arch/i386/kernel/setup.c

   1 /*
   2  *  linux/arch/i386/kernel/setup.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *
   6  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   7  *
   8  *  Memory region support
   9  *      David Parsons <orc@pell.chi.il.us>, July-August 1999
  10  *
  11  *  Added E820 sanitization routine (removes overlapping memory regions);
  12  *  Brian Moyle <bmoyle@mvista.com>, February 2001
  13  *
  14  * Moved CPU detection code to cpu/${cpu}.c
  15  *    Patrick Mochel <mochel@osdl.org>, March 2002
  16  *
  17  *  Provisions for empty E820 memory regions (reported by certain BIOSes).
  18  *  Alex Achenbach <xela@slit.de>, December 2002.
  19  *
  20  */
  21
  22 /*
  23  * This file handles the architecture-dependent parts of initialization
  24  */
  25
  26 #include <linux/sched.h>
  27 #include <linux/mm.h>
  28 #include <linux/mmzone.h>
  29 #include <linux/screen_info.h>
  30 #include <linux/ioport.h>
  31 #include <linux/acpi.h>
  32 #include <linux/apm_bios.h>
  33 #include <linux/initrd.h>
  34 #include <linux/bootmem.h>
  35 #include <linux/seq_file.h>
  36 #include <linux/platform_device.h>
  37 #include <linux/console.h>
  38 #include <linux/mca.h>
  39 #include <linux/root_dev.h>
  40 #include <linux/highmem.h>
  41 #include <linux/module.h>
  42 #include <linux/efi.h>
  43 #include <linux/init.h>
  44 #include <linux/edd.h>
  45 #include <linux/nodemask.h>
  46 #include <linux/kexec.h>
  47 #include <linux/crash_dump.h>
  48 #include <linux/dmi.h>
  49 #include <linux/pfn.h>
  50
  51 #include <video/edid.h>
  52
  53 #include <asm/apic.h>
  54 #include <asm/e820.h>
  55 #include <asm/mpspec.h>
  56 #include <asm/setup.h>
  57 #include <asm/arch_hooks.h>
  58 #include <asm/sections.h>
  59 #include <asm/io_apic.h>
  60 #include <asm/ist.h>
  61 #include <asm/io.h>
  62 #include <setup_arch.h>
  63 #include <bios_ebda.h>
  64
  65 /* Forward Declaration. */
  66 void __init find_max_pfn(void);
  67
  68 /* This value is set up by the early boot code to point to the value
  69    immediately after the boot time page tables.  It contains a *physical*
  70    address, and must not be in the .bss segment! */
  71 unsigned long init_pg_tables_end __initdata = ~0UL;
  72
  73 int disable_pse __devinitdata = 0;
  74
  75 /*
  76  * Machine setup..
  77  */
  78
  79 #ifdef CONFIG_EFI
  80 int efi_enabled = 0;
  81 EXPORT_SYMBOL(efi_enabled);
  82 #endif
  83
  84 /* cpu data as detected by the assembly code in head.S */
  85 struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
  86 /* common cpu data for all cpus */
  87 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
  88 EXPORT_SYMBOL(boot_cpu_data);
  89
  90 unsigned long mmu_cr4_features;
  91
  92 #ifdef  CONFIG_ACPI
  93         int acpi_disabled = 0;
  94 #else
  95         int acpi_disabled = 1;
  96 #endif
  97 EXPORT_SYMBOL(acpi_disabled);
  98
  99 #ifdef  CONFIG_ACPI
 100 int __initdata acpi_force = 0;
 101 extern acpi_interrupt_flags     acpi_sci_flags;
 102 #endif
 103
 104 /* for MCA, but anyone else can use it if they want */
 105 unsigned int machine_id;
 106 #ifdef CONFIG_MCA
 107 EXPORT_SYMBOL(machine_id);
 108 #endif
 109 unsigned int machine_submodel_id;
 110 unsigned int BIOS_revision;
 111 unsigned int mca_pentium_flag;
 112
 113 /* For PCI or other memory-mapped resources */
 114 unsigned long pci_mem_start = 0x10000000;
 115 #ifdef CONFIG_PCI
 116 EXPORT_SYMBOL(pci_mem_start);
 117 #endif
 118
 119 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 120 int bootloader_type;
 121
 122 /* user-defined highmem size */
 123 static unsigned int highmem_pages = -1;
 124
 125 /*
 126  * Setup options
 127  */
 128 struct drive_info_struct { char dummy[32]; } drive_info;
 129 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
 130     defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
 131 EXPORT_SYMBOL(drive_info);
 132 #endif
 133 struct screen_info screen_info;
 134 EXPORT_SYMBOL(screen_info);
 135 struct apm_info apm_info;
 136 EXPORT_SYMBOL(apm_info);
 137 struct sys_desc_table_struct {
 138         unsigned short length;
 139         unsigned char table[0];
 140 };
 141 struct edid_info edid_info;
 142 EXPORT_SYMBOL_GPL(edid_info);
 143 struct ist_info ist_info;
 144 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 145         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 146 EXPORT_SYMBOL(ist_info);
 147 #endif
 148 struct e820map e820;
 149
 150 extern void early_cpu_init(void);
 151 extern void generic_apic_probe(char *);
 152 extern int root_mountflags;
 153
 154 unsigned long saved_videomode;
 155
 156 #define RAMDISK_IMAGE_START_MASK        0x07FF
 157 #define RAMDISK_PROMPT_FLAG             0x8000
 158 #define RAMDISK_LOAD_FLAG               0x4000
 159
 160 static char command_line[COMMAND_LINE_SIZE];
 161
 162 unsigned char __initdata boot_params[PARAM_SIZE];
 163
 164 static struct resource data_resource = {
 165         .name   = "Kernel data",
 166         .start  = 0,
 167         .end    = 0,
 168         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 169 };
 170
 171 static struct resource code_resource = {
 172         .name   = "Kernel code",
 173         .start  = 0,
 174         .end    = 0,
 175         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 176 };
 177
 178 static struct resource system_rom_resource = {
 179         .name   = "System ROM",
 180         .start  = 0xf0000,
 181         .end    = 0xfffff,
 182         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 183 };
 184
 185 static struct resource extension_rom_resource = {
 186         .name   = "Extension ROM",
 187         .start  = 0xe0000,
 188         .end    = 0xeffff,
 189         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 190 };
 191
 192 static struct resource adapter_rom_resources[] = { {
 193         .name   = "Adapter ROM",
 194         .start  = 0xc8000,
 195         .end    = 0,
 196         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 197 }, {
 198         .name   = "Adapter ROM",
 199         .start  = 0,
 200         .end    = 0,
 201         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 202 }, {
 203         .name   = "Adapter ROM",
 204         .start  = 0,
 205         .end    = 0,
 206         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 207 }, {
 208         .name   = "Adapter ROM",
 209         .start  = 0,
 210         .end    = 0,
 211         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 212 }, {
 213         .name   = "Adapter ROM",
 214         .start  = 0,
 215         .end    = 0,
 216         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 217 }, {
 218         .name   = "Adapter ROM",
 219         .start  = 0,
 220         .end    = 0,
 221         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 222 } };
 223
