memory_hotplug.c

Version: - FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
    1 /*
    2  *  linux/mm/memory_hotplug.c
    3  *
    4  *  Copyright (C)
    5  */
    6 
    7 #include <linux/stddef.h>
    8 #include <linux/mm.h>
    9 #include <linux/swap.h>
   10 #include <linux/interrupt.h>
   11 #include <linux/pagemap.h>
   12 #include <linux/bootmem.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/memory_hotplug.h>
   22 #include <linux/highmem.h>
   23 #include <linux/vmalloc.h>
   24 #include <linux/ioport.h>
   25 #include <linux/delay.h>
   26 #include <linux/migrate.h>
   27 #include <linux/page-isolation.h>
   28 #include <linux/pfn.h>
   29 #include <linux/suspend.h>
   30 #include <linux/mm_inline.h>
   31 #include <linux/firmware-map.h>
   32 
   33 #include <asm/tlbflush.h>
   34 
   35 #include "internal.h"
   36 
   37 /*
   38  * online_page_callback contains pointer to current page onlining function.
   39  * Initially it is generic_online_page(). If it is required it could be
   40  * changed by calling set_online_page_callback() for callback registration
   41  * and restore_online_page_callback() for generic callback restore.
   42  */
   43 
   44 static void generic_online_page(struct page *page);
   45 
   46 static online_page_callback_t online_page_callback = generic_online_page;
   47 
   48 DEFINE_MUTEX(mem_hotplug_mutex);
   49 
   50 void lock_memory_hotplug(void)
   51 {
   52         mutex_lock(&mem_hotplug_mutex);
   53 
   54         /* for exclusive hibernation if CONFIG_HIBERNATION=y */
   55         lock_system_sleep();
   56 }
   57 
   58 void unlock_memory_hotplug(void)
   59 {
   60         unlock_system_sleep();
   61         mutex_unlock(&mem_hotplug_mutex);
   62 }
   63 
   64 
   65 /* add this memory to iomem resource */
   66 static struct resource *register_memory_resource(u64 start, u64 size)
   67 {
   68         struct resource *res;
   69         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
   70         BUG_ON(!res);
   71 
   72         res->name = "System RAM";
   73         res->start = start;
   74         res->end = start + size - 1;
   75         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
   76         if (request_resource(&iomem_resource, res) < 0) {
   77                 printk("System RAM resource %pR cannot be added\n", res);
   78                 kfree(res);
   79                 res = NULL;
   80         }
   81         return res;
   82 }
   83 
   84 static void release_memory_resource(struct resource *res)
   85 {
   86         if (!res)
   87                 return;
   88         release_resource(res);
   89         kfree(res);
   90         return;
   91 }
   92 
   93 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
   94 #ifndef CONFIG_SPARSEMEM_VMEMMAP
   95 static void get_page_bootmem(unsigned long info,  struct page *page,
   96                              unsigned long type)
   97 {
   98         page->lru.next = (struct list_head *) type;
   99         SetPagePrivate(page);
  100         set_page_private(page, info);
  101         atomic_inc(&page->_count);
  102 }
  103 
  104 /* reference to __meminit __free_pages_bootmem is valid
  105  * so use __ref to tell modpost not to generate a warning */
  106 void __ref put_page_bootmem(struct page *page)
  107 {
  108         unsigned long type;
  109         static DEFINE_MUTEX(ppb_lock);
  110 
  111         type = (unsigned long) page->lru.next;
  112         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
  113                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
  114 
  115         if (atomic_dec_return(&page->_count) == 1) {
  116                 ClearPagePrivate(page);
  117                 set_page_private(page, 0);
  118                 INIT_LIST_HEAD(&page->lru);
  119 
  120                 /*
  121                  * Please refer to comment for __free_pages_bootmem()
  122                  * for why we serialize here.
  123                  */
  124                 mutex_lock(&ppb_lock);
  125                 __free_pages_bootmem(page, 0);
  126                 mutex_unlock(&ppb_lock);
  127         }
  128 
  129 }
  130 
  131 static void register_page_bootmem_info_section(unsigned long start_pfn)
  132 {
  133         unsigned long *usemap, mapsize, section_nr, i;
  134         struct mem_section *ms;
  135         struct page *page, *memmap;
  136 
  137         section_nr = pfn_to_section_nr(start_pfn);
  138         ms = __nr_to_section(section_nr);
  139 
  140         /* Get section's memmap address */
  141         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
  142 
  143         /*
  144          * Get page for the memmap's phys address
  145          * XXX: need more consideration for sparse_vmemmap...
  146          */
  147         page = virt_to_page(memmap);
  148         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
  149         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
  150 
  151         /* remember memmap's page */
  152         for (i = 0; i < mapsize; i++, page++)
  153                 get_page_bootmem(section_nr, page, SECTION_INFO);
  154 
  155         usemap = __nr_to_section(section_nr)->pageblock_flags;
  156         page = virt_to_page(usemap);
  157 
  158         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
  159 
  160         for (i = 0; i < mapsize; i++, page++)
  161                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
  162 
  163 }
  164 
  165 void register_page_bootmem_info_node(struct pglist_data *pgdat)
  166 {
  167         unsigned long i, pfn, end_pfn, nr_pages;
  168         int node = pgdat->node_id;
  169         struct page *page;
  170         struct zone *zone;
  171 
  172         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
  173         page = virt_to_page(pgdat);
  174 
  175         for (i = 0; i < nr_pages; i++, page++)
  176                 get_page_bootmem(node, page, NODE_INFO);
  177 
  178         zone = &pgdat->node_zones[0];
  179         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
  180                 if (zone->wait_table) {
  181                         nr_pages = zone->wait_table_hash_nr_entries
  182                                 * sizeof(wait_queue_head_t);
  183                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
  184                         page = virt_to_page(zone->wait_table);
  185 
  186                         for (i = 0; i < nr_pages; i++, page++)
  187                                 get_page_bootmem(node, page, NODE_INFO);
  188                 }
  189         }
  190 
  191         pfn = pgdat->node_start_pfn;
  192         end_pfn = pfn + pgdat->node_spanned_pages;
  193 
  194         /* register_section info */
  195         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
  196                 /*
  197                  * Some platforms can assign the same pfn to multiple nodes - on
  198                  * node0 as well as nodeN.  To avoid registering a pfn against
  199                  * multiple nodes we check that this pfn does not already
  200                  * reside in some other node.
