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/mmzone.h>
  57 #include <asm/setup.h>
  58 #include <asm/arch_hooks.h>
  59 #include <asm/sections.h>
  60 #include <asm/io_apic.h>
  61 #include <asm/ist.h>
  62 #include <asm/io.h>
  63 #include <setup_arch.h>
  64 #include <bios_ebda.h>
  65
  66 /* Forward Declaration. */
  67 void __init find_max_pfn(void);
  68
  69 /* This value is set up by the early boot code to point to the value
  70    immediately after the boot time page tables.  It contains a *physical*
  71    address, and must not be in the .bss segment! */
  72 unsigned long init_pg_tables_end __initdata = ~0UL;
  73
  74 int disable_pse __devinitdata = 0;
  75
  76 /*
  77  * Machine setup..
  78  */
  79
  80 #ifdef CONFIG_EFI
  81 int efi_enabled = 0;
  82 EXPORT_SYMBOL(efi_enabled);
  83 #endif
  84
  85 /* cpu data as detected by the assembly code in head.S */
  86 struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
  87 /* common cpu data for all cpus */
  88 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
  89 EXPORT_SYMBOL(boot_cpu_data);
  90
  91 unsigned long mmu_cr4_features;
  92
  93 #ifdef  CONFIG_ACPI
  94         int acpi_disabled = 0;
  95 #else
  96         int acpi_disabled = 1;
  97 #endif
  98 EXPORT_SYMBOL(acpi_disabled);
  99
 100 #ifdef  CONFIG_ACPI
 101 int __initdata acpi_force = 0;
 102 extern acpi_interrupt_flags     acpi_sci_flags;
 103 #endif
 104
 105 /* for MCA, but anyone else can use it if they want */
 106 unsigned int machine_id;
 107 #ifdef CONFIG_MCA
 108 EXPORT_SYMBOL(machine_id);
 109 #endif
 110 unsigned int machine_submodel_id;
 111 unsigned int BIOS_revision;
 112 unsigned int mca_pentium_flag;
 113
 114 /* For PCI or other memory-mapped resources */
 115 unsigned long pci_mem_start = 0x10000000;
 116 #ifdef CONFIG_PCI
 117 EXPORT_SYMBOL(pci_mem_start);
 118 #endif
 119
 120 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 121 int bootloader_type;
 122
 123 /* user-defined highmem size */
 124 static unsigned int highmem_pages = -1;
 125
 126 /*
 127  * Setup options
 128  */
 129 struct drive_info_struct { char dummy[32]; } drive_info;
 130 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
 131     defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
 132 EXPORT_SYMBOL(drive_info);
 133 #endif
 134 struct screen_info screen_info;
 135 EXPORT_SYMBOL(screen_info);
 136 struct apm_info apm_info;
 137 EXPORT_SYMBOL(apm_info);
 138 struct sys_desc_table_struct {
 139         unsigned short length;
 140         unsigned char table[0];
 141 };
 142 struct edid_info edid_info;
 143 EXPORT_SYMBOL_GPL(edid_info);
 144 struct ist_info ist_info;
 145 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 146         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 147 EXPORT_SYMBOL(ist_info);
 148 #endif
 149 struct e820map e820;
 150
 151 extern void early_cpu_init(void);
 152 extern void generic_apic_probe(char *);
 153 extern int root_mountflags;
 154
 155 unsigned long saved_videomode;
 156
 157 #define RAMDISK_IMAGE_START_MASK        0x07FF
 158 #define RAMDISK_PROMPT_FLAG             0x8000
 159 #define RAMDISK_LOAD_FLAG               0x4000
 160
 161 static char command_line[COMMAND_LINE_SIZE];
 162
 163 unsigned char __initdata boot_params[PARAM_SIZE];
 164
 165 static struct resource data_resource = {
 166         .name   = "Kernel data",
 167         .start  = 0,
 168         .end    = 0,
 169         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 170 };
 171
 172 static struct resource code_resource = {
 173         .name   = "Kernel code",
 174         .start  = 0,
 175         .end    = 0,
 176         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 177 };
 178
 179 static struct resource system_rom_resource = {
 180         .name   = "System ROM",
 181         .start  = 0xf0000,
 182         .end    = 0xfffff,
 183         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 184 };
 185
 186 static struct resource extension_rom_resource = {
 187         .name   = "Extension ROM",
 188         .start  = 0xe0000,
 189         .end    = 0xeffff,
 190         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 191 };
 192
 193 static struct resource adapter_rom_resources[] = { {
 194         .name   = "Adapter ROM",
 195         .start  = 0xc8000,
 196         .end    = 0,
 197         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 198 }, {
 199         .name   = "Adapter ROM",
 200         .start  = 0,
 201         .end    = 0,
 202         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 203 }, {
 204         .name   = "Adapter ROM",
 205         .start  = 0,
 206         .end    = 0,
 207         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 208 }, {
 209         .name   = "Adapter ROM",
 210         .start  = 0,
 211         .end    = 0,
 212         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 213 }, {
 214         .name   = "Adapter ROM",
 215         .start  = 0,
 216         .end    = 0,
 217         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 218 }, {
 219         .name   = "Adapter ROM",
 220         .start  = 0,
 221         .end    = 0,
 222         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 223 } };
 224
 225 #define ADAPTER_ROM_RESOURCES \
 226         (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
 227
 228 static struct resource video_rom_resource = {
 229         .name   = "Video ROM",
 230         .start  = 0xc0000,
 231         .end    = 0xc7fff,
 232         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 233 };
 234
 235 static struct resource video_ram_resource = {
 236         .name   = "Video RAM area",
 237         .start  = 0xa0000,
 238         .end    = 0xbffff,
 239         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 240 };
 241
 242 static struct resource standard_io_resources[] = { {
 243         .name   = "dma1",
 244         .start  = 0x0000,
 245         .end    = 0x001f,
 246         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 247 }, {
 248         .name   = "pic1",
 249         .start  = 0x0020,
 250         .end    = 0x0021,
 251         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 252 }, {
 253         .name   = "timer0",
 254         .start  = 0x0040,
 255         .end    = 0x0043,
 256         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 257 }, {
 258         .name   = "timer1",
 259         .start  = 0x0050,
 260         .end    = 0x0053,
 261         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 262 }, {
 263         .name   = "keyboard",
 264         .start  = 0x0060,
 265         .end    = 0x006f,
 266         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 267 }, {
 268         .name   = "dma page reg",
 269         .start  = 0x0080,
 270         .end    = 0x008f,
 271         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 272 }, {
 273         .name   = "pic2",
 274         .start  = 0x00a0,
 275         .end    = 0x00a1,
 276         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 277 }, {
 278         .name   = "dma2",
 279         .start  = 0x00c0,
 280         .end    = 0x00df,
