arch/i386/mm/discontig.c

   1 /*
   2  * Written by: Patricia Gaughen <gone@us.ibm.com>, IBM Corporation
   3  * August 2002: added remote node KVA remap - Martin J. Bligh
   4  *
   5  * Copyright (C) 2002, IBM Corp.
   6  *
   7  * All rights reserved.
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  17  * NON INFRINGEMENT.  See the GNU General Public License for more
  18  * details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23  */
  24
  25 #include <linux/config.h>
  26 #include <linux/mm.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/mmzone.h>
  29 #include <linux/highmem.h>
  30 #include <linux/initrd.h>
  31 #include <linux/nodemask.h>
  32 #include <asm/e820.h>
  33 #include <asm/setup.h>
  34 #include <asm/mmzone.h>
  35 #include <bios_ebda.h>
  36
  37 struct pglist_data *node_data[MAX_NUMNODES];
  38 bootmem_data_t node0_bdata;
  39
  40 /*
  41  * numa interface - we expect the numa architecture specfic code to have
  42  *                  populated the following initialisation.
  43  *
  44  * 1) node_online_map  - the map of all nodes configured (online) in the system
  45  * 2) physnode_map     - the mapping between a pfn and owning node
  46  * 3) node_start_pfn   - the starting page frame number for a node
  47  * 3) node_end_pfn     - the ending page fram number for a node
  48  */
  49
  50 /*
  51  * physnode_map keeps track of the physical memory layout of a generic
  52  * numa node on a 256Mb break (each element of the array will
  53  * represent 256Mb of memory and will be marked by the node id.  so,
  54  * if the first gig is on node 0, and the second gig is on node 1
  55  * physnode_map will contain:
  56  *
  57  *     physnode_map[0-3] = 0;
  58  *     physnode_map[4-7] = 1;
  59  *     physnode_map[8- ] = -1;
  60  */
  61 s8 physnode_map[MAX_ELEMENTS] = { [0 ... (MAX_ELEMENTS - 1)] = -1};
  62
  63 void memory_present(int nid, unsigned long start, unsigned long end)
  64 {
  65         unsigned long pfn;
  66
  67         printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
  68                         nid, start, end);
  69         printk(KERN_DEBUG "  Setting physnode_map array to node %d for pfns:\n", nid);
  70         printk(KERN_DEBUG "  ");
  71         for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
  72                 physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
  73                 printk("%ld ", pfn);
  74         }
  75         printk("\n");
  76 }
  77
  78 unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
  79                                               unsigned long end_pfn)
  80 {
  81         unsigned long nr_pages = end_pfn - start_pfn;
  82
  83         if (!nr_pages)
  84                 return 0;
  85
  86         return (nr_pages + 1) * sizeof(struct page);
  87 }
  88
  89 unsigned long node_start_pfn[MAX_NUMNODES];
  90 unsigned long node_end_pfn[MAX_NUMNODES];
  91
  92 extern unsigned long find_max_low_pfn(void);
  93 extern void find_max_pfn(void);
  94 extern void one_highpage_init(struct page *, int, int);
  95
  96 extern struct e820map e820;
  97 extern unsigned long init_pg_tables_end;
  98 extern unsigned long highend_pfn, highstart_pfn;
  99 extern unsigned long max_low_pfn;
 100 extern unsigned long totalram_pages;
 101 extern unsigned long totalhigh_pages;
 102
 103 #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
 104
 105 unsigned long node_remap_start_pfn[MAX_NUMNODES];
 106 unsigned long node_remap_size[MAX_NUMNODES];
 107 unsigned long node_remap_offset[MAX_NUMNODES];
 108 void *node_remap_start_vaddr[MAX_NUMNODES];
 109 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
 110
 111 /*
 112  * FLAT - support for basic PC memory model with discontig enabled, essentially
 113  *        a single node with all available processors in it with a flat
 114  *        memory map.
 115  */
 116 int __init get_memcfg_numa_flat(void)
 117 {
 118         printk("NUMA - single node, flat memory mode\n");
 119
 120         /* Run the memory configuration and find the top of memory. */
 121         find_max_pfn();
 122         node_start_pfn[0] = 0;
 123         node_end_pfn[0] = max_pfn;
 124         memory_present(0, 0, max_pfn);
 125
 126         /* Indicate there is one node available. */
 127         nodes_clear(node_online_map);
 128         node_set_online(0);
 129         return 1;
 130 }
 131
 132 /*
 133  * Find the highest page frame number we have available for the node
 134  */
 135 static void __init find_max_pfn_node(int nid)
 136 {
 137         if (node_end_pfn[nid] > max_pfn)
 138                 node_end_pfn[nid] = max_pfn;
 139         /*
 140          * if a user has given mem=XXXX, then we need to make sure
 141          * that the node _starts_ before that, too, not just ends
 142          */
 143         if (node_start_pfn[nid] > max_pfn)
 144                 node_start_pfn[nid] = max_pfn;
 145         if (node_start_pfn[nid] > node_end_pfn[nid])
 146                 BUG();
 147 }
 148
 149 /*
 150  * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
 151  * method.  For node zero take this from the bottom of memory, for
 152  * subsequent nodes place them at node_remap_start_vaddr which contains
 153  * node local data in physically node local memory.  See setup_memory()
 154  * for details.
