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/config.h>
  27 #include <linux/sched.h>
  28 #include <linux/mm.h>
  29 #include <linux/mmzone.h>
  30 #include <linux/tty.h>
  31 #include <linux/ioport.h>
  32 #include <linux/acpi.h>
  33 #include <linux/apm_bios.h>
  34 #include <linux/initrd.h>
  35 #include <linux/bootmem.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/platform_device.h>
  38 #include <linux/console.h>
  39 #include <linux/mca.h>
  40 #include <linux/root_dev.h>
  41 #include <linux/highmem.h>
  42 #include <linux/module.h>
  43 #include <linux/efi.h>
  44 #include <linux/init.h>
  45 #include <linux/edd.h>
  46 #include <linux/nodemask.h>
  47 #include <linux/kexec.h>
  48 #include <linux/crash_dump.h>
  49 #include <linux/dmi.h>
  50 #include <linux/pfn.h>
  51 #include <linux/suspend.h>
  52
  53 #include <video/edid.h>
  54
  55 #include <asm/apic.h>
  56 #include <asm/e820.h>
  57 #include <asm/mpspec.h>
  58 #include <asm/setup.h>
  59 #include <asm/arch_hooks.h>
  60 #include <asm/sections.h>
  61 #include <asm/io_apic.h>
  62 #include <asm/ist.h>
  63 #include <asm/io.h>
  64 #include <setup_arch.h>
  65 #include <bios_ebda.h>
  66
  67 /* Forward Declaration. */
  68 void __init find_max_pfn(void);
  69
  70 /* This value is set up by the early boot code to point to the value
  71    immediately after the boot time page tables.  It contains a *physical*
  72    address, and must not be in the .bss segment! */
  73 unsigned long init_pg_tables_end __initdata = ~0UL;
  74
  75 int disable_pse __devinitdata = 0;
  76
  77 /*
  78  * Machine setup..
  79  */
  80
  81 #ifdef CONFIG_EFI
  82 int efi_enabled = 0;
  83 EXPORT_SYMBOL(efi_enabled);
  84 #endif
  85
  86 /* cpu data as detected by the assembly code in head.S */
  87 struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
  88 /* common cpu data for all cpus */
  89 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
  90 EXPORT_SYMBOL(boot_cpu_data);
  91
  92 unsigned long mmu_cr4_features;
  93
  94 #ifdef  CONFIG_ACPI
  95         int acpi_disabled = 0;
  96 #else
  97         int acpi_disabled = 1;
  98 #endif
  99 EXPORT_SYMBOL(acpi_disabled);
 100
 101 #ifdef  CONFIG_ACPI
 102 int __initdata acpi_force = 0;
 103 extern acpi_interrupt_flags     acpi_sci_flags;
 104 #endif
 105
 106 /* for MCA, but anyone else can use it if they want */
 107 unsigned int machine_id;
 108 #ifdef CONFIG_MCA
 109 EXPORT_SYMBOL(machine_id);
 110 #endif
 111 unsigned int machine_submodel_id;
 112 unsigned int BIOS_revision;
 113 unsigned int mca_pentium_flag;
 114
 115 /* For PCI or other memory-mapped resources */
 116 unsigned long pci_mem_start = 0x10000000;
 117 #ifdef CONFIG_PCI
 118 EXPORT_SYMBOL(pci_mem_start);
 119 #endif
 120
 121 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 122 int bootloader_type;
 123
 124 /* user-defined highmem size */
 125 static unsigned int highmem_pages = -1;
 126
 127 /*
 128  * Setup options
 129  */
 130 struct drive_info_struct { char dummy[32]; } drive_info;
 131 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
 132     defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
 133 EXPORT_SYMBOL(drive_info);
 134 #endif
 135 struct screen_info screen_info;
 136 EXPORT_SYMBOL(screen_info);
 137 struct apm_info apm_info;
 138 EXPORT_SYMBOL(apm_info);
 139 struct sys_desc_table_struct {
 140         unsigned short length;
 141         unsigned char table[0];
 142 };
 143 struct edid_info edid_info;
 144 EXPORT_SYMBOL_GPL(edid_info);
 145 struct ist_info ist_info;
 146 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 147         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 148 EXPORT_SYMBOL(ist_info);
 149 #endif
 150 struct e820map e820;
 151
 152 extern void early_cpu_init(void);
 153 extern void generic_apic_probe(char *);
 154 extern int root_mountflags;
 155
 156 unsigned long saved_videomode;
 157
 158 #define RAMDISK_IMAGE_START_MASK        0x07FF
 159 #define RAMDISK_PROMPT_FLAG             0x8000
 160 #define RAMDISK_LOAD_FLAG               0x4000
 161
 162 static char command_line[COMMAND_LINE_SIZE];
 163
 164 unsigned char __initdata boot_params[PARAM_SIZE];
 165
 166 static struct resource data_resource = {
 167         .name   = "Kernel data",
 168         .start  = 0,
 169         .end    = 0,
 170         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 171 };
 172
 173 static struct resource code_resource = {
 174         .name   = "Kernel code",
 175         .start  = 0,
 176         .end    = 0,
 177         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 178 };
 179
 180 static struct resource system_rom_resource = {
 181         .name   = "System ROM",
 182         .start  = 0xf0000,
 183         .end    = 0xfffff,
 184         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 185 };
 186
 187 static struct resource extension_rom_resource = {
 188         .name   = "Extension ROM",
 189         .start  = 0xe0000,
 190         .end    = 0xeffff,
 191         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 192 };
 193
 194 static struct resource adapter_rom_resources[] = { {
 195         .name   = "Adapter ROM",
 196         .start  = 0xc8000,
 197         .end    = 0,
 198         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 199 }, {
 200         .name   = "Adapter ROM",
 201         .start  = 0,
 202         .end    = 0,
 203         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 204 }, {
 205         .name   = "Adapter ROM",
 206         .start  = 0,
 207         .end    = 0,
 208         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 209 }, {
 210         .name   = "Adapter ROM",
 211         .start  = 0,
 212         .end    = 0,
 213         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 214 }, {
 215         .name   = "Adapter ROM",
 216         .start  = 0,
 217         .end    = 0,
 218         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 219 }, {
 220         .name   = "Adapter ROM",
 221         .start  = 0,
 222         .end    = 0,
 223         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 224 } };
 225
 226 #define ADAPTER_ROM_RESOURCES \
 227         (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
 228
 229 static struct resource video_rom_resource = {
 230         .name   = "Video ROM",
 231         .start  = 0xc0000,
 232         .end    = 0xc7fff,
 233         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 234 };
 235
 236 static struct resource video_ram_resource = {
 237         .name   = "Video RAM area",
 238         .start  = 0xa0000,
 239         .end    = 0xbffff,
 240         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 241 };
 242
 243 static struct resource standard_io_resources[] = { {
 244         .name   = "dma1",
 245         .start  = 0x0000,
 246         .end    = 0x001f,
 247         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 248 }, {
 249         .name   = "pic1",
 250         .start  = 0x0020,
 251         .end    = 0x0021,
 252         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 253 }, {
 254         .name   = "timer0",
 255         .start  = 0x0040,
 256         .end    = 0x0043,
 257         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 258 }, {
 259         .name   = "timer1",
 260         .start  = 0x0050,
 261         .end    = 0x0053,
 262         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 263 }, {
 264         .name   = "keyboard",
 265         .start  = 0x0060,
 266         .end    = 0x006f,
 267         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 268 }, {
