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/sfi.h>
  31 #include <linux/apm_bios.h>
  32 #include <linux/initrd.h>
  33 #include <linux/bootmem.h>
  34 #include <linux/seq_file.h>
  35 #include <linux/console.h>
  36 #include <linux/mca.h>
  37 #include <linux/root_dev.h>
  38 #include <linux/highmem.h>
  39 #include <linux/module.h>
  40 #include <linux/efi.h>
  41 #include <linux/init.h>
  42 #include <linux/edd.h>
  43 #include <linux/iscsi_ibft.h>
  44 #include <linux/nodemask.h>
  45 #include <linux/kexec.h>
  46 #include <linux/dmi.h>
  47 #include <linux/pfn.h>
  48 #include <linux/pci.h>
  49 #include <asm/pci-direct.h>
  50 #include <linux/init_ohci1394_dma.h>
  51 #include <linux/kvm_para.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/kernel.h>
  55 #include <linux/stddef.h>
  56 #include <linux/unistd.h>
  57 #include <linux/ptrace.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 #include <linux/tboot.h>
  70
  71 #include <video/edid.h>
  72
  73 #include <asm/mtrr.h>
  74 #include <asm/apic.h>
  75 #include <asm/trampoline.h>
  76 #include <asm/e820.h>
  77 #include <asm/mpspec.h>
  78 #include <asm/setup.h>
  79 #include <asm/efi.h>
  80 #include <asm/timer.h>
  81 #include <asm/i8259.h>
  82 #include <asm/sections.h>
  83 #include <asm/dmi.h>
  84 #include <asm/io_apic.h>
  85 #include <asm/ist.h>
  86 #include <asm/vmi.h>
  87 #include <asm/setup_arch.h>
  88 #include <asm/bios_ebda.h>
  89 #include <asm/cacheflush.h>
  90 #include <asm/processor.h>
  91 #include <asm/bugs.h>
  92
  93 #include <asm/system.h>
  94 #include <asm/vsyscall.h>
  95 #include <asm/cpu.h>
  96 #include <asm/desc.h>
  97 #include <asm/dma.h>
  98 #include <asm/iommu.h>
  99 #include <asm/gart.h>
 100 #include <asm/mmu_context.h>
 101 #include <asm/proto.h>
 102
 103 #include <asm/paravirt.h>
 104 #include <asm/hypervisor.h>
 105 #include <asm/olpc_ofw.h>
 106
 107 #include <asm/percpu.h>
 108 #include <asm/topology.h>
 109 #include <asm/apicdef.h>
 110 #include <asm/k8.h>
 111 #ifdef CONFIG_X86_64
 112 #include <asm/numa_64.h>
 113 #endif
 114 #include <asm/mce.h>
 115
 116 /*
 117  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
 118  * The direct mapping extends to max_pfn_mapped, so that we can directly access
 119  * apertures, ACPI and other tables without having to play with fixmaps.
 120  */
 121 unsigned long max_low_pfn_mapped;
 122 unsigned long max_pfn_mapped;
 123
 124 #ifdef CONFIG_DMI
 125 RESERVE_BRK(dmi_alloc, 65536);
 126 #endif
 127
 128 unsigned int boot_cpu_id __read_mostly;
 129
 130 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 131 unsigned long _brk_end = (unsigned long)__brk_base;
 132
 133 #ifdef CONFIG_X86_64
 134 int default_cpu_present_to_apicid(int mps_cpu)
 135 {
 136         return __default_cpu_present_to_apicid(mps_cpu);
 137 }
 138
 139 int default_check_phys_apicid_present(int phys_apicid)
 140 {
 141         return __default_check_phys_apicid_present(phys_apicid);
 142 }
 143 #endif
 144
 145 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 146 struct boot_params __initdata boot_params;
 147 #else
 148 struct boot_params boot_params;
 149 #endif
 150
 151 /*
 152  * Machine setup..