  201                  */
  202                 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
  203                         register_page_bootmem_info_section(pfn);
  204         }
  205 }
  206 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
  207 
  208 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
  209                            unsigned long end_pfn)
  210 {
  211         unsigned long old_zone_end_pfn;
  212 
  213         zone_span_writelock(zone);
  214 
  215         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
  216         if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn)
  217                 zone->zone_start_pfn = start_pfn;
  218 
  219         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
  220                                 zone->zone_start_pfn;
  221 
  222         zone_span_writeunlock(zone);
  223 }
  224 
  225 static void resize_zone(struct zone *zone, unsigned long start_pfn,
  226                 unsigned long end_pfn)
  227 {
  228         zone_span_writelock(zone);
  229 
  230         if (end_pfn - start_pfn) {
  231                 zone->zone_start_pfn = start_pfn;
  232                 zone->spanned_pages = end_pfn - start_pfn;
  233         } else {
  234                 /*
  235                  * make it consist as free_area_init_core(),
  236                  * if spanned_pages = 0, then keep start_pfn = 0
  237                  */
  238                 zone->zone_start_pfn = 0;
  239                 zone->spanned_pages = 0;
  240         }
  241 
  242         zone_span_writeunlock(zone);
  243 }
  244 
  245 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
  246                 unsigned long end_pfn)
  247 {
  248         enum zone_type zid = zone_idx(zone);
  249         int nid = zone->zone_pgdat->node_id;
  250         unsigned long pfn;
  251 
  252         for (pfn = start_pfn; pfn < end_pfn; pfn++)
  253                 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
  254 }
  255 
  256 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
  257                 unsigned long start_pfn, unsigned long end_pfn)
  258 {
  259         int ret;
  260         unsigned long flags;
  261         unsigned long z1_start_pfn;
  262 
  263         if (!z1->wait_table) {
  264                 ret = init_currently_empty_zone(z1, start_pfn,
  265                         end_pfn - start_pfn, MEMMAP_HOTPLUG);
  266                 if (ret)
  267                         return ret;
  268         }
  269 
  270         pgdat_resize_lock(z1->zone_pgdat, &flags);
  271 
  272         /* can't move pfns which are higher than @z2 */
  273         if (end_pfn > z2->zone_start_pfn + z2->spanned_pages)
  274                 goto out_fail;
  275         /* the move out part mast at the left most of @z2 */
  276         if (start_pfn > z2->zone_start_pfn)
  277                 goto out_fail;
  278         /* must included/overlap */
  279         if (end_pfn <= z2->zone_start_pfn)
  280                 goto out_fail;
  281 
  282         /* use start_pfn for z1's start_pfn if z1 is empty */
  283         if (z1->spanned_pages)
  284                 z1_start_pfn = z1->zone_start_pfn;
  285         else
  286                 z1_start_pfn = start_pfn;
  287 
  288         resize_zone(z1, z1_start_pfn, end_pfn);
  289         resize_zone(z2, end_pfn, z2->zone_start_pfn + z2->spanned_pages);
  290 
  291         pgdat_resize_unlock(z1->zone_pgdat, &flags);
  292 
  293         fix_zone_id(z1, start_pfn, end_pfn);
  294 
  295         return 0;
  296 out_fail:
  297         pgdat_resize_unlock(z1->zone_pgdat, &flags);
  298         return -1;
  299 }
  300 
  301 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
  302                 unsigned long start_pfn, unsigned long end_pfn)
  303 {
  304         int ret;
  305         unsigned long flags;
  306         unsigned long z2_end_pfn;
  307 
  308         if (!z2->wait_table) {
  309                 ret = init_currently_empty_zone(z2, start_pfn,
  310                         end_pfn - start_pfn, MEMMAP_HOTPLUG);
  311                 if (ret)
  312                         return ret;
  313         }
  314 
  315         pgdat_resize_lock(z1->zone_pgdat, &flags);
  316 
  317         /* can't move pfns which are lower than @z1 */
  318         if (z1->zone_start_pfn > start_pfn)
  319                 goto out_fail;
  320         /* the move out part mast at the right most of @z1 */
  321         if (z1->zone_start_pfn + z1->spanned_pages >  end_pfn)
  322                 goto out_fail;
  323         /* must included/overlap */
  324         if (start_pfn >= z1->zone_start_pfn + z1->spanned_pages)
  325                 goto out_fail;
  326 
  327         /* use end_pfn for z2's end_pfn if z2 is empty */
  328         if (z2->spanned_pages)
  329                 z2_end_pfn = z2->zone_start_pfn + z2->spanned_pages;
  330         else
  331                 z2_end_pfn = end_pfn;
  332 
  333         resize_zone(z1, z1->zone_start_pfn, start_pfn);
  334         resize_zone(z2, start_pfn, z2_end_pfn);
  335 
  336         pgdat_resize_unlock(z1->zone_pgdat, &flags);
  337 
  338         fix_zone_id(z2, start_pfn, end_pfn);
  339 
  340         return 0;
  341 out_fail:
  342         pgdat_resize_unlock(z1->zone_pgdat, &flags);
  343         return -1;
  344 }
  345 
  346 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
  347                             unsigned long end_pfn)
  348 {
  349         unsigned long old_pgdat_end_pfn =
  350                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
  351 
  352         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
  353                 pgdat->node_start_pfn = start_pfn;
  354 
  355         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
  356                                         pgdat->node_start_pfn;
  357 }
  358 
  359 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
  360 {
  361         struct pglist_data *pgdat = zone->zone_pgdat;
  362         int nr_pages = PAGES_PER_SECTION;
  363         int nid = pgdat->node_id;
  364         int zone_type;
  365         unsigned long flags;
  366 
  367         zone_type = zone - pgdat->node_zones;
  368         if (!zone->wait_table) {
  369                 int ret;
  370 
  371                 ret = init_currently_empty_zone(zone, phys_start_pfn,
  372                                                 nr_pages, MEMMAP_HOTPLUG);
  373                 if (ret)
  374                         return ret;
  375         }
  376         pgdat_resize_lock(zone->zone_pgdat, &flags);
  377         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
  378         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
  379                         phys_start_pfn + nr_pages);
  380         pgdat_resize_unlock(zone->zone_pgdat, &flags);
  381         memmap_init_zone(nr_pages, nid, zone_type,
  382                          phys_start_pfn, MEMMAP_HOTPLUG);
  383         return 0;
  384 }
  385 
  386 static int __meminit __add_section(int nid, struct zone *zone,
  387                                         unsigned long phys_start_pfn)
  388 {
  389         int nr_pages = PAGES_PER_SECTION;
  390         int ret;
  391 
  392         if (pfn_valid(phys_start_pfn))
  393                 return -EEXIST;
  394 
  395         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
  396 
  397         if (ret < 0)
  398                 return ret;
  399 
  400         ret = __add_zone(zone, phys_start_pfn);
  401 
  402         if (ret < 0)
  403                 return ret;
  404 
  405         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
  406 }
  407 
  408 #ifdef CONFIG_SPARSEMEM_VMEMMAP
  409 static int __remove_section(struct zone *zone, struct mem_section *ms)
  410 {
  411         /*
  412          * XXX: Freeing memmap with vmemmap is not implement yet.