 155  */
 156 static void __init allocate_pgdat(int nid)
 157 {
 158         if (nid && node_has_online_mem(nid))
 159                 NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
 160         else {
 161                 NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
 162                 min_low_pfn += PFN_UP(sizeof(pg_data_t));
 163         }
 164 }
 165
 166 void __init remap_numa_kva(void)
 167 {
 168         void *vaddr;
 169         unsigned long pfn;
 170         int node;
 171
 172         for_each_online_node(node) {
 173                 if (node == 0)
 174                         continue;
 175                 for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
 176                         vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
 177                         set_pmd_pfn((ulong) vaddr,
 178                                 node_remap_start_pfn[node] + pfn,
 179                                 PAGE_KERNEL_LARGE);
 180                 }
 181         }
 182 }
 183
 184 static unsigned long calculate_numa_remap_pages(void)
 185 {
 186         int nid;
 187         unsigned long size, reserve_pages = 0;
 188
 189         for_each_online_node(nid) {
 190                 if (nid == 0)
 191                         continue;
 192                 if (!node_remap_size[nid])
 193                         continue;
 194
 195                 /*
 196                  * The acpi/srat node info can show hot-add memroy zones
 197                  * where memory could be added but not currently present.
 198                  */
 199                 if (node_start_pfn[nid] > max_pfn)
 200                         continue;
 201                 if (node_end_pfn[nid] > max_pfn)
 202                         node_end_pfn[nid] = max_pfn;
 203
 204                 /* ensure the remap includes space for the pgdat. */
 205                 size = node_remap_size[nid] + sizeof(pg_data_t);
 206
 207                 /* convert size to large (pmd size) pages, rounding up */
 208                 size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
 209                 /* now the roundup is correct, convert to PAGE_SIZE pages */
 210                 size = size * PTRS_PER_PTE;
 211                 printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
 212                                 size, nid);
 213                 node_remap_size[nid] = size;
 214                 reserve_pages += size;
 215                 node_remap_offset[nid] = reserve_pages;
 216                 printk("Shrinking node %d from %ld pages to %ld pages\n",
 217                         nid, node_end_pfn[nid], node_end_pfn[nid] - size);
 218                 node_end_pfn[nid] -= size;
 219                 node_remap_start_pfn[nid] = node_end_pfn[nid];
 220         }
 221         printk("Reserving total of %ld pages for numa KVA remap\n",
 222                         reserve_pages);
 223         return reserve_pages;
 224 }
 225
 226 extern void setup_bootmem_allocator(void);
 227 unsigned long __init setup_memory(void)
 228 {
 229         int nid;
 230         unsigned long system_start_pfn, system_max_low_pfn;
 231         unsigned long reserve_pages;
 232
 233         /*
 234          * When mapping a NUMA machine we allocate the node_mem_map arrays
 235          * from node local memory.  They are then mapped directly into KVA
 236          * between zone normal and vmalloc space.  Calculate the size of
 237          * this space and use it to adjust the boundry between ZONE_NORMAL
 238          * and ZONE_HIGHMEM.
