Back to home page

LXR

 
 

    


0001 /*
0002  *  linux/mm/memory_hotplug.c
0003  *
0004  *  Copyright (C)
0005  */
0006 
0007 #include <linux/stddef.h>
0008 #include <linux/mm.h>
0009 #include <linux/swap.h>
0010 #include <linux/interrupt.h>
0011 #include <linux/pagemap.h>
0012 #include <linux/compiler.h>
0013 #include <linux/export.h>
0014 #include <linux/pagevec.h>
0015 #include <linux/writeback.h>
0016 #include <linux/slab.h>
0017 #include <linux/sysctl.h>
0018 #include <linux/cpu.h>
0019 #include <linux/memory.h>
0020 #include <linux/memremap.h>
0021 #include <linux/memory_hotplug.h>
0022 #include <linux/highmem.h>
0023 #include <linux/vmalloc.h>
0024 #include <linux/ioport.h>
0025 #include <linux/delay.h>
0026 #include <linux/migrate.h>
0027 #include <linux/page-isolation.h>
0028 #include <linux/pfn.h>
0029 #include <linux/suspend.h>
0030 #include <linux/mm_inline.h>
0031 #include <linux/firmware-map.h>
0032 #include <linux/stop_machine.h>
0033 #include <linux/hugetlb.h>
0034 #include <linux/memblock.h>
0035 #include <linux/bootmem.h>
0036 #include <linux/compaction.h>
0037 
0038 #include <asm/tlbflush.h>
0039 
0040 #include "internal.h"
0041 
0042 /*
0043  * online_page_callback contains pointer to current page onlining function.
0044  * Initially it is generic_online_page(). If it is required it could be
0045  * changed by calling set_online_page_callback() for callback registration
0046  * and restore_online_page_callback() for generic callback restore.
0047  */
0048 
0049 static void generic_online_page(struct page *page);
0050 
0051 static online_page_callback_t online_page_callback = generic_online_page;
0052 static DEFINE_MUTEX(online_page_callback_lock);
0053 
0054 /* The same as the cpu_hotplug lock, but for memory hotplug. */
0055 static struct {
0056     struct task_struct *active_writer;
0057     struct mutex lock; /* Synchronizes accesses to refcount, */
0058     /*
0059      * Also blocks the new readers during
0060      * an ongoing mem hotplug operation.
0061      */
0062     int refcount;
0063 
0064 #ifdef CONFIG_DEBUG_LOCK_ALLOC
0065     struct lockdep_map dep_map;
0066 #endif
0067 } mem_hotplug = {
0068     .active_writer = NULL,
0069     .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
0070     .refcount = 0,
0071 #ifdef CONFIG_DEBUG_LOCK_ALLOC
0072     .dep_map = {.name = "mem_hotplug.lock" },
0073 #endif
0074 };
0075 
0076 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
0077 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
0078 #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
0079 #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
0080 
0081 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
0082 bool memhp_auto_online;
0083 #else
0084 bool memhp_auto_online = true;
0085 #endif
0086 EXPORT_SYMBOL_GPL(memhp_auto_online);
0087 
0088 static int __init setup_memhp_default_state(char *str)
0089 {
0090     if (!strcmp(str, "online"))
0091         memhp_auto_online = true;
0092     else if (!strcmp(str, "offline"))
0093         memhp_auto_online = false;
0094 
0095     return 1;
0096 }
0097 __setup("memhp_default_state=", setup_memhp_default_state);
0098 
0099 void get_online_mems(void)
0100 {
0101     might_sleep();
0102     if (mem_hotplug.active_writer == current)
0103         return;
0104     memhp_lock_acquire_read();
0105     mutex_lock(&mem_hotplug.lock);
0106     mem_hotplug.refcount++;
0107     mutex_unlock(&mem_hotplug.lock);
0108 
0109 }
0110 
0111 void put_online_mems(void)
0112 {
0113     if (mem_hotplug.active_writer == current)
0114         return;
0115     mutex_lock(&mem_hotplug.lock);
0116 
0117     if (WARN_ON(!mem_hotplug.refcount))
0118         mem_hotplug.refcount++; /* try to fix things up */
0119 
0120     if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
0121         wake_up_process(mem_hotplug.active_writer);
0122     mutex_unlock(&mem_hotplug.lock);
0123     memhp_lock_release();
0124 
0125 }
0126 
0127 void mem_hotplug_begin(void)
0128 {
0129     mem_hotplug.active_writer = current;
0130 
0131     memhp_lock_acquire();
0132     for (;;) {
0133         mutex_lock(&mem_hotplug.lock);
0134         if (likely(!mem_hotplug.refcount))
0135             break;
0136         __set_current_state(TASK_UNINTERRUPTIBLE);
0137         mutex_unlock(&mem_hotplug.lock);
0138         schedule();
0139     }
0140 }
0141 
0142 void mem_hotplug_done(void)
0143 {
0144     mem_hotplug.active_writer = NULL;
0145     mutex_unlock(&mem_hotplug.lock);
0146     memhp_lock_release();
0147 }
0148 
0149 /* add this memory to iomem resource */
0150 static struct resource *register_memory_resource(u64 start, u64 size)
0151 {
0152     struct resource *res;
0153     res = kzalloc(sizeof(struct resource), GFP_KERNEL);
0154     if (!res)
0155         return ERR_PTR(-ENOMEM);
0156 
0157     res->name = "System RAM";
0158     res->start = start;
0159     res->end = start + size - 1;
0160     res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
0161     if (request_resource(&iomem_resource, res) < 0) {
0162         pr_debug("System RAM resource %pR cannot be added\n", res);
0163         kfree(res);
0164         return ERR_PTR(-EEXIST);
0165     }
0166     return res;
0167 }
0168 
0169 static void release_memory_resource(struct resource *res)
0170 {
0171     if (!res)
0172         return;
0173     release_resource(res);
0174     kfree(res);
0175     return;
0176 }
0177 
0178 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
0179 void get_page_bootmem(unsigned long info,  struct page *page,
0180               unsigned long type)
0181 {
0182     page->lru.next = (struct list_head *) type;
0183     SetPagePrivate(page);
0184     set_page_private(page, info);
0185     page_ref_inc(page);
0186 }
0187 
0188 void put_page_bootmem(struct page *page)
0189 {
0190     unsigned long type;
0191 
0192     type = (unsigned long) page->lru.next;
0193     BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
0194            type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
0195 
0196     if (page_ref_dec_return(page) == 1) {
0197         ClearPagePrivate(page);
0198         set_page_private(page, 0);
0199         INIT_LIST_HEAD(&page->lru);
0200         free_reserved_page(page);
0201     }
0202 }
0203 
0204 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
0205 #ifndef CONFIG_SPARSEMEM_VMEMMAP
0206 static void register_page_bootmem_info_section(unsigned long start_pfn)
0207 {
0208     unsigned long *usemap, mapsize, section_nr, i;
0209     struct mem_section *ms;
0210     struct page *page, *memmap;
0211 
0212     section_nr = pfn_to_section_nr(start_pfn);
0213     ms = __nr_to_section(section_nr);
0214 
0215     /* Get section's memmap address */
0216     memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
0217 
0218     /*
0219      * Get page for the memmap's phys address
0220      * XXX: need more consideration for sparse_vmemmap...
0221      */
0222     page = virt_to_page(memmap);
0223     mapsize = sizeof(struct page) * PAGES_PER_SECTION;
0224     mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
0225 
0226     /* remember memmap's page */
0227     for (i = 0; i < mapsize; i++, page++)
0228         get_page_bootmem(section_nr, page, SECTION_INFO);
0229 
0230     usemap = __nr_to_section(section_nr)->pageblock_flags;
0231     page = virt_to_page(usemap);
0232 
0233     mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
0234 
0235     for (i = 0; i < mapsize; i++, page++)
0236         get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
0237 
0238 }
0239 #else /* CONFIG_SPARSEMEM_VMEMMAP */
0240 static void register_page_bootmem_info_section(unsigned long start_pfn)
0241 {
0242     unsigned long *usemap, mapsize, section_nr, i;
0243     struct mem_section *ms;
0244     struct page *page, *memmap;
0245 
0246     if (!pfn_valid(start_pfn))
0247         return;
0248 
0249     section_nr = pfn_to_section_nr(start_pfn);
0250     ms = __nr_to_section(section_nr);
0251 
0252     memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
0253 
0254     register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
0255 
0256     usemap = __nr_to_section(section_nr)->pageblock_flags;
0257     page = virt_to_page(usemap);
0258 
0259     mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
0260 
0261     for (i = 0; i < mapsize; i++, page++)
0262         get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
0263 }
0264 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
0265 
0266 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
0267 {
0268     unsigned long i, pfn, end_pfn, nr_pages;
0269     int node = pgdat->node_id;
0270     struct page *page;
0271 
0272     nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
0273     page = virt_to_page(pgdat);
0274 
0275     for (i = 0; i < nr_pages; i++, page++)
0276         get_page_bootmem(node, page, NODE_INFO);
0277 
0278     pfn = pgdat->node_start_pfn;
0279     end_pfn = pgdat_end_pfn(pgdat);
0280 
0281     /* register section info */
0282     for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
0283         /*
0284          * Some platforms can assign the same pfn to multiple nodes - on
0285          * node0 as well as nodeN.  To avoid registering a pfn against
0286          * multiple nodes we check that this pfn does not already
0287          * reside in some other nodes.