  413          *      This should be removed later.
  414          */
  415         return -EBUSY;
  416 }
  417 #else
  418 static int __remove_section(struct zone *zone, struct mem_section *ms)
  419 {
  420         unsigned long flags;
  421         struct pglist_data *pgdat = zone->zone_pgdat;
  422         int ret = -EINVAL;
  423 
  424         if (!valid_section(ms))
  425                 return ret;
  426 
  427         ret = unregister_memory_section(ms);
  428         if (ret)
  429                 return ret;
  430 
  431         pgdat_resize_lock(pgdat, &flags);
  432         sparse_remove_one_section(zone, ms);
  433         pgdat_resize_unlock(pgdat, &flags);
  434         return 0;
  435 }
  436 #endif
  437 
  438 /*
  439  * Reasonably generic function for adding memory.  It is
  440  * expected that archs that support memory hotplug will
  441  * call this function after deciding the zone to which to
  442  * add the new pages.
  443  */
  444 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
  445                         unsigned long nr_pages)
  446 {
  447         unsigned long i;
  448         int err = 0;
  449         int start_sec, end_sec;
  450         /* during initialize mem_map, align hot-added range to section */
  451         start_sec = pfn_to_section_nr(phys_start_pfn);
  452         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
  453 
  454         for (i = start_sec; i <= end_sec; i++) {
  455                 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
  456 
  457                 /*
  458                  * EEXIST is finally dealt with by ioresource collision
  459                  * check. see add_memory() => register_memory_resource()
  460                  * Warning will be printed if there is collision.
  461                  */
  462                 if (err && (err != -EEXIST))
  463                         break;
  464                 err = 0;
  465         }
  466 
  467         return err;
  468 }
  469 EXPORT_SYMBOL_GPL(__add_pages);
  470 
  471 /**
  472  * __remove_pages() - remove sections of pages from a zone
  473  * @zone: zone from which pages need to be removed
  474  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
  475  * @nr_pages: number of pages to remove (must be multiple of section size)
  476  *
  477  * Generic helper function to remove section mappings and sysfs entries
  478  * for the section of the memory we are removing. Caller needs to make
  479  * sure that pages are marked reserved and zones are adjust properly by
  480  * calling offline_pages().
  481  */
  482 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
  483                  unsigned long nr_pages)
  484 {
  485         unsigned long i, ret = 0;
  486         int sections_to_remove;
  487 
  488         /*
  489          * We can only remove entire sections
  490          */
  491         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
  492         BUG_ON(nr_pages % PAGES_PER_SECTION);
  493 
  494         release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
  495 
  496         sections_to_remove = nr_pages / PAGES_PER_SECTION;
  497         for (i = 0; i < sections_to_remove; i++) {
  498                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
  499                 ret = __remove_section(zone, __pfn_to_section(pfn));
  500                 if (ret)
  501                         break;
  502         }
  503         return ret;
  504 }
  505 EXPORT_SYMBOL_GPL(__remove_pages);
  506 
  507 int set_online_page_callback(online_page_callback_t callback)
  508 {
  509         int rc = -EINVAL;
  510 
  511         lock_memory_hotplug();
  512 
  513         if (online_page_callback == generic_online_page) {
  514                 online_page_callback = callback;
  515                 rc = 0;
  516         }
  517 
  518         unlock_memory_hotplug();
  519 
  520         return rc;
  521 }
  522 EXPORT_SYMBOL_GPL(set_online_page_callback);
  523 
  524 int restore_online_page_callback(online_page_callback_t callback)
  525 {
  526         int rc = -EINVAL;
  527 
  528         lock_memory_hotplug();
  529 
  530         if (online_page_callback == callback) {
  531                 online_page_callback = generic_online_page;
  532                 rc = 0;
  533         }
  534 
  535         unlock_memory_hotplug();
  536 
  537         return rc;
  538 }
  539 EXPORT_SYMBOL_GPL(restore_online_page_callback);
  540 
  541 void __online_page_set_limits(struct page *page)
  542 {
  543         unsigned long pfn = page_to_pfn(page);
  544 
  545         if (pfn >= num_physpages)
  546                 num_physpages = pfn + 1;
  547 }
  548 EXPORT_SYMBOL_GPL(__online_page_set_limits);
  549 
  550 void __online_page_increment_counters(struct page *page)
  551 {
  552         totalram_pages++;
  553 
  554 #ifdef CONFIG_HIGHMEM
  555         if (PageHighMem(page))
  556                 totalhigh_pages++;
  557 #endif
  558 }
  559 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
  560 
  561 void __online_page_free(struct page *page)
  562 {
  563         ClearPageReserved(page);
  564         init_page_count(page);
  565         __free_page(page);
  566 }
  567 EXPORT_SYMBOL_GPL(__online_page_free);
  568 
  569 static void generic_online_page(struct page *page)
  570 {
  571         __online_page_set_limits(page);
  572         __online_page_increment_counters(page);
  573         __online_page_free(page);
  574 }
  575 
  576 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
  577                         void *arg)
  578 {
  579         unsigned long i;
  580         unsigned long onlined_pages = *(unsigned long *)arg;
  581         struct page *page;
  582         if (PageReserved(pfn_to_page(start_pfn)))
  583                 for (i = 0; i < nr_pages; i++) {
  584                         page = pfn_to_page(start_pfn + i);
  585                         (*online_page_callback)(page);
  586                         onlined_pages++;
  587                 }
  588         *(unsigned long *)arg = onlined_pages;
  589         return 0;
  590 }
  591 
  592 #ifdef CONFIG_MOVABLE_NODE
  593 /*
  594  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
  595  * normal memory.
