1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/mm/memory_hotplug.c
8 #include <linux/stddef.h>
10 #include <linux/sched/signal.h>
11 #include <linux/swap.h>
12 #include <linux/interrupt.h>
13 #include <linux/pagemap.h>
14 #include <linux/compiler.h>
15 #include <linux/export.h>
16 #include <linux/pagevec.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sysctl.h>
20 #include <linux/cpu.h>
21 #include <linux/memory.h>
22 #include <linux/memremap.h>
23 #include <linux/memory_hotplug.h>
24 #include <linux/highmem.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ioport.h>
27 #include <linux/delay.h>
28 #include <linux/migrate.h>
29 #include <linux/page-isolation.h>
30 #include <linux/pfn.h>
31 #include <linux/suspend.h>
32 #include <linux/mm_inline.h>
33 #include <linux/firmware-map.h>
34 #include <linux/stop_machine.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memblock.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
52 static online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
55 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
57 void get_online_mems(void)
59 percpu_down_read(&mem_hotplug_lock);
62 void put_online_mems(void)
64 percpu_up_read(&mem_hotplug_lock);
67 bool movable_node_enabled = false;
69 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
70 int memhp_default_online_type = MMOP_OFFLINE;
72 int memhp_default_online_type = MMOP_ONLINE;
75 static int __init setup_memhp_default_state(char *str)
77 const int online_type = memhp_online_type_from_str(str);
80 memhp_default_online_type = online_type;
84 __setup("memhp_default_state=", setup_memhp_default_state);
86 void mem_hotplug_begin(void)
89 percpu_down_write(&mem_hotplug_lock);
92 void mem_hotplug_done(void)
94 percpu_up_write(&mem_hotplug_lock);
98 u64 max_mem_size = U64_MAX;
100 /* add this memory to iomem resource */
101 static struct resource *register_memory_resource(u64 start, u64 size,
102 const char *resource_name)
104 struct resource *res;
105 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
107 if (strcmp(resource_name, "System RAM"))
108 flags |= IORESOURCE_MEM_DRIVER_MANAGED;
111 * Make sure value parsed from 'mem=' only restricts memory adding
112 * while booting, so that memory hotplug won't be impacted. Please
113 * refer to document of 'mem=' in kernel-parameters.txt for more
116 if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
117 return ERR_PTR(-E2BIG);
120 * Request ownership of the new memory range. This might be
121 * a child of an existing resource that was present but
122 * not marked as busy.
124 res = __request_region(&iomem_resource, start, size,
125 resource_name, flags);
128 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
129 start, start + size);
130 return ERR_PTR(-EEXIST);
135 static void release_memory_resource(struct resource *res)
139 release_resource(res);
143 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
144 void get_page_bootmem(unsigned long info, struct page *page,
147 page->freelist = (void *)type;
148 SetPagePrivate(page);
149 set_page_private(page, info);
153 void put_page_bootmem(struct page *page)
157 type = (unsigned long) page->freelist;
158 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
159 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
161 if (page_ref_dec_return(page) == 1) {
162 page->freelist = NULL;
163 ClearPagePrivate(page);
164 set_page_private(page, 0);
165 INIT_LIST_HEAD(&page->lru);
166 free_reserved_page(page);
170 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
171 #ifndef CONFIG_SPARSEMEM_VMEMMAP
172 static void register_page_bootmem_info_section(unsigned long start_pfn)
174 unsigned long mapsize, section_nr, i;
175 struct mem_section *ms;
176 struct page *page, *memmap;
177 struct mem_section_usage *usage;
179 section_nr = pfn_to_section_nr(start_pfn);
180 ms = __nr_to_section(section_nr);
182 /* Get section's memmap address */
183 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
186 * Get page for the memmap's phys address
187 * XXX: need more consideration for sparse_vmemmap...
189 page = virt_to_page(memmap);
190 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
191 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
193 /* remember memmap's page */
194 for (i = 0; i < mapsize; i++, page++)
195 get_page_bootmem(section_nr, page, SECTION_INFO);
198 page = virt_to_page(usage);
200 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
202 for (i = 0; i < mapsize; i++, page++)
203 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
206 #else /* CONFIG_SPARSEMEM_VMEMMAP */
207 static void register_page_bootmem_info_section(unsigned long start_pfn)
209 unsigned long mapsize, section_nr, i;
210 struct mem_section *ms;
211 struct page *page, *memmap;
212 struct mem_section_usage *usage;
214 section_nr = pfn_to_section_nr(start_pfn);
215 ms = __nr_to_section(section_nr);
217 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
219 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
222 page = virt_to_page(usage);
224 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
226 for (i = 0; i < mapsize; i++, page++)
227 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
229 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
231 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
233 unsigned long i, pfn, end_pfn, nr_pages;
234 int node = pgdat->node_id;
237 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
238 page = virt_to_page(pgdat);
240 for (i = 0; i < nr_pages; i++, page++)
241 get_page_bootmem(node, page, NODE_INFO);
243 pfn = pgdat->node_start_pfn;
244 end_pfn = pgdat_end_pfn(pgdat);
246 /* register section info */
247 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
249 * Some platforms can assign the same pfn to multiple nodes - on
250 * node0 as well as nodeN. To avoid registering a pfn against
251 * multiple nodes we check that this pfn does not already
252 * reside in some other nodes.
