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 #ifndef ARCH_SETUP
 112 #define ARCH_SETUP
 113 #endif
 114
 115 /*
 116  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
 117  * The direct mapping extends to max_pfn_mapped, so that we can directly access
 118  * apertures, ACPI and other tables without having to play with fixmaps.
 119  */
 120 unsigned long max_low_pfn_mapped;
 121 unsigned long max_pfn_mapped;
 122
 123 RESERVE_BRK(dmi_alloc, 65536);
 124
 125 unsigned int boot_cpu_id __read_mostly;
 126
 127 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 128 unsigned long _brk_end = (unsigned long)__brk_base;
 129
 130 #ifdef CONFIG_X86_64
 131 int default_cpu_present_to_apicid(int mps_cpu)
 132 {
 133         return __default_cpu_present_to_apicid(mps_cpu);
 134 }
 135
 136 int default_check_phys_apicid_present(int boot_cpu_physical_apicid)
 137 {
 138         return __default_check_phys_apicid_present(boot_cpu_physical_apicid);
 139 }
 140 #endif
 141
 142 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 143 struct boot_params __initdata boot_params;
 144 #else
 145 struct boot_params boot_params;
 146 #endif
 147
 148 /*
 149  * Machine setup..
 150  */
 151 static struct resource data_resource = {
 152         .name   = "Kernel data",
 153         .start  = 0,
 154         .end    = 0,
 155         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 156 };
 157
 158 static struct resource code_resource = {
 159         .name   = "Kernel code",
 160         .start  = 0,
 161         .end    = 0,
 162         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 163 };
 164
 165 static struct resource bss_resource = {
 166         .name   = "Kernel bss",
 167         .start  = 0,
 168         .end    = 0,
 169         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 170 };
 171
 172
 173 #ifdef CONFIG_X86_32
 174 static struct resource video_ram_resource = {
 175         .name   = "Video RAM area",
 176         .start  = 0xa0000,
 177         .end    = 0xbffff,
 178         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 179 };
 180
 181 /* cpu data as detected by the assembly code in head.S */
 182 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 183 /* common cpu data for all cpus */
 184 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 185 EXPORT_SYMBOL(boot_cpu_data);
 186 static void set_mca_bus(int x)
 187 {
 188 #ifdef CONFIG_MCA
 189         MCA_bus = x;
 190 #endif
 191 }
 192
 193 unsigned int def_to_bigsmp;
 194
 195 /* for MCA, but anyone else can use it if they want */
 196 unsigned int machine_id;
 197 unsigned int machine_submodel_id;
 198 unsigned int BIOS_revision;
 199
 200 struct apm_info apm_info;
 201 EXPORT_SYMBOL(apm_info);
 202
 203 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 204         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 205 struct ist_info ist_info;
 206 EXPORT_SYMBOL(ist_info);
 207 #else
 208 struct ist_info ist_info;
 209 #endif
 210
 211 #else
 212 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 213         .x86_phys_bits = MAX_PHYSMEM_BITS,
 214 };
 215 EXPORT_SYMBOL(boot_cpu_data);
 216 #endif
 217
 218
 219 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 220 unsigned long mmu_cr4_features;
 221 #else
 222 unsigned long mmu_cr4_features = X86_CR4_PAE;
 223 #endif
 224
 225 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
 226 int bootloader_type, bootloader_version;
 227
 228 /*
 229  * Setup options
 230  */
 231 struct screen_info screen_info;
 232 EXPORT_SYMBOL(screen_info);
 233 struct edid_info edid_info;
 234 EXPORT_SYMBOL_GPL(edid_info);
 235
 236 extern int root_mountflags;
 237
 238 unsigned long saved_video_mode;
 239
 240 #define RAMDISK_IMAGE_START_MASK        0x07FF
 241 #define RAMDISK_PROMPT_FLAG             0x8000
 242 #define RAMDISK_LOAD_FLAG               0x4000
 243
 244 static char __initdata command_line[COMMAND_LINE_SIZE];
 245 #ifdef CONFIG_CMDLINE_BOOL
 246 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 247 #endif
 248
 249 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 250 struct edd edd;
 251 #ifdef CONFIG_EDD_MODULE
 252 EXPORT_SYMBOL(edd);
 253 #endif
 254 /**
 255  * copy_edd() - Copy the BIOS EDD information
 256  *              from boot_params into a safe place.