 224 #define ADAPTER_ROM_RESOURCES \
 225         (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
 226
 227 static struct resource video_rom_resource = {
 228         .name   = "Video ROM",
 229         .start  = 0xc0000,
 230         .end    = 0xc7fff,
 231         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 232 };
 233
 234 static struct resource video_ram_resource = {
 235         .name   = "Video RAM area",
 236         .start  = 0xa0000,
 237         .end    = 0xbffff,
 238         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 239 };
 240
 241 static struct resource standard_io_resources[] = { {
 242         .name   = "dma1",
 243         .start  = 0x0000,
 244         .end    = 0x001f,
 245         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 246 }, {
 247         .name   = "pic1",
 248         .start  = 0x0020,
 249         .end    = 0x0021,
 250         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 251 }, {
 252         .name   = "timer0",
 253         .start  = 0x0040,
 254         .end    = 0x0043,
 255         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 256 }, {
 257         .name   = "timer1",
 258         .start  = 0x0050,
 259         .end    = 0x0053,
 260         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 261 }, {
 262         .name   = "keyboard",
 263         .start  = 0x0060,
 264         .end    = 0x006f,
 265         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 266 }, {
 267         .name   = "dma page reg",
 268         .start  = 0x0080,
 269         .end    = 0x008f,
 270         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 271 }, {
 272         .name   = "pic2",
 273         .start  = 0x00a0,
 274         .end    = 0x00a1,
 275         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 276 }, {
 277         .name   = "dma2",
 278         .start  = 0x00c0,
 279         .end    = 0x00df,
 280         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 281 }, {
 282         .name   = "fpu",
 283         .start  = 0x00f0,
 284         .end    = 0x00ff,
 285         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 286 } };
 287
 288 #define STANDARD_IO_RESOURCES \
 289         (sizeof standard_io_resources / sizeof standard_io_resources[0])
 290
 291 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
 292
 293 static int __init romchecksum(unsigned char *rom, unsigned long length)
 294 {
 295         unsigned char *p, sum = 0;
 296
 297         for (p = rom; p < rom + length; p++)
 298                 sum += *p;
 299         return sum == 0;
 300 }
 301
 302 static void __init probe_roms(void)
 303 {
 304         unsigned long start, length, upper;
 305         unsigned char *rom;
 306         int           i;
 307
 308         /* video rom */
 309         upper = adapter_rom_resources[0].start;
 310         for (start = video_rom_resource.start; start < upper; start += 2048) {
 311                 rom = isa_bus_to_virt(start);
 312                 if (!romsignature(rom))
 313                         continue;
 314
 315                 video_rom_resource.start = start;
 316
 317                 /* 0 < length <= 0x7f * 512, historically */
 318                 length = rom[2] * 512;
 319
 320                 /* if checksum okay, trust length byte */
 321                 if (length && romchecksum(rom, length))
 322                         video_rom_resource.end = start + length - 1;
 323
 324                 request_resource(&iomem_resource, &video_rom_resource);
 325                 break;
 326         }
 327
 328         start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
 329         if (start < upper)
 330                 start = upper;
 331
 332         /* system rom */
 333         request_resource(&iomem_resource, &system_rom_resource);
 334         upper = system_rom_resource.start;
 335
 336         /* check for extension rom (ignore length byte!) */
 337         rom = isa_bus_to_virt(extension_rom_resource.start);
 338         if (romsignature(rom)) {
 339                 length = extension_rom_resource.end - extension_rom_resource.start + 1;
 340                 if (romchecksum(rom, length)) {
 341                         request_resource(&iomem_resource, &extension_rom_resource);
 342                         upper = extension_rom_resource.start;
 343                 }
 344         }
 345
 346         /* check for adapter roms on 2k boundaries */
 347         for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
 348                 rom = isa_bus_to_virt(start);
 349                 if (!romsignature(rom))
 350                         continue;
 351
 352                 /* 0 < length <= 0x7f * 512, historically */
 353                 length = rom[2] * 512;
 354
 355                 /* but accept any length that fits if checksum okay */
 356                 if (!length || start + length > upper || !romchecksum(rom, length))
 357                         continue;
 358
 359                 adapter_rom_resources[i].start = start;
 360                 adapter_rom_resources[i].end = start + length - 1;
 361                 request_resource(&iomem_resource, &adapter_rom_resources[i]);
 362
 363                 start = adapter_rom_resources[i++].end & ~2047UL;
 364         }
 365 }
 366
 367 static void __init limit_regions(unsigned long long size)
 368 {
 369         unsigned long long current_addr = 0;
 370         int i;
 371
 372         if (efi_enabled) {
 373                 efi_memory_desc_t *md;
 374                 void *p;
 375
 376                 for (p = memmap.map, i = 0; p < memmap.map_end;
 377                         p += memmap.desc_size, i++) {
 378                         md = p;
 379                         current_addr = md->phys_addr + (md->num_pages << 12);
 380                         if (md->type == EFI_CONVENTIONAL_MEMORY) {
 381                                 if (current_addr >= size) {
 382                                         md->num_pages -=
 383                                                 (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT);
 384                                         memmap.nr_map = i + 1;
 385                                         return;
 386                                 }
 387                         }
 388                 }
 389         }
 390         for (i = 0; i < e820.nr_map; i++) {
 391                 current_addr = e820.map[i].addr + e820.map[i].size;
 392                 if (current_addr < size)
 393                         continue;
 394
 395                 if (e820.map[i].type != E820_RAM)
 396                         continue;
 397
 398                 if (e820.map[i].addr >= size) {
 399                         /*
 400                          * This region starts past the end of the
 401                          * requested size, skip it completely.
