1 // SPDX-License-Identifier: GPL-2.0
3 * Memory subsystem support
5 * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
6 * Dave Hansen <haveblue@us.ibm.com>
8 * This file provides the necessary infrastructure to represent
9 * a SPARSEMEM-memory-model system's physical memory in /sysfs.
10 * All arch-independent code that assumes MEMORY_HOTPLUG requires
11 * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
14 #include <linux/module.h>
15 #include <linux/init.h>
16 #include <linux/topology.h>
17 #include <linux/capability.h>
18 #include <linux/device.h>
19 #include <linux/memory.h>
20 #include <linux/memory_hotplug.h>
22 #include <linux/stat.h>
23 #include <linux/slab.h>
24 #include <linux/xarray.h>
26 #include <linux/atomic.h>
27 #include <linux/uaccess.h>
29 #define MEMORY_CLASS_NAME "memory"
31 static const char *const online_type_to_str[] = {
32 [MMOP_OFFLINE] = "offline",
33 [MMOP_ONLINE] = "online",
34 [MMOP_ONLINE_KERNEL] = "online_kernel",
35 [MMOP_ONLINE_MOVABLE] = "online_movable",
38 int mhp_online_type_from_str(const char *str)
42 for (i = 0; i < ARRAY_SIZE(online_type_to_str); i++) {
43 if (sysfs_streq(str, online_type_to_str[i]))
49 #define to_memory_block(dev) container_of(dev, struct memory_block, dev)
51 static int sections_per_block;
53 static inline unsigned long memory_block_id(unsigned long section_nr)
55 return section_nr / sections_per_block;
58 static inline unsigned long pfn_to_block_id(unsigned long pfn)
60 return memory_block_id(pfn_to_section_nr(pfn));
63 static inline unsigned long phys_to_block_id(unsigned long phys)
65 return pfn_to_block_id(PFN_DOWN(phys));
68 static int memory_subsys_online(struct device *dev);
69 static int memory_subsys_offline(struct device *dev);
71 static struct bus_type memory_subsys = {
72 .name = MEMORY_CLASS_NAME,
73 .dev_name = MEMORY_CLASS_NAME,
74 .online = memory_subsys_online,
75 .offline = memory_subsys_offline,
79 * Memory blocks are cached in a local radix tree to avoid
80 * a costly linear search for the corresponding device on
83 static DEFINE_XARRAY(memory_blocks);
86 * Memory groups, indexed by memory group id (mgid).
88 static DEFINE_XARRAY_FLAGS(memory_groups, XA_FLAGS_ALLOC);
89 #define MEMORY_GROUP_MARK_DYNAMIC XA_MARK_1
91 static BLOCKING_NOTIFIER_HEAD(memory_chain);
93 int register_memory_notifier(struct notifier_block *nb)
95 return blocking_notifier_chain_register(&memory_chain, nb);
97 EXPORT_SYMBOL(register_memory_notifier);
99 void unregister_memory_notifier(struct notifier_block *nb)
101 blocking_notifier_chain_unregister(&memory_chain, nb);
103 EXPORT_SYMBOL(unregister_memory_notifier);
105 static void memory_block_release(struct device *dev)
107 struct memory_block *mem = to_memory_block(dev);
112 unsigned long __weak memory_block_size_bytes(void)
114 return MIN_MEMORY_BLOCK_SIZE;
116 EXPORT_SYMBOL_GPL(memory_block_size_bytes);
119 * Show the first physical section index (number) of this memory block.
121 static ssize_t phys_index_show(struct device *dev,
122 struct device_attribute *attr, char *buf)
124 struct memory_block *mem = to_memory_block(dev);
125 unsigned long phys_index;
127 phys_index = mem->start_section_nr / sections_per_block;
129 return sysfs_emit(buf, "%08lx\n", phys_index);
133 * Legacy interface that we cannot remove. Always indicate "removable"
134 * with CONFIG_MEMORY_HOTREMOVE - bad heuristic.
136 static ssize_t removable_show(struct device *dev, struct device_attribute *attr,
139 return sysfs_emit(buf, "%d\n", (int)IS_ENABLED(CONFIG_MEMORY_HOTREMOVE));
143 * online, offline, going offline, etc.
