2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/bootmem.h>
37 #include <linux/compaction.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page *page);
52 static online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
55 /* The same as the cpu_hotplug lock, but for memory hotplug. */
57 struct task_struct *active_writer;
58 struct mutex lock; /* Synchronizes accesses to refcount, */
60 * Also blocks the new readers during
61 * an ongoing mem hotplug operation.
65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
66 struct lockdep_map dep_map;
69 .active_writer = NULL,
70 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
72 #ifdef CONFIG_DEBUG_LOCK_ALLOC
73 .dep_map = {.name = "mem_hotplug.lock" },
77 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
78 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
79 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
80 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
82 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
83 bool memhp_auto_online;
85 bool memhp_auto_online = true;
87 EXPORT_SYMBOL_GPL(memhp_auto_online);
89 static int __init setup_memhp_default_state(char *str)
91 if (!strcmp(str, "online"))
92 memhp_auto_online = true;
93 else if (!strcmp(str, "offline"))
94 memhp_auto_online = false;
98 __setup("memhp_default_state=", setup_memhp_default_state);
100 void get_online_mems(void)
103 if (mem_hotplug.active_writer == current)
105 memhp_lock_acquire_read();
106 mutex_lock(&mem_hotplug.lock);
107 mem_hotplug.refcount++;
108 mutex_unlock(&mem_hotplug.lock);
112 void put_online_mems(void)
114 if (mem_hotplug.active_writer == current)
116 mutex_lock(&mem_hotplug.lock);
118 if (WARN_ON(!mem_hotplug.refcount))
119 mem_hotplug.refcount++; /* try to fix things up */
121 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
122 wake_up_process(mem_hotplug.active_writer);
123 mutex_unlock(&mem_hotplug.lock);
124 memhp_lock_release();
128 /* Serializes write accesses to mem_hotplug.active_writer. */
129 static DEFINE_MUTEX(memory_add_remove_lock);
131 void mem_hotplug_begin(void)
133 mutex_lock(&memory_add_remove_lock);
135 mem_hotplug.active_writer = current;
137 memhp_lock_acquire();
139 mutex_lock(&mem_hotplug.lock);
140 if (likely(!mem_hotplug.refcount))
142 __set_current_state(TASK_UNINTERRUPTIBLE);
143 mutex_unlock(&mem_hotplug.lock);
148 void mem_hotplug_done(void)
150 mem_hotplug.active_writer = NULL;
151 mutex_unlock(&mem_hotplug.lock);
152 memhp_lock_release();
153 mutex_unlock(&memory_add_remove_lock);
156 /* add this memory to iomem resource */
157 static struct resource *register_memory_resource(u64 start, u64 size)
159 struct resource *res;
160 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
162 return ERR_PTR(-ENOMEM);
164 res->name = "System RAM";
166 res->end = start + size - 1;
167 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
168 if (request_resource(&iomem_resource, res) < 0) {
169 pr_debug("System RAM resource %pR cannot be added\n", res);
171 return ERR_PTR(-EEXIST);
176 static void release_memory_resource(struct resource *res)
180 release_resource(res);
185 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
186 void get_page_bootmem(unsigned long info, struct page *page,
189 page->freelist = (void *)type;
190 SetPagePrivate(page);
191 set_page_private(page, info);
195 void put_page_bootmem(struct page *page)
199 type = (unsigned long) page->freelist;
200 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
201 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
203 if (page_ref_dec_return(page) == 1) {
204 page->freelist = NULL;
205 ClearPagePrivate(page);
206 set_page_private(page, 0);
207 INIT_LIST_HEAD(&page->lru);
208 free_reserved_page(page);
212 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
213 #ifndef CONFIG_SPARSEMEM_VMEMMAP
214 static void register_page_bootmem_info_section(unsigned long start_pfn)
216 unsigned long *usemap, mapsize, section_nr, i;
217 struct mem_section *ms;
218 struct page *page, *memmap;
220 section_nr = pfn_to_section_nr(start_pfn);
221 ms = __nr_to_section(section_nr);
223 /* Get section's memmap address */
224 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
227 * Get page for the memmap's phys address
228 * XXX: need more consideration for sparse_vmemmap...
230 page = virt_to_page(memmap);
231 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
232 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
234 /* remember memmap's page */
235 for (i = 0; i < mapsize; i++, page++)
236 get_page_bootmem(section_nr, page, SECTION_INFO);
238 usemap = __nr_to_section(section_nr)->pageblock_flags;
239 page = virt_to_page(usemap);
241 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
243 for (i = 0; i < mapsize; i++, page++)
244 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
247 #else /* CONFIG_SPARSEMEM_VMEMMAP */
248 static void register_page_bootmem_info_section(unsigned long start_pfn)
250 unsigned long *usemap, mapsize, section_nr, i;
251 struct mem_section *ms;
252 struct page *page, *memmap;
254 if (!pfn_valid(start_pfn))
257 section_nr = pfn_to_section_nr(start_pfn);
258 ms = __nr_to_section(section_nr);
260 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
262 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
264 usemap = __nr_to_section(section_nr)->pageblock_flags;
265 page = virt_to_page(usemap);
267 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
269 for (i = 0; i < mapsize; i++, page++)
270 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
272 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
274 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
276 unsigned long i, pfn, end_pfn, nr_pages;
277 int node = pgdat->node_id;
280 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
281 page = virt_to_page(pgdat);
283 for (i = 0; i < nr_pages; i++, page++)
284 get_page_bootmem(node, page, NODE_INFO);
286 pfn = pgdat->node_start_pfn;
287 end_pfn = pgdat_end_pfn(pgdat);
289 /* register section info */
290 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
292 * Some platforms can assign the same pfn to multiple nodes - on
293 * node0 as well as nodeN. To avoid registering a pfn against
294 * multiple nodes we check that this pfn does not already
295 * reside in some other nodes.
