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/arch_hooks.h>
  78 #include <asm/efi.h>
  79 #include <asm/sections.h>
  80 #include <asm/dmi.h>
  81 #include <asm/io_apic.h>
  82 #include <asm/ist.h>
  83 #include <asm/vmi.h>
  84 #include <setup_arch.h>
  85 #include <asm/bios_ebda.h>
  86 #include <asm/cacheflush.h>
  87 #include <asm/processor.h>
  88 #include <asm/bugs.h>
  89
  90 #include <asm/system.h>
  91 #include <asm/vsyscall.h>
  92 #include <asm/smp.h>
  93 #include <asm/desc.h>
  94 #include <asm/dma.h>
  95 #include <asm/iommu.h>
  96 #include <asm/mmu_context.h>
  97 #include <asm/proto.h>
  98
  99 #include <mach_apic.h>
 100 #include <asm/paravirt.h>
 101
 102 #include <asm/percpu.h>
 103 #include <asm/topology.h>
 104 #include <asm/apicdef.h>
 105 #ifdef CONFIG_X86_64
 106 #include <asm/numa_64.h>
 107 #endif
 108
 109 #ifndef ARCH_SETUP
 110 #define ARCH_SETUP
 111 #endif
 112
 113 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 114 struct boot_params __initdata boot_params;
 115 #else
 116 struct boot_params boot_params;
 117 #endif
 118
 119 /*
 120  * Machine setup..
 121  */
 122 static struct resource data_resource = {
 123         .name   = "Kernel data",
 124         .start  = 0,
 125         .end    = 0,
 126         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 127 };
 128
 129 static struct resource code_resource = {
 130         .name   = "Kernel code",
 131         .start  = 0,
 132         .end    = 0,
 133         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 134 };
 135
 136 static struct resource bss_resource = {
 137         .name   = "Kernel bss",
 138         .start  = 0,
 139         .end    = 0,
 140         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 141 };
 142
 143
 144 #ifdef CONFIG_X86_32
 145 /* This value is set up by the early boot code to point to the value
 146    immediately after the boot time page tables.  It contains a *physical*
 147    address, and must not be in the .bss segment! */
 148 unsigned long init_pg_tables_start __initdata = ~0UL;
 149 unsigned long init_pg_tables_end __initdata = ~0UL;
 150
 151 static struct resource video_ram_resource = {
 152         .name   = "Video RAM area",
 153         .start  = 0xa0000,
 154         .end    = 0xbffff,
 155         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 156 };
 157
 158 /* cpu data as detected by the assembly code in head.S */
 159 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 160 /* common cpu data for all cpus */
 161 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 162 EXPORT_SYMBOL(boot_cpu_data);
 163 static void set_mca_bus(int x)
 164 {
 165 #ifdef CONFIG_MCA
 166         MCA_bus = x;
 167 #endif
 168 }
 169
 170 unsigned int def_to_bigsmp;
 171
 172 /* for MCA, but anyone else can use it if they want */
 173 unsigned int machine_id;
 174 unsigned int machine_submodel_id;
 175 unsigned int BIOS_revision;
 176
 177 struct apm_info apm_info;
 178 EXPORT_SYMBOL(apm_info);
 179
 180 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 181         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 182 struct ist_info ist_info;
 183 EXPORT_SYMBOL(ist_info);
 184 #else
 185 struct ist_info ist_info;
 186 #endif
 187
 188 #else
 189 struct cpuinfo_x86 boot_cpu_data __read_mostly;
 190 EXPORT_SYMBOL(boot_cpu_data);
 191 #endif
 192
 193
 194 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 195 unsigned long mmu_cr4_features;
 196 #else
 197 unsigned long mmu_cr4_features = X86_CR4_PAE;
 198 #endif
 199
 200 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 201 int bootloader_type;
 202
 203 /*
 204  * Early DMI memory
 205  */
 206 int dmi_alloc_index;
 207 char dmi_alloc_data[DMI_MAX_DATA];
 208
 209 /*
 210  * Setup options
 211  */
 212 struct screen_info screen_info;
 213 EXPORT_SYMBOL(screen_info);
 214 struct edid_info edid_info;
 215 EXPORT_SYMBOL_GPL(edid_info);
 216
 217 extern int root_mountflags;
 218
 219 unsigned long saved_video_mode;
 220
 221 #define RAMDISK_IMAGE_START_MASK        0x07FF
 222 #define RAMDISK_PROMPT_FLAG             0x8000
 223 #define RAMDISK_LOAD_FLAG               0x4000
 224
 225 static char __initdata command_line[COMMAND_LINE_SIZE];
 226 #ifdef CONFIG_CMDLINE_BOOL
 227 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 228 #endif
 229
 230 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 231 struct edd edd;
 232 #ifdef CONFIG_EDD_MODULE
 233 EXPORT_SYMBOL(edd);
 234 #endif
 235 /**
 236  * copy_edd() - Copy the BIOS EDD information
 237  *              from boot_params into a safe place.