 281         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 282 }, {
 283         .name   = "fpu",
 284         .start  = 0x00f0,
 285         .end    = 0x00ff,
 286         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 287 } };
 288
 289 #define STANDARD_IO_RESOURCES \
 290         (sizeof standard_io_resources / sizeof standard_io_resources[0])
 291
 292 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
 293
 294 static int __init romchecksum(unsigned char *rom, unsigned long length)
 295 {
 296         unsigned char *p, sum = 0;
 297
 298         for (p = rom; p < rom + length; p++)
 299                 sum += *p;
 300         return sum == 0;
 301 }
 302
 303 static void __init probe_roms(void)
 304 {
 305         unsigned long start, length, upper;
 306         unsigned char *rom;
 307         int           i;
 308
 309         /* video rom */
 310         upper = adapter_rom_resources[0].start;
 311         for (start = video_rom_resource.start; start < upper; start += 2048) {
 312                 rom = isa_bus_to_virt(start);
 313                 if (!romsignature(rom))
 314                         continue;
 315
 316                 video_rom_resource.start = start;
 317
 318                 /* 0 < length <= 0x7f * 512, historically */
 319                 length = rom[2] * 512;
 320
 321                 /* if checksum okay, trust length byte */
 322                 if (length && romchecksum(rom, length))
 323                         video_rom_resource.end = start + length - 1;
 324
 325                 request_resource(&iomem_resource, &video_rom_resource);
 326                 break;
 327         }
 328
 329         start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
 330         if (start < upper)
 331                 start = upper;
 332
 333         /* system rom */
 334         request_resource(&iomem_resource, &system_rom_resource);
 335         upper = system_rom_resource.start;
 336
 337         /* check for extension rom (ignore length byte!) */
 338         rom = isa_bus_to_virt(extension_rom_resource.start);
 339         if (romsignature(rom)) {
 340                 length = extension_rom_resource.end - extension_rom_resource.start + 1;
 341                 if (romchecksum(rom, length)) {
 342                         request_resource(&iomem_resource, &extension_rom_resource);
 343                         upper = extension_rom_resource.start;
 344                 }
 345         }
 346
 347         /* check for adapter roms on 2k boundaries */
 348         for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
 349                 rom = isa_bus_to_virt(start);
 350                 if (!romsignature(rom))
 351                         continue;
 352
 353                 /* 0 < length <= 0x7f * 512, historically */
 354                 length = rom[2] * 512;
 355
 356                 /* but accept any length that fits if checksum okay */
 357                 if (!length || start + length > upper || !romchecksum(rom, length))
 358                         continue;
 359
 360                 adapter_rom_resources[i].start = start;
 361                 adapter_rom_resources[i].end = start + length - 1;
 362                 request_resource(&iomem_resource, &adapter_rom_resources[i]);
 363
 364                 start = adapter_rom_resources[i++].end & ~2047UL;
 365         }
 366 }
 367
 368 static void __init limit_regions(unsigned long long size)
 369 {
 370         unsigned long long current_addr = 0;
 371         int i;
 372
 373         if (efi_enabled) {
 374                 efi_memory_desc_t *md;
 375                 void *p;
 376
 377                 for (p = memmap.map, i = 0; p < memmap.map_end;
 378                         p += memmap.desc_size, i++) {
 379                         md = p;
 380                         current_addr = md->phys_addr + (md->num_pages << 12);
 381                         if (md->type == EFI_CONVENTIONAL_MEMORY) {
 382                                 if (current_addr >= size) {
 383                                         md->num_pages -=
 384                                                 (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT);
 385                                         memmap.nr_map = i + 1;
 386                                         return;
 387                                 }
 388                         }
 389                 }
 390         }
 391         for (i = 0; i < e820.nr_map; i++) {
 392                 current_addr = e820.map[i].addr + e820.map[i].size;
 393                 if (current_addr < size)
 394                         continue;
 395
 396                 if (e820.map[i].type != E820_RAM)
 397                         continue;
 398
 399                 if (e820.map[i].addr >= size) {
 400                         /*
 401                          * This region starts past the end of the
 402                          * requested size, skip it completely.
 403                          */
 404                         e820.nr_map = i;
 405                 } else {
 406                         e820.nr_map = i + 1;
 407                         e820.map[i].size -= current_addr - size;
 408                 }
 409                 return;
 410         }
 411 }
 412
 413 void __init add_memory_region(unsigned long long start,
 414                               unsigned long long size, int type)
 415 {
 416         int x;
 417
 418         if (!efi_enabled) {
 419                 x = e820.nr_map;
 420
 421                 if (x == E820MAX) {
 422                     printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 423                     return;
 424                 }
 425
 426                 e820.map[x].addr = start;
 427                 e820.map[x].size = size;
 428                 e820.map[x].type = type;
 429                 e820.nr_map++;
 430         }
 431 } /* add_memory_region */
 432
 433 #define E820_DEBUG      1
 434
 435 static void __init print_memory_map(char *who)
 436 {
 437         int i;
 438
 439         for (i = 0; i < e820.nr_map; i++) {
 440                 printk(" %s: %016Lx - %016Lx ", who,
 441                         e820.map[i].addr,
 442                         e820.map[i].addr + e820.map[i].size);
 443                 switch (e820.map[i].type) {
 444                 case E820_RAM:  printk("(usable)\n");
 445                                 break;
 446                 case E820_RESERVED:
 447                                 printk("(reserved)\n");
 448                                 break;
 449                 case E820_ACPI:
 450                                 printk("(ACPI data)\n");
 451                                 break;
 452                 case E820_NVS:
 453                                 printk("(ACPI NVS)\n");
 454                                 break;
 455                 default:        printk("type %lu\n", e820.map[i].type);
 456                                 break;
 457                 }
 458         }
 459 }
 460
 461 /*
 462  * Sanitize the BIOS e820 map.
 463  *
 464  * Some e820 responses include overlapping entries.  The following
 465  * replaces the original e820 map with a new one, removing overlaps.