145 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
148 struct memory_block *mem = to_memory_block(dev);
152 * We can probably put these states in a nice little array
153 * so that they're not open-coded
155 switch (mem->state) {
162 case MEM_GOING_OFFLINE:
163 output = "going-offline";
167 return sysfs_emit(buf, "ERROR-UNKNOWN-%ld\n", mem->state);
170 return sysfs_emit(buf, "%s\n", output);
173 int memory_notify(unsigned long val, void *v)
175 return blocking_notifier_call_chain(&memory_chain, val, v);
178 static int memory_block_online(struct memory_block *mem)
180 unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
181 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
182 unsigned long nr_vmemmap_pages = mem->nr_vmemmap_pages;
186 zone = zone_for_pfn_range(mem->online_type, mem->nid, mem->group,
187 start_pfn, nr_pages);
190 * Although vmemmap pages have a different lifecycle than the pages
191 * they describe (they remain until the memory is unplugged), doing
192 * their initialization and accounting at memory onlining/offlining
193 * stage helps to keep accounting easier to follow - e.g vmemmaps
194 * belong to the same zone as the memory they backed.
196 if (nr_vmemmap_pages) {
197 ret = mhp_init_memmap_on_memory(start_pfn, nr_vmemmap_pages, zone);
202 ret = online_pages(start_pfn + nr_vmemmap_pages,
203 nr_pages - nr_vmemmap_pages, zone, mem->group);
205 if (nr_vmemmap_pages)
206 mhp_deinit_memmap_on_memory(start_pfn, nr_vmemmap_pages);
211 * Account once onlining succeeded. If the zone was unpopulated, it is
212 * now already properly populated.
214 if (nr_vmemmap_pages)
215 adjust_present_page_count(pfn_to_page(start_pfn), mem->group,
221 static int memory_block_offline(struct memory_block *mem)
223 unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
224 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
225 unsigned long nr_vmemmap_pages = mem->nr_vmemmap_pages;
229 * Unaccount before offlining, such that unpopulated zone and kthreads
230 * can properly be torn down in offline_pages().
232 if (nr_vmemmap_pages)
233 adjust_present_page_count(pfn_to_page(start_pfn), mem->group,
236 ret = offline_pages(start_pfn + nr_vmemmap_pages,
237 nr_pages - nr_vmemmap_pages, mem->group);
239 /* offline_pages() failed. Account back. */
240 if (nr_vmemmap_pages)
241 adjust_present_page_count(pfn_to_page(start_pfn),
242 mem->group, nr_vmemmap_pages);
246 if (nr_vmemmap_pages)
247 mhp_deinit_memmap_on_memory(start_pfn, nr_vmemmap_pages);
253 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
254 * OK to have direct references to sparsemem variables in here.
257 memory_block_action(struct memory_block *mem, unsigned long action)
263 ret = memory_block_online(mem);
266 ret = memory_block_offline(mem);
269 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
270 "%ld\n", __func__, mem->start_section_nr, action, action);
277 static int memory_block_change_state(struct memory_block *mem,
278 unsigned long to_state, unsigned long from_state_req)
282 if (mem->state != from_state_req)
285 if (to_state == MEM_OFFLINE)
286 mem->state = MEM_GOING_OFFLINE;
288 ret = memory_block_action(mem, to_state);
289 mem->state = ret ? from_state_req : to_state;
294 /* The device lock serializes operations on memory_subsys_[online|offline] */
295 static int memory_subsys_online(struct device *dev)
297 struct memory_block *mem = to_memory_block(dev);
300 if (mem->state == MEM_ONLINE)
304 * When called via device_online() without configuring the online_type,
305 * we want to default to MMOP_ONLINE.
307 if (mem->online_type == MMOP_OFFLINE)
308 mem->online_type = MMOP_ONLINE;
310 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
311 mem->online_type = MMOP_OFFLINE;
316 static int memory_subsys_offline(struct device *dev)
318 struct memory_block *mem = to_memory_block(dev);
320 if (mem->state == MEM_OFFLINE)
323 return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
326 static ssize_t state_store(struct device *dev, struct device_attribute *attr,
327 const char *buf, size_t count)
329 const int online_type = mhp_online_type_from_str(buf);
330 struct memory_block *mem = to_memory_block(dev);
336 ret = lock_device_hotplug_sysfs();
340 switch (online_type) {
341 case MMOP_ONLINE_KERNEL:
342 case MMOP_ONLINE_MOVABLE:
344 /* mem->online_type is protected by device_hotplug_lock */
345 mem->online_type = online_type;
346 ret = device_online(&mem->dev);
349 ret = device_offline(&mem->dev);
352 ret = -EINVAL; /* should never happen */
355 unlock_device_hotplug();
366 * Legacy interface that we cannot remove: s390x exposes the storage increment
367 * covered by a memory block, allowing for identifying which memory blocks
368 * comprise a storage increment. Since a memory block spans complete
369 * storage increments nowadays, this interface is basically unused. Other
370 * archs never exposed != 0.