 238  *
 239  */
 240 static inline void copy_edd(void)
 241 {
 242      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 243             sizeof(edd.mbr_signature));
 244      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 245      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 246      edd.edd_info_nr = boot_params.eddbuf_entries;
 247 }
 248 #else
 249 static inline void copy_edd(void)
 250 {
 251 }
 252 #endif
 253
 254 #ifdef CONFIG_BLK_DEV_INITRD
 255
 256 #ifdef CONFIG_X86_32
 257
 258 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 259 static void __init relocate_initrd(void)
 260 {
 261
 262         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 263         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 264         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 265         u64 ramdisk_here;
 266         unsigned long slop, clen, mapaddr;
 267         char *p, *q;
 268
 269         /* We need to move the initrd down into lowmem */
 270         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 271                                          PAGE_SIZE);
 272
 273         if (ramdisk_here == -1ULL)
 274                 panic("Cannot find place for new RAMDISK of size %lld\n",
 275                          ramdisk_size);
 276
 277         /* Note: this includes all the lowmem currently occupied by
 278            the initrd, we rely on that fact to keep the data intact. */
 279         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 280                          "NEW RAMDISK");
 281         initrd_start = ramdisk_here + PAGE_OFFSET;
 282         initrd_end   = initrd_start + ramdisk_size;
 283         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 284                          ramdisk_here, ramdisk_here + ramdisk_size);
 285
 286         q = (char *)initrd_start;
 287
 288         /* Copy any lowmem portion of the initrd */
 289         if (ramdisk_image < end_of_lowmem) {
 290                 clen = end_of_lowmem - ramdisk_image;
 291                 p = (char *)__va(ramdisk_image);
 292                 memcpy(q, p, clen);
 293                 q += clen;
 294                 ramdisk_image += clen;
 295                 ramdisk_size  -= clen;
 296         }
 297
 298         /* Copy the highmem portion of the initrd */
 299         while (ramdisk_size) {
 300                 slop = ramdisk_image & ~PAGE_MASK;
 301                 clen = ramdisk_size;
 302                 if (clen > MAX_MAP_CHUNK-slop)
 303                         clen = MAX_MAP_CHUNK-slop;
 304                 mapaddr = ramdisk_image & PAGE_MASK;
 305                 p = early_memremap(mapaddr, clen+slop);
 306                 memcpy(q, p+slop, clen);
 307                 early_iounmap(p, clen+slop);
 308                 q += clen;
 309                 ramdisk_image += clen;
 310                 ramdisk_size  -= clen;
 311         }
 312         /* high pages is not converted by early_res_to_bootmem */
 313         ramdisk_image = boot_params.hdr.ramdisk_image;
 314         ramdisk_size  = boot_params.hdr.ramdisk_size;
 315         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 316                 " %08llx - %08llx\n",
 317                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 318                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 319 }
 320 #endif
 321
 322 static void __init reserve_initrd(void)
 323 {
 324         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 325         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 326         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 327         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 328
 329         if (!boot_params.hdr.type_of_loader ||
 330             !ramdisk_image || !ramdisk_size)
 331                 return;         /* No initrd provided by bootloader */
 332
 333         initrd_start = 0;
 334
 335         if (ramdisk_size >= (end_of_lowmem>>1)) {