 466  *
 467  */
 468 struct change_member {
 469         struct e820entry *pbios; /* pointer to original bios entry */
 470         unsigned long long addr; /* address for this change point */
 471 };
 472 static struct change_member change_point_list[2*E820MAX] __initdata;
 473 static struct change_member *change_point[2*E820MAX] __initdata;
 474 static struct e820entry *overlap_list[E820MAX] __initdata;
 475 static struct e820entry new_bios[E820MAX] __initdata;
 476
 477 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
 478 {
 479         struct change_member *change_tmp;
 480         unsigned long current_type, last_type;
 481         unsigned long long last_addr;
 482         int chgidx, still_changing;
 483         int overlap_entries;
 484         int new_bios_entry;
 485         int old_nr, new_nr, chg_nr;
 486         int i;
 487
 488         /*
 489                 Visually we're performing the following (1,2,3,4 = memory types)...
 490
 491                 Sample memory map (w/overlaps):
 492                    ____22__________________
 493                    ______________________4_
 494                    ____1111________________
 495                    _44_____________________
 496                    11111111________________
 497                    ____________________33__
 498                    ___________44___________
 499                    __________33333_________
 500                    ______________22________
 501                    ___________________2222_
 502                    _________111111111______
 503                    _____________________11_
 504                    _________________4______
 505
 506                 Sanitized equivalent (no overlap):
 507                    1_______________________
 508                    _44_____________________
 509                    ___1____________________
 510                    ____22__________________
 511                    ______11________________
 512                    _________1______________
 513                    __________3_____________
 514                    ___________44___________
 515                    _____________33_________
 516                    _______________2________
 517                    ________________1_______
 518                    _________________4______
 519                    ___________________2____
 520                    ____________________33__
 521                    ______________________4_
 522         */
 523
 524         /* if there's only one memory region, don't bother */
 525         if (*pnr_map < 2)
 526                 return -1;
 527
 528         old_nr = *pnr_map;
 529
 530         /* bail out if we find any unreasonable addresses in bios map */
 531         for (i=0; i<old_nr; i++)
 532                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 533                         return -1;
 534
 535         /* create pointers for initial change-point information (for sorting) */
 536         for (i=0; i < 2*old_nr; i++)
 537                 change_point[i] = &change_point_list[i];
 538
 539         /* record all known change-points (starting and ending addresses),
 540            omitting those that are for empty memory regions */
 541         chgidx = 0;
 542         for (i=0; i < old_nr; i++)      {
 543                 if (biosmap[i].size != 0) {
 544                         change_point[chgidx]->addr = biosmap[i].addr;
 545                         change_point[chgidx++]->pbios = &biosmap[i];
 546                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
 547                         change_point[chgidx++]->pbios = &biosmap[i];
 548                 }
 549         }
 550         chg_nr = chgidx;        /* true number of change-points */
 551
 552         /* sort change-point list by memory addresses (low -> high) */
 553         still_changing = 1;
 554         while (still_changing)  {
 555                 still_changing = 0;
 556                 for (i=1; i < chg_nr; i++)  {
 557                         /* if <current_addr> > <last_addr>, swap */
 558                         /* or, if current=<start_addr> & last=<end_addr>, swap */
 559                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
 560                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
 561                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
 562                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
 563                            )
 564                         {
 565                                 change_tmp = change_point[i];
 566                                 change_point[i] = change_point[i-1];
 567                                 change_point[i-1] = change_tmp;
 568                                 still_changing=1;
 569                         }
 570                 }
 571         }
 572
 573         /* create a new bios memory map, removing overlaps */
 574         overlap_entries=0;       /* number of entries in the overlap table */
 575         new_bios_entry=0;        /* index for creating new bios map entries */
 576         last_type = 0;           /* start with undefined memory type */
 577         last_addr = 0;           /* start with 0 as last starting address */
 578         /* loop through change-points, determining affect on the new bios map */
 579         for (chgidx=0; chgidx < chg_nr; chgidx++)
 580         {
 581                 /* keep track of all overlapping bios entries */
 582                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
 583                 {
 584                         /* add map entry to overlap list (> 1 entry implies an overlap) */
 585                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
 586                 }
 587                 else
 588                 {
 589                         /* remove entry from list (order independent, so swap with last) */
 590                         for (i=0; i<overlap_entries; i++)
 591                         {
 592                                 if (overlap_list[i] == change_point[chgidx]->pbios)
 593                                         overlap_list[i] = overlap_list[overlap_entries-1];
 594                         }
 595                         overlap_entries--;
 596                 }
 597                 /* if there are overlapping entries, decide which "type" to use */
 598                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
 599                 current_type = 0;
 600                 for (i=0; i<overlap_entries; i++)
 601                         if (overlap_list[i]->type > current_type)
 602                                 current_type = overlap_list[i]->type;
 603                 /* continue building up new bios map based on this information */
 604                 if (current_type != last_type)  {
 605                         if (last_type != 0)      {
 606                                 new_bios[new_bios_entry].size =
 607                                         change_point[chgidx]->addr - last_addr;
 608                                 /* move forward only if the new size was non-zero */
 609                                 if (new_bios[new_bios_entry].size != 0)
 610                                         if (++new_bios_entry >= E820MAX)
 611                                                 break;  /* no more space left for new bios entries */
 612                         }
 613                         if (current_type != 0)  {
 614                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
 615                                 new_bios[new_bios_entry].type = current_type;
 616                                 last_addr=change_point[chgidx]->addr;
 617                         }
 618                         last_type = current_type;
 619                 }
 620         }
 621         new_nr = new_bios_entry;   /* retain count for new bios entries */
 622
 623         /* copy new bios mapping into original location */
 624         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
 625         *pnr_map = new_nr;
 626
 627         return 0;
 628 }
 629
 630 /*
 631  * Copy the BIOS e820 map into a safe place.
 632  *
 633  * Sanity-check it while we're at it..
 634  *
 635  * If we're lucky and live on a modern system, the setup code
 636  * will have given us a memory map that we can use to properly
 637  * set up memory.  If we aren't, we'll fake a memory map.
 638  *
 639  * We check to see that the memory map contains at least 2 elements
 640  * before we'll use it, because the detection code in setup.S may
 641  * not be perfect and most every PC known to man has two memory
 642  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
 643  * thinkpad 560x, for example, does not cooperate with the memory
 644  * detection code.)