  596  */
  597 static bool can_online_high_movable(struct zone *zone)
  598 {
  599         return true;
  600 }
  601 #else /* CONFIG_MOVABLE_NODE */
  602 /* ensure every online node has NORMAL memory */
  603 static bool can_online_high_movable(struct zone *zone)
  604 {
  605         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
  606 }
  607 #endif /* CONFIG_MOVABLE_NODE */
  608 
  609 /* check which state of node_states will be changed when online memory */
  610 static void node_states_check_changes_online(unsigned long nr_pages,
  611         struct zone *zone, struct memory_notify *arg)
  612 {
  613         int nid = zone_to_nid(zone);
  614         enum zone_type zone_last = ZONE_NORMAL;
  615 
  616         /*
  617          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
  618          * contains nodes which have zones of 0...ZONE_NORMAL,
  619          * set zone_last to ZONE_NORMAL.
  620          *
  621          * If we don't have HIGHMEM nor movable node,
  622          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
  623          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
  624          */
  625         if (N_MEMORY == N_NORMAL_MEMORY)
  626                 zone_last = ZONE_MOVABLE;
  627 
  628         /*
  629          * if the memory to be online is in a zone of 0...zone_last, and
  630          * the zones of 0...zone_last don't have memory before online, we will
  631          * need to set the node to node_states[N_NORMAL_MEMORY] after
  632          * the memory is online.
  633          */
  634         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
  635                 arg->status_change_nid_normal = nid;
  636         else
  637                 arg->status_change_nid_normal = -1;
  638 
  639 #ifdef CONFIG_HIGHMEM
  640         /*
  641          * If we have movable node, node_states[N_HIGH_MEMORY]
  642          * contains nodes which have zones of 0...ZONE_HIGHMEM,
  643          * set zone_last to ZONE_HIGHMEM.
  644          *
  645          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
  646          * contains nodes which have zones of 0...ZONE_MOVABLE,
  647          * set zone_last to ZONE_MOVABLE.
  648          */
  649         zone_last = ZONE_HIGHMEM;
  650         if (N_MEMORY == N_HIGH_MEMORY)
  651                 zone_last = ZONE_MOVABLE;
  652 
  653         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
  654                 arg->status_change_nid_high = nid;
  655         else
  656                 arg->status_change_nid_high = -1;
  657 #else
  658         arg->status_change_nid_high = arg->status_change_nid_normal;
  659 #endif
  660 
  661         /*
  662          * if the node don't have memory befor online, we will need to
  663          * set the node to node_states[N_MEMORY] after the memory
  664          * is online.
  665          */
  666         if (!node_state(nid, N_MEMORY))
  667                 arg->status_change_nid = nid;
  668         else
  669                 arg->status_change_nid = -1;
  670 }
  671 
  672 static void node_states_set_node(int node, struct memory_notify *arg)
  673 {
  674         if (arg->status_change_nid_normal >= 0)
  675                 node_set_state(node, N_NORMAL_MEMORY);
  676 
  677         if (arg->status_change_nid_high >= 0)
  678                 node_set_state(node, N_HIGH_MEMORY);
  679 
  680         node_set_state(node, N_MEMORY);
  681 }
  682 
  683 
  684 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
  685 {
  686         unsigned long onlined_pages = 0;
  687         struct zone *zone;
  688         int need_zonelists_rebuild = 0;
  689         int nid;
  690         int ret;
  691         struct memory_notify arg;
  692 
  693         lock_memory_hotplug();
  694         /*
  695          * This doesn't need a lock to do pfn_to_page().
  696          * The section can't be removed here because of the
  697          * memory_block->state_mutex.
  698          */
  699         zone = page_zone(pfn_to_page(pfn));
  700 
  701         if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
  702             !can_online_high_movable(zone)) {
  703                 unlock_memory_hotplug();
  704                 return -1;
  705         }
  706 
  707         if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
  708                 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
  709                         unlock_memory_hotplug();
  710                         return -1;
  711                 }
  712         }
  713         if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
  714                 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
  715                         unlock_memory_hotplug();
  716                         return -1;
  717                 }
  718         }
  719 
  720         /* Previous code may changed the zone of the pfn range */
  721         zone = page_zone(pfn_to_page(pfn));
  722 
  723         arg.start_pfn = pfn;
  724         arg.nr_pages = nr_pages;
  725         node_states_check_changes_online(nr_pages, zone, &arg);
  726 
  727         nid = page_to_nid(pfn_to_page(pfn));
  728 
  729         ret = memory_notify(MEM_GOING_ONLINE, &arg);
  730         ret = notifier_to_errno(ret);
  731         if (ret) {
  732                 memory_notify(MEM_CANCEL_ONLINE, &arg);
  733                 unlock_memory_hotplug();
  734                 return ret;
  735         }
  736         /*
  737          * If this zone is not populated, then it is not in zonelist.