 336                 free_early(ramdisk_image, ramdisk_end);
 337                 printk(KERN_ERR "initrd too large to handle, "
 338                        "disabling initrd\n");
 339                 return;
 340         }
 341
 342         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 343                         ramdisk_end);
 344
 345
 346         if (ramdisk_end <= end_of_lowmem) {
 347                 /* All in lowmem, easy case */
 348                 /*
 349                  * don't need to reserve again, already reserved early
 350                  * in i386_start_kernel
 351                  */
 352                 initrd_start = ramdisk_image + PAGE_OFFSET;
 353                 initrd_end = initrd_start + ramdisk_size;
 354                 return;
 355         }
 356
 357 #ifdef CONFIG_X86_32
 358         relocate_initrd();
 359 #else
 360         printk(KERN_ERR "initrd extends beyond end of memory "
 361                "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
 362                ramdisk_end, end_of_lowmem);
 363         initrd_start = 0;
 364 #endif
 365         free_early(ramdisk_image, ramdisk_end);
 366 }
 367 #else
 368 static void __init reserve_initrd(void)
 369 {
 370 }
 371 #endif /* CONFIG_BLK_DEV_INITRD */
 372
 373 static void __init parse_setup_data(void)
 374 {
 375         struct setup_data *data;
 376         u64 pa_data;
 377
 378         if (boot_params.hdr.version < 0x0209)
 379                 return;
 380         pa_data = boot_params.hdr.setup_data;
 381         while (pa_data) {
 382                 data = early_memremap(pa_data, PAGE_SIZE);
 383                 switch (data->type) {
 384                 case SETUP_E820_EXT:
 385                         parse_e820_ext(data, pa_data);
 386                         break;
 387                 default:
 388                         break;
 389                 }
 390                 pa_data = data->next;
 391                 early_iounmap(data, PAGE_SIZE);
 392         }
 393 }
 394
 395 static void __init e820_reserve_setup_data(void)
 396 {
 397         struct setup_data *data;
 398         u64 pa_data;
 399         int found = 0;
 400
 401         if (boot_params.hdr.version < 0x0209)
 402                 return;
 403         pa_data = boot_params.hdr.setup_data;
 404         while (pa_data) {
 405                 data = early_memremap(pa_data, sizeof(*data));
 406                 e820_update_range(pa_data, sizeof(*data)+data->len,
 407                          E820_RAM, E820_RESERVED_KERN);
 408                 found = 1;
 409                 pa_data = data->next;
 410                 early_iounmap(data, sizeof(*data));
 411         }
 412         if (!found)
 413                 return;
 414
 415         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 416         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 417         printk(KERN_INFO "extended physical RAM map:\n");
 418         e820_print_map("reserve setup_data");
 419 }
 420
 421 static void __init reserve_early_setup_data(void)
 422 {
 423         struct setup_data *data;
 424         u64 pa_data;
 425         char buf[32];
 426
 427         if (boot_params.hdr.version < 0x0209)
 428                 return;
 429         pa_data = boot_params.hdr.setup_data;
 430         while (pa_data) {
 431                 data = early_memremap(pa_data, sizeof(*data));
 432                 sprintf(buf, "setup data %x", data->type);
 433                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 434                 pa_data = data->next;
 435                 early_iounmap(data, sizeof(*data));
 436         }
 437 }
 438
 439 /*
 440  * --------- Crashkernel reservation ------------------------------
 441  */
 442
 443 #ifdef CONFIG_KEXEC
 444
 445 /**
 446  * Reserve @size bytes of crashkernel memory at any suitable offset.
 447  *
 448  * @size: Size of the crashkernel memory to reserve.
 449  * Returns the base address on success, and -1ULL on failure.