 239          */
 240         find_max_pfn();
 241         get_memcfg_numa();
 242
 243         reserve_pages = calculate_numa_remap_pages();
 244
 245         /* partially used pages are not usable - thus round upwards */
 246         system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
 247
 248         system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages;
 249         printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
 250                         reserve_pages, max_low_pfn + reserve_pages);
 251         printk("max_pfn = %ld\n", max_pfn);
 252 #ifdef CONFIG_HIGHMEM
 253         highstart_pfn = highend_pfn = max_pfn;
 254         if (max_pfn > system_max_low_pfn)
 255                 highstart_pfn = system_max_low_pfn;
 256         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 257                pages_to_mb(highend_pfn - highstart_pfn));
 258 #endif
 259         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 260                         pages_to_mb(system_max_low_pfn));
 261         printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
 262                         min_low_pfn, max_low_pfn, highstart_pfn);
 263
 264         printk("Low memory ends at vaddr %08lx\n",
 265                         (ulong) pfn_to_kaddr(max_low_pfn));
 266         for_each_online_node(nid) {
 267                 node_remap_start_vaddr[nid] = pfn_to_kaddr(
 268                         (highstart_pfn + reserve_pages) - node_remap_offset[nid]);
 269                 allocate_pgdat(nid);
 270                 printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
 271                         (ulong) node_remap_start_vaddr[nid],
 272                         (ulong) pfn_to_kaddr(highstart_pfn + reserve_pages
 273                             - node_remap_offset[nid] + node_remap_size[nid]));
 274         }
 275         printk("High memory starts at vaddr %08lx\n",
 276                         (ulong) pfn_to_kaddr(highstart_pfn));
 277         vmalloc_earlyreserve = reserve_pages * PAGE_SIZE;
 278         for_each_online_node(nid)
 279                 find_max_pfn_node(nid);
 280
 281         memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
 282         NODE_DATA(0)->bdata = &node0_bdata;
 283         setup_bootmem_allocator();
 284         return max_low_pfn;
 285 }
 286
 287 void __init zone_sizes_init(void)
 288 {
 289         int nid;
 290
 291         /*
 292          * Insert nodes into pgdat_list backward so they appear in order.
 293          * Clobber node 0's links and NULL out pgdat_list before starting.
 294          */
 295         pgdat_list = NULL;
 296         for (nid = MAX_NUMNODES - 1; nid >= 0; nid--) {
 297                 if (!node_online(nid))
 298                         continue;
 299                 NODE_DATA(nid)->pgdat_next = pgdat_list;
 300                 pgdat_list = NODE_DATA(nid);
 301         }
 302
 303         for_each_online_node(nid) {
 304                 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
 305                 unsigned long *zholes_size;
 306                 unsigned int max_dma;
 307
 308                 unsigned long low = max_low_pfn;
 309                 unsigned long start = node_start_pfn[nid];
 310                 unsigned long high = node_end_pfn[nid];
 311
 312                 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 313
 314                 if (node_has_online_mem(nid)){
 315                         if (start > low) {
 316 #ifdef CONFIG_HIGHMEM
 317                                 BUG_ON(start > high);
 318                                 zones_size[ZONE_HIGHMEM] = high - start;
 319 #endif
 320                         } else {
 321                                 if (low < max_dma)
 322                                         zones_size[ZONE_DMA] = low;
 323                                 else {
 324                                         BUG_ON(max_dma > low);
 325                                         BUG_ON(low > high);
 326                                         zones_size[ZONE_DMA] = max_dma;
 327                                         zones_size[ZONE_NORMAL] = low - max_dma;
 328 #ifdef CONFIG_HIGHMEM
 329                                         zones_size[ZONE_HIGHMEM] = high - low;
 330 #endif
 331                                 }
 332                         }
 333                 }
 334
 335                 zholes_size = get_zholes_size(nid);
 336                 /*
 337                  * We let the lmem_map for node 0 be allocated from the
 338                  * normal bootmem allocator, but other nodes come from the
 339                  * remapped KVA area - mbligh
 340                  */
 341                 if (!nid)
 342                         free_area_init_node(nid, NODE_DATA(nid),
 343                                         zones_size, start, zholes_size);
 344                 else {
 345                         unsigned long lmem_map;
 346                         lmem_map = (unsigned long)node_remap_start_vaddr[nid];
 347                         lmem_map += sizeof(pg_data_t) + PAGE_SIZE - 1;
 348                         lmem_map &= PAGE_MASK;
 349                         NODE_DATA(nid)->node_mem_map = (struct page *)lmem_map;
 350                         free_area_init_node(nid, NODE_DATA(nid), zones_size,
 351                                 start, zholes_size);
 352                 }
 353         }
 354         return;
 355 }
 356
 357 void __init set_highmem_pages_init(int bad_ppro)
 358 {
 359 #ifdef CONFIG_HIGHMEM
 360         struct zone *zone;
 361
 362         for_each_zone(zone) {
 363                 unsigned long node_pfn, node_high_size, zone_start_pfn;
 364                 struct page * zone_mem_map;
 365
 366                 if (!is_highmem(zone))
 367                         continue;
 368
 369                 printk("Initializing %s for node %d\n", zone->name,
 370                         zone->zone_pgdat->node_id);
 371
 372                 node_high_size = zone->spanned_pages;
 373                 zone_mem_map = zone->zone_mem_map;
 374                 zone_start_pfn = zone->zone_start_pfn;
 375
 376                 for (node_pfn = 0; node_pfn < node_high_size; node_pfn++) {
 377                         one_highpage_init((struct page *)(zone_mem_map + node_pfn),
 378                                           zone_start_pfn + node_pfn, bad_ppro);
 379                 }
 380         }
 381         totalram_pages += totalhigh_pages;
 382 #endif
 383 }