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_SYSRAM_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();
353 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
354 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
355 unsigned long start_pfn,
356 unsigned long end_pfn)
358 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
359 if (unlikely(!pfn_to_online_page(start_pfn)))
362 if (unlikely(pfn_to_nid(start_pfn) != nid))
365 if (zone != page_zone(pfn_to_page(start_pfn)))
374 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
375 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
376 unsigned long start_pfn,
377 unsigned long end_pfn)
381 /* pfn is the end pfn of a memory section. */
383 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
384 if (unlikely(!pfn_to_online_page(pfn)))
387 if (unlikely(pfn_to_nid(pfn) != nid))
390 if (zone != page_zone(pfn_to_page(pfn)))
399 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
400 unsigned long end_pfn)
403 int nid = zone_to_nid(zone);
405 zone_span_writelock(zone);
406 if (zone->zone_start_pfn == start_pfn) {
408 * If the section is smallest section in the zone, it need
409 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
410 * In this case, we find second smallest valid mem_section
411 * for shrinking zone.
413 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
416 zone->spanned_pages = zone_end_pfn(zone) - pfn;
417 zone->zone_start_pfn = pfn;
419 zone->zone_start_pfn = 0;
420 zone->spanned_pages = 0;
422 } else if (zone_end_pfn(zone) == end_pfn) {
424 * If the section is biggest section in the zone, it need
425 * shrink zone->spanned_pages.
426 * In this case, we find second biggest valid mem_section for
429 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
432 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
434 zone->zone_start_pfn = 0;
435 zone->spanned_pages = 0;
438 zone_span_writeunlock(zone);
441 static void update_pgdat_span(struct pglist_data *pgdat)
443 unsigned long node_start_pfn = 0, node_end_pfn = 0;
446 for (zone = pgdat->node_zones;
447 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
448 unsigned long zone_end_pfn = zone->zone_start_pfn +
451 /* No need to lock the zones, they can't change. */
452 if (!zone->spanned_pages)
455 node_start_pfn = zone->zone_start_pfn;
456 node_end_pfn = zone_end_pfn;
460 if (zone_end_pfn > node_end_pfn)
461 node_end_pfn = zone_end_pfn;
462 if (zone->zone_start_pfn < node_start_pfn)
463 node_start_pfn = zone->zone_start_pfn;
466 pgdat->node_start_pfn = node_start_pfn;
467 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
470 void __ref remove_pfn_range_from_zone(struct zone *zone,
471 unsigned long start_pfn,
472 unsigned long nr_pages)
474 const unsigned long end_pfn = start_pfn + nr_pages;
475 struct pglist_data *pgdat = zone->zone_pgdat;
476 unsigned long pfn, cur_nr_pages, flags;
478 /* Poison struct pages because they are now uninitialized again. */
479 for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
482 /* Select all remaining pages up to the next section boundary */
484 min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
485 page_init_poison(pfn_to_page(pfn),
486 sizeof(struct page) * cur_nr_pages);
489 #ifdef CONFIG_ZONE_DEVICE
491 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
492 * we will not try to shrink the zones - which is okay as
493 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
495 if (zone_idx(zone) == ZONE_DEVICE)
499 clear_zone_contiguous(zone);
501 pgdat_resize_lock(zone->zone_pgdat, &flags);
502 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
503 update_pgdat_span(pgdat);
504 pgdat_resize_unlock(zone->zone_pgdat, &flags);
506 set_zone_contiguous(zone);
509 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
510 unsigned long map_offset,
511 struct vmem_altmap *altmap)
513 struct mem_section *ms = __pfn_to_section(pfn);
515 if (WARN_ON_ONCE(!valid_section(ms)))
518 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
522 * __remove_pages() - remove sections of pages
523 * @pfn: starting pageframe (must be aligned to start of a section)
524 * @nr_pages: number of pages to remove (must be multiple of section size)
525 * @altmap: alternative device page map or %NULL if default memmap is used
527 * Generic helper function to remove section mappings and sysfs entries
528 * for the section of the memory we are removing. Caller needs to make
529 * sure that pages are marked reserved and zones are adjust properly by
530 * calling offline_pages().