 645  */
 646 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
 647 {
 648         /* Only one memory region (or negative)? Ignore it */
 649         if (nr_map < 2)
 650                 return -1;
 651
 652         do {
 653                 unsigned long long start = biosmap->addr;
 654                 unsigned long long size = biosmap->size;
 655                 unsigned long long end = start + size;
 656                 unsigned long type = biosmap->type;
 657
 658                 /* Overflow in 64 bits? Ignore the memory map. */
 659                 if (start > end)
 660                         return -1;
 661
 662                 /*
 663                  * Some BIOSes claim RAM in the 640k - 1M region.
 664                  * Not right. Fix it up.
 665                  */
 666                 if (type == E820_RAM) {
 667                         if (start < 0x100000ULL && end > 0xA0000ULL) {
 668                                 if (start < 0xA0000ULL)
 669                                         add_memory_region(start, 0xA0000ULL-start, type);
 670                                 if (end <= 0x100000ULL)
 671                                         continue;
 672                                 start = 0x100000ULL;
 673                                 size = end - start;
 674                         }
 675                 }
 676                 add_memory_region(start, size, type);
 677         } while (biosmap++,--nr_map);
 678         return 0;
 679 }
 680
 681 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 682 struct edd edd;
 683 #ifdef CONFIG_EDD_MODULE
 684 EXPORT_SYMBOL(edd);
 685 #endif
 686 /**
 687  * copy_edd() - Copy the BIOS EDD information
 688  *              from boot_params into a safe place.
 689  *
 690  */
 691 static inline void copy_edd(void)
 692 {
 693      memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
 694      memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
 695      edd.mbr_signature_nr = EDD_MBR_SIG_NR;
 696      edd.edd_info_nr = EDD_NR;
 697 }
 698 #else
 699 static inline void copy_edd(void)
 700 {
 701 }
 702 #endif
 703
 704 static void __init parse_cmdline_early (char ** cmdline_p)
 705 {
 706         char c = ' ', *to = command_line, *from = saved_command_line;
 707         int len = 0;
 708         int userdef = 0;
 709
 710         /* Save unparsed command line copy for /proc/cmdline */
 711         saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
 712
 713         for (;;) {
 714                 if (c != ' ')
 715                         goto next_char;
 716                 /*
 717                  * "mem=nopentium" disables the 4MB page tables.
 718                  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
 719                  * to <mem>, overriding the bios size.
 720                  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
 721                  * <start> to <start>+<mem>, overriding the bios size.
 722                  *
 723                  * HPA tells me bootloaders need to parse mem=, so no new
 724                  * option should be mem=  [also see Documentation/i386/boot.txt]
 725                  */
 726                 if (!memcmp(from, "mem=", 4)) {
 727                         if (to != command_line)
 728                                 to--;
 729                         if (!memcmp(from+4, "nopentium", 9)) {
 730                                 from += 9+4;
 731                                 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
 732                                 disable_pse = 1;
 733                         } else {
 734                                 /* If the user specifies memory size, we
 735                                  * limit the BIOS-provided memory map to
 736                                  * that size. exactmap can be used to specify
 737                                  * the exact map. mem=number can be used to
 738                                  * trim the existing memory map.
 739                                  */
 740                                 unsigned long long mem_size;
 741
 742                                 mem_size = memparse(from+4, &from);
 743                                 limit_regions(mem_size);
 744                                 userdef=1;
 745                         }
 746                 }
 747
 748                 else if (!memcmp(from, "memmap=", 7)) {
 749                         if (to != command_line)
 750                                 to--;
 751                         if (!memcmp(from+7, "exactmap", 8)) {
 752 #ifdef CONFIG_CRASH_DUMP
 753                                 /* If we are doing a crash dump, we
 754                                  * still need to know the real mem
 755                                  * size before original memory map is
 756                                  * reset.
 757                                  */
 758                                 find_max_pfn();
 759                                 saved_max_pfn = max_pfn;
 760 #endif
 761                                 from += 8+7;
 762                                 e820.nr_map = 0;
 763                                 userdef = 1;
 764                         } else {
 765                                 /* If the user specifies memory size, we
 766                                  * limit the BIOS-provided memory map to
 767                                  * that size. exactmap can be used to specify
 768                                  * the exact map. mem=number can be used to
 769                                  * trim the existing memory map.
 770                                  */
 771                                 unsigned long long start_at, mem_size;
 772
 773                                 mem_size = memparse(from+7, &from);
 774                                 if (*from == '@') {
 775                                         start_at = memparse(from+1, &from);
 776                                         add_memory_region(start_at, mem_size, E820_RAM);
 777                                 } else if (*from == '#') {
 778                                         start_at = memparse(from+1, &from);
 779                                         add_memory_region(start_at, mem_size, E820_ACPI);
 780                                 } else if (*from == '$') {
 781                                         start_at = memparse(from+1, &from);
 782                                         add_memory_region(start_at, mem_size, E820_RESERVED);
 783                                 } else {
 784                                         limit_regions(mem_size);
 785                                         userdef=1;
 786                                 }
 787                         }
 788                 }
 789
 790                 else if (!memcmp(from, "noexec=", 7))
 791                         noexec_setup(from + 7);
 792
 793
 794 #ifdef  CONFIG_X86_SMP
 795                 /*
 796                  * If the BIOS enumerates physical processors before logical,
 797                  * maxcpus=N at enumeration-time can be used to disable HT.
