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 <asm/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/cpu.h>
  93 #include <asm/desc.h>
  94 #include <asm/dma.h>
  95 #include <asm/iommu.h>
  96 #include <asm/gart.h>
  97 #include <asm/mmu_context.h>
  98 #include <asm/proto.h>
  99
 100 #include <asm/paravirt.h>
 101 #include <asm/hypervisor.h>
 102
 103 #include <asm/percpu.h>
 104 #include <asm/topology.h>
 105 #include <asm/apicdef.h>
 106 #ifdef CONFIG_X86_64
 107 #include <asm/numa_64.h>
 108 #endif
 109
 110 #ifndef ARCH_SETUP
 111 #define ARCH_SETUP
 112 #endif
 113
 114 unsigned int boot_cpu_id __read_mostly;
 115
 116 #ifdef CONFIG_X86_64
 117 int default_cpu_present_to_apicid(int mps_cpu)
 118 {
 119         return __default_cpu_present_to_apicid(mps_cpu);
 120 }
 121
 122 int default_check_phys_apicid_present(int boot_cpu_physical_apicid)
 123 {
 124         return __default_check_phys_apicid_present(boot_cpu_physical_apicid);
 125 }
 126 #endif
 127
 128 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 129 struct boot_params __initdata boot_params;
 130 #else
 131 struct boot_params boot_params;
 132 #endif
 133
 134 /*
 135  * Machine setup..
 136  */
 137 static struct resource data_resource = {
 138         .name   = "Kernel data",
 139         .start  = 0,
 140         .end    = 0,
 141         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 142 };
 143
 144 static struct resource code_resource = {
 145         .name   = "Kernel code",
 146         .start  = 0,
 147         .end    = 0,
 148         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 149 };
 150
 151 static struct resource bss_resource = {
 152         .name   = "Kernel bss",
 153         .start  = 0,
 154         .end    = 0,
 155         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 156 };
 157
 158
 159 #ifdef CONFIG_X86_32
 160 /* This value is set up by the early boot code to point to the value
 161    immediately after the boot time page tables.  It contains a *physical*
 162    address, and must not be in the .bss segment! */
 163 unsigned long init_pg_tables_start __initdata = ~0UL;
 164 unsigned long init_pg_tables_end __initdata = ~0UL;
 165
 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 EXPORT_SYMBOL(boot_cpu_data);
 206 #endif
 207
 208
 209 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 210 unsigned long mmu_cr4_features;
 211 #else
 212 unsigned long mmu_cr4_features = X86_CR4_PAE;
 213 #endif
 214
 215 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 216 int bootloader_type;
 217
 218 /*
 219  * Early DMI memory
 220  */
 221 int dmi_alloc_index;
 222 char dmi_alloc_data[DMI_MAX_DATA];
 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 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 copy_edd(void)
 265 {
 266 }
 267 #endif
 268
 269 #ifdef CONFIG_BLK_DEV_INITRD
 270
 271 #ifdef CONFIG_X86_32
 272
 273 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 274 static void __init relocate_initrd(void)
 275 {
 276
 277         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 278         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 279         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 280         u64 ramdisk_here;
 281         unsigned long slop, clen, mapaddr;
 282         char *p, *q;
 283
 284         /* We need to move the initrd down into lowmem */
 285         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 286                                          PAGE_SIZE);
 287
 288         if (ramdisk_here == -1ULL)
 289                 panic("Cannot find place for new RAMDISK of size %lld\n",
 290                          ramdisk_size);
 291
 292         /* Note: this includes all the lowmem currently occupied by
 293            the initrd, we rely on that fact to keep the data intact. */
 294         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 295                          "NEW RAMDISK");
 296         initrd_start = ramdisk_here + PAGE_OFFSET;
 297         initrd_end   = initrd_start + ramdisk_size;
 298         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 299                          ramdisk_here, ramdisk_here + ramdisk_size);
 300
 301         q = (char *)initrd_start;
 302
 303         /* Copy any lowmem portion of the initrd */
 304         if (ramdisk_image < end_of_lowmem) {
 305                 clen = end_of_lowmem - ramdisk_image;
 306                 p = (char *)__va(ramdisk_image);
 307                 memcpy(q, p, clen);
 308                 q += clen;
 309                 ramdisk_image += clen;
 310                 ramdisk_size  -= clen;
 311         }
 312
 313         /* Copy the highmem portion of the initrd */
 314         while (ramdisk_size) {
 315                 slop = ramdisk_image & ~PAGE_MASK;
 316                 clen = ramdisk_size;
 317                 if (clen > MAX_MAP_CHUNK-slop)
 318                         clen = MAX_MAP_CHUNK-slop;
 319                 mapaddr = ramdisk_image & PAGE_MASK;
 320                 p = early_memremap(mapaddr, clen+slop);
 321                 memcpy(q, p+slop, clen);