 269         .name   = "dma page reg",
 270         .start  = 0x0080,
 271         .end    = 0x008f,
 272         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 273 }, {
 274         .name   = "pic2",
 275         .start  = 0x00a0,
 276         .end    = 0x00a1,
 277         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 278 }, {
 279         .name   = "dma2",
 280         .start  = 0x00c0,
 281         .end    = 0x00df,
 282         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 283 }, {
 284         .name   = "fpu",
 285         .start  = 0x00f0,
 286         .end    = 0x00ff,
 287         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 288 } };
 289
 290 #define STANDARD_IO_RESOURCES \
 291         (sizeof standard_io_resources / sizeof standard_io_resources[0])
 292
 293 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
 294
 295 static int __init romchecksum(unsigned char *rom, unsigned long length)
 296 {
 297         unsigned char *p, sum = 0;
 298
 299         for (p = rom; p < rom + length; p++)
 300                 sum += *p;
 301         return sum == 0;
 302 }
 303
 304 static void __init probe_roms(void)
 305 {
 306         unsigned long start, length, upper;
 307         unsigned char *rom;
 308         int           i;
 309
 310         /* video rom */
 311         upper = adapter_rom_resources[0].start;
 312         for (start = video_rom_resource.start; start < upper; start += 2048) {
 313                 rom = isa_bus_to_virt(start);
 314                 if (!romsignature(rom))
 315                         continue;
 316
 317                 video_rom_resource.start = start;
 318
 319                 /* 0 < length <= 0x7f * 512, historically */
 320                 length = rom[2] * 512;
 321
 322                 /* if checksum okay, trust length byte */
 323                 if (length && romchecksum(rom, length))
 324                         video_rom_resource.end = start + length - 1;
 325
 326                 request_resource(&iomem_resource, &video_rom_resource);
 327                 break;
 328         }
 329
 330         start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
 331         if (start < upper)
 332                 start = upper;
 333
 334         /* system rom */
 335         request_resource(&iomem_resource, &system_rom_resource);
 336         upper = system_rom_resource.start;
 337
 338         /* check for extension rom (ignore length byte!) */
 339         rom = isa_bus_to_virt(extension_rom_resource.start);
 340         if (romsignature(rom)) {
 341                 length = extension_rom_resource.end - extension_rom_resource.start + 1;
 342                 if (romchecksum(rom, length)) {
 343                         request_resource(&iomem_resource, &extension_rom_resource);
 344                         upper = extension_rom_resource.start;
 345                 }
 346         }
 347
 348         /* check for adapter roms on 2k boundaries */
 349         for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
 350                 rom = isa_bus_to_virt(start);
 351                 if (!romsignature(rom))
 352                         continue;
 353
 354                 /* 0 < length <= 0x7f * 512, historically */
 355                 length = rom[2] * 512;
 356
 357                 /* but accept any length that fits if checksum okay */
 358                 if (!length || start + length > upper || !romchecksum(rom, length))
 359                         continue;
 360
 361                 adapter_rom_resources[i].start = start;
 362                 adapter_rom_resources[i].end = start + length - 1;
 363                 request_resource(&iomem_resource, &adapter_rom_resources[i]);
 364
 365                 start = adapter_rom_resources[i++].end & ~2047UL;
 366         }
 367 }
 368
 369 static void __init limit_regions(unsigned long long size)
 370 {
 371         unsigned long long current_addr = 0;
 372         int i;
 373
 374         if (efi_enabled) {
 375                 efi_memory_desc_t *md;
 376                 void *p;
 377
 378                 for (p = memmap.map, i = 0; p < memmap.map_end;
 379                         p += memmap.desc_size, i++) {
 380                         md = p;
 381                         current_addr = md->phys_addr + (md->num_pages << 12);
 382                         if (md->type == EFI_CONVENTIONAL_MEMORY) {
 383                                 if (current_addr >= size) {
 384                                         md->num_pages -=
 385                                                 (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT);
 386                                         memmap.nr_map = i + 1;
 387                                         return;
 388                                 }
 389                         }
 390                 }
 391         }
 392         for (i = 0; i < e820.nr_map; i++) {
 393                 current_addr = e820.map[i].addr + e820.map[i].size;
 394                 if (current_addr < size)
 395                         continue;
 396
 397                 if (e820.map[i].type != E820_RAM)
 398                         continue;
 399
 400                 if (e820.map[i].addr >= size) {
 401                         /*
 402                          * This region starts past the end of the
 403                          * requested size, skip it completely.
 404                          */
 405                         e820.nr_map = i;
 406                 } else {
 407                         e820.nr_map = i + 1;
 408                         e820.map[i].size -= current_addr - size;
 409                 }
 410                 return;
 411         }
 412 }
 413
 414 void __init add_memory_region(unsigned long long start,
 415                               unsigned long long size, int type)
 416 {
 417         int x;
 418
 419         if (!efi_enabled) {
 420                 x = e820.nr_map;
 421
 422                 if (x == E820MAX) {
 423                     printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
 424                     return;
 425                 }
 426
 427                 e820.map[x].addr = start;
 428                 e820.map[x].size = size;
 429                 e820.map[x].type = type;
 430                 e820.nr_map++;
 431         }
 432 } /* add_memory_region */
 433
 434 #define E820_DEBUG      1
 435
 436 static void __init print_memory_map(char *who)
 437 {
 438         int i;
 439
 440         for (i = 0; i < e820.nr_map; i++) {
 441                 printk(" %s: %016Lx - %016Lx ", who,
 442                         e820.map[i].addr,
 443                         e820.map[i].addr + e820.map[i].size);
 444                 switch (e820.map[i].type) {
 445                 case E820_RAM:  printk("(usable)\n");
 446                                 break;
 447                 case E820_RESERVED:
 448                                 printk("(reserved)\n");
 449                                 break;
 450                 case E820_ACPI:
 451                                 printk("(ACPI data)\n");
 452                                 break;
 453                 case E820_NVS:
 454                                 printk("(ACPI NVS)\n");
 455                                 break;
 456                 default:        printk("type %lu\n", e820.map[i].type);
 457                                 break;
 458                 }
 459         }
 460 }
 461
 462 /*
 463  * Sanitize the BIOS e820 map.
 464  *
 465  * Some e820 responses include overlapping entries.  The following
 466  * replaces the original e820 map with a new one, removing overlaps.
