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,
222 static int memory_block_offline(struct memory_block *mem)
224 unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
225 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
226 unsigned long nr_vmemmap_pages = mem->nr_vmemmap_pages;
233 * Unaccount before offlining, such that unpopulated zone and kthreads
234 * can properly be torn down in offline_pages().
236 if (nr_vmemmap_pages)
237 adjust_present_page_count(pfn_to_page(start_pfn), mem->group,
240 ret = offline_pages(start_pfn + nr_vmemmap_pages,
241 nr_pages - nr_vmemmap_pages, mem->zone, mem->group);
243 /* offline_pages() failed. Account back. */
244 if (nr_vmemmap_pages)
245 adjust_present_page_count(pfn_to_page(start_pfn),
246 mem->group, nr_vmemmap_pages);
250 if (nr_vmemmap_pages)
251 mhp_deinit_memmap_on_memory(start_pfn, nr_vmemmap_pages);
258 * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
259 * OK to have direct references to sparsemem variables in here.
262 memory_block_action(struct memory_block *mem, unsigned long action)
268 ret = memory_block_online(mem);
271 ret = memory_block_offline(mem);
274 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
275 "%ld\n", __func__, mem->start_section_nr, action, action);
282 static int memory_block_change_state(struct memory_block *mem,
283 unsigned long to_state, unsigned long from_state_req)
287 if (mem->state != from_state_req)
290 if (to_state == MEM_OFFLINE)
291 mem->state = MEM_GOING_OFFLINE;
293 ret = memory_block_action(mem, to_state);
294 mem->state = ret ? from_state_req : to_state;
299 /* The device lock serializes operations on memory_subsys_[online|offline] */
300 static int memory_subsys_online(struct device *dev)
302 struct memory_block *mem = to_memory_block(dev);
305 if (mem->state == MEM_ONLINE)
309 * When called via device_online() without configuring the online_type,
310 * we want to default to MMOP_ONLINE.
312 if (mem->online_type == MMOP_OFFLINE)
313 mem->online_type = MMOP_ONLINE;
315 ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
316 mem->online_type = MMOP_OFFLINE;
321 static int memory_subsys_offline(struct device *dev)
323 struct memory_block *mem = to_memory_block(dev);
325 if (mem->state == MEM_OFFLINE)
328 return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
331 static ssize_t state_store(struct device *dev, struct device_attribute *attr,
332 const char *buf, size_t count)
334 const int online_type = mhp_online_type_from_str(buf);
335 struct memory_block *mem = to_memory_block(dev);
341 ret = lock_device_hotplug_sysfs();
345 switch (online_type) {
346 case MMOP_ONLINE_KERNEL:
347 case MMOP_ONLINE_MOVABLE:
349 /* mem->online_type is protected by device_hotplug_lock */
350 mem->online_type = online_type;
351 ret = device_online(&mem->dev);
354 ret = device_offline(&mem->dev);
357 ret = -EINVAL; /* should never happen */
360 unlock_device_hotplug();
371 * Legacy interface that we cannot remove: s390x exposes the storage increment
372 * covered by a memory block, allowing for identifying which memory blocks
373 * comprise a storage increment. Since a memory block spans complete
374 * storage increments nowadays, this interface is basically unused. Other
375 * archs never exposed != 0.
377 static ssize_t phys_device_show(struct device *dev,
378 struct device_attribute *attr, char *buf)
380 struct memory_block *mem = to_memory_block(dev);
381 unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
383 return sysfs_emit(buf, "%d\n",
384 arch_get_memory_phys_device(start_pfn));
387 #ifdef CONFIG_MEMORY_HOTREMOVE
388 static int print_allowed_zone(char *buf, int len, int nid,
389 struct memory_group *group,
390 unsigned long start_pfn, unsigned long nr_pages,
391 int online_type, struct zone *default_zone)
395 zone = zone_for_pfn_range(online_type, nid, group, start_pfn, nr_pages);
396 if (zone == default_zone)
399 return sysfs_emit_at(buf, len, " %s", zone->name);
402 static ssize_t valid_zones_show(struct device *dev,
403 struct device_attribute *attr, char *buf)
405 struct memory_block *mem = to_memory_block(dev);
406 unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
407 unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
408 struct memory_group *group = mem->group;
409 struct zone *default_zone;
414 * Check the existing zone. Make sure that we do that only on the
415 * online nodes otherwise the page_zone is not reliable
417 if (mem->state == MEM_ONLINE) {
419 * If !mem->zone, the memory block spans multiple zones and
420 * cannot get offlined.
