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 RESERVE_BRK(dmi_alloc, 65536);
 116
 117 unsigned int boot_cpu_id __read_mostly;
 118
 119 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 120 unsigned long _brk_end = (unsigned long)__brk_base;
 121
 122 #ifdef CONFIG_X86_64
 123 int default_cpu_present_to_apicid(int mps_cpu)
 124 {
 125         return __default_cpu_present_to_apicid(mps_cpu);
 126 }
 127
 128 int default_check_phys_apicid_present(int boot_cpu_physical_apicid)
 129 {
 130         return __default_check_phys_apicid_present(boot_cpu_physical_apicid);
 131 }
 132 #endif
 133
 134 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 135 struct boot_params __initdata boot_params;
 136 #else
 137 struct boot_params boot_params;
 138 #endif
 139
 140 /*
 141  * Machine setup..
 142  */
 143 static struct resource data_resource = {
 144         .name   = "Kernel data",
 145         .start  = 0,
 146         .end    = 0,
 147         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 148 };
 149
 150 static struct resource code_resource = {
 151         .name   = "Kernel code",
 152         .start  = 0,
 153         .end    = 0,
 154         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 155 };
 156
 157 static struct resource bss_resource = {
 158         .name   = "Kernel bss",
 159         .start  = 0,
 160         .end    = 0,
 161         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 162 };
 163
 164
 165 #ifdef CONFIG_X86_32
 166 static struct resource video_ram_resource = {
 167         .name   = "Video RAM area",
 168         .start  = 0xa0000,
 169         .end    = 0xbffff,
 170         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 171 };
 172
 173 /* cpu data as detected by the assembly code in head.S */
 174 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 175 /* common cpu data for all cpus */
 176 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 177 EXPORT_SYMBOL(boot_cpu_data);
 178 static void set_mca_bus(int x)
 179 {
 180 #ifdef CONFIG_MCA
 181         MCA_bus = x;
 182 #endif
 183 }
 184
 185 unsigned int def_to_bigsmp;
 186
 187 /* for MCA, but anyone else can use it if they want */
 188 unsigned int machine_id;
 189 unsigned int machine_submodel_id;
 190 unsigned int BIOS_revision;
 191
 192 struct apm_info apm_info;
 193 EXPORT_SYMBOL(apm_info);
 194
 195 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 196         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 197 struct ist_info ist_info;
 198 EXPORT_SYMBOL(ist_info);
 199 #else
 200 struct ist_info ist_info;
 201 #endif
 202
 203 #else
 204 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 205         .x86_phys_bits = MAX_PHYSMEM_BITS,
 206 };
 207 EXPORT_SYMBOL(boot_cpu_data);
 208 #endif
 209
 210
 211 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 212 unsigned long mmu_cr4_features;
 213 #else
 214 unsigned long mmu_cr4_features = X86_CR4_PAE;
 215 #endif
 216
 217 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 218 int bootloader_type;
 219
 220 /*
 221  * Setup options
 222  */
 223 struct screen_info screen_info;
 224 EXPORT_SYMBOL(screen_info);
 225 struct edid_info edid_info;
 226 EXPORT_SYMBOL_GPL(edid_info);
 227
 228 extern int root_mountflags;
 229
 230 unsigned long saved_video_mode;
 231
 232 #define RAMDISK_IMAGE_START_MASK        0x07FF
 233 #define RAMDISK_PROMPT_FLAG             0x8000
 234 #define RAMDISK_LOAD_FLAG               0x4000
 235
 236 static char __initdata command_line[COMMAND_LINE_SIZE];
 237 #ifdef CONFIG_CMDLINE_BOOL
 238 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 239 #endif
 240
 241 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 242 struct edd edd;
 243 #ifdef CONFIG_EDD_MODULE
 244 EXPORT_SYMBOL(edd);
 245 #endif
 246 /**
 247  * copy_edd() - Copy the BIOS EDD information
 248  *              from boot_params into a safe place.