 257  *
 258  */
 259 static inline void copy_edd(void)
 260 {
 261      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 262             sizeof(edd.mbr_signature));
 263      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 264      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 265      edd.edd_info_nr = boot_params.eddbuf_entries;
 266 }
 267 #else
 268 static inline void copy_edd(void)
 269 {
 270 }
 271 #endif
 272
 273 void * __init extend_brk(size_t size, size_t align)
 274 {
 275         size_t mask = align - 1;
 276         void *ret;
 277
 278         BUG_ON(_brk_start == 0);
 279         BUG_ON(align & mask);
 280
 281         _brk_end = (_brk_end + mask) & ~mask;
 282         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 283
 284         ret = (void *)_brk_end;
 285         _brk_end += size;
 286
 287         memset(ret, 0, size);
 288
 289         return ret;
 290 }
 291
 292 static void __init reserve_brk(void)
 293 {
 294         if (_brk_end > _brk_start)
 295                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 296
 297         /* Mark brk area as locked down and no longer taking any
 298            new allocations */
 299         _brk_start = 0;
 300 }
 301
 302 #ifdef CONFIG_BLK_DEV_INITRD
 303
 304 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 305 static void __init relocate_initrd(void)
 306 {
 307
 308         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 309         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 310         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 311         u64 ramdisk_here;
 312         unsigned long slop, clen, mapaddr;
 313         char *p, *q;
 314
 315         /* We need to move the initrd down into lowmem */
 316         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 317                                          PAGE_SIZE);
 318
 319         if (ramdisk_here == -1ULL)
 320                 panic("Cannot find place for new RAMDISK of size %lld\n",
 321                          ramdisk_size);
 322
 323         /* Note: this includes all the lowmem currently occupied by
 324            the initrd, we rely on that fact to keep the data intact. */
 325         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 326                          "NEW RAMDISK");
 327         initrd_start = ramdisk_here + PAGE_OFFSET;
 328         initrd_end   = initrd_start + ramdisk_size;
 329         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 330                          ramdisk_here, ramdisk_here + ramdisk_size);
 331
 332         q = (char *)initrd_start;
 333
 334         /* Copy any lowmem portion of the initrd */
 335         if (ramdisk_image < end_of_lowmem) {
 336                 clen = end_of_lowmem - ramdisk_image;
 337                 p = (char *)__va(ramdisk_image);
 338                 memcpy(q, p, clen);
 339                 q += clen;
 340                 ramdisk_image += clen;
 341                 ramdisk_size  -= clen;
 342         }
 343
 344         /* Copy the highmem portion of the initrd */
 345         while (ramdisk_size) {
 346                 slop = ramdisk_image & ~PAGE_MASK;
 347                 clen = ramdisk_size;
 348                 if (clen > MAX_MAP_CHUNK-slop)
 349                         clen = MAX_MAP_CHUNK-slop;
 350                 mapaddr = ramdisk_image & PAGE_MASK;
 351                 p = early_memremap(mapaddr, clen+slop);
 352                 memcpy(q, p+slop, clen);
 353                 early_iounmap(p, clen+slop);
 354                 q += clen;
 355                 ramdisk_image += clen;
 356                 ramdisk_size  -= clen;
 357         }
 358         /* high pages is not converted by early_res_to_bootmem */
 359         ramdisk_image = boot_params.hdr.ramdisk_image;
 360         ramdisk_size  = boot_params.hdr.ramdisk_size;
 361         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 362                 " %08llx - %08llx\n",
 363                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 364                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 365 }
 366
 367 static void __init reserve_initrd(void)
 368 {
