mm/memory_hotplug.c

   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/module.h>
  15 #include <linux/pagevec.h>
  16 #include <linux/slab.h>
  17 #include <linux/sysctl.h>
  18 #include <linux/cpu.h>
  19 #include <linux/memory.h>
  20 #include <linux/memory_hotplug.h>
  21 #include <linux/highmem.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/ioport.h>
  24
  25 #include <asm/tlbflush.h>
  26
  27 extern void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
  28                           unsigned long size);
  29 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
  30 {
  31         struct pglist_data *pgdat = zone->zone_pgdat;
  32         int nr_pages = PAGES_PER_SECTION;
  33         int nid = pgdat->node_id;
  34         int zone_type;
  35
  36         zone_type = zone - pgdat->node_zones;
  37         if (!populated_zone(zone)) {
  38                 int ret = 0;
  39                 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
  40                 if (ret < 0)
  41                         return ret;
  42         }
  43         memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
  44         zonetable_add(zone, nid, zone_type, phys_start_pfn, nr_pages);
  45         return 0;
  46 }
  47
  48 extern int sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
  49                                   int nr_pages);
  50 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
  51 {
  52         int nr_pages = PAGES_PER_SECTION;
  53         int ret;
  54
  55         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
  56
  57         if (ret < 0)
  58                 return ret;
  59
  60         ret = __add_zone(zone, phys_start_pfn);
  61
  62         if (ret < 0)
  63                 return ret;
  64
  65         return register_new_memory(__pfn_to_section(phys_start_pfn));
  66 }
  67
  68 /*
  69  * Reasonably generic function for adding memory.  It is
  70  * expected that archs that support memory hotplug will
  71  * call this function after deciding the zone to which to
  72  * add the new pages.
  73  */
  74 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
  75                  unsigned long nr_pages)
  76 {
  77         unsigned long i;
  78         int err = 0;
  79
  80         for (i = 0; i < nr_pages; i += PAGES_PER_SECTION) {
  81                 err = __add_section(zone, phys_start_pfn + i);
  82
  83                 /* We want to keep adding the rest of the
  84                  * sections if the first ones already exist
  85                  */
  86                 if (err && (err != -EEXIST))
  87                         break;
  88         }
  89
  90         return err;
  91 }
  92 EXPORT_SYMBOL_GPL(__add_pages);
  93
  94 static void grow_zone_span(struct zone *zone,
  95                 unsigned long start_pfn, unsigned long end_pfn)
  96 {
  97         unsigned long old_zone_end_pfn;
  98
  99         zone_span_writelock(zone);
 100
 101         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 102         if (start_pfn < zone->zone_start_pfn)
 103                 zone->zone_start_pfn = start_pfn;
 104
 105         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 106                                 zone->zone_start_pfn;
 107
 108         zone_span_writeunlock(zone);
 109 }
 110
 111 static void grow_pgdat_span(struct pglist_data *pgdat,
 112                 unsigned long start_pfn, unsigned long end_pfn)
 113 {
 114         unsigned long old_pgdat_end_pfn =
 115                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 116
 117         if (start_pfn < pgdat->node_start_pfn)
 118                 pgdat->node_start_pfn = start_pfn;
 119
 120         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 121                                         pgdat->node_start_pfn;
 122 }
 123
 124 int online_pages(unsigned long pfn, unsigned long nr_pages)
 125 {
 126         unsigned long i;
 127         unsigned long flags;
 128         unsigned long onlined_pages = 0;
 129         struct resource res;
 130         u64 section_end;
 131         unsigned long start_pfn;
 132         struct zone *zone;
 133         int need_zonelists_rebuild = 0;
 134
 135         /*
 136          * This doesn't need a lock to do pfn_to_page().
 137          * The section can't be removed here because of the
 138          * memory_block->state_sem.
 139          */
 140         zone = page_zone(pfn_to_page(pfn));
 141         pgdat_resize_lock(zone->zone_pgdat, &flags);
 142         grow_zone_span(zone, pfn, pfn + nr_pages);
 143         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
 144         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 145
 146         /*
 147          * If this zone is not populated, then it is not in zonelist.
 148          * This means the page allocator ignores this zone.
 149          * So, zonelist must be updated after online.