297 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
298 register_page_bootmem_info_section(pfn);
301 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
303 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
304 unsigned long end_pfn)
306 unsigned long old_zone_end_pfn;
308 zone_span_writelock(zone);
310 old_zone_end_pfn = zone_end_pfn(zone);
311 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
312 zone->zone_start_pfn = start_pfn;
314 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
315 zone->zone_start_pfn;
317 zone_span_writeunlock(zone);
320 static void resize_zone(struct zone *zone, unsigned long start_pfn,
321 unsigned long end_pfn)
323 zone_span_writelock(zone);
325 if (end_pfn - start_pfn) {
326 zone->zone_start_pfn = start_pfn;
327 zone->spanned_pages = end_pfn - start_pfn;
330 * make it consist as free_area_init_core(),
331 * if spanned_pages = 0, then keep start_pfn = 0
333 zone->zone_start_pfn = 0;
334 zone->spanned_pages = 0;
337 zone_span_writeunlock(zone);
340 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
341 unsigned long end_pfn)
343 enum zone_type zid = zone_idx(zone);
344 int nid = zone->zone_pgdat->node_id;
347 for (pfn = start_pfn; pfn < end_pfn; pfn++)
348 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
351 static void __ref ensure_zone_is_initialized(struct zone *zone,
352 unsigned long start_pfn, unsigned long num_pages)
354 if (!zone_is_initialized(zone))
355 init_currently_empty_zone(zone, start_pfn, num_pages);
358 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
359 unsigned long start_pfn, unsigned long end_pfn)
362 unsigned long z1_start_pfn;
364 ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
366 pgdat_resize_lock(z1->zone_pgdat, &flags);
368 /* can't move pfns which are higher than @z2 */
369 if (end_pfn > zone_end_pfn(z2))
371 /* the move out part must be at the left most of @z2 */
372 if (start_pfn > z2->zone_start_pfn)
374 /* must included/overlap */
375 if (end_pfn <= z2->zone_start_pfn)
378 /* use start_pfn for z1's start_pfn if z1 is empty */
379 if (!zone_is_empty(z1))
380 z1_start_pfn = z1->zone_start_pfn;
382 z1_start_pfn = start_pfn;
384 resize_zone(z1, z1_start_pfn, end_pfn);
385 resize_zone(z2, end_pfn, zone_end_pfn(z2));
387 pgdat_resize_unlock(z1->zone_pgdat, &flags);
389 fix_zone_id(z1, start_pfn, end_pfn);
393 pgdat_resize_unlock(z1->zone_pgdat, &flags);
397 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
398 unsigned long start_pfn, unsigned long end_pfn)
401 unsigned long z2_end_pfn;
403 ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
405 pgdat_resize_lock(z1->zone_pgdat, &flags);
407 /* can't move pfns which are lower than @z1 */
408 if (z1->zone_start_pfn > start_pfn)
410 /* the move out part mast at the right most of @z1 */
411 if (zone_end_pfn(z1) > end_pfn)
413 /* must included/overlap */
414 if (start_pfn >= zone_end_pfn(z1))
417 /* use end_pfn for z2's end_pfn if z2 is empty */
418 if (!zone_is_empty(z2))
419 z2_end_pfn = zone_end_pfn(z2);
421 z2_end_pfn = end_pfn;
423 resize_zone(z1, z1->zone_start_pfn, start_pfn);
424 resize_zone(z2, start_pfn, z2_end_pfn);
426 pgdat_resize_unlock(z1->zone_pgdat, &flags);
428 fix_zone_id(z2, start_pfn, end_pfn);
432 pgdat_resize_unlock(z1->zone_pgdat, &flags);
436 static struct zone * __meminit move_pfn_range(int zone_shift,
437 unsigned long start_pfn, unsigned long end_pfn)
439 struct zone *zone = page_zone(pfn_to_page(start_pfn));
443 ret = move_pfn_range_left(zone + zone_shift, zone,
446 ret = move_pfn_range_right(zone, zone + zone_shift,
452 return zone + zone_shift;
455 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
456 unsigned long end_pfn)
458 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
460 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
461 pgdat->node_start_pfn = start_pfn;
463 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
464 pgdat->node_start_pfn;
467 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
469 struct pglist_data *pgdat = zone->zone_pgdat;
470 int nr_pages = PAGES_PER_SECTION;
471 int nid = pgdat->node_id;
473 unsigned long flags, pfn;
475 zone_type = zone - pgdat->node_zones;
476 ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
478 pgdat_resize_lock(zone->zone_pgdat, &flags);
479 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
480 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
481 phys_start_pfn + nr_pages);
482 pgdat_resize_unlock(zone->zone_pgdat, &flags);
483 memmap_init_zone(nr_pages, nid, zone_type,
484 phys_start_pfn, MEMMAP_HOTPLUG);
486 /* online_page_range is called later and expects pages reserved */
487 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
491 SetPageReserved(pfn_to_page(pfn));
496 static int __meminit __add_section(int nid, struct zone *zone,
497 unsigned long phys_start_pfn, bool want_memblock)
501 if (pfn_valid(phys_start_pfn))
504 ret = sparse_add_one_section(zone, phys_start_pfn);
509 ret = __add_zone(zone, phys_start_pfn);
517 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
521 * Reasonably generic function for adding memory. It is
522 * expected that archs that support memory hotplug will
523 * call this function after deciding the zone to which to
526 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
527 unsigned long nr_pages, bool want_memblock)
531 int start_sec, end_sec;
532 struct vmem_altmap *altmap;
534 clear_zone_contiguous(zone);
536 /* during initialize mem_map, align hot-added range to section */
537 start_sec = pfn_to_section_nr(phys_start_pfn);
538 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
540 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
543 * Validate altmap is within bounds of the total request
545 if (altmap->base_pfn != phys_start_pfn
546 || vmem_altmap_offset(altmap) > nr_pages) {
547 pr_warn_once("memory add fail, invalid altmap\n");
554 for (i = start_sec; i <= end_sec; i++) {
555 err = __add_section(nid, zone, section_nr_to_pfn(i), want_memblock);
558 * EEXIST is finally dealt with by ioresource collision
559 * check. see add_memory() => register_memory_resource()
560 * Warning will be printed if there is collision.