532 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
533 struct vmem_altmap *altmap)
535 const unsigned long end_pfn = pfn + nr_pages;
536 unsigned long cur_nr_pages;
537 unsigned long map_offset = 0;
539 map_offset = vmem_altmap_offset(altmap);
541 if (check_pfn_span(pfn, nr_pages, "remove"))
544 for (; pfn < end_pfn; pfn += cur_nr_pages) {
546 /* Select all remaining pages up to the next section boundary */
547 cur_nr_pages = min(end_pfn - pfn,
548 SECTION_ALIGN_UP(pfn + 1) - pfn);
549 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
554 int set_online_page_callback(online_page_callback_t callback)
559 mutex_lock(&online_page_callback_lock);
561 if (online_page_callback == generic_online_page) {
562 online_page_callback = callback;
566 mutex_unlock(&online_page_callback_lock);
571 EXPORT_SYMBOL_GPL(set_online_page_callback);
573 int restore_online_page_callback(online_page_callback_t callback)
578 mutex_lock(&online_page_callback_lock);
580 if (online_page_callback == callback) {
581 online_page_callback = generic_online_page;
585 mutex_unlock(&online_page_callback_lock);
590 EXPORT_SYMBOL_GPL(restore_online_page_callback);
592 void generic_online_page(struct page *page, unsigned int order)
595 * Freeing the page with debug_pagealloc enabled will try to unmap it,
596 * so we should map it first. This is better than introducing a special
597 * case in page freeing fast path.
599 if (debug_pagealloc_enabled_static())
600 kernel_map_pages(page, 1 << order, 1);
601 __free_pages_core(page, order);
602 totalram_pages_add(1UL << order);
603 #ifdef CONFIG_HIGHMEM
604 if (PageHighMem(page))
605 totalhigh_pages_add(1UL << order);
608 EXPORT_SYMBOL_GPL(generic_online_page);
610 static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
612 const unsigned long end_pfn = start_pfn + nr_pages;
616 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
617 * decide to not expose all pages to the buddy (e.g., expose them
618 * later). We account all pages as being online and belonging to this
621 for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES)
622 (*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1);
624 /* mark all involved sections as online */
625 online_mem_sections(start_pfn, end_pfn);
628 /* check which state of node_states will be changed when online memory */
629 static void node_states_check_changes_online(unsigned long nr_pages,
630 struct zone *zone, struct memory_notify *arg)
632 int nid = zone_to_nid(zone);
634 arg->status_change_nid = NUMA_NO_NODE;
635 arg->status_change_nid_normal = NUMA_NO_NODE;
636 arg->status_change_nid_high = NUMA_NO_NODE;
638 if (!node_state(nid, N_MEMORY))
639 arg->status_change_nid = nid;
640 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
641 arg->status_change_nid_normal = nid;
642 #ifdef CONFIG_HIGHMEM
643 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
644 arg->status_change_nid_high = nid;
648 static void node_states_set_node(int node, struct memory_notify *arg)
650 if (arg->status_change_nid_normal >= 0)
651 node_set_state(node, N_NORMAL_MEMORY);
653 if (arg->status_change_nid_high >= 0)
654 node_set_state(node, N_HIGH_MEMORY);
656 if (arg->status_change_nid >= 0)
657 node_set_state(node, N_MEMORY);
660 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
661 unsigned long nr_pages)
663 unsigned long old_end_pfn = zone_end_pfn(zone);
665 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
666 zone->zone_start_pfn = start_pfn;
668 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
671 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
672 unsigned long nr_pages)
674 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
676 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
677 pgdat->node_start_pfn = start_pfn;
679 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
683 * Associate the pfn range with the given zone, initializing the memmaps
684 * and resizing the pgdat/zone data to span the added pages. After this
685 * call, all affected pages are PG_reserved.
687 * All aligned pageblocks are initialized to the specified migratetype
688 * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
689 * zone stats (e.g., nr_isolate_pageblock) are touched.
691 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
692 unsigned long nr_pages,
693 struct vmem_altmap *altmap, int migratetype)
695 struct pglist_data *pgdat = zone->zone_pgdat;
696 int nid = pgdat->node_id;
699 clear_zone_contiguous(zone);
701 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
702 pgdat_resize_lock(pgdat, &flags);
703 zone_span_writelock(zone);
704 if (zone_is_empty(zone))
705 init_currently_empty_zone(zone, start_pfn, nr_pages);
706 resize_zone_range(zone, start_pfn, nr_pages);
707 zone_span_writeunlock(zone);
708 resize_pgdat_range(pgdat, start_pfn, nr_pages);
709 pgdat_resize_unlock(pgdat, &flags);
712 * TODO now we have a visible range of pages which are not associated
713 * with their zone properly. Not nice but set_pfnblock_flags_mask
714 * expects the zone spans the pfn range. All the pages in the range
715 * are reserved so nobody should be touching them so we should be safe
717 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, 0,
718 MEMINIT_HOTPLUG, altmap, migratetype);
720 set_zone_contiguous(zone);
724 * Returns a default kernel memory zone for the given pfn range.
725 * If no kernel zone covers this pfn range it will automatically go
726 * to the ZONE_NORMAL.