 402                          */
 403                         e820.nr_map = i;
 404                 } else {
 405                         e820.nr_map = i + 1;
 406                         e820.map[i].size -= current_addr - size;
 407                 }
 408                 return;
 409         }
 410 }
 411
 412 void __init add_memory_region(unsigned long long start,
 413                               unsigned long long size, int type)
 414 {
 415         int x;
 416
 417         if (!efi_enabled) {
 418                 x = e820.nr_map;
 419
 420                 if (x == E820MAX) {
 421                     printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 422                     return;
 423                 }
 424
 425                 e820.map[x].addr = start;
 426                 e820.map[x].size = size;
 427                 e820.map[x].type = type;
 428                 e820.nr_map++;
 429         }
 430 } /* add_memory_region */
 431
 432 #define E820_DEBUG      1
 433
 434 static void __init print_memory_map(char *who)
 435 {
 436         int i;
 437
 438         for (i = 0; i < e820.nr_map; i++) {
 439                 printk(" %s: %016Lx - %016Lx ", who,
 440                         e820.map[i].addr,
 441                         e820.map[i].addr + e820.map[i].size);
 442                 switch (e820.map[i].type) {
 443                 case E820_RAM:  printk("(usable)\n");
 444                                 break;
 445                 case E820_RESERVED:
 446                                 printk("(reserved)\n");
 447                                 break;
 448                 case E820_ACPI:
 449                                 printk("(ACPI data)\n");
 450                                 break;
 451                 case E820_NVS:
 452                                 printk("(ACPI NVS)\n");
 453                                 break;
 454                 default:        printk("type %lu\n", e820.map[i].type);
 455                                 break;
 456                 }
 457         }
 458 }
 459
 460 /*
 461  * Sanitize the BIOS e820 map.
 462  *
 463  * Some e820 responses include overlapping entries.  The following
 464  * replaces the original e820 map with a new one, removing overlaps.
 465  *
 466  */
 467 struct change_member {
 468         struct e820entry *pbios; /* pointer to original bios entry */
 469         unsigned long long addr; /* address for this change point */
 470 };
 471 static struct change_member change_point_list[2*E820MAX] __initdata;
 472 static struct change_member *change_point[2*E820MAX] __initdata;
 473 static struct e820entry *overlap_list[E820MAX] __initdata;
 474 static struct e820entry new_bios[E820MAX] __initdata;
 475
 476 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
 477 {
 478         struct change_member *change_tmp;
 479         unsigned long current_type, last_type;
 480         unsigned long long last_addr;
 481         int chgidx, still_changing;
 482         int overlap_entries;
 483         int new_bios_entry;
 484         int old_nr, new_nr, chg_nr;
 485         int i;
 486
 487         /*
 488                 Visually we're performing the following (1,2,3,4 = memory types)...
 489
 490                 Sample memory map (w/overlaps):
 491                    ____22__________________
 492                    ______________________4_
 493                    ____1111________________
 494                    _44_____________________
 495                    11111111________________
 496                    ____________________33__
 497                    ___________44___________
 498                    __________33333_________
 499                    ______________22________
 500                    ___________________2222_
 501                    _________111111111______
 502                    _____________________11_
 503                    _________________4______
 504
 505                 Sanitized equivalent (no overlap):
 506                    1_______________________
 507                    _44_____________________
 508                    ___1____________________
 509                    ____22__________________
 510                    ______11________________
 511                    _________1______________
 512                    __________3_____________
 513                    ___________44___________
 514                    _____________33_________
 515                    _______________2________
 516                    ________________1_______
 517                    _________________4______
 518                    ___________________2____
 519                    ____________________33__
 520                    ______________________4_
 521         */
 522
 523         /* if there's only one memory region, don't bother */
 524         if (*pnr_map < 2)
 525                 return -1;
 526
 527         old_nr = *pnr_map;
 528
 529         /* bail out if we find any unreasonable addresses in bios map */
 530         for (i=0; i<old_nr; i++)
 531                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 532                         return -1;
 533
 534         /* create pointers for initial change-point information (for sorting) */
 535         for (i=0; i < 2*old_nr; i++)
 536                 change_point[i] = &change_point_list[i];
 537
 538         /* record all known change-points (starting and ending addresses),
 539            omitting those that are for empty memory regions */
 540         chgidx = 0;
 541         for (i=0; i < old_nr; i++)      {
 542                 if (biosmap[i].size != 0) {
 543                         change_point[chgidx]->addr = biosmap[i].addr;
 544                         change_point[chgidx++]->pbios = &biosmap[i];
 545                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
 546                         change_point[chgidx++]->pbios = &biosmap[i];
 547                 }
 548         }
 549         chg_nr = chgidx;        /* true number of change-points */
 550
 551         /* sort change-point list by memory addresses (low -> high) */
 552         still_changing = 1;
 553         while (still_changing)  {
 554                 still_changing = 0;
 555                 for (i=1; i < chg_nr; i++)  {
 556                         /* if <current_addr> > <last_addr>, swap */
 557                         /* or, if current=<start_addr> & last=<end_addr>, swap */
 558                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
 559                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
 560                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
 561                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
 562                            )
 563                         {
 564                                 change_tmp = change_point[i];
 565                                 change_point[i] = change_point[i-1];
 566                                 change_point[i-1] = change_tmp;
 567                                 still_changing=1;
 568                         }
 569                 }
 570         }
 571
 572         /* create a new bios memory map, removing overlaps */
 573         overlap_entries=0;       /* number of entries in the overlap table */
 574         new_bios_entry=0;        /* index for creating new bios map entries */
 575         last_type = 0;           /* start with undefined memory type */
 576         last_addr = 0;           /* start with 0 as last starting address */
 577         /* loop through change-points, determining affect on the new bios map */
 578         for (chgidx=0; chgidx < chg_nr; chgidx++)
 579         {
 580                 /* keep track of all overlapping bios entries */
 581                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
 582                 {
 583                         /* add map entry to overlap list (> 1 entry implies an overlap) */
 584                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
 585                 }
 586                 else
 587                 {
 588                         /* remove entry from list (order independent, so swap with last) */
 589                         for (i=0; i<overlap_entries; i++)
 590                         {
 591                                 if (overlap_list[i] == change_point[chgidx]->pbios)
 592                                         overlap_list[i] = overlap_list[overlap_entries-1];
 593                         }
 594                         overlap_entries--;
 595                 }
 596                 /* if there are overlapping entries, decide which "type" to use */
 597                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
 598                 current_type = 0;
 599                 for (i=0; i<overlap_entries; i++)
 600                         if (overlap_list[i]->type > current_type)
 601                                 current_type = overlap_list[i]->type;
 602                 /* continue building up new bios map based on this information */
 603                 if (current_type != last_type)  {
 604                         if (last_type != 0)      {
 605                                 new_bios[new_bios_entry].size =
 606                                         change_point[chgidx]->addr - last_addr;
 607                                 /* move forward only if the new size was non-zero */
 608                                 if (new_bios[new_bios_entry].size != 0)
 609                                         if (++new_bios_entry >= E820MAX)
 610                                                 break;  /* no more space left for new bios entries */
 611                         }
 612                         if (current_type != 0)  {
 613                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
 614                                 new_bios[new_bios_entry].type = current_type;
 615                                 last_addr=change_point[chgidx]->addr;
 616                         }
 617                         last_type = current_type;
 618                 }
 619         }
 620         new_nr = new_bios_entry;   /* retain count for new bios entries */
 621
 622         /* copy new bios mapping into original location */
 623         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
 624         *pnr_map = new_nr;
 625
 626         return 0;
 627 }
 628
 629 /*
 630  * Copy the BIOS e820 map into a safe place.