  738          * This means the page allocator ignores this zone.
  739          * So, zonelist must be updated after online.
  740          */
  741         mutex_lock(&zonelists_mutex);
  742         if (!populated_zone(zone)) {
  743                 need_zonelists_rebuild = 1;
  744                 build_all_zonelists(NULL, zone);
  745         }
  746 
  747         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
  748                 online_pages_range);
  749         if (ret) {
  750                 if (need_zonelists_rebuild)
  751                         zone_pcp_reset(zone);
  752                 mutex_unlock(&zonelists_mutex);
  753                 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
  754                        (unsigned long long) pfn << PAGE_SHIFT,
  755                        (((unsigned long long) pfn + nr_pages)
  756                             << PAGE_SHIFT) - 1);
  757                 memory_notify(MEM_CANCEL_ONLINE, &arg);
  758                 unlock_memory_hotplug();
  759                 return ret;
  760         }
  761 
  762         zone->managed_pages += onlined_pages;
  763         zone->present_pages += onlined_pages;
  764         zone->zone_pgdat->node_present_pages += onlined_pages;
  765         if (onlined_pages) {
  766                 node_states_set_node(zone_to_nid(zone), &arg);
  767                 if (need_zonelists_rebuild)
  768                         build_all_zonelists(NULL, NULL);
  769                 else
  770                         zone_pcp_update(zone);
  771         }
  772 
  773         mutex_unlock(&zonelists_mutex);
  774 
  775         init_per_zone_wmark_min();
  776 
  777         if (onlined_pages)
  778                 kswapd_run(zone_to_nid(zone));
  779 
  780         vm_total_pages = nr_free_pagecache_pages();
  781 
  782         writeback_set_ratelimit();
  783 
  784         if (onlined_pages)
  785                 memory_notify(MEM_ONLINE, &arg);
  786         unlock_memory_hotplug();
  787 
  788         return 0;
  789 }
  790 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
  791 
  792 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
  793 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
  794 {
  795         struct pglist_data *pgdat;
  796         unsigned long zones_size[MAX_NR_ZONES] = {0};
  797         unsigned long zholes_size[MAX_NR_ZONES] = {0};
  798         unsigned long start_pfn = start >> PAGE_SHIFT;
  799 
  800         pgdat = arch_alloc_nodedata(nid);
  801         if (!pgdat)
  802                 return NULL;
  803 
  804         arch_refresh_nodedata(nid, pgdat);
  805 
  806         /* we can use NODE_DATA(nid) from here */
  807 
  808         /* init node's zones as empty zones, we don't have any present pages.*/
  809         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
  810 
  811         /*
  812          * The node we allocated has no zone fallback lists. For avoiding
  813          * to access not-initialized zonelist, build here.
  814          */
  815         mutex_lock(&zonelists_mutex);
  816         build_all_zonelists(pgdat, NULL);
  817         mutex_unlock(&zonelists_mutex);
  818 
  819         return pgdat;
  820 }
  821 
  822 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
  823 {
  824         arch_refresh_nodedata(nid, NULL);
  825         arch_free_nodedata(pgdat);
  826         return;
  827 }
  828 
  829 
  830 /*
  831  * called by cpu_up() to online a node without onlined memory.
  832  */
  833 int mem_online_node(int nid)
  834 {
  835         pg_data_t       *pgdat;
  836         int     ret;
  837 
  838         lock_memory_hotplug();
  839         pgdat = hotadd_new_pgdat(nid, 0);
  840         if (!pgdat) {
  841                 ret = -ENOMEM;
  842                 goto out;
  843         }
  844         node_set_online(nid);
  845         ret = register_one_node(nid);
  846         BUG_ON(ret);
  847 
  848 out:
  849         unlock_memory_hotplug();
  850         return ret;
  851 }
  852 
  853 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
  854 int __ref add_memory(int nid, u64 start, u64 size)
  855 {
  856         pg_data_t *pgdat = NULL;
  857         int new_pgdat = 0;
  858         struct resource *res;
  859         int ret;
  860 
  861         lock_memory_hotplug();
  862 
  863         res = register_memory_resource(start, size);
  864         ret = -EEXIST;
  865         if (!res)
  866                 goto out;
  867 
  868         if (!node_online(nid)) {
  869                 pgdat = hotadd_new_pgdat(nid, start);
  870                 ret = -ENOMEM;
  871                 if (!pgdat)
  872                         goto error;
  873                 new_pgdat = 1;
  874         }
  875 
  876         /* call arch's memory hotadd */
  877         ret = arch_add_memory(nid, start, size);
  878 
  879         if (ret < 0)
  880                 goto error;
  881 
  882         /* we online node here. we can't roll back from here. */
  883         node_set_online(nid);
  884 
  885         if (new_pgdat) {
  886                 ret = register_one_node(nid);
  887                 /*
  888                  * If sysfs file of new node can't create, cpu on the node
  889                  * can't be hot-added. There is no rollback way now.
  890                  * So, check by BUG_ON() to catch it reluctantly..
  891                  */
  892                 BUG_ON(ret);
  893         }
  894 
  895         /* create new memmap entry */
  896         firmware_map_add_hotplug(start, start + size, "System RAM");
  897 
  898         goto out;
  899 
  900 error:
  901         /* rollback pgdat allocation and others */
  902         if (new_pgdat)
  903                 rollback_node_hotadd(nid, pgdat);
  904         if (res)
  905                 release_memory_resource(res);
  906 
  907 out:
  908         unlock_memory_hotplug();
  909         return ret;
  910 }
  911 EXPORT_SYMBOL_GPL(add_memory);
  912 
  913 #ifdef CONFIG_MEMORY_HOTREMOVE
  914 /*
  915  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
  916  * set and the size of the free page is given by page_order(). Using this,
  917  * the function determines if the pageblock contains only free pages.
