arch/x86/kernel/setup.c

   1 /*
   2  *  Copyright (C) 1995  Linus Torvalds
   3  *
   4  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   5  *
   6  *  Memory region support
   7  *      David Parsons <orc@pell.chi.il.us>, July-August 1999
   8  *
   9  *  Added E820 sanitization routine (removes overlapping memory regions);
  10  *  Brian Moyle <bmoyle@mvista.com>, February 2001
  11  *
  12  * Moved CPU detection code to cpu/${cpu}.c
  13  *    Patrick Mochel <mochel@osdl.org>, March 2002
  14  *
  15  *  Provisions for empty E820 memory regions (reported by certain BIOSes).
  16  *  Alex Achenbach <xela@slit.de>, December 2002.
  17  *
  18  */
  19
  20 /*
  21  * This file handles the architecture-dependent parts of initialization
  22  */
  23
  24 #include <linux/sched.h>
  25 #include <linux/mm.h>
  26 #include <linux/mmzone.h>
  27 #include <linux/screen_info.h>
  28 #include <linux/ioport.h>
  29 #include <linux/acpi.h>
  30 #include <linux/apm_bios.h>
  31 #include <linux/initrd.h>
  32 #include <linux/bootmem.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/console.h>
  35 #include <linux/mca.h>
  36 #include <linux/root_dev.h>
  37 #include <linux/highmem.h>
  38 #include <linux/module.h>
  39 #include <linux/efi.h>
  40 #include <linux/init.h>
  41 #include <linux/edd.h>
  42 #include <linux/iscsi_ibft.h>
  43 #include <linux/nodemask.h>
  44 #include <linux/kexec.h>
  45 #include <linux/dmi.h>
  46 #include <linux/pfn.h>
  47 #include <linux/pci.h>
  48 #include <asm/pci-direct.h>
  49 #include <linux/init_ohci1394_dma.h>
  50 #include <linux/kvm_para.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/kernel.h>
  54 #include <linux/stddef.h>
  55 #include <linux/unistd.h>
  56 #include <linux/ptrace.h>
  57 #include <linux/slab.h>
  58 #include <linux/user.h>
  59 #include <linux/delay.h>
  60
  61 #include <linux/kallsyms.h>
  62 #include <linux/cpufreq.h>
  63 #include <linux/dma-mapping.h>
  64 #include <linux/ctype.h>
  65 #include <linux/uaccess.h>
  66
  67 #include <linux/percpu.h>
  68 #include <linux/crash_dump.h>
  69
  70 #include <video/edid.h>
  71
  72 #include <asm/mtrr.h>
  73 #include <asm/apic.h>
  74 #include <asm/e820.h>
  75 #include <asm/mpspec.h>
  76 #include <asm/setup.h>
  77 #include <asm/efi.h>
  78 #include <asm/timer.h>
  79 #include <asm/i8259.h>
  80 #include <asm/sections.h>
  81 #include <asm/dmi.h>
  82 #include <asm/io_apic.h>
  83 #include <asm/ist.h>
  84 #include <asm/vmi.h>
  85 #include <asm/setup_arch.h>
  86 #include <asm/bios_ebda.h>
  87 #include <asm/cacheflush.h>
  88 #include <asm/processor.h>
  89 #include <asm/bugs.h>
  90
  91 #include <asm/system.h>
  92 #include <asm/vsyscall.h>
  93 #include <asm/cpu.h>
  94 #include <asm/desc.h>
  95 #include <asm/dma.h>
  96 #include <asm/iommu.h>
  97 #include <asm/gart.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/proto.h>
 100
 101 #include <asm/paravirt.h>
 102 #include <asm/hypervisor.h>
 103
 104 #include <asm/percpu.h>
 105 #include <asm/topology.h>
 106 #include <asm/apicdef.h>
 107 #ifdef CONFIG_X86_64
 108 #include <asm/numa_64.h>
 109 #endif
 110
 111 /*
 112  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
 113  * The direct mapping extends to max_pfn_mapped, so that we can directly access
 114  * apertures, ACPI and other tables without having to play with fixmaps.