0288          */
0289         if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
0290             register_page_bootmem_info_section(pfn);
0291     }
0292 }
0293 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
0294 
0295 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
0296                      unsigned long end_pfn)
0297 {
0298     unsigned long old_zone_end_pfn;
0299 
0300     zone_span_writelock(zone);
0301 
0302     old_zone_end_pfn = zone_end_pfn(zone);
0303     if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
0304         zone->zone_start_pfn = start_pfn;
0305 
0306     zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
0307                 zone->zone_start_pfn;
0308 
0309     zone_span_writeunlock(zone);
0310 }
0311 
0312 static void resize_zone(struct zone *zone, unsigned long start_pfn,
0313         unsigned long end_pfn)
0314 {
0315     zone_span_writelock(zone);
0316 
0317     if (end_pfn - start_pfn) {
0318         zone->zone_start_pfn = start_pfn;
0319         zone->spanned_pages = end_pfn - start_pfn;
0320     } else {
0321         /*
0322          * make it consist as free_area_init_core(),
0323          * if spanned_pages = 0, then keep start_pfn = 0
0324          */
0325         zone->zone_start_pfn = 0;
0326         zone->spanned_pages = 0;
0327     }
0328 
0329     zone_span_writeunlock(zone);
0330 }
0331 
0332 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
0333         unsigned long end_pfn)
0334 {
0335     enum zone_type zid = zone_idx(zone);
0336     int nid = zone->zone_pgdat->node_id;
0337     unsigned long pfn;
0338 
0339     for (pfn = start_pfn; pfn < end_pfn; pfn++)
0340         set_page_links(pfn_to_page(pfn), zid, nid, pfn);
0341 }
0342 
0343 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
0344  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
0345 static int __ref ensure_zone_is_initialized(struct zone *zone,
0346             unsigned long start_pfn, unsigned long num_pages)
0347 {
0348     if (!zone_is_initialized(zone))
0349         return init_currently_empty_zone(zone, start_pfn, num_pages);
0350 
0351     return 0;
0352 }
0353 
0354 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
0355         unsigned long start_pfn, unsigned long end_pfn)
0356 {
0357     int ret;
0358     unsigned long flags;
0359     unsigned long z1_start_pfn;
0360 
0361     ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
0362     if (ret)
0363         return ret;
0364 
0365     pgdat_resize_lock(z1->zone_pgdat, &flags);
0366 
0367     /* can't move pfns which are higher than @z2 */
0368     if (end_pfn > zone_end_pfn(z2))
0369         goto out_fail;
0370     /* the move out part must be at the left most of @z2 */
0371     if (start_pfn > z2->zone_start_pfn)
0372         goto out_fail;
0373     /* must included/overlap */
0374     if (end_pfn <= z2->zone_start_pfn)
0375         goto out_fail;
0376 
0377     /* use start_pfn for z1's start_pfn if z1 is empty */
0378     if (!zone_is_empty(z1))
0379         z1_start_pfn = z1->zone_start_pfn;
0380     else
0381         z1_start_pfn = start_pfn;
0382 
0383     resize_zone(z1, z1_start_pfn, end_pfn);
0384     resize_zone(z2, end_pfn, zone_end_pfn(z2));
0385 
0386     pgdat_resize_unlock(z1->zone_pgdat, &flags);
0387 
0388     fix_zone_id(z1, start_pfn, end_pfn);
0389 
0390     return 0;
0391 out_fail:
0392     pgdat_resize_unlock(z1->zone_pgdat, &flags);
0393     return -1;
0394 }
0395 
0396 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
0397         unsigned long start_pfn, unsigned long end_pfn)
0398 {
0399     int ret;
0400     unsigned long flags;
0401     unsigned long z2_end_pfn;
0402 
0403     ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
0404     if (ret)
0405         return ret;
0406 
0407     pgdat_resize_lock(z1->zone_pgdat, &flags);
0408 
0409     /* can't move pfns which are lower than @z1 */
0410     if (z1->zone_start_pfn > start_pfn)
0411         goto out_fail;
0412     /* the move out part mast at the right most of @z1 */
0413     if (zone_end_pfn(z1) >  end_pfn)
0414         goto out_fail;
0415     /* must included/overlap */
0416     if (start_pfn >= zone_end_pfn(z1))
0417         goto out_fail;
0418 
0419     /* use end_pfn for z2's end_pfn if z2 is empty */
0420     if (!zone_is_empty(z2))
0421         z2_end_pfn = zone_end_pfn(z2);
0422     else
0423         z2_end_pfn = end_pfn;
0424 
0425     resize_zone(z1, z1->zone_start_pfn, start_pfn);
0426     resize_zone(z2, start_pfn, z2_end_pfn);
0427 
0428     pgdat_resize_unlock(z1->zone_pgdat, &flags);
0429 
0430     fix_zone_id(z2, start_pfn, end_pfn);
0431 
0432     return 0;
0433 out_fail:
0434     pgdat_resize_unlock(z1->zone_pgdat, &flags);
0435     return -1;
0436 }
0437 
0438 static struct zone * __meminit move_pfn_range(int zone_shift,
0439         unsigned long start_pfn, unsigned long end_pfn)
0440 {
0441     struct zone *zone = page_zone(pfn_to_page(start_pfn));
0442     int ret = 0;
0443 
0444     if (zone_shift < 0)
0445         ret = move_pfn_range_left(zone + zone_shift, zone,
0446                       start_pfn, end_pfn);
0447     else if (zone_shift)
0448         ret = move_pfn_range_right(zone, zone + zone_shift,
0449                        start_pfn, end_pfn);
0450 
0451     if (ret)
0452         return NULL;
0453 
0454     return zone + zone_shift;
0455 }
0456 
0457 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
0458                       unsigned long end_pfn)
0459 {
0460     unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
0461 
0462     if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
0463         pgdat->node_start_pfn = start_pfn;
0464 
0465     pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
0466                     pgdat->node_start_pfn;
0467 }
0468 
0469 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
0470 {
0471     struct pglist_data *pgdat = zone->zone_pgdat;
0472     int nr_pages = PAGES_PER_SECTION;
0473     int nid = pgdat->node_id;
0474     int zone_type;
0475     unsigned long flags, pfn;
0476     int ret;
0477 
0478     zone_type = zone - pgdat->node_zones;
0479     ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
0480     if (ret)
0481         return ret;
0482 
0483     pgdat_resize_lock(zone->zone_pgdat, &flags);
0484     grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
0485     grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
0486             phys_start_pfn + nr_pages);
0487     pgdat_resize_unlock(zone->zone_pgdat, &flags);
0488     memmap_init_zone(nr_pages, nid, zone_type,
0489              phys_start_pfn, MEMMAP_HOTPLUG);
0490 
0491     /* online_page_range is called later and expects pages reserved */
0492     for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
0493         if (!pfn_valid(pfn))
0494             continue;
0495 
0496         SetPageReserved(pfn_to_page(pfn));
0497     }
0498     return 0;
0499 }
0500 
0501 static int __meminit __add_section(int nid, struct zone *zone,
0502                     unsigned long phys_start_pfn)
0503 {
0504     int ret;
0505 
0506     if (pfn_valid(phys_start_pfn))
0507         return -EEXIST;
0508 
0509     ret = sparse_add_one_section(zone, phys_start_pfn);
0510 
0511     if (ret < 0)
0512         return ret;
0513 
0514     ret = __add_zone(zone, phys_start_pfn);
0515 
0516     if (ret < 0)
0517         return ret;
0518 
0519     return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
0520 }
0521 
0522 /*
0523  * Reasonably generic function for adding memory.  It is
0524  * expected that archs that support memory hotplug will
0525  * call this function after deciding the zone to which to
0526  * add the new pages.