 467  *
 468  */
 469 struct change_member {
 470         struct e820entry *pbios; /* pointer to original bios entry */
 471         unsigned long long addr; /* address for this change point */
 472 };
 473 static struct change_member change_point_list[2*E820MAX] __initdata;
 474 static struct change_member *change_point[2*E820MAX] __initdata;
 475 static struct e820entry *overlap_list[E820MAX] __initdata;
 476 static struct e820entry new_bios[E820MAX] __initdata;
 477
 478 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
 479 {
 480         struct change_member *change_tmp;
 481         unsigned long current_type, last_type;
 482         unsigned long long last_addr;
 483         int chgidx, still_changing;
 484         int overlap_entries;
 485         int new_bios_entry;
 486         int old_nr, new_nr, chg_nr;
 487         int i;
 488
 489         /*
 490                 Visually we're performing the following (1,2,3,4 = memory types)...
 491
 492                 Sample memory map (w/overlaps):
 493                    ____22__________________
 494                    ______________________4_
 495                    ____1111________________
 496                    _44_____________________
 497                    11111111________________
 498                    ____________________33__
 499                    ___________44___________
 500                    __________33333_________
 501                    ______________22________
 502                    ___________________2222_
 503                    _________111111111______
 504                    _____________________11_
 505                    _________________4______
 506
 507                 Sanitized equivalent (no overlap):
 508                    1_______________________
 509                    _44_____________________
 510                    ___1____________________
 511                    ____22__________________
 512                    ______11________________
 513                    _________1______________
 514                    __________3_____________
 515                    ___________44___________
 516                    _____________33_________
 517                    _______________2________
 518                    ________________1_______
 519                    _________________4______
 520                    ___________________2____
 521                    ____________________33__
 522                    ______________________4_
 523         */
 524
 525         /* if there's only one memory region, don't bother */
 526         if (*pnr_map < 2)
 527                 return -1;
 528
 529         old_nr = *pnr_map;
 530
 531         /* bail out if we find any unreasonable addresses in bios map */
 532         for (i=0; i<old_nr; i++)
 533                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
 534                         return -1;
 535
 536         /* create pointers for initial change-point information (for sorting) */
 537         for (i=0; i < 2*old_nr; i++)
 538                 change_point[i] = &change_point_list[i];
 539
 540         /* record all known change-points (starting and ending addresses),
 541            omitting those that are for empty memory regions */
 542         chgidx = 0;
 543         for (i=0; i < old_nr; i++)      {
 544                 if (biosmap[i].size != 0) {
 545                         change_point[chgidx]->addr = biosmap[i].addr;
 546                         change_point[chgidx++]->pbios = &biosmap[i];
 547                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
 548                         change_point[chgidx++]->pbios = &biosmap[i];
 549                 }
 550         }
 551         chg_nr = chgidx;        /* true number of change-points */
 552
 553         /* sort change-point list by memory addresses (low -> high) */
 554         still_changing = 1;
 555         while (still_changing)  {
 556                 still_changing = 0;
 557                 for (i=1; i < chg_nr; i++)  {
 558                         /* if <current_addr> > <last_addr>, swap */
 559                         /* or, if current=<start_addr> & last=<end_addr>, swap */
 560                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
 561                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
 562                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
 563                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
 564                            )
 565                         {
 566                                 change_tmp = change_point[i];
 567                                 change_point[i] = change_point[i-1];
 568                                 change_point[i-1] = change_tmp;
 569                                 still_changing=1;
 570                         }
 571                 }
 572         }
 573
 574         /* create a new bios memory map, removing overlaps */
 575         overlap_entries=0;       /* number of entries in the overlap table */
 576         new_bios_entry=0;        /* index for creating new bios map entries */
 577         last_type = 0;           /* start with undefined memory type */
 578         last_addr = 0;           /* start with 0 as last starting address */
 579         /* loop through change-points, determining affect on the new bios map */
 580         for (chgidx=0; chgidx < chg_nr; chgidx++)
 581         {
 582                 /* keep track of all overlapping bios entries */
 583                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
 584                 {
 585                         /* add map entry to overlap list (> 1 entry implies an overlap) */
 586                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
 587                 }
 588                 else
 589                 {
 590                         /* remove entry from list (order independent, so swap with last) */
 591                         for (i=0; i<overlap_entries; i++)
 592                         {
 593                                 if (overlap_list[i] == change_point[chgidx]->pbios)
 594                                         overlap_list[i] = overlap_list[overlap_entries-1];
 595                         }
 596                         overlap_entries--;
 597                 }
 598                 /* if there are overlapping entries, decide which "type" to use */
 599                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
 600                 current_type = 0;
 601                 for (i=0; i<overlap_entries; i++)
 602                         if (overlap_list[i]->type > current_type)
 603                                 current_type = overlap_list[i]->type;
 604                 /* continue building up new bios map based on this information */
 605                 if (current_type != last_type)  {
 606                         if (last_type != 0)      {
 607                                 new_bios[new_bios_entry].size =
 608                                         change_point[chgidx]->addr - last_addr;
 609                                 /* move forward only if the new size was non-zero */
 610                                 if (new_bios[new_bios_entry].size != 0)
 611                                         if (++new_bios_entry >= E820MAX)
 612                                                 break;  /* no more space left for new bios entries */
 613                         }
 614                         if (current_type != 0)  {
 615                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
 616                                 new_bios[new_bios_entry].type = current_type;
 617                                 last_addr=change_point[chgidx]->addr;
 618                         }
 619                         last_type = current_type;
 620                 }
 621         }
 622         new_nr = new_bios_entry;   /* retain count for new bios entries */
 623
 624         /* copy new bios mapping into original location */
 625         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
 626         *pnr_map = new_nr;
 627
 628         return 0;
 629 }
 630
 631 /*
 632  * Copy the BIOS e820 map into a safe place.
 633  *
 634  * Sanity-check it while we're at it..
 635  *
 636  * If we're lucky and live on a modern system, the setup code
 637  * will have given us a memory map that we can use to properly
 638  * set up memory.  If we aren't, we'll fake a memory map.
 639  *
 640  * We check to see that the memory map contains at least 2 elements
 641  * before we'll use it, because the detection code in setup.S may
 642  * not be perfect and most every PC known to man has two memory
 643  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
 644  * thinkpad 560x, for example, does not cooperate with the memory
 645  * detection code.)
 646  */
 647 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
 648 {
 649         /* Only one memory region (or negative)? Ignore it */
 650         if (nr_map < 2)
 651                 return -1;
 652
 653         do {
 654                 unsigned long long start = biosmap->addr;
 655                 unsigned long long size = biosmap->size;
 656                 unsigned long long end = start + size;
 657                 unsigned long type = biosmap->type;
 658
 659                 /* Overflow in 64 bits? Ignore the memory map. */
 660                 if (start > end)
 661                         return -1;
 662
 663                 /*
 664                  * Some BIOSes claim RAM in the 640k - 1M region.
 665                  * Not right. Fix it up.
