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 mhp_default_online_type = MMOP_OFFLINE;
72 int mhp_default_online_type = MMOP_ONLINE;
75 static int __init setup_memhp_default_state(char *str)
77 const int online_type = mhp_online_type_from_str(str);
80 mhp_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_SYSRAM_DRIVER_MANAGED;
110 if (!mhp_range_allowed(start, size, true))
111 return ERR_PTR(-E2BIG);
114 * Make sure value parsed from 'mem=' only restricts memory adding
115 * while booting, so that memory hotplug won't be impacted. Please
116 * refer to document of 'mem=' in kernel-parameters.txt for more
119 if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
120 return ERR_PTR(-E2BIG);
123 * Request ownership of the new memory range. This might be
124 * a child of an existing resource that was present but
125 * not marked as busy.
127 res = __request_region(&iomem_resource, start, size,
128 resource_name, flags);
131 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
132 start, start + size);
133 return ERR_PTR(-EEXIST);
138 static void release_memory_resource(struct resource *res)
142 release_resource(res);
146 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
147 void get_page_bootmem(unsigned long info, struct page *page,
150 page->freelist = (void *)type;
151 SetPagePrivate(page);
152 set_page_private(page, info);
156 void put_page_bootmem(struct page *page)
160 type = (unsigned long) page->freelist;
161 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
162 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
164 if (page_ref_dec_return(page) == 1) {
165 page->freelist = NULL;
166 ClearPagePrivate(page);
167 set_page_private(page, 0);
168 INIT_LIST_HEAD(&page->lru);
169 free_reserved_page(page);
173 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
174 #ifndef CONFIG_SPARSEMEM_VMEMMAP
175 static void register_page_bootmem_info_section(unsigned long start_pfn)
177 unsigned long mapsize, section_nr, i;
178 struct mem_section *ms;
179 struct page *page, *memmap;
180 struct mem_section_usage *usage;
182 section_nr = pfn_to_section_nr(start_pfn);
183 ms = __nr_to_section(section_nr);
185 /* Get section's memmap address */
186 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
189 * Get page for the memmap's phys address
190 * XXX: need more consideration for sparse_vmemmap...
192 page = virt_to_page(memmap);
193 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
194 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
196 /* remember memmap's page */
197 for (i = 0; i < mapsize; i++, page++)
198 get_page_bootmem(section_nr, page, SECTION_INFO);
201 page = virt_to_page(usage);
203 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
205 for (i = 0; i < mapsize; i++, page++)
206 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
209 #else /* CONFIG_SPARSEMEM_VMEMMAP */
210 static void register_page_bootmem_info_section(unsigned long start_pfn)
212 unsigned long mapsize, section_nr, i;
213 struct mem_section *ms;
214 struct page *page, *memmap;
215 struct mem_section_usage *usage;
217 section_nr = pfn_to_section_nr(start_pfn);
218 ms = __nr_to_section(section_nr);
220 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
222 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
225 page = virt_to_page(usage);
227 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
229 for (i = 0; i < mapsize; i++, page++)
230 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
232 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
234 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
236 unsigned long i, pfn, end_pfn, nr_pages;
237 int node = pgdat->node_id;
240 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
241 page = virt_to_page(pgdat);
243 for (i = 0; i < nr_pages; i++, page++)
244 get_page_bootmem(node, page, NODE_INFO);
246 pfn = pgdat->node_start_pfn;
247 end_pfn = pgdat_end_pfn(pgdat);
249 /* register section info */
250 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
252 * Some platforms can assign the same pfn to multiple nodes - on
253 * node0 as well as nodeN. To avoid registering a pfn against
254 * multiple nodes we check that this pfn does not already
255 * reside in some other nodes.
257 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
258 register_page_bootmem_info_section(pfn);
261 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
263 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
267 * Disallow all operations smaller than a sub-section and only
268 * allow operations smaller than a section for
269 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
270 * enforces a larger memory_block_size_bytes() granularity for
271 * memory that will be marked online, so this check should only
272 * fire for direct arch_{add,remove}_memory() users outside of
273 * add_memory_resource().
275 unsigned long min_align;
277 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
278 min_align = PAGES_PER_SUBSECTION;
280 min_align = PAGES_PER_SECTION;
281 if (!IS_ALIGNED(pfn, min_align)
282 || !IS_ALIGNED(nr_pages, min_align)) {
283 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
284 reason, pfn, pfn + nr_pages - 1);
291 * Return page for the valid pfn only if the page is online. All pfn
292 * walkers which rely on the fully initialized page->flags and others
293 * should use this rather than pfn_valid && pfn_to_page
295 struct page *pfn_to_online_page(unsigned long pfn)
297 unsigned long nr = pfn_to_section_nr(pfn);
298 struct dev_pagemap *pgmap;
299 struct mem_section *ms;
301 if (nr >= NR_MEM_SECTIONS)
304 ms = __nr_to_section(nr);
305 if (!online_section(ms))
309 * Save some code text when online_section() +
310 * pfn_section_valid() are sufficient.
312 if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn))
315 if (!pfn_section_valid(ms, pfn))
318 if (!online_device_section(ms))
319 return pfn_to_page(pfn);
322 * Slowpath: when ZONE_DEVICE collides with
323 * ZONE_{NORMAL,MOVABLE} within the same section some pfns in
324 * the section may be 'offline' but 'valid'. Only
325 * get_dev_pagemap() can determine sub-section online status.