 369         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 370         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 371         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 372         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 373
 374         if (!boot_params.hdr.type_of_loader ||
 375             !ramdisk_image || !ramdisk_size)
 376                 return;         /* No initrd provided by bootloader */
 377
 378         initrd_start = 0;
 379
 380         if (ramdisk_size >= (end_of_lowmem>>1)) {
 381                 free_early(ramdisk_image, ramdisk_end);
 382                 printk(KERN_ERR "initrd too large to handle, "
 383                        "disabling initrd\n");
 384                 return;
 385         }
 386
 387         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 388                         ramdisk_end);
 389
 390
 391         if (ramdisk_end <= end_of_lowmem) {
 392                 /* All in lowmem, easy case */
 393                 /*
 394                  * don't need to reserve again, already reserved early
 395                  * in i386_start_kernel
 396                  */
 397                 initrd_start = ramdisk_image + PAGE_OFFSET;
 398                 initrd_end = initrd_start + ramdisk_size;
 399                 return;
 400         }
 401
 402         relocate_initrd();
 403
 404         free_early(ramdisk_image, ramdisk_end);
 405 }
 406 #else
 407 static void __init reserve_initrd(void)
 408 {
 409 }
 410 #endif /* CONFIG_BLK_DEV_INITRD */
 411
 412 static void __init parse_setup_data(void)
 413 {
 414         struct setup_data *data;
 415         u64 pa_data;
 416
 417         if (boot_params.hdr.version < 0x0209)
 418                 return;
 419         pa_data = boot_params.hdr.setup_data;
 420         while (pa_data) {
 421                 data = early_memremap(pa_data, PAGE_SIZE);
 422                 switch (data->type) {
 423                 case SETUP_E820_EXT:
 424                         parse_e820_ext(data, pa_data);
 425                         break;
 426                 default:
 427                         break;
 428                 }
 429                 pa_data = data->next;
 430                 early_iounmap(data, PAGE_SIZE);
 431         }
 432 }
 433
 434 static void __init e820_reserve_setup_data(void)
 435 {
 436         struct setup_data *data;
 437         u64 pa_data;
 438         int found = 0;
 439
 440         if (boot_params.hdr.version < 0x0209)
 441                 return;
 442         pa_data = boot_params.hdr.setup_data;
 443         while (pa_data) {
 444                 data = early_memremap(pa_data, sizeof(*data));
 445                 e820_update_range(pa_data, sizeof(*data)+data->len,
 446                          E820_RAM, E820_RESERVED_KERN);
 447                 found = 1;
 448                 pa_data = data->next;
 449                 early_iounmap(data, sizeof(*data));
 450         }
 451         if (!found)
 452                 return;
 453
 454         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 455         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 456         printk(KERN_INFO "extended physical RAM map:\n");
 457         e820_print_map("reserve setup_data");
 458 }
 459
 460 static void __init reserve_early_setup_data(void)
 461 {
 462         struct setup_data *data;
 463         u64 pa_data;
 464         char buf[32];
 465
 466         if (boot_params.hdr.version < 0x0209)
 467                 return;
 468         pa_data = boot_params.hdr.setup_data;
 469         while (pa_data) {
 470                 data = early_memremap(pa_data, sizeof(*data));
 471                 sprintf(buf, "setup data %x", data->type);
 472                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 473                 pa_data = data->next;
 474                 early_iounmap(data, sizeof(*data));
 475         }
 476 }
 477
 478 /*
 479  * --------- Crashkernel reservation ------------------------------
 480  */
 481
 482 #ifdef CONFIG_KEXEC
 483
 484 /**
 485  * Reserve @size bytes of crashkernel memory at any suitable offset.
 486  *
 487  * @size: Size of the crashkernel memory to reserve.
 488  * Returns the base address on success, and -1ULL on failure.
