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/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/cpuset.h>
  26
  27 #include <asm/tlbflush.h>
  28
  29 /* add this memory to iomem resource */
  30 static struct resource *register_memory_resource(u64 start, u64 size)
  31 {
  32         struct resource *res;
  33         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
  34         BUG_ON(!res);
  35
  36         res->name = "System RAM";
  37         res->start = start;
  38         res->end = start + size - 1;
  39         res->flags = IORESOURCE_MEM;
  40         if (request_resource(&iomem_resource, res) < 0) {
  41                 printk("System RAM resource %llx - %llx cannot be added\n",
  42                 (unsigned long long)res->start, (unsigned long long)res->end);
  43                 kfree(res);
  44                 res = NULL;
  45         }
  46         return res;
  47 }
  48
  49 static void release_memory_resource(struct resource *res)
  50 {
  51         if (!res)
  52                 return;
  53         release_resource(res);
  54         kfree(res);
  55         return;
  56 }
  57
  58
  59 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  60 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
  61 {
  62         struct pglist_data *pgdat = zone->zone_pgdat;
  63         int nr_pages = PAGES_PER_SECTION;
  64         int nid = pgdat->node_id;
  65         int zone_type;
  66
  67         zone_type = zone - pgdat->node_zones;
  68         if (!populated_zone(zone)) {
  69                 int ret = 0;
  70                 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
  71                 if (ret < 0)
  72                         return ret;
  73         }
  74         memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
  75         zonetable_add(zone, nid, zone_type, phys_start_pfn, nr_pages);
  76         return 0;
  77 }
  78
  79 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
  80 {
  81         int nr_pages = PAGES_PER_SECTION;
  82         int ret;
  83
  84         if (pfn_valid(phys_start_pfn))
  85                 return -EEXIST;
  86
  87         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
  88
  89         if (ret < 0)
  90                 return ret;
  91
  92         ret = __add_zone(zone, phys_start_pfn);
  93
  94         if (ret < 0)
  95                 return ret;
  96
  97         return register_new_memory(__pfn_to_section(phys_start_pfn));
  98 }
  99
 100 /*
 101  * Reasonably generic function for adding memory.  It is
 102  * expected that archs that support memory hotplug will
 103  * call this function after deciding the zone to which to
 104  * add the new pages.
 105  */
 106 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
 107                  unsigned long nr_pages)
 108 {
 109         unsigned long i;
 110         int err = 0;
 111         int start_sec, end_sec;
 112         /* during initialize mem_map, align hot-added range to section */
 113         start_sec = pfn_to_section_nr(phys_start_pfn);
 114         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 115
 116         for (i = start_sec; i <= end_sec; i++) {
 117                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
 118
 119                 /*
 120                  * EEXIST is finally dealed with by ioresource collision
 121                  * check. see add_memory() => register_memory_resource()
 122                  * Warning will be printed if there is collision.
 123                  */
 124                 if (err && (err != -EEXIST))
 125                         break;
 126                 err = 0;
 127         }
 128
 129         return err;
 130 }
 131 EXPORT_SYMBOL_GPL(__add_pages);
 132
 133 static void grow_zone_span(struct zone *zone,
 134                 unsigned long start_pfn, unsigned long end_pfn)
 135 {
 136         unsigned long old_zone_end_pfn;
 137
 138         zone_span_writelock(zone);
 139
 140         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 141         if (start_pfn < zone->zone_start_pfn)
 142                 zone->zone_start_pfn = start_pfn;
 143
 144         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 145                                 zone->zone_start_pfn;
 146
 147         zone_span_writeunlock(zone);
 148 }
 149
 150 static void grow_pgdat_span(struct pglist_data *pgdat,
 151                 unsigned long start_pfn, unsigned long end_pfn)
 152 {
 153         unsigned long old_pgdat_end_pfn =
 154                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 155
 156         if (start_pfn < pgdat->node_start_pfn)
 157                 pgdat->node_start_pfn = start_pfn;
 158
 159         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 160                                         pgdat->node_start_pfn;
 161 }
 162
 163 int online_pages(unsigned long pfn, unsigned long nr_pages)
 164 {
 165         unsigned long i;
 166         unsigned long flags;
 167         unsigned long onlined_pages = 0;
 168         struct resource res;
 169         u64 section_end;
 170         unsigned long start_pfn;
 171         struct zone *zone;
 172         int need_zonelists_rebuild = 0;
 173
 174         /*
 175          * This doesn't need a lock to do pfn_to_page().