327 pgmap = get_dev_pagemap(pfn, NULL);
328 put_dev_pagemap(pgmap);
330 /* The presence of a pgmap indicates ZONE_DEVICE offline pfn */
334 return pfn_to_page(pfn);
336 EXPORT_SYMBOL_GPL(pfn_to_online_page);
339 * Reasonably generic function for adding memory. It is
340 * expected that archs that support memory hotplug will
341 * call this function after deciding the zone to which to
344 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
345 struct mhp_params *params)
347 const unsigned long end_pfn = pfn + nr_pages;
348 unsigned long cur_nr_pages;
350 struct vmem_altmap *altmap = params->altmap;
352 if (WARN_ON_ONCE(!params->pgprot.pgprot))
355 VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false));
359 * Validate altmap is within bounds of the total request
361 if (altmap->base_pfn != pfn
362 || vmem_altmap_offset(altmap) > nr_pages) {
363 pr_warn_once("memory add fail, invalid altmap\n");
369 err = check_pfn_span(pfn, nr_pages, "add");
373 for (; pfn < end_pfn; pfn += cur_nr_pages) {
374 /* Select all remaining pages up to the next section boundary */
375 cur_nr_pages = min(end_pfn - pfn,
376 SECTION_ALIGN_UP(pfn + 1) - pfn);
377 err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
382 vmemmap_populate_print_last();
386 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
387 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
388 unsigned long start_pfn,
389 unsigned long end_pfn)
391 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
392 if (unlikely(!pfn_to_online_page(start_pfn)))
395 if (unlikely(pfn_to_nid(start_pfn) != nid))
398 if (zone != page_zone(pfn_to_page(start_pfn)))
407 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
408 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
409 unsigned long start_pfn,
410 unsigned long end_pfn)
414 /* pfn is the end pfn of a memory section. */
416 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
417 if (unlikely(!pfn_to_online_page(pfn)))
420 if (unlikely(pfn_to_nid(pfn) != nid))
423 if (zone != page_zone(pfn_to_page(pfn)))
432 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
433 unsigned long end_pfn)
436 int nid = zone_to_nid(zone);
438 zone_span_writelock(zone);
439 if (zone->zone_start_pfn == start_pfn) {
441 * If the section is smallest section in the zone, it need
442 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
443 * In this case, we find second smallest valid mem_section
444 * for shrinking zone.
446 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
449 zone->spanned_pages = zone_end_pfn(zone) - pfn;
450 zone->zone_start_pfn = pfn;
452 zone->zone_start_pfn = 0;
453 zone->spanned_pages = 0;
455 } else if (zone_end_pfn(zone) == end_pfn) {
457 * If the section is biggest section in the zone, it need
458 * shrink zone->spanned_pages.
459 * In this case, we find second biggest valid mem_section for
462 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
465 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
467 zone->zone_start_pfn = 0;
468 zone->spanned_pages = 0;
471 zone_span_writeunlock(zone);
474 static void update_pgdat_span(struct pglist_data *pgdat)
476 unsigned long node_start_pfn = 0, node_end_pfn = 0;
479 for (zone = pgdat->node_zones;
480 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
481 unsigned long end_pfn = zone_end_pfn(zone);
483 /* No need to lock the zones, they can't change. */
484 if (!zone->spanned_pages)
487 node_start_pfn = zone->zone_start_pfn;
488 node_end_pfn = end_pfn;
492 if (end_pfn > node_end_pfn)
493 node_end_pfn = end_pfn;
494 if (zone->zone_start_pfn < node_start_pfn)
495 node_start_pfn = zone->zone_start_pfn;
498 pgdat->node_start_pfn = node_start_pfn;
499 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
502 void __ref remove_pfn_range_from_zone(struct zone *zone,
503 unsigned long start_pfn,
504 unsigned long nr_pages)
506 const unsigned long end_pfn = start_pfn + nr_pages;
507 struct pglist_data *pgdat = zone->zone_pgdat;
508 unsigned long pfn, cur_nr_pages, flags;
510 /* Poison struct pages because they are now uninitialized again. */
511 for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
514 /* Select all remaining pages up to the next section boundary */
516 min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
517 page_init_poison(pfn_to_page(pfn),
518 sizeof(struct page) * cur_nr_pages);
521 #ifdef CONFIG_ZONE_DEVICE
523 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
524 * we will not try to shrink the zones - which is okay as
525 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
527 if (zone_idx(zone) == ZONE_DEVICE)
531 clear_zone_contiguous(zone);
533 pgdat_resize_lock(zone->zone_pgdat, &flags);
534 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
535 update_pgdat_span(pgdat);
536 pgdat_resize_unlock(zone->zone_pgdat, &flags);
538 set_zone_contiguous(zone);
541 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
542 unsigned long map_offset,
543 struct vmem_altmap *altmap)
545 struct mem_section *ms = __pfn_to_section(pfn);
547 if (WARN_ON_ONCE(!valid_section(ms)))
550 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
554 * __remove_pages() - remove sections of pages
555 * @pfn: starting pageframe (must be aligned to start of a section)
556 * @nr_pages: number of pages to remove (must be multiple of section size)
557 * @altmap: alternative device page map or %NULL if default memmap is used
559 * Generic helper function to remove section mappings and sysfs entries
560 * for the section of the memory we are removing. Caller needs to make
561 * sure that pages are marked reserved and zones are adjust properly by
562 * calling offline_pages().