562 if (err && (err != -EEXIST))
566 vmemmap_populate_print_last();
568 set_zone_contiguous(zone);
571 EXPORT_SYMBOL_GPL(__add_pages);
573 #ifdef CONFIG_MEMORY_HOTREMOVE
574 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
575 static int find_smallest_section_pfn(int nid, struct zone *zone,
576 unsigned long start_pfn,
577 unsigned long end_pfn)
579 struct mem_section *ms;
581 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
582 ms = __pfn_to_section(start_pfn);
584 if (unlikely(!valid_section(ms)))
587 if (unlikely(pfn_to_nid(start_pfn) != nid))
590 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
599 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
600 static int find_biggest_section_pfn(int nid, struct zone *zone,
601 unsigned long start_pfn,
602 unsigned long end_pfn)
604 struct mem_section *ms;
607 /* pfn is the end pfn of a memory section. */
609 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
610 ms = __pfn_to_section(pfn);
612 if (unlikely(!valid_section(ms)))
615 if (unlikely(pfn_to_nid(pfn) != nid))
618 if (zone && zone != page_zone(pfn_to_page(pfn)))
627 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
628 unsigned long end_pfn)
630 unsigned long zone_start_pfn = zone->zone_start_pfn;
631 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
632 unsigned long zone_end_pfn = z;
634 struct mem_section *ms;
635 int nid = zone_to_nid(zone);
637 zone_span_writelock(zone);
638 if (zone_start_pfn == start_pfn) {
640 * If the section is smallest section in the zone, it need
641 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
642 * In this case, we find second smallest valid mem_section
643 * for shrinking zone.
645 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
648 zone->zone_start_pfn = pfn;
649 zone->spanned_pages = zone_end_pfn - pfn;
651 } else if (zone_end_pfn == end_pfn) {
653 * If the section is biggest section in the zone, it need
654 * shrink zone->spanned_pages.
655 * In this case, we find second biggest valid mem_section for
658 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
661 zone->spanned_pages = pfn - zone_start_pfn + 1;
665 * The section is not biggest or smallest mem_section in the zone, it
666 * only creates a hole in the zone. So in this case, we need not
667 * change the zone. But perhaps, the zone has only hole data. Thus
668 * it check the zone has only hole or not.
670 pfn = zone_start_pfn;
671 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
672 ms = __pfn_to_section(pfn);
674 if (unlikely(!valid_section(ms)))
677 if (page_zone(pfn_to_page(pfn)) != zone)
680 /* If the section is current section, it continues the loop */
681 if (start_pfn == pfn)
684 /* If we find valid section, we have nothing to do */
685 zone_span_writeunlock(zone);
689 /* The zone has no valid section */
690 zone->zone_start_pfn = 0;
691 zone->spanned_pages = 0;
692 zone_span_writeunlock(zone);
695 static void shrink_pgdat_span(struct pglist_data *pgdat,
696 unsigned long start_pfn, unsigned long end_pfn)
698 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
699 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
700 unsigned long pgdat_end_pfn = p;
702 struct mem_section *ms;
703 int nid = pgdat->node_id;
705 if (pgdat_start_pfn == start_pfn) {
707 * If the section is smallest section in the pgdat, it need
708 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
709 * In this case, we find second smallest valid mem_section
710 * for shrinking zone.
712 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
715 pgdat->node_start_pfn = pfn;
716 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
718 } else if (pgdat_end_pfn == end_pfn) {
720 * If the section is biggest section in the pgdat, it need
721 * shrink pgdat->node_spanned_pages.
722 * In this case, we find second biggest valid mem_section for
725 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
728 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
732 * If the section is not biggest or smallest mem_section in the pgdat,
733 * it only creates a hole in the pgdat. So in this case, we need not
735 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
736 * has only hole or not.
738 pfn = pgdat_start_pfn;
739 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
740 ms = __pfn_to_section(pfn);
742 if (unlikely(!valid_section(ms)))
745 if (pfn_to_nid(pfn) != nid)
748 /* If the section is current section, it continues the loop */
749 if (start_pfn == pfn)
752 /* If we find valid section, we have nothing to do */
756 /* The pgdat has no valid section */
757 pgdat->node_start_pfn = 0;
758 pgdat->node_spanned_pages = 0;
761 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
763 struct pglist_data *pgdat = zone->zone_pgdat;
764 int nr_pages = PAGES_PER_SECTION;
768 zone_type = zone - pgdat->node_zones;
770 pgdat_resize_lock(zone->zone_pgdat, &flags);
771 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
772 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
773 pgdat_resize_unlock(zone->zone_pgdat, &flags);
776 static int __remove_section(struct zone *zone, struct mem_section *ms,
777 unsigned long map_offset)
779 unsigned long start_pfn;
783 if (!valid_section(ms))
786 ret = unregister_memory_section(ms);
790 scn_nr = __section_nr(ms);
791 start_pfn = section_nr_to_pfn(scn_nr);
792 __remove_zone(zone, start_pfn);
794 sparse_remove_one_section(zone, ms, map_offset);
799 * __remove_pages() - remove sections of pages from a zone
800 * @zone: zone from which pages need to be removed
801 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
802 * @nr_pages: number of pages to remove (must be multiple of section size)
804 * Generic helper function to remove section mappings and sysfs entries
805 * for the section of the memory we are removing. Caller needs to make
806 * sure that pages are marked reserved and zones are adjust properly by
807 * calling offline_pages().