728 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
729 unsigned long nr_pages)
731 struct pglist_data *pgdat = NODE_DATA(nid);
734 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
735 struct zone *zone = &pgdat->node_zones[zid];
737 if (zone_intersects(zone, start_pfn, nr_pages))
741 return &pgdat->node_zones[ZONE_NORMAL];
744 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
745 unsigned long nr_pages)
747 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
749 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
750 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
751 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
754 * We inherit the existing zone in a simple case where zones do not
755 * overlap in the given range
757 if (in_kernel ^ in_movable)
758 return (in_kernel) ? kernel_zone : movable_zone;
761 * If the range doesn't belong to any zone or two zones overlap in the
762 * given range then we use movable zone only if movable_node is
763 * enabled because we always online to a kernel zone by default.
765 return movable_node_enabled ? movable_zone : kernel_zone;
768 struct zone *zone_for_pfn_range(int online_type, int nid,
769 unsigned long start_pfn, unsigned long nr_pages)
771 if (online_type == MMOP_ONLINE_KERNEL)
772 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
774 if (online_type == MMOP_ONLINE_MOVABLE)
775 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
777 return default_zone_for_pfn(nid, start_pfn, nr_pages);
780 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
781 int online_type, int nid)
785 int need_zonelists_rebuild = 0;
787 struct memory_notify arg;
789 /* We can only online full sections (e.g., SECTION_IS_ONLINE) */
790 if (WARN_ON_ONCE(!nr_pages ||
791 !IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION)))
796 /* associate pfn range with the zone */
797 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
798 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
801 arg.nr_pages = nr_pages;
802 node_states_check_changes_online(nr_pages, zone, &arg);
804 ret = memory_notify(MEM_GOING_ONLINE, &arg);
805 ret = notifier_to_errno(ret);
807 goto failed_addition;
810 * Fixup the number of isolated pageblocks before marking the sections
811 * onlining, such that undo_isolate_page_range() works correctly.
813 spin_lock_irqsave(&zone->lock, flags);
814 zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
815 spin_unlock_irqrestore(&zone->lock, flags);
818 * If this zone is not populated, then it is not in zonelist.
819 * This means the page allocator ignores this zone.
820 * So, zonelist must be updated after online.
822 if (!populated_zone(zone)) {
823 need_zonelists_rebuild = 1;
824 setup_zone_pageset(zone);
827 online_pages_range(pfn, nr_pages);
828 zone->present_pages += nr_pages;
830 pgdat_resize_lock(zone->zone_pgdat, &flags);
831 zone->zone_pgdat->node_present_pages += nr_pages;
832 pgdat_resize_unlock(zone->zone_pgdat, &flags);
834 node_states_set_node(nid, &arg);
835 if (need_zonelists_rebuild)
836 build_all_zonelists(NULL);
837 zone_pcp_update(zone);
839 /* Basic onlining is complete, allow allocation of onlined pages. */
840 undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
843 * Freshly onlined pages aren't shuffled (e.g., all pages are placed to
844 * the tail of the freelist when undoing isolation). Shuffle the whole
845 * zone to make sure the just onlined pages are properly distributed
846 * across the whole freelist - to create an initial shuffle.
850 init_per_zone_wmark_min();
855 writeback_set_ratelimit();
857 memory_notify(MEM_ONLINE, &arg);
862 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
863 (unsigned long long) pfn << PAGE_SHIFT,
864 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
865 memory_notify(MEM_CANCEL_ONLINE, &arg);
866 remove_pfn_range_from_zone(zone, pfn, nr_pages);
870 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
872 static void reset_node_present_pages(pg_data_t *pgdat)
876 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
877 z->present_pages = 0;
879 pgdat->node_present_pages = 0;
882 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
883 static pg_data_t __ref *hotadd_new_pgdat(int nid)
885 struct pglist_data *pgdat;
887 pgdat = NODE_DATA(nid);
889 pgdat = arch_alloc_nodedata(nid);
893 pgdat->per_cpu_nodestats =
894 alloc_percpu(struct per_cpu_nodestat);
895 arch_refresh_nodedata(nid, pgdat);
899 * Reset the nr_zones, order and highest_zoneidx before reuse.
900 * Note that kswapd will init kswapd_highest_zoneidx properly
901 * when it starts in the near future.
904 pgdat->kswapd_order = 0;
905 pgdat->kswapd_highest_zoneidx = 0;
906 for_each_online_cpu(cpu) {
907 struct per_cpu_nodestat *p;
909 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
910 memset(p, 0, sizeof(*p));
914 /* we can use NODE_DATA(nid) from here */
915 pgdat->node_id = nid;
916 pgdat->node_start_pfn = 0;
918 /* init node's zones as empty zones, we don't have any present pages.*/
919 free_area_init_core_hotplug(nid);
922 * The node we allocated has no zone fallback lists. For avoiding
923 * to access not-initialized zonelist, build here.