0527  */
0528 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
0529             unsigned long nr_pages)
0530 {
0531     unsigned long i;
0532     int err = 0;
0533     int start_sec, end_sec;
0534     struct vmem_altmap *altmap;
0535 
0536     clear_zone_contiguous(zone);
0537 
0538     /* during initialize mem_map, align hot-added range to section */
0539     start_sec = pfn_to_section_nr(phys_start_pfn);
0540     end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
0541 
0542     altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
0543     if (altmap) {
0544         /*
0545          * Validate altmap is within bounds of the total request
0546          */
0547         if (altmap->base_pfn != phys_start_pfn
0548                 || vmem_altmap_offset(altmap) > nr_pages) {
0549             pr_warn_once("memory add fail, invalid altmap\n");
0550             err = -EINVAL;
0551             goto out;
0552         }
0553         altmap->alloc = 0;
0554     }
0555 
0556     for (i = start_sec; i <= end_sec; i++) {
0557         err = __add_section(nid, zone, section_nr_to_pfn(i));
0558 
0559         /*
0560          * EEXIST is finally dealt with by ioresource collision
0561          * check. see add_memory() => register_memory_resource()
0562          * Warning will be printed if there is collision.
0563          */
0564         if (err && (err != -EEXIST))
0565             break;
0566         err = 0;
0567     }
0568     vmemmap_populate_print_last();
0569 out:
0570     set_zone_contiguous(zone);
0571     return err;
0572 }
0573 EXPORT_SYMBOL_GPL(__add_pages);
0574 
0575 #ifdef CONFIG_MEMORY_HOTREMOVE
0576 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
0577 static int find_smallest_section_pfn(int nid, struct zone *zone,
0578                      unsigned long start_pfn,
0579                      unsigned long end_pfn)
0580 {
0581     struct mem_section *ms;
0582 
0583     for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
0584         ms = __pfn_to_section(start_pfn);
0585 
0586         if (unlikely(!valid_section(ms)))
0587             continue;
0588 
0589         if (unlikely(pfn_to_nid(start_pfn) != nid))
0590             continue;
0591 
0592         if (zone && zone != page_zone(pfn_to_page(start_pfn)))
0593             continue;
0594 
0595         return start_pfn;
0596     }
0597 
0598     return 0;
0599 }
0600 
0601 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
0602 static int find_biggest_section_pfn(int nid, struct zone *zone,
0603                     unsigned long start_pfn,
0604                     unsigned long end_pfn)
0605 {
0606     struct mem_section *ms;
0607     unsigned long pfn;
0608 
0609     /* pfn is the end pfn of a memory section. */
0610     pfn = end_pfn - 1;
0611     for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
0612         ms = __pfn_to_section(pfn);
0613 
0614         if (unlikely(!valid_section(ms)))
0615             continue;
0616 
0617         if (unlikely(pfn_to_nid(pfn) != nid))
0618             continue;
0619 
0620         if (zone && zone != page_zone(pfn_to_page(pfn)))
0621             continue;
0622 
0623         return pfn;
0624     }
0625 
0626     return 0;
0627 }
0628 
0629 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
0630                  unsigned long end_pfn)
0631 {
0632     unsigned long zone_start_pfn = zone->zone_start_pfn;
0633     unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
0634     unsigned long zone_end_pfn = z;
0635     unsigned long pfn;
0636     struct mem_section *ms;
0637     int nid = zone_to_nid(zone);
0638 
0639     zone_span_writelock(zone);
0640     if (zone_start_pfn == start_pfn) {
0641         /*
0642          * If the section is smallest section in the zone, it need
0643          * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
0644          * In this case, we find second smallest valid mem_section
0645          * for shrinking zone.
0646          */
0647         pfn = find_smallest_section_pfn(nid, zone, end_pfn,
0648                         zone_end_pfn);
0649         if (pfn) {
0650             zone->zone_start_pfn = pfn;
0651             zone->spanned_pages = zone_end_pfn - pfn;
0652         }
0653     } else if (zone_end_pfn == end_pfn) {
0654         /*
0655          * If the section is biggest section in the zone, it need
0656          * shrink zone->spanned_pages.
0657          * In this case, we find second biggest valid mem_section for
0658          * shrinking zone.
0659          */
0660         pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
0661                            start_pfn);
0662         if (pfn)
0663             zone->spanned_pages = pfn - zone_start_pfn + 1;
0664     }
0665 
0666     /*
0667      * The section is not biggest or smallest mem_section in the zone, it
0668      * only creates a hole in the zone. So in this case, we need not
0669      * change the zone. But perhaps, the zone has only hole data. Thus
0670      * it check the zone has only hole or not.
0671      */
0672     pfn = zone_start_pfn;
0673     for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
0674         ms = __pfn_to_section(pfn);
0675 
0676         if (unlikely(!valid_section(ms)))
0677             continue;
0678 
0679         if (page_zone(pfn_to_page(pfn)) != zone)
0680             continue;
0681 
0682          /* If the section is current section, it continues the loop */
0683         if (start_pfn == pfn)
0684             continue;
0685 
0686         /* If we find valid section, we have nothing to do */
0687         zone_span_writeunlock(zone);
0688         return;
0689     }
0690 
0691     /* The zone has no valid section */
0692     zone->zone_start_pfn = 0;
0693     zone->spanned_pages = 0;
0694     zone_span_writeunlock(zone);
0695 }
0696 
0697 static void shrink_pgdat_span(struct pglist_data *pgdat,
0698                   unsigned long start_pfn, unsigned long end_pfn)
0699 {
0700     unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
0701     unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
0702     unsigned long pgdat_end_pfn = p;
0703     unsigned long pfn;
0704     struct mem_section *ms;
0705     int nid = pgdat->node_id;
0706 
0707     if (pgdat_start_pfn == start_pfn) {
0708         /*
0709          * If the section is smallest section in the pgdat, it need
0710          * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
0711          * In this case, we find second smallest valid mem_section
0712          * for shrinking zone.
0713          */
0714         pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
0715                         pgdat_end_pfn);
0716         if (pfn) {
0717             pgdat->node_start_pfn = pfn;
0718             pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
0719         }
0720     } else if (pgdat_end_pfn == end_pfn) {
0721         /*
0722          * If the section is biggest section in the pgdat, it need
0723          * shrink pgdat->node_spanned_pages.
0724          * In this case, we find second biggest valid mem_section for
0725          * shrinking zone.
0726          */
0727         pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
0728                            start_pfn);
0729         if (pfn)
0730             pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
0731     }
0732 
0733     /*
0734      * If the section is not biggest or smallest mem_section in the pgdat,
0735      * it only creates a hole in the pgdat. So in this case, we need not
0736      * change the pgdat.
0737      * But perhaps, the pgdat has only hole data. Thus it check the pgdat
0738      * has only hole or not.
0739      */
0740     pfn = pgdat_start_pfn;
0741     for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
0742         ms = __pfn_to_section(pfn);
0743 
0744         if (unlikely(!valid_section(ms)))
0745             continue;
0746 
0747         if (pfn_to_nid(pfn) != nid)
0748             continue;
0749 
0750          /* If the section is current section, it continues the loop */
0751         if (start_pfn == pfn)
0752             continue;
0753 
0754         /* If we find valid section, we have nothing to do */
0755         return;
0756     }
0757 
0758     /* The pgdat has no valid section */
0759     pgdat->node_start_pfn = 0;
0760     pgdat->node_spanned_pages = 0;
0761 }
0762 
0763 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
0764 {
0765     struct pglist_data *pgdat = zone->zone_pgdat;
0766     int nr_pages = PAGES_PER_SECTION;
0767     int zone_type;
0768     unsigned long flags;
0769 
0770     zone_type = zone - pgdat->node_zones;
0771 
0772     pgdat_resize_lock(zone->zone_pgdat, &flags);
0773     shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
0774     shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
0775     pgdat_resize_unlock(zone->zone_pgdat, &flags);
0776 }
0777 
0778 static int __remove_section(struct zone *zone, struct mem_section *ms,
0779         unsigned long map_offset)
0780 {
0781     unsigned long start_pfn;
0782     int scn_nr;
0783     int ret = -EINVAL;
0784 
0785     if (!valid_section(ms))
0786         return ret;
0787 
0788     ret = unregister_memory_section(ms);
0789     if (ret)
0790         return ret;
0791 
0792     scn_nr = __section_nr(ms);
0793     start_pfn = section_nr_to_pfn(scn_nr);
0794     __remove_zone(zone, start_pfn);
0795 
0796     sparse_remove_one_section(zone, ms, map_offset);
0797     return 0;
0798 }
0799 
0800 /**
0801  * __remove_pages() - remove sections of pages from a zone
0802  * @zone: zone from which pages need to be removed
0803  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
0804  * @nr_pages: number of pages to remove (must be multiple of section size)
0805  *
0806  * Generic helper function to remove section mappings and sysfs entries
0807  * for the section of the memory we are removing. Caller needs to make
0808  * sure that pages are marked reserved and zones are adjust properly by
0809  * calling offline_pages().