 322                 early_iounmap(p, clen+slop);
 323                 q += clen;
 324                 ramdisk_image += clen;
 325                 ramdisk_size  -= clen;
 326         }
 327         /* high pages is not converted by early_res_to_bootmem */
 328         ramdisk_image = boot_params.hdr.ramdisk_image;
 329         ramdisk_size  = boot_params.hdr.ramdisk_size;
 330         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 331                 " %08llx - %08llx\n",
 332                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 333                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 334 }
 335 #endif
 336
 337 static void __init reserve_initrd(void)
 338 {
 339         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 340         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 341         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 342         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 343
 344         if (!boot_params.hdr.type_of_loader ||
 345             !ramdisk_image || !ramdisk_size)
 346                 return;         /* No initrd provided by bootloader */
 347
 348         initrd_start = 0;
 349
 350         if (ramdisk_size >= (end_of_lowmem>>1)) {
 351                 free_early(ramdisk_image, ramdisk_end);
 352                 printk(KERN_ERR "initrd too large to handle, "
 353                        "disabling initrd\n");
 354                 return;
 355         }
 356
 357         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 358                         ramdisk_end);
 359
 360
 361         if (ramdisk_end <= end_of_lowmem) {
 362                 /* All in lowmem, easy case */
 363                 /*
 364                  * don't need to reserve again, already reserved early
 365                  * in i386_start_kernel
 366                  */
 367                 initrd_start = ramdisk_image + PAGE_OFFSET;
 368                 initrd_end = initrd_start + ramdisk_size;
 369                 return;
 370         }
 371
 372 #ifdef CONFIG_X86_32
 373         relocate_initrd();
 374 #else
 375         printk(KERN_ERR "initrd extends beyond end of memory "
 376                "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
 377                ramdisk_end, end_of_lowmem);
 378         initrd_start = 0;
 379 #endif
 380         free_early(ramdisk_image, ramdisk_end);
 381 }
 382 #else
 383 static void __init reserve_initrd(void)
 384 {
 385 }
 386 #endif /* CONFIG_BLK_DEV_INITRD */
 387
 388 static void __init parse_setup_data(void)
 389 {
 390         struct setup_data *data;
 391         u64 pa_data;
 392
 393         if (boot_params.hdr.version < 0x0209)
 394                 return;
 395         pa_data = boot_params.hdr.setup_data;
 396         while (pa_data) {
 397                 data = early_memremap(pa_data, PAGE_SIZE);
 398                 switch (data->type) {
 399                 case SETUP_E820_EXT:
 400                         parse_e820_ext(data, pa_data);
 401                         break;
 402                 default:
 403                         break;
 404                 }
 405                 pa_data = data->next;
 406                 early_iounmap(data, PAGE_SIZE);
 407         }
 408 }
 409
 410 static void __init e820_reserve_setup_data(void)
 411 {
 412         struct setup_data *data;
 413         u64 pa_data;
 414         int found = 0;
 415
 416         if (boot_params.hdr.version < 0x0209)
 417                 return;
 418         pa_data = boot_params.hdr.setup_data;
 419         while (pa_data) {
 420                 data = early_memremap(pa_data, sizeof(*data));
 421                 e820_update_range(pa_data, sizeof(*data)+data->len,
 422                          E820_RAM, E820_RESERVED_KERN);
 423                 found = 1;
 424                 pa_data = data->next;
 425                 early_iounmap(data, sizeof(*data));
 426         }
 427         if (!found)
 428                 return;
 429
 430         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 431         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 432         printk(KERN_INFO "extended physical RAM map:\n");
 433         e820_print_map("reserve setup_data");
 434 }
 435
 436 static void __init reserve_early_setup_data(void)
 437 {
 438         struct setup_data *data;
 439         u64 pa_data;
 440         char buf[32];
 441
 442         if (boot_params.hdr.version < 0x0209)
 443                 return;
 444         pa_data = boot_params.hdr.setup_data;
 445         while (pa_data) {
 446                 data = early_memremap(pa_data, sizeof(*data));
 447                 sprintf(buf, "setup data %x", data->type);
 448                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 449                 pa_data = data->next;
 450                 early_iounmap(data, sizeof(*data));
 451         }
 452 }
 453
 454 /*
 455  * --------- Crashkernel reservation ------------------------------
 456  */
 457
 458 #ifdef CONFIG_KEXEC
 459
 460 /**
 461  * Reserve @size bytes of crashkernel memory at any suitable offset.