 666                  */
 667                 if (type == E820_RAM) {
 668                         if (start < 0x100000ULL && end > 0xA0000ULL) {
 669                                 if (start < 0xA0000ULL)
 670                                         add_memory_region(start, 0xA0000ULL-start, type);
 671                                 if (end <= 0x100000ULL)
 672                                         continue;
 673                                 start = 0x100000ULL;
 674                                 size = end - start;
 675                         }
 676                 }
 677                 add_memory_region(start, size, type);
 678         } while (biosmap++,--nr_map);
 679         return 0;
 680 }
 681
 682 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 683 struct edd edd;
 684 #ifdef CONFIG_EDD_MODULE
 685 EXPORT_SYMBOL(edd);
 686 #endif
 687 /**
 688  * copy_edd() - Copy the BIOS EDD information
 689  *              from boot_params into a safe place.
 690  *
 691  */
 692 static inline void copy_edd(void)
 693 {
 694      memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
 695      memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
 696      edd.mbr_signature_nr = EDD_MBR_SIG_NR;
 697      edd.edd_info_nr = EDD_NR;
 698 }
 699 #else
 700 static inline void copy_edd(void)
 701 {
 702 }
 703 #endif
 704
 705 static void __init parse_cmdline_early (char ** cmdline_p)
 706 {
 707         char c = ' ', *to = command_line, *from = saved_command_line;
 708         int len = 0;
 709         int userdef = 0;
 710
 711         /* Save unparsed command line copy for /proc/cmdline */
 712         saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
 713
 714         for (;;) {
 715                 if (c != ' ')
 716                         goto next_char;
 717                 /*
 718                  * "mem=nopentium" disables the 4MB page tables.
 719                  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
 720                  * to <mem>, overriding the bios size.
 721                  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
 722                  * <start> to <start>+<mem>, overriding the bios size.
 723                  *
 724                  * HPA tells me bootloaders need to parse mem=, so no new
 725                  * option should be mem=  [also see Documentation/i386/boot.txt]
 726                  */
 727                 if (!memcmp(from, "mem=", 4)) {
 728                         if (to != command_line)
 729                                 to--;
 730                         if (!memcmp(from+4, "nopentium", 9)) {
 731                                 from += 9+4;
 732                                 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
 733                                 disable_pse = 1;
 734                         } else {
 735                                 /* If the user specifies memory size, we
 736                                  * limit the BIOS-provided memory map to
 737                                  * that size. exactmap can be used to specify
 738                                  * the exact map. mem=number can be used to
 739                                  * trim the existing memory map.
 740                                  */
 741                                 unsigned long long mem_size;
 742
 743                                 mem_size = memparse(from+4, &from);
 744                                 limit_regions(mem_size);
 745                                 userdef=1;
 746                         }
 747                 }
 748
 749                 else if (!memcmp(from, "memmap=", 7)) {
 750                         if (to != command_line)
 751                                 to--;
 752                         if (!memcmp(from+7, "exactmap", 8)) {
 753 #ifdef CONFIG_CRASH_DUMP
 754                                 /* If we are doing a crash dump, we
 755                                  * still need to know the real mem
 756                                  * size before original memory map is
 757                                  * reset.
 758                                  */
 759                                 find_max_pfn();
 760                                 saved_max_pfn = max_pfn;
 761 #endif
 762                                 from += 8+7;
 763                                 e820.nr_map = 0;
 764                                 userdef = 1;
 765                         } else {
 766                                 /* If the user specifies memory size, we
 767                                  * limit the BIOS-provided memory map to
 768                                  * that size. exactmap can be used to specify
 769                                  * the exact map. mem=number can be used to
 770                                  * trim the existing memory map.
 771                                  */
 772                                 unsigned long long start_at, mem_size;
 773
 774                                 mem_size = memparse(from+7, &from);
 775                                 if (*from == '@') {
 776                                         start_at = memparse(from+1, &from);
 777                                         add_memory_region(start_at, mem_size, E820_RAM);
 778                                 } else if (*from == '#') {
 779                                         start_at = memparse(from+1, &from);
 780                                         add_memory_region(start_at, mem_size, E820_ACPI);
 781                                 } else if (*from == '$') {
 782                                         start_at = memparse(from+1, &from);
 783                                         add_memory_region(start_at, mem_size, E820_RESERVED);
 784                                 } else {
 785                                         limit_regions(mem_size);
 786                                         userdef=1;
 787                                 }
 788                         }
 789                 }
 790
 791                 else if (!memcmp(from, "noexec=", 7))
 792                         noexec_setup(from + 7);
 793
 794
 795 #ifdef  CONFIG_X86_SMP
 796                 /*
 797                  * If the BIOS enumerates physical processors before logical,
 798                  * maxcpus=N at enumeration-time can be used to disable HT.
 799                  */
 800                 else if (!memcmp(from, "maxcpus=", 8)) {
 801                         extern unsigned int maxcpus;
 802
 803                         maxcpus = simple_strtoul(from + 8, NULL, 0);
 804                 }
 805 #endif
 806
 807 #ifdef CONFIG_ACPI
 808                 /* "acpi=off" disables both ACPI table parsing and interpreter */
 809                 else if (!memcmp(from, "acpi=off", 8)) {
 810                         disable_acpi();
 811                 }
 812
 813                 /* acpi=force to over-ride black-list */
 814                 else if (!memcmp(from, "acpi=force", 10)) {
 815                         acpi_force = 1;
 816                         acpi_ht = 1;
 817                         acpi_disabled = 0;
 818                 }
 819
 820                 /* acpi=strict disables out-of-spec workarounds */
 821                 else if (!memcmp(from, "acpi=strict", 11)) {
 822                         acpi_strict = 1;
 823                 }
 824
 825                 /* Limit ACPI just to boot-time to enable HT */
 826                 else if (!memcmp(from, "acpi=ht", 7)) {
 827                         if (!acpi_force)
 828                                 disable_acpi();
 829                         acpi_ht = 1;
 830                 }
 831
 832                 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
 833                 else if (!memcmp(from, "pci=noacpi", 10)) {
 834                         acpi_disable_pci();
 835                 }
 836                 /* "acpi=noirq" disables ACPI interrupt routing */
 837                 else if (!memcmp(from, "acpi=noirq", 10)) {
 838                         acpi_noirq_set();
 839                 }
 840
 841                 else if (!memcmp(from, "acpi_sci=edge", 13))
 842                         acpi_sci_flags.trigger =  1;
 843
 844                 else if (!memcmp(from, "acpi_sci=level", 14))
 845                         acpi_sci_flags.trigger = 3;
 846
 847                 else if (!memcmp(from, "acpi_sci=high", 13))
 848                         acpi_sci_flags.polarity = 1;
 849
 850                 else if (!memcmp(from, "acpi_sci=low", 12))
 851                         acpi_sci_flags.polarity = 3;
 852
 853 #ifdef CONFIG_X86_IO_APIC
 854                 else if (!memcmp(from, "acpi_skip_timer_override", 24))
 855                         acpi_skip_timer_override = 1;
 856
 857                 if (!memcmp(from, "disable_timer_pin_1", 19))
 858                         disable_timer_pin_1 = 1;
 859                 if (!memcmp(from, "enable_timer_pin_1", 18))
 860                         disable_timer_pin_1 = -1;
 861
 862                 /* disable IO-APIC */
 863                 else if (!memcmp(from, "noapic", 6))
 864                         disable_ioapic_setup();
 865 #endif /* CONFIG_X86_IO_APIC */
 866 #endif /* CONFIG_ACPI */
 867
 868 #ifdef CONFIG_X86_LOCAL_APIC
 869                 /* enable local APIC */
 870                 else if (!memcmp(from, "lapic", 5))
 871                         lapic_enable();
 872
 873                 /* disable local APIC */
 874                 else if (!memcmp(from, "nolapic", 6))
 875                         lapic_disable();
 876 #endif /* CONFIG_X86_LOCAL_APIC */
 877
 878 #ifdef CONFIG_KEXEC
 879                 /* crashkernel=size@addr specifies the location to reserve for
 880                  * a crash kernel.  By reserving this memory we guarantee
 881                  * that linux never set's it up as a DMA target.