925 build_all_zonelists(pgdat);
928 * When memory is hot-added, all the memory is in offline state. So
929 * clear all zones' present_pages because they will be updated in
930 * online_pages() and offline_pages().
932 reset_node_managed_pages(pgdat);
933 reset_node_present_pages(pgdat);
938 static void rollback_node_hotadd(int nid)
940 pg_data_t *pgdat = NODE_DATA(nid);
942 arch_refresh_nodedata(nid, NULL);
943 free_percpu(pgdat->per_cpu_nodestats);
944 arch_free_nodedata(pgdat);
949 * try_online_node - online a node if offlined
951 * @set_node_online: Whether we want to online the node
952 * called by cpu_up() to online a node without onlined memory.
955 * 1 -> a new node has been allocated
956 * 0 -> the node is already online
957 * -ENOMEM -> the node could not be allocated
959 static int __try_online_node(int nid, bool set_node_online)
964 if (node_online(nid))
967 pgdat = hotadd_new_pgdat(nid);
969 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
974 if (set_node_online) {
975 node_set_online(nid);
976 ret = register_one_node(nid);
984 * Users of this function always want to online/register the node
986 int try_online_node(int nid)
991 ret = __try_online_node(nid, true);
996 static int check_hotplug_memory_range(u64 start, u64 size)
998 /* memory range must be block size aligned */
999 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1000 !IS_ALIGNED(size, memory_block_size_bytes())) {
1001 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1002 memory_block_size_bytes(), start, size);
1009 static int online_memory_block(struct memory_block *mem, void *arg)
1011 mem->online_type = memhp_default_online_type;
1012 return device_online(&mem->dev);
1016 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1017 * and online/offline operations (triggered e.g. by sysfs).
1019 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1021 int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1023 struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
1025 bool new_node = false;
1029 size = resource_size(res);
1031 ret = check_hotplug_memory_range(start, size);
1035 if (!node_possible(nid)) {
1036 WARN(1, "node %d was absent from the node_possible_map\n", nid);
1040 mem_hotplug_begin();
1042 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1043 memblock_add_node(start, size, nid);
1045 ret = __try_online_node(nid, false);
1050 /* call arch's memory hotadd */
1051 ret = arch_add_memory(nid, start, size, ¶ms);
1055 /* create memory block devices after memory was added */
1056 ret = create_memory_block_devices(start, size);
1058 arch_remove_memory(nid, start, size, NULL);
1063 /* If sysfs file of new node can't be created, cpu on the node
1064 * can't be hot-added. There is no rollback way now.
1065 * So, check by BUG_ON() to catch it reluctantly..
1066 * We online node here. We can't roll back from here.
1068 node_set_online(nid);
1069 ret = __register_one_node(nid);
1073 /* link memory sections under this node.*/
1074 link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1077 /* create new memmap entry */
1078 if (!strcmp(res->name, "System RAM"))
1079 firmware_map_add_hotplug(start, start + size, "System RAM");
1081 /* device_online() will take the lock when calling online_pages() */
1085 * In case we're allowed to merge the resource, flag it and trigger
1086 * merging now that adding succeeded.
1088 if (mhp_flags & MEMHP_MERGE_RESOURCE)
1089 merge_system_ram_resource(res);
1091 /* online pages if requested */
1092 if (memhp_default_online_type != MMOP_OFFLINE)
1093 walk_memory_blocks(start, size, NULL, online_memory_block);
1097 /* rollback pgdat allocation and others */
1099 rollback_node_hotadd(nid);
1100 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1101 memblock_remove(start, size);
1106 /* requires device_hotplug_lock, see add_memory_resource() */
1107 int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1109 struct resource *res;
1112 res = register_memory_resource(start, size, "System RAM");
1114 return PTR_ERR(res);
1116 ret = add_memory_resource(nid, res, mhp_flags);
1118 release_memory_resource(res);
1122 int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1126 lock_device_hotplug();
1127 rc = __add_memory(nid, start, size, mhp_flags);
1128 unlock_device_hotplug();
1132 EXPORT_SYMBOL_GPL(add_memory);
1135 * Add special, driver-managed memory to the system as system RAM. Such
1136 * memory is not exposed via the raw firmware-provided memmap as system
1137 * RAM, instead, it is detected and added by a driver - during cold boot,
1138 * after a reboot, and after kexec.
1140 * Reasons why this memory should not be used for the initial memmap of a
1141 * kexec kernel or for placing kexec images:
1142 * - The booting kernel is in charge of determining how this memory will be
1143 * used (e.g., use persistent memory as system RAM)
1144 * - Coordination with a hypervisor is required before this memory
1145 * can be used (e.g., inaccessible parts).