 115  */
 116 unsigned long max_low_pfn_mapped;
 117 unsigned long max_pfn_mapped;
 118
 119 RESERVE_BRK(dmi_alloc, 65536);
 120
 121 unsigned int boot_cpu_id __read_mostly;
 122
 123 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 124 unsigned long _brk_end = (unsigned long)__brk_base;
 125
 126 #ifdef CONFIG_X86_64
 127 int default_cpu_present_to_apicid(int mps_cpu)
 128 {
 129         return __default_cpu_present_to_apicid(mps_cpu);
 130 }
 131
 132 int default_check_phys_apicid_present(int boot_cpu_physical_apicid)
 133 {
 134         return __default_check_phys_apicid_present(boot_cpu_physical_apicid);
 135 }
 136 #endif
 137
 138 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 139 struct boot_params __initdata boot_params;
 140 #else
 141 struct boot_params boot_params;
 142 #endif
 143
 144 /*
 145  * Machine setup..
 146  */
 147 static struct resource data_resource = {
 148         .name   = "Kernel data",
 149         .start  = 0,
 150         .end    = 0,
 151         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 152 };
 153
 154 static struct resource code_resource = {
 155         .name   = "Kernel code",
 156         .start  = 0,
 157         .end    = 0,
 158         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 159 };
 160
 161 static struct resource bss_resource = {
 162         .name   = "Kernel bss",
 163         .start  = 0,
 164         .end    = 0,
 165         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 166 };
 167
 168
 169 #ifdef CONFIG_X86_32
 170 /* cpu data as detected by the assembly code in head.S */
 171 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 172 /* common cpu data for all cpus */
 173 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 174 EXPORT_SYMBOL(boot_cpu_data);
 175 static void set_mca_bus(int x)
 176 {
 177 #ifdef CONFIG_MCA
 178         MCA_bus = x;
 179 #endif
 180 }
 181
 182 unsigned int def_to_bigsmp;
 183
 184 /* for MCA, but anyone else can use it if they want */
 185 unsigned int machine_id;
 186 unsigned int machine_submodel_id;
 187 unsigned int BIOS_revision;
 188
 189 struct apm_info apm_info;
 190 EXPORT_SYMBOL(apm_info);
 191
 192 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 193         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 194 struct ist_info ist_info;
 195 EXPORT_SYMBOL(ist_info);
 196 #else
 197 struct ist_info ist_info;
 198 #endif
 199
 200 #else
 201 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 202         .x86_phys_bits = MAX_PHYSMEM_BITS,
 203 };
 204 EXPORT_SYMBOL(boot_cpu_data);
 205 #endif
 206
 207
 208 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 209 unsigned long mmu_cr4_features;
 210 #else
 211 unsigned long mmu_cr4_features = X86_CR4_PAE;
 212 #endif
 213
 214 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
 215 int bootloader_type, bootloader_version;
 216
 217 /*
 218  * Setup options
 219  */
 220 struct screen_info screen_info;
 221 EXPORT_SYMBOL(screen_info);
 222 struct edid_info edid_info;
 223 EXPORT_SYMBOL_GPL(edid_info);
 224
 225 extern int root_mountflags;
 226
 227 unsigned long saved_video_mode;
 228
 229 #define RAMDISK_IMAGE_START_MASK        0x07FF
 230 #define RAMDISK_PROMPT_FLAG             0x8000
 231 #define RAMDISK_LOAD_FLAG               0x4000
 232
 233 static char __initdata command_line[COMMAND_LINE_SIZE];
 234 #ifdef CONFIG_CMDLINE_BOOL
 235 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 236 #endif
 237
 238 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 239 struct edd edd;
 240 #ifdef CONFIG_EDD_MODULE
 241 EXPORT_SYMBOL(edd);
 242 #endif
 243 /**
 244  * copy_edd() - Copy the BIOS EDD information
 245  *              from boot_params into a safe place.