 882                  * Useful for holding code to do something appropriate
 883                  * after a kernel panic.
 884                  */
 885                 else if (!memcmp(from, "crashkernel=", 12)) {
 886                         unsigned long size, base;
 887                         size = memparse(from+12, &from);
 888                         if (*from == '@') {
 889                                 base = memparse(from+1, &from);
 890                                 /* FIXME: Do I want a sanity check
 891                                  * to validate the memory range?
 892                                  */
 893                                 crashk_res.start = base;
 894                                 crashk_res.end   = base + size - 1;
 895                         }
 896                 }
 897 #endif
 898 #ifdef CONFIG_PROC_VMCORE
 899                 /* elfcorehdr= specifies the location of elf core header
 900                  * stored by the crashed kernel.
 901                  */
 902                 else if (!memcmp(from, "elfcorehdr=", 11))
 903                         elfcorehdr_addr = memparse(from+11, &from);
 904 #endif
 905
 906                 /*
 907                  * highmem=size forces highmem to be exactly 'size' bytes.
 908                  * This works even on boxes that have no highmem otherwise.
 909                  * This also works to reduce highmem size on bigger boxes.
 910                  */
 911                 else if (!memcmp(from, "highmem=", 8))
 912                         highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;
 913
 914                 /*
 915                  * vmalloc=size forces the vmalloc area to be exactly 'size'
 916                  * bytes. This can be used to increase (or decrease) the
 917                  * vmalloc area - the default is 128m.
 918                  */
 919                 else if (!memcmp(from, "vmalloc=", 8))
 920                         __VMALLOC_RESERVE = memparse(from+8, &from);
 921
 922         next_char:
 923                 c = *(from++);
 924                 if (!c)
 925                         break;
 926                 if (COMMAND_LINE_SIZE <= ++len)
 927                         break;
 928                 *(to++) = c;
 929         }
 930         *to = '\0';
 931         *cmdline_p = command_line;
 932         if (userdef) {
 933                 printk(KERN_INFO "user-defined physical RAM map:\n");
 934                 print_memory_map("user");
 935         }
 936 }
 937
 938 /*
 939  * Callback for efi_memory_walk.
 940  */
 941 static int __init
 942 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
 943 {
 944         unsigned long *max_pfn = arg, pfn;
 945
 946         if (start < end) {
 947                 pfn = PFN_UP(end -1);
 948                 if (pfn > *max_pfn)
 949                         *max_pfn = pfn;
 950         }
 951         return 0;
 952 }
 953
 954 static int __init
 955 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
 956 {
 957         memory_present(0, start, end);
 958         return 0;
 959 }
 960
 961  /*
 962   * This function checks if the entire range <start,end> is mapped with type.
 963   *
 964   * Note: this function only works correct if the e820 table is sorted and
 965   * not-overlapping, which is the case
 966   */
 967 int __init
 968 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
 969 {
 970         u64 start = s;
 971         u64 end = e;
 972         int i;
 973         for (i = 0; i < e820.nr_map; i++) {
 974                 struct e820entry *ei = &e820.map[i];
 975                 if (type && ei->type != type)
 976                         continue;
 977                 /* is the region (part) in overlap with the current region ?*/
 978                 if (ei->addr >= end || ei->addr + ei->size <= start)
 979                         continue;
 980                 /* if the region is at the beginning of <start,end> we move
 981                  * start to the end of the region since it's ok until there
 982                  */
 983                 if (ei->addr <= start)
 984                         start = ei->addr + ei->size;
 985                 /* if start is now at or beyond end, we're done, full
 986                  * coverage */
 987                 if (start >= end)
 988                         return 1; /* we're done */
 989         }
 990         return 0;
 991 }
 992
 993 /*
 994  * Find the highest page frame number we have available
 995  */
 996 void __init find_max_pfn(void)
 997 {
 998         int i;
 999
1000         max_pfn = 0;
1001         if (efi_enabled) {
1002                 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
1003                 efi_memmap_walk(efi_memory_present_wrapper, NULL);
1004                 return;
1005         }
1006
1007         for (i = 0; i < e820.nr_map; i++) {
1008                 unsigned long start, end;
1009                 /* RAM? */
1010                 if (e820.map[i].type != E820_RAM)
1011                         continue;
1012                 start = PFN_UP(e820.map[i].addr);
1013                 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1014                 if (start >= end)
1015                         continue;
1016                 if (end > max_pfn)
1017                         max_pfn = end;
1018                 memory_present(0, start, end);
1019         }
1020 }
1021
1022 /*
1023  * Determine low and high memory ranges:
1024  */
1025 unsigned long __init find_max_low_pfn(void)
1026 {
1027         unsigned long max_low_pfn;
1028
1029         max_low_pfn = max_pfn;
1030         if (max_low_pfn > MAXMEM_PFN) {
1031                 if (highmem_pages == -1)
1032                         highmem_pages = max_pfn - MAXMEM_PFN;
1033                 if (highmem_pages + MAXMEM_PFN < max_pfn)
1034                         max_pfn = MAXMEM_PFN + highmem_pages;
1035                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
1036                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
1037                         highmem_pages = 0;
1038                 }
1039                 max_low_pfn = MAXMEM_PFN;
1040 #ifndef CONFIG_HIGHMEM
1041                 /* Maximum memory usable is what is directly addressable */
1042                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
1043                                         MAXMEM>>20);
1044                 if (max_pfn > MAX_NONPAE_PFN)
1045                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1046                 else
1047                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
1048                 max_pfn = MAXMEM_PFN;
1049 #else /* !CONFIG_HIGHMEM */
1050 #ifndef CONFIG_X86_PAE
1051                 if (max_pfn > MAX_NONPAE_PFN) {
1052                         max_pfn = MAX_NONPAE_PFN;
1053                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
1054                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1055                 }
1056 #endif /* !CONFIG_X86_PAE */
1057 #endif /* !CONFIG_HIGHMEM */
1058         } else {
1059                 if (highmem_pages == -1)
1060                         highmem_pages = 0;
1061 #ifdef CONFIG_HIGHMEM
1062                 if (highmem_pages >= max_pfn) {
1063                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
1064                         highmem_pages = 0;
1065                 }
1066                 if (highmem_pages) {
1067                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
1068                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
1069                                 highmem_pages = 0;
1070                         }
1071                         max_low_pfn -= highmem_pages;
1072                 }
1073 #else
1074                 if (highmem_pages)
1075                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
1076 #endif
1077         }
1078         return max_low_pfn;
1079 }
1080
1081 /*
1082  * Free all available memory for boot time allocation.  Used
1083  * as a callback function by efi_memory_walk()
1084  */
1085
1086 static int __init
1087 free_available_memory(unsigned long start, unsigned long end, void *arg)
1088 {
1089         /* check max_low_pfn */
1090         if (start >= (max_low_pfn << PAGE_SHIFT))
1091                 return 0;
1092         if (end >= (max_low_pfn << PAGE_SHIFT))
1093                 end = max_low_pfn << PAGE_SHIFT;
1094         if (start < end)
1095                 free_bootmem(start, end - start);
1096
1097         return 0;
1098 }
1099 /*
1100  * Register fully available low RAM pages with the bootmem allocator.