0810  */
0811 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
0812          unsigned long nr_pages)
0813 {
0814     unsigned long i;
0815     unsigned long map_offset = 0;
0816     int sections_to_remove, ret = 0;
0817 
0818     /* In the ZONE_DEVICE case device driver owns the memory region */
0819     if (is_dev_zone(zone)) {
0820         struct page *page = pfn_to_page(phys_start_pfn);
0821         struct vmem_altmap *altmap;
0822 
0823         altmap = to_vmem_altmap((unsigned long) page);
0824         if (altmap)
0825             map_offset = vmem_altmap_offset(altmap);
0826     } else {
0827         resource_size_t start, size;
0828 
0829         start = phys_start_pfn << PAGE_SHIFT;
0830         size = nr_pages * PAGE_SIZE;
0831 
0832         ret = release_mem_region_adjustable(&iomem_resource, start,
0833                     size);
0834         if (ret) {
0835             resource_size_t endres = start + size - 1;
0836 
0837             pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
0838                     &start, &endres, ret);
0839         }
0840     }
0841 
0842     clear_zone_contiguous(zone);
0843 
0844     /*
0845      * We can only remove entire sections
0846      */
0847     BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
0848     BUG_ON(nr_pages % PAGES_PER_SECTION);
0849 
0850     sections_to_remove = nr_pages / PAGES_PER_SECTION;
0851     for (i = 0; i < sections_to_remove; i++) {
0852         unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
0853 
0854         ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
0855         map_offset = 0;
0856         if (ret)
0857             break;
0858     }
0859 
0860     set_zone_contiguous(zone);
0861 
0862     return ret;
0863 }
0864 EXPORT_SYMBOL_GPL(__remove_pages);
0865 #endif /* CONFIG_MEMORY_HOTREMOVE */
0866 
0867 int set_online_page_callback(online_page_callback_t callback)
0868 {
0869     int rc = -EINVAL;
0870 
0871     get_online_mems();
0872     mutex_lock(&online_page_callback_lock);
0873 
0874     if (online_page_callback == generic_online_page) {
0875         online_page_callback = callback;
0876         rc = 0;
0877     }
0878 
0879     mutex_unlock(&online_page_callback_lock);
0880     put_online_mems();
0881 
0882     return rc;
0883 }
0884 EXPORT_SYMBOL_GPL(set_online_page_callback);
0885 
0886 int restore_online_page_callback(online_page_callback_t callback)
0887 {
0888     int rc = -EINVAL;
0889 
0890     get_online_mems();
0891     mutex_lock(&online_page_callback_lock);
0892 
0893     if (online_page_callback == callback) {
0894         online_page_callback = generic_online_page;
0895         rc = 0;
0896     }
0897 
0898     mutex_unlock(&online_page_callback_lock);
0899     put_online_mems();
0900 
0901     return rc;
0902 }
0903 EXPORT_SYMBOL_GPL(restore_online_page_callback);
0904 
0905 void __online_page_set_limits(struct page *page)
0906 {
0907 }
0908 EXPORT_SYMBOL_GPL(__online_page_set_limits);
0909 
0910 void __online_page_increment_counters(struct page *page)
0911 {
0912     adjust_managed_page_count(page, 1);
0913 }
0914 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
0915 
0916 void __online_page_free(struct page *page)
0917 {
0918     __free_reserved_page(page);
0919 }
0920 EXPORT_SYMBOL_GPL(__online_page_free);
0921 
0922 static void generic_online_page(struct page *page)
0923 {
0924     __online_page_set_limits(page);
0925     __online_page_increment_counters(page);
0926     __online_page_free(page);
0927 }
0928 
0929 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
0930             void *arg)
0931 {
0932     unsigned long i;
0933     unsigned long onlined_pages = *(unsigned long *)arg;
0934     struct page *page;
0935     if (PageReserved(pfn_to_page(start_pfn)))
0936         for (i = 0; i < nr_pages; i++) {
0937             page = pfn_to_page(start_pfn + i);
0938             (*online_page_callback)(page);
0939             onlined_pages++;
0940         }
0941     *(unsigned long *)arg = onlined_pages;
0942     return 0;
0943 }
0944 
0945 #ifdef CONFIG_MOVABLE_NODE
0946 /*
0947  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
0948  * normal memory.
0949  */
0950 static bool can_online_high_movable(struct zone *zone)
0951 {
0952     return true;
0953 }
0954 #else /* CONFIG_MOVABLE_NODE */
0955 /* ensure every online node has NORMAL memory */
0956 static bool can_online_high_movable(struct zone *zone)
0957 {
0958     return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
0959 }
0960 #endif /* CONFIG_MOVABLE_NODE */
0961 
0962 /* check which state of node_states will be changed when online memory */
0963 static void node_states_check_changes_online(unsigned long nr_pages,
0964     struct zone *zone, struct memory_notify *arg)
0965 {
0966     int nid = zone_to_nid(zone);
0967     enum zone_type zone_last = ZONE_NORMAL;
0968 
0969     /*
0970      * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
0971      * contains nodes which have zones of 0...ZONE_NORMAL,
0972      * set zone_last to ZONE_NORMAL.
0973      *
0974      * If we don't have HIGHMEM nor movable node,
0975      * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
0976      * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
0977      */
0978     if (N_MEMORY == N_NORMAL_MEMORY)
0979         zone_last = ZONE_MOVABLE;
0980 
0981     /*
0982      * if the memory to be online is in a zone of 0...zone_last, and
0983      * the zones of 0...zone_last don't have memory before online, we will
0984      * need to set the node to node_states[N_NORMAL_MEMORY] after
0985      * the memory is online.
0986      */
0987     if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
0988         arg->status_change_nid_normal = nid;
0989     else
0990         arg->status_change_nid_normal = -1;
0991 
0992 #ifdef CONFIG_HIGHMEM
0993     /*
0994      * If we have movable node, node_states[N_HIGH_MEMORY]
0995      * contains nodes which have zones of 0...ZONE_HIGHMEM,
0996      * set zone_last to ZONE_HIGHMEM.
0997      *
0998      * If we don't have movable node, node_states[N_NORMAL_MEMORY]
0999      * contains nodes which have zones of 0...ZONE_MOVABLE,
1000      * set zone_last to ZONE_MOVABLE.
1001      */
1002     zone_last = ZONE_HIGHMEM;
1003     if (N_MEMORY == N_HIGH_MEMORY)
1004         zone_last = ZONE_MOVABLE;
1005 
1006     if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1007         arg->status_change_nid_high = nid;
1008     else
1009         arg->status_change_nid_high = -1;
1010 #else
1011     arg->status_change_nid_high = arg->status_change_nid_normal;
1012 #endif
1013 
1014     /*
1015      * if the node don't have memory befor online, we will need to
1016      * set the node to node_states[N_MEMORY] after the memory
1017      * is online.
1018      */
1019     if (!node_state(nid, N_MEMORY))
1020         arg->status_change_nid = nid;
1021     else
1022         arg->status_change_nid = -1;
1023 }
1024 
1025 static void node_states_set_node(int node, struct memory_notify *arg)
1026 {
1027     if (arg->status_change_nid_normal >= 0)
1028         node_set_state(node, N_NORMAL_MEMORY);
1029 
1030     if (arg->status_change_nid_high >= 0)
1031         node_set_state(node, N_HIGH_MEMORY);
1032 
1033     node_set_state(node, N_MEMORY);
1034 }
1035 
1036 bool zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1037            enum zone_type target, int *zone_shift)
1038 {
1039     struct zone *zone = page_zone(pfn_to_page(pfn));
1040     enum zone_type idx = zone_idx(zone);
1041     int i;
1042 
1043     *zone_shift = 0;
1044 
1045     if (idx < target) {
1046         /* pages must be at end of current zone */
1047         if (pfn + nr_pages != zone_end_pfn(zone))
1048             return false;
1049 
1050         /* no zones in use between current zone and target */
1051         for (i = idx + 1; i < target; i++)
1052             if (zone_is_initialized(zone - idx + i))
1053                 return false;
1054     }
1055 
1056     if (target < idx) {
1057         /* pages must be at beginning of current zone */
1058         if (pfn != zone->zone_start_pfn)
1059             return false;
1060 
1061         /* no zones in use between current zone and target */
1062         for (i = target + 1; i < idx; i++)
1063             if (zone_is_initialized(zone - idx + i))
1064                 return false;
1065     }
1066 
1067     *zone_shift = target - idx;
1068     return true;
1069 }
1070 
1071 /* Must be protected by mem_hotplug_begin() */
1072 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1073 {
1074     unsigned long flags;
1075     unsigned long onlined_pages = 0;
1076     struct zone *zone;
1077     int need_zonelists_rebuild = 0;
1078     int nid;
1079     int ret;
1080     struct memory_notify arg;
1081     int zone_shift = 0;
1082 
1083     /*
1084      * This doesn't need a lock to do pfn_to_page().
