf, len, "\n");
561 static DEVICE_ATTR(distance, 0444, node_read_distance, NULL);
563 static struct attribute *node_dev_attrs[] = {
564 &dev_attr_meminfo.attr,
565 &dev_attr_numastat.attr,
566 &dev_attr_distance.attr,
567 &dev_attr_vmstat.attr,
571 static struct bin_attribute *node_dev_bin_attrs[] = {
577 static const struct attribute_group node_dev_group = {
578 .attrs = node_dev_attrs,
579 .bin_attrs = node_dev_bin_attrs
582 static const struct attribute_group *node_dev_groups[] = {
587 #ifdef CONFIG_HUGETLBFS
589 * hugetlbfs per node attributes registration interface:
590 * When/if hugetlb[fs] subsystem initializes [sometime after this module],
591 * it will register its per node attributes for all online nodes with
592 * memory. It will also call register_hugetlbfs_with_node(), below, to
593 * register its attribute registration functions with this node driver.
594 * Once these hooks have been initialized, the node driver will call into
595 * the hugetlb module to [un]register attributes for hot-plugged nodes.
597 static node_registration_func_t __hugetlb_register_node;
598 static node_registration_func_t __hugetlb_unregister_node;
600 static inline bool hugetlb_register_node(struct node *node)
602 if (__hugetlb_register_node &&
603 node_state(node->dev.id, N_MEMORY)) {
604 __hugetlb_register_node(node);
610 static inline void hugetlb_unregister_node(struct node *node)
612 if (__hugetlb_unregister_node)
613 __hugetlb_unregister_node(node);
616 void register_hugetlbfs_with_node(node_registration_func_t doregister,
617 node_registration_func_t unregister)
619 __hugetlb_register_node = doregister;
620 __hugetlb_unregister_node = unregister;
623 static inline void hugetlb_register_node(struct node *node) {}
625 static inline void hugetlb_unregister_node(struct node *node) {}
628 static void node_device_release(struct device *dev)
630 struct node *node = to_node(dev);
632 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HUGETLBFS)
634 * We schedule the work only when a memory section is
635 * onlined/offlined on this node. When we come here,
636 * all the memory on this node has been offlined,
637 * so we won't enqueue new work to this work.
639 * The work is using node->node_work, so we should
640 * flush work before freeing the memory.
642 flush_work(&node->node_work);
648 * register_node - Setup a sysfs device for a node.
649 * @num - Node number to use when creating the device.
651 * Initialize and register the node device.
653 static int register_node(struct node *node, int num)
658 node->dev.bus = &node_subsys;
659 node->dev.release = node_device_release;
660 node->dev.groups = node_dev_groups;
661 error = device_register(&node->dev);
664 put_device(&node->dev);
666 hugetlb_register_node(node);
668 compaction_register_node(node);
674 * unregister_node - unregister a node device
675 * @node: node going away
677 * Unregisters a node device @node. All the devices on the node must be
678 * unregistered before calling this function.
680 void unregister_node(struct node *node)
682 compaction_unregister_node(node);
683 hugetlb_unregister_node(node); /* no-op, if memoryless node */
684 node_remove_accesses(node);
685 node_remove_caches(node);
686 device_unregister(&node->dev);
689 struct node *node_devices[MAX_NUMNODES];
692 * register cpu under node
694 int register_cpu_under_node(unsigned int cpu, unsigned int nid)
699 if (!node_online(nid))
702 obj = get_cpu_device(cpu);
706 ret = sysfs_create_link(&node_devices[nid]->dev.kobj,
708 kobject_name(&obj->kobj));
712 return sysfs_create_link(&obj->kobj,
713 &node_devices[nid]->dev.kobj,
714 kobject_name(&node_devices[nid]->dev.kobj));
718 * register_memory_node_under_compute_node - link memory node to its compute
719 * node for a given access class.
720 * @mem_nid: Memory node number
721 * @cpu_nid: Cpu node number
722 * @access: Access class to register
725 * For use with platforms that may have separate memory and compute nodes.