 246  *
 247  */
 248 static inline void copy_edd(void)
 249 {
 250      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 251             sizeof(edd.mbr_signature));
 252      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 253      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 254      edd.edd_info_nr = boot_params.eddbuf_entries;
 255 }
 256 #else
 257 static inline void copy_edd(void)
 258 {
 259 }
 260 #endif
 261
 262 void * __init extend_brk(size_t size, size_t align)
 263 {
 264         size_t mask = align - 1;
 265         void *ret;
 266
 267         BUG_ON(_brk_start == 0);
 268         BUG_ON(align & mask);
 269
 270         _brk_end = (_brk_end + mask) & ~mask;
 271         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 272
 273         ret = (void *)_brk_end;
 274         _brk_end += size;
 275
 276         memset(ret, 0, size);
 277
 278         return ret;
 279 }
 280
 281 #ifdef CONFIG_X86_64
 282 static void __init init_gbpages(void)
 283 {
 284         if (direct_gbpages && cpu_has_gbpages)
 285                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 286         else
 287                 direct_gbpages = 0;
 288 }
 289 #else
 290 static inline void init_gbpages(void)
 291 {
 292 }
 293 #endif
 294
 295 static void __init reserve_brk(void)
 296 {
 297         if (_brk_end > _brk_start)
 298                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 299
 300         /* Mark brk area as locked down and no longer taking any
 301            new allocations */
 302         _brk_start = 0;
 303 }
 304
 305 #ifdef CONFIG_BLK_DEV_INITRD
 306
 307 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 308 static void __init relocate_initrd(void)
 309 {
 310
 311         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 312         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 313         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 314         u64 ramdisk_here;
 315         unsigned long slop, clen, mapaddr;
 316         char *p, *q;
 317
 318         /* We need to move the initrd down into lowmem */
 319         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 320                                          PAGE_SIZE);
 321
 322         if (ramdisk_here == -1ULL)
 323                 panic("Cannot find place for new RAMDISK of size %lld\n",
 324                          ramdisk_size);
 325
 326         /* Note: this includes all the lowmem currently occupied by
 327            the initrd, we rely on that fact to keep the data intact. */
 328         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 329                          "NEW RAMDISK");
 330         initrd_start = ramdisk_here + PAGE_OFFSET;
 331         initrd_end   = initrd_start + ramdisk_size;
 332         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 333                          ramdisk_here, ramdisk_here + ramdisk_size);
 334
 335         q = (char *)initrd_start;
 336
 337         /* Copy any lowmem portion of the initrd */
 338         if (ramdisk_image < end_of_lowmem) {
 339                 clen = end_of_lowmem - ramdisk_image;
 340                 p = (char *)__va(ramdisk_image);
 341                 memcpy(q, p, clen);
 342                 q += clen;
 343                 ramdisk_image += clen;
 344                 ramdisk_size  -= clen;
 345         }
 346
 347         /* Copy the highmem portion of the initrd */
 348         while (ramdisk_size) {
 349                 slop = ramdisk_image & ~PAGE_MASK;
 350                 clen = ramdisk_size;
 351                 if (clen > MAX_MAP_CHUNK-slop)
 352                         clen = MAX_MAP_CHUNK-slop;
 353                 mapaddr = ramdisk_image & PAGE_MASK;
 354                 p = early_memremap(mapaddr, clen+slop);
 355                 memcpy(q, p+slop, clen);
 356                 early_iounmap(p, clen+slop);
 357                 q += clen;
 358                 ramdisk_image += clen;
 359                 ramdisk_size  -= clen;
 360         }
 361         /* high pages is not converted by early_res_to_bootmem */
 362         ramdisk_image = boot_params.hdr.ramdisk_image;
 363         ramdisk_size  = boot_params.hdr.ramdisk_size;
 364         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 365                 " %08llx - %08llx\n",
 366                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 367                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 368 }