 489  */
 490 static
 491 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 492 {
 493         const unsigned long long alignment = 16<<20;    /* 16M */
 494         unsigned long long start = 0LL;
 495
 496         while (1) {
 497                 int ret;
 498
 499                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 500                 if (start == -1ULL)
 501                         return start;
 502
 503                 /* try to reserve it */
 504                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 505                 if (ret >= 0)
 506                         return start;
 507
 508                 start += alignment;
 509         }
 510 }
 511
 512 static inline unsigned long long get_total_mem(void)
 513 {
 514         unsigned long long total;
 515
 516         total = max_low_pfn - min_low_pfn;
 517 #ifdef CONFIG_HIGHMEM
 518         total += highend_pfn - highstart_pfn;
 519 #endif
 520
 521         return total << PAGE_SHIFT;
 522 }
 523
 524 static void __init reserve_crashkernel(void)
 525 {
 526         unsigned long long total_mem;
 527         unsigned long long crash_size, crash_base;
 528         int ret;
 529
 530         total_mem = get_total_mem();
 531
 532         ret = parse_crashkernel(boot_command_line, total_mem,
 533                         &crash_size, &crash_base);
 534         if (ret != 0 || crash_size <= 0)
 535                 return;
 536
 537         /* 0 means: find the address automatically */
 538         if (crash_base <= 0) {
 539                 crash_base = find_and_reserve_crashkernel(crash_size);
 540                 if (crash_base == -1ULL) {
 541                         pr_info("crashkernel reservation failed. "
 542                                 "No suitable area found.\n");
 543                         return;
 544                 }
 545         } else {
 546                 ret = reserve_bootmem_generic(crash_base, crash_size,
 547                                         BOOTMEM_EXCLUSIVE);
 548                 if (ret < 0) {
 549                         pr_info("crashkernel reservation failed - "
 550                                 "memory is in use\n");
 551                         return;
 552                 }
 553         }
 554
 555         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 556                         "for crashkernel (System RAM: %ldMB)\n",
 557                         (unsigned long)(crash_size >> 20),
 558                         (unsigned long)(crash_base >> 20),
 559                         (unsigned long)(total_mem >> 20));
 560
 561         crashk_res.start = crash_base;
 562         crashk_res.end   = crash_base + crash_size - 1;
 563         insert_resource(&iomem_resource, &crashk_res);
 564 }
 565 #else
 566 static void __init reserve_crashkernel(void)
 567 {
 568 }
 569 #endif
 570
 571 static struct resource standard_io_resources[] = {
 572         { .name = "dma1", .start = 0x00, .end = 0x1f,
 573                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 574         { .name = "pic1", .start = 0x20, .end = 0x21,
 575                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 576         { .name = "timer0", .start = 0x40, .end = 0x43,
 577                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 578         { .name = "timer1", .start = 0x50, .end = 0x53,
 579                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 580         { .name = "keyboard", .start = 0x60, .end = 0x60,
 581                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 582         { .name = "keyboard", .start = 0x64, .end = 0x64,
 583                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 584         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 585                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 586         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 587                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 588         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 589                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 590         { .name = "fpu", .start = 0xf0, .end = 0xff,
 591                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 592 };
 593
 594 static void __init reserve_standard_io_resources(void)
 595 {
 596         int i;
 597
 598         /* request I/O space for devices used on all i[345]86 PCs */
 599         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 600                 request_resource(&ioport_resource, &standard_io_resources[i]);
 601
 602 }
 603
 604 /*
 605  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 606  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 607  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 608  */
 609
 610 #ifdef CONFIG_CRASH_DUMP
 611 /* elfcorehdr= specifies the location of elf core header
 612  * stored by the crashed kernel. This option will be passed
 613  * by kexec loader to the capture kernel.
 614  */
 615 static int __init setup_elfcorehdr(char *arg)
 616 {
 617         char *end;
 618         if (!arg)
 619                 return -EINVAL;
 620         elfcorehdr_addr = memparse(arg, &end);
 621         return end > arg ? 0 : -EINVAL;
 622 }
 623 early_param("elfcorehdr", setup_elfcorehdr);
 624 #endif
 625
 626 static struct x86_quirks default_x86_quirks __initdata;
 627
 628 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 629
 630 #ifdef CONFIG_X86_RESERVE_LOW_64K
 631 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 632 {
 633         printk(KERN_NOTICE
 634                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 635                 d->ident);
 636
 637         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 638         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 639
 640         return 0;
 641 }
 642 #endif
 643
 644 /* List of systems that have known low memory corruption BIOS problems */
 645 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 646 #ifdef CONFIG_X86_RESERVE_LOW_64K
 647         {
 648                 .callback = dmi_low_memory_corruption,
 649                 .ident = "AMI BIOS",
 650                 .matches = {
 651                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 652                 },
 653         },
 654         {
 655                 .callback = dmi_low_memory_corruption,
 656                 .ident = "Phoenix BIOS",
 657                 .matches = {
 658                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 659                 },
 660         },
 661 #endif
 662         {}
 663 };
 664
 665 /*
 666  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 667  * passed the efi memmap, systab, etc., so we should use these data structures
 668  * for initialization.  Note, the efi init code path is determined by the
 669  * global efi_enabled. This allows the same kernel image to be used on existing
 670  * systems (with a traditional BIOS) as well as on EFI systems.