254 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
255 register_page_bootmem_info_section(pfn);
258 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
260 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
264 * Disallow all operations smaller than a sub-section and only
265 * allow operations smaller than a section for
266 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
267 * enforces a larger memory_block_size_bytes() granularity for
268 * memory that will be marked online, so this check should only
269 * fire for direct arch_{add,remove}_memory() users outside of
270 * add_memory_resource().
272 unsigned long min_align;
274 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
275 min_align = PAGES_PER_SUBSECTION;
277 min_align = PAGES_PER_SECTION;
278 if (!IS_ALIGNED(pfn, min_align)
279 || !IS_ALIGNED(nr_pages, min_align)) {
280 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
281 reason, pfn, pfn + nr_pages - 1);
287 static int check_hotplug_memory_addressable(unsigned long pfn,
288 unsigned long nr_pages)
290 const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1;
292 if (max_addr >> MAX_PHYSMEM_BITS) {
293 const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1;
295 "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
296 (u64)PFN_PHYS(pfn), max_addr, max_allowed);
304 * Reasonably generic function for adding memory. It is
305 * expected that archs that support memory hotplug will
306 * call this function after deciding the zone to which to
309 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
310 struct mhp_params *params)
312 const unsigned long end_pfn = pfn + nr_pages;
313 unsigned long cur_nr_pages;
315 struct vmem_altmap *altmap = params->altmap;
317 if (WARN_ON_ONCE(!params->pgprot.pgprot))
320 err = check_hotplug_memory_addressable(pfn, nr_pages);
326 * Validate altmap is within bounds of the total request
328 if (altmap->base_pfn != pfn
329 || vmem_altmap_offset(altmap) > nr_pages) {
330 pr_warn_once("memory add fail, invalid altmap\n");
336 err = check_pfn_span(pfn, nr_pages, "add");
340 for (; pfn < end_pfn; pfn += cur_nr_pages) {
341 /* Select all remaining pages up to the next section boundary */
342 cur_nr_pages = min(end_pfn - pfn,
343 SECTION_ALIGN_UP(pfn + 1) - pfn);
344 err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
349 vmemmap_populate_print_last();
354 int __weak memory_add_physaddr_to_nid(u64 start)
356 pr_info_once("Unknown online node for memory at 0x%llx, assuming node 0\n",
360 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
362 int __weak phys_to_target_node(u64 start)
364 pr_info_once("Unknown target node for memory at 0x%llx, assuming node 0\n",
368 EXPORT_SYMBOL_GPL(phys_to_target_node);
371 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
372 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
373 unsigned long start_pfn,
374 unsigned long end_pfn)
376 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
377 if (unlikely(!pfn_to_online_page(start_pfn)))
380 if (unlikely(pfn_to_nid(start_pfn) != nid))
383 if (zone != page_zone(pfn_to_page(start_pfn)))
392 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
393 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
394 unsigned long start_pfn,
395 unsigned long end_pfn)
399 /* pfn is the end pfn of a memory section. */
401 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
402 if (unlikely(!pfn_to_online_page(pfn)))
405 if (unlikely(pfn_to_nid(pfn) != nid))
408 if (zone != page_zone(pfn_to_page(pfn)))
417 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
418 unsigned long end_pfn)
421 int nid = zone_to_nid(zone);
423 zone_span_writelock(zone);
424 if (zone->zone_start_pfn == start_pfn) {
426 * If the section is smallest section in the zone, it need
427 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
428 * In this case, we find second smallest valid mem_section
429 * for shrinking zone.
431 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
434 zone->spanned_pages = zone_end_pfn(zone) - pfn;
435 zone->zone_start_pfn = pfn;
437 zone->zone_start_pfn = 0;
438 zone->spanned_pages = 0;
440 } else if (zone_end_pfn(zone) == end_pfn) {
442 * If the section is biggest section in the zone, it need
443 * shrink zone->spanned_pages.
444 * In this case, we find second biggest valid mem_section for
447 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
450 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
452 zone->zone_start_pfn = 0;
453 zone->spanned_pages = 0;
456 zone_span_writeunlock(zone);
459 static void update_pgdat_span(struct pglist_data *pgdat)
461 unsigned long node_start_pfn = 0, node_end_pfn = 0;
464 for (zone = pgdat->node_zones;
465 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
466 unsigned long zone_end_pfn = zone->zone_start_pfn +
469 /* No need to lock the zones, they can't change. */
470 if (!zone->spanned_pages)
473 node_start_pfn = zone->zone_start_pfn;
474 node_end_pfn = zone_end_pfn;
478 if (zone_end_pfn > node_end_pfn)
479 node_end_pfn = zone_end_pfn;
480 if (zone->zone_start_pfn < node_start_pfn)
481 node_start_pfn = zone->zone_start_pfn;
484 pgdat->node_start_pfn = node_start_pfn;
485 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
488 void __ref remove_pfn_range_from_zone(struct zone *zone,
489 unsigned long start_pfn,
490 unsigned long nr_pages)
492 const unsigned long end_pfn = start_pfn + nr_pages;
493 struct pglist_data *pgdat = zone->zone_pgdat;
494 unsigned long pfn, cur_nr_pages, flags;
496 /* Poison struct pages because they are now uninitialized again. */
497 for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
500 /* Select all remaining pages up to the next section boundary */
502 min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
503 page_init_poison(pfn_to_page(pfn),
504 sizeof(struct page) * cur_nr_pages);
507 #ifdef CONFIG_ZONE_DEVICE
509 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
510 * we will not try to shrink the zones - which is okay as
511 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
513 if (zone_idx(zone) == ZONE_DEVICE)
517 clear_zone_contiguous(zone);
519 pgdat_resize_lock(zone->zone_pgdat, &flags);
520 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
521 update_pgdat_span(pgdat);
522 pgdat_resize_unlock(zone->zone_pgdat, &flags);
524 set_zone_contiguous(zone);
527 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
528 unsigned long map_offset,
529 struct vmem_altmap *altmap)
531 struct mem_section *ms = __pfn_to_section(pfn);
533 if (WARN_ON_ONCE(!valid_section(ms)))
536 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
540 * __remove_pages() - remove sections of pages
541 * @pfn: starting pageframe (must be aligned to start of a section)
542 * @nr_pages: number of pages to remove (must be multiple of section size)
543 * @altmap: alternative device page map or %NULL if default memmap is used
545 * Generic helper function to remove section mappings and sysfs entries
546 * for the section of the memory we are removing. Caller needs to make
547 * sure that pages are marked reserved and zones are adjust properly by
548 * calling offline_pages().