1085      * The section can't be removed here because of the
1086      * memory_block->state_mutex.
1087      */
1088     zone = page_zone(pfn_to_page(pfn));
1089 
1090     if ((zone_idx(zone) > ZONE_NORMAL ||
1091         online_type == MMOP_ONLINE_MOVABLE) &&
1092         !can_online_high_movable(zone))
1093         return -EINVAL;
1094 
1095     if (online_type == MMOP_ONLINE_KERNEL) {
1096         if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift))
1097             return -EINVAL;
1098     } else if (online_type == MMOP_ONLINE_MOVABLE) {
1099         if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift))
1100             return -EINVAL;
1101     }
1102 
1103     zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1104     if (!zone)
1105         return -EINVAL;
1106 
1107     arg.start_pfn = pfn;
1108     arg.nr_pages = nr_pages;
1109     node_states_check_changes_online(nr_pages, zone, &arg);
1110 
1111     nid = zone_to_nid(zone);
1112 
1113     ret = memory_notify(MEM_GOING_ONLINE, &arg);
1114     ret = notifier_to_errno(ret);
1115     if (ret)
1116         goto failed_addition;
1117 
1118     /*
1119      * If this zone is not populated, then it is not in zonelist.
1120      * This means the page allocator ignores this zone.
1121      * So, zonelist must be updated after online.
1122      */
1123     mutex_lock(&zonelists_mutex);
1124     if (!populated_zone(zone)) {
1125         need_zonelists_rebuild = 1;
1126         build_all_zonelists(NULL, zone);
1127     }
1128 
1129     ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1130         online_pages_range);
1131     if (ret) {
1132         if (need_zonelists_rebuild)
1133             zone_pcp_reset(zone);
1134         mutex_unlock(&zonelists_mutex);
1135         goto failed_addition;
1136     }
1137 
1138     zone->present_pages += onlined_pages;
1139 
1140     pgdat_resize_lock(zone->zone_pgdat, &flags);
1141     zone->zone_pgdat->node_present_pages += onlined_pages;
1142     pgdat_resize_unlock(zone->zone_pgdat, &flags);
1143 
1144     if (onlined_pages) {
1145         node_states_set_node(nid, &arg);
1146         if (need_zonelists_rebuild)
1147             build_all_zonelists(NULL, NULL);
1148         else
1149             zone_pcp_update(zone);
1150     }
1151 
1152     mutex_unlock(&zonelists_mutex);
1153 
1154     init_per_zone_wmark_min();
1155 
1156     if (onlined_pages) {
1157         kswapd_run(nid);
1158         kcompactd_run(nid);
1159     }
1160 
1161     vm_total_pages = nr_free_pagecache_pages();
1162 
1163     writeback_set_ratelimit();
1164 
1165     if (onlined_pages)
1166         memory_notify(MEM_ONLINE, &arg);
1167     return 0;
1168 
1169 failed_addition:
1170     pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1171          (unsigned long long) pfn << PAGE_SHIFT,
1172          (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1173     memory_notify(MEM_CANCEL_ONLINE, &arg);
1174     return ret;
1175 }
1176 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1177 
1178 static void reset_node_present_pages(pg_data_t *pgdat)
1179 {
1180     struct zone *z;
1181 
1182     for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1183         z->present_pages = 0;
1184 
1185     pgdat->node_present_pages = 0;
1186 }
1187 
1188 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1189 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1190 {
1191     struct pglist_data *pgdat;
1192     unsigned long zones_size[MAX_NR_ZONES] = {0};
1193     unsigned long zholes_size[MAX_NR_ZONES] = {0};
1194     unsigned long start_pfn = PFN_DOWN(start);
1195 
1196     pgdat = NODE_DATA(nid);
1197     if (!pgdat) {
1198         pgdat = arch_alloc_nodedata(nid);
1199         if (!pgdat)
1200             return NULL;
1201 
1202         arch_refresh_nodedata(nid, pgdat);
1203     } else {
1204         /* Reset the nr_zones, order and classzone_idx before reuse */
1205         pgdat->nr_zones = 0;
1206         pgdat->kswapd_order = 0;
1207         pgdat->kswapd_classzone_idx = 0;
1208     }
1209 
1210     /* we can use NODE_DATA(nid) from here */
1211 
1212     /* init node's zones as empty zones, we don't have any present pages.*/
1213     free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1214     pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1215 
1216     /*
1217      * The node we allocated has no zone fallback lists. For avoiding
1218      * to access not-initialized zonelist, build here.
1219      */
1220     mutex_lock(&zonelists_mutex);
1221     build_all_zonelists(pgdat, NULL);
1222     mutex_unlock(&zonelists_mutex);
1223 
1224     /*
1225      * zone->managed_pages is set to an approximate value in
1226      * free_area_init_core(), which will cause
1227      * /sys/device/system/node/nodeX/meminfo has wrong data.
1228      * So reset it to 0 before any memory is onlined.
1229      */
1230     reset_node_managed_pages(pgdat);
1231 
1232     /*
1233      * When memory is hot-added, all the memory is in offline state. So
1234      * clear all zones' present_pages because they will be updated in
1235      * online_pages() and offline_pages().
1236      */
1237     reset_node_present_pages(pgdat);
1238 
1239     return pgdat;
1240 }
1241 
1242 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1243 {
1244     arch_refresh_nodedata(nid, NULL);
1245     free_percpu(pgdat->per_cpu_nodestats);
1246     arch_free_nodedata(pgdat);
1247     return;
1248 }
1249 
1250 
1251 /**
1252  * try_online_node - online a node if offlined
1253  *
1254  * called by cpu_up() to online a node without onlined memory.
1255  */
1256 int try_online_node(int nid)
1257 {
1258     pg_data_t   *pgdat;
1259     int ret;
1260 
1261     if (node_online(nid))
1262         return 0;
1263 
1264     mem_hotplug_begin();
1265     pgdat = hotadd_new_pgdat(nid, 0);
1266     if (!pgdat) {
1267         pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1268         ret = -ENOMEM;
1269         goto out;
1270     }
1271     node_set_online(nid);
1272     ret = register_one_node(nid);
1273     BUG_ON(ret);
1274 
1275     if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1276         mutex_lock(&zonelists_mutex);
1277         build_all_zonelists(NULL, NULL);
1278         mutex_unlock(&zonelists_mutex);
1279     }
1280 
1281 out:
1282     mem_hotplug_done();
1283     return ret;
1284 }
1285 
1286 static int check_hotplug_memory_range(u64 start, u64 size)
1287 {
1288     u64 start_pfn = PFN_DOWN(start);
1289     u64 nr_pages = size >> PAGE_SHIFT;
1290 
1291     /* Memory range must be aligned with section */
1292     if ((start_pfn & ~PAGE_SECTION_MASK) ||
1293         (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1294         pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1295                 (unsigned long long)start,
1296                 (unsigned long long)size);
1297         return -EINVAL;
1298     }
1299 
1300     return 0;
1301 }
1302 
1303 /*
1304  * If movable zone has already been setup, newly added memory should be check.
1305  * If its address is higher than movable zone, it should be added as movable.
1306  * Without this check, movable zone may overlap with other zone.
1307  */
1308 static int should_add_memory_movable(int nid, u64 start, u64 size)
1309 {
1310     unsigned long start_pfn = start >> PAGE_SHIFT;
1311     pg_data_t *pgdat = NODE_DATA(nid);
1312     struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1313 
1314     if (zone_is_empty(movable_zone))
1315         return 0;
1316 
1317     if (movable_zone->zone_start_pfn <= start_pfn)
1318         return 1;
1319 
1320     return 0;
1321 }
1322 
1323 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1324         bool for_device)
1325 {
1326 #ifdef CONFIG_ZONE_DEVICE
1327     if (for_device)
1328         return ZONE_DEVICE;
1329 #endif
1330     if (should_add_memory_movable(nid, start, size))
1331         return ZONE_MOVABLE;
1332 
1333     return zone_default;
1334 }
1335 
1336 static int online_memory_block(struct memory_block *mem, void *arg)
1337 {
1338     return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1339 }
1340 
1341 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1342 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1343 {
1344     u64 start, size;
1345     pg_data_t *pgdat = NULL;
1346     bool new_pgdat;
1347     bool new_node;
1348     int ret;
1349 
1350     start = res->start;
1351     size = resource_size(res);
1352 
1353     ret = check_hotplug_memory_range(start, size);
1354     if (ret)
1355         return ret;
1356 
1357     {   /* Stupid hack to suppress address-never-null warning */
1358         void *p = NODE_DATA(nid);
1359         new_pgdat = !p;
1360     }
1361 
1362     mem_hotplug_begin();
1363 
1364     /*
1365      * Add new range to memblock so that when hotadd_new_pgdat() is called
1366      * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1367      * this new range and calculate total pages correctly.  The range will
1368      * be removed at hot-remove time.