 631  *
 632  * Sanity-check it while we're at it..
 633  *
 634  * If we're lucky and live on a modern system, the setup code
 635  * will have given us a memory map that we can use to properly
 636  * set up memory.  If we aren't, we'll fake a memory map.
 637  *
 638  * We check to see that the memory map contains at least 2 elements
 639  * before we'll use it, because the detection code in setup.S may
 640  * not be perfect and most every PC known to man has two memory
 641  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
 642  * thinkpad 560x, for example, does not cooperate with the memory
 643  * detection code.)
 644  */
 645 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
 646 {
 647         /* Only one memory region (or negative)? Ignore it */
 648         if (nr_map < 2)
 649                 return -1;
 650
 651         do {
 652                 unsigned long long start = biosmap->addr;
 653                 unsigned long long size = biosmap->size;
 654                 unsigned long long end = start + size;
 655                 unsigned long type = biosmap->type;
 656
 657                 /* Overflow in 64 bits? Ignore the memory map. */
 658                 if (start > end)
 659                         return -1;
 660
 661                 /*
 662                  * Some BIOSes claim RAM in the 640k - 1M region.
 663                  * Not right. Fix it up.
 664                  */
 665                 if (type == E820_RAM) {
 666                         if (start < 0x100000ULL && end > 0xA0000ULL) {
 667                                 if (start < 0xA0000ULL)
 668                                         add_memory_region(start, 0xA0000ULL-start, type);
 669                                 if (end <= 0x100000ULL)
 670                                         continue;
 671                                 start = 0x100000ULL;
 672                                 size = end - start;
 673                         }
 674                 }
 675                 add_memory_region(start, size, type);
 676         } while (biosmap++,--nr_map);
 677         return 0;
 678 }
 679
 680 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 681 struct edd edd;
 682 #ifdef CONFIG_EDD_MODULE
 683 EXPORT_SYMBOL(edd);
 684 #endif
 685 /**
 686  * copy_edd() - Copy the BIOS EDD information
 687  *              from boot_params into a safe place.
 688  *
 689  */
 690 static inline void copy_edd(void)
 691 {
 692      memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
 693      memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
 694      edd.mbr_signature_nr = EDD_MBR_SIG_NR;
 695      edd.edd_info_nr = EDD_NR;
 696 }
 697 #else
 698 static inline void copy_edd(void)
 699 {
 700 }
 701 #endif
 702
 703 static void __init parse_cmdline_early (char ** cmdline_p)
 704 {
 705         char c = ' ', *to = command_line, *from = saved_command_line;
 706         int len = 0;
 707         int userdef = 0;
 708
 709         /* Save unparsed command line copy for /proc/cmdline */
 710         saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
 711
 712         for (;;) {
 713                 if (c != ' ')
 714                         goto next_char;
 715                 /*
 716                  * "mem=nopentium" disables the 4MB page tables.
 717                  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
 718                  * to <mem>, overriding the bios size.
 719                  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
 720                  * <start> to <start>+<mem>, overriding the bios size.
 721                  *
 722                  * HPA tells me bootloaders need to parse mem=, so no new
 723                  * option should be mem=  [also see Documentation/i386/boot.txt]
 724                  */
 725                 if (!memcmp(from, "mem=", 4)) {
 726                         if (to != command_line)
 727                                 to--;
 728                         if (!memcmp(from+4, "nopentium", 9)) {
 729                                 from += 9+4;
 730                                 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
 731                                 disable_pse = 1;
 732                         } else {
 733                                 /* If the user specifies memory size, we
 734                                  * limit the BIOS-provided memory map to
 735                                  * that size. exactmap can be used to specify
 736                                  * the exact map. mem=number can be used to
 737                                  * trim the existing memory map.
 738                                  */
 739                                 unsigned long long mem_size;
 740
 741                                 mem_size = memparse(from+4, &from);
 742                                 limit_regions(mem_size);
 743                                 userdef=1;
 744                         }
 745                 }
 746
 747                 else if (!memcmp(from, "memmap=", 7)) {
 748                         if (to != command_line)
 749                                 to--;
 750                         if (!memcmp(from+7, "exactmap", 8)) {
 751 #ifdef CONFIG_CRASH_DUMP
 752                                 /* If we are doing a crash dump, we
 753                                  * still need to know the real mem
 754                                  * size before original memory map is
 755                                  * reset.
 756                                  */
 757                                 find_max_pfn();
 758                                 saved_max_pfn = max_pfn;
 759 #endif
 760                                 from += 8+7;
 761                                 e820.nr_map = 0;
 762                                 userdef = 1;
 763                         } else {
 764                                 /* If the user specifies memory size, we
 765                                  * limit the BIOS-provided memory map to
 766                                  * that size. exactmap can be used to specify
 767                                  * the exact map. mem=number can be used to
 768                                  * trim the existing memory map.
 769                                  */
 770                                 unsigned long long start_at, mem_size;
 771
 772                                 mem_size = memparse(from+7, &from);
 773                                 if (*from == '@') {
 774                                         start_at = memparse(from+1, &from);
 775                                         add_memory_region(start_at, mem_size, E820_RAM);
 776                                 } else if (*from == '#') {
 777                                         start_at = memparse(from+1, &from);
 778                                         add_memory_region(start_at, mem_size, E820_ACPI);
 779                                 } else if (*from == '$') {
 780                                         start_at = memparse(from+1, &from);
 781                                         add_memory_region(start_at, mem_size, E820_RESERVED);
 782                                 } else {
 783                                         limit_regions(mem_size);
 784                                         userdef=1;
 785                                 }
 786                         }
 787                 }
 788
 789                 else if (!memcmp(from, "noexec=", 7))
 790                         noexec_setup(from + 7);
 791
 792
 793 #ifdef  CONFIG_X86_SMP
 794                 /*
 795                  * If the BIOS enumerates physical processors before logical,
 796                  * maxcpus=N at enumeration-time can be used to disable HT.