809 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
810 unsigned long nr_pages)
813 unsigned long map_offset = 0;
814 int sections_to_remove, ret = 0;
816 /* In the ZONE_DEVICE case device driver owns the memory region */
817 if (is_dev_zone(zone)) {
818 struct page *page = pfn_to_page(phys_start_pfn);
819 struct vmem_altmap *altmap;
821 altmap = to_vmem_altmap((unsigned long) page);
823 map_offset = vmem_altmap_offset(altmap);
825 resource_size_t start, size;
827 start = phys_start_pfn << PAGE_SHIFT;
828 size = nr_pages * PAGE_SIZE;
830 ret = release_mem_region_adjustable(&iomem_resource, start,
833 resource_size_t endres = start + size - 1;
835 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
836 &start, &endres, ret);
840 clear_zone_contiguous(zone);
843 * We can only remove entire sections
845 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
846 BUG_ON(nr_pages % PAGES_PER_SECTION);
848 sections_to_remove = nr_pages / PAGES_PER_SECTION;
849 for (i = 0; i < sections_to_remove; i++) {
850 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
852 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
858 set_zone_contiguous(zone);
862 #endif /* CONFIG_MEMORY_HOTREMOVE */
864 int set_online_page_callback(online_page_callback_t callback)
869 mutex_lock(&online_page_callback_lock);
871 if (online_page_callback == generic_online_page) {
872 online_page_callback = callback;
876 mutex_unlock(&online_page_callback_lock);
881 EXPORT_SYMBOL_GPL(set_online_page_callback);
883 int restore_online_page_callback(online_page_callback_t callback)
888 mutex_lock(&online_page_callback_lock);
890 if (online_page_callback == callback) {
891 online_page_callback = generic_online_page;
895 mutex_unlock(&online_page_callback_lock);
900 EXPORT_SYMBOL_GPL(restore_online_page_callback);
902 void __online_page_set_limits(struct page *page)
905 EXPORT_SYMBOL_GPL(__online_page_set_limits);
907 void __online_page_increment_counters(struct page *page)
909 adjust_managed_page_count(page, 1);
911 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
913 void __online_page_free(struct page *page)
915 __free_reserved_page(page);
917 EXPORT_SYMBOL_GPL(__online_page_free);
919 static void generic_online_page(struct page *page)
921 __online_page_set_limits(page);
922 __online_page_increment_counters(page);
923 __online_page_free(page);
926 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
930 unsigned long onlined_pages = *(unsigned long *)arg;
933 if (PageReserved(pfn_to_page(start_pfn)))
934 for (i = 0; i < nr_pages; i++) {
935 page = pfn_to_page(start_pfn + i);
936 (*online_page_callback)(page);
940 online_mem_sections(start_pfn, start_pfn + nr_pages);
942 *(unsigned long *)arg = onlined_pages;
946 #ifdef CONFIG_MOVABLE_NODE
948 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
951 static bool can_online_high_movable(int nid)
955 #else /* CONFIG_MOVABLE_NODE */
956 /* ensure every online node has NORMAL memory */
957 static bool can_online_high_movable(int nid)
959 return node_state(nid, N_NORMAL_MEMORY);
961 #endif /* CONFIG_MOVABLE_NODE */
963 /* check which state of node_states will be changed when online memory */
964 static void node_states_check_changes_online(unsigned long nr_pages,
965 struct zone *zone, struct memory_notify *arg)
967 int nid = zone_to_nid(zone);
968 enum zone_type zone_last = ZONE_NORMAL;
971 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
972 * contains nodes which have zones of 0...ZONE_NORMAL,
973 * set zone_last to ZONE_NORMAL.
975 * If we don't have HIGHMEM nor movable node,
976 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
977 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
979 if (N_MEMORY == N_NORMAL_MEMORY)
980 zone_last = ZONE_MOVABLE;
983 * if the memory to be online is in a zone of 0...zone_last, and
984 * the zones of 0...zone_last don't have memory before online, we will
985 * need to set the node to node_states[N_NORMAL_MEMORY] after
986 * the memory is online.
988 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
989 arg->status_change_nid_normal = nid;
991 arg->status_change_nid_normal = -1;
993 #ifdef CONFIG_HIGHMEM
995 * If we have movable node, node_states[N_HIGH_MEMORY]
996 * contains nodes which have zones of 0...ZONE_HIGHMEM,
997 * set zone_last to ZONE_HIGHMEM.
999 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1000 * contains nodes which have zones of 0...ZONE_MOVABLE,
1001 * set zone_last to ZONE_MOVABLE.
1003 zone_last = ZONE_HIGHMEM;
1004 if (N_MEMORY == N_HIGH_MEMORY)
1005 zone_last = ZONE_MOVABLE;
1007 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1008 arg->status_change_nid_high = nid;
1010 arg->status_change_nid_high = -1;
1012 arg->status_change_nid_high = arg->status_change_nid_normal;
1016 * if the node don't have memory befor online, we will need to
1017 * set the node to node_states[N_MEMORY] after the memory
1020 if (!node_state(nid, N_MEMORY))
1021 arg->status_change_nid = nid;
1023 arg->status_change_nid = -1;
1026 static void node_states_set_node(int node, struct memory_notify *arg)
1028 if (arg->status_change_nid_normal >= 0)
1029 node_set_state(node, N_NORMAL_MEMORY);
1031 if (arg->status_change_nid_high >= 0)
1032 node_set_state(node, N_HIGH_MEMORY);
1034 node_set_state(node, N_MEMORY);
1037 bool zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1038 enum zone_type target, int *zone_shift)
1040 struct zone *zone = page_zone(pfn_to_page(pfn));
1041 enum zone_type idx = zone_idx(zone);
1047 /* pages must be at end of current zone */
1048 if (pfn + nr_pages != zone_end_pfn(zone))
1051 /* no zones in use between current zone and target */
1052 for (i = idx + 1; i < target; i++)
1053 if (zone_is_initialized(zone - idx + i))
1058 /* pages must be at beginning of current zone */
1059 if (pfn != zone->zone_start_pfn)
1062 /* no zones in use between current zone and target */
1063 for (i = target + 1; i < idx; i++)
1064 if (zone_is_initialized(zone - idx + i))
1068 *zone_shift = target - idx;
1072 /* Must be protected by mem_hotplug_begin() */
1073 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1075 unsigned long flags;
1076 unsigned long onlined_pages = 0;
1078 int need_zonelists_rebuild = 0;
1081 struct memory_notify arg;
1085 * This doesn't need a lock to do pfn_to_page().
1086 * The section can't be removed here because of the
1087 * memory_block->state_mutex.
1089 zone = page_zone(pfn_to_page(pfn));
1091 if ((zone_idx(zone) > ZONE_NORMAL ||
1092 online_type == MMOP_ONLINE_MOVABLE) &&
1093 !can_online_high_movable(pfn_to_nid(pfn)))
1096 if (online_type == MMOP_ONLINE_KERNEL) {
1097 if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift))
1099 } else if (online_type == MMOP_ONLINE_MOVABLE) {
1100 if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift))
1104 zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1108 arg.start_pfn = pfn;
1109 arg.nr_pages = nr_pages;
1110 node_states_check_changes_online(nr_pages, zone, &arg);
1112 nid = zone_to_nid(zone);
1114 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1115 ret = notifier_to_errno(ret);
1117 goto failed_addition;
1120 * If this zone is not populated, then it is not in zonelist.