 798                  */
 799                 else if (!memcmp(from, "maxcpus=", 8)) {
 800                         extern unsigned int maxcpus;
 801
 802                         maxcpus = simple_strtoul(from + 8, NULL, 0);
 803                 }
 804 #endif
 805
 806 #ifdef CONFIG_ACPI
 807                 /* "acpi=off" disables both ACPI table parsing and interpreter */
 808                 else if (!memcmp(from, "acpi=off", 8)) {
 809                         disable_acpi();
 810                 }
 811
 812                 /* acpi=force to over-ride black-list */
 813                 else if (!memcmp(from, "acpi=force", 10)) {
 814                         acpi_force = 1;
 815                         acpi_ht = 1;
 816                         acpi_disabled = 0;
 817                 }
 818
 819                 /* acpi=strict disables out-of-spec workarounds */
 820                 else if (!memcmp(from, "acpi=strict", 11)) {
 821                         acpi_strict = 1;
 822                 }
 823
 824                 /* Limit ACPI just to boot-time to enable HT */
 825                 else if (!memcmp(from, "acpi=ht", 7)) {
 826                         if (!acpi_force)
 827                                 disable_acpi();
 828                         acpi_ht = 1;
 829                 }
 830
 831                 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
 832                 else if (!memcmp(from, "pci=noacpi", 10)) {
 833                         acpi_disable_pci();
 834                 }
 835                 /* "acpi=noirq" disables ACPI interrupt routing */
 836                 else if (!memcmp(from, "acpi=noirq", 10)) {
 837                         acpi_noirq_set();
 838                 }
 839
 840                 else if (!memcmp(from, "acpi_sci=edge", 13))
 841                         acpi_sci_flags.trigger =  1;
 842
 843                 else if (!memcmp(from, "acpi_sci=level", 14))
 844                         acpi_sci_flags.trigger = 3;
 845
 846                 else if (!memcmp(from, "acpi_sci=high", 13))
 847                         acpi_sci_flags.polarity = 1;
 848
 849                 else if (!memcmp(from, "acpi_sci=low", 12))
 850                         acpi_sci_flags.polarity = 3;
 851
 852 #ifdef CONFIG_X86_IO_APIC
 853                 else if (!memcmp(from, "acpi_skip_timer_override", 24))
 854                         acpi_skip_timer_override = 1;
 855
 856                 if (!memcmp(from, "disable_timer_pin_1", 19))
 857                         disable_timer_pin_1 = 1;
 858                 if (!memcmp(from, "enable_timer_pin_1", 18))
 859                         disable_timer_pin_1 = -1;
 860
 861                 /* disable IO-APIC */
 862                 else if (!memcmp(from, "noapic", 6))
 863                         disable_ioapic_setup();
 864 #endif /* CONFIG_X86_IO_APIC */
 865 #endif /* CONFIG_ACPI */
 866
 867 #ifdef CONFIG_X86_LOCAL_APIC
 868                 /* enable local APIC */
 869                 else if (!memcmp(from, "lapic", 5))
 870                         lapic_enable();
 871
 872                 /* disable local APIC */
 873                 else if (!memcmp(from, "nolapic", 6))
 874                         lapic_disable();
 875 #endif /* CONFIG_X86_LOCAL_APIC */
 876
 877 #ifdef CONFIG_KEXEC
 878                 /* crashkernel=size@addr specifies the location to reserve for
 879                  * a crash kernel.  By reserving this memory we guarantee
 880                  * that linux never set's it up as a DMA target.
 881                  * Useful for holding code to do something appropriate
 882                  * after a kernel panic.
 883                  */
 884                 else if (!memcmp(from, "crashkernel=", 12)) {
 885                         unsigned long size, base;
 886                         size = memparse(from+12, &from);
 887                         if (*from == '@') {
 888                                 base = memparse(from+1, &from);
 889                                 /* FIXME: Do I want a sanity check
 890                                  * to validate the memory range?
 891                                  */
 892                                 crashk_res.start = base;
 893                                 crashk_res.end   = base + size - 1;
 894                         }
 895                 }
 896 #endif
 897 #ifdef CONFIG_PROC_VMCORE
 898                 /* elfcorehdr= specifies the location of elf core header
 899                  * stored by the crashed kernel.
 900                  */
 901                 else if (!memcmp(from, "elfcorehdr=", 11))
 902                         elfcorehdr_addr = memparse(from+11, &from);
 903 #endif
 904
 905                 /*
 906                  * highmem=size forces highmem to be exactly 'size' bytes.
 907                  * This works even on boxes that have no highmem otherwise.
 908                  * This also works to reduce highmem size on bigger boxes.
 909                  */
 910                 else if (!memcmp(from, "highmem=", 8))
 911                         highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;
 912
 913                 /*
 914                  * vmalloc=size forces the vmalloc area to be exactly 'size'
 915                  * bytes. This can be used to increase (or decrease) the
 916                  * vmalloc area - the default is 128m.
 917                  */
 918                 else if (!memcmp(from, "vmalloc=", 8))
 919                         __VMALLOC_RESERVE = memparse(from+8, &from);
 920
 921         next_char:
 922                 c = *(from++);
 923                 if (!c)
 924                         break;
 925                 if (COMMAND_LINE_SIZE <= ++len)
 926                         break;
 927                 *(to++) = c;
 928         }
 929         *to = '\0';
 930         *cmdline_p = command_line;
 931         if (userdef) {
 932                 printk(KERN_INFO "user-defined physical RAM map:\n");
 933                 print_memory_map("user");
 934         }
 935 }
 936
 937 /*
 938  * Callback for efi_memory_walk.
 939  */
 940 static int __init
 941 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
 942 {
 943         unsigned long *max_pfn = arg, pfn;
 944
 945         if (start < end) {
 946                 pfn = PFN_UP(end -1);
 947                 if (pfn > *max_pfn)
 948                         *max_pfn = pfn;
 949         }
 950         return 0;
 951 }
 952
 953 static int __init
 954 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
 955 {
 956         memory_present(0, start, end);
 957         return 0;
 958 }
 959
 960  /*
 961   * This function checks if the entire range <start,end> is mapped with type.