550 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
551 struct vmem_altmap *altmap)
553 const unsigned long end_pfn = pfn + nr_pages;
554 unsigned long cur_nr_pages;
555 unsigned long map_offset = 0;
557 map_offset = vmem_altmap_offset(altmap);
559 if (check_pfn_span(pfn, nr_pages, "remove"))
562 for (; pfn < end_pfn; pfn += cur_nr_pages) {
564 /* Select all remaining pages up to the next section boundary */
565 cur_nr_pages = min(end_pfn - pfn,
566 SECTION_ALIGN_UP(pfn + 1) - pfn);
567 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
572 int set_online_page_callback(online_page_callback_t callback)
577 mutex_lock(&online_page_callback_lock);
579 if (online_page_callback == generic_online_page) {
580 online_page_callback = callback;
584 mutex_unlock(&online_page_callback_lock);
589 EXPORT_SYMBOL_GPL(set_online_page_callback);
591 int restore_online_page_callback(online_page_callback_t callback)
596 mutex_lock(&online_page_callback_lock);
598 if (online_page_callback == callback) {
599 online_page_callback = generic_online_page;
603 mutex_unlock(&online_page_callback_lock);
608 EXPORT_SYMBOL_GPL(restore_online_page_callback);
610 void generic_online_page(struct page *page, unsigned int order)
613 * Freeing the page with debug_pagealloc enabled will try to unmap it,
614 * so we should map it first. This is better than introducing a special
615 * case in page freeing fast path.
617 if (debug_pagealloc_enabled_static())
618 kernel_map_pages(page, 1 << order, 1);
619 __free_pages_core(page, order);
620 totalram_pages_add(1UL << order);
621 #ifdef CONFIG_HIGHMEM
622 if (PageHighMem(page))
623 totalhigh_pages_add(1UL << order);
626 EXPORT_SYMBOL_GPL(generic_online_page);
628 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
631 const unsigned long end_pfn = start_pfn + nr_pages;
636 * Online the pages. The callback might decide to keep some pages
637 * PG_reserved (to add them to the buddy later), but we still account
638 * them as being online/belonging to this zone ("present").
640 for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
641 order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
642 /* __free_pages_core() wants pfns to be aligned to the order */
643 if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order)))
645 (*online_page_callback)(pfn_to_page(pfn), order);
648 /* mark all involved sections as online */
649 online_mem_sections(start_pfn, end_pfn);
651 *(unsigned long *)arg += nr_pages;
655 /* check which state of node_states will be changed when online memory */
656 static void node_states_check_changes_online(unsigned long nr_pages,
657 struct zone *zone, struct memory_notify *arg)
659 int nid = zone_to_nid(zone);
661 arg->status_change_nid = NUMA_NO_NODE;
662 arg->status_change_nid_normal = NUMA_NO_NODE;
663 arg->status_change_nid_high = NUMA_NO_NODE;
665 if (!node_state(nid, N_MEMORY))
666 arg->status_change_nid = nid;
667 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
668 arg->status_change_nid_normal = nid;
669 #ifdef CONFIG_HIGHMEM
670 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
671 arg->status_change_nid_high = nid;
675 static void node_states_set_node(int node, struct memory_notify *arg)
677 if (arg->status_change_nid_normal >= 0)
678 node_set_state(node, N_NORMAL_MEMORY);
680 if (arg->status_change_nid_high >= 0)
681 node_set_state(node, N_HIGH_MEMORY);
683 if (arg->status_change_nid >= 0)
684 node_set_state(node, N_MEMORY);
687 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
688 unsigned long nr_pages)
690 unsigned long old_end_pfn = zone_end_pfn(zone);
692 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
693 zone->zone_start_pfn = start_pfn;
695 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
698 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
699 unsigned long nr_pages)
701 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
703 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
704 pgdat->node_start_pfn = start_pfn;
706 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
710 * Associate the pfn range with the given zone, initializing the memmaps
711 * and resizing the pgdat/zone data to span the added pages. After this
712 * call, all affected pages are PG_reserved.