  918  * Due to buddy contraints, a free page at least the size of a pageblock will
  919  * be located at the start of the pageblock
  920  */
  921 static inline int pageblock_free(struct page *page)
  922 {
  923         return PageBuddy(page) && page_order(page) >= pageblock_order;
  924 }
  925 
  926 /* Return the start of the next active pageblock after a given page */
  927 static struct page *next_active_pageblock(struct page *page)
  928 {
  929         /* Ensure the starting page is pageblock-aligned */
  930         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
  931 
  932         /* If the entire pageblock is free, move to the end of free page */
  933         if (pageblock_free(page)) {
  934                 int order;
  935                 /* be careful. we don't have locks, page_order can be changed.*/
  936                 order = page_order(page);
  937                 if ((order < MAX_ORDER) && (order >= pageblock_order))
  938                         return page + (1 << order);
  939         }
  940 
  941         return page + pageblock_nr_pages;
  942 }
  943 
  944 /* Checks if this range of memory is likely to be hot-removable. */
  945 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
  946 {
  947         struct page *page = pfn_to_page(start_pfn);
  948         struct page *end_page = page + nr_pages;
  949 
  950         /* Check the starting page of each pageblock within the range */
  951         for (; page < end_page; page = next_active_pageblock(page)) {
  952                 if (!is_pageblock_removable_nolock(page))
  953                         return 0;
  954                 cond_resched();
  955         }
  956 
  957         /* All pageblocks in the memory block are likely to be hot-removable */
  958         return 1;
  959 }
  960 
  961 /*
  962  * Confirm all pages in a range [start, end) is belongs to the same zone.
  963  */
  964 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
  965 {
  966         unsigned long pfn;
  967         struct zone *zone = NULL;
  968         struct page *page;
  969         int i;
  970         for (pfn = start_pfn;
  971              pfn < end_pfn;
  972              pfn += MAX_ORDER_NR_PAGES) {
  973                 i = 0;
  974                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
  975                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
  976                         i++;
  977                 if (i == MAX_ORDER_NR_PAGES)
  978                         continue;
  979                 page = pfn_to_page(pfn + i);
  980                 if (zone && page_zone(page) != zone)
  981                         return 0;
  982                 zone = page_zone(page);
  983         }
  984         return 1;
  985 }
  986 
  987 /*
  988  * Scanning pfn is much easier than scanning lru list.
  989  * Scan pfn from start to end and Find LRU page.
  990  */
  991 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
  992 {
  993         unsigned long pfn;
  994         struct page *page;
  995         for (pfn = start; pfn < end; pfn++) {
  996                 if (pfn_valid(pfn)) {
  997                         page = pfn_to_page(pfn);
  998                         if (PageLRU(page))
  999                                 return pfn;
 1000                 }
 1001         }
 1002         return 0;
 1003 }
 1004 
 1005 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 1006 static int
 1007 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 1008 {
 1009         unsigned long pfn;
 1010         struct page *page;
 1011         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 1012         int not_managed = 0;
 1013         int ret = 0;
 1014         LIST_HEAD(source);
 1015 
 1016         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 1017                 if (!pfn_valid(pfn))
 1018                         continue;
 1019                 page = pfn_to_page(pfn);
 1020                 if (!get_page_unless_zero(page))
 1021                         continue;
 1022                 /*
 1023                  * We can skip free pages. And we can only deal with pages on
 1024                  * LRU.
 1025                  */
 1026                 ret = isolate_lru_page(page);
 1027                 if (!ret) { /* Success */
 1028                         put_page(page);
 1029                         list_add_tail(&page->lru, &source);
 1030                         move_pages--;
 1031                         inc_zone_page_state(page, NR_ISOLATED_ANON +
 1032                                             page_is_file_cache(page));
 1033 
 1034                 } else {
 1035 #ifdef CONFIG_DEBUG_VM
 1036                         printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
 1037                                pfn);
 1038                         dump_page(page);
 1039 #endif
 1040                         put_page(page);
 1041                         /* Because we don't have big zone->lock. we should
 1042                            check this again here. */
 1043                         if (page_count(page)) {
 1044                                 not_managed++;
 1045                                 ret = -EBUSY;
 1046                                 break;
 1047                         }
 1048                 }
 1049         }
 1050         if (!list_empty(&source)) {
 1051                 if (not_managed) {
 1052                         putback_lru_pages(&source);
 1053                         goto out;
 1054                 }
 1055 
 1056                 /*
 1057                  * alloc_migrate_target should be improooooved!!
 1058                  * migrate_pages returns # of failed pages.
 1059                  */
 1060                 ret = migrate_pages(&source, alloc_migrate_target, 0,
 1061                                                         true, MIGRATE_SYNC,
 1062                                                         MR_MEMORY_HOTPLUG);
 1063                 if (ret)
 1064                         putback_lru_pages(&source);
 1065         }
 1066 out:
 1067         return ret;
 1068 }
 1069 
 1070 /*
 1071  * remove from free_area[] and mark all as Reserved.
 1072  */
 1073 static int
 1074 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 1075                         void *data)
 1076 {
 1077         __offline_isolated_pages(start, start + nr_pages);
 1078         return 0;
 1079 }
 1080 
 1081 static void
 1082 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 1083 {
 1084         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
 1085                                 offline_isolated_pages_cb);
 1086 }
 1087 
 1088 /*
 1089  * Check all pages in range, recoreded as memory resource, are isolated.
 1090  */
 1091 static int
 1092 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 1093                         void *data)
 1094 {
 1095         int ret;
 1096         long offlined = *(long *)data;
 1097         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
 1098         offlined = nr_pages;
 1099         if (!ret)
 1100                 *(long *)data += offlined;
 1101         return ret;
 1102 }
 1103 
 1104 static long
 1105 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 1106 {
 1107         long offlined = 0;
 1108         int ret;
 1109 
 1110         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
 1111                         check_pages_isolated_cb);
 1112         if (ret < 0)
 1113                 offlined = (long)ret;
 1114         return offlined;
 1115 }
 1116 
 1117 #ifdef CONFIG_MOVABLE_NODE
 1118 /*
 1119  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
 1120  * normal memory.