 249  *
 250  */
 251 static inline void copy_edd(void)
 252 {
 253      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 254             sizeof(edd.mbr_signature));
 255      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 256      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 257      edd.edd_info_nr = boot_params.eddbuf_entries;
 258 }
 259 #else
 260 static inline void copy_edd(void)
 261 {
 262 }
 263 #endif
 264
 265 void * __init extend_brk(size_t size, size_t align)
 266 {
 267         size_t mask = align - 1;
 268         void *ret;
 269
 270         BUG_ON(_brk_start == 0);
 271         BUG_ON(align & mask);
 272
 273         _brk_end = (_brk_end + mask) & ~mask;
 274         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 275
 276         ret = (void *)_brk_end;
 277         _brk_end += size;
 278
 279         memset(ret, 0, size);
 280
 281         return ret;
 282 }
 283
 284 static void __init reserve_brk(void)
 285 {
 286         if (_brk_end > _brk_start)
 287                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 288
 289         /* Mark brk area as locked down and no longer taking any
 290            new allocations */
 291         _brk_start = 0;
 292 }
 293
 294 #ifdef CONFIG_BLK_DEV_INITRD
 295
 296 #ifdef CONFIG_X86_32
 297
 298 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 299 static void __init relocate_initrd(void)
 300 {
 301
 302         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 303         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 304         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 305         u64 ramdisk_here;
 306         unsigned long slop, clen, mapaddr;
 307         char *p, *q;
 308
 309         /* We need to move the initrd down into lowmem */
 310         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 311                                          PAGE_SIZE);
 312
 313         if (ramdisk_here == -1ULL)
 314                 panic("Cannot find place for new RAMDISK of size %lld\n",
 315                          ramdisk_size);
 316
 317         /* Note: this includes all the lowmem currently occupied by
 318            the initrd, we rely on that fact to keep the data intact. */
 319         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 320                          "NEW RAMDISK");
 321         initrd_start = ramdisk_here + PAGE_OFFSET;
 322         initrd_end   = initrd_start + ramdisk_size;
 323         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 324                          ramdisk_here, ramdisk_here + ramdisk_size);
 325
 326         q = (char *)initrd_start;
 327
 328         /* Copy any lowmem portion of the initrd */
 329         if (ramdisk_image < end_of_lowmem) {
 330                 clen = end_of_lowmem - ramdisk_image;
 331                 p = (char *)__va(ramdisk_image);
 332                 memcpy(q, p, clen);
 333                 q += clen;
 334                 ramdisk_image += clen;
 335                 ramdisk_size  -= clen;
 336         }
 337
 338         /* Copy the highmem portion of the initrd */
 339         while (ramdisk_size) {
 340                 slop = ramdisk_image & ~PAGE_MASK;
 341                 clen = ramdisk_size;
 342                 if (clen > MAX_MAP_CHUNK-slop)
 343                         clen = MAX_MAP_CHUNK-slop;
 344                 mapaddr = ramdisk_image & PAGE_MASK;
 345                 p = early_memremap(mapaddr, clen+slop);
 346                 memcpy(q, p+slop, clen);
 347                 early_iounmap(p, clen+slop);
 348                 q += clen;
 349                 ramdisk_image += clen;
 350                 ramdisk_size  -= clen;
 351         }
 352         /* high pages is not converted by early_res_to_bootmem */
 353         ramdisk_image = boot_params.hdr.ramdisk_image;
 354         ramdisk_size  = boot_params.hdr.ramdisk_size;
 355         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 356                 " %08llx - %08llx\n",
 357                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 358                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 359 }
 360 #endif
 361
 362 static void __init reserve_initrd(void)
 363 {
 364         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 365         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 366         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 367         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 368