422 default_zone = mem->zone;
424 return sysfs_emit(buf, "%s\n", "none");
425 len += sysfs_emit_at(buf, len, "%s", default_zone->name);
429 default_zone = zone_for_pfn_range(MMOP_ONLINE, nid, group,
430 start_pfn, nr_pages);
432 len += sysfs_emit_at(buf, len, "%s", default_zone->name);
433 len += print_allowed_zone(buf, len, nid, group, start_pfn, nr_pages,
434 MMOP_ONLINE_KERNEL, default_zone);
435 len += print_allowed_zone(buf, len, nid, group, start_pfn, nr_pages,
436 MMOP_ONLINE_MOVABLE, default_zone);
438 len += sysfs_emit_at(buf, len, "\n");
441 static DEVICE_ATTR_RO(valid_zones);
444 static DEVICE_ATTR_RO(phys_index);
445 static DEVICE_ATTR_RW(state);
446 static DEVICE_ATTR_RO(phys_device);
447 static DEVICE_ATTR_RO(removable);
450 * Show the memory block size (shared by all memory blocks).
452 static ssize_t block_size_bytes_show(struct device *dev,
453 struct device_attribute *attr, char *buf)
455 return sysfs_emit(buf, "%lx\n", memory_block_size_bytes());
458 static DEVICE_ATTR_RO(block_size_bytes);
461 * Memory auto online policy.
464 static ssize_t auto_online_blocks_show(struct device *dev,
465 struct device_attribute *attr, char *buf)
467 return sysfs_emit(buf, "%s\n",
468 online_type_to_str[mhp_default_online_type]);
471 static ssize_t auto_online_blocks_store(struct device *dev,
472 struct device_attribute *attr,
473 const char *buf, size_t count)
475 const int online_type = mhp_online_type_from_str(buf);
480 mhp_default_online_type = online_type;
484 static DEVICE_ATTR_RW(auto_online_blocks);
487 * Some architectures will have custom drivers to do this, and
488 * will not need to do it from userspace. The fake hot-add code
489 * as well as ppc64 will do all of their discovery in userspace
490 * and will require this interface.
492 #ifdef CONFIG_ARCH_MEMORY_PROBE
493 static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
494 const char *buf, size_t count)
498 unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
500 ret = kstrtoull(buf, 0, &phys_addr);
504 if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
507 ret = lock_device_hotplug_sysfs();
511 nid = memory_add_physaddr_to_nid(phys_addr);
512 ret = __add_memory(nid, phys_addr,
513 MIN_MEMORY_BLOCK_SIZE * sections_per_block,
521 unlock_device_hotplug();
525 static DEVICE_ATTR_WO(probe);
528 #ifdef CONFIG_MEMORY_FAILURE
530 * Support for offlining pages of memory
533 /* Soft offline a page */
534 static ssize_t soft_offline_page_store(struct device *dev,
535 struct device_attribute *attr,
536 const char *buf, size_t count)
540 if (!capable(CAP_SYS_ADMIN))
542 if (kstrtoull(buf, 0, &pfn) < 0)
545 ret = soft_offline_page(pfn, 0);
546 return ret == 0 ? count : ret;
549 /* Forcibly offline a page, including killing processes. */
550 static ssize_t hard_offline_page_store(struct device *dev,
551 struct device_attribute *attr,
552 const char *buf, size_t count)
556 if (!capable(CAP_SYS_ADMIN))
558 if (kstrtoull(buf, 0, &pfn) < 0)
561 ret = memory_failure(pfn, MF_SW_SIMULATED);
562 if (ret == -EOPNOTSUPP)
564 return ret ? ret : count;
567 static DEVICE_ATTR_WO(soft_offline_page);
568 static DEVICE_ATTR_WO(hard_offline_page);
571 /* See phys_device_show(). */
572 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
578 * A reference for the returned memory block device is acquired.
580 * Called under device_hotplug_lock.
582 static struct memory_block *find_memory_block_by_id(unsigned long block_id)
584 struct memory_block *mem;
586 mem = xa_load(&memory_blocks, block_id);
588 get_device(&mem->dev);
593 * Called under device_hotplug_lock.