1101  */
1102 static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
1103 {
1104         int i;
1105
1106         if (efi_enabled) {
1107                 efi_memmap_walk(free_available_memory, NULL);
1108                 return;
1109         }
1110         for (i = 0; i < e820.nr_map; i++) {
1111                 unsigned long curr_pfn, last_pfn, size;
1112                 /*
1113                  * Reserve usable low memory
1114                  */
1115                 if (e820.map[i].type != E820_RAM)
1116                         continue;
1117                 /*
1118                  * We are rounding up the start address of usable memory:
1119                  */
1120                 curr_pfn = PFN_UP(e820.map[i].addr);
1121                 if (curr_pfn >= max_low_pfn)
1122                         continue;
1123                 /*
1124                  * ... and at the end of the usable range downwards:
1125                  */
1126                 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1127
1128                 if (last_pfn > max_low_pfn)
1129                         last_pfn = max_low_pfn;
1130
1131                 /*
1132                  * .. finally, did all the rounding and playing
1133                  * around just make the area go away?
1134                  */
1135                 if (last_pfn <= curr_pfn)
1136                         continue;
1137
1138                 size = last_pfn - curr_pfn;
1139                 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1140         }
1141 }
1142
1143 /*
1144  * workaround for Dell systems that neglect to reserve EBDA
1145  */
1146 static void __init reserve_ebda_region(void)
1147 {
1148         unsigned int addr;
1149         addr = get_bios_ebda();
1150         if (addr)
1151                 reserve_bootmem(addr, PAGE_SIZE);
1152 }
1153
1154 #ifndef CONFIG_NEED_MULTIPLE_NODES
1155 void __init setup_bootmem_allocator(void);
1156 static unsigned long __init setup_memory(void)
1157 {
1158         /*
1159          * partially used pages are not usable - thus
1160          * we are rounding upwards:
1161          */
1162         min_low_pfn = PFN_UP(init_pg_tables_end);
1163
1164         find_max_pfn();
1165
1166         max_low_pfn = find_max_low_pfn();
1167
1168 #ifdef CONFIG_HIGHMEM
1169         highstart_pfn = highend_pfn = max_pfn;
1170         if (max_pfn > max_low_pfn) {
1171                 highstart_pfn = max_low_pfn;
1172         }
1173         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1174                 pages_to_mb(highend_pfn - highstart_pfn));
1175 #endif
1176         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1177                         pages_to_mb(max_low_pfn));
1178
1179         setup_bootmem_allocator();
1180
1181         return max_low_pfn;
1182 }
1183
1184 void __init zone_sizes_init(void)
1185 {
1186         unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
1187         unsigned int max_dma, low;
1188
1189         max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1190         low = max_low_pfn;
1191
1192         if (low < max_dma)
1193                 zones_size[ZONE_DMA] = low;
1194         else {
1195                 zones_size[ZONE_DMA] = max_dma;
1196                 zones_size[ZONE_NORMAL] = low - max_dma;
1197 #ifdef CONFIG_HIGHMEM
1198                 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1199 #endif
1200         }
1201         free_area_init(zones_size);
1202 }
1203 #else
1204 extern unsigned long __init setup_memory(void);
1205 extern void zone_sizes_init(void);
1206 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1207
1208 void __init setup_bootmem_allocator(void)
1209 {
1210         unsigned long bootmap_size;
1211         /*
1212          * Initialize the boot-time allocator (with low memory only):
1213          */
1214         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1215
1216         register_bootmem_low_pages(max_low_pfn);
1217
1218         /*
1219          * Reserve the bootmem bitmap itself as well. We do this in two
1220          * steps (first step was init_bootmem()) because this catches
1221          * the (very unlikely) case of us accidentally initializing the
1222          * bootmem allocator with an invalid RAM area.
1223          */
1224         reserve_bootmem(__PHYSICAL_START, (PFN_PHYS(min_low_pfn) +
1225                          bootmap_size + PAGE_SIZE-1) - (__PHYSICAL_START));
1226
1227         /*
1228          * reserve physical page 0 - it's a special BIOS page on many boxes,
1229          * enabling clean reboots, SMP operation, laptop functions.
1230          */
1231         reserve_bootmem(0, PAGE_SIZE);
1232
1233         /* reserve EBDA region, it's a 4K region */
1234         reserve_ebda_region();
1235
1236     /* could be an AMD 768MPX chipset. Reserve a page  before VGA to prevent
1237        PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1238        unless you have no PS/2 mouse plugged in. */
1239         if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1240             boot_cpu_data.x86 == 6)
1241              reserve_bootmem(0xa0000 - 4096, 4096);
1242
1243 #ifdef CONFIG_SMP
1244         /*
1245          * But first pinch a few for the stack/trampoline stuff
1246          * FIXME: Don't need the extra page at 4K, but need to fix
1247          * trampoline before removing it. (see the GDT stuff)
1248          */
1249         reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1250 #endif
1251 #ifdef CONFIG_ACPI_SLEEP
1252         /*
1253          * Reserve low memory region for sleep support.
1254          */
1255         acpi_reserve_bootmem();
1256 #endif
1257 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1258         /*
1259          * Find and reserve possible boot-time SMP configuration:
1260          */
1261         find_smp_config();
1262 #endif
1263
1264 #ifdef CONFIG_BLK_DEV_INITRD
1265         if (LOADER_TYPE && INITRD_START) {
1266                 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1267                         reserve_bootmem(INITRD_START, INITRD_SIZE);
1268                         initrd_start =
1269                                 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1270                         initrd_end = initrd_start+INITRD_SIZE;
1271                 }
1272                 else {
1273                         printk(KERN_ERR "initrd extends beyond end of memory "
1274                             "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1275                             INITRD_START + INITRD_SIZE,
1276                             max_low_pfn << PAGE_SHIFT);
1277                         initrd_start = 0;
1278                 }
1279         }
1280 #endif
1281 #ifdef CONFIG_KEXEC
1282         if (crashk_res.start != crashk_res.end)
1283                 reserve_bootmem(crashk_res.start,
1284                         crashk_res.end - crashk_res.start + 1);
1285 #endif
1286 }
1287
1288 /*
1289  * The node 0 pgdat is initialized before all of these because
1290  * it's needed for bootmem.  node>0 pgdats have their virtual
1291  * space allocated before the pagetables are in place to access
1292  * them, so they can't be cleared then.