 369         if (!boot_params.hdr.type_of_loader ||
 370             !ramdisk_image || !ramdisk_size)
 371                 return;         /* No initrd provided by bootloader */
 372
 373         initrd_start = 0;
 374
 375         if (ramdisk_size >= (end_of_lowmem>>1)) {
 376                 free_early(ramdisk_image, ramdisk_end);
 377                 printk(KERN_ERR "initrd too large to handle, "
 378                        "disabling initrd\n");
 379                 return;
 380         }
 381
 382         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 383                         ramdisk_end);
 384
 385
 386         if (ramdisk_end <= end_of_lowmem) {
 387                 /* All in lowmem, easy case */
 388                 /*
 389                  * don't need to reserve again, already reserved early
 390                  * in i386_start_kernel
 391                  */
 392                 initrd_start = ramdisk_image + PAGE_OFFSET;
 393                 initrd_end = initrd_start + ramdisk_size;
 394                 return;
 395         }
 396
 397 #ifdef CONFIG_X86_32
 398         relocate_initrd();
 399 #else
 400         printk(KERN_ERR "initrd extends beyond end of memory "
 401                "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
 402                ramdisk_end, end_of_lowmem);
 403         initrd_start = 0;
 404 #endif
 405         free_early(ramdisk_image, ramdisk_end);
 406 }
 407 #else
 408 static void __init reserve_initrd(void)
 409 {
 410 }
 411 #endif /* CONFIG_BLK_DEV_INITRD */
 412
 413 static void __init parse_setup_data(void)
 414 {
 415         struct setup_data *data;
 416         u64 pa_data;
 417
 418         if (boot_params.hdr.version < 0x0209)
 419                 return;
 420         pa_data = boot_params.hdr.setup_data;
 421         while (pa_data) {
 422                 data = early_memremap(pa_data, PAGE_SIZE);
 423                 switch (data->type) {
 424                 case SETUP_E820_EXT:
 425                         parse_e820_ext(data, pa_data);
 426                         break;
 427                 default:
 428                         break;
 429                 }
 430                 pa_data = data->next;
 431                 early_iounmap(data, PAGE_SIZE);
 432         }
 433 }
 434
 435 static void __init e820_reserve_setup_data(void)
 436 {
 437         struct setup_data *data;
 438         u64 pa_data;
 439         int found = 0;
 440
 441         if (boot_params.hdr.version < 0x0209)
 442                 return;
 443         pa_data = boot_params.hdr.setup_data;
 444         while (pa_data) {
 445                 data = early_memremap(pa_data, sizeof(*data));
 446                 e820_update_range(pa_data, sizeof(*data)+data->len,
 447                          E820_RAM, E820_RESERVED_KERN);
 448                 found = 1;
 449                 pa_data = data->next;
 450                 early_iounmap(data, sizeof(*data));
 451         }
 452         if (!found)
 453                 return;
 454
 455         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 456         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 457         printk(KERN_INFO "extended physical RAM map:\n");
 458         e820_print_map("reserve setup_data");
 459 }
 460
 461 static void __init reserve_early_setup_data(void)
 462 {
 463         struct setup_data *data;
 464         u64 pa_data;
 465         char buf[32];
 466
 467         if (boot_params.hdr.version < 0x0209)
 468                 return;
 469         pa_data = boot_params.hdr.setup_data;
 470         while (pa_data) {
 471                 data = early_memremap(pa_data, sizeof(*data));
 472                 sprintf(buf, "setup data %x", data->type);
 473                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 474                 pa_data = data->next;
 475                 early_iounmap(data, sizeof(*data));
 476         }
 477 }
 478
 479 /*
 480  * --------- Crashkernel reservation ------------------------------
 481  */
 482
 483 #ifdef CONFIG_KEXEC
 484
 485 /**
 486  * Reserve @size bytes of crashkernel memory at any suitable offset.
 487  *
 488  * @size: Size of the crashkernel memory to reserve.
 489  * Returns the base address on success, and -1ULL on failure.