 369
 370 static void __init reserve_initrd(void)
 371 {
 372         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 373         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 374         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 375         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 376
 377         if (!boot_params.hdr.type_of_loader ||
 378             !ramdisk_image || !ramdisk_size)
 379                 return;         /* No initrd provided by bootloader */
 380
 381         initrd_start = 0;
 382
 383         if (ramdisk_size >= (end_of_lowmem>>1)) {
 384                 free_early(ramdisk_image, ramdisk_end);
 385                 printk(KERN_ERR "initrd too large to handle, "
 386                        "disabling initrd\n");
 387                 return;
 388         }
 389
 390         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 391                         ramdisk_end);
 392
 393
 394         if (ramdisk_end <= end_of_lowmem) {
 395                 /* All in lowmem, easy case */
 396                 /*
 397                  * don't need to reserve again, already reserved early
 398                  * in i386_start_kernel
 399                  */
 400                 initrd_start = ramdisk_image + PAGE_OFFSET;
 401                 initrd_end = initrd_start + ramdisk_size;
 402                 return;
 403         }
 404
 405         relocate_initrd();
 406
 407         free_early(ramdisk_image, ramdisk_end);
 408 }
 409 #else
 410 static void __init reserve_initrd(void)
 411 {
 412 }
 413 #endif /* CONFIG_BLK_DEV_INITRD */
 414
 415 static void __init parse_setup_data(void)
 416 {
 417         struct setup_data *data;
 418         u64 pa_data;
 419
 420         if (boot_params.hdr.version < 0x0209)
 421                 return;
 422         pa_data = boot_params.hdr.setup_data;
 423         while (pa_data) {
 424                 data = early_memremap(pa_data, PAGE_SIZE);
 425                 switch (data->type) {
 426                 case SETUP_E820_EXT:
 427                         parse_e820_ext(data, pa_data);
 428                         break;
 429                 default:
 430                         break;
 431                 }
 432                 pa_data = data->next;
 433                 early_iounmap(data, PAGE_SIZE);
 434         }
 435 }
 436
 437 static void __init e820_reserve_setup_data(void)
 438 {
 439         struct setup_data *data;
 440         u64 pa_data;
 441         int found = 0;
 442
 443         if (boot_params.hdr.version < 0x0209)
 444                 return;
 445         pa_data = boot_params.hdr.setup_data;
 446         while (pa_data) {
 447                 data = early_memremap(pa_data, sizeof(*data));
 448                 e820_update_range(pa_data, sizeof(*data)+data->len,
 449                          E820_RAM, E820_RESERVED_KERN);
 450                 found = 1;
 451                 pa_data = data->next;
 452                 early_iounmap(data, sizeof(*data));
 453         }
 454         if (!found)
 455                 return;
 456
 457         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 458         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 459         printk(KERN_INFO "extended physical RAM map:\n");
 460         e820_print_map("reserve setup_data");
 461 }
 462
 463 static void __init reserve_early_setup_data(void)
 464 {
 465         struct setup_data *data;
 466         u64 pa_data;
 467         char buf[32];
 468
 469         if (boot_params.hdr.version < 0x0209)
 470                 return;
 471         pa_data = boot_params.hdr.setup_data;
 472         while (pa_data) {
 473                 data = early_memremap(pa_data, sizeof(*data));
 474                 sprintf(buf, "setup data %x", data->type);
 475                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 476                 pa_data = data->next;
 477                 early_iounmap(data, sizeof(*data));
 478         }
 479 }
 480
 481 /*
 482  * --------- Crashkernel reservation ------------------------------
 483  */
 484
 485 #ifdef CONFIG_KEXEC
 486
 487 /**
 488  * Reserve @size bytes of crashkernel memory at any suitable offset.
 489  *
 490  * @size: Size of the crashkernel memory to reserve.
 491  * Returns the base address on success, and -1ULL on failure.