 671  */
 672 /*
 673  * setup_arch - architecture-specific boot-time initializations
 674  *
 675  * Note: On x86_64, fixmaps are ready for use even before this is called.
 676  */
 677
 678 void __init setup_arch(char **cmdline_p)
 679 {
 680 #ifdef CONFIG_X86_32
 681         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 682         visws_early_detect();
 683 #else
 684         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 685 #endif
 686
 687         /* VMI may relocate the fixmap; do this before touching ioremap area */
 688         vmi_init();
 689
 690         early_cpu_init();
 691         early_ioremap_init();
 692
 693         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 694         screen_info = boot_params.screen_info;
 695         edid_info = boot_params.edid_info;
 696 #ifdef CONFIG_X86_32
 697         apm_info.bios = boot_params.apm_bios_info;
 698         ist_info = boot_params.ist_info;
 699         if (boot_params.sys_desc_table.length != 0) {
 700                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 701                 machine_id = boot_params.sys_desc_table.table[0];
 702                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 703                 BIOS_revision = boot_params.sys_desc_table.table[2];
 704         }
 705 #endif
 706         saved_video_mode = boot_params.hdr.vid_mode;
 707         bootloader_type = boot_params.hdr.type_of_loader;
 708         if ((bootloader_type >> 4) == 0xe) {
 709                 bootloader_type &= 0xf;
 710                 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
 711         }
 712         bootloader_version  = bootloader_type & 0xf;
 713         bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
 714
 715 #ifdef CONFIG_BLK_DEV_RAM
 716         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 717         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 718         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 719 #endif
 720 #ifdef CONFIG_EFI
 721         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 722 #ifdef CONFIG_X86_32
 723                      "EL32",
 724 #else
 725                      "EL64",
 726 #endif
 727          4)) {
 728                 efi_enabled = 1;
 729                 efi_reserve_early();
 730         }
 731 #endif
 732
 733         ARCH_SETUP
 734
 735         setup_memory_map();
 736         parse_setup_data();
 737         /* update the e820_saved too */
 738         e820_reserve_setup_data();
 739
 740         copy_edd();
 741
 742         if (!boot_params.hdr.root_flags)
 743                 root_mountflags &= ~MS_RDONLY;
 744         init_mm.start_code = (unsigned long) _text;
 745         init_mm.end_code = (unsigned long) _etext;
 746         init_mm.end_data = (unsigned long) _edata;
 747         init_mm.brk = _brk_end;
 748
 749         code_resource.start = virt_to_phys(_text);
 750         code_resource.end = virt_to_phys(_etext)-1;
 751         data_resource.start = virt_to_phys(_etext);
 752         data_resource.end = virt_to_phys(_edata)-1;
 753         bss_resource.start = virt_to_phys(&__bss_start);
 754         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 755
 756 #ifdef CONFIG_CMDLINE_BOOL
 757 #ifdef CONFIG_CMDLINE_OVERRIDE
 758         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 759 #else
 760         if (builtin_cmdline[0]) {
 761                 /* append boot loader cmdline to builtin */
 762                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 763                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 764                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 765         }
 766 #endif
 767 #endif
 768
 769         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 770         *cmdline_p = command_line;
 771
 772         parse_early_param();
 773
 774 #ifdef CONFIG_X86_64
 775         check_efer();
 776 #endif
 777
 778         /* Must be before kernel pagetables are setup */
 779         vmi_activate();
 780
 781         /* after early param, so could get panic from serial */
 782         reserve_early_setup_data();
 783
 784         if (acpi_mps_check()) {
 785 #ifdef CONFIG_X86_LOCAL_APIC
 786                 disable_apic = 1;
 787 #endif
 788                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 789         }
 790
 791 #ifdef CONFIG_PCI
 792         if (pci_early_dump_regs)
 793                 early_dump_pci_devices();
 794 #endif
 795
 796         finish_e820_parsing();
 797
 798         if (efi_enabled)
 799                 efi_init();
 800
 801         dmi_scan_machine();
 802
 803         dmi_check_system(bad_bios_dmi_table);
 804
 805         /*
 806          * VMware detection requires dmi to be available, so this
 807          * needs to be done after dmi_scan_machine, for the BP.