 450  */
 451 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 452 {
 453         const unsigned long long alignment = 16<<20;    /* 16M */
 454         unsigned long long start = 0LL;
 455
 456         while (1) {
 457                 int ret;
 458
 459                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 460                 if (start == -1ULL)
 461                         return start;
 462
 463                 /* try to reserve it */
 464                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 465                 if (ret >= 0)
 466                         return start;
 467
 468                 start += alignment;
 469         }
 470 }
 471
 472 static inline unsigned long long get_total_mem(void)
 473 {
 474         unsigned long long total;
 475
 476         total = max_low_pfn - min_low_pfn;
 477 #ifdef CONFIG_HIGHMEM
 478         total += highend_pfn - highstart_pfn;
 479 #endif
 480
 481         return total << PAGE_SHIFT;
 482 }
 483
 484 static void __init reserve_crashkernel(void)
 485 {
 486         unsigned long long total_mem;
 487         unsigned long long crash_size, crash_base;
 488         int ret;
 489
 490         total_mem = get_total_mem();
 491
 492         ret = parse_crashkernel(boot_command_line, total_mem,
 493                         &crash_size, &crash_base);
 494         if (ret != 0 || crash_size <= 0)
 495                 return;
 496
 497         /* 0 means: find the address automatically */
 498         if (crash_base <= 0) {
 499                 crash_base = find_and_reserve_crashkernel(crash_size);
 500                 if (crash_base == -1ULL) {
 501                         pr_info("crashkernel reservation failed. "
 502                                 "No suitable area found.\n");
 503                         return;
 504                 }
 505         } else {
 506                 ret = reserve_bootmem_generic(crash_base, crash_size,
 507                                         BOOTMEM_EXCLUSIVE);
 508                 if (ret < 0) {
 509                         pr_info("crashkernel reservation failed - "
 510                                 "memory is in use\n");
 511                         return;
 512                 }
 513         }
 514
 515         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 516                         "for crashkernel (System RAM: %ldMB)\n",
 517                         (unsigned long)(crash_size >> 20),
 518                         (unsigned long)(crash_base >> 20),
 519                         (unsigned long)(total_mem >> 20));
 520
 521         crashk_res.start = crash_base;
 522         crashk_res.end   = crash_base + crash_size - 1;
 523         insert_resource(&iomem_resource, &crashk_res);
 524 }
 525 #else
 526 static void __init reserve_crashkernel(void)
 527 {
 528 }
 529 #endif
 530
 531 static struct resource standard_io_resources[] = {
 532         { .name = "dma1", .start = 0x00, .end = 0x1f,
 533                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 534         { .name = "pic1", .start = 0x20, .end = 0x21,
 535                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 536         { .name = "timer0", .start = 0x40, .end = 0x43,
 537                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 538         { .name = "timer1", .start = 0x50, .end = 0x53,
 539                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 540         { .name = "keyboard", .start = 0x60, .end = 0x60,
 541                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 542         { .name = "keyboard", .start = 0x64, .end = 0x64,
 543                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 544         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 545                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 546         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 547                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 548         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 549                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 550         { .name = "fpu", .start = 0xf0, .end = 0xff,
 551                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 552 };
 553
 554 static void __init reserve_standard_io_resources(void)
 555 {
 556         int i;
 557
 558         /* request I/O space for devices used on all i[345]86 PCs */
 559         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 560                 request_resource(&ioport_resource, &standard_io_resources[i]);
 561
 562 }
 563
 564 /*
 565  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 566  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 567  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 568  */
 569
 570 #ifdef CONFIG_CRASH_DUMP
 571 /* elfcorehdr= specifies the location of elf core header
 572  * stored by the crashed kernel. This option will be passed
 573  * by kexec loader to the capture kernel.
 574  */
 575 static int __init setup_elfcorehdr(char *arg)
 576 {
 577         char *end;
 578         if (!arg)
 579                 return -EINVAL;
 580         elfcorehdr_addr = memparse(arg, &end);
 581         return end > arg ? 0 : -EINVAL;
 582 }
 583 early_param("elfcorehdr", setup_elfcorehdr);
 584 #endif
 585
 586 static int __init default_update_genapic(void)
 587 {
 588 #ifdef CONFIG_X86_SMP
 589 # if defined(CONFIG_X86_GENERICARCH) || defined(CONFIG_X86_64)
 590         genapic->wakeup_cpu = wakeup_secondary_cpu_via_init;
 591 # endif
 592 #endif
 593
 594         return 0;
 595 }
 596
 597 static struct x86_quirks default_x86_quirks __initdata = {
 598         .update_genapic         = default_update_genapic,
 599 };
 600
 601 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 602
 603 /*
 604  * Some BIOSes seem to corrupt the low 64k of memory during events
 605  * like suspend/resume and unplugging an HDMI cable.  Reserve all
 606  * remaining free memory in that area and fill it with a distinct
 607  * pattern.