 492  */
 493 static
 494 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 495 {
 496         const unsigned long long alignment = 16<<20;    /* 16M */
 497         unsigned long long start = 0LL;
 498
 499         while (1) {
 500                 int ret;
 501
 502                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 503                 if (start == -1ULL)
 504                         return start;
 505
 506                 /* try to reserve it */
 507                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 508                 if (ret >= 0)
 509                         return start;
 510
 511                 start += alignment;
 512         }
 513 }
 514
 515 static inline unsigned long long get_total_mem(void)
 516 {
 517         unsigned long long total;
 518
 519         total = max_low_pfn - min_low_pfn;
 520 #ifdef CONFIG_HIGHMEM
 521         total += highend_pfn - highstart_pfn;
 522 #endif
 523
 524         return total << PAGE_SHIFT;
 525 }
 526
 527 static void __init reserve_crashkernel(void)
 528 {
 529         unsigned long long total_mem;
 530         unsigned long long crash_size, crash_base;
 531         int ret;
 532
 533         total_mem = get_total_mem();
 534
 535         ret = parse_crashkernel(boot_command_line, total_mem,
 536                         &crash_size, &crash_base);
 537         if (ret != 0 || crash_size <= 0)
 538                 return;
 539
 540         /* 0 means: find the address automatically */
 541         if (crash_base <= 0) {
 542                 crash_base = find_and_reserve_crashkernel(crash_size);
 543                 if (crash_base == -1ULL) {
 544                         pr_info("crashkernel reservation failed. "
 545                                 "No suitable area found.\n");
 546                         return;
 547                 }
 548         } else {
 549                 ret = reserve_bootmem_generic(crash_base, crash_size,
 550                                         BOOTMEM_EXCLUSIVE);
 551                 if (ret < 0) {
 552                         pr_info("crashkernel reservation failed - "
 553                                 "memory is in use\n");
 554                         return;
 555                 }
 556         }
 557
 558         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 559                         "for crashkernel (System RAM: %ldMB)\n",
 560                         (unsigned long)(crash_size >> 20),
 561                         (unsigned long)(crash_base >> 20),
 562                         (unsigned long)(total_mem >> 20));
 563
 564         crashk_res.start = crash_base;
 565         crashk_res.end   = crash_base + crash_size - 1;
 566         insert_resource(&iomem_resource, &crashk_res);
 567 }
 568 #else
 569 static void __init reserve_crashkernel(void)
 570 {
 571 }
 572 #endif
 573
 574 static struct resource standard_io_resources[] = {
 575         { .name = "dma1", .start = 0x00, .end = 0x1f,
 576                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 577         { .name = "pic1", .start = 0x20, .end = 0x21,
 578                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 579         { .name = "timer0", .start = 0x40, .end = 0x43,
 580                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 581         { .name = "timer1", .start = 0x50, .end = 0x53,
 582                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 583         { .name = "keyboard", .start = 0x60, .end = 0x60,
 584                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 585         { .name = "keyboard", .start = 0x64, .end = 0x64,
 586                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 587         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 588                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 589         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 590                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 591         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 592                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 593         { .name = "fpu", .start = 0xf0, .end = 0xff,
 594                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 595 };
 596
 597 void __init reserve_standard_io_resources(void)
 598 {
 599         int i;
 600
 601         /* request I/O space for devices used on all i[345]86 PCs */
 602         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 603                 request_resource(&ioport_resource, &standard_io_resources[i]);
 604
 605 }
 606
 607 /*
 608  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 609  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 610  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 611  */
 612
 613 #ifdef CONFIG_CRASH_DUMP
 614 /* elfcorehdr= specifies the location of elf core header
 615  * stored by the crashed kernel. This option will be passed
 616  * by kexec loader to the capture kernel.
 617  */
 618 static int __init setup_elfcorehdr(char *arg)
 619 {
 620         char *end;
 621         if (!arg)
 622                 return -EINVAL;
 623         elfcorehdr_addr = memparse(arg, &end);
 624         return end > arg ? 0 : -EINVAL;
 625 }
 626 early_param("elfcorehdr", setup_elfcorehdr);
 627 #endif
 628
 629 static struct x86_quirks default_x86_quirks __initdata;
 630
 631 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 632
 633 #ifdef CONFIG_X86_RESERVE_LOW_64K
 634 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 635 {
 636         printk(KERN_NOTICE
 637                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 638                 d->ident);
 639
 640         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 641         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 642
 643         return 0;
 644 }
 645 #endif
 646
 647 /* List of systems that have known low memory corruption BIOS problems */
 648 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 649 #ifdef CONFIG_X86_RESERVE_LOW_64K
 650         {
 651                 .callback = dmi_low_memory_corruption,
 652                 .ident = "AMI BIOS",
 653                 .matches = {
 654                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 655                 },
 656         },
 657         {
 658                 .callback = dmi_low_memory_corruption,
 659                 .ident = "Phoenix BIOS",
 660                 .matches = {
 661                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 662                 },
 663         },
 664         {
 665         /*
 666          * AMI BIOS with low memory corruption was found on Intel DG45ID board.