714 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
715 unsigned long nr_pages, struct vmem_altmap *altmap)
717 struct pglist_data *pgdat = zone->zone_pgdat;
718 int nid = pgdat->node_id;
721 clear_zone_contiguous(zone);
723 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
724 pgdat_resize_lock(pgdat, &flags);
725 zone_span_writelock(zone);
726 if (zone_is_empty(zone))
727 init_currently_empty_zone(zone, start_pfn, nr_pages);
728 resize_zone_range(zone, start_pfn, nr_pages);
729 zone_span_writeunlock(zone);
730 resize_pgdat_range(pgdat, start_pfn, nr_pages);
731 pgdat_resize_unlock(pgdat, &flags);
734 * TODO now we have a visible range of pages which are not associated
735 * with their zone properly. Not nice but set_pfnblock_flags_mask
736 * expects the zone spans the pfn range. All the pages in the range
737 * are reserved so nobody should be touching them so we should be safe
739 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
740 MEMINIT_HOTPLUG, altmap);
742 set_zone_contiguous(zone);
746 * Returns a default kernel memory zone for the given pfn range.
747 * If no kernel zone covers this pfn range it will automatically go
748 * to the ZONE_NORMAL.
750 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
751 unsigned long nr_pages)
753 struct pglist_data *pgdat = NODE_DATA(nid);
756 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
757 struct zone *zone = &pgdat->node_zones[zid];
759 if (zone_intersects(zone, start_pfn, nr_pages))
763 return &pgdat->node_zones[ZONE_NORMAL];
766 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
767 unsigned long nr_pages)
769 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
771 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
772 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
773 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
776 * We inherit the existing zone in a simple case where zones do not
777 * overlap in the given range
779 if (in_kernel ^ in_movable)
780 return (in_kernel) ? kernel_zone : movable_zone;
783 * If the range doesn't belong to any zone or two zones overlap in the
784 * given range then we use movable zone only if movable_node is
785 * enabled because we always online to a kernel zone by default.
787 return movable_node_enabled ? movable_zone : kernel_zone;
790 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
791 unsigned long nr_pages)
793 if (online_type == MMOP_ONLINE_KERNEL)
794 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
796 if (online_type == MMOP_ONLINE_MOVABLE)
797 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
799 return default_zone_for_pfn(nid, start_pfn, nr_pages);
802 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
803 int online_type, int nid)
806 unsigned long onlined_pages = 0;
808 int need_zonelists_rebuild = 0;
810 struct memory_notify arg;
814 /* associate pfn range with the zone */
815 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
816 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
819 arg.nr_pages = nr_pages;
820 node_states_check_changes_online(nr_pages, zone, &arg);
822 ret = memory_notify(MEM_GOING_ONLINE, &arg);
823 ret = notifier_to_errno(ret);
825 goto failed_addition;
828 * If this zone is not populated, then it is not in zonelist.
829 * This means the page allocator ignores this zone.
830 * So, zonelist must be updated after online.
832 if (!populated_zone(zone)) {
833 need_zonelists_rebuild = 1;
834 setup_zone_pageset(zone);
837 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
840 /* not a single memory resource was applicable */
841 if (need_zonelists_rebuild)
842 zone_pcp_reset(zone);
843 goto failed_addition;
846 zone->present_pages += onlined_pages;
848 pgdat_resize_lock(zone->zone_pgdat, &flags);
849 zone->zone_pgdat->node_present_pages += onlined_pages;
850 pgdat_resize_unlock(zone->zone_pgdat, &flags);
853 * When exposing larger, physically contiguous memory areas to the
854 * buddy, shuffling in the buddy (when freeing onlined pages, putting
855 * them either to the head or the tail of the freelist) is only helpful
856 * for maintaining the shuffle, but not for creating the initial
857 * shuffle. Shuffle the whole zone to make sure the just onlined pages
858 * are properly distributed across the whole freelist.
862 node_states_set_node(nid, &arg);
863 if (need_zonelists_rebuild)
864 build_all_zonelists(NULL);
865 zone_pcp_update(zone);
867 init_per_zone_wmark_min();
872 writeback_set_ratelimit();
874 memory_notify(MEM_ONLINE, &arg);
879 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
880 (unsigned long long) pfn << PAGE_SHIFT,
881 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
882 memory_notify(MEM_CANCEL_ONLINE, &arg);
883 remove_pfn_range_from_zone(zone, pfn, nr_pages);
887 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
889 static void reset_node_present_pages(pg_data_t *pgdat)
893 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
894 z->present_pages = 0;
896 pgdat->node_present_pages = 0;
899 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
900 static pg_data_t __ref *hotadd_new_pgdat(int nid)
902 struct pglist_data *pgdat;
904 pgdat = NODE_DATA(nid);
906 pgdat = arch_alloc_nodedata(nid);
910 pgdat->per_cpu_nodestats =
911 alloc_percpu(struct per_cpu_nodestat);
912 arch_refresh_nodedata(nid, pgdat);
916 * Reset the nr_zones, order and highest_zoneidx before reuse.
917 * Note that kswapd will init kswapd_highest_zoneidx properly
918 * when it starts in the near future.
921 pgdat->kswapd_order = 0;
922 pgdat->kswapd_highest_zoneidx = 0;
923 for_each_online_cpu(cpu) {
924 struct per_cpu_nodestat *p;
926 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
927 memset(p, 0, sizeof(*p));
931 /* we can use NODE_DATA(nid) from here */
932 pgdat->node_id = nid;
933 pgdat->node_start_pfn = 0;
935 /* init node's zones as empty zones, we don't have any present pages.*/
936 free_area_init_core_hotplug(nid);
939 * The node we allocated has no zone fallback lists. For avoiding
940 * to access not-initialized zonelist, build here.
942 build_all_zonelists(pgdat);
945 * When memory is hot-added, all the memory is in offline state. So
946 * clear all zones' present_pages because they will be updated in
947 * online_pages() and offline_pages().