595 struct memory_block *find_memory_block(unsigned long section_nr)
597 unsigned long block_id = memory_block_id(section_nr);
599 return find_memory_block_by_id(block_id);
602 static struct attribute *memory_memblk_attrs[] = {
603 &dev_attr_phys_index.attr,
604 &dev_attr_state.attr,
605 &dev_attr_phys_device.attr,
606 &dev_attr_removable.attr,
607 #ifdef CONFIG_MEMORY_HOTREMOVE
608 &dev_attr_valid_zones.attr,
613 static const struct attribute_group memory_memblk_attr_group = {
614 .attrs = memory_memblk_attrs,
617 static const struct attribute_group *memory_memblk_attr_groups[] = {
618 &memory_memblk_attr_group,
622 static int __add_memory_block(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 struct zone *early_node_zone_for_memory_block(struct memory_block *mem,
648 const unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
649 const unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
650 struct zone *zone, *matching_zone = NULL;
651 pg_data_t *pgdat = NODE_DATA(nid);
655 * This logic only works for early memory, when the applicable zones
656 * already span the memory block. We don't expect overlapping zones on
657 * a single node for early memory. So if we're told that some PFNs
658 * of a node fall into this memory block, we can assume that all node
659 * zones that intersect with the memory block are actually applicable.
660 * No need to look at the memmap.
662 for (i = 0; i < MAX_NR_ZONES; i++) {
663 zone = pgdat->node_zones + i;
664 if (!populated_zone(zone))
666 if (!zone_intersects(zone, start_pfn, nr_pages))
668 if (!matching_zone) {
669 matching_zone = zone;
672 /* Spans multiple zones ... */
673 matching_zone = NULL;
676 return matching_zone;
681 * memory_block_add_nid() - Indicate that system RAM falling into this memory
682 * block device (partially) belongs to the given node.
683 * @mem: The memory block device.
685 * @context: The memory initialization context.
687 * Indicate that system RAM falling into this memory block (partially) belongs
688 * to the given node. If the context indicates ("early") that we are adding the
689 * node during node device subsystem initialization, this will also properly
690 * set/adjust mem->zone based on the zone ranges of the given node.
692 void memory_block_add_nid(struct memory_block *mem, int nid,
693 enum meminit_context context)
695 if (context == MEMINIT_EARLY && mem->nid != nid) {
697 * For early memory we have to determine the zone when setting
698 * the node id and handle multiple nodes spanning a single
699 * memory block by indicate via zone == NULL that we're not
700 * dealing with a single zone. So if we're setting the node id
701 * the first time, determine if there is a single zone. If we're
702 * setting the node id a second time to a different node,
703 * invalidate the single detected zone.
705 if (mem->nid == NUMA_NO_NODE)
706 mem->zone = early_node_zone_for_memory_block(mem, nid);
712 * If this memory block spans multiple nodes, we only indicate
713 * the last processed node. If we span multiple nodes (not applicable
714 * to hotplugged memory), zone == NULL will prohibit memory offlining
715 * and consequently unplug.
721 static int add_memory_block(unsigned long block_id, unsigned long state,
722 unsigned long nr_vmemmap_pages,
723 struct memory_group *group)
725 struct memory_block *mem;
728 mem = find_memory_block_by_id(block_id);
730 put_device(&mem->dev);
733 mem = kzalloc(sizeof(*mem), GFP_KERNEL);
737 mem->start_section_nr = block_id * sections_per_block;
739 mem->nid = NUMA_NO_NODE;
740 mem->nr_vmemmap_pages = nr_vmemmap_pages;
741 INIT_LIST_HEAD(&mem->group_next);
744 if (state == MEM_ONLINE)
746 * MEM_ONLINE at this point implies early memory. With NUMA,
747 * we'll determine the zone when setting the node id via
748 * memory_block_add_nid(). Memory hotplug updated the zone
749 * manually when memory onlining/offlining succeeds.
751 mem->zone = early_node_zone_for_memory_block(mem, NUMA_NO_NODE);
752 #endif /* CONFIG_NUMA */
754 ret = __add_memory_block(mem);
760 list_add(&mem->group_next, &group->memory_blocks);
766 static int __init add_boot_memory_block(unsigned long base_section_nr)
768 int section_count = 0;
771 for (nr = base_section_nr; nr < base_section_nr + sections_per_block;
773 if (present_section_nr(nr))
776 if (section_count == 0)
778 return add_memory_block(memory_block_id(base_section_nr),
779 MEM_ONLINE, 0, NULL);
782 static int add_hotplug_memory_block(unsigned long block_id,
783 unsigned long nr_vmemmap_pages,
784 struct memory_group *group)
786 return add_memory_block(block_id, MEM_OFFLINE, nr_vmemmap_pages, group);
789 static void remove_memory_block(struct memory_block *memory)
791 if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys))
794 WARN_ON(xa_erase(&memory_blocks, memory->dev.id) == NULL);
797 list_del(&memory->group_next);
798 memory->group = NULL;
801 /* drop the ref. we got via find_memory_block() */
802 put_device(&memory->dev);
803 device_unregister(&memory->dev);
807 * Create memory block devices for the given memory area. Start and size
808 * have to be aligned to memory block granularity. Memory block devices
809 * will be initialized as offline.