564 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
565 struct vmem_altmap *altmap)
567 const unsigned long end_pfn = pfn + nr_pages;
568 unsigned long cur_nr_pages;
569 unsigned long map_offset = 0;
571 map_offset = vmem_altmap_offset(altmap);
573 if (check_pfn_span(pfn, nr_pages, "remove"))
576 for (; pfn < end_pfn; pfn += cur_nr_pages) {
578 /* Select all remaining pages up to the next section boundary */
579 cur_nr_pages = min(end_pfn - pfn,
580 SECTION_ALIGN_UP(pfn + 1) - pfn);
581 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
586 int set_online_page_callback(online_page_callback_t callback)
591 mutex_lock(&online_page_callback_lock);
593 if (online_page_callback == generic_online_page) {
594 online_page_callback = callback;
598 mutex_unlock(&online_page_callback_lock);
603 EXPORT_SYMBOL_GPL(set_online_page_callback);
605 int restore_online_page_callback(online_page_callback_t callback)
610 mutex_lock(&online_page_callback_lock);
612 if (online_page_callback == callback) {
613 online_page_callback = generic_online_page;
617 mutex_unlock(&online_page_callback_lock);
622 EXPORT_SYMBOL_GPL(restore_online_page_callback);
624 void generic_online_page(struct page *page, unsigned int order)
627 * Freeing the page with debug_pagealloc enabled will try to unmap it,
628 * so we should map it first. This is better than introducing a special
629 * case in page freeing fast path.
631 debug_pagealloc_map_pages(page, 1 << order);
632 __free_pages_core(page, order);
633 totalram_pages_add(1UL << order);
634 #ifdef CONFIG_HIGHMEM
635 if (PageHighMem(page))
636 totalhigh_pages_add(1UL << order);
639 EXPORT_SYMBOL_GPL(generic_online_page);
641 static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
643 const unsigned long end_pfn = start_pfn + nr_pages;
647 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
648 * decide to not expose all pages to the buddy (e.g., expose them
649 * later). We account all pages as being online and belonging to this
652 for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES)
653 (*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1);
655 /* mark all involved sections as online */
656 online_mem_sections(start_pfn, end_pfn);
659 /* check which state of node_states will be changed when online memory */
660 static void node_states_check_changes_online(unsigned long nr_pages,
661 struct zone *zone, struct memory_notify *arg)
663 int nid = zone_to_nid(zone);
665 arg->status_change_nid = NUMA_NO_NODE;
666 arg->status_change_nid_normal = NUMA_NO_NODE;
667 arg->status_change_nid_high = NUMA_NO_NODE;
669 if (!node_state(nid, N_MEMORY))
670 arg->status_change_nid = nid;
671 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
672 arg->status_change_nid_normal = nid;
673 #ifdef CONFIG_HIGHMEM
674 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
675 arg->status_change_nid_high = nid;
679 static void node_states_set_node(int node, struct memory_notify *arg)
681 if (arg->status_change_nid_normal >= 0)
682 node_set_state(node, N_NORMAL_MEMORY);
684 if (arg->status_change_nid_high >= 0)
685 node_set_state(node, N_HIGH_MEMORY);
687 if (arg->status_change_nid >= 0)
688 node_set_state(node, N_MEMORY);
691 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
692 unsigned long nr_pages)
694 unsigned long old_end_pfn = zone_end_pfn(zone);
696 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
697 zone->zone_start_pfn = start_pfn;
699 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
702 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
703 unsigned long nr_pages)
705 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
707 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
708 pgdat->node_start_pfn = start_pfn;
710 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
714 static void section_taint_zone_device(unsigned long pfn)
716 struct mem_section *ms = __pfn_to_section(pfn);
718 ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE;
722 * Associate the pfn range with the given zone, initializing the memmaps
723 * and resizing the pgdat/zone data to span the added pages. After this
724 * call, all affected pages are PG_reserved.
726 * All aligned pageblocks are initialized to the specified migratetype
727 * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
728 * zone stats (e.g., nr_isolate_pageblock) are touched.
730 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
731 unsigned long nr_pages,
732 struct vmem_altmap *altmap, int migratetype)
734 struct pglist_data *pgdat = zone->zone_pgdat;
735 int nid = pgdat->node_id;
738 clear_zone_contiguous(zone);
740 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
741 pgdat_resize_lock(pgdat, &flags);
742 zone_span_writelock(zone);
743 if (zone_is_empty(zone))
744 init_currently_empty_zone(zone, start_pfn, nr_pages);
745 resize_zone_range(zone, start_pfn, nr_pages);
746 zone_span_writeunlock(zone);
747 resize_pgdat_range(pgdat, start_pfn, nr_pages);
748 pgdat_resize_unlock(pgdat, &flags);
751 * Subsection population requires care in pfn_to_online_page().
752 * Set the taint to enable the slow path detection of
753 * ZONE_DEVICE pages in an otherwise ZONE_{NORMAL,MOVABLE}
756 if (zone_is_zone_device(zone)) {
757 if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION))
758 section_taint_zone_device(start_pfn);
759 if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))
760 section_taint_zone_device(start_pfn + nr_pages);
764 * TODO now we have a visible range of pages which are not associated
765 * with their zone properly. Not nice but set_pfnblock_flags_mask
766 * expects the zone spans the pfn range. All the pages in the range
767 * are reserved so nobody should be touching them so we should be safe
769 memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0,
770 MEMINIT_HOTPLUG, altmap, migratetype);
772 set_zone_contiguous(zone);
776 * Returns a default kernel memory zone for the given pfn range.
777 * If no kernel zone covers this pfn range it will automatically go
778 * to the ZONE_NORMAL.
780 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
781 unsigned long nr_pages)
783 struct pglist_data *pgdat = NODE_DATA(nid);
786 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
787 struct zone *zone = &pgdat->node_zones[zid];
789 if (zone_intersects(zone, start_pfn, nr_pages))
793 return &pgdat->node_zones[ZONE_NORMAL];
796 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
797 unsigned long nr_pages)
799 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
801 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
802 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
803 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
806 * We inherit the existing zone in a simple case where zones do not
807 * overlap in the given range
809 if (in_kernel ^ in_movable)
810 return (in_kernel) ? kernel_zone : movable_zone;
813 * If the range doesn't belong to any zone or two zones overlap in the
814 * given range then we use movable zone only if movable_node is
815 * enabled because we always online to a kernel zone by default.