 667          * It hase different DMI_BIOS_VENDOR = "Intel Corp.", for now we will
 668          * match only DMI_BOARD_NAME and see if there is more bad products
 669          * with this vendor.
 670          */
 671                 .callback = dmi_low_memory_corruption,
 672                 .ident = "AMI BIOS",
 673                 .matches = {
 674                         DMI_MATCH(DMI_BOARD_NAME, "DG45ID"),
 675                 },
 676         },
 677 #endif
 678         {}
 679 };
 680
 681 /*
 682  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 683  * passed the efi memmap, systab, etc., so we should use these data structures
 684  * for initialization.  Note, the efi init code path is determined by the
 685  * global efi_enabled. This allows the same kernel image to be used on existing
 686  * systems (with a traditional BIOS) as well as on EFI systems.
 687  */
 688 /*
 689  * setup_arch - architecture-specific boot-time initializations
 690  *
 691  * Note: On x86_64, fixmaps are ready for use even before this is called.
 692  */
 693
 694 void __init setup_arch(char **cmdline_p)
 695 {
 696 #ifdef CONFIG_X86_32
 697         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 698         visws_early_detect();
 699 #else
 700         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 701 #endif
 702
 703         /* VMI may relocate the fixmap; do this before touching ioremap area */
 704         vmi_init();
 705
 706         early_cpu_init();
 707         early_ioremap_init();
 708
 709         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 710         screen_info = boot_params.screen_info;
 711         edid_info = boot_params.edid_info;
 712 #ifdef CONFIG_X86_32
 713         apm_info.bios = boot_params.apm_bios_info;
 714         ist_info = boot_params.ist_info;
 715         if (boot_params.sys_desc_table.length != 0) {
 716                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 717                 machine_id = boot_params.sys_desc_table.table[0];
 718                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 719                 BIOS_revision = boot_params.sys_desc_table.table[2];
 720         }
 721 #endif
 722         saved_video_mode = boot_params.hdr.vid_mode;
 723         bootloader_type = boot_params.hdr.type_of_loader;
 724         if ((bootloader_type >> 4) == 0xe) {
 725                 bootloader_type &= 0xf;
 726                 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
 727         }
 728         bootloader_version  = bootloader_type & 0xf;
 729         bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
 730
 731 #ifdef CONFIG_BLK_DEV_RAM
 732         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 733         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 734         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 735 #endif
 736 #ifdef CONFIG_EFI
 737         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 738 #ifdef CONFIG_X86_32
 739                      "EL32",
 740 #else
 741                      "EL64",
 742 #endif
 743          4)) {
 744                 efi_enabled = 1;
 745                 efi_reserve_early();
 746         }
 747 #endif
 748
 749         x86_init.oem.arch_setup();
 750
 751         setup_memory_map();
 752         parse_setup_data();
 753         /* update the e820_saved too */
 754         e820_reserve_setup_data();
 755
 756         copy_edd();
 757
 758         if (!boot_params.hdr.root_flags)
 759                 root_mountflags &= ~MS_RDONLY;
 760         init_mm.start_code = (unsigned long) _text;
 761         init_mm.end_code = (unsigned long) _etext;
 762         init_mm.end_data = (unsigned long) _edata;
 763         init_mm.brk = _brk_end;
 764
 765         code_resource.start = virt_to_phys(_text);
 766         code_resource.end = virt_to_phys(_etext)-1;
 767         data_resource.start = virt_to_phys(_etext);
 768         data_resource.end = virt_to_phys(_edata)-1;
 769         bss_resource.start = virt_to_phys(&__bss_start);
 770         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 771
 772 #ifdef CONFIG_CMDLINE_BOOL
 773 #ifdef CONFIG_CMDLINE_OVERRIDE
 774         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 775 #else
 776         if (builtin_cmdline[0]) {
 777                 /* append boot loader cmdline to builtin */
 778                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 779                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 780                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 781         }
 782 #endif
 783 #endif
 784
 785         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 786         *cmdline_p = command_line;
 787
 788         parse_early_param();
 789
 790 #ifdef CONFIG_X86_64
 791         check_efer();
 792 #endif
 793
 794         /* Must be before kernel pagetables are setup */
 795         vmi_activate();
 796
 797         /* after early param, so could get panic from serial */
 798         reserve_early_setup_data();
 799
 800         if (acpi_mps_check()) {
 801 #ifdef CONFIG_X86_LOCAL_APIC
 802                 disable_apic = 1;
 803 #endif
 804                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 805         }
 806
 807 #ifdef CONFIG_PCI
 808         if (pci_early_dump_regs)
 809                 early_dump_pci_devices();
 810 #endif
 811
 812         finish_e820_parsing();
 813
 814         if (efi_enabled)
 815                 efi_init();
 816
 817         dmi_scan_machine();
 818
 819         dmi_check_system(bad_bios_dmi_table);
 820
 821         /*
 822          * VMware detection requires dmi to be available, so this
 823          * needs to be done after dmi_scan_machine, for the BP.