1147 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1148 * memory map") are created. Also, the created memory resource is flagged
1149 * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
1150 * this memory as well (esp., not place kexec images onto it).
1152 * The resource_name (visible via /proc/iomem) has to have the format
1153 * "System RAM ($DRIVER)".
1155 int add_memory_driver_managed(int nid, u64 start, u64 size,
1156 const char *resource_name, mhp_t mhp_flags)
1158 struct resource *res;
1161 if (!resource_name ||
1162 strstr(resource_name, "System RAM (") != resource_name ||
1163 resource_name[strlen(resource_name) - 1] != ')')
1166 lock_device_hotplug();
1168 res = register_memory_resource(start, size, resource_name);
1174 rc = add_memory_resource(nid, res, mhp_flags);
1176 release_memory_resource(res);
1179 unlock_device_hotplug();
1182 EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1184 #ifdef CONFIG_MEMORY_HOTREMOVE
1186 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1187 * memory holes). When true, return the zone.
1189 struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1190 unsigned long end_pfn)
1192 unsigned long pfn, sec_end_pfn;
1193 struct zone *zone = NULL;
1196 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1198 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1199 /* Make sure the memory section is present first */
1200 if (!present_section_nr(pfn_to_section_nr(pfn)))
1202 for (; pfn < sec_end_pfn && pfn < end_pfn;
1203 pfn += MAX_ORDER_NR_PAGES) {
1205 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1206 while ((i < MAX_ORDER_NR_PAGES) &&
1207 !pfn_valid_within(pfn + i))
1209 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1211 /* Check if we got outside of the zone */
1212 if (zone && !zone_spans_pfn(zone, pfn + i))
1214 page = pfn_to_page(pfn + i);
1215 if (zone && page_zone(page) != zone)
1217 zone = page_zone(page);
1225 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1226 * non-lru movable pages and hugepages). Will skip over most unmovable
1227 * pages (esp., pages that can be skipped when offlining), but bail out on
1228 * definitely unmovable pages.
1231 * 0 in case a movable page is found and movable_pfn was updated.
1232 * -ENOENT in case no movable page was found.
1233 * -EBUSY in case a definitely unmovable page was found.
1235 static int scan_movable_pages(unsigned long start, unsigned long end,
1236 unsigned long *movable_pfn)
1240 for (pfn = start; pfn < end; pfn++) {
1241 struct page *page, *head;
1244 if (!pfn_valid(pfn))
1246 page = pfn_to_page(pfn);
1249 if (__PageMovable(page))
1253 * PageOffline() pages that are not marked __PageMovable() and
1254 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1255 * definitely unmovable. If their reference count would be 0,
1256 * they could at least be skipped when offlining memory.
1258 if (PageOffline(page) && page_count(page))
1261 if (!PageHuge(page))
1263 head = compound_head(page);
1264 if (page_huge_active(head))
1266 skip = compound_nr(head) - (pfn - page_to_pfn(head));
1276 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1279 struct page *page, *head;
1282 static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL,
1283 DEFAULT_RATELIMIT_BURST);
1285 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1286 if (!pfn_valid(pfn))
1288 page = pfn_to_page(pfn);
1289 head = compound_head(page);
1291 if (PageHuge(page)) {
1292 pfn = page_to_pfn(head) + compound_nr(head) - 1;
1293 isolate_hugetlb(head, &source);
1295 } else if (PageTransHuge(page))
1296 pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1299 * HWPoison pages have elevated reference counts so the migration would
1300 * fail on them. It also doesn't make any sense to migrate them in the
1301 * first place. Still try to unmap such a page in case it is still mapped
1302 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1303 * the unmap as the catch all safety net).
1305 if (PageHWPoison(page)) {
1306 if (WARN_ON(PageLRU(page)))
1307 isolate_lru_page(page);
1308 if (page_mapped(page))
1309 try_to_unmap(page, TTU_IGNORE_MLOCK);
1313 if (!get_page_unless_zero(page))
1316 * We can skip free pages. And we can deal with pages on
1317 * LRU and non-lru movable pages.
1320 ret = isolate_lru_page(page);
1322 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1323 if (!ret) { /* Success */
1324 list_add_tail(&page->lru, &source);
1325 if (!__PageMovable(page))
1326 inc_node_page_state(page, NR_ISOLATED_ANON +
1327 page_is_file_lru(page));
1330 if (__ratelimit(&migrate_rs)) {
1331 pr_warn("failed to isolate pfn %lx\n", pfn);
1332 dump_page(page, "isolation failed");
1337 if (!list_empty(&source)) {
1338 nodemask_t nmask = node_states[N_MEMORY];
1339 struct migration_target_control mtc = {
1341 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1345 * We have checked that migration range is on a single zone so
1346 * we can use the nid of the first page to all the others.