817 return movable_node_enabled ? movable_zone : kernel_zone;
820 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
821 unsigned long nr_pages)
823 if (online_type == MMOP_ONLINE_KERNEL)
824 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
826 if (online_type == MMOP_ONLINE_MOVABLE)
827 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
829 return default_zone_for_pfn(nid, start_pfn, nr_pages);
832 static void adjust_present_page_count(struct zone *zone, long nr_pages)
836 zone->present_pages += nr_pages;
837 pgdat_resize_lock(zone->zone_pgdat, &flags);
838 zone->zone_pgdat->node_present_pages += nr_pages;
839 pgdat_resize_unlock(zone->zone_pgdat, &flags);
842 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
843 int online_type, int nid)
847 int need_zonelists_rebuild = 0;
849 struct memory_notify arg;
852 * {on,off}lining is constrained to full memory sections (or more
853 * precisly to memory blocks from the user space POV).
854 * memmap_on_memory is an exception because it reserves initial part
855 * of the physical memory space for vmemmaps. That space is pageblock
858 if (WARN_ON_ONCE(!nr_pages ||
859 !IS_ALIGNED(pfn, pageblock_nr_pages) ||
860 !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION)))
865 /* associate pfn range with the zone */
866 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
867 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
870 arg.nr_pages = nr_pages;
871 node_states_check_changes_online(nr_pages, zone, &arg);
873 ret = memory_notify(MEM_GOING_ONLINE, &arg);
874 ret = notifier_to_errno(ret);
876 goto failed_addition;
879 * Fixup the number of isolated pageblocks before marking the sections
880 * onlining, such that undo_isolate_page_range() works correctly.
882 spin_lock_irqsave(&zone->lock, flags);
883 zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
884 spin_unlock_irqrestore(&zone->lock, flags);
887 * If this zone is not populated, then it is not in zonelist.
888 * This means the page allocator ignores this zone.
889 * So, zonelist must be updated after online.
891 if (!populated_zone(zone)) {
892 need_zonelists_rebuild = 1;
893 setup_zone_pageset(zone);
896 online_pages_range(pfn, nr_pages);
897 adjust_present_page_count(zone, nr_pages);
899 node_states_set_node(nid, &arg);
900 if (need_zonelists_rebuild)
901 build_all_zonelists(NULL);
902 zone_pcp_update(zone);
904 /* Basic onlining is complete, allow allocation of onlined pages. */
905 undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
908 * Freshly onlined pages aren't shuffled (e.g., all pages are placed to
909 * the tail of the freelist when undoing isolation). Shuffle the whole
910 * zone to make sure the just onlined pages are properly distributed
911 * across the whole freelist - to create an initial shuffle.
915 init_per_zone_wmark_min();
920 writeback_set_ratelimit();
922 memory_notify(MEM_ONLINE, &arg);
927 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
928 (unsigned long long) pfn << PAGE_SHIFT,
929 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
930 memory_notify(MEM_CANCEL_ONLINE, &arg);
931 remove_pfn_range_from_zone(zone, pfn, nr_pages);
935 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
937 static void reset_node_present_pages(pg_data_t *pgdat)
941 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
942 z->present_pages = 0;
944 pgdat->node_present_pages = 0;
947 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
948 static pg_data_t __ref *hotadd_new_pgdat(int nid)
950 struct pglist_data *pgdat;
952 pgdat = NODE_DATA(nid);
954 pgdat = arch_alloc_nodedata(nid);
958 pgdat->per_cpu_nodestats =
959 alloc_percpu(struct per_cpu_nodestat);
960 arch_refresh_nodedata(nid, pgdat);
964 * Reset the nr_zones, order and highest_zoneidx before reuse.
965 * Note that kswapd will init kswapd_highest_zoneidx properly
966 * when it starts in the near future.
969 pgdat->kswapd_order = 0;
970 pgdat->kswapd_highest_zoneidx = 0;
971 for_each_online_cpu(cpu) {
972 struct per_cpu_nodestat *p;
974 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
975 memset(p, 0, sizeof(*p));
979 /* we can use NODE_DATA(nid) from here */
980 pgdat->node_id = nid;
981 pgdat->node_start_pfn = 0;
983 /* init node's zones as empty zones, we don't have any present pages.*/
984 free_area_init_core_hotplug(nid);
987 * The node we allocated has no zone fallback lists. For avoiding
988 * to access not-initialized zonelist, build here.
990 build_all_zonelists(pgdat);
993 * When memory is hot-added, all the memory is in offline state. So
994 * clear all zones' present_pages because they will be updated in
995 * online_pages() and offline_pages().
997 reset_node_managed_pages(pgdat);
998 reset_node_present_pages(pgdat);
1003 static void rollback_node_hotadd(int nid)
1005 pg_data_t *pgdat = NODE_DATA(nid);
1007 arch_refresh_nodedata(nid, NULL);
1008 free_percpu(pgdat->per_cpu_nodestats);
1009 arch_free_nodedata(pgdat);
1014 * try_online_node - online a node if offlined
1016 * @set_node_online: Whether we want to online the node
1017 * called by cpu_up() to online a node without onlined memory.
1020 * 1 -> a new node has been allocated
1021 * 0 -> the node is already online
1022 * -ENOMEM -> the node could not be allocated
1024 static int __try_online_node(int nid, bool set_node_online)
1029 if (node_online(nid))
1032 pgdat = hotadd_new_pgdat(nid);
1034 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1039 if (set_node_online) {
1040 node_set_online(nid);
1041 ret = register_one_node(nid);
1049 * Users of this function always want to online/register the node
1051 int try_online_node(int nid)
1055 mem_hotplug_begin();
1056 ret = __try_online_node(nid, true);
1061 static int check_hotplug_memory_range(u64 start, u64 size)
1063 /* memory range must be block size aligned */
1064 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1065 !IS_ALIGNED(size, memory_block_size_bytes())) {
1066 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1067 memory_block_size_bytes(), start, size);
1074 static int online_memory_block(struct memory_block *mem, void *arg)
1076 mem->online_type = mhp_default_online_type;
1077 return device_online(&mem->dev);
1081 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1082 * and online/offline operations (triggered e.g. by sysfs).