372 static ssize_t phys_device_show(struct device *dev,
373 struct device_attribute *attr, char *buf)
375 struct memory_block *mem = to_memory_block(dev);
376 unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
378 return sysfs_emit(buf, "%d\n",
379 arch_get_memory_phys_device(start_pfn));
382 #ifdef CONFIG_MEMORY_HOTREMOVE
383 static int print_allowed_zone(char *buf, int len, int nid,
384 struct memory_group *group,
385 unsigned long start_pfn, unsigned long nr_pages,
386 int online_type, struct zone *default_zone)
390 zone = zone_for_pfn_range(online_type, nid, group, start_pfn, nr_pages);
391 if (zone == default_zone)
394 return sysfs_emit_at(buf, len, " %s", zone->name);
397 static ssize_t valid_zones_show(struct device *dev,
398 struct device_attribute *attr, char *buf)
400 struct memory_block *mem = to_memory_block(dev);
401 unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
402 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
403 struct memory_group *group = mem->group;
404 struct zone *default_zone;
409 * Check the existing zone. Make sure that we do that only on the
410 * online nodes otherwise the page_zone is not reliable
412 if (mem->state == MEM_ONLINE) {
414 * The block contains more than one zone can not be offlined.
415 * This can happen e.g. for ZONE_DMA and ZONE_DMA32
417 default_zone = test_pages_in_a_zone(start_pfn,
418 start_pfn + nr_pages);
420 return sysfs_emit(buf, "%s\n", "none");
421 len += sysfs_emit_at(buf, len, "%s", default_zone->name);
425 default_zone = zone_for_pfn_range(MMOP_ONLINE, nid, group,
426 start_pfn, nr_pages);
428 len += sysfs_emit_at(buf, len, "%s", default_zone->name);
429 len += print_allowed_zone(buf, len, nid, group, start_pfn, nr_pages,
430 MMOP_ONLINE_KERNEL, default_zone);
431 len += print_allowed_zone(buf, len, nid, group, start_pfn, nr_pages,
432 MMOP_ONLINE_MOVABLE, default_zone);
434 len += sysfs_emit_at(buf, len, "\n");
437 static DEVICE_ATTR_RO(valid_zones);
440 static DEVICE_ATTR_RO(phys_index);
441 static DEVICE_ATTR_RW(state);
442 static DEVICE_ATTR_RO(phys_device);
443 static DEVICE_ATTR_RO(removable);
446 * Show the memory block size (shared by all memory blocks).
448 static ssize_t block_size_bytes_show(struct device *dev,
449 struct device_attribute *attr, char *buf)
451 return sysfs_emit(buf, "%lx\n", memory_block_size_bytes());
454 static DEVICE_ATTR_RO(block_size_bytes);
457 * Memory auto online policy.
460 static ssize_t auto_online_blocks_show(struct device *dev,
461 struct device_attribute *attr, char *buf)
463 return sysfs_emit(buf, "%s\n",
464 online_type_to_str[mhp_default_online_type]);
467 static ssize_t auto_online_blocks_store(struct device *dev,
468 struct device_attribute *attr,
469 const char *buf, size_t count)
471 const int online_type = mhp_online_type_from_str(buf);
476 mhp_default_online_type = online_type;
480 static DEVICE_ATTR_RW(auto_online_blocks);
483 * Some architectures will have custom drivers to do this, and
484 * will not need to do it from userspace. The fake hot-add code
485 * as well as ppc64 will do all of their discovery in userspace
486 * and will require this interface.