 176          * The section can't be removed here because of the
 177          * memory_block->state_sem.
 178          */
 179         zone = page_zone(pfn_to_page(pfn));
 180         pgdat_resize_lock(zone->zone_pgdat, &flags);
 181         grow_zone_span(zone, pfn, pfn + nr_pages);
 182         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
 183         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 184
 185         /*
 186          * If this zone is not populated, then it is not in zonelist.
 187          * This means the page allocator ignores this zone.
 188          * So, zonelist must be updated after online.
 189          */
 190         if (!populated_zone(zone))
 191                 need_zonelists_rebuild = 1;
 192
 193         res.start = (u64)pfn << PAGE_SHIFT;
 194         res.end = res.start + ((u64)nr_pages << PAGE_SHIFT) - 1;
 195         res.flags = IORESOURCE_MEM; /* we just need system ram */
 196         section_end = res.end;
 197
 198         while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
 199                 start_pfn = (unsigned long)(res.start >> PAGE_SHIFT);
 200                 nr_pages = (unsigned long)
 201                            ((res.end + 1 - res.start) >> PAGE_SHIFT);
 202
 203                 if (PageReserved(pfn_to_page(start_pfn))) {
 204                         /* this region's page is not onlined now */
 205                         for (i = 0; i < nr_pages; i++) {
 206                                 struct page *page = pfn_to_page(start_pfn + i);
 207                                 online_page(page);
 208                                 onlined_pages++;
 209                         }
 210                 }
 211
 212                 res.start = res.end + 1;
 213                 res.end = section_end;
 214         }
 215         zone->present_pages += onlined_pages;
 216         zone->zone_pgdat->node_present_pages += onlined_pages;
 217
 218         setup_per_zone_pages_min();
 219
 220         if (need_zonelists_rebuild)
 221                 build_all_zonelists();
 222         vm_total_pages = nr_free_pagecache_pages();
 223         writeback_set_ratelimit();
 224         return 0;
 225 }
 226 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 227
 228 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 229 {
 230         struct pglist_data *pgdat;
 231         unsigned long zones_size[MAX_NR_ZONES] = {0};
 232         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 233         unsigned long start_pfn = start >> PAGE_SHIFT;
 234
 235         pgdat = arch_alloc_nodedata(nid);
 236         if (!pgdat)
 237                 return NULL;
 238
 239         arch_refresh_nodedata(nid, pgdat);
 240
 241         /* we can use NODE_DATA(nid) from here */
 242
 243         /* init node's zones as empty zones, we don't have any present pages.*/
 244         free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
 245
 246         return pgdat;
 247 }
 248
 249 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 250 {
 251         arch_refresh_nodedata(nid, NULL);
 252         arch_free_nodedata(pgdat);
 253         return;
 254 }
 255
 256
 257 int add_memory(int nid, u64 start, u64 size)
 258 {
 259         pg_data_t *pgdat = NULL;
 260         int new_pgdat = 0;
 261         struct resource *res;
 262         int ret;
 263
 264         res = register_memory_resource(start, size);
 265         if (!res)
 266                 return -EEXIST;
 267
 268         if (!node_online(nid)) {
 269                 pgdat = hotadd_new_pgdat(nid, start);
 270                 if (!pgdat)
 271                         return -ENOMEM;
 272                 new_pgdat = 1;
 273                 ret = kswapd_run(nid);
 274                 if (ret)
 275                         goto error;
 276         }
 277
 278         /* call arch's memory hotadd */
 279         ret = arch_add_memory(nid, start, size);
 280
 281         if (ret < 0)
 282                 goto error;
 283
 284         /* we online node here. we can't roll back from here. */
 285         node_set_online(nid);
 286
 287         cpuset_track_online_nodes();
 288
 289         if (new_pgdat) {
 290                 ret = register_one_node(nid);
 291                 /*
 292                  * If sysfs file of new node can't create, cpu on the node
 293                  * can't be hot-added. There is no rollback way now.
 294                  * So, check by BUG_ON() to catch it reluctantly..
 295                  */
 296                 BUG_ON(ret);
 297         }
 298
 299         return ret;
 300 error:
 301         /* rollback pgdat allocation and others */
 302         if (new_pgdat)
 303                 rollback_node_hotadd(nid, pgdat);
 304         if (res)
 305                 release_memory_resource(res);
 306
 307         return ret;
 308 }
 309 EXPORT_SYMBOL_GPL(add_memory);