 153  */
 154 static struct resource data_resource = {
 155         .name   = "Kernel data",
 156         .start  = 0,
 157         .end    = 0,
 158         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 159 };
 160
 161 static struct resource code_resource = {
 162         .name   = "Kernel code",
 163         .start  = 0,
 164         .end    = 0,
 165         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 166 };
 167
 168 static struct resource bss_resource = {
 169         .name   = "Kernel bss",
 170         .start  = 0,
 171         .end    = 0,
 172         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 173 };
 174
 175
 176 #ifdef CONFIG_X86_32
 177 /* cpu data as detected by the assembly code in head.S */
 178 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 179 /* common cpu data for all cpus */
 180 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 181 EXPORT_SYMBOL(boot_cpu_data);
 182 static void set_mca_bus(int x)
 183 {
 184 #ifdef CONFIG_MCA
 185         MCA_bus = x;
 186 #endif
 187 }
 188
 189 unsigned int def_to_bigsmp;
 190
 191 /* for MCA, but anyone else can use it if they want */
 192 unsigned int machine_id;
 193 unsigned int machine_submodel_id;
 194 unsigned int BIOS_revision;
 195
 196 struct apm_info apm_info;
 197 EXPORT_SYMBOL(apm_info);
 198
 199 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 200         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 201 struct ist_info ist_info;
 202 EXPORT_SYMBOL(ist_info);
 203 #else
 204 struct ist_info ist_info;
 205 #endif
 206
 207 #else
 208 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 209         .x86_phys_bits = MAX_PHYSMEM_BITS,
 210 };
 211 EXPORT_SYMBOL(boot_cpu_data);
 212 #endif
 213
 214
 215 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 216 unsigned long mmu_cr4_features;
 217 #else
 218 unsigned long mmu_cr4_features = X86_CR4_PAE;
 219 #endif
 220
 221 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
 222 int bootloader_type, bootloader_version;
 223
 224 /*
 225  * Setup options
 226  */
 227 struct screen_info screen_info;
 228 EXPORT_SYMBOL(screen_info);
 229 struct edid_info edid_info;
 230 EXPORT_SYMBOL_GPL(edid_info);
 231
 232 extern int root_mountflags;
 233
 234 unsigned long saved_video_mode;
 235
 236 #define RAMDISK_IMAGE_START_MASK        0x07FF
 237 #define RAMDISK_PROMPT_FLAG             0x8000
 238 #define RAMDISK_LOAD_FLAG               0x4000
 239
 240 static char __initdata command_line[COMMAND_LINE_SIZE];
 241 #ifdef CONFIG_CMDLINE_BOOL
 242 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 243 #endif
 244
 245 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 246 struct edd edd;
 247 #ifdef CONFIG_EDD_MODULE
 248 EXPORT_SYMBOL(edd);
 249 #endif
 250 /**
 251  * copy_edd() - Copy the BIOS EDD information
 252  *              from boot_params into a safe place.
 253  *
 254  */
 255 static inline void __init copy_edd(void)
 256 {
 257      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 258             sizeof(edd.mbr_signature));
 259      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 260      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 261      edd.edd_info_nr = boot_params.eddbuf_entries;
 262 }
 263 #else
 264 static inline void __init copy_edd(void)
 265 {
 266 }
 267 #endif
 268
 269 void * __init extend_brk(size_t size, size_t align)
 270 {
 271         size_t mask = align - 1;
 272         void *ret;
 273
 274         BUG_ON(_brk_start == 0);
 275         BUG_ON(align & mask);
 276
 277         _brk_end = (_brk_end + mask) & ~mask;
 278         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 279
 280         ret = (void *)_brk_end;
 281         _brk_end += size;
 282
 283         memset(ret, 0, size);
 284
 285         return ret;
 286 }
 287
 288 #ifdef CONFIG_X86_64
 289 static void __init init_gbpages(void)
 290 {
 291         if (direct_gbpages && cpu_has_gbpages)
 292                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 293         else
 294                 direct_gbpages = 0;
 295 }
 296 #else
 297 static inline void init_gbpages(void)
 298 {
 299 }
 300 #endif
 301
 302 static void __init reserve_brk(void)
 303 {
 304         if (_brk_end > _brk_start)
 305                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 306
 307         /* Mark brk area as locked down and no longer taking any
 308            new allocations */
 309         _brk_start = 0;
 310 }
 311
 312 #ifdef CONFIG_BLK_DEV_INITRD
 313
 314 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 315 static void __init relocate_initrd(void)
 316 {
 317         /* Assume only end is not page aligned */