1293  *
1294  * This should all compile down to nothing when NUMA is off.
1295  */
1296 void __init remapped_pgdat_init(void)
1297 {
1298         int nid;
1299
1300         for_each_online_node(nid) {
1301                 if (nid != 0)
1302                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1303         }
1304 }
1305
1306 /*
1307  * Request address space for all standard RAM and ROM resources
1308  * and also for regions reported as reserved by the e820.
1309  */
1310 static void __init
1311 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1312 {
1313         int i;
1314
1315         probe_roms();
1316         for (i = 0; i < e820.nr_map; i++) {
1317                 struct resource *res;
1318                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1319                         continue;
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                 request_resource(&iomem_resource, res);
1331                 if (e820.map[i].type == E820_RAM) {
1332                         /*
1333                          *  We don't know which RAM region contains kernel data,
1334                          *  so we try it repeatedly and let the resource manager
1335                          *  test it.
1336                          */
1337                         request_resource(res, code_resource);
1338                         request_resource(res, data_resource);
1339 #ifdef CONFIG_KEXEC
1340                         request_resource(res, &crashk_res);
1341 #endif
1342                 }
1343         }
1344 }
1345
1346 /*
1347  * Request address space for all standard resources
1348  *
1349  * This is called just before pcibios_init(), which is also a
1350  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
1351  */
1352 static int __init request_standard_resources(void)
1353 {
1354         int i;
1355
1356         printk("Setting up standard PCI resources\n");
1357         if (efi_enabled)
1358                 efi_initialize_iomem_resources(&code_resource, &data_resource);
1359         else
1360                 legacy_init_iomem_resources(&code_resource, &data_resource);
1361
1362         /* EFI systems may still have VGA */
1363         request_resource(&iomem_resource, &video_ram_resource);
1364
1365         /* request I/O space for devices used on all i[345]86 PCs */
1366         for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1367                 request_resource(&ioport_resource, &standard_io_resources[i]);
1368         return 0;
1369 }
1370
1371 subsys_initcall(request_standard_resources);
1372
1373 static void __init register_memory(void)
1374 {
1375         unsigned long gapstart, gapsize, round;
1376         unsigned long long last;
1377         int i;
1378
1379         /*
1380          * Search for the bigest gap in the low 32 bits of the e820
1381          * memory space.
1382          */
1383         last = 0x100000000ull;
1384         gapstart = 0x10000000;
1385         gapsize = 0x400000;
1386         i = e820.nr_map;
1387         while (--i >= 0) {
1388                 unsigned long long start = e820.map[i].addr;
1389                 unsigned long long end = start + e820.map[i].size;
1390
1391                 /*
1392                  * Since "last" is at most 4GB, we know we'll
1393                  * fit in 32 bits if this condition is true
1394                  */
1395                 if (last > end) {
1396                         unsigned long gap = last - end;
1397
1398                         if (gap > gapsize) {
1399                                 gapsize = gap;
1400                                 gapstart = end;
1401                         }
1402                 }
1403                 if (start < last)
1404                         last = start;
1405         }
1406
1407         /*
1408          * See how much we want to round up: start off with
1409          * rounding to the next 1MB area.
1410          */
1411         round = 0x100000;
1412         while ((gapsize >> 4) > round)
1413                 round += round;
1414         /* Fun with two's complement */
1415         pci_mem_start = (gapstart + round) & -round;
1416
1417         printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1418                 pci_mem_start, gapstart, gapsize);
1419 }
1420
1421 #ifdef CONFIG_MCA
1422 static void set_mca_bus(int x)
1423 {
1424         MCA_bus = x;
1425 }
1426 #else
1427 static void set_mca_bus(int x) { }
1428 #endif
1429
1430 #ifdef CONFIG_SOFTWARE_SUSPEND
1431 static void __init mark_nosave_page_range(unsigned long start, unsigned long end)
1432 {
1433         struct page *page;
1434         while (start <= end) {
1435                 page = pfn_to_page(start);
1436                 SetPageNosave(page);
1437                 start++;
1438         }
1439 }
1440
1441 static void __init e820_nosave_reserved_pages(void)
1442 {
1443         int i;
1444         unsigned long r_start = 0, r_end = 0;
1445
1446         /* Assume e820 map is sorted */
1447         for (i = 0; i < e820.nr_map; i++) {
1448                 struct e820entry *ei = &e820.map[i];
1449                 unsigned long start, end;
1450
1451                 start = PFN_DOWN(ei->addr);
1452                 end = PFN_UP(ei->addr + ei->size);
1453                 if (start >= end)
1454                         continue;
1455                 if (ei->type == E820_RESERVED)
1456                         continue;
1457                 r_end = start;
1458                 /*
1459                  * Highmem 'Reserved' pages are marked as reserved, swsusp
1460                  * will not save/restore them, so we ignore these pages here.
1461                  */
1462                 if (r_end > max_low_pfn)
1463                         r_end = max_low_pfn;
1464                 if (r_end > r_start)
1465                         mark_nosave_page_range(r_start, r_end-1);
1466                 if (r_end >= max_low_pfn)
1467                         break;
1468                 r_start = end;
1469         }
1470 }
1471
1472 static void __init e820_save_acpi_pages(void)
1473 {
1474         int i;
1475
1476         /* Assume e820 map is sorted */
1477         for (i = 0; i < e820.nr_map; i++) {
1478                 struct e820entry *ei = &e820.map[i];
1479                 unsigned long start, end;
1480
1481                 start = ei->addr;
1482                 end = ei->addr + ei->size;
1483                 if (start >= end)
1484                         continue;
1485                 if (ei->type != E820_ACPI && ei->type != E820_NVS)
1486                         continue;
1487                 /*
1488                  * If the region is below max_low_pfn, it will be
1489                  * saved/restored by swsusp follow 'RAM' type.
1490                  */
1491                 if (start < (max_low_pfn << PAGE_SHIFT))
1492                         start = max_low_pfn << PAGE_SHIFT;
1493                 /*
1494                  * Highmem pages (ACPI NVS/Data) are reserved, but swsusp
1495                  * highmem save/restore will not save/restore them. We marked
1496                  * them as arch saveable pages here
1497                  */
1498                 if (end > start)
1499                         swsusp_add_arch_pages(start, end);
1500         }
1501 }
1502
1503 extern char __start_rodata, __end_rodata;
1504 /*
1505  * BIOS reserved region/hole - no save/restore
1506  * ACPI NVS - save/restore
1507  * ACPI Data - this is a little tricky, the mem could be used by OS after OS
1508  * reads tables from the region, but anyway save/restore the memory hasn't any
1509  * side effect and Linux runtime module load/unload might use it.