949 reset_node_managed_pages(pgdat);
950 reset_node_present_pages(pgdat);
955 static void rollback_node_hotadd(int nid)
957 pg_data_t *pgdat = NODE_DATA(nid);
959 arch_refresh_nodedata(nid, NULL);
960 free_percpu(pgdat->per_cpu_nodestats);
961 arch_free_nodedata(pgdat);
966 * try_online_node - online a node if offlined
968 * @set_node_online: Whether we want to online the node
969 * called by cpu_up() to online a node without onlined memory.
972 * 1 -> a new node has been allocated
973 * 0 -> the node is already online
974 * -ENOMEM -> the node could not be allocated
976 static int __try_online_node(int nid, bool set_node_online)
981 if (node_online(nid))
984 pgdat = hotadd_new_pgdat(nid);
986 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
991 if (set_node_online) {
992 node_set_online(nid);
993 ret = register_one_node(nid);
1001 * Users of this function always want to online/register the node
1003 int try_online_node(int nid)
1007 mem_hotplug_begin();
1008 ret = __try_online_node(nid, true);
1013 static int check_hotplug_memory_range(u64 start, u64 size)
1015 /* memory range must be block size aligned */
1016 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1017 !IS_ALIGNED(size, memory_block_size_bytes())) {
1018 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1019 memory_block_size_bytes(), start, size);
1026 static int online_memory_block(struct memory_block *mem, void *arg)
1028 mem->online_type = memhp_default_online_type;
1029 return device_online(&mem->dev);
1033 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1034 * and online/offline operations (triggered e.g. by sysfs).
1036 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1038 int __ref add_memory_resource(int nid, struct resource *res)
1040 struct mhp_params params = { .pgprot = PAGE_KERNEL };
1042 bool new_node = false;
1046 size = resource_size(res);
1048 ret = check_hotplug_memory_range(start, size);
1052 if (!node_possible(nid)) {
1053 WARN(1, "node %d was absent from the node_possible_map\n", nid);
1057 mem_hotplug_begin();
1059 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1060 memblock_add_node(start, size, nid);
1062 ret = __try_online_node(nid, false);
1067 /* call arch's memory hotadd */
1068 ret = arch_add_memory(nid, start, size, ¶ms);
1072 /* create memory block devices after memory was added */
1073 ret = create_memory_block_devices(start, size);
1075 arch_remove_memory(nid, start, size, NULL);
1080 /* If sysfs file of new node can't be created, cpu on the node
1081 * can't be hot-added. There is no rollback way now.
1082 * So, check by BUG_ON() to catch it reluctantly..
1083 * We online node here. We can't roll back from here.
1085 node_set_online(nid);
1086 ret = __register_one_node(nid);
1090 /* link memory sections under this node.*/
1091 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1095 /* create new memmap entry */
1096 if (!strcmp(res->name, "System RAM"))
1097 firmware_map_add_hotplug(start, start + size, "System RAM");
1099 /* device_online() will take the lock when calling online_pages() */
1102 /* online pages if requested */
1103 if (memhp_default_online_type != MMOP_OFFLINE)
1104 walk_memory_blocks(start, size, NULL, online_memory_block);
1108 /* rollback pgdat allocation and others */
1110 rollback_node_hotadd(nid);
1111 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1112 memblock_remove(start, size);
1117 /* requires device_hotplug_lock, see add_memory_resource() */
1118 int __ref __add_memory(int nid, u64 start, u64 size)
1120 struct resource *res;
1123 res = register_memory_resource(start, size, "System RAM");
1125 return PTR_ERR(res);
1127 ret = add_memory_resource(nid, res);
1129 release_memory_resource(res);
1133 int add_memory(int nid, u64 start, u64 size)
1137 lock_device_hotplug();
1138 rc = __add_memory(nid, start, size);
1139 unlock_device_hotplug();
1143 EXPORT_SYMBOL_GPL(add_memory);
1146 * Add special, driver-managed memory to the system as system RAM. Such
1147 * memory is not exposed via the raw firmware-provided memmap as system
1148 * RAM, instead, it is detected and added by a driver - during cold boot,
1149 * after a reboot, and after kexec.
1151 * Reasons why this memory should not be used for the initial memmap of a
1152 * kexec kernel or for placing kexec images:
1153 * - The booting kernel is in charge of determining how this memory will be
1154 * used (e.g., use persistent memory as system RAM)
1155 * - Coordination with a hypervisor is required before this memory
1156 * can be used (e.g., inaccessible parts).
1158 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1159 * memory map") are created. Also, the created memory resource is flagged
1160 * with IORESOURCE_MEM_DRIVER_MANAGED, so in-kernel users can special-case
1161 * this memory as well (esp., not place kexec images onto it).
1163 * The resource_name (visible via /proc/iomem) has to have the format
1164 * "System RAM ($DRIVER)".
1166 int add_memory_driver_managed(int nid, u64 start, u64 size,
1167 const char *resource_name)
1169 struct resource *res;
1172 if (!resource_name ||
1173 strstr(resource_name, "System RAM (") != resource_name ||
1174 resource_name[strlen(resource_name) - 1] != ')')
1177 lock_device_hotplug();
1179 res = register_memory_resource(start, size, resource_name);
1185 rc = add_memory_resource(nid, res);
1187 release_memory_resource(res);
1190 unlock_device_hotplug();
1193 EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1195 #ifdef CONFIG_MEMORY_HOTREMOVE
1197 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1198 * memory holes). When true, return the zone.