811 * Called under device_hotplug_lock.
813 int create_memory_block_devices(unsigned long start, unsigned long size,
814 unsigned long vmemmap_pages,
815 struct memory_group *group)
817 const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
818 unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
819 struct memory_block *mem;
820 unsigned long block_id;
823 if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
824 !IS_ALIGNED(size, memory_block_size_bytes())))
827 for (block_id = start_block_id; block_id != end_block_id; block_id++) {
828 ret = add_hotplug_memory_block(block_id, vmemmap_pages, group);
833 end_block_id = block_id;
834 for (block_id = start_block_id; block_id != end_block_id;
836 mem = find_memory_block_by_id(block_id);
837 if (WARN_ON_ONCE(!mem))
839 remove_memory_block(mem);
846 * Remove memory block devices for the given memory area. Start and size
847 * have to be aligned to memory block granularity. Memory block devices
848 * have to be offline.
850 * Called under device_hotplug_lock.
852 void remove_memory_block_devices(unsigned long start, unsigned long size)
854 const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
855 const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
856 struct memory_block *mem;
857 unsigned long block_id;
859 if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
860 !IS_ALIGNED(size, memory_block_size_bytes())))
863 for (block_id = start_block_id; block_id != end_block_id; block_id++) {
864 mem = find_memory_block_by_id(block_id);
865 if (WARN_ON_ONCE(!mem))
867 unregister_memory_block_under_nodes(mem);
868 remove_memory_block(mem);
872 static struct attribute *memory_root_attrs[] = {
873 #ifdef CONFIG_ARCH_MEMORY_PROBE
874 &dev_attr_probe.attr,
877 #ifdef CONFIG_MEMORY_FAILURE
878 &dev_attr_soft_offline_page.attr,
879 &dev_attr_hard_offline_page.attr,
882 &dev_attr_block_size_bytes.attr,
883 &dev_attr_auto_online_blocks.attr,
887 static const struct attribute_group memory_root_attr_group = {
888 .attrs = memory_root_attrs,
891 static const struct attribute_group *memory_root_attr_groups[] = {
892 &memory_root_attr_group,
897 * Initialize the sysfs support for memory devices. At the time this function
898 * is called, we cannot have concurrent creation/deletion of memory block
899 * devices, the device_hotplug_lock is not needed.
901 void __init memory_dev_init(void)
904 unsigned long block_sz, nr;
906 /* Validate the configured memory block size */
907 block_sz = memory_block_size_bytes();
908 if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE)
909 panic("Memory block size not suitable: 0x%lx\n", block_sz);
910 sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
912 ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
914 panic("%s() failed to register subsystem: %d\n", __func__, ret);
917 * Create entries for memory sections that were found
918 * during boot and have been initialized
920 for (nr = 0; nr <= __highest_present_section_nr;
921 nr += sections_per_block) {
922 ret = add_boot_memory_block(nr);
924 panic("%s() failed to add memory block: %d\n", __func__,
930 * walk_memory_blocks - walk through all present memory blocks overlapped
931 * by the range [start, start + size)
933 * @start: start address of the memory range
934 * @size: size of the memory range
935 * @arg: argument passed to func
936 * @func: callback for each memory section walked
938 * This function walks through all present memory blocks overlapped by the
939 * range [start, start + size), calling func on each memory block.
941 * In case func() returns an error, walking is aborted and the error is
944 * Called under device_hotplug_lock.
946 int walk_memory_blocks(unsigned long start, unsigned long size,
947 void *arg, walk_memory_blocks_func_t func)
949 const unsigned long start_block_id = phys_to_block_id(start);
950 const unsigned long end_block_id = phys_to_block_id(start + size - 1);
951 struct memory_block *mem;
952 unsigned long block_id;
958 for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
959 mem = find_memory_block_by_id(block_id);
963 ret = func(mem, arg);
964 put_device(&mem->dev);
971 struct for_each_memory_block_cb_data {
972 walk_memory_blocks_func_t func;
976 static int for_each_memory_block_cb(struct device *dev, void *data)
978 struct memory_block *mem = to_memory_block(dev);
979 struct for_each_memory_block_cb_data *cb_data = data;
981 return cb_data->func(mem, cb_data->arg);
985 * for_each_memory_block - walk through all present memory blocks
987 * @arg: argument passed to func
988 * @func: callback for each memory block walked
990 * This function walks through all present memory blocks, calling func on
993 * In case func() returns an error, walking is aborted and the error is
996 int for_each_memory_block(void *arg, walk_memory_blocks_func_t func)
998 struct for_each_memory_block_cb_data cb_data = {
1003 return bus_for_each_dev(&memory_subsys, NULL, &cb_data,
1004 for_each_memory_block_cb);
1008 * This is an internal helper to unify allocation and initialization of
1009 * memory groups. Note that the passed memory group will be copied to a
1010 * dynamically allocated memory group. After this call, the passed
1011 * memory group should no longer be used.