 797                  */
 798                 else if (!memcmp(from, "maxcpus=", 8)) {
 799                         extern unsigned int maxcpus;
 800
 801                         maxcpus = simple_strtoul(from + 8, NULL, 0);
 802                 }
 803 #endif
 804
 805 #ifdef CONFIG_ACPI
 806                 /* "acpi=off" disables both ACPI table parsing and interpreter */
 807                 else if (!memcmp(from, "acpi=off", 8)) {
 808                         disable_acpi();
 809                 }
 810
 811                 /* acpi=force to over-ride black-list */
 812                 else if (!memcmp(from, "acpi=force", 10)) {
 813                         acpi_force = 1;
 814                         acpi_ht = 1;
 815                         acpi_disabled = 0;
 816                 }
 817
 818                 /* acpi=strict disables out-of-spec workarounds */
 819                 else if (!memcmp(from, "acpi=strict", 11)) {
 820                         acpi_strict = 1;
 821                 }
 822
 823                 /* Limit ACPI just to boot-time to enable HT */
 824                 else if (!memcmp(from, "acpi=ht", 7)) {
 825                         if (!acpi_force)
 826                                 disable_acpi();
 827                         acpi_ht = 1;
 828                 }
 829
 830                 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
 831                 else if (!memcmp(from, "pci=noacpi", 10)) {
 832                         acpi_disable_pci();
 833                 }
 834                 /* "acpi=noirq" disables ACPI interrupt routing */
 835                 else if (!memcmp(from, "acpi=noirq", 10)) {
 836                         acpi_noirq_set();
 837                 }
 838
 839                 else if (!memcmp(from, "acpi_sci=edge", 13))
 840                         acpi_sci_flags.trigger =  1;
 841
 842                 else if (!memcmp(from, "acpi_sci=level", 14))
 843                         acpi_sci_flags.trigger = 3;
 844
 845                 else if (!memcmp(from, "acpi_sci=high", 13))
 846                         acpi_sci_flags.polarity = 1;
 847
 848                 else if (!memcmp(from, "acpi_sci=low", 12))
 849                         acpi_sci_flags.polarity = 3;
 850
 851 #ifdef CONFIG_X86_IO_APIC
 852                 else if (!memcmp(from, "acpi_skip_timer_override", 24))
 853                         acpi_skip_timer_override = 1;
 854
 855                 if (!memcmp(from, "disable_timer_pin_1", 19))
 856                         disable_timer_pin_1 = 1;
 857                 if (!memcmp(from, "enable_timer_pin_1", 18))
 858                         disable_timer_pin_1 = -1;
 859
 860                 /* disable IO-APIC */
 861                 else if (!memcmp(from, "noapic", 6))
 862                         disable_ioapic_setup();
 863 #endif /* CONFIG_X86_IO_APIC */
 864 #endif /* CONFIG_ACPI */
 865
 866 #ifdef CONFIG_X86_LOCAL_APIC
 867                 /* enable local APIC */
 868                 else if (!memcmp(from, "lapic", 5))
 869                         lapic_enable();
 870
 871                 /* disable local APIC */
 872                 else if (!memcmp(from, "nolapic", 6))
 873                         lapic_disable();
 874 #endif /* CONFIG_X86_LOCAL_APIC */
 875
 876 #ifdef CONFIG_KEXEC
 877                 /* crashkernel=size@addr specifies the location to reserve for
 878                  * a crash kernel.  By reserving this memory we guarantee
 879                  * that linux never set's it up as a DMA target.
 880                  * Useful for holding code to do something appropriate
 881                  * after a kernel panic.
 882                  */
 883                 else if (!memcmp(from, "crashkernel=", 12)) {
 884                         unsigned long size, base;
 885                         size = memparse(from+12, &from);
 886                         if (*from == '@') {
 887                                 base = memparse(from+1, &from);
 888                                 /* FIXME: Do I want a sanity check
 889                                  * to validate the memory range?
 890                                  */
 891                                 crashk_res.start = base;
 892                                 crashk_res.end   = base + size - 1;
 893                         }
 894                 }
 895 #endif
 896 #ifdef CONFIG_PROC_VMCORE
 897                 /* elfcorehdr= specifies the location of elf core header
 898                  * stored by the crashed kernel.
 899                  */
 900                 else if (!memcmp(from, "elfcorehdr=", 11))
 901                         elfcorehdr_addr = memparse(from+11, &from);
 902 #endif
 903
 904                 /*
 905                  * highmem=size forces highmem to be exactly 'size' bytes.
 906                  * This works even on boxes that have no highmem otherwise.
 907                  * This also works to reduce highmem size on bigger boxes.
 908                  */
 909                 else if (!memcmp(from, "highmem=", 8))
 910                         highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;
 911
 912                 /*
 913                  * vmalloc=size forces the vmalloc area to be exactly 'size'
 914                  * bytes. This can be used to increase (or decrease) the
 915                  * vmalloc area - the default is 128m.
 916                  */
 917                 else if (!memcmp(from, "vmalloc=", 8))
 918                         __VMALLOC_RESERVE = memparse(from+8, &from);
 919
 920         next_char:
 921                 c = *(from++);
 922                 if (!c)
 923                         break;
 924                 if (COMMAND_LINE_SIZE <= ++len)
 925                         break;
 926                 *(to++) = c;
 927         }
 928         *to = '\0';
 929         *cmdline_p = command_line;
 930         if (userdef) {
 931                 printk(KERN_INFO "user-defined physical RAM map:\n");
 932                 print_memory_map("user");
 933         }
 934 }
 935
 936 /*
 937  * Callback for efi_memory_walk.
 938  */
 939 static int __init
 940 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
 941 {
 942         unsigned long *max_pfn = arg, pfn;
 943
 944         if (start < end) {
 945                 pfn = PFN_UP(end -1);
 946                 if (pfn > *max_pfn)
 947                         *max_pfn = pfn;
 948         }
 949         return 0;
 950 }
 951
 952 static int __init
 953 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
 954 {
 955         memory_present(0, start, end);
 956         return 0;
 957 }
 958
 959  /*
 960   * This function checks if the entire range <start,end> is mapped with type.