1369      */
1370     memblock_add_node(start, size, nid);
1371 
1372     new_node = !node_online(nid);
1373     if (new_node) {
1374         pgdat = hotadd_new_pgdat(nid, start);
1375         ret = -ENOMEM;
1376         if (!pgdat)
1377             goto error;
1378     }
1379 
1380     /* call arch's memory hotadd */
1381     ret = arch_add_memory(nid, start, size, false);
1382 
1383     if (ret < 0)
1384         goto error;
1385 
1386     /* we online node here. we can't roll back from here. */
1387     node_set_online(nid);
1388 
1389     if (new_node) {
1390         ret = register_one_node(nid);
1391         /*
1392          * If sysfs file of new node can't create, cpu on the node
1393          * can't be hot-added. There is no rollback way now.
1394          * So, check by BUG_ON() to catch it reluctantly..
1395          */
1396         BUG_ON(ret);
1397     }
1398 
1399     /* create new memmap entry */
1400     firmware_map_add_hotplug(start, start + size, "System RAM");
1401 
1402     /* online pages if requested */
1403     if (online)
1404         walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1405                   NULL, online_memory_block);
1406 
1407     goto out;
1408 
1409 error:
1410     /* rollback pgdat allocation and others */
1411     if (new_pgdat)
1412         rollback_node_hotadd(nid, pgdat);
1413     memblock_remove(start, size);
1414 
1415 out:
1416     mem_hotplug_done();
1417     return ret;
1418 }
1419 EXPORT_SYMBOL_GPL(add_memory_resource);
1420 
1421 int __ref add_memory(int nid, u64 start, u64 size)
1422 {
1423     struct resource *res;
1424     int ret;
1425 
1426     res = register_memory_resource(start, size);
1427     if (IS_ERR(res))
1428         return PTR_ERR(res);
1429 
1430     ret = add_memory_resource(nid, res, memhp_auto_online);
1431     if (ret < 0)
1432         release_memory_resource(res);
1433     return ret;
1434 }
1435 EXPORT_SYMBOL_GPL(add_memory);
1436 
1437 #ifdef CONFIG_MEMORY_HOTREMOVE
1438 /*
1439  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1440  * set and the size of the free page is given by page_order(). Using this,
1441  * the function determines if the pageblock contains only free pages.
1442  * Due to buddy contraints, a free page at least the size of a pageblock will
1443  * be located at the start of the pageblock
1444  */
1445 static inline int pageblock_free(struct page *page)
1446 {
1447     return PageBuddy(page) && page_order(page) >= pageblock_order;
1448 }
1449 
1450 /* Return the start of the next active pageblock after a given page */
1451 static struct page *next_active_pageblock(struct page *page)
1452 {
1453     /* Ensure the starting page is pageblock-aligned */
1454     BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1455 
1456     /* If the entire pageblock is free, move to the end of free page */
1457     if (pageblock_free(page)) {
1458         int order;
1459         /* be careful. we don't have locks, page_order can be changed.*/
1460         order = page_order(page);
1461         if ((order < MAX_ORDER) && (order >= pageblock_order))
1462             return page + (1 << order);
1463     }
1464 
1465     return page + pageblock_nr_pages;
1466 }
1467 
1468 /* Checks if this range of memory is likely to be hot-removable. */
1469 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1470 {
1471     struct page *page = pfn_to_page(start_pfn);
1472     struct page *end_page = page + nr_pages;
1473 
1474     /* Check the starting page of each pageblock within the range */
1475     for (; page < end_page; page = next_active_pageblock(page)) {
1476         if (!is_pageblock_removable_nolock(page))
1477             return false;
1478         cond_resched();
1479     }
1480 
1481     /* All pageblocks in the memory block are likely to be hot-removable */
1482     return true;
1483 }
1484 
1485 /*
1486  * Confirm all pages in a range [start, end) belong to the same zone.
1487  * When true, return its valid [start, end).
1488  */
1489 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1490              unsigned long *valid_start, unsigned long *valid_end)
1491 {
1492     unsigned long pfn, sec_end_pfn;
1493     unsigned long start, end;
1494     struct zone *zone = NULL;
1495     struct page *page;
1496     int i;
1497     for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1498          pfn < end_pfn;
1499          pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1500         /* Make sure the memory section is present first */
1501         if (!present_section_nr(pfn_to_section_nr(pfn)))
1502             continue;
1503         for (; pfn < sec_end_pfn && pfn < end_pfn;
1504              pfn += MAX_ORDER_NR_PAGES) {
1505             i = 0;
1506             /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1507             while ((i < MAX_ORDER_NR_PAGES) &&
1508                 !pfn_valid_within(pfn + i))
1509                 i++;
1510             if (i == MAX_ORDER_NR_PAGES)
1511                 continue;
1512             page = pfn_to_page(pfn + i);
1513             if (zone && page_zone(page) != zone)
1514                 return 0;
1515             if (!zone)
1516                 start = pfn + i;
1517             zone = page_zone(page);
1518             end = pfn + MAX_ORDER_NR_PAGES;
1519         }
1520     }
1521 
1522     if (zone) {
1523         *valid_start = start;
1524         *valid_end = end;
1525         return 1;
1526     } else {
1527         return 0;
1528     }
1529 }
1530 
1531 /*
1532  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1533  * and hugepages). We scan pfn because it's much easier than scanning over
1534  * linked list. This function returns the pfn of the first found movable
1535  * page if it's found, otherwise 0.
1536  */
1537 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1538 {
1539     unsigned long pfn;
1540     struct page *page;
1541     for (pfn = start; pfn < end; pfn++) {
1542         if (pfn_valid(pfn)) {
1543             page = pfn_to_page(pfn);
1544             if (PageLRU(page))
1545                 return pfn;
1546             if (PageHuge(page)) {
1547                 if (page_huge_active(page))
1548                     return pfn;
1549                 else
1550                     pfn = round_up(pfn + 1,
1551                         1 << compound_order(page)) - 1;
1552             }
1553         }
1554     }
1555     return 0;
1556 }
1557 
1558 static struct page *new_node_page(struct page *page, unsigned long private,
1559         int **result)
1560 {
1561     gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1562     int nid = page_to_nid(page);
1563     nodemask_t nmask = node_states[N_MEMORY];
1564     struct page *new_page = NULL;
1565 
1566     /*
1567      * TODO: allocate a destination hugepage from a nearest neighbor node,
1568      * accordance with memory policy of the user process if possible. For
1569      * now as a simple work-around, we use the next node for destination.
1570      */
1571     if (PageHuge(page))
1572         return alloc_huge_page_node(page_hstate(compound_head(page)),
1573                     next_node_in(nid, nmask));
1574 
1575     node_clear(nid, nmask);
1576 
1577     if (PageHighMem(page)
1578         || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1579         gfp_mask |= __GFP_HIGHMEM;
1580 
1581     if (!nodes_empty(nmask))
1582         new_page = __alloc_pages_nodemask(gfp_mask, 0,
1583                     node_zonelist(nid, gfp_mask), &nmask);
1584     if (!new_page)
1585         new_page = __alloc_pages(gfp_mask, 0,
1586                     node_zonelist(nid, gfp_mask));
1587 
1588     return new_page;
1589 }
1590 
1591 #define NR_OFFLINE_AT_ONCE_PAGES    (256)
1592 static int
1593 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1594 {
1595     unsigned long pfn;
1596     struct page *page;
1597     int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1598     int not_managed = 0;
1599     int ret = 0;
1600     LIST_HEAD(source);
1601 
1602     for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1603         if (!pfn_valid(pfn))
1604             continue;
1605         page = pfn_to_page(pfn);
1606 
1607         if (PageHuge(page)) {
1608             struct page *head = compound_head(page);
1609             pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1610             if (compound_order(head) > PFN_SECTION_SHIFT) {
1611                 ret = -EBUSY;
1612                 break;
1613             }
1614             if (isolate_huge_page(page, &source))
1615                 move_pages -= 1 << compound_order(head);
1616             continue;
1617         }
1618 
1619         if (!get_page_unless_zero(page))
1620             continue;
1621         /*
1622          * We can skip free pages. And we can only deal with pages on
1623          * LRU.