488 #ifdef CONFIG_ARCH_MEMORY_PROBE
489 static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
490 const char *buf, size_t count)
494 unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
496 ret = kstrtoull(buf, 0, &phys_addr);
500 if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
503 ret = lock_device_hotplug_sysfs();
507 nid = memory_add_physaddr_to_nid(phys_addr);
508 ret = __add_memory(nid, phys_addr,
509 MIN_MEMORY_BLOCK_SIZE * sections_per_block,
517 unlock_device_hotplug();
521 static DEVICE_ATTR_WO(probe);
524 #ifdef CONFIG_MEMORY_FAILURE
526 * Support for offlining pages of memory
529 /* Soft offline a page */
530 static ssize_t soft_offline_page_store(struct device *dev,
531 struct device_attribute *attr,
532 const char *buf, size_t count)
536 if (!capable(CAP_SYS_ADMIN))
538 if (kstrtoull(buf, 0, &pfn) < 0)
541 ret = soft_offline_page(pfn, 0);
542 return ret == 0 ? count : ret;
545 /* Forcibly offline a page, including killing processes. */
546 static ssize_t hard_offline_page_store(struct device *dev,
547 struct device_attribute *attr,
548 const char *buf, size_t count)
552 if (!capable(CAP_SYS_ADMIN))
554 if (kstrtoull(buf, 0, &pfn) < 0)
557 ret = memory_failure(pfn, 0);
558 if (ret == -EOPNOTSUPP)
560 return ret ? ret : count;
563 static DEVICE_ATTR_WO(soft_offline_page);
564 static DEVICE_ATTR_WO(hard_offline_page);
567 /* See phys_device_show(). */
568 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
574 * A reference for the returned memory block device is acquired.
576 * Called under device_hotplug_lock.
578 static struct memory_block *find_memory_block_by_id(unsigned long block_id)
580 struct memory_block *mem;
582 mem = xa_load(&memory_blocks, block_id);
584 get_device(&mem->dev);
589 * Called under device_hotplug_lock.
591 struct memory_block *find_memory_block(unsigned long section_nr)
593 unsigned long block_id = memory_block_id(section_nr);
595 return find_memory_block_by_id(block_id);
598 static struct attribute *memory_memblk_attrs[] = {
599 &dev_attr_phys_index.attr,
600 &dev_attr_state.attr,
601 &dev_attr_phys_device.attr,
602 &dev_attr_removable.attr,
603 #ifdef CONFIG_MEMORY_HOTREMOVE
604 &dev_attr_valid_zones.attr,
609 static const struct attribute_group memory_memblk_attr_group = {
610 .attrs = memory_memblk_attrs,
613 static const struct attribute_group *memory_memblk_attr_groups[] = {
614 &memory_memblk_attr_group,
619 * register_memory - Setup a sysfs device for a memory block
622 int register_memory(struct memory_block *memory)
626 memory->dev.bus = &memory_subsys;
627 memory->dev.id = memory->start_section_nr / sections_per_block;
628 memory->dev.release = memory_block_release;
629 memory->dev.groups = memory_memblk_attr_groups;
630 memory->dev.offline = memory->state == MEM_OFFLINE;
632 ret = device_register(&memory->dev);
634 put_device(&memory->dev);
637 ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
640 device_unregister(&memory->dev);
645 static int init_memory_block(unsigned long block_id, unsigned long state,
646 unsigned long nr_vmemmap_pages,
647 struct memory_group *group)
649 struct memory_block *mem;
652 mem = find_memory_block_by_id(block_id);
654 put_device(&mem->dev);
657 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
661 mem->start_section_nr = block_id * sections_per_block;
663 mem->nid = NUMA_NO_NODE;
664 mem->nr_vmemmap_pages = nr_vmemmap_pages;
665 INIT_LIST_HEAD(&mem->group_next);
667 ret = register_memory(mem);
673 list_add(&mem->group_next, &group->memory_blocks);
679 static int add_memory_block(unsigned long base_section_nr)
681 int section_count = 0;
684 for (nr = base_section_nr; nr < base_section_nr + sections_per_block;
686 if (present_section_nr(nr))
689 if (section_count == 0)
691 return init_memory_block(memory_block_id(base_section_nr),
692 MEM_ONLINE, 0, NULL);
695 static void unregister_memory(struct memory_block *memory)
697 if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys))
700 WARN_ON(xa_erase(&memory_blocks, memory->dev.id) == NULL);
703 list_del(&memory->group_next);
704 memory->group = NULL;
707 /* drop the ref. we got via find_memory_block() */
708 put_device(&memory->dev);
709 device_unregister(&memory->dev);
713 * Create memory block devices for the given memory area. Start and size
714 * have to be aligned to memory block granularity. Memory block devices
715 * will be initialized as offline.
717 * Called under device_hotplug_lock.