 962   *
 963   * Note: this function only works correct if the e820 table is sorted and
 964   * not-overlapping, which is the case
 965   */
 966 int __init
 967 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
 968 {
 969         u64 start = s;
 970         u64 end = e;
 971         int i;
 972         for (i = 0; i < e820.nr_map; i++) {
 973                 struct e820entry *ei = &e820.map[i];
 974                 if (type && ei->type != type)
 975                         continue;
 976                 /* is the region (part) in overlap with the current region ?*/
 977                 if (ei->addr >= end || ei->addr + ei->size <= start)
 978                         continue;
 979                 /* if the region is at the beginning of <start,end> we move
 980                  * start to the end of the region since it's ok until there
 981                  */
 982                 if (ei->addr <= start)
 983                         start = ei->addr + ei->size;
 984                 /* if start is now at or beyond end, we're done, full
 985                  * coverage */
 986                 if (start >= end)
 987                         return 1; /* we're done */
 988         }
 989         return 0;
 990 }
 991
 992 /*
 993  * Find the highest page frame number we have available
 994  */
 995 void __init find_max_pfn(void)
 996 {
 997         int i;
 998
 999         max_pfn = 0;
1000         if (efi_enabled) {
1001                 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
1002                 efi_memmap_walk(efi_memory_present_wrapper, NULL);
1003                 return;
1004         }
1005
1006         for (i = 0; i < e820.nr_map; i++) {
1007                 unsigned long start, end;
1008                 /* RAM? */
1009                 if (e820.map[i].type != E820_RAM)
1010                         continue;
1011                 start = PFN_UP(e820.map[i].addr);
1012                 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1013                 if (start >= end)
1014                         continue;
1015                 if (end > max_pfn)
1016                         max_pfn = end;
1017                 memory_present(0, start, end);
1018         }
1019 }
1020
1021 /*
1022  * Determine low and high memory ranges:
1023  */
1024 unsigned long __init find_max_low_pfn(void)
1025 {
1026         unsigned long max_low_pfn;
1027
1028         max_low_pfn = max_pfn;
1029         if (max_low_pfn > MAXMEM_PFN) {
1030                 if (highmem_pages == -1)
1031                         highmem_pages = max_pfn - MAXMEM_PFN;
1032                 if (highmem_pages + MAXMEM_PFN < max_pfn)
1033                         max_pfn = MAXMEM_PFN + highmem_pages;
1034                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
1035                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
1036                         highmem_pages = 0;
1037                 }
1038                 max_low_pfn = MAXMEM_PFN;
1039 #ifndef CONFIG_HIGHMEM
1040                 /* Maximum memory usable is what is directly addressable */
1041                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
1042                                         MAXMEM>>20);
1043                 if (max_pfn > MAX_NONPAE_PFN)
1044                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1045                 else
1046                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
1047                 max_pfn = MAXMEM_PFN;
1048 #else /* !CONFIG_HIGHMEM */
1049 #ifndef CONFIG_X86_PAE
1050                 if (max_pfn > MAX_NONPAE_PFN) {
1051                         max_pfn = MAX_NONPAE_PFN;
1052                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
1053                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1054                 }
1055 #endif /* !CONFIG_X86_PAE */
1056 #endif /* !CONFIG_HIGHMEM */
1057         } else {
1058                 if (highmem_pages == -1)
1059                         highmem_pages = 0;
1060 #ifdef CONFIG_HIGHMEM
1061                 if (highmem_pages >= max_pfn) {
1062                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
1063                         highmem_pages = 0;
1064                 }
1065                 if (highmem_pages) {
1066                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
1067                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
1068                                 highmem_pages = 0;
1069                         }
1070                         max_low_pfn -= highmem_pages;
1071                 }
1072 #else
1073                 if (highmem_pages)
1074                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
1075 #endif
1076         }
1077         return max_low_pfn;
1078 }
1079
1080 /*
1081  * Free all available memory for boot time allocation.  Used
1082  * as a callback function by efi_memory_walk()
1083  */
1084
1085 static int __init
1086 free_available_memory(unsigned long start, unsigned long end, void *arg)
1087 {
1088         /* check max_low_pfn */
1089         if (start >= (max_low_pfn << PAGE_SHIFT))
1090                 return 0;
1091         if (end >= (max_low_pfn << PAGE_SHIFT))
1092                 end = max_low_pfn << PAGE_SHIFT;
1093         if (start < end)
1094                 free_bootmem(start, end - start);
1095
1096         return 0;
1097 }
1098 /*
1099  * Register fully available low RAM pages with the bootmem allocator.
1100  */
1101 static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
1102 {
1103         int i;
1104
1105         if (efi_enabled) {
1106                 efi_memmap_walk(free_available_memory, NULL);
1107                 return;
1108         }
1109         for (i = 0; i < e820.nr_map; i++) {
1110                 unsigned long curr_pfn, last_pfn, size;
1111                 /*
1112                  * Reserve usable low memory
1113                  */
1114                 if (e820.map[i].type != E820_RAM)
1115                         continue;
1116                 /*
1117                  * We are rounding up the start address of usable memory:
1118                  */
1119                 curr_pfn = PFN_UP(e820.map[i].addr);
1120                 if (curr_pfn >= max_low_pfn)
1121                         continue;
1122                 /*
1123                  * ... and at the end of the usable range downwards:
1124                  */
1125                 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1126
1127                 if (last_pfn > max_low_pfn)
1128                         last_pfn = max_low_pfn;
1129
1130                 /*
1131                  * .. finally, did all the rounding and playing
1132                  * around just make the area go away?
1133                  */
1134                 if (last_pfn <= curr_pfn)
1135                         continue;
1136
1137                 size = last_pfn - curr_pfn;
1138                 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1139         }
1140 }
1141
1142 /*
1143  * workaround for Dell systems that neglect to reserve EBDA
1144  */
1145 static void __init reserve_ebda_region(void)
1146 {
1147         unsigned int addr;
1148         addr = get_bios_ebda();
1149         if (addr)
1150                 reserve_bootmem(addr, PAGE_SIZE);
1151 }
1152
1153 #ifndef CONFIG_NEED_MULTIPLE_NODES
1154 void __init setup_bootmem_allocator(void);
1155 static unsigned long __init setup_memory(void)
1156 {
1157         /*
1158          * partially used pages are not usable - thus
1159          * we are rounding upwards:
1160          */
1161         min_low_pfn = PFN_UP(init_pg_tables_end);
1162
1163         find_max_pfn();
1164
1165         max_low_pfn = find_max_low_pfn();
1166
1167 #ifdef CONFIG_HIGHMEM
1168         highstart_pfn = highend_pfn = max_pfn;
1169         if (max_pfn > max_low_pfn) {
1170                 highstart_pfn = max_low_pfn;
1171         }
1172         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1173                 pages_to_mb(highend_pfn - highstart_pfn));
1174 #endif
1175         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1176                         pages_to_mb(max_low_pfn));
1177
1178         setup_bootmem_allocator();
1179
1180         return max_low_pfn;
1181 }
1182
1183 void __init zone_sizes_init(void)
1184 {
1185         unsigned long zones_size[MAX_NR_ZONES] = { 0, };
1186         unsigned int max_dma, low;
1187
1188         max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1189         low = max_low_pfn;
1190
1191         if (low < max_dma)
1192                 zones_size[ZONE_DMA] = low;
1193         else {
1194                 zones_size[ZONE_DMA] = max_dma;
1195                 zones_size[ZONE_NORMAL] = low - max_dma;
1196 #ifdef CONFIG_HIGHMEM
1197                 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1198 #endif
1199         }
1200         free_area_init(zones_size);
1201 }
1202 #else
1203 extern unsigned long __init setup_memory(void);
1204 extern void zone_sizes_init(void);
1205 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1206
1207 void __init setup_bootmem_allocator(void)
1208 {
1209         unsigned long bootmap_size;
1210         /*
1211          * Initialize the boot-time allocator (with low memory only):
1212          */
1213         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1214
1215         register_bootmem_low_pages(max_low_pfn);
1216
1217         /*
1218          * Reserve the bootmem bitmap itself as well. We do this in two
1219          * steps (first step was init_bootmem()) because this catches
1220          * the (very unlikely) case of us accidentally initializing the
1221          * bootmem allocator with an invalid RAM area.