1624          */
1625         ret = isolate_lru_page(page);
1626         if (!ret) { /* Success */
1627             put_page(page);
1628             list_add_tail(&page->lru, &source);
1629             move_pages--;
1630             inc_node_page_state(page, NR_ISOLATED_ANON +
1631                         page_is_file_cache(page));
1632 
1633         } else {
1634 #ifdef CONFIG_DEBUG_VM
1635             pr_alert("removing pfn %lx from LRU failed\n", pfn);
1636             dump_page(page, "failed to remove from LRU");
1637 #endif
1638             put_page(page);
1639             /* Because we don't have big zone->lock. we should
1640                check this again here. */
1641             if (page_count(page)) {
1642                 not_managed++;
1643                 ret = -EBUSY;
1644                 break;
1645             }
1646         }
1647     }
1648     if (!list_empty(&source)) {
1649         if (not_managed) {
1650             putback_movable_pages(&source);
1651             goto out;
1652         }
1653 
1654         /* Allocate a new page from the nearest neighbor node */
1655         ret = migrate_pages(&source, new_node_page, NULL, 0,
1656                     MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1657         if (ret)
1658             putback_movable_pages(&source);
1659     }
1660 out:
1661     return ret;
1662 }
1663 
1664 /*
1665  * remove from free_area[] and mark all as Reserved.
1666  */
1667 static int
1668 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1669             void *data)
1670 {
1671     __offline_isolated_pages(start, start + nr_pages);
1672     return 0;
1673 }
1674 
1675 static void
1676 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1677 {
1678     walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1679                 offline_isolated_pages_cb);
1680 }
1681 
1682 /*
1683  * Check all pages in range, recoreded as memory resource, are isolated.
1684  */
1685 static int
1686 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1687             void *data)
1688 {
1689     int ret;
1690     long offlined = *(long *)data;
1691     ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1692     offlined = nr_pages;
1693     if (!ret)
1694         *(long *)data += offlined;
1695     return ret;
1696 }
1697 
1698 static long
1699 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1700 {
1701     long offlined = 0;
1702     int ret;
1703 
1704     ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1705             check_pages_isolated_cb);
1706     if (ret < 0)
1707         offlined = (long)ret;
1708     return offlined;
1709 }
1710 
1711 #ifdef CONFIG_MOVABLE_NODE
1712 /*
1713  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1714  * normal memory.
1715  */
1716 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1717 {
1718     return true;
1719 }
1720 #else /* CONFIG_MOVABLE_NODE */
1721 /* ensure the node has NORMAL memory if it is still online */
1722 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1723 {
1724     struct pglist_data *pgdat = zone->zone_pgdat;
1725     unsigned long present_pages = 0;
1726     enum zone_type zt;
1727 
1728     for (zt = 0; zt <= ZONE_NORMAL; zt++)
1729         present_pages += pgdat->node_zones[zt].present_pages;
1730 
1731     if (present_pages > nr_pages)
1732         return true;
1733 
1734     present_pages = 0;
1735     for (; zt <= ZONE_MOVABLE; zt++)
1736         present_pages += pgdat->node_zones[zt].present_pages;
1737 
1738     /*
1739      * we can't offline the last normal memory until all
1740      * higher memory is offlined.
1741      */
1742     return present_pages == 0;
1743 }
1744 #endif /* CONFIG_MOVABLE_NODE */
1745 
1746 static int __init cmdline_parse_movable_node(char *p)
1747 {
1748 #ifdef CONFIG_MOVABLE_NODE
1749     movable_node_enabled = true;
1750 #else
1751     pr_warn("movable_node option not supported\n");
1752 #endif
1753     return 0;
1754 }
1755 early_param("movable_node", cmdline_parse_movable_node);
1756 
1757 /* check which state of node_states will be changed when offline memory */
1758 static void node_states_check_changes_offline(unsigned long nr_pages,
1759         struct zone *zone, struct memory_notify *arg)
1760 {
1761     struct pglist_data *pgdat = zone->zone_pgdat;
1762     unsigned long present_pages = 0;
1763     enum zone_type zt, zone_last = ZONE_NORMAL;
1764 
1765     /*
1766      * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1767      * contains nodes which have zones of 0...ZONE_NORMAL,
1768      * set zone_last to ZONE_NORMAL.
1769      *
1770      * If we don't have HIGHMEM nor movable node,
1771      * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1772      * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1773      */
1774     if (N_MEMORY == N_NORMAL_MEMORY)
1775         zone_last = ZONE_MOVABLE;
1776 
1777     /*
1778      * check whether node_states[N_NORMAL_MEMORY] will be changed.
1779      * If the memory to be offline is in a zone of 0...zone_last,
1780      * and it is the last present memory, 0...zone_last will
1781      * become empty after offline , thus we can determind we will
1782      * need to clear the node from node_states[N_NORMAL_MEMORY].
1783      */
1784     for (zt = 0; zt <= zone_last; zt++)
1785         present_pages += pgdat->node_zones[zt].present_pages;
1786     if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1787         arg->status_change_nid_normal = zone_to_nid(zone);
1788     else
1789         arg->status_change_nid_normal = -1;
1790 
1791 #ifdef CONFIG_HIGHMEM
1792     /*
1793      * If we have movable node, node_states[N_HIGH_MEMORY]
1794      * contains nodes which have zones of 0...ZONE_HIGHMEM,
1795      * set zone_last to ZONE_HIGHMEM.
1796      *
1797      * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1798      * contains nodes which have zones of 0...ZONE_MOVABLE,
1799      * set zone_last to ZONE_MOVABLE.
1800      */
1801     zone_last = ZONE_HIGHMEM;
1802     if (N_MEMORY == N_HIGH_MEMORY)
1803         zone_last = ZONE_MOVABLE;
1804 
1805     for (; zt <= zone_last; zt++)
1806         present_pages += pgdat->node_zones[zt].present_pages;
1807     if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1808         arg->status_change_nid_high = zone_to_nid(zone);
1809     else
1810         arg->status_change_nid_high = -1;
1811 #else
1812     arg->status_change_nid_high = arg->status_change_nid_normal;
1813 #endif
1814 
1815     /*
1816      * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1817      */
1818     zone_last = ZONE_MOVABLE;
1819 
1820     /*
1821      * check whether node_states[N_HIGH_MEMORY] will be changed
1822      * If we try to offline the last present @nr_pages from the node,
1823      * we can determind we will need to clear the node from
1824      * node_states[N_HIGH_MEMORY].
1825      */
1826     for (; zt <= zone_last; zt++)
1827         present_pages += pgdat->node_zones[zt].present_pages;
1828     if (nr_pages >= present_pages)
1829         arg->status_change_nid = zone_to_nid(zone);
1830     else
1831         arg->status_change_nid = -1;
1832 }
1833 
1834 static void node_states_clear_node(int node, struct memory_notify *arg)
1835 {
1836     if (arg->status_change_nid_normal >= 0)
1837         node_clear_state(node, N_NORMAL_MEMORY);
1838 
1839     if ((N_MEMORY != N_NORMAL_MEMORY) &&
1840         (arg->status_change_nid_high >= 0))
1841         node_clear_state(node, N_HIGH_MEMORY);
1842 
1843     if ((N_MEMORY != N_HIGH_MEMORY) &&
1844         (arg->status_change_nid >= 0))
1845         node_clear_state(node, N_MEMORY);
1846 }
1847 
1848 static int __ref __offline_pages(unsigned long start_pfn,
1849           unsigned long end_pfn, unsigned long timeout)
1850 {
1851     unsigned long pfn, nr_pages, expire;
1852     long offlined_pages;
1853     int ret, drain, retry_max, node;
1854     unsigned long flags;
1855     unsigned long valid_start, valid_end;
1856     struct zone *zone;
1857     struct memory_notify arg;
1858 
1859     /* at least, alignment against pageblock is necessary */
1860     if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1861         return -EINVAL;
1862     if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1863         return -EINVAL;
1864     /* This makes hotplug much easier...and readable.