 961   *
 962   * Note: this function only works correct if the e820 table is sorted and
 963   * not-overlapping, which is the case
 964   */
 965 int __init
 966 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
 967 {
 968         u64 start = s;
 969         u64 end = e;
 970         int i;
 971         for (i = 0; i < e820.nr_map; i++) {
 972                 struct e820entry *ei = &e820.map[i];
 973                 if (type && ei->type != type)
 974                         continue;
 975                 /* is the region (part) in overlap with the current region ?*/
 976                 if (ei->addr >= end || ei->addr + ei->size <= start)
 977                         continue;
 978                 /* if the region is at the beginning of <start,end> we move
 979                  * start to the end of the region since it's ok until there
 980                  */
 981                 if (ei->addr <= start)
 982                         start = ei->addr + ei->size;
 983                 /* if start is now at or beyond end, we're done, full
 984                  * coverage */
 985                 if (start >= end)
 986                         return 1; /* we're done */
 987         }
 988         return 0;
 989 }
 990
 991 /*
 992  * Find the highest page frame number we have available
 993  */
 994 void __init find_max_pfn(void)
 995 {
 996         int i;
 997
 998         max_pfn = 0;
 999         if (efi_enabled) {
1000                 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
1001                 efi_memmap_walk(efi_memory_present_wrapper, NULL);
1002                 return;
1003         }
1004
1005         for (i = 0; i < e820.nr_map; i++) {
1006                 unsigned long start, end;
1007                 /* RAM? */
1008                 if (e820.map[i].type != E820_RAM)
1009                         continue;
1010                 start = PFN_UP(e820.map[i].addr);
1011                 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1012                 if (start >= end)
1013                         continue;
1014                 if (end > max_pfn)
1015                         max_pfn = end;
1016                 memory_present(0, start, end);
1017         }
1018 }
1019
1020 /*
1021  * Determine low and high memory ranges:
1022  */
1023 unsigned long __init find_max_low_pfn(void)
1024 {
1025         unsigned long max_low_pfn;
1026
1027         max_low_pfn = max_pfn;
1028         if (max_low_pfn > MAXMEM_PFN) {
1029                 if (highmem_pages == -1)
1030                         highmem_pages = max_pfn - MAXMEM_PFN;
1031                 if (highmem_pages + MAXMEM_PFN < max_pfn)
1032                         max_pfn = MAXMEM_PFN + highmem_pages;
1033                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
1034                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
1035                         highmem_pages = 0;
1036                 }
1037                 max_low_pfn = MAXMEM_PFN;
1038 #ifndef CONFIG_HIGHMEM
1039                 /* Maximum memory usable is what is directly addressable */
1040                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
1041                                         MAXMEM>>20);
1042                 if (max_pfn > MAX_NONPAE_PFN)
1043                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1044                 else
1045                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
1046                 max_pfn = MAXMEM_PFN;
1047 #else /* !CONFIG_HIGHMEM */
1048 #ifndef CONFIG_X86_PAE
1049                 if (max_pfn > MAX_NONPAE_PFN) {
1050                         max_pfn = MAX_NONPAE_PFN;
1051                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
1052                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1053                 }
1054 #endif /* !CONFIG_X86_PAE */
1055 #endif /* !CONFIG_HIGHMEM */
1056         } else {
1057                 if (highmem_pages == -1)
1058                         highmem_pages = 0;
1059 #ifdef CONFIG_HIGHMEM
1060                 if (highmem_pages >= max_pfn) {
1061                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
1062                         highmem_pages = 0;
1063                 }
1064                 if (highmem_pages) {
1065                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
1066                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
1067                                 highmem_pages = 0;
1068                         }
1069                         max_low_pfn -= highmem_pages;
1070                 }
1071 #else
1072                 if (highmem_pages)
1073                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
1074 #endif
1075         }
1076         return max_low_pfn;
1077 }
1078
1079 /*
1080  * Free all available memory for boot time allocation.  Used
1081  * as a callback function by efi_memory_walk()
1082  */
1083
1084 static int __init
1085 free_available_memory(unsigned long start, unsigned long end, void *arg)
1086 {
1087         /* check max_low_pfn */
1088         if (start >= (max_low_pfn << PAGE_SHIFT))
1089                 return 0;
1090         if (end >= (max_low_pfn << PAGE_SHIFT))
1091                 end = max_low_pfn << PAGE_SHIFT;
1092         if (start < end)
1093                 free_bootmem(start, end - start);
1094
1095         return 0;
1096 }
1097 /*
1098  * Register fully available low RAM pages with the bootmem allocator.
1099  */
1100 static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
1101 {
1102         int i;
1103
1104         if (efi_enabled) {
1105                 efi_memmap_walk(free_available_memory, NULL);
1106                 return;
1107         }
1108         for (i = 0; i < e820.nr_map; i++) {
1109                 unsigned long curr_pfn, last_pfn, size;
1110                 /*
1111                  * Reserve usable low memory
1112                  */
1113                 if (e820.map[i].type != E820_RAM)
1114                         continue;
1115                 /*
1116                  * We are rounding up the start address of usable memory:
1117                  */
1118                 curr_pfn = PFN_UP(e820.map[i].addr);
1119                 if (curr_pfn >= max_low_pfn)
1120                         continue;
1121                 /*
1122                  * ... and at the end of the usable range downwards:
1123                  */
1124                 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1125
1126                 if (last_pfn > max_low_pfn)
1127                         last_pfn = max_low_pfn;
1128
1129                 /*
1130                  * .. finally, did all the rounding and playing
1131                  * around just make the area go away?
1132                  */
1133                 if (last_pfn <= curr_pfn)
1134                         continue;
1135
1136                 size = last_pfn - curr_pfn;
1137                 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1138         }
1139 }
1140
1141 /*
1142  * workaround for Dell systems that neglect to reserve EBDA
1143  */
1144 static void __init reserve_ebda_region(void)
1145 {
1146         unsigned int addr;
1147         addr = get_bios_ebda();
1148         if (addr)
1149                 reserve_bootmem(addr, PAGE_SIZE);
1150 }
1151
1152 #ifndef CONFIG_NEED_MULTIPLE_NODES
1153 void __init setup_bootmem_allocator(void);
1154 static unsigned long __init setup_memory(void)
1155 {
1156         /*
1157          * partially used pages are not usable - thus
1158          * we are rounding upwards:
1159          */
1160         min_low_pfn = PFN_UP(init_pg_tables_end);
1161
1162         find_max_pfn();
1163
1164         max_low_pfn = find_max_low_pfn();
1165
1166 #ifdef CONFIG_HIGHMEM
1167         highstart_pfn = highend_pfn = max_pfn;
1168         if (max_pfn > max_low_pfn) {
1169                 highstart_pfn = max_low_pfn;
1170         }
1171         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1172                 pages_to_mb(highend_pfn - highstart_pfn));
1173 #endif
1174         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1175                         pages_to_mb(max_low_pfn));
1176
1177         setup_bootmem_allocator();
1178
1179         return max_low_pfn;
1180 }
1181
1182 void __init zone_sizes_init(void)
1183 {
1184         unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
1185         unsigned int max_dma, low;
1186
1187         max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1188         low = max_low_pfn;
1189
1190         if (low < max_dma)
1191                 zones_size[ZONE_DMA] = low;
1192         else {
1193                 zones_size[ZONE_DMA] = max_dma;
1194                 zones_size[ZONE_NORMAL] = low - max_dma;
1195 #ifdef CONFIG_HIGHMEM
1196                 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1197 #endif
1198         }
1199         free_area_init(zones_size);
1200 }
1201 #else
1202 extern unsigned long __init setup_memory(void);
1203 extern void zone_sizes_init(void);
1204 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1205
1206 void __init setup_bootmem_allocator(void)
1207 {
1208         unsigned long bootmap_size;
1209         /*
1210          * Initialize the boot-time allocator (with low memory only):
1211          */
1212         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1213
1214         register_bootmem_low_pages(max_low_pfn);
1215
1216         /*
1217          * Reserve the bootmem bitmap itself as well. We do this in two
1218          * steps (first step was init_bootmem()) because this catches
1219          * the (very unlikely) case of us accidentally initializing the
1220          * bootmem allocator with an invalid RAM area.