1222          */
1223         reserve_bootmem(__PHYSICAL_START, (PFN_PHYS(min_low_pfn) +
1224                          bootmap_size + PAGE_SIZE-1) - (__PHYSICAL_START));
1225
1226         /*
1227          * reserve physical page 0 - it's a special BIOS page on many boxes,
1228          * enabling clean reboots, SMP operation, laptop functions.
1229          */
1230         reserve_bootmem(0, PAGE_SIZE);
1231
1232         /* reserve EBDA region, it's a 4K region */
1233         reserve_ebda_region();
1234
1235     /* could be an AMD 768MPX chipset. Reserve a page  before VGA to prevent
1236        PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1237        unless you have no PS/2 mouse plugged in. */
1238         if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1239             boot_cpu_data.x86 == 6)
1240              reserve_bootmem(0xa0000 - 4096, 4096);
1241
1242 #ifdef CONFIG_SMP
1243         /*
1244          * But first pinch a few for the stack/trampoline stuff
1245          * FIXME: Don't need the extra page at 4K, but need to fix
1246          * trampoline before removing it. (see the GDT stuff)
1247          */
1248         reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1249 #endif
1250 #ifdef CONFIG_ACPI_SLEEP
1251         /*
1252          * Reserve low memory region for sleep support.
1253          */
1254         acpi_reserve_bootmem();
1255 #endif
1256 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1257         /*
1258          * Find and reserve possible boot-time SMP configuration:
1259          */
1260         find_smp_config();
1261 #endif
1262         numa_kva_reserve();
1263 #ifdef CONFIG_BLK_DEV_INITRD
1264         if (LOADER_TYPE && INITRD_START) {
1265                 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1266                         reserve_bootmem(INITRD_START, INITRD_SIZE);
1267                         initrd_start =
1268                                 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1269                         initrd_end = initrd_start+INITRD_SIZE;
1270                 }
1271                 else {
1272                         printk(KERN_ERR "initrd extends beyond end of memory "
1273                             "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1274                             INITRD_START + INITRD_SIZE,
1275                             max_low_pfn << PAGE_SHIFT);
1276                         initrd_start = 0;
1277                 }
1278         }
1279 #endif
1280 #ifdef CONFIG_KEXEC
1281         if (crashk_res.start != crashk_res.end)
1282                 reserve_bootmem(crashk_res.start,
1283                         crashk_res.end - crashk_res.start + 1);
1284 #endif
1285 }
1286
1287 /*
1288  * The node 0 pgdat is initialized before all of these because
1289  * it's needed for bootmem.  node>0 pgdats have their virtual
1290  * space allocated before the pagetables are in place to access
1291  * them, so they can't be cleared then.
1292  *
1293  * This should all compile down to nothing when NUMA is off.
1294  */
1295 void __init remapped_pgdat_init(void)
1296 {
1297         int nid;
1298
1299         for_each_online_node(nid) {
1300                 if (nid != 0)
1301                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1302         }
1303 }
1304
1305 /*
1306  * Request address space for all standard RAM and ROM resources
1307  * and also for regions reported as reserved by the e820.
1308  */
1309 static void __init
1310 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1311 {
1312         int i;
1313
1314         probe_roms();
1315         for (i = 0; i < e820.nr_map; i++) {
1316                 struct resource *res;
1317 #ifndef CONFIG_RESOURCES_64BIT
1318                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1319                         continue;
1320 #endif
1321                 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
1322                 switch (e820.map[i].type) {
1323                 case E820_RAM:  res->name = "System RAM"; break;
1324                 case E820_ACPI: res->name = "ACPI Tables"; break;
1325                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
1326                 default:        res->name = "reserved";
1327                 }
1328                 res->start = e820.map[i].addr;
1329                 res->end = res->start + e820.map[i].size - 1;
1330                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1331                 if (request_resource(&iomem_resource, res)) {
1332                         kfree(res);
1333                         continue;
1334                 }
1335                 if (e820.map[i].type == E820_RAM) {
1336                         /*
1337                          *  We don't know which RAM region contains kernel data,
1338                          *  so we try it repeatedly and let the resource manager
1339                          *  test it.
1340                          */
1341                         request_resource(res, code_resource);
1342                         request_resource(res, data_resource);
1343 #ifdef CONFIG_KEXEC
1344                         request_resource(res, &crashk_res);
1345 #endif
1346                 }
1347         }
1348 }
1349
1350 /*
1351  * Request address space for all standard resources
1352  *
1353  * This is called just before pcibios_init(), which is also a
1354  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
1355  */
1356 static int __init request_standard_resources(void)
1357 {
1358         int i;
1359
1360         printk("Setting up standard PCI resources\n");
1361         if (efi_enabled)
1362                 efi_initialize_iomem_resources(&code_resource, &data_resource);
1363         else
1364                 legacy_init_iomem_resources(&code_resource, &data_resource);
1365
1366         /* EFI systems may still have VGA */
1367         request_resource(&iomem_resource, &video_ram_resource);
1368
1369         /* request I/O space for devices used on all i[345]86 PCs */
1370         for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1371                 request_resource(&ioport_resource, &standard_io_resources[i]);
1372         return 0;
1373 }
1374
1375 subsys_initcall(request_standard_resources);
1376
1377 static void __init register_memory(void)
1378 {
1379         unsigned long gapstart, gapsize, round;
1380         unsigned long long last;
1381         int i;
1382
1383         /*
1384          * Search for the bigest gap in the low 32 bits of the e820
1385          * memory space.
1386          */
1387         last = 0x100000000ull;
1388         gapstart = 0x10000000;
1389         gapsize = 0x400000;
1390         i = e820.nr_map;
1391         while (--i >= 0) {
1392                 unsigned long long start = e820.map[i].addr;
1393                 unsigned long long end = start + e820.map[i].size;
1394
1395                 /*
1396                  * Since "last" is at most 4GB, we know we'll
1397                  * fit in 32 bits if this condition is true
1398                  */
1399                 if (last > end) {
1400                         unsigned long gap = last - end;
1401
1402                         if (gap > gapsize) {
1403                                 gapsize = gap;
1404                                 gapstart = end;
1405                         }
1406                 }
1407                 if (start < last)
1408                         last = start;
1409         }
1410
1411         /*
1412          * See how much we want to round up: start off with
1413          * rounding to the next 1MB area.