 490  */
 491 static
 492 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 493 {
 494         const unsigned long long alignment = 16<<20;    /* 16M */
 495         unsigned long long start = 0LL;
 496
 497         while (1) {
 498                 int ret;
 499
 500                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 501                 if (start == -1ULL)
 502                         return start;
 503
 504                 /* try to reserve it */
 505                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 506                 if (ret >= 0)
 507                         return start;
 508
 509                 start += alignment;
 510         }
 511 }
 512
 513 static inline unsigned long long get_total_mem(void)
 514 {
 515         unsigned long long total;
 516
 517         total = max_low_pfn - min_low_pfn;
 518 #ifdef CONFIG_HIGHMEM
 519         total += highend_pfn - highstart_pfn;
 520 #endif
 521
 522         return total << PAGE_SHIFT;
 523 }
 524
 525 static void __init reserve_crashkernel(void)
 526 {
 527         unsigned long long total_mem;
 528         unsigned long long crash_size, crash_base;
 529         int ret;
 530
 531         total_mem = get_total_mem();
 532
 533         ret = parse_crashkernel(boot_command_line, total_mem,
 534                         &crash_size, &crash_base);
 535         if (ret != 0 || crash_size <= 0)
 536                 return;
 537
 538         /* 0 means: find the address automatically */
 539         if (crash_base <= 0) {
 540                 crash_base = find_and_reserve_crashkernel(crash_size);
 541                 if (crash_base == -1ULL) {
 542                         pr_info("crashkernel reservation failed. "
 543                                 "No suitable area found.\n");
 544                         return;
 545                 }
 546         } else {
 547                 ret = reserve_bootmem_generic(crash_base, crash_size,
 548                                         BOOTMEM_EXCLUSIVE);
 549                 if (ret < 0) {
 550                         pr_info("crashkernel reservation failed - "
 551                                 "memory is in use\n");
 552                         return;
 553                 }
 554         }
 555
 556         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 557                         "for crashkernel (System RAM: %ldMB)\n",
 558                         (unsigned long)(crash_size >> 20),
 559                         (unsigned long)(crash_base >> 20),
 560                         (unsigned long)(total_mem >> 20));
 561
 562         crashk_res.start = crash_base;
 563         crashk_res.end   = crash_base + crash_size - 1;
 564         insert_resource(&iomem_resource, &crashk_res);
 565 }
 566 #else
 567 static void __init reserve_crashkernel(void)
 568 {
 569 }
 570 #endif
 571
 572 static struct resource standard_io_resources[] = {
 573         { .name = "dma1", .start = 0x00, .end = 0x1f,
 574                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 575         { .name = "pic1", .start = 0x20, .end = 0x21,
 576                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 577         { .name = "timer0", .start = 0x40, .end = 0x43,
 578                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 579         { .name = "timer1", .start = 0x50, .end = 0x53,
 580                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 581         { .name = "keyboard", .start = 0x60, .end = 0x60,
 582                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 583         { .name = "keyboard", .start = 0x64, .end = 0x64,
 584                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 585         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 586                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 587         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 588                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 589         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 590                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 591         { .name = "fpu", .start = 0xf0, .end = 0xff,
 592                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 593 };
 594
 595 static void __init reserve_standard_io_resources(void)
 596 {
 597         int i;
 598
 599         /* request I/O space for devices used on all i[345]86 PCs */
 600         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 601                 request_resource(&ioport_resource, &standard_io_resources[i]);
 602
 603 }
 604
 605 /*
 606  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 607  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 608  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 609  */
 610
 611 #ifdef CONFIG_CRASH_DUMP
 612 /* elfcorehdr= specifies the location of elf core header
 613  * stored by the crashed kernel. This option will be passed
 614  * by kexec loader to the capture kernel.
 615  */
 616 static int __init setup_elfcorehdr(char *arg)
 617 {
 618         char *end;
 619         if (!arg)
 620                 return -EINVAL;
 621         elfcorehdr_addr = memparse(arg, &end);
 622         return end > arg ? 0 : -EINVAL;
 623 }
 624 early_param("elfcorehdr", setup_elfcorehdr);
 625 #endif
 626
 627 static struct x86_quirks default_x86_quirks __initdata;
 628
 629 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 630
 631 #ifdef CONFIG_X86_RESERVE_LOW_64K
 632 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 633 {
 634         printk(KERN_NOTICE
 635                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 636                 d->ident);
 637
 638         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 639         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 640
 641         return 0;
 642 }
 643 #endif
 644
 645 /* List of systems that have known low memory corruption BIOS problems */
 646 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 647 #ifdef CONFIG_X86_RESERVE_LOW_64K
 648         {
 649                 .callback = dmi_low_memory_corruption,
 650                 .ident = "AMI BIOS",
 651                 .matches = {
 652                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 653                 },
 654         },
 655         {
 656                 .callback = dmi_low_memory_corruption,
 657                 .ident = "Phoenix BIOS",
 658                 .matches = {
 659                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 660                 },
 661         },
 662 #endif
 663         {}
 664 };
 665
 666 /*
 667  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 668  * passed the efi memmap, systab, etc., so we should use these data structures
 669  * for initialization.  Note, the efi init code path is determined by the
 670  * global efi_enabled. This allows the same kernel image to be used on existing
 671  * systems (with a traditional BIOS) as well as on EFI systems.