1200 struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1201 unsigned long end_pfn)
1203 unsigned long pfn, sec_end_pfn;
1204 struct zone *zone = NULL;
1207 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1209 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1210 /* Make sure the memory section is present first */
1211 if (!present_section_nr(pfn_to_section_nr(pfn)))
1213 for (; pfn < sec_end_pfn && pfn < end_pfn;
1214 pfn += MAX_ORDER_NR_PAGES) {
1216 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1217 while ((i < MAX_ORDER_NR_PAGES) &&
1218 !pfn_valid_within(pfn + i))
1220 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1222 /* Check if we got outside of the zone */
1223 if (zone && !zone_spans_pfn(zone, pfn + i))
1225 page = pfn_to_page(pfn + i);
1226 if (zone && page_zone(page) != zone)
1228 zone = page_zone(page);
1236 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1237 * non-lru movable pages and hugepages). Will skip over most unmovable
1238 * pages (esp., pages that can be skipped when offlining), but bail out on
1239 * definitely unmovable pages.
1242 * 0 in case a movable page is found and movable_pfn was updated.
1243 * -ENOENT in case no movable page was found.
1244 * -EBUSY in case a definitely unmovable page was found.
1246 static int scan_movable_pages(unsigned long start, unsigned long end,
1247 unsigned long *movable_pfn)
1251 for (pfn = start; pfn < end; pfn++) {
1252 struct page *page, *head;
1255 if (!pfn_valid(pfn))
1257 page = pfn_to_page(pfn);
1260 if (__PageMovable(page))
1264 * PageOffline() pages that are not marked __PageMovable() and
1265 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1266 * definitely unmovable. If their reference count would be 0,
1267 * they could at least be skipped when offlining memory.
1269 if (PageOffline(page) && page_count(page))
1272 if (!PageHuge(page))
1274 head = compound_head(page);
1275 if (page_huge_active(head))
1277 skip = compound_nr(head) - (page - head);
1286 static struct page *new_node_page(struct page *page, unsigned long private)
1288 nodemask_t nmask = node_states[N_MEMORY];
1289 struct migration_target_control mtc = {
1290 .nid = page_to_nid(page),
1292 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1296 * try to allocate from a different node but reuse this node if there
1297 * are no other online nodes to be used (e.g. we are offlining a part
1298 * of the only existing node)
1300 node_clear(mtc.nid, nmask);
1301 if (nodes_empty(nmask))
1302 node_set(mtc.nid, nmask);
1304 return alloc_migration_target(page, (unsigned long)&mtc);
1308 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1311 struct page *page, *head;
1315 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1316 if (!pfn_valid(pfn))
1318 page = pfn_to_page(pfn);
1319 head = compound_head(page);
1321 if (PageHuge(page)) {
1322 pfn = page_to_pfn(head) + compound_nr(head) - 1;
1323 isolate_huge_page(head, &source);
1325 } else if (PageTransHuge(page))
1326 pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1329 * HWPoison pages have elevated reference counts so the migration would
1330 * fail on them. It also doesn't make any sense to migrate them in the
1331 * first place. Still try to unmap such a page in case it is still mapped
1332 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1333 * the unmap as the catch all safety net).
1335 if (PageHWPoison(page)) {
1336 if (WARN_ON(PageLRU(page)))
1337 isolate_lru_page(page);
1338 if (page_mapped(page))
1339 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1343 if (!get_page_unless_zero(page))
1346 * We can skip free pages. And we can deal with pages on
1347 * LRU and non-lru movable pages.
1350 ret = isolate_lru_page(page);
1352 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1353 if (!ret) { /* Success */
1354 list_add_tail(&page->lru, &source);
1355 if (!__PageMovable(page))
1356 inc_node_page_state(page, NR_ISOLATED_ANON +
1357 page_is_file_lru(page));
1360 pr_warn("failed to isolate pfn %lx\n", pfn);
1361 dump_page(page, "isolation failed");
1365 if (!list_empty(&source)) {
1366 /* Allocate a new page from the nearest neighbor node */
1367 ret = migrate_pages(&source, new_node_page, NULL, 0,
1368 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1370 list_for_each_entry(page, &source, lru) {
1371 pr_warn("migrating pfn %lx failed ret:%d ",
1372 page_to_pfn(page), ret);
1373 dump_page(page, "migration failure");
1375 putback_movable_pages(&source);
1382 /* Mark all sections offline and remove all free pages from the buddy. */
1384 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1387 unsigned long *offlined_pages = (unsigned long *)data;
1389 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1394 * Check all pages in range, recorded as memory resource, are isolated.
1397 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1400 return test_pages_isolated(start_pfn, start_pfn + nr_pages,
1404 static int __init cmdline_parse_movable_node(char *p)
1406 movable_node_enabled = true;
1409 early_param("movable_node", cmdline_parse_movable_node);
1411 /* check which state of node_states will be changed when offline memory */
1412 static void node_states_check_changes_offline(unsigned long nr_pages,
1413 struct zone *zone, struct memory_notify *arg)
1415 struct pglist_data *pgdat = zone->zone_pgdat;
1416 unsigned long present_pages = 0;
1419 arg->status_change_nid = NUMA_NO_NODE;
1420 arg->status_change_nid_normal = NUMA_NO_NODE;
1421 arg->status_change_nid_high = NUMA_NO_NODE;
1424 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1425 * If the memory to be offline is within the range
1426 * [0..ZONE_NORMAL], and it is the last present memory there,
1427 * the zones in that range will become empty after the offlining,
1428 * thus we can determine that we need to clear the node from
1429 * node_states[N_NORMAL_MEMORY].