 608  */
 609 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 610 #define MAX_SCAN_AREAS  8
 611
 612 static int __read_mostly memory_corruption_check = -1;
 613
 614 static unsigned __read_mostly corruption_check_size = 64*1024;
 615 static unsigned __read_mostly corruption_check_period = 60; /* seconds */
 616
 617 static struct e820entry scan_areas[MAX_SCAN_AREAS];
 618 static int num_scan_areas;
 619
 620
 621 static int set_corruption_check(char *arg)
 622 {
 623         char *end;
 624
 625         memory_corruption_check = simple_strtol(arg, &end, 10);
 626
 627         return (*end == 0) ? 0 : -EINVAL;
 628 }
 629 early_param("memory_corruption_check", set_corruption_check);
 630
 631 static int set_corruption_check_period(char *arg)
 632 {
 633         char *end;
 634
 635         corruption_check_period = simple_strtoul(arg, &end, 10);
 636
 637         return (*end == 0) ? 0 : -EINVAL;
 638 }
 639 early_param("memory_corruption_check_period", set_corruption_check_period);
 640
 641 static int set_corruption_check_size(char *arg)
 642 {
 643         char *end;
 644         unsigned size;
 645
 646         size = memparse(arg, &end);
 647
 648         if (*end == '\0')
 649                 corruption_check_size = size;
 650
 651         return (size == corruption_check_size) ? 0 : -EINVAL;
 652 }
 653 early_param("memory_corruption_check_size", set_corruption_check_size);
 654
 655
 656 static void __init setup_bios_corruption_check(void)
 657 {
 658         u64 addr = PAGE_SIZE;   /* assume first page is reserved anyway */
 659
 660         if (memory_corruption_check == -1) {
 661                 memory_corruption_check =
 662 #ifdef CONFIG_X86_BOOTPARAM_MEMORY_CORRUPTION_CHECK
 663                         1
 664 #else
 665                         0
 666 #endif
 667                         ;
 668         }
 669
 670         if (corruption_check_size == 0)
 671                 memory_corruption_check = 0;
 672
 673         if (!memory_corruption_check)
 674                 return;
 675
 676         corruption_check_size = round_up(corruption_check_size, PAGE_SIZE);
 677
 678         while(addr < corruption_check_size && num_scan_areas < MAX_SCAN_AREAS) {
 679                 u64 size;
 680                 addr = find_e820_area_size(addr, &size, PAGE_SIZE);
 681
 682                 if (addr == 0)
 683                         break;
 684
 685                 if ((addr + size) > corruption_check_size)
 686                         size = corruption_check_size - addr;
 687
 688                 if (size == 0)
 689                         break;
 690
 691                 e820_update_range(addr, size, E820_RAM, E820_RESERVED);
 692                 scan_areas[num_scan_areas].addr = addr;
 693                 scan_areas[num_scan_areas].size = size;
 694                 num_scan_areas++;
 695
 696                 /* Assume we've already mapped this early memory */
 697                 memset(__va(addr), 0, size);
 698
 699                 addr += size;
 700         }
 701
 702         printk(KERN_INFO "Scanning %d areas for low memory corruption\n",
 703                num_scan_areas);
 704         update_e820();
 705 }
 706
 707 static struct timer_list periodic_check_timer;
 708
 709 void check_for_bios_corruption(void)
 710 {
 711         int i;
 712         int corruption = 0;
 713
 714         if (!memory_corruption_check)
 715                 return;
 716
 717         for(i = 0; i < num_scan_areas; i++) {
 718                 unsigned long *addr = __va(scan_areas[i].addr);
 719                 unsigned long size = scan_areas[i].size;
 720
 721                 for(; size; addr++, size -= sizeof(unsigned long)) {
 722                         if (!*addr)
 723                                 continue;
 724                         printk(KERN_ERR "Corrupted low memory at %p (%lx phys) = %08lx\n",
 725                                addr, __pa(addr), *addr);
 726                         corruption = 1;
 727                         *addr = 0;
 728                 }
 729         }
 730
 731         WARN(corruption, KERN_ERR "Memory corruption detected in low memory\n");
 732 }
 733
 734 static void periodic_check_for_corruption(unsigned long data)
 735 {
 736         check_for_bios_corruption();
 737         mod_timer(&periodic_check_timer, round_jiffies(jiffies + corruption_check_period*HZ));
 738 }
 739
 740 void start_periodic_check_for_corruption(void)
 741 {
 742         if (!memory_corruption_check || corruption_check_period == 0)
 743                 return;
 744
 745         printk(KERN_INFO "Scanning for low memory corruption every %d seconds\n",
 746                corruption_check_period);