1121 * This means the page allocator ignores this zone.
1122 * So, zonelist must be updated after online.
1124 mutex_lock(&zonelists_mutex);
1125 if (!populated_zone(zone)) {
1126 need_zonelists_rebuild = 1;
1127 build_all_zonelists(NULL, zone);
1130 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1131 online_pages_range);
1133 if (need_zonelists_rebuild)
1134 zone_pcp_reset(zone);
1135 mutex_unlock(&zonelists_mutex);
1136 goto failed_addition;
1139 zone->present_pages += onlined_pages;
1141 pgdat_resize_lock(zone->zone_pgdat, &flags);
1142 zone->zone_pgdat->node_present_pages += onlined_pages;
1143 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1145 if (onlined_pages) {
1146 node_states_set_node(nid, &arg);
1147 if (need_zonelists_rebuild)
1148 build_all_zonelists(NULL, NULL);
1150 zone_pcp_update(zone);
1153 mutex_unlock(&zonelists_mutex);
1155 init_per_zone_wmark_min();
1157 if (onlined_pages) {
1162 vm_total_pages = nr_free_pagecache_pages();
1164 writeback_set_ratelimit();
1167 memory_notify(MEM_ONLINE, &arg);
1171 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1172 (unsigned long long) pfn << PAGE_SHIFT,
1173 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1174 memory_notify(MEM_CANCEL_ONLINE, &arg);
1177 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1179 static void reset_node_present_pages(pg_data_t *pgdat)
1183 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1184 z->present_pages = 0;
1186 pgdat->node_present_pages = 0;
1189 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1190 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1192 struct pglist_data *pgdat;
1193 unsigned long zones_size[MAX_NR_ZONES] = {0};
1194 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1195 unsigned long start_pfn = PFN_DOWN(start);
1197 pgdat = NODE_DATA(nid);
1199 pgdat = arch_alloc_nodedata(nid);
1203 arch_refresh_nodedata(nid, pgdat);
1206 * Reset the nr_zones, order and classzone_idx before reuse.
1207 * Note that kswapd will init kswapd_classzone_idx properly
1208 * when it starts in the near future.
1210 pgdat->nr_zones = 0;
1211 pgdat->kswapd_order = 0;
1212 pgdat->kswapd_classzone_idx = 0;
1215 /* we can use NODE_DATA(nid) from here */
1217 /* init node's zones as empty zones, we don't have any present pages.*/
1218 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1219 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1222 * The node we allocated has no zone fallback lists. For avoiding
1223 * to access not-initialized zonelist, build here.
1225 mutex_lock(&zonelists_mutex);
1226 build_all_zonelists(pgdat, NULL);
1227 mutex_unlock(&zonelists_mutex);
1230 * zone->managed_pages is set to an approximate value in
1231 * free_area_init_core(), which will cause
1232 * /sys/device/system/node/nodeX/meminfo has wrong data.
1233 * So reset it to 0 before any memory is onlined.
1235 reset_node_managed_pages(pgdat);
1238 * When memory is hot-added, all the memory is in offline state. So
1239 * clear all zones' present_pages because they will be updated in
1240 * online_pages() and offline_pages().
1242 reset_node_present_pages(pgdat);
1247 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1249 arch_refresh_nodedata(nid, NULL);
1250 free_percpu(pgdat->per_cpu_nodestats);
1251 arch_free_nodedata(pgdat);
1257 * try_online_node - online a node if offlined
1259 * called by cpu_up() to online a node without onlined memory.
1261 int try_online_node(int nid)
1266 if (node_online(nid))
1269 mem_hotplug_begin();
1270 pgdat = hotadd_new_pgdat(nid, 0);
1272 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1276 node_set_online(nid);
1277 ret = register_one_node(nid);
1280 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1281 mutex_lock(&zonelists_mutex);
1282 build_all_zonelists(NULL, NULL);
1283 mutex_unlock(&zonelists_mutex);
1291 static int check_hotplug_memory_range(u64 start, u64 size)
1293 u64 start_pfn = PFN_DOWN(start);
1294 u64 nr_pages = size >> PAGE_SHIFT;
1296 /* Memory range must be aligned with section */
1297 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1298 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1299 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1300 (unsigned long long)start,
1301 (unsigned long long)size);
1309 * If movable zone has already been setup, newly added memory should be check.
1310 * If its address is higher than movable zone, it should be added as movable.
1311 * Without this check, movable zone may overlap with other zone.
1313 static int should_add_memory_movable(int nid, u64 start, u64 size)
1315 unsigned long start_pfn = start >> PAGE_SHIFT;
1316 pg_data_t *pgdat = NODE_DATA(nid);
1317 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1319 if (zone_is_empty(movable_zone))
1322 if (movable_zone->zone_start_pfn <= start_pfn)
1328 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1331 #ifdef CONFIG_ZONE_DEVICE
1335 if (should_add_memory_movable(nid, start, size))
1336 return ZONE_MOVABLE;
1338 return zone_default;
1341 static int online_memory_block(struct memory_block *mem, void *arg)
1343 return device_online(&mem->dev);
1346 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1347 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1350 pg_data_t *pgdat = NULL;
1356 size = resource_size(res);
1358 ret = check_hotplug_memory_range(start, size);
1362 { /* Stupid hack to suppress address-never-null warning */
1363 void *p = NODE_DATA(nid);
1367 mem_hotplug_begin();
1370 * Add new range to memblock so that when hotadd_new_pgdat() is called
1371 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1372 * this new range and calculate total pages correctly. The range will
1373 * be removed at hot-remove time.
1375 memblock_add_node(start, size, nid);
1377 new_node = !node_online(nid);
1379 pgdat = hotadd_new_pgdat(nid, start);
1385 /* call arch's memory hotadd */
1386 ret = arch_add_memory(nid, start, size, false);
1391 /* we online node here. we can't roll back from here. */
1392 node_set_online(nid);
1395 unsigned long start_pfn = start >> PAGE_SHIFT;
1396 unsigned long nr_pages = size >> PAGE_SHIFT;
1398 ret = __register_one_node(nid);
1403 * link memory sections under this node. This is already
1404 * done when creatig memory section in register_new_memory
1405 * but that depends to have the node registered so offline
1406 * nodes have to go through register_node.