 462  *
 463  * @size: Size of the crashkernel memory to reserve.
 464  * Returns the base address on success, and -1ULL on failure.
 465  */
 466 static
 467 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 468 {
 469         const unsigned long long alignment = 16<<20;    /* 16M */
 470         unsigned long long start = 0LL;
 471
 472         while (1) {
 473                 int ret;
 474
 475                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 476                 if (start == -1ULL)
 477                         return start;
 478
 479                 /* try to reserve it */
 480                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 481                 if (ret >= 0)
 482                         return start;
 483
 484                 start += alignment;
 485         }
 486 }
 487
 488 static inline unsigned long long get_total_mem(void)
 489 {
 490         unsigned long long total;
 491
 492         total = max_low_pfn - min_low_pfn;
 493 #ifdef CONFIG_HIGHMEM
 494         total += highend_pfn - highstart_pfn;
 495 #endif
 496
 497         return total << PAGE_SHIFT;
 498 }
 499
 500 static void __init reserve_crashkernel(void)
 501 {
 502         unsigned long long total_mem;
 503         unsigned long long crash_size, crash_base;
 504         int ret;
 505
 506         total_mem = get_total_mem();
 507
 508         ret = parse_crashkernel(boot_command_line, total_mem,
 509                         &crash_size, &crash_base);
 510         if (ret != 0 || crash_size <= 0)
 511                 return;
 512
 513         /* 0 means: find the address automatically */
 514         if (crash_base <= 0) {
 515                 crash_base = find_and_reserve_crashkernel(crash_size);
 516                 if (crash_base == -1ULL) {
 517                         pr_info("crashkernel reservation failed. "
 518                                 "No suitable area found.\n");
 519                         return;
 520                 }
 521         } else {
 522                 ret = reserve_bootmem_generic(crash_base, crash_size,
 523                                         BOOTMEM_EXCLUSIVE);
 524                 if (ret < 0) {
 525                         pr_info("crashkernel reservation failed - "
 526                                 "memory is in use\n");
 527                         return;
 528                 }
 529         }
 530
 531         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 532                         "for crashkernel (System RAM: %ldMB)\n",
 533                         (unsigned long)(crash_size >> 20),
 534                         (unsigned long)(crash_base >> 20),
 535                         (unsigned long)(total_mem >> 20));
 536
 537         crashk_res.start = crash_base;
 538         crashk_res.end   = crash_base + crash_size - 1;
 539         insert_resource(&iomem_resource, &crashk_res);
 540 }
 541 #else
 542 static void __init reserve_crashkernel(void)
 543 {
 544 }
 545 #endif
 546
 547 static struct resource standard_io_resources[] = {
 548         { .name = "dma1", .start = 0x00, .end = 0x1f,
 549                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 550         { .name = "pic1", .start = 0x20, .end = 0x21,
 551                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 552         { .name = "timer0", .start = 0x40, .end = 0x43,
 553                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 554         { .name = "timer1", .start = 0x50, .end = 0x53,
 555                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 556         { .name = "keyboard", .start = 0x60, .end = 0x60,
 557                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 558         { .name = "keyboard", .start = 0x64, .end = 0x64,
 559                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 560         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 561                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 562         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 563                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 564         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 565                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 566         { .name = "fpu", .start = 0xf0, .end = 0xff,
 567                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 568 };
 569
 570 static void __init reserve_standard_io_resources(void)
 571 {
 572         int i;
 573
 574         /* request I/O space for devices used on all i[345]86 PCs */
 575         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 576                 request_resource(&ioport_resource, &standard_io_resources[i]);
 577
 578 }
 579
 580 /*
 581  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 582  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 583  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 584  */
 585
 586 #ifdef CONFIG_CRASH_DUMP
 587 /* elfcorehdr= specifies the location of elf core header
 588  * stored by the crashed kernel. This option will be passed
 589  * by kexec loader to the capture kernel.