1221          */
1222         reserve_bootmem(__PHYSICAL_START, (PFN_PHYS(min_low_pfn) +
1223                          bootmap_size + PAGE_SIZE-1) - (__PHYSICAL_START));
1224
1225         /*
1226          * reserve physical page 0 - it's a special BIOS page on many boxes,
1227          * enabling clean reboots, SMP operation, laptop functions.
1228          */
1229         reserve_bootmem(0, PAGE_SIZE);
1230
1231         /* reserve EBDA region, it's a 4K region */
1232         reserve_ebda_region();
1233
1234     /* could be an AMD 768MPX chipset. Reserve a page  before VGA to prevent
1235        PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1236        unless you have no PS/2 mouse plugged in. */
1237         if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1238             boot_cpu_data.x86 == 6)
1239              reserve_bootmem(0xa0000 - 4096, 4096);
1240
1241 #ifdef CONFIG_SMP
1242         /*
1243          * But first pinch a few for the stack/trampoline stuff
1244          * FIXME: Don't need the extra page at 4K, but need to fix
1245          * trampoline before removing it. (see the GDT stuff)
1246          */
1247         reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1248 #endif
1249 #ifdef CONFIG_ACPI_SLEEP
1250         /*
1251          * Reserve low memory region for sleep support.
1252          */
1253         acpi_reserve_bootmem();
1254 #endif
1255 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1256         /*
1257          * Find and reserve possible boot-time SMP configuration:
1258          */
1259         find_smp_config();
1260 #endif
1261
1262 #ifdef CONFIG_BLK_DEV_INITRD
1263         if (LOADER_TYPE && INITRD_START) {
1264                 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1265                         reserve_bootmem(INITRD_START, INITRD_SIZE);
1266                         initrd_start =
1267                                 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1268                         initrd_end = initrd_start+INITRD_SIZE;
1269                 }
1270                 else {
1271                         printk(KERN_ERR "initrd extends beyond end of memory "
1272                             "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1273                             INITRD_START + INITRD_SIZE,
1274                             max_low_pfn << PAGE_SHIFT);
1275                         initrd_start = 0;
1276                 }
1277         }
1278 #endif
1279 #ifdef CONFIG_KEXEC
1280         if (crashk_res.start != crashk_res.end)
1281                 reserve_bootmem(crashk_res.start,
1282                         crashk_res.end - crashk_res.start + 1);
1283 #endif
1284 }
1285
1286 /*
1287  * The node 0 pgdat is initialized before all of these because
1288  * it's needed for bootmem.  node>0 pgdats have their virtual
1289  * space allocated before the pagetables are in place to access
1290  * them, so they can't be cleared then.
1291  *
1292  * This should all compile down to nothing when NUMA is off.
1293  */
1294 void __init remapped_pgdat_init(void)
1295 {
1296         int nid;
1297
1298         for_each_online_node(nid) {
1299                 if (nid != 0)
1300                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1301         }
1302 }
1303
1304 /*
1305  * Request address space for all standard RAM and ROM resources
1306  * and also for regions reported as reserved by the e820.
1307  */
1308 static void __init
1309 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1310 {
1311         int i;
1312
1313         probe_roms();
1314         for (i = 0; i < e820.nr_map; i++) {
1315                 struct resource *res;
1316 #ifndef CONFIG_RESOURCES_64BIT
1317                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1318                         continue;
1319 #endif
1320                 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
1321                 switch (e820.map[i].type) {
1322                 case E820_RAM:  res->name = "System RAM"; break;
1323                 case E820_ACPI: res->name = "ACPI Tables"; break;
1324                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
1325                 default:        res->name = "reserved";
1326                 }
1327                 res->start = e820.map[i].addr;
1328                 res->end = res->start + e820.map[i].size - 1;
1329                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1330                 if (request_resource(&iomem_resource, res)) {
1331                         kfree(res);
1332                         continue;
1333                 }
1334                 if (e820.map[i].type == E820_RAM) {
1335                         /*
1336                          *  We don't know which RAM region contains kernel data,
1337                          *  so we try it repeatedly and let the resource manager
1338                          *  test it.
1339                          */
1340                         request_resource(res, code_resource);
1341                         request_resource(res, data_resource);
1342 #ifdef CONFIG_KEXEC
1343                         request_resource(res, &crashk_res);
1344 #endif
1345                 }
1346         }
1347 }
1348
1349 /*
1350  * Request address space for all standard resources
1351  *
1352  * This is called just before pcibios_init(), which is also a
1353  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
1354  */
1355 static int __init request_standard_resources(void)
1356 {
1357         int i;
1358
1359         printk("Setting up standard PCI resources\n");
1360         if (efi_enabled)
1361                 efi_initialize_iomem_resources(&code_resource, &data_resource);
1362         else
1363                 legacy_init_iomem_resources(&code_resource, &data_resource);
1364
1365         /* EFI systems may still have VGA */
1366         request_resource(&iomem_resource, &video_ram_resource);
1367
1368         /* request I/O space for devices used on all i[345]86 PCs */
1369         for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1370                 request_resource(&ioport_resource, &standard_io_resources[i]);
1371         return 0;
1372 }
1373
1374 subsys_initcall(request_standard_resources);
1375
1376 static void __init register_memory(void)
1377 {
1378         unsigned long gapstart, gapsize, round;
1379         unsigned long long last;
1380         int i;
1381
1382         /*
1383          * Search for the bigest gap in the low 32 bits of the e820
1384          * memory space.
1385          */
1386         last = 0x100000000ull;
1387         gapstart = 0x10000000;
1388         gapsize = 0x400000;
1389         i = e820.nr_map;
1390         while (--i >= 0) {
1391                 unsigned long long start = e820.map[i].addr;
1392                 unsigned long long end = start + e820.map[i].size;
1393
1394                 /*
1395                  * Since "last" is at most 4GB, we know we'll
1396                  * fit in 32 bits if this condition is true
1397                  */
1398                 if (last > end) {
1399                         unsigned long gap = last - end;
1400
1401                         if (gap > gapsize) {
1402                                 gapsize = gap;
1403                                 gapstart = end;
1404                         }
1405                 }
1406                 if (start < last)
1407                         last = start;
1408         }
1409
1410         /*
1411          * See how much we want to round up: start off with
1412          * rounding to the next 1MB area.