719 int create_memory_block_devices(unsigned long start, unsigned long size,
720 unsigned long vmemmap_pages,
721 struct memory_group *group)
723 const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
724 unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
725 struct memory_block *mem;
726 unsigned long block_id;
729 if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
730 !IS_ALIGNED(size, memory_block_size_bytes())))
733 for (block_id = start_block_id; block_id != end_block_id; block_id++) {
734 ret = init_memory_block(block_id, MEM_OFFLINE, vmemmap_pages,
740 end_block_id = block_id;
741 for (block_id = start_block_id; block_id != end_block_id;
743 mem = find_memory_block_by_id(block_id);
744 if (WARN_ON_ONCE(!mem))
746 unregister_memory(mem);
753 * Remove memory block devices for the given memory area. Start and size
754 * have to be aligned to memory block granularity. Memory block devices
755 * have to be offline.
757 * Called under device_hotplug_lock.
759 void remove_memory_block_devices(unsigned long start, unsigned long size)
761 const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
762 const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
763 struct memory_block *mem;
764 unsigned long block_id;
766 if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
767 !IS_ALIGNED(size, memory_block_size_bytes())))
770 for (block_id = start_block_id; block_id != end_block_id; block_id++) {
771 mem = find_memory_block_by_id(block_id);
772 if (WARN_ON_ONCE(!mem))
774 unregister_memory_block_under_nodes(mem);
775 unregister_memory(mem);
779 /* return true if the memory block is offlined, otherwise, return false */
780 bool is_memblock_offlined(struct memory_block *mem)
782 return mem->state == MEM_OFFLINE;
785 static struct attribute *memory_root_attrs[] = {
786 #ifdef CONFIG_ARCH_MEMORY_PROBE
787 &dev_attr_probe.attr,
790 #ifdef CONFIG_MEMORY_FAILURE
791 &dev_attr_soft_offline_page.attr,
792 &dev_attr_hard_offline_page.attr,
795 &dev_attr_block_size_bytes.attr,
796 &dev_attr_auto_online_blocks.attr,
800 static const struct attribute_group memory_root_attr_group = {
801 .attrs = memory_root_attrs,
804 static const struct attribute_group *memory_root_attr_groups[] = {
805 &memory_root_attr_group,
810 * Initialize the sysfs support for memory devices. At the time this function
811 * is called, we cannot have concurrent creation/deletion of memory block
812 * devices, the device_hotplug_lock is not needed.
814 void __init memory_dev_init(void)
817 unsigned long block_sz, nr;
819 /* Validate the configured memory block size */
820 block_sz = memory_block_size_bytes();
821 if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE)
822 panic("Memory block size not suitable: 0x%lx\n", block_sz);
823 sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
825 ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
827 panic("%s() failed to register subsystem: %d\n", __func__, ret);
830 * Create entries for memory sections that were found
831 * during boot and have been initialized
833 for (nr = 0; nr <= __highest_present_section_nr;
834 nr += sections_per_block) {
835 ret = add_memory_block(nr);
837 panic("%s() failed to add memory block: %d\n", __func__,
843 * walk_memory_blocks - walk through all present memory blocks overlapped
844 * by the range [start, start + size)
846 * @start: start address of the memory range
847 * @size: size of the memory range
848 * @arg: argument passed to func
849 * @func: callback for each memory section walked
851 * This function walks through all present memory blocks overlapped by the
852 * range [start, start + size), calling func on each memory block.
854 * In case func() returns an error, walking is aborted and the error is
857 * Called under device_hotplug_lock.
859 int walk_memory_blocks(unsigned long start, unsigned long size,
860 void *arg, walk_memory_blocks_func_t func)
862 const unsigned long start_block_id = phys_to_block_id(start);
863 const unsigned long end_block_id = phys_to_block_id(start + size - 1);
864 struct memory_block *mem;
865 unsigned long block_id;
871 for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
872 mem = find_memory_block_by_id(block_id);
876 ret = func(mem, arg);
877 put_device(&mem->dev);
884 struct for_each_memory_block_cb_data {
885 walk_memory_blocks_func_t func;
889 static int for_each_memory_block_cb(struct device *dev, void *data)
891 struct memory_block *mem = to_memory_block(dev);
892 struct for_each_memory_block_cb_data *cb_data = data;
894 return cb_data->func(mem, cb_data->arg);
898 * for_each_memory_block - walk through all present memory blocks
900 * @arg: argument passed to func
901 * @func: callback for each memory block walked
903 * This function walks through all present memory blocks, calling func on
906 * In case func() returns an error, walking is aborted and the error is
909 int for_each_memory_block(void *arg, walk_memory_blocks_func_t func)
911 struct for_each_memory_block_cb_data cb_data = {
916 return bus_for_each_dev(&memory_subsys, NULL, &cb_data,
917 for_each_memory_block_cb);
921 * This is an internal helper to unify allocation and initialization of
922 * memory groups. Note that the passed memory group will be copied to a
923 * dynamically allocated memory group. After this call, the passed
924 * memory group should no longer be used.