1407 * TODO clean up this mess.
1409 ret = link_mem_sections(nid, start_pfn, nr_pages);
1412 * If sysfs file of new node can't create, cpu on the node
1413 * can't be hot-added. There is no rollback way now.
1414 * So, check by BUG_ON() to catch it reluctantly..
1419 /* create new memmap entry */
1420 firmware_map_add_hotplug(start, start + size, "System RAM");
1422 /* online pages if requested */
1424 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1425 NULL, online_memory_block);
1430 /* rollback pgdat allocation and others */
1432 rollback_node_hotadd(nid, pgdat);
1433 memblock_remove(start, size);
1439 EXPORT_SYMBOL_GPL(add_memory_resource);
1441 int __ref add_memory(int nid, u64 start, u64 size)
1443 struct resource *res;
1446 res = register_memory_resource(start, size);
1448 return PTR_ERR(res);
1450 ret = add_memory_resource(nid, res, memhp_auto_online);
1452 release_memory_resource(res);
1455 EXPORT_SYMBOL_GPL(add_memory);
1457 #ifdef CONFIG_MEMORY_HOTREMOVE
1459 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1460 * set and the size of the free page is given by page_order(). Using this,
1461 * the function determines if the pageblock contains only free pages.
1462 * Due to buddy contraints, a free page at least the size of a pageblock will
1463 * be located at the start of the pageblock
1465 static inline int pageblock_free(struct page *page)
1467 return PageBuddy(page) && page_order(page) >= pageblock_order;
1470 /* Return the start of the next active pageblock after a given page */
1471 static struct page *next_active_pageblock(struct page *page)
1473 /* Ensure the starting page is pageblock-aligned */
1474 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1476 /* If the entire pageblock is free, move to the end of free page */
1477 if (pageblock_free(page)) {
1479 /* be careful. we don't have locks, page_order can be changed.*/
1480 order = page_order(page);
1481 if ((order < MAX_ORDER) && (order >= pageblock_order))
1482 return page + (1 << order);
1485 return page + pageblock_nr_pages;
1488 /* Checks if this range of memory is likely to be hot-removable. */
1489 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1491 struct page *page = pfn_to_page(start_pfn);
1492 struct page *end_page = page + nr_pages;
1494 /* Check the starting page of each pageblock within the range */
1495 for (; page < end_page; page = next_active_pageblock(page)) {
1496 if (!is_pageblock_removable_nolock(page))
1501 /* All pageblocks in the memory block are likely to be hot-removable */
1506 * Confirm all pages in a range [start, end) belong to the same zone.
1507 * When true, return its valid [start, end).
1509 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1510 unsigned long *valid_start, unsigned long *valid_end)
1512 unsigned long pfn, sec_end_pfn;
1513 unsigned long start, end;
1514 struct zone *zone = NULL;
1517 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1519 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1520 /* Make sure the memory section is present first */
1521 if (!present_section_nr(pfn_to_section_nr(pfn)))
1523 for (; pfn < sec_end_pfn && pfn < end_pfn;
1524 pfn += MAX_ORDER_NR_PAGES) {
1526 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1527 while ((i < MAX_ORDER_NR_PAGES) &&
1528 !pfn_valid_within(pfn + i))
1530 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1532 page = pfn_to_page(pfn + i);
1533 if (zone && page_zone(page) != zone)
1537 zone = page_zone(page);
1538 end = pfn + MAX_ORDER_NR_PAGES;
1543 *valid_start = start;
1544 *valid_end = min(end, end_pfn);
1552 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1553 * non-lru movable pages and hugepages). We scan pfn because it's much
1554 * easier than scanning over linked list. This function returns the pfn
1555 * of the first found movable page if it's found, otherwise 0.
1557 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1561 for (pfn = start; pfn < end; pfn++) {
1562 if (pfn_valid(pfn)) {
1563 page = pfn_to_page(pfn);
1566 if (__PageMovable(page))
1568 if (PageHuge(page)) {
1569 if (page_huge_active(page))
1572 pfn = round_up(pfn + 1,
1573 1 << compound_order(page)) - 1;
1580 static struct page *new_node_page(struct page *page, unsigned long private,
1583 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1584 int nid = page_to_nid(page);
1585 nodemask_t nmask = node_states[N_MEMORY];
1586 struct page *new_page = NULL;
1589 * TODO: allocate a destination hugepage from a nearest neighbor node,
1590 * accordance with memory policy of the user process if possible. For
1591 * now as a simple work-around, we use the next node for destination.
1594 return alloc_huge_page_node(page_hstate(compound_head(page)),
1595 next_node_in(nid, nmask));
1597 node_clear(nid, nmask);
1599 if (PageHighMem(page)
1600 || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1601 gfp_mask |= __GFP_HIGHMEM;
1603 if (!nodes_empty(nmask))
1604 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1605 node_zonelist(nid, gfp_mask), &nmask);
1607 new_page = __alloc_pages(gfp_mask, 0,
1608 node_zonelist(nid, gfp_mask));
1613 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1615 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1619 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1620 int not_managed = 0;
1624 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1625 if (!pfn_valid(pfn))
1627 page = pfn_to_page(pfn);
1629 if (PageHuge(page)) {
1630 struct page *head = compound_head(page);
1631 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1632 if (compound_order(head) > PFN_SECTION_SHIFT) {
1636 if (isolate_huge_page(page, &source))
1637 move_pages -= 1 << compound_order(head);
1641 if (!get_page_unless_zero(page))
1644 * We can skip free pages. And we can deal with pages on
1645 * LRU and non-lru movable pages.
1648 ret = isolate_lru_page(page);
1650 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1651 if (!ret) { /* Success */
1653 list_add_tail(&page->lru, &source);
1655 if (!__PageMovable(page))
1656 inc_node_page_state(page, NR_ISOLATED_ANON +
1657 page_is_file_cache(page));
1660 #ifdef CONFIG_DEBUG_VM
1661 pr_alert("failed to isolate pfn %lx\n", pfn);
1662 dump_page(page, "isolation failed");
1665 /* Because we don't have big zone->lock. we should
1666 check this again here. */
1667 if (page_count(page)) {
1674 if (!list_empty(&source)) {
1676 putback_movable_pages(&source);
1680 /* Allocate a new page from the nearest neighbor node */
1681 ret = migrate_pages(&source, new_node_page, NULL, 0,
1682 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1684 putback_movable_pages(&source);
1691 * remove from free_area[] and mark all as Reserved.