 672  */
 673 /*
 674  * setup_arch - architecture-specific boot-time initializations
 675  *
 676  * Note: On x86_64, fixmaps are ready for use even before this is called.
 677  */
 678
 679 void __init setup_arch(char **cmdline_p)
 680 {
 681 #ifdef CONFIG_X86_32
 682         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 683         visws_early_detect();
 684 #else
 685         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 686 #endif
 687
 688         /* VMI may relocate the fixmap; do this before touching ioremap area */
 689         vmi_init();
 690
 691         early_cpu_init();
 692         early_ioremap_init();
 693
 694         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 695         screen_info = boot_params.screen_info;
 696         edid_info = boot_params.edid_info;
 697 #ifdef CONFIG_X86_32
 698         apm_info.bios = boot_params.apm_bios_info;
 699         ist_info = boot_params.ist_info;
 700         if (boot_params.sys_desc_table.length != 0) {
 701                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 702                 machine_id = boot_params.sys_desc_table.table[0];
 703                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 704                 BIOS_revision = boot_params.sys_desc_table.table[2];
 705         }
 706 #endif
 707         saved_video_mode = boot_params.hdr.vid_mode;
 708         bootloader_type = boot_params.hdr.type_of_loader;
 709
 710 #ifdef CONFIG_BLK_DEV_RAM
 711         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 712         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 713         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 714 #endif
 715 #ifdef CONFIG_EFI
 716         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 717 #ifdef CONFIG_X86_32
 718                      "EL32",
 719 #else
 720                      "EL64",
 721 #endif
 722          4)) {
 723                 efi_enabled = 1;
 724                 efi_reserve_early();
 725         }
 726 #endif
 727
 728         ARCH_SETUP
 729
 730         setup_memory_map();
 731         parse_setup_data();
 732         /* update the e820_saved too */
 733         e820_reserve_setup_data();
 734
 735         copy_edd();
 736
 737         if (!boot_params.hdr.root_flags)
 738                 root_mountflags &= ~MS_RDONLY;
 739         init_mm.start_code = (unsigned long) _text;
 740         init_mm.end_code = (unsigned long) _etext;
 741         init_mm.end_data = (unsigned long) _edata;
 742         init_mm.brk = _brk_end;
 743
 744         code_resource.start = virt_to_phys(_text);
 745         code_resource.end = virt_to_phys(_etext)-1;
 746         data_resource.start = virt_to_phys(_etext);
 747         data_resource.end = virt_to_phys(_edata)-1;
 748         bss_resource.start = virt_to_phys(&__bss_start);
 749         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 750
 751 #ifdef CONFIG_CMDLINE_BOOL
 752 #ifdef CONFIG_CMDLINE_OVERRIDE
 753         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 754 #else
 755         if (builtin_cmdline[0]) {
 756                 /* append boot loader cmdline to builtin */
 757                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 758                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 759                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 760         }
 761 #endif
 762 #endif
 763
 764         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 765         *cmdline_p = command_line;
 766
 767         parse_early_param();
 768
 769 #ifdef CONFIG_X86_64
 770         check_efer();
 771 #endif
 772
 773         /* Must be before kernel pagetables are setup */
 774         vmi_activate();
 775
 776         /* after early param, so could get panic from serial */
 777         reserve_early_setup_data();
 778
 779         if (acpi_mps_check()) {
 780 #ifdef CONFIG_X86_LOCAL_APIC
 781                 disable_apic = 1;
 782 #endif
 783                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 784         }
 785
 786 #ifdef CONFIG_PCI
 787         if (pci_early_dump_regs)
 788                 early_dump_pci_devices();
 789 #endif
 790
 791         finish_e820_parsing();
 792
 793         if (efi_enabled)
 794                 efi_init();
 795
 796         dmi_scan_machine();
 797
 798         dmi_check_system(bad_bios_dmi_table);
 799
 800         /*
 801          * VMware detection requires dmi to be available, so this
 802          * needs to be done after dmi_scan_machine, for the BP.