 590  */
 591 static int __init setup_elfcorehdr(char *arg)
 592 {
 593         char *end;
 594         if (!arg)
 595                 return -EINVAL;
 596         elfcorehdr_addr = memparse(arg, &end);
 597         return end > arg ? 0 : -EINVAL;
 598 }
 599 early_param("elfcorehdr", setup_elfcorehdr);
 600 #endif
 601
 602 static int __init default_update_apic(void)
 603 {
 604 #ifdef CONFIG_SMP
 605         if (!apic->wakeup_cpu)
 606                 apic->wakeup_cpu = wakeup_secondary_cpu_via_init;
 607 #endif
 608
 609         return 0;
 610 }
 611
 612 static struct x86_quirks default_x86_quirks __initdata = {
 613         .update_apic         = default_update_apic,
 614 };
 615
 616 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 617
 618 #ifdef CONFIG_X86_RESERVE_LOW_64K
 619 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 620 {
 621         printk(KERN_NOTICE
 622                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 623                 d->ident);
 624
 625         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 626         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 627
 628         return 0;
 629 }
 630 #endif
 631
 632 /* List of systems that have known low memory corruption BIOS problems */
 633 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 634 #ifdef CONFIG_X86_RESERVE_LOW_64K
 635         {
 636                 .callback = dmi_low_memory_corruption,
 637                 .ident = "AMI BIOS",
 638                 .matches = {
 639                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 640                 },
 641         },
 642         {
 643                 .callback = dmi_low_memory_corruption,
 644                 .ident = "Phoenix BIOS",
 645                 .matches = {
 646                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 647                 },
 648         },
 649 #endif
 650         {}
 651 };
 652
 653 /*
 654  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 655  * passed the efi memmap, systab, etc., so we should use these data structures
 656  * for initialization.  Note, the efi init code path is determined by the
 657  * global efi_enabled. This allows the same kernel image to be used on existing
 658  * systems (with a traditional BIOS) as well as on EFI systems.
 659  */
 660 /*
 661  * setup_arch - architecture-specific boot-time initializations
 662  *
 663  * Note: On x86_64, fixmaps are ready for use even before this is called.
 664  */
 665
 666 void __init setup_arch(char **cmdline_p)
 667 {
 668 #ifdef CONFIG_X86_32
 669         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 670         visws_early_detect();
 671         pre_setup_arch_hook();
 672 #else
 673         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 674 #endif
 675
 676         /* VMI may relocate the fixmap; do this before touching ioremap area */
 677         vmi_init();
 678
 679         early_cpu_init();
 680         early_ioremap_init();
 681
 682         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 683         screen_info = boot_params.screen_info;
 684         edid_info = boot_params.edid_info;
 685 #ifdef CONFIG_X86_32
 686         apm_info.bios = boot_params.apm_bios_info;
 687         ist_info = boot_params.ist_info;
 688         if (boot_params.sys_desc_table.length != 0) {
 689                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 690                 machine_id = boot_params.sys_desc_table.table[0];
 691                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 692                 BIOS_revision = boot_params.sys_desc_table.table[2];
 693         }
 694 #endif
 695         saved_video_mode = boot_params.hdr.vid_mode;
 696         bootloader_type = boot_params.hdr.type_of_loader;
 697
 698 #ifdef CONFIG_BLK_DEV_RAM
 699         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 700         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 701         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 702 #endif
 703 #ifdef CONFIG_EFI
 704         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 705 #ifdef CONFIG_X86_32
 706                      "EL32",
 707 #else
 708                      "EL64",
 709 #endif
 710          4)) {