1348 mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1351 * try to allocate from a different node but reuse this node
1352 * if there are no other online nodes to be used (e.g. we are
1353 * offlining a part of the only existing node)
1355 node_clear(mtc.nid, nmask);
1356 if (nodes_empty(nmask))
1357 node_set(mtc.nid, nmask);
1358 ret = migrate_pages(&source, alloc_migration_target, NULL,
1359 (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1361 list_for_each_entry(page, &source, lru) {
1362 if (__ratelimit(&migrate_rs)) {
1363 pr_warn("migrating pfn %lx failed ret:%d\n",
1364 page_to_pfn(page), ret);
1365 dump_page(page, "migration failure");
1368 putback_movable_pages(&source);
1375 static int __init cmdline_parse_movable_node(char *p)
1377 movable_node_enabled = true;
1380 early_param("movable_node", cmdline_parse_movable_node);
1382 /* check which state of node_states will be changed when offline memory */
1383 static void node_states_check_changes_offline(unsigned long nr_pages,
1384 struct zone *zone, struct memory_notify *arg)
1386 struct pglist_data *pgdat = zone->zone_pgdat;
1387 unsigned long present_pages = 0;
1390 arg->status_change_nid = NUMA_NO_NODE;
1391 arg->status_change_nid_normal = NUMA_NO_NODE;
1392 arg->status_change_nid_high = NUMA_NO_NODE;
1395 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1396 * If the memory to be offline is within the range
1397 * [0..ZONE_NORMAL], and it is the last present memory there,
1398 * the zones in that range will become empty after the offlining,
1399 * thus we can determine that we need to clear the node from
1400 * node_states[N_NORMAL_MEMORY].
1402 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1403 present_pages += pgdat->node_zones[zt].present_pages;
1404 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1405 arg->status_change_nid_normal = zone_to_nid(zone);
1407 #ifdef CONFIG_HIGHMEM
1409 * node_states[N_HIGH_MEMORY] contains nodes which
1410 * have normal memory or high memory.
1411 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1412 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1413 * we determine that the zones in that range become empty,
1414 * we need to clear the node for N_HIGH_MEMORY.
1416 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1417 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1418 arg->status_change_nid_high = zone_to_nid(zone);
1422 * We have accounted the pages from [0..ZONE_NORMAL), and
1423 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1425 * Here we count the possible pages from ZONE_MOVABLE.
1426 * If after having accounted all the pages, we see that the nr_pages
1427 * to be offlined is over or equal to the accounted pages,
1428 * we know that the node will become empty, and so, we can clear
1429 * it for N_MEMORY as well.
1431 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1433 if (nr_pages >= present_pages)
1434 arg->status_change_nid = zone_to_nid(zone);
1437 static void node_states_clear_node(int node, struct memory_notify *arg)
1439 if (arg->status_change_nid_normal >= 0)
1440 node_clear_state(node, N_NORMAL_MEMORY);
1442 if (arg->status_change_nid_high >= 0)
1443 node_clear_state(node, N_HIGH_MEMORY);
1445 if (arg->status_change_nid >= 0)
1446 node_clear_state(node, N_MEMORY);
1449 static int count_system_ram_pages_cb(unsigned long start_pfn,
1450 unsigned long nr_pages, void *data)
1452 unsigned long *nr_system_ram_pages = data;
1454 *nr_system_ram_pages += nr_pages;
1458 int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1460 const unsigned long end_pfn = start_pfn + nr_pages;
1461 unsigned long pfn, system_ram_pages = 0;
1462 unsigned long flags;
1464 struct memory_notify arg;
1468 /* We can only offline full sections (e.g., SECTION_IS_ONLINE) */
1469 if (WARN_ON_ONCE(!nr_pages ||
1470 !IS_ALIGNED(start_pfn | nr_pages, PAGES_PER_SECTION)))
1473 mem_hotplug_begin();
1476 * Don't allow to offline memory blocks that contain holes.
1477 * Consequently, memory blocks with holes can never get onlined
1478 * via the hotplug path - online_pages() - as hotplugged memory has
1479 * no holes. This way, we e.g., don't have to worry about marking
1480 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1481 * avoid using walk_system_ram_range() later.
1483 walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1484 count_system_ram_pages_cb);
1485 if (system_ram_pages != nr_pages) {
1487 reason = "memory holes";
1488 goto failed_removal;
1491 /* This makes hotplug much easier...and readable.