 803          */
 804         init_hypervisor(&boot_cpu_data);
 805
 806 #ifdef CONFIG_X86_32
 807         probe_roms();
 808 #endif
 809
 810         /* after parse_early_param, so could debug it */
 811         insert_resource(&iomem_resource, &code_resource);
 812         insert_resource(&iomem_resource, &data_resource);
 813         insert_resource(&iomem_resource, &bss_resource);
 814
 815
 816 #ifdef CONFIG_X86_32
 817         if (ppro_with_ram_bug()) {
 818                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 819                                   E820_RESERVED);
 820                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 821                 printk(KERN_INFO "fixed physical RAM map:\n");
 822                 e820_print_map("bad_ppro");
 823         }
 824 #else
 825         early_gart_iommu_check();
 826 #endif
 827
 828         /*
 829          * partially used pages are not usable - thus
 830          * we are rounding upwards:
 831          */
 832         max_pfn = e820_end_of_ram_pfn();
 833
 834         /* preallocate 4k for mptable mpc */
 835         early_reserve_e820_mpc_new();
 836         /* update e820 for memory not covered by WB MTRRs */
 837         mtrr_bp_init();
 838         if (mtrr_trim_uncached_memory(max_pfn))
 839                 max_pfn = e820_end_of_ram_pfn();
 840
 841 #ifdef CONFIG_X86_32
 842         /* max_low_pfn get updated here */
 843         find_low_pfn_range();
 844 #else
 845         num_physpages = max_pfn;
 846
 847         check_x2apic();
 848
 849         /* How many end-of-memory variables you have, grandma! */
 850         /* need this before calling reserve_initrd */
 851         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 852                 max_low_pfn = e820_end_of_low_ram_pfn();
 853         else
 854                 max_low_pfn = max_pfn;
 855
 856         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 857 #endif
 858
 859 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 860         setup_bios_corruption_check();
 861 #endif
 862
 863         reserve_brk();
 864
 865         /* max_pfn_mapped is updated here */
 866         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 867         max_pfn_mapped = max_low_pfn_mapped;
 868
 869 #ifdef CONFIG_X86_64
 870         if (max_pfn > max_low_pfn) {
 871                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 872                                                      max_pfn<<PAGE_SHIFT);
 873                 /* can we preseve max_low_pfn ?*/
 874                 max_low_pfn = max_pfn;
 875         }
 876 #endif
 877
 878         /*
 879          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 880          */
 881
 882 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 883         if (init_ohci1394_dma_early)
 884                 init_ohci1394_dma_on_all_controllers();
 885 #endif
 886
 887         reserve_initrd();
 888
 889         vsmp_init();
 890
 891         io_delay_init();
 892
 893         /*
 894          * Parse the ACPI tables for possible boot-time SMP configuration.
 895          */
 896         acpi_boot_table_init();
 897
 898         early_acpi_boot_init();
 899
 900 #ifdef CONFIG_ACPI_NUMA
 901         /*
 902          * Parse SRAT to discover nodes.
 903          */
 904         acpi_numa_init();
 905 #endif
 906
 907         initmem_init(0, max_pfn);
 908
 909 #ifdef CONFIG_ACPI_SLEEP
 910         /*
 911          * Reserve low memory region for sleep support.
 912          */
 913         acpi_reserve_bootmem();
 914 #endif
 915         /*
 916          * Find and reserve possible boot-time SMP configuration:
 917          */
 918         find_smp_config();
 919
 920         reserve_crashkernel();
 921
 922 #ifdef CONFIG_X86_64
 923         /*
 924          * dma32_reserve_bootmem() allocates bootmem which may conflict
 925          * with the crashkernel command line, so do that after
 926          * reserve_crashkernel()
 927          */
 928         dma32_reserve_bootmem();
 929 #endif
 930
 931         reserve_ibft_region();
 932
 933 #ifdef CONFIG_KVM_CLOCK
 934         kvmclock_init();
 935 #endif
 936
 937         paravirt_pagetable_setup_start(swapper_pg_dir);
 938         paging_init();
 939         paravirt_pagetable_setup_done(swapper_pg_dir);
 940         paravirt_post_allocator_init();
 941
 942 #ifdef CONFIG_X86_64
 943         map_vsyscall();
 944 #endif
 945
 946         generic_apic_probe();
 947
 948         early_quirks();
 949
 950         /*
 951          * Read APIC and some other early information from ACPI tables.