1865        we assume this for now. .*/
1866     if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1867         return -EINVAL;
1868 
1869     zone = page_zone(pfn_to_page(valid_start));
1870     node = zone_to_nid(zone);
1871     nr_pages = end_pfn - start_pfn;
1872 
1873     if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1874         return -EINVAL;
1875 
1876     /* set above range as isolated */
1877     ret = start_isolate_page_range(start_pfn, end_pfn,
1878                        MIGRATE_MOVABLE, true);
1879     if (ret)
1880         return ret;
1881 
1882     arg.start_pfn = start_pfn;
1883     arg.nr_pages = nr_pages;
1884     node_states_check_changes_offline(nr_pages, zone, &arg);
1885 
1886     ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1887     ret = notifier_to_errno(ret);
1888     if (ret)
1889         goto failed_removal;
1890 
1891     pfn = start_pfn;
1892     expire = jiffies + timeout;
1893     drain = 0;
1894     retry_max = 5;
1895 repeat:
1896     /* start memory hot removal */
1897     ret = -EAGAIN;
1898     if (time_after(jiffies, expire))
1899         goto failed_removal;
1900     ret = -EINTR;
1901     if (signal_pending(current))
1902         goto failed_removal;
1903     ret = 0;
1904     if (drain) {
1905         lru_add_drain_all();
1906         cond_resched();
1907         drain_all_pages(zone);
1908     }
1909 
1910     pfn = scan_movable_pages(start_pfn, end_pfn);
1911     if (pfn) { /* We have movable pages */
1912         ret = do_migrate_range(pfn, end_pfn);
1913         if (!ret) {
1914             drain = 1;
1915             goto repeat;
1916         } else {
1917             if (ret < 0)
1918                 if (--retry_max == 0)
1919                     goto failed_removal;
1920             yield();
1921             drain = 1;
1922             goto repeat;
1923         }
1924     }
1925     /* drain all zone's lru pagevec, this is asynchronous... */
1926     lru_add_drain_all();
1927     yield();
1928     /* drain pcp pages, this is synchronous. */
1929     drain_all_pages(zone);
1930     /*
1931      * dissolve free hugepages in the memory block before doing offlining
1932      * actually in order to make hugetlbfs's object counting consistent.
1933      */
1934     ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1935     if (ret)
1936         goto failed_removal;
1937     /* check again */
1938     offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1939     if (offlined_pages < 0) {
1940         ret = -EBUSY;
1941         goto failed_removal;
1942     }
1943     pr_info("Offlined Pages %ld\n", offlined_pages);
1944     /* Ok, all of our target is isolated.
1945        We cannot do rollback at this point. */
1946     offline_isolated_pages(start_pfn, end_pfn);
1947     /* reset pagetype flags and makes migrate type to be MOVABLE */
1948     undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1949     /* removal success */
1950     adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1951     zone->present_pages -= offlined_pages;
1952 
1953     pgdat_resize_lock(zone->zone_pgdat, &flags);
1954     zone->zone_pgdat->node_present_pages -= offlined_pages;
1955     pgdat_resize_unlock(zone->zone_pgdat, &flags);
1956 
1957     init_per_zone_wmark_min();
1958 
1959     if (!populated_zone(zone)) {
1960         zone_pcp_reset(zone);
1961         mutex_lock(&zonelists_mutex);
1962         build_all_zonelists(NULL, NULL);
1963         mutex_unlock(&zonelists_mutex);
1964     } else
1965         zone_pcp_update(zone);
1966 
1967     node_states_clear_node(node, &arg);
1968     if (arg.status_change_nid >= 0) {
1969         kswapd_stop(node);
1970         kcompactd_stop(node);
1971     }
1972 
1973     vm_total_pages = nr_free_pagecache_pages();
1974     writeback_set_ratelimit();
1975 
1976     memory_notify(MEM_OFFLINE, &arg);
1977     return 0;
1978 
1979 failed_removal:
1980     pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1981          (unsigned long long) start_pfn << PAGE_SHIFT,
1982          ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1983     memory_notify(MEM_CANCEL_OFFLINE, &arg);
1984     /* pushback to free area */
1985     undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1986     return ret;
1987 }
1988 
1989 /* Must be protected by mem_hotplug_begin() */
1990 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1991 {
1992     return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1993 }
1994 #endif /* CONFIG_MEMORY_HOTREMOVE */
1995 
1996 /**
1997  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1998  * @start_pfn: start pfn of the memory range
1999  * @end_pfn: end pfn of the memory range
2000  * @arg: argument passed to func
2001  * @func: callback for each memory section walked
2002  *
2003  * This function walks through all present mem sections in range
2004  * [start_pfn, end_pfn) and call func on each mem section.
2005  *
2006  * Returns the return value of func.
2007  */
2008 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2009         void *arg, int (*func)(struct memory_block *, void *))
2010 {
2011     struct memory_block *mem = NULL;
2012     struct mem_section *section;
2013     unsigned long pfn, section_nr;
2014     int ret;
2015 
2016     for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2017         section_nr = pfn_to_section_nr(pfn);
2018         if (!present_section_nr(section_nr))
2019             continue;
2020 
2021         section = __nr_to_section(section_nr);
2022         /* same memblock? */
2023         if (mem)
2024             if ((section_nr >= mem->start_section_nr) &&
2025                 (section_nr <= mem->end_section_nr))
2026                 continue;
2027 
2028         mem = find_memory_block_hinted(section, mem);
2029         if (!mem)
2030             continue;
2031 
2032         ret = func(mem, arg);
2033         if (ret) {
2034             kobject_put(&mem->dev.kobj);
2035             return ret;
2036         }
2037     }
2038 
2039     if (mem)
2040         kobject_put(&mem->dev.kobj);
2041 
2042     return 0;
2043 }
2044 
2045 #ifdef CONFIG_MEMORY_HOTREMOVE
2046 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2047 {
2048     int ret = !is_memblock_offlined(mem);
2049 
2050     if (unlikely(ret)) {
2051         phys_addr_t beginpa, endpa;
2052 
2053         beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2054         endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2055         pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2056             &beginpa, &endpa);
2057     }
2058 
2059     return ret;
2060 }
2061 
2062 static int check_cpu_on_node(pg_data_t *pgdat)
2063 {
2064     int cpu;
2065 
2066     for_each_present_cpu(cpu) {
2067         if (cpu_to_node(cpu) == pgdat->node_id)
2068             /*
2069              * the cpu on this node isn't removed, and we can't
2070              * offline this node.
2071              */
2072             return -EBUSY;
2073     }
2074 
2075     return 0;
2076 }
2077 
2078 static void unmap_cpu_on_node(pg_data_t *pgdat)
2079 {
2080 #ifdef CONFIG_ACPI_NUMA
2081     int cpu;
2082 
2083     for_each_possible_cpu(cpu)
2084         if (cpu_to_node(cpu) == pgdat->node_id)
2085             numa_clear_node(cpu);
2086 #endif
2087 }
2088 
2089 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2090 {
2091     int ret;
2092 
2093     ret = check_cpu_on_node(pgdat);
2094     if (ret)
2095         return ret;
2096 
2097     /*
2098      * the node will be offlined when we come here, so we can clear
2099      * the cpu_to_node() now.
2100      */
2101 
2102     unmap_cpu_on_node(pgdat);
2103     return 0;
2104 }
2105 
2106 /**
2107  * try_offline_node
2108  *
2109  * Offline a node if all memory sections and cpus of the node are removed.
2110  *
2111  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2112  * and online/offline operations before this call.
2113  */
2114 void try_offline_node(int nid)
2115 {
2116     pg_data_t *pgdat = NODE_DATA(nid);
2117     unsigned long start_pfn = pgdat->node_start_pfn;
2118     unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2119     unsigned long pfn;
2120 
2121     for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2122         unsigned long section_nr = pfn_to_section_nr(pfn);
2123 
2124         if (!present_section_nr(section_nr))
2125             continue;
2126 
2127         if (pfn_to_nid(pfn) != nid)
2128             continue;
2129 
2130         /*
2131          * some memory sections of this node are not removed, and we
2132          * can't offline node now.
2133          */
2134         return;
2135     }
2136 
2137     if (check_and_unmap_cpu_on_node(pgdat))
2138         return;
2139 
2140     /*
2141      * all memory/cpu of this node are removed, we can offline this
2142      * node now.
2143      */
2144     node_set_offline(nid);
2145     unregister_one_node(nid);
2146 }
2147 EXPORT_SYMBOL(try_offline_node);
2148 
2149 /**
2150  * remove_memory
2151  *
2152  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2153  * and online/offline operations before this call, as required by
2154  * try_offline_node().
2155  */
2156 void __ref remove_memory(int nid, u64 start, u64 size)
2157 {
2158     int ret;
2159 
2160     BUG_ON(check_hotplug_memory_range(start, size));
2161 
2162     mem_hotplug_begin();
2163 
2164     /*
2165      * All memory blocks must be offlined before removing memory.  Check
2166      * whether all memory blocks in question are offline and trigger a BUG()
2167      * if this is not the case.
2168      */
2169     ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2170                 check_memblock_offlined_cb);
2171     if (ret)
2172         BUG();
2173 
2174     /* remove memmap entry */
2175     firmware_map_remove(start, start + size, "System RAM");
2176     memblock_free(start, size);
2177     memblock_remove(start, size);
2178 
2179     arch_remove_memory(start, size);
2180 
2181     try_offline_node(nid);
2182 
2183     mem_hotplug_done();
2184 }
2185 EXPORT_SYMBOL_GPL(remove_memory);
2186 #endif /* CONFIG_MEMORY_HOTREMOVE */