1492 we assume this for now. .*/
1493 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1496 reason = "multizone range";
1497 goto failed_removal;
1499 node = zone_to_nid(zone);
1501 /* set above range as isolated */
1502 ret = start_isolate_page_range(start_pfn, end_pfn,
1504 MEMORY_OFFLINE | REPORT_FAILURE);
1506 reason = "failure to isolate range";
1507 goto failed_removal;
1510 arg.start_pfn = start_pfn;
1511 arg.nr_pages = nr_pages;
1512 node_states_check_changes_offline(nr_pages, zone, &arg);
1514 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1515 ret = notifier_to_errno(ret);
1517 reason = "notifier failure";
1518 goto failed_removal_isolated;
1524 if (signal_pending(current)) {
1526 reason = "signal backoff";
1527 goto failed_removal_isolated;
1531 lru_add_drain_all();
1533 ret = scan_movable_pages(pfn, end_pfn, &pfn);
1536 * TODO: fatal migration failures should bail
1539 do_migrate_range(pfn, end_pfn);
1543 if (ret != -ENOENT) {
1544 reason = "unmovable page";
1545 goto failed_removal_isolated;
1549 * Dissolve free hugepages in the memory block before doing
1550 * offlining actually in order to make hugetlbfs's object
1551 * counting consistent.
1553 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1555 reason = "failure to dissolve huge pages";
1556 goto failed_removal_isolated;
1560 * per-cpu pages are drained in start_isolate_page_range, but if
1561 * there are still pages that are not free, make sure that we
1562 * drain again, because when we isolated range we might
1563 * have raced with another thread that was adding pages to pcp
1566 * Forward progress should be still guaranteed because
1567 * pages on the pcp list can only belong to MOVABLE_ZONE
1568 * because has_unmovable_pages explicitly checks for
1569 * PageBuddy on freed pages on other zones.
1571 ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
1573 drain_all_pages(zone);
1576 /* Mark all sections offline and remove free pages from the buddy. */
1577 __offline_isolated_pages(start_pfn, end_pfn);
1578 pr_info("Offlined Pages %ld\n", nr_pages);
1581 * The memory sections are marked offline, and the pageblock flags
1582 * effectively stale; nobody should be touching them. Fixup the number
1583 * of isolated pageblocks, memory onlining will properly revert this.
1585 spin_lock_irqsave(&zone->lock, flags);
1586 zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
1587 spin_unlock_irqrestore(&zone->lock, flags);
1589 /* removal success */
1590 adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1591 zone->present_pages -= nr_pages;
1593 pgdat_resize_lock(zone->zone_pgdat, &flags);
1594 zone->zone_pgdat->node_present_pages -= nr_pages;
1595 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1597 init_per_zone_wmark_min();
1599 if (!populated_zone(zone)) {
1600 zone_pcp_reset(zone);
1601 build_all_zonelists(NULL);
1603 zone_pcp_update(zone);
1605 node_states_clear_node(node, &arg);
1606 if (arg.status_change_nid >= 0) {
1608 kcompactd_stop(node);
1611 writeback_set_ratelimit();
1613 memory_notify(MEM_OFFLINE, &arg);
1614 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1618 failed_removal_isolated:
1619 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1620 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1622 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1623 (unsigned long long) start_pfn << PAGE_SHIFT,
1624 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1626 /* pushback to free area */
1631 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1633 int ret = !is_memblock_offlined(mem);
1635 if (unlikely(ret)) {
1636 phys_addr_t beginpa, endpa;
1638 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1639 endpa = beginpa + memory_block_size_bytes() - 1;
1640 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1648 static int check_cpu_on_node(pg_data_t *pgdat)
1652 for_each_present_cpu(cpu) {
1653 if (cpu_to_node(cpu) == pgdat->node_id)
1655 * the cpu on this node isn't removed, and we can't
1656 * offline this node.
1664 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1666 int nid = *(int *)arg;
1669 * If a memory block belongs to multiple nodes, the stored nid is not
1670 * reliable. However, such blocks are always online (e.g., cannot get
1671 * offlined) and, therefore, are still spanned by the node.
1673 return mem->nid == nid ? -EEXIST : 0;
1680 * Offline a node if all memory sections and cpus of the node are removed.
1682 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1683 * and online/offline operations before this call.
1685 void try_offline_node(int nid)
1687 pg_data_t *pgdat = NODE_DATA(nid);
1691 * If the node still spans pages (especially ZONE_DEVICE), don't
1692 * offline it. A node spans memory after move_pfn_range_to_zone(),
1693 * e.g., after the memory block was onlined.
1695 if (pgdat->node_spanned_pages)
1699 * Especially offline memory blocks might not be spanned by the
1700 * node. They will get spanned by the node once they get onlined.