 318         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 319         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 320         u64 area_size     = PAGE_ALIGN(ramdisk_size);
 321         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 322         u64 ramdisk_here;
 323         unsigned long slop, clen, mapaddr;
 324         char *p, *q;
 325
 326         /* We need to move the initrd down into lowmem */
 327         ramdisk_here = find_e820_area(0, end_of_lowmem, area_size,
 328                                          PAGE_SIZE);
 329
 330         if (ramdisk_here == -1ULL)
 331                 panic("Cannot find place for new RAMDISK of size %lld\n",
 332                          ramdisk_size);
 333
 334         /* Note: this includes all the lowmem currently occupied by
 335            the initrd, we rely on that fact to keep the data intact. */
 336         reserve_early(ramdisk_here, ramdisk_here + area_size,
 337                          "NEW RAMDISK");
 338         initrd_start = ramdisk_here + PAGE_OFFSET;
 339         initrd_end   = initrd_start + ramdisk_size;
 340         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 341                          ramdisk_here, ramdisk_here + ramdisk_size);
 342
 343         q = (char *)initrd_start;
 344
 345         /* Copy any lowmem portion of the initrd */
 346         if (ramdisk_image < end_of_lowmem) {
 347                 clen = end_of_lowmem - ramdisk_image;
 348                 p = (char *)__va(ramdisk_image);
 349                 memcpy(q, p, clen);
 350                 q += clen;
 351                 ramdisk_image += clen;
 352                 ramdisk_size  -= clen;
 353         }
 354
 355         /* Copy the highmem portion of the initrd */
 356         while (ramdisk_size) {
 357                 slop = ramdisk_image & ~PAGE_MASK;
 358                 clen = ramdisk_size;
 359                 if (clen > MAX_MAP_CHUNK-slop)
 360                         clen = MAX_MAP_CHUNK-slop;
 361                 mapaddr = ramdisk_image & PAGE_MASK;
 362                 p = early_memremap(mapaddr, clen+slop);
 363                 memcpy(q, p+slop, clen);
 364                 early_iounmap(p, clen+slop);
 365                 q += clen;
 366                 ramdisk_image += clen;
 367                 ramdisk_size  -= clen;
 368         }
 369         /* high pages is not converted by early_res_to_bootmem */
 370         ramdisk_image = boot_params.hdr.ramdisk_image;
 371         ramdisk_size  = boot_params.hdr.ramdisk_size;
 372         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 373                 " %08llx - %08llx\n",
 374                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 375                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 376 }
 377
 378 static void __init reserve_initrd(void)
 379 {
 380         /* Assume only end is not page aligned */
 381         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 382         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 383         u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
 384         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 385
 386         if (!boot_params.hdr.type_of_loader ||
 387             !ramdisk_image || !ramdisk_size)
 388                 return;         /* No initrd provided by bootloader */
 389
 390         initrd_start = 0;
 391
 392         if (ramdisk_size >= (end_of_lowmem>>1)) {
 393                 free_early(ramdisk_image, ramdisk_end);
 394                 printk(KERN_ERR "initrd too large to handle, "
 395                        "disabling initrd\n");
 396                 return;
 397         }
 398
 399         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 400                         ramdisk_end);
 401
 402
 403         if (ramdisk_end <= end_of_lowmem) {
 404                 /* All in lowmem, easy case */
 405                 /*
 406                  * don't need to reserve again, already reserved early
 407                  * in i386_start_kernel
 408                  */
 409                 initrd_start = ramdisk_image + PAGE_OFFSET;
 410                 initrd_end = initrd_start + ramdisk_size;
 411                 return;
 412         }
 413
 414         relocate_initrd();
 415
 416         free_early(ramdisk_image, ramdisk_end);
 417 }
 418 #else
 419 static void __init reserve_initrd(void)
 420 {
 421 }
 422 #endif /* CONFIG_BLK_DEV_INITRD */
 423
 424 static void __init parse_setup_data(void)
 425 {
 426         struct setup_data *data;
 427         u64 pa_data;
 428
 429         if (boot_params.hdr.version < 0x0209)
 430                 return;
 431         pa_data = boot_params.hdr.setup_data;
 432         while (pa_data) {
 433                 data = early_memremap(pa_data, PAGE_SIZE);
 434                 switch (data->type) {