 824          */
 825         init_hypervisor(&boot_cpu_data);
 826
 827         x86_init.resources.probe_roms();
 828
 829         /* after parse_early_param, so could debug it */
 830         insert_resource(&iomem_resource, &code_resource);
 831         insert_resource(&iomem_resource, &data_resource);
 832         insert_resource(&iomem_resource, &bss_resource);
 833
 834
 835 #ifdef CONFIG_X86_32
 836         if (ppro_with_ram_bug()) {
 837                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 838                                   E820_RESERVED);
 839                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 840                 printk(KERN_INFO "fixed physical RAM map:\n");
 841                 e820_print_map("bad_ppro");
 842         }
 843 #else
 844         early_gart_iommu_check();
 845 #endif
 846
 847         /*
 848          * partially used pages are not usable - thus
 849          * we are rounding upwards:
 850          */
 851         max_pfn = e820_end_of_ram_pfn();
 852
 853         /* preallocate 4k for mptable mpc */
 854         early_reserve_e820_mpc_new();
 855         /* update e820 for memory not covered by WB MTRRs */
 856         mtrr_bp_init();
 857         if (mtrr_trim_uncached_memory(max_pfn))
 858                 max_pfn = e820_end_of_ram_pfn();
 859
 860 #ifdef CONFIG_X86_32
 861         /* max_low_pfn get updated here */
 862         find_low_pfn_range();
 863 #else
 864         num_physpages = max_pfn;
 865
 866         check_x2apic();
 867
 868         /* How many end-of-memory variables you have, grandma! */
 869         /* need this before calling reserve_initrd */
 870         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 871                 max_low_pfn = e820_end_of_low_ram_pfn();
 872         else
 873                 max_low_pfn = max_pfn;
 874
 875         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 876         max_pfn_mapped = KERNEL_IMAGE_SIZE >> PAGE_SHIFT;
 877 #endif
 878
 879 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 880         setup_bios_corruption_check();
 881 #endif
 882
 883         printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
 884                         max_pfn_mapped<<PAGE_SHIFT);
 885
 886         reserve_brk();
 887
 888         init_gbpages();
 889
 890         /* max_pfn_mapped is updated here */
 891         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 892         max_pfn_mapped = max_low_pfn_mapped;
 893
 894 #ifdef CONFIG_X86_64
 895         if (max_pfn > max_low_pfn) {
 896                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 897                                                      max_pfn<<PAGE_SHIFT);
 898                 /* can we preseve max_low_pfn ?*/
 899                 max_low_pfn = max_pfn;
 900         }
 901 #endif
 902
 903         /*
 904          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 905          */
 906
 907 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 908         if (init_ohci1394_dma_early)
 909                 init_ohci1394_dma_on_all_controllers();
 910 #endif
 911
 912         reserve_initrd();
 913
 914         vsmp_init();
 915
 916         io_delay_init();
 917
 918         /*
 919          * Parse the ACPI tables for possible boot-time SMP configuration.
 920          */
 921         acpi_boot_table_init();
 922
 923         early_acpi_boot_init();
 924
 925 #ifdef CONFIG_ACPI_NUMA
 926         /*
 927          * Parse SRAT to discover nodes.