1431 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1432 present_pages += pgdat->node_zones[zt].present_pages;
1433 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1434 arg->status_change_nid_normal = zone_to_nid(zone);
1436 #ifdef CONFIG_HIGHMEM
1438 * node_states[N_HIGH_MEMORY] contains nodes which
1439 * have normal memory or high memory.
1440 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1441 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1442 * we determine that the zones in that range become empty,
1443 * we need to clear the node for N_HIGH_MEMORY.
1445 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1446 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1447 arg->status_change_nid_high = zone_to_nid(zone);
1451 * We have accounted the pages from [0..ZONE_NORMAL), and
1452 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1454 * Here we count the possible pages from ZONE_MOVABLE.
1455 * If after having accounted all the pages, we see that the nr_pages
1456 * to be offlined is over or equal to the accounted pages,
1457 * we know that the node will become empty, and so, we can clear
1458 * it for N_MEMORY as well.
1460 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1462 if (nr_pages >= present_pages)
1463 arg->status_change_nid = zone_to_nid(zone);
1466 static void node_states_clear_node(int node, struct memory_notify *arg)
1468 if (arg->status_change_nid_normal >= 0)
1469 node_clear_state(node, N_NORMAL_MEMORY);
1471 if (arg->status_change_nid_high >= 0)
1472 node_clear_state(node, N_HIGH_MEMORY);
1474 if (arg->status_change_nid >= 0)
1475 node_clear_state(node, N_MEMORY);
1478 static int count_system_ram_pages_cb(unsigned long start_pfn,
1479 unsigned long nr_pages, void *data)
1481 unsigned long *nr_system_ram_pages = data;
1483 *nr_system_ram_pages += nr_pages;
1487 static int __ref __offline_pages(unsigned long start_pfn,
1488 unsigned long end_pfn)
1490 unsigned long pfn, nr_pages = 0;
1491 unsigned long offlined_pages = 0;
1492 int ret, node, nr_isolate_pageblock;
1493 unsigned long flags;
1495 struct memory_notify arg;
1498 mem_hotplug_begin();
1501 * Don't allow to offline memory blocks that contain holes.
1502 * Consequently, memory blocks with holes can never get onlined
1503 * via the hotplug path - online_pages() - as hotplugged memory has
1504 * no holes. This way, we e.g., don't have to worry about marking
1505 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1506 * avoid using walk_system_ram_range() later.
1508 walk_system_ram_range(start_pfn, end_pfn - start_pfn, &nr_pages,
1509 count_system_ram_pages_cb);
1510 if (nr_pages != end_pfn - start_pfn) {
1512 reason = "memory holes";
1513 goto failed_removal;
1516 /* This makes hotplug much easier...and readable.
1517 we assume this for now. .*/
1518 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1521 reason = "multizone range";
1522 goto failed_removal;
1524 node = zone_to_nid(zone);
1526 /* set above range as isolated */
1527 ret = start_isolate_page_range(start_pfn, end_pfn,
1529 MEMORY_OFFLINE | REPORT_FAILURE);
1531 reason = "failure to isolate range";
1532 goto failed_removal;
1534 nr_isolate_pageblock = ret;
1536 arg.start_pfn = start_pfn;
1537 arg.nr_pages = nr_pages;
1538 node_states_check_changes_offline(nr_pages, zone, &arg);
1540 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1541 ret = notifier_to_errno(ret);
1543 reason = "notifier failure";
1544 goto failed_removal_isolated;
1550 if (signal_pending(current)) {
1552 reason = "signal backoff";
1553 goto failed_removal_isolated;
1557 lru_add_drain_all();
1559 ret = scan_movable_pages(pfn, end_pfn, &pfn);
1562 * TODO: fatal migration failures should bail
1565 do_migrate_range(pfn, end_pfn);
1569 if (ret != -ENOENT) {
1570 reason = "unmovable page";
1571 goto failed_removal_isolated;
1575 * Dissolve free hugepages in the memory block before doing
1576 * offlining actually in order to make hugetlbfs's object
1577 * counting consistent.
1579 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1581 reason = "failure to dissolve huge pages";
1582 goto failed_removal_isolated;
1585 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1586 NULL, check_pages_isolated_cb);
1588 * per-cpu pages are drained in start_isolate_page_range, but if
1589 * there are still pages that are not free, make sure that we
1590 * drain again, because when we isolated range we might
1591 * have raced with another thread that was adding pages to pcp
1594 * Forward progress should be still guaranteed because
1595 * pages on the pcp list can only belong to MOVABLE_ZONE
1596 * because has_unmovable_pages explicitly checks for
1597 * PageBuddy on freed pages on other zones.
1600 drain_all_pages(zone);
1603 /* Ok, all of our target is isolated.
1604 We cannot do rollback at this point. */
1605 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1606 &offlined_pages, offline_isolated_pages_cb);
1607 pr_info("Offlined Pages %ld\n", offlined_pages);
1609 * Onlining will reset pagetype flags and makes migrate type
1610 * MOVABLE, so just need to decrease the number of isolated
1611 * pageblocks zone counter here.