 747
 748         init_timer(&periodic_check_timer);
 749         periodic_check_timer.function = &periodic_check_for_corruption;
 750         periodic_check_for_corruption(0);
 751 }
 752 #endif
 753
 754 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 755 {
 756         printk(KERN_NOTICE
 757                 "%s detected: BIOS may corrupt low RAM, working it around.\n",
 758                 d->ident);
 759
 760         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 761         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 762
 763         return 0;
 764 }
 765
 766 /* List of systems that have known low memory corruption BIOS problems */
 767 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 768 #ifdef CONFIG_X86_RESERVE_LOW_64K
 769         {
 770                 .callback = dmi_low_memory_corruption,
 771                 .ident = "AMI BIOS",
 772                 .matches = {
 773                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 774                 },
 775         },
 776         {
 777                 .callback = dmi_low_memory_corruption,
 778                 .ident = "Phoenix BIOS",
 779                 .matches = {
 780                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 781                 },
 782         },
 783 #endif
 784         {}
 785 };
 786
 787 /*
 788  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 789  * passed the efi memmap, systab, etc., so we should use these data structures
 790  * for initialization.  Note, the efi init code path is determined by the
 791  * global efi_enabled. This allows the same kernel image to be used on existing
 792  * systems (with a traditional BIOS) as well as on EFI systems.
 793  */
 794 /*
 795  * setup_arch - architecture-specific boot-time initializations
 796  *
 797  * Note: On x86_64, fixmaps are ready for use even before this is called.
 798  */
 799
 800 void __init setup_arch(char **cmdline_p)
 801 {
 802 #ifdef CONFIG_X86_32
 803         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 804         visws_early_detect();
 805         pre_setup_arch_hook();
 806 #else
 807         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 808 #endif
 809
 810         early_cpu_init();
 811         early_ioremap_init();
 812
 813         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 814         screen_info = boot_params.screen_info;
 815         edid_info = boot_params.edid_info;
 816 #ifdef CONFIG_X86_32
 817         apm_info.bios = boot_params.apm_bios_info;
 818         ist_info = boot_params.ist_info;
 819         if (boot_params.sys_desc_table.length != 0) {
 820                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 821                 machine_id = boot_params.sys_desc_table.table[0];
 822                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 823                 BIOS_revision = boot_params.sys_desc_table.table[2];
 824         }
 825 #endif
 826         saved_video_mode = boot_params.hdr.vid_mode;
 827         bootloader_type = boot_params.hdr.type_of_loader;
 828
 829 #ifdef CONFIG_BLK_DEV_RAM
 830         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 831         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 832         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 833 #endif
 834 #ifdef CONFIG_EFI
 835         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 836 #ifdef CONFIG_X86_32
 837                      "EL32",
 838 #else
 839                      "EL64",
 840 #endif
 841          4)) {
 842                 efi_enabled = 1;
 843                 efi_reserve_early();
 844         }
 845 #endif
 846
 847         ARCH_SETUP
 848
 849         setup_memory_map();
 850         parse_setup_data();
 851         /* update the e820_saved too */
 852         e820_reserve_setup_data();
 853
 854         copy_edd();
 855
 856         if (!boot_params.hdr.root_flags)
 857                 root_mountflags &= ~MS_RDONLY;
 858         init_mm.start_code = (unsigned long) _text;
 859         init_mm.end_code = (unsigned long) _etext;
 860         init_mm.end_data = (unsigned long) _edata;
 861 #ifdef CONFIG_X86_32
 862         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 863 #else
 864         init_mm.brk = (unsigned long) &_end;
 865 #endif
 866
 867         code_resource.start = virt_to_phys(_text);
 868         code_resource.end = virt_to_phys(_etext)-1;
 869         data_resource.start = virt_to_phys(_etext);
 870         data_resource.end = virt_to_phys(_edata)-1;
 871         bss_resource.start = virt_to_phys(&__bss_start);
 872         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 873