926 static int memory_group_register(struct memory_group group)
928 struct memory_group *new_group;
932 if (!node_possible(group.nid))
935 new_group = kzalloc(sizeof(group), GFP_KERNEL);
939 INIT_LIST_HEAD(&new_group->memory_blocks);
941 ret = xa_alloc(&memory_groups, &mgid, new_group, xa_limit_31b,
946 } else if (group.is_dynamic) {
947 xa_set_mark(&memory_groups, mgid, MEMORY_GROUP_MARK_DYNAMIC);
953 * memory_group_register_static() - Register a static memory group.
955 * @max_pages: The maximum number of pages we'll have in this static memory
958 * Register a new static memory group and return the memory group id.
959 * All memory in the group belongs to a single unit, such as a DIMM. All
960 * memory belonging to a static memory group is added in one go to be removed
961 * in one go -- it's static.
963 * Returns an error if out of memory, if the node id is invalid, if no new
964 * memory groups can be registered, or if max_pages is invalid (0). Otherwise,
965 * returns the new memory group id.
967 int memory_group_register_static(int nid, unsigned long max_pages)
969 struct memory_group group = {
972 .max_pages = max_pages,
978 return memory_group_register(group);
980 EXPORT_SYMBOL_GPL(memory_group_register_static);
983 * memory_group_register_dynamic() - Register a dynamic memory group.
985 * @unit_pages: Unit in pages in which is memory added/removed in this dynamic
988 * Register a new dynamic memory group and return the memory group id.
989 * Memory within a dynamic memory group is added/removed dynamically
992 * Returns an error if out of memory, if the node id is invalid, if no new
993 * memory groups can be registered, or if unit_pages is invalid (0, not a
994 * power of two, smaller than a single memory block). Otherwise, returns the
995 * new memory group id.
997 int memory_group_register_dynamic(int nid, unsigned long unit_pages)
999 struct memory_group group = {
1003 .unit_pages = unit_pages,
1007 if (!unit_pages || !is_power_of_2(unit_pages) ||
1008 unit_pages < PHYS_PFN(memory_block_size_bytes()))
1010 return memory_group_register(group);
1012 EXPORT_SYMBOL_GPL(memory_group_register_dynamic);
1015 * memory_group_unregister() - Unregister a memory group.
1016 * @mgid: the memory group id
1018 * Unregister a memory group. If any memory block still belongs to this
1019 * memory group, unregistering will fail.
1021 * Returns -EINVAL if the memory group id is invalid, returns -EBUSY if some
1022 * memory blocks still belong to this memory group and returns 0 if
1023 * unregistering succeeded.
1025 int memory_group_unregister(int mgid)
1027 struct memory_group *group;
1032 group = xa_load(&memory_groups, mgid);
1035 if (!list_empty(&group->memory_blocks))
1037 xa_erase(&memory_groups, mgid);
1041 EXPORT_SYMBOL_GPL(memory_group_unregister);
1044 * This is an internal helper only to be used in core memory hotplug code to
1045 * lookup a memory group. We don't care about locking, as we don't expect a
1046 * memory group to get unregistered while adding memory to it -- because
1047 * the group and the memory is managed by the same driver.
1049 struct memory_group *memory_group_find_by_id(int mgid)
1051 return xa_load(&memory_groups, mgid);
1055 * This is an internal helper only to be used in core memory hotplug code to
1056 * walk all dynamic memory groups excluding a given memory group, either
1057 * belonging to a specific node, or belonging to any node.
1059 int walk_dynamic_memory_groups(int nid, walk_memory_groups_func_t func,
1060 struct memory_group *excluded, void *arg)
1062 struct memory_group *group;
1063 unsigned long index;
1066 xa_for_each_marked(&memory_groups, index, group,
1067 MEMORY_GROUP_MARK_DYNAMIC) {
1068 if (group == excluded)
1071 if (nid != NUMA_NO_NODE && group->nid != nid)
1073 #endif /* CONFIG_NUMA */
1074 ret = func(group, arg);