 150          */
 151         if (!populated_zone(zone))
 152                 need_zonelists_rebuild = 1;
 153
 154         res.start = (u64)pfn << PAGE_SHIFT;
 155         res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
 156         res.flags = IORESOURCE_MEM; /* we just need system ram */
 157         section_end = res.end;
 158
 159         while (find_next_system_ram(&res) >= 0) {
 160                 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
 161                 nr_pages = (unsigned long)
 162                            ((res.end + 1 - res.start) >> PAGE_SHIFT);
 163
 164                 if (PageReserved(pfn_to_page(start_pfn))) {
 165                         /* this region's page is not onlined now */
 166                         for (i = 0; i < nr_pages; i++) {
 167                                 struct page *page = pfn_to_page(start_pfn + i);
 168                                 online_page(page);
 169                                 onlined_pages++;
 170                         }
 171                 }
 172
 173                 res.start = res.end + 1;
 174                 res.end = section_end;
 175         }
 176         zone->present_pages += onlined_pages;
 177         zone->zone_pgdat->node_present_pages += onlined_pages;
 178
 179         setup_per_zone_pages_min();
 180
 181         if (need_zonelists_rebuild)
 182                 build_all_zonelists();
 183         vm_total_pages = nr_free_pagecache_pages();
 184         return 0;
 185 }
 186
 187 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 188 {
 189         struct pglist_data *pgdat;
 190         unsigned long zones_size[MAX_NR_ZONES] = {0};
 191         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 192         unsigned long start_pfn = start >> PAGE_SHIFT;
 193
 194         pgdat = arch_alloc_nodedata(nid);
 195         if (!pgdat)
 196                 return NULL;
 197
 198         arch_refresh_nodedata(nid, pgdat);
 199
 200         /* we can use NODE_DATA(nid) from here */
 201
 202         /* init node's zones as empty zones, we don't have any present pages.*/
 203         free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
 204
 205         return pgdat;
 206 }
 207
 208 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 209 {
 210         arch_refresh_nodedata(nid, NULL);
 211         arch_free_nodedata(pgdat);
 212         return;
 213 }
 214
 215 /* add this memory to iomem resource */
 216 static void register_memory_resource(u64 start, u64 size)
 217 {
 218         struct resource *res;
 219
 220         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 221         BUG_ON(!res);
 222
 223         res->name = "System RAM";
 224         res->start = start;
 225         res->end = start + size - 1;
 226         res->flags = IORESOURCE_MEM;
 227         if (request_resource(&iomem_resource, res) < 0) {
 228                 printk("System RAM resource %llx - %llx cannot be added\n",
 229                 (unsigned long long)res->start, (unsigned long long)res->end);
 230                 kfree(res);
 231         }
 232 }
 233
 234
 235
 236 int add_memory(int nid, u64 start, u64 size)
 237 {
 238         pg_data_t *pgdat = NULL;
 239         int new_pgdat = 0;
 240         int ret;
 241
 242         if (!node_online(nid)) {
 243                 pgdat = hotadd_new_pgdat(nid, start);
 244                 if (!pgdat)
 245                         return -ENOMEM;
 246                 new_pgdat = 1;
 247                 ret = kswapd_run(nid);
 248                 if (ret)
 249                         goto error;
 250         }
 251
 252         /* call arch's memory hotadd */
 253         ret = arch_add_memory(nid, start, size);
 254
 255         if (ret < 0)
 256                 goto error;
 257
 258         /* we online node here. we can't roll back from here. */
 259         node_set_online(nid);
 260
 261         if (new_pgdat) {
 262                 ret = register_one_node(nid);
 263                 /*
 264                  * If sysfs file of new node can't create, cpu on the node
 265                  * can't be hot-added. There is no rollback way now.
 266                  * So, check by BUG_ON() to catch it reluctantly..
 267                  */
 268                 BUG_ON(ret);
 269         }
 270
 271         /* register this memory as resource */
 272         register_memory_resource(start, size);
 273
 274         return ret;
 275 error:
 276         /* rollback pgdat allocation and others */
 277         if (new_pgdat)
 278                 rollback_node_hotadd(nid, pgdat);
 279
 280         return ret;
 281 }
 282 EXPORT_SYMBOL_GPL(add_memory);