 808          */
 809         init_hypervisor(&boot_cpu_data);
 810
 811 #ifdef CONFIG_X86_32
 812         probe_roms();
 813 #endif
 814
 815         /* after parse_early_param, so could debug it */
 816         insert_resource(&iomem_resource, &code_resource);
 817         insert_resource(&iomem_resource, &data_resource);
 818         insert_resource(&iomem_resource, &bss_resource);
 819
 820
 821 #ifdef CONFIG_X86_32
 822         if (ppro_with_ram_bug()) {
 823                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 824                                   E820_RESERVED);
 825                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 826                 printk(KERN_INFO "fixed physical RAM map:\n");
 827                 e820_print_map("bad_ppro");
 828         }
 829 #else
 830         early_gart_iommu_check();
 831 #endif
 832
 833         /*
 834          * partially used pages are not usable - thus
 835          * we are rounding upwards:
 836          */
 837         max_pfn = e820_end_of_ram_pfn();
 838
 839         /* preallocate 4k for mptable mpc */
 840         early_reserve_e820_mpc_new();
 841         /* update e820 for memory not covered by WB MTRRs */
 842         mtrr_bp_init();
 843         if (mtrr_trim_uncached_memory(max_pfn))
 844                 max_pfn = e820_end_of_ram_pfn();
 845
 846 #ifdef CONFIG_X86_32
 847         /* max_low_pfn get updated here */
 848         find_low_pfn_range();
 849 #else
 850         num_physpages = max_pfn;
 851
 852         check_x2apic();
 853
 854         /* How many end-of-memory variables you have, grandma! */
 855         /* need this before calling reserve_initrd */
 856         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 857                 max_low_pfn = e820_end_of_low_ram_pfn();
 858         else
 859                 max_low_pfn = max_pfn;
 860
 861         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 862         max_pfn_mapped = KERNEL_IMAGE_SIZE >> PAGE_SHIFT;
 863 #endif
 864
 865 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 866         setup_bios_corruption_check();
 867 #endif
 868
 869         printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
 870                         max_pfn_mapped<<PAGE_SHIFT);
 871
 872         reserve_brk();
 873
 874         /* max_pfn_mapped is updated here */
 875         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 876         max_pfn_mapped = max_low_pfn_mapped;
 877
 878 #ifdef CONFIG_X86_64
 879         if (max_pfn > max_low_pfn) {
 880                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 881                                                      max_pfn<<PAGE_SHIFT);
 882                 /* can we preseve max_low_pfn ?*/
 883                 max_low_pfn = max_pfn;
 884         }
 885 #endif
 886
 887         /*
 888          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 889          */
 890
 891 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 892         if (init_ohci1394_dma_early)
 893                 init_ohci1394_dma_on_all_controllers();
 894 #endif
 895
 896         reserve_initrd();
 897
 898         vsmp_init();
 899
 900         io_delay_init();
 901
 902         /*
 903          * Parse the ACPI tables for possible boot-time SMP configuration.
 904          */
 905         acpi_boot_table_init();
 906
 907         early_acpi_boot_init();
 908
 909 #ifdef CONFIG_ACPI_NUMA
 910         /*
 911          * Parse SRAT to discover nodes.
 912          */
 913         acpi_numa_init();
 914 #endif
 915
 916         initmem_init(0, max_pfn);
 917
 918 #ifdef CONFIG_ACPI_SLEEP
 919         /*
 920          * Reserve low memory region for sleep support.