1694 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1697 __offline_isolated_pages(start, start + nr_pages);
1702 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1704 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1705 offline_isolated_pages_cb);
1709 * Check all pages in range, recoreded as memory resource, are isolated.
1712 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1716 long offlined = *(long *)data;
1717 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1718 offlined = nr_pages;
1720 *(long *)data += offlined;
1725 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1730 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1731 check_pages_isolated_cb);
1733 offlined = (long)ret;
1737 #ifdef CONFIG_MOVABLE_NODE
1739 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1742 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1746 #else /* CONFIG_MOVABLE_NODE */
1747 /* ensure the node has NORMAL memory if it is still online */
1748 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1750 struct pglist_data *pgdat = zone->zone_pgdat;
1751 unsigned long present_pages = 0;
1754 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1755 present_pages += pgdat->node_zones[zt].present_pages;
1757 if (present_pages > nr_pages)
1761 for (; zt <= ZONE_MOVABLE; zt++)
1762 present_pages += pgdat->node_zones[zt].present_pages;
1765 * we can't offline the last normal memory until all
1766 * higher memory is offlined.
1768 return present_pages == 0;
1770 #endif /* CONFIG_MOVABLE_NODE */
1772 static int __init cmdline_parse_movable_node(char *p)
1774 #ifdef CONFIG_MOVABLE_NODE
1775 movable_node_enabled = true;
1777 pr_warn("movable_node option not supported\n");
1781 early_param("movable_node", cmdline_parse_movable_node);
1783 /* check which state of node_states will be changed when offline memory */
1784 static void node_states_check_changes_offline(unsigned long nr_pages,
1785 struct zone *zone, struct memory_notify *arg)
1787 struct pglist_data *pgdat = zone->zone_pgdat;
1788 unsigned long present_pages = 0;
1789 enum zone_type zt, zone_last = ZONE_NORMAL;
1792 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1793 * contains nodes which have zones of 0...ZONE_NORMAL,
1794 * set zone_last to ZONE_NORMAL.
1796 * If we don't have HIGHMEM nor movable node,
1797 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1798 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1800 if (N_MEMORY == N_NORMAL_MEMORY)
1801 zone_last = ZONE_MOVABLE;
1804 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1805 * If the memory to be offline is in a zone of 0...zone_last,
1806 * and it is the last present memory, 0...zone_last will
1807 * become empty after offline , thus we can determind we will
1808 * need to clear the node from node_states[N_NORMAL_MEMORY].
1810 for (zt = 0; zt <= zone_last; zt++)
1811 present_pages += pgdat->node_zones[zt].present_pages;
1812 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1813 arg->status_change_nid_normal = zone_to_nid(zone);
1815 arg->status_change_nid_normal = -1;
1817 #ifdef CONFIG_HIGHMEM
1819 * If we have movable node, node_states[N_HIGH_MEMORY]
1820 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1821 * set zone_last to ZONE_HIGHMEM.
1823 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1824 * contains nodes which have zones of 0...ZONE_MOVABLE,
1825 * set zone_last to ZONE_MOVABLE.
1827 zone_last = ZONE_HIGHMEM;
1828 if (N_MEMORY == N_HIGH_MEMORY)
1829 zone_last = ZONE_MOVABLE;
1831 for (; zt <= zone_last; zt++)
1832 present_pages += pgdat->node_zones[zt].present_pages;
1833 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1834 arg->status_change_nid_high = zone_to_nid(zone);
1836 arg->status_change_nid_high = -1;
1838 arg->status_change_nid_high = arg->status_change_nid_normal;
1842 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1844 zone_last = ZONE_MOVABLE;
1847 * check whether node_states[N_HIGH_MEMORY] will be changed
1848 * If we try to offline the last present @nr_pages from the node,
1849 * we can determind we will need to clear the node from
1850 * node_states[N_HIGH_MEMORY].
1852 for (; zt <= zone_last; zt++)
1853 present_pages += pgdat->node_zones[zt].present_pages;
1854 if (nr_pages >= present_pages)
1855 arg->status_change_nid = zone_to_nid(zone);
1857 arg->status_change_nid = -1;
1860 static void node_states_clear_node(int node, struct memory_notify *arg)
1862 if (arg->status_change_nid_normal >= 0)
1863 node_clear_state(node, N_NORMAL_MEMORY);
1865 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1866 (arg->status_change_nid_high >= 0))
1867 node_clear_state(node, N_HIGH_MEMORY);
1869 if ((N_MEMORY != N_HIGH_MEMORY) &&
1870 (arg->status_change_nid >= 0))
1871 node_clear_state(node, N_MEMORY);
1874 static int __ref __offline_pages(unsigned long start_pfn,
1875 unsigned long end_pfn, unsigned long timeout)
1877 unsigned long pfn, nr_pages, expire;
1878 long offlined_pages;
1879 int ret, drain, retry_max, node;
1880 unsigned long flags;
1881 unsigned long valid_start, valid_end;
1883 struct memory_notify arg;
1885 /* at least, alignment against pageblock is necessary */
1886 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1888 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1890 /* This makes hotplug much easier...and readable.