 928          */
 929         acpi_numa_init();
 930 #endif
 931
 932         initmem_init(0, max_pfn);
 933
 934 #ifdef CONFIG_ACPI_SLEEP
 935         /*
 936          * Reserve low memory region for sleep support.
 937          */
 938         acpi_reserve_bootmem();
 939 #endif
 940         /*
 941          * Find and reserve possible boot-time SMP configuration:
 942          */
 943         find_smp_config();
 944
 945         reserve_crashkernel();
 946
 947 #ifdef CONFIG_X86_64
 948         /*
 949          * dma32_reserve_bootmem() allocates bootmem which may conflict
 950          * with the crashkernel command line, so do that after
 951          * reserve_crashkernel()
 952          */
 953         dma32_reserve_bootmem();
 954 #endif
 955
 956         reserve_ibft_region();
 957
 958 #ifdef CONFIG_KVM_CLOCK
 959         kvmclock_init();
 960 #endif
 961
 962         x86_init.paging.pagetable_setup_start(swapper_pg_dir);
 963         paging_init();
 964         x86_init.paging.pagetable_setup_done(swapper_pg_dir);
 965
 966 #ifdef CONFIG_X86_64
 967         map_vsyscall();
 968 #endif
 969
 970         generic_apic_probe();
 971
 972         early_quirks();
 973
 974         /*
 975          * Read APIC and some other early information from ACPI tables.
 976          */
 977         acpi_boot_init();
 978
 979         /*
 980          * get boot-time SMP configuration:
 981          */
 982         if (smp_found_config)
 983                 get_smp_config();
 984
 985         prefill_possible_map();
 986
 987 #ifdef CONFIG_X86_64
 988         init_cpu_to_node();
 989 #endif
 990
 991         init_apic_mappings();
 992         ioapic_init_mappings();
 993
 994         /* need to wait for io_apic is mapped */
 995         probe_nr_irqs_gsi();
 996
 997         kvm_guest_init();
 998
 999         e820_reserve_resources();
1000         e820_mark_nosave_regions(max_low_pfn);
1001
1002         x86_init.resources.reserve_resources();
1003
1004         e820_setup_gap();
1005
1006 #ifdef CONFIG_VT
1007 #if defined(CONFIG_VGA_CONSOLE)
1008         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1009                 conswitchp = &vga_con;
1010 #elif defined(CONFIG_DUMMY_CONSOLE)
1011         conswitchp = &dummy_con;
1012 #endif
1013 #endif
1014         x86_init.oem.banner();
1015 }
1016
1017 #ifdef CONFIG_X86_32
1018
1019 static struct irqaction irq0  = {
1020         .handler = timer_interrupt,
1021         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1022         .name = "timer"
1023 };
1024
1025 /**
1026  * x86_quirk_pre_time_init - do any specific initialisations before.
1027  *
1028  **/
1029 void __init x86_quirk_pre_time_init(void)
1030 {
1031         if (x86_quirks->arch_pre_time_init)
1032                 x86_quirks->arch_pre_time_init();
1033 }
1034
1035 /**
1036  * x86_quirk_time_init - do any specific initialisations for the system timer.
1037  *
1038  * Description:
1039  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1040  *      in irq_vectors.h:TIMER_IRQ
1041  **/
1042 void __init x86_quirk_time_init(void)
1043 {
1044         if (x86_quirks->arch_time_init) {
1045                 /*
1046                  * A nonzero return code does not mean failure, it means
1047                  * that the architecture quirk does not want any
1048                  * generic (timer) setup to be performed after this:
1049                  */
1050                 if (x86_quirks->arch_time_init())
1051                         return;
1052         }
1053
1054         irq0.mask = cpumask_of_cpu(0);
1055         setup_irq(0, &irq0);
1056 }
1057
1058 static struct resource video_ram_resource = {
1059         .name   = "Video RAM area",
1060         .start  = 0xa0000,
1061         .end    = 0xbffff,
1062         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
1063 };
1064
1065 void __init i386_reserve_resources(void)
1066 {
1067         request_resource(&iomem_resource, &video_ram_resource);
1068         reserve_standard_io_resources();
1069 }
1070
1071 #endif /* CONFIG_X86_32 */