726 * This function will export node relationships linking which memory
727 * initiator nodes can access memory targets at a given ranked access
730 int register_memory_node_under_compute_node(unsigned int mem_nid,
731 unsigned int cpu_nid,
734 struct node *init_node, *targ_node;
735 struct node_access_nodes *initiator, *target;
738 if (!node_online(cpu_nid) || !node_online(mem_nid))
741 init_node = node_devices[cpu_nid];
742 targ_node = node_devices[mem_nid];
743 initiator = node_init_node_access(init_node, access);
744 target = node_init_node_access(targ_node, access);
745 if (!initiator || !target)
748 ret = sysfs_add_link_to_group(&initiator->dev.kobj, "targets",
749 &targ_node->dev.kobj,
750 dev_name(&targ_node->dev));
754 ret = sysfs_add_link_to_group(&target->dev.kobj, "initiators",
755 &init_node->dev.kobj,
756 dev_name(&init_node->dev));
762 sysfs_remove_link_from_group(&initiator->dev.kobj, "targets",
763 dev_name(&targ_node->dev));
767 int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
771 if (!node_online(nid))
774 obj = get_cpu_device(cpu);
778 sysfs_remove_link(&node_devices[nid]->dev.kobj,
779 kobject_name(&obj->kobj));
780 sysfs_remove_link(&obj->kobj,
781 kobject_name(&node_devices[nid]->dev.kobj));
786 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
787 static int __ref get_nid_for_pfn(unsigned long pfn)
789 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
790 if (system_state < SYSTEM_RUNNING)
791 return early_pfn_to_nid(pfn);
793 return pfn_to_nid(pfn);
796 static void do_register_memory_block_under_node(int nid,
797 struct memory_block *mem_blk)
802 * If this memory block spans multiple nodes, we only indicate
803 * the last processed node.
807 ret = sysfs_create_link_nowarn(&node_devices[nid]->dev.kobj,
809 kobject_name(&mem_blk->dev.kobj));
810 if (ret && ret != -EEXIST)
811 dev_err_ratelimited(&node_devices[nid]->dev,
812 "can't create link to %s in sysfs (%d)\n",
813 kobject_name(&mem_blk->dev.kobj), ret);
815 ret = sysfs_create_link_nowarn(&mem_blk->dev.kobj,
816 &node_devices[nid]->dev.kobj,
817 kobject_name(&node_devices[nid]->dev.kobj));
818 if (ret && ret != -EEXIST)
819 dev_err_ratelimited(&mem_blk->dev,
820 "can't create link to %s in sysfs (%d)\n",
821 kobject_name(&node_devices[nid]->dev.kobj),
825 /* register memory section under specified node if it spans that node */
826 static int register_mem_block_under_node_early(struct memory_block *mem_blk,
829 unsigned long memory_block_pfns = memory_block_size_bytes() / PAGE_SIZE;
830 unsigned long start_pfn = section_nr_to_pfn(mem_blk->start_section_nr);
831 unsigned long end_pfn = start_pfn + memory_block_pfns - 1;
832 int nid = *(int *)arg;
835 for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
839 * memory block could have several absent sections from start.
840 * skip pfn range from absent section
842 if (!pfn_in_present_section(pfn)) {
843 pfn = round_down(pfn + PAGES_PER_SECTION,
844 PAGES_PER_SECTION) - 1;
849 * We need to check if page belongs to nid only at the boot
850 * case because node's ranges can be interleaved.
852 page_nid = get_nid_for_pfn(pfn);
858 do_register_memory_block_under_node(nid, mem_blk);
861 /* mem section does not span the specified node */
866 * During hotplug we know that all pages in the memory block belong to the same
869 static int register_mem_block_under_node_hotplug(struct memory_block *mem_blk,
872 int nid = *(int *)arg;
874 do_register_memory_block_under_node(nid, mem_blk);
879 * Unregister a memory block device under the node it spans. Memory blocks
880 * with multiple nodes cannot be offlined and therefore also never be removed.
882 void unregister_memory_block_under_nodes(struct memory_block *mem_blk)
884 if (mem_blk->nid == NUMA_NO_NODE)
887 sysfs_remove_link(&node_devices[mem_blk->nid]->dev.kobj,
888 kobject_name(&mem_blk->dev.kobj));
889 sysfs_remove_link(&mem_blk->dev.kobj,
890 kobject_name(&node_devices[mem_blk->nid]->dev.kobj));
893 void link_mem_sections(int nid, unsigned long start_pfn, unsigned long end_pfn,
894 enum meminit_context context)
896 walk_memory_blocks_func_t func;
898 if (context == MEMINIT_HOTPLUG)
899 func = register_mem_block_under_node_hotplug;
901 func = register_mem_block_under_node_early;
903 walk_memory_blocks(PFN_PHYS(start_pfn), PFN_PHYS(end_pfn - start_pfn),
908 #ifdef CONFIG_HUGETLBFS
910 * Handle per node hstate attribute [un]registration on transistions
911 * to/from memoryless state.
913 static void node_hugetlb_work(struct work_struct *work)
915 struct node *node = container_of(work, struct node, node_work);
918 * We only get here when a node transitions to/from memoryless state.