1084 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1086 int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1088 struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
1090 bool new_node = false;
1094 size = resource_size(res);
1096 ret = check_hotplug_memory_range(start, size);
1100 if (!node_possible(nid)) {
1101 WARN(1, "node %d was absent from the node_possible_map\n", nid);
1105 mem_hotplug_begin();
1107 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1108 memblock_add_node(start, size, nid);
1110 ret = __try_online_node(nid, false);
1115 /* call arch's memory hotadd */
1116 ret = arch_add_memory(nid, start, size, ¶ms);
1120 /* create memory block devices after memory was added */
1121 ret = create_memory_block_devices(start, size);
1123 arch_remove_memory(nid, start, size, NULL);
1128 /* If sysfs file of new node can't be created, cpu on the node
1129 * can't be hot-added. There is no rollback way now.
1130 * So, check by BUG_ON() to catch it reluctantly..
1131 * We online node here. We can't roll back from here.
1133 node_set_online(nid);
1134 ret = __register_one_node(nid);
1138 /* link memory sections under this node.*/
1139 link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1142 /* create new memmap entry */
1143 if (!strcmp(res->name, "System RAM"))
1144 firmware_map_add_hotplug(start, start + size, "System RAM");
1146 /* device_online() will take the lock when calling online_pages() */
1150 * In case we're allowed to merge the resource, flag it and trigger
1151 * merging now that adding succeeded.
1153 if (mhp_flags & MHP_MERGE_RESOURCE)
1154 merge_system_ram_resource(res);
1156 /* online pages if requested */
1157 if (mhp_default_online_type != MMOP_OFFLINE)
1158 walk_memory_blocks(start, size, NULL, online_memory_block);
1162 /* rollback pgdat allocation and others */
1164 rollback_node_hotadd(nid);
1165 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1166 memblock_remove(start, size);
1171 /* requires device_hotplug_lock, see add_memory_resource() */
1172 int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1174 struct resource *res;
1177 res = register_memory_resource(start, size, "System RAM");
1179 return PTR_ERR(res);
1181 ret = add_memory_resource(nid, res, mhp_flags);
1183 release_memory_resource(res);
1187 int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1191 lock_device_hotplug();
1192 rc = __add_memory(nid, start, size, mhp_flags);
1193 unlock_device_hotplug();
1197 EXPORT_SYMBOL_GPL(add_memory);
1200 * Add special, driver-managed memory to the system as system RAM. Such
1201 * memory is not exposed via the raw firmware-provided memmap as system
1202 * RAM, instead, it is detected and added by a driver - during cold boot,
1203 * after a reboot, and after kexec.
1205 * Reasons why this memory should not be used for the initial memmap of a
1206 * kexec kernel or for placing kexec images:
1207 * - The booting kernel is in charge of determining how this memory will be
1208 * used (e.g., use persistent memory as system RAM)
1209 * - Coordination with a hypervisor is required before this memory
1210 * can be used (e.g., inaccessible parts).
1212 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1213 * memory map") are created. Also, the created memory resource is flagged
1214 * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
1215 * this memory as well (esp., not place kexec images onto it).
1217 * The resource_name (visible via /proc/iomem) has to have the format
1218 * "System RAM ($DRIVER)".
1220 int add_memory_driver_managed(int nid, u64 start, u64 size,
1221 const char *resource_name, mhp_t mhp_flags)
1223 struct resource *res;
1226 if (!resource_name ||
1227 strstr(resource_name, "System RAM (") != resource_name ||
1228 resource_name[strlen(resource_name) - 1] != ')')
1231 lock_device_hotplug();
1233 res = register_memory_resource(start, size, resource_name);
1239 rc = add_memory_resource(nid, res, mhp_flags);
1241 release_memory_resource(res);
1244 unlock_device_hotplug();
1247 EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1250 * Platforms should define arch_get_mappable_range() that provides
1251 * maximum possible addressable physical memory range for which the
1252 * linear mapping could be created. The platform returned address
1253 * range must adhere to these following semantics.
1255 * - range.start <= range.end
1256 * - Range includes both end points [range.start..range.end]
1258 * There is also a fallback definition provided here, allowing the
1259 * entire possible physical address range in case any platform does
1260 * not define arch_get_mappable_range().
1262 struct range __weak arch_get_mappable_range(void)
1264 struct range mhp_range = {
1271 struct range mhp_get_pluggable_range(bool need_mapping)
1273 const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1;
1274 struct range mhp_range;
1277 mhp_range = arch_get_mappable_range();
1278 if (mhp_range.start > max_phys) {
1279 mhp_range.start = 0;
1282 mhp_range.end = min_t(u64, mhp_range.end, max_phys);
1284 mhp_range.start = 0;
1285 mhp_range.end = max_phys;
1289 EXPORT_SYMBOL_GPL(mhp_get_pluggable_range);
1291 bool mhp_range_allowed(u64 start, u64 size, bool need_mapping)
1293 struct range mhp_range = mhp_get_pluggable_range(need_mapping);
1294 u64 end = start + size;
1296 if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end)
1299 pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n",
1300 start, end, mhp_range.start, mhp_range.end);
1304 #ifdef CONFIG_MEMORY_HOTREMOVE
1306 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1307 * memory holes). When true, return the zone.