 1121  */
 1122 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
 1123 {
 1124         return true;
 1125 }
 1126 #else /* CONFIG_MOVABLE_NODE */
 1127 /* ensure the node has NORMAL memory if it is still online */
 1128 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
 1129 {
 1130         struct pglist_data *pgdat = zone->zone_pgdat;
 1131         unsigned long present_pages = 0;
 1132         enum zone_type zt;
 1133 
 1134         for (zt = 0; zt <= ZONE_NORMAL; zt++)
 1135                 present_pages += pgdat->node_zones[zt].present_pages;
 1136 
 1137         if (present_pages > nr_pages)
 1138                 return true;
 1139 
 1140         present_pages = 0;
 1141         for (; zt <= ZONE_MOVABLE; zt++)
 1142                 present_pages += pgdat->node_zones[zt].present_pages;
 1143 
 1144         /*
 1145          * we can't offline the last normal memory until all
 1146          * higher memory is offlined.
 1147          */
 1148         return present_pages == 0;
 1149 }
 1150 #endif /* CONFIG_MOVABLE_NODE */
 1151 
 1152 /* check which state of node_states will be changed when offline memory */
 1153 static void node_states_check_changes_offline(unsigned long nr_pages,
 1154                 struct zone *zone, struct memory_notify *arg)
 1155 {
 1156         struct pglist_data *pgdat = zone->zone_pgdat;
 1157         unsigned long present_pages = 0;
 1158         enum zone_type zt, zone_last = ZONE_NORMAL;
 1159 
 1160         /*
 1161          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
 1162          * contains nodes which have zones of 0...ZONE_NORMAL,
 1163          * set zone_last to ZONE_NORMAL.
 1164          *
 1165          * If we don't have HIGHMEM nor movable node,
 1166          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
 1167          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
 1168          */
 1169         if (N_MEMORY == N_NORMAL_MEMORY)
 1170                 zone_last = ZONE_MOVABLE;
 1171 
 1172         /*
 1173          * check whether node_states[N_NORMAL_MEMORY] will be changed.
 1174          * If the memory to be offline is in a zone of 0...zone_last,
 1175          * and it is the last present memory, 0...zone_last will
 1176          * become empty after offline , thus we can determind we will
 1177          * need to clear the node from node_states[N_NORMAL_MEMORY].
 1178          */
 1179         for (zt = 0; zt <= zone_last; zt++)
 1180                 present_pages += pgdat->node_zones[zt].present_pages;
 1181         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
 1182                 arg->status_change_nid_normal = zone_to_nid(zone);
 1183         else
 1184                 arg->status_change_nid_normal = -1;
 1185 
 1186 #ifdef CONFIG_HIGHMEM
 1187         /*
 1188          * If we have movable node, node_states[N_HIGH_MEMORY]
 1189          * contains nodes which have zones of 0...ZONE_HIGHMEM,
 1190          * set zone_last to ZONE_HIGHMEM.
 1191          *
 1192          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
 1193          * contains nodes which have zones of 0...ZONE_MOVABLE,
 1194          * set zone_last to ZONE_MOVABLE.
 1195          */
 1196         zone_last = ZONE_HIGHMEM;
 1197         if (N_MEMORY == N_HIGH_MEMORY)
 1198                 zone_last = ZONE_MOVABLE;
 1199 
 1200         for (; zt <= zone_last; zt++)
 1201                 present_pages += pgdat->node_zones[zt].present_pages;
 1202         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
 1203                 arg->status_change_nid_high = zone_to_nid(zone);
 1204         else
 1205                 arg->status_change_nid_high = -1;
 1206 #else
 1207         arg->status_change_nid_high = arg->status_change_nid_normal;
 1208 #endif
 1209 
 1210         /*
 1211          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
 1212          */
 1213         zone_last = ZONE_MOVABLE;
 1214 
 1215         /*
 1216          * check whether node_states[N_HIGH_MEMORY] will be changed
 1217          * If we try to offline the last present @nr_pages from the node,
 1218          * we can determind we will need to clear the node from
 1219          * node_states[N_HIGH_MEMORY].
 1220          */
 1221         for (; zt <= zone_last; zt++)
 1222                 present_pages += pgdat->node_zones[zt].present_pages;
 1223         if (nr_pages >= present_pages)
 1224                 arg->status_change_nid = zone_to_nid(zone);
 1225         else
 1226                 arg->status_change_nid = -1;
 1227 }
 1228 
 1229 static void node_states_clear_node(int node, struct memory_notify *arg)
 1230 {
 1231         if (arg->status_change_nid_normal >= 0)
 1232                 node_clear_state(node, N_NORMAL_MEMORY);
 1233 
 1234         if ((N_MEMORY != N_NORMAL_MEMORY) &&
 1235             (arg->status_change_nid_high >= 0))
 1236                 node_clear_state(node, N_HIGH_MEMORY);
 1237 
 1238         if ((N_MEMORY != N_HIGH_MEMORY) &&
 1239             (arg->status_change_nid >= 0))
 1240                 node_clear_state(node, N_MEMORY);
 1241 }
 1242 
 1243 static int __ref __offline_pages(unsigned long start_pfn,
 1244                   unsigned long end_pfn, unsigned long timeout)
 1245 {
 1246         unsigned long pfn, nr_pages, expire;
 1247         long offlined_pages;
 1248         int ret, drain, retry_max, node;
 1249         struct zone *zone;
 1250         struct memory_notify arg;
 1251 
 1252         BUG_ON(start_pfn >= end_pfn);
 1253         /* at least, alignment against pageblock is necessary */
 1254         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 1255                 return -EINVAL;
 1256         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 1257                 return -EINVAL;
 1258         /* This makes hotplug much easier...and readable.