919 * We can detect which transition occurred by examining whether the
920 * node has memory now. hugetlb_register_node() already check this
921 * so we try to register the attributes. If that fails, then the
922 * node has transitioned to memoryless, try to unregister the
925 if (!hugetlb_register_node(node))
926 hugetlb_unregister_node(node);
929 static void init_node_hugetlb_work(int nid)
931 INIT_WORK(&node_devices[nid]->node_work, node_hugetlb_work);
934 static int node_memory_callback(struct notifier_block *self,
935 unsigned long action, void *arg)
937 struct memory_notify *mnb = arg;
938 int nid = mnb->status_change_nid;
944 * offload per node hstate [un]registration to a work thread
945 * when transitioning to/from memoryless state.
947 if (nid != NUMA_NO_NODE)
948 schedule_work(&node_devices[nid]->node_work);
951 case MEM_GOING_ONLINE:
952 case MEM_GOING_OFFLINE:
953 case MEM_CANCEL_ONLINE:
954 case MEM_CANCEL_OFFLINE:
961 #endif /* CONFIG_HUGETLBFS */
962 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
964 #if !defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || \
965 !defined(CONFIG_HUGETLBFS)
966 static inline int node_memory_callback(struct notifier_block *self,
967 unsigned long action, void *arg)
972 static void init_node_hugetlb_work(int nid) { }
976 int __register_one_node(int nid)
981 node_devices[nid] = kzalloc(sizeof(struct node), GFP_KERNEL);
982 if (!node_devices[nid])
985 error = register_node(node_devices[nid], nid);
987 /* link cpu under this node */
988 for_each_present_cpu(cpu) {
989 if (cpu_to_node(cpu) == nid)
990 register_cpu_under_node(cpu, nid);
993 INIT_LIST_HEAD(&node_devices[nid]->access_list);
994 /* initialize work queue for memory hot plug */
995 init_node_hugetlb_work(nid);
996 node_init_caches(nid);
1001 void unregister_one_node(int nid)
1003 if (!node_devices[nid])
1006 unregister_node(node_devices[nid]);
1007 node_devices[nid] = NULL;
1011 * node states attributes
1015 struct device_attribute attr;
1016 enum node_states state;
1019 static ssize_t show_node_state(struct device *dev,
1020 struct device_attribute *attr, char *buf)
1022 struct node_attr *na = container_of(attr, struct node_attr, attr);
1024 return sysfs_emit(buf, "%*pbl\n",
1025 nodemask_pr_args(&node_states[na->state]));
1028 #define _NODE_ATTR(name, state) \
1029 { __ATTR(name, 0444, show_node_state, NULL), state }
1031 static struct node_attr node_state_attr[] = {
1032 [N_POSSIBLE] = _NODE_ATTR(possible, N_POSSIBLE),
1033 [N_ONLINE] = _NODE_ATTR(online, N_ONLINE),
1034 [N_NORMAL_MEMORY] = _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
1035 #ifdef CONFIG_HIGHMEM
1036 [N_HIGH_MEMORY] = _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
1038 [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY),
1039 [N_CPU] = _NODE_ATTR(has_cpu, N_CPU),
1040 [N_GENERIC_INITIATOR] = _NODE_ATTR(has_generic_initiator,
1041 N_GENERIC_INITIATOR),
1044 static struct attribute *node_state_attrs[] = {
1045 &node_state_attr[N_POSSIBLE].attr.attr,
1046 &node_state_attr[N_ONLINE].attr.attr,
1047 &node_state_attr[N_NORMAL_MEMORY].attr.attr,
1048 #ifdef CONFIG_HIGHMEM
1049 &node_state_attr[N_HIGH_MEMORY].attr.attr,
1051 &node_state_attr[N_MEMORY].attr.attr,
1052 &node_state_attr[N_CPU].attr.attr,
1053 &node_state_attr[N_GENERIC_INITIATOR].attr.attr,
1057 static const struct attribute_group memory_root_attr_group = {
1058 .attrs = node_state_attrs,
1061 static const struct attribute_group *cpu_root_attr_groups[] = {
1062 &memory_root_attr_group,
1066 #define NODE_CALLBACK_PRI 2 /* lower than SLAB */
1067 static int __init register_node_type(void)
1071 BUILD_BUG_ON(ARRAY_SIZE(node_state_attr) != NR_NODE_STATES);
1072 BUILD_BUG_ON(ARRAY_SIZE(node_state_attrs)-1 != NR_NODE_STATES);
1074 ret = subsys_system_register(&node_subsys, cpu_root_attr_groups);
1076 static struct notifier_block node_memory_callback_nb = {
1077 .notifier_call = node_memory_callback,
1078 .priority = NODE_CALLBACK_PRI,
1080 register_hotmemory_notifier(&node_memory_callback_nb);
1084 * Note: we're not going to unregister the node class if we fail
1085 * to register the node state class attribute files.
1089 postcore_initcall(register_node_type);