 874 #ifdef CONFIG_CMDLINE_BOOL
 875 #ifdef CONFIG_CMDLINE_OVERRIDE
 876         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 877 #else
 878         if (builtin_cmdline[0]) {
 879                 /* append boot loader cmdline to builtin */
 880                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 881                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 882                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 883         }
 884 #endif
 885 #endif
 886
 887         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 888         *cmdline_p = command_line;
 889
 890         parse_early_param();
 891
 892 #ifdef CONFIG_X86_64
 893         check_efer();
 894 #endif
 895
 896 #if defined(CONFIG_VMI) && defined(CONFIG_X86_32)
 897         /*
 898          * Must be before kernel pagetables are setup
 899          * or fixmap area is touched.
 900          */
 901         vmi_init();
 902 #endif
 903
 904         /* after early param, so could get panic from serial */
 905         reserve_early_setup_data();
 906
 907         if (acpi_mps_check()) {
 908 #ifdef CONFIG_X86_LOCAL_APIC
 909                 disable_apic = 1;
 910 #endif
 911                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 912         }
 913
 914 #ifdef CONFIG_PCI
 915         if (pci_early_dump_regs)
 916                 early_dump_pci_devices();
 917 #endif
 918
 919         finish_e820_parsing();
 920
 921         dmi_scan_machine();
 922
 923         dmi_check_system(bad_bios_dmi_table);
 924
 925 #ifdef CONFIG_X86_32
 926         probe_roms();
 927 #endif
 928
 929         /* after parse_early_param, so could debug it */
 930         insert_resource(&iomem_resource, &code_resource);
 931         insert_resource(&iomem_resource, &data_resource);
 932         insert_resource(&iomem_resource, &bss_resource);
 933
 934         if (efi_enabled)
 935                 efi_init();
 936
 937 #ifdef CONFIG_X86_32
 938         if (ppro_with_ram_bug()) {
 939                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 940                                   E820_RESERVED);
 941                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 942                 printk(KERN_INFO "fixed physical RAM map:\n");
 943                 e820_print_map("bad_ppro");
 944         }
 945 #else
 946         early_gart_iommu_check();
 947 #endif
 948
 949         /*
 950          * partially used pages are not usable - thus
 951          * we are rounding upwards:
 952          */
 953         max_pfn = e820_end_of_ram_pfn();
 954
 955         /* preallocate 4k for mptable mpc */
 956         early_reserve_e820_mpc_new();
 957         /* update e820 for memory not covered by WB MTRRs */
 958         mtrr_bp_init();
 959         if (mtrr_trim_uncached_memory(max_pfn))
 960                 max_pfn = e820_end_of_ram_pfn();
 961
 962 #ifdef CONFIG_X86_32
 963         /* max_low_pfn get updated here */
 964         find_low_pfn_range();
 965 #else
 966         num_physpages = max_pfn;
 967
 968         if (cpu_has_x2apic)
 969                 check_x2apic();
 970
 971         /* How many end-of-memory variables you have, grandma! */
 972         /* need this before calling reserve_initrd */
 973         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 974                 max_low_pfn = e820_end_of_low_ram_pfn();
 975         else
 976                 max_low_pfn = max_pfn;
 977
 978         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 979 #endif
 980
 981 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 982         setup_bios_corruption_check();
 983 #endif
 984
 985         /* max_pfn_mapped is updated here */
 986         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 987         max_pfn_mapped = max_low_pfn_mapped;
 988
 989 #ifdef CONFIG_X86_64
 990         if (max_pfn > max_low_pfn) {
 991                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 992                                                      max_pfn<<PAGE_SHIFT);
 993                 /* can we preseve max_low_pfn ?*/
 994                 max_low_pfn = max_pfn;
 995         }
 996 #endif
 997
 998         /*
 999          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1000          */
1001
1002 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1003         if (init_ohci1394_dma_early)
1004                 init_ohci1394_dma_on_all_controllers();
1005 #endif
1006
1007         reserve_initrd();
1008
1009 #ifdef CONFIG_X86_64
1010         vsmp_init();
1011 #endif
1012
1013         io_delay_init();
1014
1015         /*
1016          * Parse the ACPI tables for possible boot-time SMP configuration.