1613 spin_lock_irqsave(&zone->lock, flags);
1614 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1615 spin_unlock_irqrestore(&zone->lock, flags);
1617 /* removal success */
1618 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1619 zone->present_pages -= offlined_pages;
1621 pgdat_resize_lock(zone->zone_pgdat, &flags);
1622 zone->zone_pgdat->node_present_pages -= offlined_pages;
1623 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1625 init_per_zone_wmark_min();
1627 if (!populated_zone(zone)) {
1628 zone_pcp_reset(zone);
1629 build_all_zonelists(NULL);
1631 zone_pcp_update(zone);
1633 node_states_clear_node(node, &arg);
1634 if (arg.status_change_nid >= 0) {
1636 kcompactd_stop(node);
1639 writeback_set_ratelimit();
1641 memory_notify(MEM_OFFLINE, &arg);
1642 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1646 failed_removal_isolated:
1647 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1648 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1650 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1651 (unsigned long long) start_pfn << PAGE_SHIFT,
1652 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1654 /* pushback to free area */
1659 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1661 return __offline_pages(start_pfn, start_pfn + nr_pages);
1664 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1666 int ret = !is_memblock_offlined(mem);
1668 if (unlikely(ret)) {
1669 phys_addr_t beginpa, endpa;
1671 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1672 endpa = beginpa + memory_block_size_bytes() - 1;
1673 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1681 static int check_cpu_on_node(pg_data_t *pgdat)
1685 for_each_present_cpu(cpu) {
1686 if (cpu_to_node(cpu) == pgdat->node_id)
1688 * the cpu on this node isn't removed, and we can't
1689 * offline this node.
1697 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1699 int nid = *(int *)arg;
1702 * If a memory block belongs to multiple nodes, the stored nid is not
1703 * reliable. However, such blocks are always online (e.g., cannot get
1704 * offlined) and, therefore, are still spanned by the node.
1706 return mem->nid == nid ? -EEXIST : 0;
1713 * Offline a node if all memory sections and cpus of the node are removed.
1715 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1716 * and online/offline operations before this call.
1718 void try_offline_node(int nid)
1720 pg_data_t *pgdat = NODE_DATA(nid);
1724 * If the node still spans pages (especially ZONE_DEVICE), don't
1725 * offline it. A node spans memory after move_pfn_range_to_zone(),
1726 * e.g., after the memory block was onlined.
1728 if (pgdat->node_spanned_pages)
1732 * Especially offline memory blocks might not be spanned by the
1733 * node. They will get spanned by the node once they get onlined.
1734 * However, they link to the node in sysfs and can get onlined later.
1736 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1740 if (check_cpu_on_node(pgdat))
1744 * all memory/cpu of this node are removed, we can offline this
1747 node_set_offline(nid);
1748 unregister_one_node(nid);
1750 EXPORT_SYMBOL(try_offline_node);
1752 static void __release_memory_resource(resource_size_t start,
1753 resource_size_t size)
1758 * When removing memory in the same granularity as it was added,
1759 * this function never fails. It might only fail if resources
1760 * have to be adjusted or split. We'll ignore the error, as
1761 * removing of memory cannot fail.
1763 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1765 resource_size_t endres = start + size - 1;
1767 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1768 &start, &endres, ret);
1772 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1776 BUG_ON(check_hotplug_memory_range(start, size));
1779 * All memory blocks must be offlined before removing memory. Check
1780 * whether all memory blocks in question are offline and return error
1781 * if this is not the case.
1783 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1787 /* remove memmap entry */
1788 firmware_map_remove(start, start + size, "System RAM");
1791 * Memory block device removal under the device_hotplug_lock is
1792 * a barrier against racing online attempts.
1794 remove_memory_block_devices(start, size);
1796 mem_hotplug_begin();
1798 arch_remove_memory(nid, start, size, NULL);
1800 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1801 memblock_free(start, size);
1802 memblock_remove(start, size);
1805 __release_memory_resource(start, size);
1807 try_offline_node(nid);
1816 * @start: physical address of the region to remove
1817 * @size: size of the region to remove
1819 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1820 * and online/offline operations before this call, as required by
1821 * try_offline_node().
1823 void __remove_memory(int nid, u64 start, u64 size)
1827 * trigger BUG() if some memory is not offlined prior to calling this
1830 if (try_remove_memory(nid, start, size))
1835 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1836 * some memory is not offline
1838 int remove_memory(int nid, u64 start, u64 size)
1842 lock_device_hotplug();
1843 rc = try_remove_memory(nid, start, size);
1844 unlock_device_hotplug();
1848 EXPORT_SYMBOL_GPL(remove_memory);
1851 * Try to offline and remove a memory block. Might take a long time to
1852 * finish in case memory is still in use. Primarily useful for memory devices
1853 * that logically unplugged all memory (so it's no longer in use) and want to
1854 * offline + remove the memory block.
1856 int offline_and_remove_memory(int nid, u64 start, u64 size)
1858 struct memory_block *mem;
1861 if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1862 size != memory_block_size_bytes())
1865 lock_device_hotplug();
1866 mem = find_memory_block(__pfn_to_section(PFN_DOWN(start)));
1868 rc = device_offline(&mem->dev);
1869 /* Ignore if the device is already offline. */
1874 * In case we succeeded to offline the memory block, remove it.
1875 * This cannot fail as it cannot get onlined in the meantime.
1878 rc = try_remove_memory(nid, start, size);
1881 unlock_device_hotplug();
1885 EXPORT_SYMBOL_GPL(offline_and_remove_memory);
1886 #endif /* CONFIG_MEMORY_HOTREMOVE */