1413          */
1414         round = 0x100000;
1415         while ((gapsize >> 4) > round)
1416                 round += round;
1417         /* Fun with two's complement */
1418         pci_mem_start = (gapstart + round) & -round;
1419
1420         printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1421                 pci_mem_start, gapstart, gapsize);
1422 }
1423
1424 #ifdef CONFIG_MCA
1425 static void set_mca_bus(int x)
1426 {
1427         MCA_bus = x;
1428 }
1429 #else
1430 static void set_mca_bus(int x) { }
1431 #endif
1432
1433 /*
1434  * Determine if we were loaded by an EFI loader.  If so, then we have also been
1435  * passed the efi memmap, systab, etc., so we should use these data structures
1436  * for initialization.  Note, the efi init code path is determined by the
1437  * global efi_enabled. This allows the same kernel image to be used on existing
1438  * systems (with a traditional BIOS) as well as on EFI systems.
1439  */
1440 void __init setup_arch(char **cmdline_p)
1441 {
1442         unsigned long max_low_pfn;
1443
1444         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1445         pre_setup_arch_hook();
1446         early_cpu_init();
1447
1448         /*
1449          * FIXME: This isn't an official loader_type right
1450          * now but does currently work with elilo.
1451          * If we were configured as an EFI kernel, check to make
1452          * sure that we were loaded correctly from elilo and that
1453          * the system table is valid.  If not, then initialize normally.
1454          */
1455 #ifdef CONFIG_EFI
1456         if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1457                 efi_enabled = 1;
1458 #endif
1459
1460         ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1461         drive_info = DRIVE_INFO;
1462         screen_info = SCREEN_INFO;
1463         edid_info = EDID_INFO;
1464         apm_info.bios = APM_BIOS_INFO;
1465         ist_info = IST_INFO;
1466         saved_videomode = VIDEO_MODE;
1467         if( SYS_DESC_TABLE.length != 0 ) {
1468                 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1469                 machine_id = SYS_DESC_TABLE.table[0];
1470                 machine_submodel_id = SYS_DESC_TABLE.table[1];
1471                 BIOS_revision = SYS_DESC_TABLE.table[2];
1472         }
1473         bootloader_type = LOADER_TYPE;
1474
1475 #ifdef CONFIG_BLK_DEV_RAM
1476         rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1477         rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1478         rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1479 #endif
1480         ARCH_SETUP
1481         if (efi_enabled)
1482                 efi_init();
1483         else {
1484                 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1485                 print_memory_map(machine_specific_memory_setup());
1486         }
1487
1488         copy_edd();
1489
1490         if (!MOUNT_ROOT_RDONLY)
1491                 root_mountflags &= ~MS_RDONLY;
1492         init_mm.start_code = (unsigned long) _text;
1493         init_mm.end_code = (unsigned long) _etext;
1494         init_mm.end_data = (unsigned long) _edata;
1495         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1496
1497         code_resource.start = virt_to_phys(_text);
1498         code_resource.end = virt_to_phys(_etext)-1;
1499         data_resource.start = virt_to_phys(_etext);
1500         data_resource.end = virt_to_phys(_edata)-1;
1501
1502         parse_cmdline_early(cmdline_p);
1503
1504 #ifdef CONFIG_EARLY_PRINTK
1505         {
1506                 char *s = strstr(*cmdline_p, "earlyprintk=");
1507                 if (s) {
1508                         setup_early_printk(strchr(s, '=') + 1);
1509                         printk("early console enabled\n");
1510                 }
1511         }
1512 #endif
1513
1514         max_low_pfn = setup_memory();
1515
1516         /*
1517          * NOTE: before this point _nobody_ is allowed to allocate
1518          * any memory using the bootmem allocator.  Although the
1519          * alloctor is now initialised only the first 8Mb of the kernel
1520          * virtual address space has been mapped.  All allocations before
1521          * paging_init() has completed must use the alloc_bootmem_low_pages()
1522          * variant (which allocates DMA'able memory) and care must be taken
1523          * not to exceed the 8Mb limit.
1524          */
1525
1526 #ifdef CONFIG_SMP
1527         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1528 #endif
1529         paging_init();
1530         remapped_pgdat_init();
1531         sparse_init();
1532         zone_sizes_init();
1533
1534         /*
1535          * NOTE: at this point the bootmem allocator is fully available.
1536          */
1537
1538         dmi_scan_machine();
1539
1540 #ifdef CONFIG_X86_GENERICARCH
1541         generic_apic_probe(*cmdline_p);
1542 #endif
1543         if (efi_enabled)
1544                 efi_map_memmap();
1545
1546 #ifdef CONFIG_ACPI
1547         /*
1548          * Parse the ACPI tables for possible boot-time SMP configuration.
1549          */
1550         acpi_boot_table_init();
1551 #endif
1552
1553 #ifdef CONFIG_X86_IO_APIC
1554         check_acpi_pci();       /* Checks more than just ACPI actually */
1555 #endif
1556
1557 #ifdef CONFIG_ACPI
1558         acpi_boot_init();
1559
1560 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1561         if (def_to_bigsmp)
1562                 printk(KERN_WARNING "More than 8 CPUs detected and "
1563                         "CONFIG_X86_PC cannot handle it.\nUse "
1564                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1565 #endif
1566 #endif
1567 #ifdef CONFIG_X86_LOCAL_APIC
1568         if (smp_found_config)
1569                 get_smp_config();
1570 #endif
1571
1572         register_memory();
1573
1574 #ifdef CONFIG_VT
1575 #if defined(CONFIG_VGA_CONSOLE)
1576         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1577                 conswitchp = &vga_con;
1578 #elif defined(CONFIG_DUMMY_CONSOLE)
1579         conswitchp = &dummy_con;
1580 #endif
1581 #endif
1582         tsc_init();
1583 }
1584
1585 static __init int add_pcspkr(void)
1586 {
1587         struct platform_device *pd;
1588         int ret;
1589
1590         pd = platform_device_alloc("pcspkr", -1);
1591         if (!pd)
1592                 return -ENOMEM;
1593
1594         ret = platform_device_add(pd);
1595         if (ret)
1596                 platform_device_put(pd);
1597
1598         return ret;
1599 }
1600 device_initcall(add_pcspkr);
1601
1602 /*
1603  * Local Variables:
1604  * mode:c
1605  * c-file-style:"k&r"
1606  * c-basic-offset:8
1607  * End:
1608  */