1017          */
1018         acpi_boot_table_init();
1019
1020         early_acpi_boot_init();
1021
1022 #ifdef CONFIG_ACPI_NUMA
1023         /*
1024          * Parse SRAT to discover nodes.
1025          */
1026         acpi_numa_init();
1027 #endif
1028
1029         initmem_init(0, max_pfn);
1030
1031 #ifdef CONFIG_ACPI_SLEEP
1032         /*
1033          * Reserve low memory region for sleep support.
1034          */
1035         acpi_reserve_bootmem();
1036 #endif
1037 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1038         /*
1039          * Find and reserve possible boot-time SMP configuration:
1040          */
1041         find_smp_config();
1042 #endif
1043         reserve_crashkernel();
1044
1045 #ifdef CONFIG_X86_64
1046         /*
1047          * dma32_reserve_bootmem() allocates bootmem which may conflict
1048          * with the crashkernel command line, so do that after
1049          * reserve_crashkernel()
1050          */
1051         dma32_reserve_bootmem();
1052 #endif
1053
1054         reserve_ibft_region();
1055
1056 #ifdef CONFIG_KVM_CLOCK
1057         kvmclock_init();
1058 #endif
1059
1060         paravirt_pagetable_setup_start(swapper_pg_dir);
1061         paging_init();
1062         paravirt_pagetable_setup_done(swapper_pg_dir);
1063         paravirt_post_allocator_init();
1064
1065 #ifdef CONFIG_X86_64
1066         map_vsyscall();
1067 #endif
1068
1069 #ifdef CONFIG_X86_GENERICARCH
1070         generic_apic_probe();
1071 #endif
1072
1073         early_quirks();
1074
1075         /*
1076          * Read APIC and some other early information from ACPI tables.
1077          */
1078         acpi_boot_init();
1079
1080 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
1081         /*
1082          * get boot-time SMP configuration:
1083          */
1084         if (smp_found_config)
1085                 get_smp_config();
1086 #endif
1087
1088         prefill_possible_map();
1089
1090 #ifdef CONFIG_X86_64
1091         init_cpu_to_node();
1092 #endif
1093
1094         init_apic_mappings();
1095         ioapic_init_mappings();
1096
1097         /* need to wait for io_apic is mapped */
1098         nr_irqs = probe_nr_irqs();
1099
1100         kvm_guest_init();
1101
1102         e820_reserve_resources();
1103         e820_mark_nosave_regions(max_low_pfn);
1104
1105 #ifdef CONFIG_X86_32
1106         request_resource(&iomem_resource, &video_ram_resource);
1107 #endif
1108         reserve_standard_io_resources();
1109
1110         e820_setup_gap();
1111
1112 #ifdef CONFIG_VT
1113 #if defined(CONFIG_VGA_CONSOLE)
1114         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1115                 conswitchp = &vga_con;
1116 #elif defined(CONFIG_DUMMY_CONSOLE)
1117         conswitchp = &dummy_con;
1118 #endif
1119 #endif
1120 }
1121
1122