 435                 case SETUP_E820_EXT:
 436                         parse_e820_ext(data, pa_data);
 437                         break;
 438                 default:
 439                         break;
 440                 }
 441                 pa_data = data->next;
 442                 early_iounmap(data, PAGE_SIZE);
 443         }
 444 }
 445
 446 static void __init e820_reserve_setup_data(void)
 447 {
 448         struct setup_data *data;
 449         u64 pa_data;
 450         int found = 0;
 451
 452         if (boot_params.hdr.version < 0x0209)
 453                 return;
 454         pa_data = boot_params.hdr.setup_data;
 455         while (pa_data) {
 456                 data = early_memremap(pa_data, sizeof(*data));
 457                 e820_update_range(pa_data, sizeof(*data)+data->len,
 458                          E820_RAM, E820_RESERVED_KERN);
 459                 found = 1;
 460                 pa_data = data->next;
 461                 early_iounmap(data, sizeof(*data));
 462         }
 463         if (!found)
 464                 return;
 465
 466         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 467         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 468         printk(KERN_INFO "extended physical RAM map:\n");
 469         e820_print_map("reserve setup_data");
 470 }
 471
 472 static void __init reserve_early_setup_data(void)
 473 {
 474         struct setup_data *data;
 475         u64 pa_data;
 476         char buf[32];
 477
 478         if (boot_params.hdr.version < 0x0209)
 479                 return;
 480         pa_data = boot_params.hdr.setup_data;
 481         while (pa_data) {
 482                 data = early_memremap(pa_data, sizeof(*data));
 483                 sprintf(buf, "setup data %x", data->type);
 484                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 485                 pa_data = data->next;
 486                 early_iounmap(data, sizeof(*data));
 487         }
 488 }
 489
 490 /*
 491  * --------- Crashkernel reservation ------------------------------
 492  */
 493
 494 #ifdef CONFIG_KEXEC
 495
 496 static inline unsigned long long get_total_mem(void)
 497 {
 498         unsigned long long total;
 499
 500         total = max_pfn - min_low_pfn;
 501
 502         return total << PAGE_SHIFT;
 503 }
 504
 505 static void __init reserve_crashkernel(void)
 506 {
 507         unsigned long long total_mem;
 508         unsigned long long crash_size, crash_base;
 509         int ret;
 510
 511         total_mem = get_total_mem();
 512
 513         ret = parse_crashkernel(boot_command_line, total_mem,
 514                         &crash_size, &crash_base);
 515         if (ret != 0 || crash_size <= 0)
 516                 return;
 517
 518         /* 0 means: find the address automatically */
 519         if (crash_base <= 0) {
 520                 const unsigned long long alignment = 16<<20;    /* 16M */
 521
 522                 crash_base = find_e820_area(alignment, ULONG_MAX, crash_size,
 523                                  alignment);
 524                 if (crash_base == -1ULL) {
 525                         pr_info("crashkernel reservation failed - No suitable area found.\n");
 526                         return;
 527                 }
 528         } else {
 529                 unsigned long long start;
 530
 531                 start = find_e820_area(crash_base, ULONG_MAX, crash_size,
 532                                  1<<20);
 533                 if (start != crash_base) {
 534                         pr_info("crashkernel reservation failed - memory is in use.\n");
 535                         return;
 536                 }
 537         }
 538         reserve_early(crash_base, crash_base + crash_size, "CRASH KERNEL");
 539
 540         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 541                         "for crashkernel (System RAM: %ldMB)\n",
 542                         (unsigned long)(crash_size >> 20),
 543                         (unsigned long)(crash_base >> 20),
 544                         (unsigned long)(total_mem >> 20));
 545
 546         crashk_res.start = crash_base;
 547         crashk_res.end   = crash_base + crash_size - 1;
 548         insert_resource(&iomem_resource, &crashk_res);
 549 }
 550 #else
 551 static void __init reserve_crashkernel(void)
 552 {
 553 }
 554 #endif
 555
 556 static struct resource standard_io_resources[] = {
 557         { .name = "dma1", .start = 0x00, .end = 0x1f,
 558                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 559         { .name = "pic1", .start = 0x20, .end = 0x21,
 560                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 561         { .name = "timer0", .start = 0x40, .end = 0x43,