1891 we assume this for now. .*/
1892 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1895 zone = page_zone(pfn_to_page(valid_start));
1896 node = zone_to_nid(zone);
1897 nr_pages = end_pfn - start_pfn;
1899 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1902 /* set above range as isolated */
1903 ret = start_isolate_page_range(start_pfn, end_pfn,
1904 MIGRATE_MOVABLE, true);
1908 arg.start_pfn = start_pfn;
1909 arg.nr_pages = nr_pages;
1910 node_states_check_changes_offline(nr_pages, zone, &arg);
1912 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1913 ret = notifier_to_errno(ret);
1915 goto failed_removal;
1918 expire = jiffies + timeout;
1922 /* start memory hot removal */
1924 if (time_after(jiffies, expire))
1925 goto failed_removal;
1927 if (signal_pending(current))
1928 goto failed_removal;
1931 lru_add_drain_all();
1933 drain_all_pages(zone);
1936 pfn = scan_movable_pages(start_pfn, end_pfn);
1937 if (pfn) { /* We have movable pages */
1938 ret = do_migrate_range(pfn, end_pfn);
1944 if (--retry_max == 0)
1945 goto failed_removal;
1951 /* drain all zone's lru pagevec, this is asynchronous... */
1952 lru_add_drain_all();
1954 /* drain pcp pages, this is synchronous. */
1955 drain_all_pages(zone);
1957 * dissolve free hugepages in the memory block before doing offlining
1958 * actually in order to make hugetlbfs's object counting consistent.
1960 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1962 goto failed_removal;
1964 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1965 if (offlined_pages < 0) {
1967 goto failed_removal;
1969 pr_info("Offlined Pages %ld\n", offlined_pages);
1970 /* Ok, all of our target is isolated.
1971 We cannot do rollback at this point. */
1972 offline_isolated_pages(start_pfn, end_pfn);
1973 /* reset pagetype flags and makes migrate type to be MOVABLE */
1974 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1975 /* removal success */
1976 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1977 zone->present_pages -= offlined_pages;
1979 pgdat_resize_lock(zone->zone_pgdat, &flags);
1980 zone->zone_pgdat->node_present_pages -= offlined_pages;
1981 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1983 init_per_zone_wmark_min();
1985 if (!populated_zone(zone)) {
1986 zone_pcp_reset(zone);
1987 mutex_lock(&zonelists_mutex);
1988 build_all_zonelists(NULL, NULL);
1989 mutex_unlock(&zonelists_mutex);
1991 zone_pcp_update(zone);
1993 node_states_clear_node(node, &arg);
1994 if (arg.status_change_nid >= 0) {
1996 kcompactd_stop(node);
1999 vm_total_pages = nr_free_pagecache_pages();
2000 writeback_set_ratelimit();
2002 memory_notify(MEM_OFFLINE, &arg);
2006 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
2007 (unsigned long long) start_pfn << PAGE_SHIFT,
2008 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
2009 memory_notify(MEM_CANCEL_OFFLINE, &arg);
2010 /* pushback to free area */
2011 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
2015 /* Must be protected by mem_hotplug_begin() */
2016 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
2018 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
2020 #endif /* CONFIG_MEMORY_HOTREMOVE */
2023 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
2024 * @start_pfn: start pfn of the memory range
2025 * @end_pfn: end pfn of the memory range
2026 * @arg: argument passed to func
2027 * @func: callback for each memory section walked
2029 * This function walks through all present mem sections in range
2030 * [start_pfn, end_pfn) and call func on each mem section.
2032 * Returns the return value of func.
2034 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2035 void *arg, int (*func)(struct memory_block *, void *))
2037 struct memory_block *mem = NULL;
2038 struct mem_section *section;
2039 unsigned long pfn, section_nr;
2042 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2043 section_nr = pfn_to_section_nr(pfn);
2044 if (!present_section_nr(section_nr))
2047 section = __nr_to_section(section_nr);
2048 /* same memblock? */
2050 if ((section_nr >= mem->start_section_nr) &&
2051 (section_nr <= mem->end_section_nr))
2054 mem = find_memory_block_hinted(section, mem);
2058 ret = func(mem, arg);
2060 kobject_put(&mem->dev.kobj);
2066 kobject_put(&mem->dev.kobj);
2071 #ifdef CONFIG_MEMORY_HOTREMOVE
2072 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2074 int ret = !is_memblock_offlined(mem);
2076 if (unlikely(ret)) {
2077 phys_addr_t beginpa, endpa;
2079 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2080 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2081 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2088 static int check_cpu_on_node(pg_data_t *pgdat)
2092 for_each_present_cpu(cpu) {
2093 if (cpu_to_node(cpu) == pgdat->node_id)
2095 * the cpu on this node isn't removed, and we can't
2096 * offline this node.
2104 static void unmap_cpu_on_node(pg_data_t *pgdat)
2106 #ifdef CONFIG_ACPI_NUMA
2109 for_each_possible_cpu(cpu)
2110 if (cpu_to_node(cpu) == pgdat->node_id)
2111 numa_clear_node(cpu);
2115 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2119 ret = check_cpu_on_node(pgdat);
2124 * the node will be offlined when we come here, so we can clear
2125 * the cpu_to_node() now.
2128 unmap_cpu_on_node(pgdat);
2135 * Offline a node if all memory sections and cpus of the node are removed.
2137 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2138 * and online/offline operations before this call.
2140 void try_offline_node(int nid)
2142 pg_data_t *pgdat = NODE_DATA(nid);
2143 unsigned long start_pfn = pgdat->node_start_pfn;
2144 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2147 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2148 unsigned long section_nr = pfn_to_section_nr(pfn);
2150 if (!present_section_nr(section_nr))
2153 if (pfn_to_nid(pfn) != nid)
2157 * some memory sections of this node are not removed, and we
2158 * can't offline node now.
2163 if (check_and_unmap_cpu_on_node(pgdat))
2167 * all memory/cpu of this node are removed, we can offline this
2170 node_set_offline(nid);
2171 unregister_one_node(nid);
2173 EXPORT_SYMBOL(try_offline_node);
2178 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2179 * and online/offline operations before this call, as required by
2180 * try_offline_node().
2182 void __ref remove_memory(int nid, u64 start, u64 size)
2186 BUG_ON(check_hotplug_memory_range(start, size));
2188 mem_hotplug_begin();
2191 * All memory blocks must be offlined before removing memory. Check
2192 * whether all memory blocks in question are offline and trigger a BUG()
2193 * if this is not the case.
2195 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2196 check_memblock_offlined_cb);
2200 /* remove memmap entry */
2201 firmware_map_remove(start, start + size, "System RAM");
2202 memblock_free(start, size);
2203 memblock_remove(start, size);
2205 arch_remove_memory(start, size);
2207 try_offline_node(nid);
2211 EXPORT_SYMBOL_GPL(remove_memory);
2212 #endif /* CONFIG_MEMORY_HOTREMOVE */