 921          */
 922         acpi_reserve_bootmem();
 923 #endif
 924         /*
 925          * Find and reserve possible boot-time SMP configuration:
 926          */
 927         find_smp_config();
 928
 929         reserve_crashkernel();
 930
 931 #ifdef CONFIG_X86_64
 932         /*
 933          * dma32_reserve_bootmem() allocates bootmem which may conflict
 934          * with the crashkernel command line, so do that after
 935          * reserve_crashkernel()
 936          */
 937         dma32_reserve_bootmem();
 938 #endif
 939
 940         reserve_ibft_region();
 941
 942 #ifdef CONFIG_KVM_CLOCK
 943         kvmclock_init();
 944 #endif
 945
 946         paravirt_pagetable_setup_start(swapper_pg_dir);
 947         paging_init();
 948         paravirt_pagetable_setup_done(swapper_pg_dir);
 949         paravirt_post_allocator_init();
 950
 951 #ifdef CONFIG_X86_64
 952         map_vsyscall();
 953 #endif
 954
 955         generic_apic_probe();
 956
 957         early_quirks();
 958
 959         /*
 960          * Read APIC and some other early information from ACPI tables.
 961          */
 962         acpi_boot_init();
 963
 964 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 965         /*
 966          * get boot-time SMP configuration:
 967          */
 968         if (smp_found_config)
 969                 get_smp_config();
 970 #endif
 971
 972         prefill_possible_map();
 973
 974 #ifdef CONFIG_X86_64
 975         init_cpu_to_node();
 976 #endif
 977
 978         init_apic_mappings();
 979         ioapic_init_mappings();
 980
 981         /* need to wait for io_apic is mapped */
 982         probe_nr_irqs_gsi();
 983
 984         kvm_guest_init();
 985
 986         e820_reserve_resources();
 987         e820_mark_nosave_regions(max_low_pfn);
 988
 989 #ifdef CONFIG_X86_32
 990         request_resource(&iomem_resource, &video_ram_resource);
 991 #endif
 992         reserve_standard_io_resources();
 993
 994         e820_setup_gap();
 995
 996 #ifdef CONFIG_VT
 997 #if defined(CONFIG_VGA_CONSOLE)
 998         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 999                 conswitchp = &vga_con;
1000 #elif defined(CONFIG_DUMMY_CONSOLE)
1001         conswitchp = &dummy_con;
1002 #endif
1003 #endif
1004 }
1005
1006 #ifdef CONFIG_X86_32
1007
1008 /**
1009  * x86_quirk_intr_init - post gate setup interrupt initialisation
1010  *
1011  * Description:
1012  *      Fill in any interrupts that may have been left out by the general
1013  *      init_IRQ() routine.  interrupts having to do with the machine rather
1014  *      than the devices on the I/O bus (like APIC interrupts in intel MP
1015  *      systems) are started here.
1016  **/
1017 void __init x86_quirk_intr_init(void)
1018 {
1019         if (x86_quirks->arch_intr_init) {
1020                 if (x86_quirks->arch_intr_init())
1021                         return;
1022         }
1023 }
1024
1025 /**
1026  * x86_quirk_trap_init - initialise system specific traps
1027  *
1028  * Description:
1029  *      Called as the final act of trap_init().  Used in VISWS to initialise
1030  *      the various board specific APIC traps.
1031  **/
1032 void __init x86_quirk_trap_init(void)
1033 {
1034         if (x86_quirks->arch_trap_init) {
1035                 if (x86_quirks->arch_trap_init())
1036                         return;
1037         }
1038 }
1039
1040 static struct irqaction irq0  = {
1041         .handler = timer_interrupt,
1042         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1043         .name = "timer"
1044 };
1045
1046 /**
1047  * x86_quirk_pre_time_init - do any specific initialisations before.
1048  *
1049  **/
1050 void __init x86_quirk_pre_time_init(void)
1051 {
1052         if (x86_quirks->arch_pre_time_init)
1053                 x86_quirks->arch_pre_time_init();
1054 }
1055
1056 /**
1057  * x86_quirk_time_init - do any specific initialisations for the system timer.
1058  *
1059  * Description:
1060  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1061  *      in irq_vectors.h:TIMER_IRQ
1062  **/
1063 void __init x86_quirk_time_init(void)
1064 {
1065         if (x86_quirks->arch_time_init) {
1066                 /*
1067                  * A nonzero return code does not mean failure, it means
1068                  * that the architecture quirk does not want any
1069                  * generic (timer) setup to be performed after this:
1070                  */
1071                 if (x86_quirks->arch_time_init())
1072                         return;
1073         }
1074
1075         irq0.mask = cpumask_of_cpu(0);
1076         setup_irq(0, &irq0);
1077 }
1078 #endif /* CONFIG_X86_32 */