 562                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 563         { .name = "timer1", .start = 0x50, .end = 0x53,
 564                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 565         { .name = "keyboard", .start = 0x60, .end = 0x60,
 566                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 567         { .name = "keyboard", .start = 0x64, .end = 0x64,
 568                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 569         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 570                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 571         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 572                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 573         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 574                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 575         { .name = "fpu", .start = 0xf0, .end = 0xff,
 576                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 577 };
 578
 579 void __init reserve_standard_io_resources(void)
 580 {
 581         int i;
 582
 583         /* request I/O space for devices used on all i[345]86 PCs */
 584         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 585                 request_resource(&ioport_resource, &standard_io_resources[i]);
 586
 587 }
 588
 589 /*
 590  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 591  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 592  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 593  */
 594
 595 #ifdef CONFIG_CRASH_DUMP
 596 /* elfcorehdr= specifies the location of elf core header
 597  * stored by the crashed kernel. This option will be passed
 598  * by kexec loader to the capture kernel.
 599  */
 600 static int __init setup_elfcorehdr(char *arg)
 601 {
 602         char *end;
 603         if (!arg)
 604                 return -EINVAL;
 605         elfcorehdr_addr = memparse(arg, &end);
 606         return end > arg ? 0 : -EINVAL;
 607 }
 608 early_param("elfcorehdr", setup_elfcorehdr);
 609 #endif
 610
 611 static __init void reserve_ibft_region(void)
 612 {
 613         unsigned long addr, size = 0;
 614
 615         addr = find_ibft_region(&size);
 616
 617         if (size)
 618                 reserve_early_overlap_ok(addr, addr + size, "ibft");
 619 }
 620
 621 #ifdef CONFIG_X86_RESERVE_LOW_64K
 622 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 623 {
 624         printk(KERN_NOTICE
 625                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 626                 d->ident);
 627
 628         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 629         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 630
 631         return 0;
 632 }
 633 #endif
 634
 635 /* List of systems that have known low memory corruption BIOS problems */
 636 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 637 #ifdef CONFIG_X86_RESERVE_LOW_64K
 638         {
 639                 .callback = dmi_low_memory_corruption,
 640                 .ident = "AMI BIOS",
 641                 .matches = {
 642                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 643                 },
 644         },
 645         {
 646                 .callback = dmi_low_memory_corruption,
 647                 .ident = "Phoenix BIOS",
 648                 .matches = {
 649                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 650                 },
 651         },
 652         {
 653                 .callback = dmi_low_memory_corruption,
 654                 .ident = "Phoenix/MSC BIOS",
 655                 .matches = {
 656                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix/MSC"),
 657                 },
 658         },
 659         /*
 660          * AMI BIOS with low memory corruption was found on Intel DG45ID and
 661          * DG45FC boards.
 662          * It has a different DMI_BIOS_VENDOR = "Intel Corp.", for now we will
 663          * match only DMI_BOARD_NAME and see if there is more bad products
 664          * with this vendor.
 665          */
 666         {
 667                 .callback = dmi_low_memory_corruption,
 668                 .ident = "AMI BIOS",
 669                 .matches = {
 670                         DMI_MATCH(DMI_BOARD_NAME, "DG45ID"),
 671                 },
 672         },
 673         {
 674                 .callback = dmi_low_memory_corruption,
 675                 .ident = "AMI BIOS",
 676                 .matches = {
 677                         DMI_MATCH(DMI_BOARD_NAME, "DG45FC"),
 678                 },
 679         },
 680         /*
 681          * The Dell Inspiron Mini 1012 has DMI_BIOS_VENDOR = "Dell Inc.", so
 682          * match on the product name.