1013 static int memory_group_register(struct memory_group group)
1015 struct memory_group *new_group;
1019 if (!node_possible(group.nid))
1022 new_group = kzalloc(sizeof(group), GFP_KERNEL);
1026 INIT_LIST_HEAD(&new_group->memory_blocks);
1028 ret = xa_alloc(&memory_groups, &mgid, new_group, xa_limit_31b,
1033 } else if (group.is_dynamic) {
1034 xa_set_mark(&memory_groups, mgid, MEMORY_GROUP_MARK_DYNAMIC);
1040 * memory_group_register_static() - Register a static memory group.
1041 * @nid: The node id.
1042 * @max_pages: The maximum number of pages we'll have in this static memory
1045 * Register a new static memory group and return the memory group id.
1046 * All memory in the group belongs to a single unit, such as a DIMM. All
1047 * memory belonging to a static memory group is added in one go to be removed
1048 * in one go -- it's static.
1050 * Returns an error if out of memory, if the node id is invalid, if no new
1051 * memory groups can be registered, or if max_pages is invalid (0). Otherwise,
1052 * returns the new memory group id.
1054 int memory_group_register_static(int nid, unsigned long max_pages)
1056 struct memory_group group = {
1059 .max_pages = max_pages,
1065 return memory_group_register(group);
1067 EXPORT_SYMBOL_GPL(memory_group_register_static);
1070 * memory_group_register_dynamic() - Register a dynamic memory group.
1071 * @nid: The node id.
1072 * @unit_pages: Unit in pages in which is memory added/removed in this dynamic
1075 * Register a new dynamic memory group and return the memory group id.
1076 * Memory within a dynamic memory group is added/removed dynamically
1079 * Returns an error if out of memory, if the node id is invalid, if no new
1080 * memory groups can be registered, or if unit_pages is invalid (0, not a
1081 * power of two, smaller than a single memory block). Otherwise, returns the
1082 * new memory group id.
1084 int memory_group_register_dynamic(int nid, unsigned long unit_pages)
1086 struct memory_group group = {
1090 .unit_pages = unit_pages,
1094 if (!unit_pages || !is_power_of_2(unit_pages) ||
1095 unit_pages < PHYS_PFN(memory_block_size_bytes()))
1097 return memory_group_register(group);
1099 EXPORT_SYMBOL_GPL(memory_group_register_dynamic);
1102 * memory_group_unregister() - Unregister a memory group.
1103 * @mgid: the memory group id
1105 * Unregister a memory group. If any memory block still belongs to this
1106 * memory group, unregistering will fail.
1108 * Returns -EINVAL if the memory group id is invalid, returns -EBUSY if some
1109 * memory blocks still belong to this memory group and returns 0 if
1110 * unregistering succeeded.
1112 int memory_group_unregister(int mgid)
1114 struct memory_group *group;
1119 group = xa_load(&memory_groups, mgid);
1122 if (!list_empty(&group->memory_blocks))
1124 xa_erase(&memory_groups, mgid);
1128 EXPORT_SYMBOL_GPL(memory_group_unregister);
1131 * This is an internal helper only to be used in core memory hotplug code to
1132 * lookup a memory group. We don't care about locking, as we don't expect a
1133 * memory group to get unregistered while adding memory to it -- because
1134 * the group and the memory is managed by the same driver.
1136 struct memory_group *memory_group_find_by_id(int mgid)
1138 return xa_load(&memory_groups, mgid);
1142 * This is an internal helper only to be used in core memory hotplug code to
1143 * walk all dynamic memory groups excluding a given memory group, either
1144 * belonging to a specific node, or belonging to any node.
1146 int walk_dynamic_memory_groups(int nid, walk_memory_groups_func_t func,
1147 struct memory_group *excluded, void *arg)
1149 struct memory_group *group;
1150 unsigned long index;
1153 xa_for_each_marked(&memory_groups, index, group,
1154 MEMORY_GROUP_MARK_DYNAMIC) {
1155 if (group == excluded)
1158 if (nid != NUMA_NO_NODE && group->nid != nid)
1160 #endif /* CONFIG_NUMA */
1161 ret = func(group, arg);