 1259            we assume this for now. .*/
 1260         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 1261                 return -EINVAL;
 1262 
 1263         lock_memory_hotplug();
 1264 
 1265         zone = page_zone(pfn_to_page(start_pfn));
 1266         node = zone_to_nid(zone);
 1267         nr_pages = end_pfn - start_pfn;
 1268 
 1269         ret = -EINVAL;
 1270         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
 1271                 goto out;
 1272 
 1273         /* set above range as isolated */
 1274         ret = start_isolate_page_range(start_pfn, end_pfn,
 1275                                        MIGRATE_MOVABLE, true);
 1276         if (ret)
 1277                 goto out;
 1278 
 1279         arg.start_pfn = start_pfn;
 1280         arg.nr_pages = nr_pages;
 1281         node_states_check_changes_offline(nr_pages, zone, &arg);
 1282 
 1283         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 1284         ret = notifier_to_errno(ret);
 1285         if (ret)
 1286                 goto failed_removal;
 1287 
 1288         pfn = start_pfn;
 1289         expire = jiffies + timeout;
 1290         drain = 0;
 1291         retry_max = 5;
 1292 repeat:
 1293         /* start memory hot removal */
 1294         ret = -EAGAIN;
 1295         if (time_after(jiffies, expire))
 1296                 goto failed_removal;
 1297         ret = -EINTR;
 1298         if (signal_pending(current))
 1299                 goto failed_removal;
 1300         ret = 0;
 1301         if (drain) {
 1302                 lru_add_drain_all();
 1303                 cond_resched();
 1304                 drain_all_pages();
 1305         }
 1306 
 1307         pfn = scan_lru_pages(start_pfn, end_pfn);
 1308         if (pfn) { /* We have page on LRU */
 1309                 ret = do_migrate_range(pfn, end_pfn);
 1310                 if (!ret) {
 1311                         drain = 1;
 1312                         goto repeat;
 1313                 } else {
 1314                         if (ret < 0)
 1315                                 if (--retry_max == 0)
 1316                                         goto failed_removal;
 1317                         yield();
 1318                         drain = 1;
 1319                         goto repeat;
 1320                 }
 1321         }
 1322         /* drain all zone's lru pagevec, this is asynchronous... */
 1323         lru_add_drain_all();
 1324         yield();
 1325         /* drain pcp pages, this is synchronous. */
 1326         drain_all_pages();
 1327         /* check again */
 1328         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 1329         if (offlined_pages < 0) {
 1330                 ret = -EBUSY;
 1331                 goto failed_removal;
 1332         }
 1333         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 1334         /* Ok, all of our target is isolated.
 1335            We cannot do rollback at this point. */
 1336         offline_isolated_pages(start_pfn, end_pfn);
 1337         /* reset pagetype flags and makes migrate type to be MOVABLE */
 1338         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
 1339         /* removal success */
 1340         zone->managed_pages -= offlined_pages;
 1341         zone->present_pages -= offlined_pages;
 1342         zone->zone_pgdat->node_present_pages -= offlined_pages;
 1343         totalram_pages -= offlined_pages;
 1344 
 1345         init_per_zone_wmark_min();
 1346 
 1347         if (!populated_zone(zone)) {
 1348                 zone_pcp_reset(zone);
 1349                 mutex_lock(&zonelists_mutex);
 1350                 build_all_zonelists(NULL, NULL);
 1351                 mutex_unlock(&zonelists_mutex);
 1352         } else
 1353                 zone_pcp_update(zone);
 1354 
 1355         node_states_clear_node(node, &arg);
 1356         if (arg.status_change_nid >= 0)
 1357                 kswapd_stop(node);
 1358 
 1359         vm_total_pages = nr_free_pagecache_pages();
 1360         writeback_set_ratelimit();
 1361 
 1362         memory_notify(MEM_OFFLINE, &arg);
 1363         unlock_memory_hotplug();
 1364         return 0;
 1365 
 1366 failed_removal:
 1367         printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
 1368                (unsigned long long) start_pfn << PAGE_SHIFT,
 1369                ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
 1370         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 1371         /* pushback to free area */
 1372         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
 1373 
 1374 out:
 1375         unlock_memory_hotplug();
 1376         return ret;
 1377 }
 1378 
 1379 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
 1380 {
 1381         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
 1382 }
 1383 
 1384 int remove_memory(u64 start, u64 size)
 1385 {
 1386         struct memory_block *mem = NULL;
 1387         struct mem_section *section;
 1388         unsigned long start_pfn, end_pfn;
 1389         unsigned long pfn, section_nr;
 1390         int ret;
 1391 
 1392         start_pfn = PFN_DOWN(start);
 1393         end_pfn = start_pfn + PFN_DOWN(size);
 1394 
 1395         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 1396                 section_nr = pfn_to_section_nr(pfn);
 1397                 if (!present_section_nr(section_nr))
 1398                         continue;
 1399 
 1400                 section = __nr_to_section(section_nr);
 1401                 /* same memblock? */
 1402                 if (mem)
 1403                         if ((section_nr >= mem->start_section_nr) &&
 1404                             (section_nr <= mem->end_section_nr))
 1405                                 continue;
 1406 
 1407                 mem = find_memory_block_hinted(section, mem);
 1408                 if (!mem)
 1409                         continue;
 1410 
 1411                 ret = offline_memory_block(mem);
 1412                 if (ret) {
 1413                         kobject_put(&mem->dev.kobj);
 1414                         return ret;
 1415                 }
 1416         }
 1417 
 1418         if (mem)
 1419                 kobject_put(&mem->dev.kobj);
 1420 
 1421         return 0;
 1422 }
 1423 #else
 1424 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
 1425 {
 1426         return -EINVAL;
 1427 }
 1428 int remove_memory(u64 start, u64 size)
 1429 {
 1430         return -EINVAL;
 1431 }
 1432 #endif /* CONFIG_MEMORY_HOTREMOVE */
 1433 EXPORT_SYMBOL_GPL(remove_memory);
Cache object: 42b63097ed8152161a4813b9f1ef907e
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.
FreeBSD/Linux Kernel Cross Reference sys/mm/memory_hotplug.c

FreeBSD/Linux Kernel Cross Reference
sys/mm/memory_hotplug.c