 683          */
 684         {
 685                 .callback = dmi_low_memory_corruption,
 686                 .ident = "Phoenix BIOS",
 687                 .matches = {
 688                         DMI_MATCH(DMI_PRODUCT_NAME, "Inspiron 1012"),
 689                 },
 690         },
 691 #endif
 692         {}
 693 };
 694
 695 static void __init trim_bios_range(void)
 696 {
 697         /*
 698          * A special case is the first 4Kb of memory;
 699          * This is a BIOS owned area, not kernel ram, but generally
 700          * not listed as such in the E820 table.
 701          */
 702         e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
 703         /*
 704          * special case: Some BIOSen report the PC BIOS
 705          * area (640->1Mb) as ram even though it is not.
 706          * take them out.
 707          */
 708         e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
 709         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 710 }
 711
 712 /*
 713  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 714  * passed the efi memmap, systab, etc., so we should use these data structures
 715  * for initialization.  Note, the efi init code path is determined by the
 716  * global efi_enabled. This allows the same kernel image to be used on existing
 717  * systems (with a traditional BIOS) as well as on EFI systems.
 718  */
 719 /*
 720  * setup_arch - architecture-specific boot-time initializations
 721  *
 722  * Note: On x86_64, fixmaps are ready for use even before this is called.
 723  */
 724
 725 void __init setup_arch(char **cmdline_p)
 726 {
 727         int acpi = 0;
 728         int k8 = 0;
 729
 730 #ifdef CONFIG_X86_32
 731         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 732         visws_early_detect();
 733 #else
 734         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 735 #endif
 736
 737         /* VMI may relocate the fixmap; do this before touching ioremap area */
 738         vmi_init();
 739
 740         /* OFW also may relocate the fixmap */
 741         olpc_ofw_detect();
 742
 743         early_trap_init();
 744         early_cpu_init();
 745         early_ioremap_init();
 746
 747         setup_olpc_ofw_pgd();
 748
 749         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 750         screen_info = boot_params.screen_info;
 751         edid_info = boot_params.edid_info;
 752 #ifdef CONFIG_X86_32
 753         apm_info.bios = boot_params.apm_bios_info;
 754         ist_info = boot_params.ist_info;
 755         if (boot_params.sys_desc_table.length != 0) {
 756                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 757                 machine_id = boot_params.sys_desc_table.table[0];
 758                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 759                 BIOS_revision = boot_params.sys_desc_table.table[2];
 760         }
 761 #endif
 762         saved_video_mode = boot_params.hdr.vid_mode;
 763         bootloader_type = boot_params.hdr.type_of_loader;
 764         if ((bootloader_type >> 4) == 0xe) {
 765                 bootloader_type &= 0xf;
 766                 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
 767         }
 768         bootloader_version  = bootloader_type & 0xf;
 769         bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
 770
 771 #ifdef CONFIG_BLK_DEV_RAM
 772         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 773         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 774         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 775 #endif
 776 #ifdef CONFIG_EFI
 777         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 778 #ifdef CONFIG_X86_32
 779                      "EL32",
 780 #else
 781                      "EL64",
 782 #endif
 783          4)) {
 784                 efi_enabled = 1;
 785                 efi_reserve_early();
 786         }
 787 #endif
 788
 789         x86_init.oem.arch_setup();
 790
 791         iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
 792         setup_memory_map();
 793         parse_setup_data();
 794         /* update the e820_saved too */
 795         e820_reserve_setup_data();
 796
 797         copy_edd();
 798
 799         if (!boot_params.hdr.root_flags)
 800                 root_mountflags &= ~MS_RDONLY;
 801         init_mm.start_code = (unsigned long) _text;
 802         init_mm.end_code = (unsigned long) _etext;
 803         init_mm.end_data = (unsigned long) _edata;
 804         init_mm.brk = _brk_end;
 805
 806         code_resource.start = virt_to_phys(_text);
 807         code_resource.end = virt_to_phys(_etext)-1;
 808         data_resource.start = virt_to_phys(_etext);
 809         data_resource.end = virt_to_phys(_edata)-1;
 810         bss_resource.start = virt_to_phys(&__bss_start);
 811         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 812
 813 #ifdef CONFIG_CMDLINE_BOOL
 814 #ifdef CONFIG_CMDLINE_OVERRIDE
 815         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 816 #else
 817         if (builtin_cmdline[0]) {
 818                 /* append boot loader cmdline to builtin */
 819                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 820                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 821                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 822         }
 823 #endif
 824 #endif
 825
 826         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 827         *cmdline_p = command_line;
 828
 829         /*
 830          * x86_configure_nx() is called before parse_early_param() to detect
 831          * whether hardware doesn't support NX (so that the early EHCI debug
 832          * console setup can safely call set_fixmap()). It may then be called
 833          * again from within noexec_setup() during parsing early parameters
 834          * to honor the respective command line option.