1414          */
1415         round = 0x100000;
1416         while ((gapsize >> 4) > round)
1417                 round += round;
1418         /* Fun with two's complement */
1419         pci_mem_start = (gapstart + round) & -round;
1420
1421         printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1422                 pci_mem_start, gapstart, gapsize);
1423 }
1424
1425 #ifdef CONFIG_MCA
1426 static void set_mca_bus(int x)
1427 {
1428         MCA_bus = x;
1429 }
1430 #else
1431 static void set_mca_bus(int x) { }
1432 #endif
1433
1434 /*
1435  * Determine if we were loaded by an EFI loader.  If so, then we have also been
1436  * passed the efi memmap, systab, etc., so we should use these data structures
1437  * for initialization.  Note, the efi init code path is determined by the
1438  * global efi_enabled. This allows the same kernel image to be used on existing
1439  * systems (with a traditional BIOS) as well as on EFI systems.
1440  */
1441 void __init setup_arch(char **cmdline_p)
1442 {
1443         unsigned long max_low_pfn;
1444
1445         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1446         pre_setup_arch_hook();
1447         early_cpu_init();
1448
1449         /*
1450          * FIXME: This isn't an official loader_type right
1451          * now but does currently work with elilo.
1452          * If we were configured as an EFI kernel, check to make
1453          * sure that we were loaded correctly from elilo and that
1454          * the system table is valid.  If not, then initialize normally.
1455          */
1456 #ifdef CONFIG_EFI
1457         if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1458                 efi_enabled = 1;
1459 #endif
1460
1461         ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1462         drive_info = DRIVE_INFO;
1463         screen_info = SCREEN_INFO;
1464         edid_info = EDID_INFO;
1465         apm_info.bios = APM_BIOS_INFO;
1466         ist_info = IST_INFO;
1467         saved_videomode = VIDEO_MODE;
1468         if( SYS_DESC_TABLE.length != 0 ) {
1469                 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1470                 machine_id = SYS_DESC_TABLE.table[0];
1471                 machine_submodel_id = SYS_DESC_TABLE.table[1];
1472                 BIOS_revision = SYS_DESC_TABLE.table[2];
1473         }
1474         bootloader_type = LOADER_TYPE;
1475
1476 #ifdef CONFIG_BLK_DEV_RAM
1477         rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1478         rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1479         rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1480 #endif
1481         ARCH_SETUP
1482         if (efi_enabled)
1483                 efi_init();
1484         else {
1485                 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1486                 print_memory_map(machine_specific_memory_setup());
1487         }
1488
1489         copy_edd();
1490
1491         if (!MOUNT_ROOT_RDONLY)
1492                 root_mountflags &= ~MS_RDONLY;
1493         init_mm.start_code = (unsigned long) _text;
1494         init_mm.end_code = (unsigned long) _etext;
1495         init_mm.end_data = (unsigned long) _edata;
1496         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1497
1498         code_resource.start = virt_to_phys(_text);
1499         code_resource.end = virt_to_phys(_etext)-1;
1500         data_resource.start = virt_to_phys(_etext);
1501         data_resource.end = virt_to_phys(_edata)-1;
1502
1503         parse_cmdline_early(cmdline_p);
1504
1505 #ifdef CONFIG_EARLY_PRINTK
1506         {
1507                 char *s = strstr(*cmdline_p, "earlyprintk=");
1508                 if (s) {
1509                         setup_early_printk(strchr(s, '=') + 1);
1510                         printk("early console enabled\n");
1511                 }
1512         }
1513 #endif
1514
1515         max_low_pfn = setup_memory();
1516
1517         /*
1518          * NOTE: before this point _nobody_ is allowed to allocate
1519          * any memory using the bootmem allocator.  Although the
1520          * alloctor is now initialised only the first 8Mb of the kernel
1521          * virtual address space has been mapped.  All allocations before
1522          * paging_init() has completed must use the alloc_bootmem_low_pages()
1523          * variant (which allocates DMA'able memory) and care must be taken
1524          * not to exceed the 8Mb limit.
1525          */
1526
1527 #ifdef CONFIG_SMP
1528         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1529 #endif
1530         paging_init();
1531         remapped_pgdat_init();
1532         sparse_init();
1533         zone_sizes_init();
1534
1535         /*
1536          * NOTE: at this point the bootmem allocator is fully available.
1537          */
1538
1539         dmi_scan_machine();
1540
1541 #ifdef CONFIG_X86_GENERICARCH
1542         generic_apic_probe(*cmdline_p);
1543 #endif
1544         if (efi_enabled)
1545                 efi_map_memmap();
1546
1547 #ifdef CONFIG_ACPI
1548         /*
1549          * Parse the ACPI tables for possible boot-time SMP configuration.
1550          */
1551         acpi_boot_table_init();
1552 #endif
1553
1554 #ifdef CONFIG_X86_IO_APIC
1555         check_acpi_pci();       /* Checks more than just ACPI actually */
1556 #endif
1557
1558 #ifdef CONFIG_ACPI
1559         acpi_boot_init();
1560
1561 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1562         if (def_to_bigsmp)
1563                 printk(KERN_WARNING "More than 8 CPUs detected and "
1564                         "CONFIG_X86_PC cannot handle it.\nUse "
1565                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1566 #endif
1567 #endif
1568 #ifdef CONFIG_X86_LOCAL_APIC
1569         if (smp_found_config)
1570                 get_smp_config();
1571 #endif
1572
1573         register_memory();
1574
1575 #ifdef CONFIG_VT
1576 #if defined(CONFIG_VGA_CONSOLE)
1577         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1578                 conswitchp = &vga_con;
1579 #elif defined(CONFIG_DUMMY_CONSOLE)
1580         conswitchp = &dummy_con;
1581 #endif
1582 #endif
1583         tsc_init();
1584 }
1585
1586 static __init int add_pcspkr(void)
1587 {
1588         struct platform_device *pd;
1589         int ret;
1590
1591         pd = platform_device_alloc("pcspkr", -1);
1592         if (!pd)
1593                 return -ENOMEM;
1594
1595         ret = platform_device_add(pd);
1596         if (ret)
1597                 platform_device_put(pd);
1598
1599         return ret;
1600 }
1601 device_initcall(add_pcspkr);
1602
1603 /*
1604  * Local Variables:
1605  * mode:c
1606  * c-file-style:"k&r"
1607  * c-basic-offset:8
1608  * End:
1609  */