1309 struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1310 unsigned long end_pfn)
1312 unsigned long pfn, sec_end_pfn;
1313 struct zone *zone = NULL;
1316 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1318 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1319 /* Make sure the memory section is present first */
1320 if (!present_section_nr(pfn_to_section_nr(pfn)))
1322 for (; pfn < sec_end_pfn && pfn < end_pfn;
1323 pfn += MAX_ORDER_NR_PAGES) {
1325 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1326 while ((i < MAX_ORDER_NR_PAGES) &&
1327 !pfn_valid_within(pfn + i))
1329 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1331 /* Check if we got outside of the zone */
1332 if (zone && !zone_spans_pfn(zone, pfn + i))
1334 page = pfn_to_page(pfn + i);
1335 if (zone && page_zone(page) != zone)
1337 zone = page_zone(page);
1345 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1346 * non-lru movable pages and hugepages). Will skip over most unmovable
1347 * pages (esp., pages that can be skipped when offlining), but bail out on
1348 * definitely unmovable pages.
1351 * 0 in case a movable page is found and movable_pfn was updated.
1352 * -ENOENT in case no movable page was found.
1353 * -EBUSY in case a definitely unmovable page was found.
1355 static int scan_movable_pages(unsigned long start, unsigned long end,
1356 unsigned long *movable_pfn)
1360 for (pfn = start; pfn < end; pfn++) {
1361 struct page *page, *head;
1364 if (!pfn_valid(pfn))
1366 page = pfn_to_page(pfn);
1369 if (__PageMovable(page))
1373 * PageOffline() pages that are not marked __PageMovable() and
1374 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1375 * definitely unmovable. If their reference count would be 0,
1376 * they could at least be skipped when offlining memory.
1378 if (PageOffline(page) && page_count(page))
1381 if (!PageHuge(page))
1383 head = compound_head(page);
1385 * This test is racy as we hold no reference or lock. The
1386 * hugetlb page could have been free'ed and head is no longer
1387 * a hugetlb page before the following check. In such unlikely
1388 * cases false positives and negatives are possible. Calling
1389 * code must deal with these scenarios.
1391 if (HPageMigratable(head))
1393 skip = compound_nr(head) - (page - head);
1403 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1406 struct page *page, *head;
1410 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1411 if (!pfn_valid(pfn))
1413 page = pfn_to_page(pfn);
1414 head = compound_head(page);
1416 if (PageHuge(page)) {
1417 pfn = page_to_pfn(head) + compound_nr(head) - 1;
1418 isolate_huge_page(head, &source);
1420 } else if (PageTransHuge(page))
1421 pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1424 * HWPoison pages have elevated reference counts so the migration would
1425 * fail on them. It also doesn't make any sense to migrate them in the
1426 * first place. Still try to unmap such a page in case it is still mapped
1427 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1428 * the unmap as the catch all safety net).
1430 if (PageHWPoison(page)) {
1431 if (WARN_ON(PageLRU(page)))
1432 isolate_lru_page(page);
1433 if (page_mapped(page))
1434 try_to_unmap(page, TTU_IGNORE_MLOCK);
1438 if (!get_page_unless_zero(page))
1441 * We can skip free pages. And we can deal with pages on
1442 * LRU and non-lru movable pages.
1445 ret = isolate_lru_page(page);
1447 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1448 if (!ret) { /* Success */
1449 list_add_tail(&page->lru, &source);
1450 if (!__PageMovable(page))
1451 inc_node_page_state(page, NR_ISOLATED_ANON +
1452 page_is_file_lru(page));
1455 pr_warn("failed to isolate pfn %lx\n", pfn);
1456 dump_page(page, "isolation failed");
1460 if (!list_empty(&source)) {
1461 nodemask_t nmask = node_states[N_MEMORY];
1462 struct migration_target_control mtc = {
1464 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1468 * We have checked that migration range is on a single zone so
1469 * we can use the nid of the first page to all the others.
1471 mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1474 * try to allocate from a different node but reuse this node
1475 * if there are no other online nodes to be used (e.g. we are
1476 * offlining a part of the only existing node)
1478 node_clear(mtc.nid, nmask);
1479 if (nodes_empty(nmask))
1480 node_set(mtc.nid, nmask);
1481 ret = migrate_pages(&source, alloc_migration_target, NULL,
1482 (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1484 list_for_each_entry(page, &source, lru) {
1485 pr_warn("migrating pfn %lx failed ret:%d ",
1486 page_to_pfn(page), ret);
1487 dump_page(page, "migration failure");
1489 putback_movable_pages(&source);
1496 static int __init cmdline_parse_movable_node(char *p)
1498 movable_node_enabled = true;
1501 early_param("movable_node", cmdline_parse_movable_node);
1503 /* check which state of node_states will be changed when offline memory */
1504 static void node_states_check_changes_offline(unsigned long nr_pages,
1505 struct zone *zone, struct memory_notify *arg)
1507 struct pglist_data *pgdat = zone->zone_pgdat;
1508 unsigned long present_pages = 0;
1511 arg->status_change_nid = NUMA_NO_NODE;
1512 arg->status_change_nid_normal = NUMA_NO_NODE;
1513 arg->status_change_nid_high = NUMA_NO_NODE;
1516 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1517 * If the memory to be offline is within the range
1518 * [0..ZONE_NORMAL], and it is the last present memory there,
1519 * the zones in that range will become empty after the offlining,
1520 * thus we can determine that we need to clear the node from
1521 * node_states[N_NORMAL_MEMORY].
1523 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1524 present_pages += pgdat->node_zones[zt].present_pages;
1525 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1526 arg->status_change_nid_normal = zone_to_nid(zone);
1528 #ifdef CONFIG_HIGHMEM
1530 * node_states[N_HIGH_MEMORY] contains nodes which
1531 * have normal memory or high memory.
1532 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1533 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1534 * we determine that the zones in that range become empty,
1535 * we need to clear the node for N_HIGH_MEMORY.
