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         int start_sec, end_sec;
  80         /* during initialize mem_map, align hot-added range to section */
  81         start_sec = pfn_to_section_nr(phys_start_pfn);
  82         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
  83
  84         for (i = start_sec; i <= end_sec; i++) {
  85                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
  86
  87                 /*
  88                  * EEXIST is finally dealed with by ioresource collision
  89                  * check. see add_memory() => register_memory_resource()
  90                  * Warning will be printed if there is collision.
  91                  */
  92                 if (err && (err != -EEXIST))
  93                         break;
  94                 err = 0;
  95         }
  96
  97         return err;
  98 }
  99 EXPORT_SYMBOL_GPL(__add_pages);
 100
 101 static void grow_zone_span(struct zone *zone,
 102                 unsigned long start_pfn, unsigned long end_pfn)
 103 {
 104         unsigned long old_zone_end_pfn;
 105
 106         zone_span_writelock(zone);
 107
 108         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 109         if (start_pfn < zone->zone_start_pfn)
 110                 zone->zone_start_pfn = start_pfn;
 111
 112         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 113                                 zone->zone_start_pfn;
 114
 115         zone_span_writeunlock(zone);
 116 }
 117
 118 static void grow_pgdat_span(struct pglist_data *pgdat,
 119                 unsigned long start_pfn, unsigned long end_pfn)
 120 {
 121         unsigned long old_pgdat_end_pfn =
 122                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 123
 124         if (start_pfn < pgdat->node_start_pfn)
 125                 pgdat->node_start_pfn = start_pfn;
 126
 127         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 128                                         pgdat->node_start_pfn;
 129 }
 130
 131 int online_pages(unsigned long pfn, unsigned long nr_pages)
 132 {
 133         unsigned long i;
 134         unsigned long flags;
 135         unsigned long onlined_pages = 0;
 136         struct resource res;
 137         u64 section_end;
 138         unsigned long start_pfn;
 139         struct zone *zone;
 140         int need_zonelists_rebuild = 0;
 141
 142         /*
 143          * This doesn't need a lock to do pfn_to_page().
 144          * The section can't be removed here because of the
 145          * memory_block->state_sem.
 146          */
 147         zone = page_zone(pfn_to_page(pfn));
 148         pgdat_resize_lock(zone->zone_pgdat, &flags);
 149         grow_zone_span(zone, pfn, pfn + nr_pages);
 150         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
 151         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 152
 153         /*
 154          * If this zone is not populated, then it is not in zonelist.
 155          * This means the page allocator ignores this zone.
 156          * So, zonelist must be updated after online.
 157          */
 158         if (!populated_zone(zone))
 159                 need_zonelists_rebuild = 1;
 160
 161         res.start = (u64)pfn << PAGE_SHIFT;
 162         res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
 163         res.flags = IORESOURCE_MEM; /* we just need system ram */
 164         section_end = res.end;
 165
 166         while (find_next_system_ram(&res) >= 0) {
 167                 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
 168                 nr_pages = (unsigned long)
 169                            ((res.end + 1 - res.start) >> PAGE_SHIFT);
 170
 171                 if (PageReserved(pfn_to_page(start_pfn))) {
 172                         /* this region's page is not onlined now */
 173                         for (i = 0; i < nr_pages; i++) {
 174                                 struct page *page = pfn_to_page(start_pfn + i);
 175                                 online_page(page);
 176                                 onlined_pages++;
 177                         }
 178                 }
 179
 180                 res.start = res.end + 1;
 181                 res.end = section_end;
 182         }
 183         zone->present_pages += onlined_pages;
 184         zone->zone_pgdat->node_present_pages += onlined_pages;
 185
 186         setup_per_zone_pages_min();
 187
 188         if (need_zonelists_rebuild)
 189                 build_all_zonelists();
 190         vm_total_pages = nr_free_pagecache_pages();
 191         return 0;
 192 }
 193
 194 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 195 {
 196         struct pglist_data *pgdat;
 197         unsigned long zones_size[MAX_NR_ZONES] = {0};
 198         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 199         unsigned long start_pfn = start >> PAGE_SHIFT;
 200
 201         pgdat = arch_alloc_nodedata(nid);
 202         if (!pgdat)
 203                 return NULL;
 204
 205         arch_refresh_nodedata(nid, pgdat);
 206
 207         /* we can use NODE_DATA(nid) from here */
 208
 209         /* init node's zones as empty zones, we don't have any present pages.*/
 210         free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
 211
 212         return pgdat;
 213 }
 214
 215 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 216 {
 217         arch_refresh_nodedata(nid, NULL);
 218         arch_free_nodedata(pgdat);
 219         return;
 220 }
 221
 222 /* add this memory to iomem resource */
 223 static int register_memory_resource(u64 start, u64 size)
 224 {
 225         struct resource *res;
 226         int ret = 0;
 227         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 228         BUG_ON(!res);
 229
 230         res->name = "System RAM";
 231         res->start = start;
 232         res->end = start + size - 1;
 233         res->flags = IORESOURCE_MEM;
 234         if (request_resource(&iomem_resource, res) < 0) {
 235                 printk("System RAM resource %llx - %llx cannot be added\n",
 236                 (unsigned long long)res->start, (unsigned long long)res->end);
 237                 kfree(res);
 238                 ret = -EEXIST;
 239         }
 240         return ret;
 241 }
 242
 243
 244
 245 int add_memory(int nid, u64 start, u64 size)
 246 {
 247         pg_data_t *pgdat = NULL;
 248         int new_pgdat = 0;
 249         int ret;
 250
 251         if (!node_online(nid)) {
 252                 pgdat = hotadd_new_pgdat(nid, start);
 253                 if (!pgdat)
 254                         return -ENOMEM;
 255                 new_pgdat = 1;
 256                 ret = kswapd_run(nid);
 257                 if (ret)
 258                         goto error;
 259         }
 260
 261         /* call arch's memory hotadd */
 262         ret = arch_add_memory(nid, start, size);
 263
 264         if (ret < 0)
 265                 goto error;
 266
 267         /* we online node here. we can't roll back from here. */
 268         node_set_online(nid);
 269
 270         if (new_pgdat) {
 271                 ret = register_one_node(nid);
 272                 /*
 273                  * If sysfs file of new node can't create, cpu on the node
 274                  * can't be hot-added. There is no rollback way now.
 275                  * So, check by BUG_ON() to catch it reluctantly..
 276                  */
 277                 BUG_ON(ret);
 278         }
 279
 280         /* register this memory as resource */
 281         ret = register_memory_resource(start, size);
 282
 283         return ret;
 284 error:
 285         /* rollback pgdat allocation and others */
 286         if (new_pgdat)
 287                 rollback_node_hotadd(nid, pgdat);
 288
 289         return ret;
 290 }
 291 EXPORT_SYMBOL_GPL(add_memory);