 952          */
 953         acpi_boot_init();
 954
 955 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 956         /*
 957          * get boot-time SMP configuration:
 958          */
 959         if (smp_found_config)
 960                 get_smp_config();
 961 #endif
 962
 963         prefill_possible_map();
 964
 965 #ifdef CONFIG_X86_64
 966         init_cpu_to_node();
 967 #endif
 968
 969         init_apic_mappings();
 970         ioapic_init_mappings();
 971
 972         /* need to wait for io_apic is mapped */
 973         probe_nr_irqs_gsi();
 974
 975         kvm_guest_init();
 976
 977         e820_reserve_resources();
 978         e820_mark_nosave_regions(max_low_pfn);
 979
 980 #ifdef CONFIG_X86_32
 981         request_resource(&iomem_resource, &video_ram_resource);
 982 #endif
 983         reserve_standard_io_resources();
 984
 985         e820_setup_gap();
 986
 987 #ifdef CONFIG_VT
 988 #if defined(CONFIG_VGA_CONSOLE)
 989         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 990                 conswitchp = &vga_con;
 991 #elif defined(CONFIG_DUMMY_CONSOLE)
 992         conswitchp = &dummy_con;
 993 #endif
 994 #endif
 995 }
 996
 997 #ifdef CONFIG_X86_32
 998
 999 /**
1000  * x86_quirk_pre_intr_init - initialisation prior to setting up interrupt vectors
1001  *
1002  * Description:
1003  *      Perform any necessary interrupt initialisation prior to setting up
1004  *      the "ordinary" interrupt call gates.  For legacy reasons, the ISA
1005  *      interrupts should be initialised here if the machine emulates a PC
1006  *      in any way.
1007  **/
1008 void __init x86_quirk_pre_intr_init(void)
1009 {
1010         if (x86_quirks->arch_pre_intr_init) {
1011                 if (x86_quirks->arch_pre_intr_init())
1012                         return;
1013         }
1014         init_ISA_irqs();
1015 }
1016
1017 /**
1018  * x86_quirk_intr_init - post gate setup interrupt initialisation
1019  *
1020  * Description:
1021  *      Fill in any interrupts that may have been left out by the general
1022  *      init_IRQ() routine.  interrupts having to do with the machine rather
1023  *      than the devices on the I/O bus (like APIC interrupts in intel MP
1024  *      systems) are started here.
1025  **/
1026 void __init x86_quirk_intr_init(void)
1027 {
1028         if (x86_quirks->arch_intr_init) {
1029                 if (x86_quirks->arch_intr_init())
1030                         return;
1031         }
1032 }
1033
1034 /**
1035  * x86_quirk_trap_init - initialise system specific traps
1036  *
1037  * Description:
1038  *      Called as the final act of trap_init().  Used in VISWS to initialise
1039  *      the various board specific APIC traps.
1040  **/
1041 void __init x86_quirk_trap_init(void)
1042 {
1043         if (x86_quirks->arch_trap_init) {
1044                 if (x86_quirks->arch_trap_init())
1045                         return;
1046         }
1047 }
1048
1049 static struct irqaction irq0  = {
1050         .handler = timer_interrupt,
1051         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1052         .name = "timer"
1053 };
1054
1055 /**
1056  * x86_quirk_pre_time_init - do any specific initialisations before.
1057  *
1058  **/
1059 void __init x86_quirk_pre_time_init(void)
1060 {
1061         if (x86_quirks->arch_pre_time_init)
1062                 x86_quirks->arch_pre_time_init();
1063 }
1064
1065 /**
1066  * x86_quirk_time_init - do any specific initialisations for the system timer.
1067  *
1068  * Description:
1069  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1070  *      in irq_vectors.h:TIMER_IRQ
1071  **/
1072 void __init x86_quirk_time_init(void)
1073 {
1074         if (x86_quirks->arch_time_init) {
1075                 /*
1076                  * A nonzero return code does not mean failure, it means
1077                  * that the architecture quirk does not want any
1078                  * generic (timer) setup to be performed after this:
1079                  */
1080                 if (x86_quirks->arch_time_init())
1081                         return;
1082         }
1083
1084         irq0.mask = cpumask_of_cpu(0);
1085         setup_irq(0, &irq0);
1086 }
1087 #endif /* CONFIG_X86_32 */