 835          */
 836         x86_configure_nx();
 837
 838         parse_early_param();
 839
 840         x86_report_nx();
 841
 842         /* Must be before kernel pagetables are setup */
 843         vmi_activate();
 844
 845         /* after early param, so could get panic from serial */
 846         reserve_early_setup_data();
 847
 848         if (acpi_mps_check()) {
 849 #ifdef CONFIG_X86_LOCAL_APIC
 850                 disable_apic = 1;
 851 #endif
 852                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 853         }
 854
 855 #ifdef CONFIG_PCI
 856         if (pci_early_dump_regs)
 857                 early_dump_pci_devices();
 858 #endif
 859
 860         finish_e820_parsing();
 861
 862         if (efi_enabled)
 863                 efi_init();
 864
 865         dmi_scan_machine();
 866
 867         dmi_check_system(bad_bios_dmi_table);
 868
 869         /*
 870          * VMware detection requires dmi to be available, so this
 871          * needs to be done after dmi_scan_machine, for the BP.
 872          */
 873         init_hypervisor_platform();
 874
 875         x86_init.resources.probe_roms();
 876
 877         /* after parse_early_param, so could debug it */
 878         insert_resource(&iomem_resource, &code_resource);
 879         insert_resource(&iomem_resource, &data_resource);
 880         insert_resource(&iomem_resource, &bss_resource);
 881
 882         trim_bios_range();
 883 #ifdef CONFIG_X86_32
 884         if (ppro_with_ram_bug()) {
 885                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 886                                   E820_RESERVED);
 887                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 888                 printk(KERN_INFO "fixed physical RAM map:\n");
 889                 e820_print_map("bad_ppro");
 890         }
 891 #else
 892         early_gart_iommu_check();
 893 #endif
 894
 895         /*
 896          * partially used pages are not usable - thus
 897          * we are rounding upwards:
 898          */
 899         max_pfn = e820_end_of_ram_pfn();
 900
 901         /* preallocate 4k for mptable mpc */
 902         early_reserve_e820_mpc_new();
 903         /* update e820 for memory not covered by WB MTRRs */
 904         mtrr_bp_init();
 905         if (mtrr_trim_uncached_memory(max_pfn))
 906                 max_pfn = e820_end_of_ram_pfn();
 907
 908 #ifdef CONFIG_X86_32
 909         /* max_low_pfn get updated here */
 910         find_low_pfn_range();
 911 #else
 912         num_physpages = max_pfn;
 913
 914         check_x2apic();
 915
 916         /* How many end-of-memory variables you have, grandma! */
 917         /* need this before calling reserve_initrd */
 918         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 919                 max_low_pfn = e820_end_of_low_ram_pfn();
 920         else
 921                 max_low_pfn = max_pfn;
 922
 923         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 924         max_pfn_mapped = KERNEL_IMAGE_SIZE >> PAGE_SHIFT;
 925 #endif
 926
 927 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 928         setup_bios_corruption_check();
 929 #endif
 930
 931         printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
 932                         max_pfn_mapped<<PAGE_SHIFT);
 933
 934         reserve_brk();
 935
 936         /*
 937          * Find and reserve possible boot-time SMP configuration:
 938          */
 939         find_smp_config();
 940
 941         reserve_ibft_region();
 942
 943         reserve_trampoline_memory();
 944
 945 #ifdef CONFIG_ACPI_SLEEP
 946         /*
 947          * Reserve low memory region for sleep support.