1510  * kernel rodata - no save/restore (kernel rodata isn't changed)
1511  */
1512 static int __init mark_nosave_pages(void)
1513 {
1514         unsigned long pfn_start, pfn_end;
1515
1516         /* FIXME: provide a version for efi BIOS */
1517         if (efi_enabled)
1518                 return 0;
1519         /* BIOS reserved regions & holes */
1520         e820_nosave_reserved_pages();
1521
1522         /* kernel rodata */
1523         pfn_start = PFN_UP(virt_to_phys(&__start_rodata));
1524         pfn_end = PFN_DOWN(virt_to_phys(&__end_rodata));
1525         mark_nosave_page_range(pfn_start, pfn_end-1);
1526
1527         /* record ACPI Data/NVS as saveable */
1528         e820_save_acpi_pages();
1529
1530         return 0;
1531 }
1532 core_initcall(mark_nosave_pages);
1533 #endif
1534
1535 /*
1536  * Determine if we were loaded by an EFI loader.  If so, then we have also been
1537  * passed the efi memmap, systab, etc., so we should use these data structures
1538  * for initialization.  Note, the efi init code path is determined by the
1539  * global efi_enabled. This allows the same kernel image to be used on existing
1540  * systems (with a traditional BIOS) as well as on EFI systems.
1541  */
1542 void __init setup_arch(char **cmdline_p)
1543 {
1544         unsigned long max_low_pfn;
1545
1546         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1547         pre_setup_arch_hook();
1548         early_cpu_init();
1549
1550         /*
1551          * FIXME: This isn't an official loader_type right
1552          * now but does currently work with elilo.
1553          * If we were configured as an EFI kernel, check to make
1554          * sure that we were loaded correctly from elilo and that
1555          * the system table is valid.  If not, then initialize normally.
1556          */
1557 #ifdef CONFIG_EFI
1558         if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1559                 efi_enabled = 1;
1560 #endif
1561
1562         ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1563         drive_info = DRIVE_INFO;
1564         screen_info = SCREEN_INFO;
1565         edid_info = EDID_INFO;
1566         apm_info.bios = APM_BIOS_INFO;
1567         ist_info = IST_INFO;
1568         saved_videomode = VIDEO_MODE;
1569         if( SYS_DESC_TABLE.length != 0 ) {
1570                 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1571                 machine_id = SYS_DESC_TABLE.table[0];
1572                 machine_submodel_id = SYS_DESC_TABLE.table[1];
1573                 BIOS_revision = SYS_DESC_TABLE.table[2];
1574         }
1575         bootloader_type = LOADER_TYPE;
1576
1577 #ifdef CONFIG_BLK_DEV_RAM
1578         rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1579         rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1580         rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1581 #endif
1582         ARCH_SETUP
1583         if (efi_enabled)
1584                 efi_init();
1585         else {
1586                 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1587                 print_memory_map(machine_specific_memory_setup());
1588         }
1589
1590         copy_edd();
1591
1592         if (!MOUNT_ROOT_RDONLY)
1593                 root_mountflags &= ~MS_RDONLY;
1594         init_mm.start_code = (unsigned long) _text;
1595         init_mm.end_code = (unsigned long) _etext;
1596         init_mm.end_data = (unsigned long) _edata;
1597         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1598
1599         code_resource.start = virt_to_phys(_text);
1600         code_resource.end = virt_to_phys(_etext)-1;
1601         data_resource.start = virt_to_phys(_etext);
1602         data_resource.end = virt_to_phys(_edata)-1;
1603
1604         parse_cmdline_early(cmdline_p);
1605
1606 #ifdef CONFIG_EARLY_PRINTK
1607         {
1608                 char *s = strstr(*cmdline_p, "earlyprintk=");
1609                 if (s) {
1610                         setup_early_printk(strchr(s, '=') + 1);
1611                         printk("early console enabled\n");
1612                 }
1613         }
1614 #endif
1615
1616         max_low_pfn = setup_memory();
1617
1618         /*
1619          * NOTE: before this point _nobody_ is allowed to allocate
1620          * any memory using the bootmem allocator.  Although the
1621          * alloctor is now initialised only the first 8Mb of the kernel
1622          * virtual address space has been mapped.  All allocations before
1623          * paging_init() has completed must use the alloc_bootmem_low_pages()
1624          * variant (which allocates DMA'able memory) and care must be taken
1625          * not to exceed the 8Mb limit.
1626          */
1627
1628 #ifdef CONFIG_SMP
1629         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1630 #endif
1631         paging_init();
1632         remapped_pgdat_init();
1633         sparse_init();
1634         zone_sizes_init();
1635
1636         /*
1637          * NOTE: at this point the bootmem allocator is fully available.
1638          */
1639
1640         dmi_scan_machine();
1641
1642 #ifdef CONFIG_X86_GENERICARCH
1643         generic_apic_probe(*cmdline_p);
1644 #endif
1645         if (efi_enabled)
1646                 efi_map_memmap();
1647
1648 #ifdef CONFIG_ACPI
1649         /*
1650          * Parse the ACPI tables for possible boot-time SMP configuration.
1651          */
1652         acpi_boot_table_init();
1653 #endif
1654
1655 #ifdef CONFIG_X86_IO_APIC
1656         check_acpi_pci();       /* Checks more than just ACPI actually */
1657 #endif
1658
1659 #ifdef CONFIG_ACPI
1660         acpi_boot_init();
1661
1662 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1663         if (def_to_bigsmp)
1664                 printk(KERN_WARNING "More than 8 CPUs detected and "
1665                         "CONFIG_X86_PC cannot handle it.\nUse "
1666                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1667 #endif
1668 #endif
1669 #ifdef CONFIG_X86_LOCAL_APIC
1670         if (smp_found_config)
1671                 get_smp_config();
1672 #endif
1673
1674         register_memory();
1675
1676 #ifdef CONFIG_VT
1677 #if defined(CONFIG_VGA_CONSOLE)
1678         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1679                 conswitchp = &vga_con;
1680 #elif defined(CONFIG_DUMMY_CONSOLE)
1681         conswitchp = &dummy_con;
1682 #endif
1683 #endif
1684 }
1685
1686 static __init int add_pcspkr(void)
1687 {
1688         struct platform_device *pd;
1689         int ret;
1690
1691         pd = platform_device_alloc("pcspkr", -1);
1692         if (!pd)
1693                 return -ENOMEM;
1694
1695         ret = platform_device_add(pd);
1696         if (ret)
1697                 platform_device_put(pd);
1698
1699         return ret;
1700 }
1701 device_initcall(add_pcspkr);
1702
1703 /*
1704  * Local Variables:
1705  * mode:c
1706  * c-file-style:"k&r"
1707  * c-basic-offset:8
1708  * End:
1709  */