1537 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1538 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1539 arg->status_change_nid_high = zone_to_nid(zone);
1543 * We have accounted the pages from [0..ZONE_NORMAL), and
1544 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1546 * Here we count the possible pages from ZONE_MOVABLE.
1547 * If after having accounted all the pages, we see that the nr_pages
1548 * to be offlined is over or equal to the accounted pages,
1549 * we know that the node will become empty, and so, we can clear
1550 * it for N_MEMORY as well.
1552 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1554 if (nr_pages >= present_pages)
1555 arg->status_change_nid = zone_to_nid(zone);
1558 static void node_states_clear_node(int node, struct memory_notify *arg)
1560 if (arg->status_change_nid_normal >= 0)
1561 node_clear_state(node, N_NORMAL_MEMORY);
1563 if (arg->status_change_nid_high >= 0)
1564 node_clear_state(node, N_HIGH_MEMORY);
1566 if (arg->status_change_nid >= 0)
1567 node_clear_state(node, N_MEMORY);
1570 static int count_system_ram_pages_cb(unsigned long start_pfn,
1571 unsigned long nr_pages, void *data)
1573 unsigned long *nr_system_ram_pages = data;
1575 *nr_system_ram_pages += nr_pages;
1579 int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1581 const unsigned long end_pfn = start_pfn + nr_pages;
1582 unsigned long pfn, system_ram_pages = 0;
1583 unsigned long flags;
1585 struct memory_notify arg;
1590 * {on,off}lining is constrained to full memory sections (or more
1591 * precisly to memory blocks from the user space POV).
1592 * memmap_on_memory is an exception because it reserves initial part
1593 * of the physical memory space for vmemmaps. That space is pageblock
1596 if (WARN_ON_ONCE(!nr_pages ||
1597 !IS_ALIGNED(start_pfn, pageblock_nr_pages) ||
1598 !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)))
1601 mem_hotplug_begin();
1604 * Don't allow to offline memory blocks that contain holes.
1605 * Consequently, memory blocks with holes can never get onlined
1606 * via the hotplug path - online_pages() - as hotplugged memory has
1607 * no holes. This way, we e.g., don't have to worry about marking
1608 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1609 * avoid using walk_system_ram_range() later.
1611 walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1612 count_system_ram_pages_cb);
1613 if (system_ram_pages != nr_pages) {
1615 reason = "memory holes";
1616 goto failed_removal;
1619 /* This makes hotplug much easier...and readable.
1620 we assume this for now. .*/
1621 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1624 reason = "multizone range";
1625 goto failed_removal;
1627 node = zone_to_nid(zone);
1630 * Disable pcplists so that page isolation cannot race with freeing
1631 * in a way that pages from isolated pageblock are left on pcplists.
1633 zone_pcp_disable(zone);
1634 lru_cache_disable();
1636 /* set above range as isolated */
1637 ret = start_isolate_page_range(start_pfn, end_pfn,
1639 MEMORY_OFFLINE | REPORT_FAILURE);
1641 reason = "failure to isolate range";
1642 goto failed_removal_pcplists_disabled;
1645 arg.start_pfn = start_pfn;
1646 arg.nr_pages = nr_pages;
1647 node_states_check_changes_offline(nr_pages, zone, &arg);
1649 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1650 ret = notifier_to_errno(ret);
1652 reason = "notifier failure";
1653 goto failed_removal_isolated;
1659 if (signal_pending(current)) {
1661 reason = "signal backoff";
1662 goto failed_removal_isolated;
1667 ret = scan_movable_pages(pfn, end_pfn, &pfn);
1670 * TODO: fatal migration failures should bail
1673 do_migrate_range(pfn, end_pfn);
1677 if (ret != -ENOENT) {
1678 reason = "unmovable page";
1679 goto failed_removal_isolated;
1683 * Dissolve free hugepages in the memory block before doing
1684 * offlining actually in order to make hugetlbfs's object
1685 * counting consistent.
1687 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1689 reason = "failure to dissolve huge pages";
1690 goto failed_removal_isolated;
1693 ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
1697 /* Mark all sections offline and remove free pages from the buddy. */
1698 __offline_isolated_pages(start_pfn, end_pfn);
1699 pr_debug("Offlined Pages %ld\n", nr_pages);
1702 * The memory sections are marked offline, and the pageblock flags
1703 * effectively stale; nobody should be touching them. Fixup the number
1704 * of isolated pageblocks, memory onlining will properly revert this.
1706 spin_lock_irqsave(&zone->lock, flags);
1707 zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
1708 spin_unlock_irqrestore(&zone->lock, flags);
1711 zone_pcp_enable(zone);
1713 /* removal success */
1714 adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1715 adjust_present_page_count(zone, -nr_pages);
1717 init_per_zone_wmark_min();
1719 if (!populated_zone(zone)) {
1720 zone_pcp_reset(zone);
1721 build_all_zonelists(NULL);
1723 zone_pcp_update(zone);
1725 node_states_clear_node(node, &arg);
1726 if (arg.status_change_nid >= 0) {
1728 kcompactd_stop(node);
1731 writeback_set_ratelimit();
1733 memory_notify(MEM_OFFLINE, &arg);
1734 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1738 failed_removal_isolated:
1739 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1740 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1741 failed_removal_pcplists_disabled:
1742 zone_pcp_enable(zone);
1744 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1745 (unsigned long long) start_pfn << PAGE_SHIFT,
1746 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1748 /* pushback to free area */
1753 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1755 int ret = !is_memblock_offlined(mem);
1757 if (unlikely(ret)) {
1758 phys_addr_t beginpa, endpa;
1760 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1761 endpa = beginpa + memory_block_size_bytes() - 1;
1762 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1770 static int check_cpu_on_node(pg_data_t *pgdat)
1774 for_each_present_cpu(cpu) {
1775 if (cpu_to_node(cpu) == pgdat->node_id)
1777 * the cpu on this node isn't removed, and we can't
1778 * offline this node.