 948          * even before init_memory_mapping
 949          */
 950         acpi_reserve_wakeup_memory();
 951 #endif
 952         init_gbpages();
 953
 954         /* max_pfn_mapped is updated here */
 955         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 956         max_pfn_mapped = max_low_pfn_mapped;
 957
 958 #ifdef CONFIG_X86_64
 959         if (max_pfn > max_low_pfn) {
 960                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 961                                                      max_pfn<<PAGE_SHIFT);
 962                 /* can we preseve max_low_pfn ?*/
 963                 max_low_pfn = max_pfn;
 964         }
 965 #endif
 966
 967         /*
 968          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 969          */
 970
 971 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 972         if (init_ohci1394_dma_early)
 973                 init_ohci1394_dma_on_all_controllers();
 974 #endif
 975
 976         reserve_initrd();
 977
 978         reserve_crashkernel();
 979
 980         vsmp_init();
 981
 982         io_delay_init();
 983
 984         /*
 985          * Parse the ACPI tables for possible boot-time SMP configuration.
 986          */
 987         acpi_boot_table_init();
 988
 989         early_acpi_boot_init();
 990
 991 #ifdef CONFIG_ACPI_NUMA
 992         /*
 993          * Parse SRAT to discover nodes.
 994          */
 995         acpi = acpi_numa_init();
 996 #endif
 997
 998 #ifdef CONFIG_K8_NUMA
 999         if (!acpi)
1000                 k8 = !k8_numa_init(0, max_pfn);
1001 #endif
1002
1003         initmem_init(0, max_pfn, acpi, k8);
1004 #ifndef CONFIG_NO_BOOTMEM
1005         early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
1006 #endif
1007
1008         dma32_reserve_bootmem();
1009
1010 #ifdef CONFIG_KVM_CLOCK
1011         kvmclock_init();
1012 #endif
1013
1014         x86_init.paging.pagetable_setup_start(swapper_pg_dir);
1015         paging_init();
1016         x86_init.paging.pagetable_setup_done(swapper_pg_dir);
1017
1018         setup_trampoline_page_table();
1019
1020         tboot_probe();
1021
1022 #ifdef CONFIG_X86_64
1023         map_vsyscall();
1024 #endif
1025
1026         generic_apic_probe();
1027
1028         early_quirks();
1029
1030         /*
1031          * Read APIC and some other early information from ACPI tables.
1032          */
1033         acpi_boot_init();
1034
1035         sfi_init();
1036
1037         /*
1038          * get boot-time SMP configuration:
1039          */
1040         if (smp_found_config)
1041                 get_smp_config();
1042
1043         prefill_possible_map();
1044
1045 #ifdef CONFIG_X86_64
1046         init_cpu_to_node();
1047 #endif
1048
1049         init_apic_mappings();
1050         ioapic_init_mappings();
1051
1052         /* need to wait for io_apic is mapped */
1053         probe_nr_irqs_gsi();
1054
1055         kvm_guest_init();
1056
1057         e820_reserve_resources();
1058         e820_mark_nosave_regions(max_low_pfn);
1059
1060         x86_init.resources.reserve_resources();
1061
1062         e820_setup_gap();
1063
1064 #ifdef CONFIG_VT
1065 #if defined(CONFIG_VGA_CONSOLE)
1066         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1067                 conswitchp = &vga_con;
1068 #elif defined(CONFIG_DUMMY_CONSOLE)
1069         conswitchp = &dummy_con;
1070 #endif
1071 #endif
1072         x86_init.oem.banner();
1073
1074         mcheck_init();
1075 }
1076
1077 #ifdef CONFIG_X86_32
1078
1079 static struct resource video_ram_resource = {
1080         .name   = "Video RAM area",
1081         .start  = 0xa0000,
1082         .end    = 0xbffff,
1083         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
1084 };
1085
1086 void __init i386_reserve_resources(void)
1087 {
1088         request_resource(&iomem_resource, &video_ram_resource);
1089         reserve_standard_io_resources();
1090 }
1091
1092 #endif /* CONFIG_X86_32 */