1701 * However, they link to the node in sysfs and can get onlined later.
1703 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1707 if (check_cpu_on_node(pgdat))
1711 * all memory/cpu of this node are removed, we can offline this
1714 node_set_offline(nid);
1715 unregister_one_node(nid);
1717 EXPORT_SYMBOL(try_offline_node);
1719 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1723 BUG_ON(check_hotplug_memory_range(start, size));
1726 * All memory blocks must be offlined before removing memory. Check
1727 * whether all memory blocks in question are offline and return error
1728 * if this is not the case.
1730 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1734 /* remove memmap entry */
1735 firmware_map_remove(start, start + size, "System RAM");
1738 * Memory block device removal under the device_hotplug_lock is
1739 * a barrier against racing online attempts.
1741 remove_memory_block_devices(start, size);
1743 mem_hotplug_begin();
1745 arch_remove_memory(nid, start, size, NULL);
1747 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1748 memblock_free(start, size);
1749 memblock_remove(start, size);
1752 release_mem_region_adjustable(start, size);
1754 try_offline_node(nid);
1763 * @start: physical address of the region to remove
1764 * @size: size of the region to remove
1766 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1767 * and online/offline operations before this call, as required by
1768 * try_offline_node().
1770 void __remove_memory(int nid, u64 start, u64 size)
1774 * trigger BUG() if some memory is not offlined prior to calling this
1777 if (try_remove_memory(nid, start, size))
1782 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1783 * some memory is not offline
1785 int remove_memory(int nid, u64 start, u64 size)
1789 lock_device_hotplug();
1790 rc = try_remove_memory(nid, start, size);
1791 unlock_device_hotplug();
1795 EXPORT_SYMBOL_GPL(remove_memory);
1797 static int try_offline_memory_block(struct memory_block *mem, void *arg)
1799 uint8_t online_type = MMOP_ONLINE_KERNEL;
1800 uint8_t **online_types = arg;
1805 * Sense the online_type via the zone of the memory block. Offlining
1806 * with multiple zones within one memory block will be rejected
1807 * by offlining code ... so we don't care about that.
1809 page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr));
1810 if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE)
1811 online_type = MMOP_ONLINE_MOVABLE;
1813 rc = device_offline(&mem->dev);
1815 * Default is MMOP_OFFLINE - change it only if offlining succeeded,
1816 * so try_reonline_memory_block() can do the right thing.
1819 **online_types = online_type;
1822 /* Ignore if already offline. */
1823 return rc < 0 ? rc : 0;
1826 static int try_reonline_memory_block(struct memory_block *mem, void *arg)
1828 uint8_t **online_types = arg;
1831 if (**online_types != MMOP_OFFLINE) {
1832 mem->online_type = **online_types;
1833 rc = device_online(&mem->dev);
1835 pr_warn("%s: Failed to re-online memory: %d",
1839 /* Continue processing all remaining memory blocks. */
1845 * Try to offline and remove memory. Might take a long time to finish in case
1846 * memory is still in use. Primarily useful for memory devices that logically
1847 * unplugged all memory (so it's no longer in use) and want to offline + remove
1850 int offline_and_remove_memory(int nid, u64 start, u64 size)
1852 const unsigned long mb_count = size / memory_block_size_bytes();
1853 uint8_t *online_types, *tmp;
1856 if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1857 !IS_ALIGNED(size, memory_block_size_bytes()) || !size)
1861 * We'll remember the old online type of each memory block, so we can
1862 * try to revert whatever we did when offlining one memory block fails
1863 * after offlining some others succeeded.
1865 online_types = kmalloc_array(mb_count, sizeof(*online_types),
1870 * Initialize all states to MMOP_OFFLINE, so when we abort processing in
1871 * try_offline_memory_block(), we'll skip all unprocessed blocks in
1872 * try_reonline_memory_block().
1874 memset(online_types, MMOP_OFFLINE, mb_count);
1876 lock_device_hotplug();
1879 rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block);
1882 * In case we succeeded to offline all memory, remove it.
1883 * This cannot fail as it cannot get onlined in the meantime.
1886 rc = try_remove_memory(nid, start, size);
1888 pr_err("%s: Failed to remove memory: %d", __func__, rc);
1892 * Rollback what we did. While memory onlining might theoretically fail
1893 * (nacked by a notifier), it barely ever happens.
1897 walk_memory_blocks(start, size, &tmp,
1898 try_reonline_memory_block);
1900 unlock_device_hotplug();
1902 kfree(online_types);
1905 EXPORT_SYMBOL_GPL(offline_and_remove_memory);
1906 #endif /* CONFIG_MEMORY_HOTREMOVE */