1786 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1788 int nid = *(int *)arg;
1791 * If a memory block belongs to multiple nodes, the stored nid is not
1792 * reliable. However, such blocks are always online (e.g., cannot get
1793 * offlined) and, therefore, are still spanned by the node.
1795 return mem->nid == nid ? -EEXIST : 0;
1802 * Offline a node if all memory sections and cpus of the node are removed.
1804 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1805 * and online/offline operations before this call.
1807 void try_offline_node(int nid)
1809 pg_data_t *pgdat = NODE_DATA(nid);
1813 * If the node still spans pages (especially ZONE_DEVICE), don't
1814 * offline it. A node spans memory after move_pfn_range_to_zone(),
1815 * e.g., after the memory block was onlined.
1817 if (pgdat->node_spanned_pages)
1821 * Especially offline memory blocks might not be spanned by the
1822 * node. They will get spanned by the node once they get onlined.
1823 * However, they link to the node in sysfs and can get onlined later.
1825 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1829 if (check_cpu_on_node(pgdat))
1833 * all memory/cpu of this node are removed, we can offline this
1836 node_set_offline(nid);
1837 unregister_one_node(nid);
1839 EXPORT_SYMBOL(try_offline_node);
1841 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1845 BUG_ON(check_hotplug_memory_range(start, size));
1848 * All memory blocks must be offlined before removing memory. Check
1849 * whether all memory blocks in question are offline and return error
1850 * if this is not the case.
1852 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1856 /* remove memmap entry */
1857 firmware_map_remove(start, start + size, "System RAM");
1860 * Memory block device removal under the device_hotplug_lock is
1861 * a barrier against racing online attempts.
1863 remove_memory_block_devices(start, size);
1865 mem_hotplug_begin();
1867 arch_remove_memory(nid, start, size, NULL);
1869 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1870 memblock_free(start, size);
1871 memblock_remove(start, size);
1874 release_mem_region_adjustable(start, size);
1876 try_offline_node(nid);
1885 * @start: physical address of the region to remove
1886 * @size: size of the region to remove
1888 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1889 * and online/offline operations before this call, as required by
1890 * try_offline_node().
1892 void __remove_memory(int nid, u64 start, u64 size)
1896 * trigger BUG() if some memory is not offlined prior to calling this
1899 if (try_remove_memory(nid, start, size))
1904 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1905 * some memory is not offline
1907 int remove_memory(int nid, u64 start, u64 size)
1911 lock_device_hotplug();
1912 rc = try_remove_memory(nid, start, size);
1913 unlock_device_hotplug();
1917 EXPORT_SYMBOL_GPL(remove_memory);
1919 static int try_offline_memory_block(struct memory_block *mem, void *arg)
1921 uint8_t online_type = MMOP_ONLINE_KERNEL;
1922 uint8_t **online_types = arg;
1927 * Sense the online_type via the zone of the memory block. Offlining
1928 * with multiple zones within one memory block will be rejected
1929 * by offlining code ... so we don't care about that.
1931 page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr));
1932 if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE)
1933 online_type = MMOP_ONLINE_MOVABLE;
1935 rc = device_offline(&mem->dev);
1937 * Default is MMOP_OFFLINE - change it only if offlining succeeded,
1938 * so try_reonline_memory_block() can do the right thing.
1941 **online_types = online_type;
1944 /* Ignore if already offline. */
1945 return rc < 0 ? rc : 0;
1948 static int try_reonline_memory_block(struct memory_block *mem, void *arg)
1950 uint8_t **online_types = arg;
1953 if (**online_types != MMOP_OFFLINE) {
1954 mem->online_type = **online_types;
1955 rc = device_online(&mem->dev);
1957 pr_warn("%s: Failed to re-online memory: %d",
1961 /* Continue processing all remaining memory blocks. */
1967 * Try to offline and remove memory. Might take a long time to finish in case
1968 * memory is still in use. Primarily useful for memory devices that logically
1969 * unplugged all memory (so it's no longer in use) and want to offline + remove
1972 int offline_and_remove_memory(int nid, u64 start, u64 size)
1974 const unsigned long mb_count = size / memory_block_size_bytes();
1975 uint8_t *online_types, *tmp;
1978 if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1979 !IS_ALIGNED(size, memory_block_size_bytes()) || !size)
1983 * We'll remember the old online type of each memory block, so we can
1984 * try to revert whatever we did when offlining one memory block fails
1985 * after offlining some others succeeded.
1987 online_types = kmalloc_array(mb_count, sizeof(*online_types),
1992 * Initialize all states to MMOP_OFFLINE, so when we abort processing in
1993 * try_offline_memory_block(), we'll skip all unprocessed blocks in
1994 * try_reonline_memory_block().
1996 memset(online_types, MMOP_OFFLINE, mb_count);
1998 lock_device_hotplug();
2001 rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block);
2004 * In case we succeeded to offline all memory, remove it.
2005 * This cannot fail as it cannot get onlined in the meantime.
2008 rc = try_remove_memory(nid, start, size);
2010 pr_err("%s: Failed to remove memory: %d", __func__, rc);
2014 * Rollback what we did. While memory onlining might theoretically fail
2015 * (nacked by a notifier), it barely ever happens.
2019 walk_memory_blocks(start, size, &tmp,
2020 try_reonline_memory_block);
2022 unlock_device_hotplug();
2024 kfree(online_types);
2027 EXPORT_SYMBOL_GPL(offline_and_remove_memory);
2028 #endif /* CONFIG_MEMORY_HOTREMOVE */