 711                 efi_enabled = 1;
 712                 efi_reserve_early();
 713         }
 714 #endif
 715
 716         ARCH_SETUP
 717
 718         setup_memory_map();
 719         parse_setup_data();
 720         /* update the e820_saved too */
 721         e820_reserve_setup_data();
 722
 723         copy_edd();
 724
 725         if (!boot_params.hdr.root_flags)
 726                 root_mountflags &= ~MS_RDONLY;
 727         init_mm.start_code = (unsigned long) _text;
 728         init_mm.end_code = (unsigned long) _etext;
 729         init_mm.end_data = (unsigned long) _edata;
 730 #ifdef CONFIG_X86_32
 731         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 732 #else
 733         init_mm.brk = (unsigned long) &_end;
 734 #endif
 735
 736         code_resource.start = virt_to_phys(_text);
 737         code_resource.end = virt_to_phys(_etext)-1;
 738         data_resource.start = virt_to_phys(_etext);
 739         data_resource.end = virt_to_phys(_edata)-1;
 740         bss_resource.start = virt_to_phys(&__bss_start);
 741         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 742
 743 #ifdef CONFIG_CMDLINE_BOOL
 744 #ifdef CONFIG_CMDLINE_OVERRIDE
 745         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 746 #else
 747         if (builtin_cmdline[0]) {
 748                 /* append boot loader cmdline to builtin */
 749                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 750                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 751                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 752         }
 753 #endif
 754 #endif
 755
 756         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 757         *cmdline_p = command_line;
 758
 759         parse_early_param();
 760
 761 #ifdef CONFIG_X86_64
 762         check_efer();
 763 #endif
 764
 765         /* Must be before kernel pagetables are setup */
 766         vmi_activate();
 767
 768         /* after early param, so could get panic from serial */
 769         reserve_early_setup_data();
 770
 771         if (acpi_mps_check()) {
 772 #ifdef CONFIG_X86_LOCAL_APIC
 773                 disable_apic = 1;
 774 #endif
 775                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 776         }
 777
 778 #ifdef CONFIG_PCI
 779         if (pci_early_dump_regs)
 780                 early_dump_pci_devices();
 781 #endif
 782
 783         finish_e820_parsing();
 784
 785         dmi_scan_machine();
 786
 787         dmi_check_system(bad_bios_dmi_table);
 788
 789         /*
 790          * VMware detection requires dmi to be available, so this
 791          * needs to be done after dmi_scan_machine, for the BP.
 792          */
 793         init_hypervisor(&boot_cpu_data);
 794
 795 #ifdef CONFIG_X86_32
 796         probe_roms();
 797 #endif
 798
 799         /* after parse_early_param, so could debug it */
 800         insert_resource(&iomem_resource, &code_resource);
 801         insert_resource(&iomem_resource, &data_resource);
 802         insert_resource(&iomem_resource, &bss_resource);
 803
 804         if (efi_enabled)
 805                 efi_init();
 806
 807 #ifdef CONFIG_X86_32
 808         if (ppro_with_ram_bug()) {
 809                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 810                                   E820_RESERVED);
 811                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 812                 printk(KERN_INFO "fixed physical RAM map:\n");
 813                 e820_print_map("bad_ppro");
 814         }
 815 #else
 816         early_gart_iommu_check();
 817 #endif
 818
 819         /*
 820          * partially used pages are not usable - thus
 821          * we are rounding upwards:
 822          */
 823         max_pfn = e820_end_of_ram_pfn();
 824
 825         /* preallocate 4k for mptable mpc */
 826         early_reserve_e820_mpc_new();
 827         /* update e820 for memory not covered by WB MTRRs */
 828         mtrr_bp_init();
 829         if (mtrr_trim_uncached_memory(max_pfn))
 830                 max_pfn = e820_end_of_ram_pfn();
 831
 832 #ifdef CONFIG_X86_32
 833         /* max_low_pfn get updated here */
 834         find_low_pfn_range();
 835 #else
 836         num_physpages = max_pfn;
 837
 838         check_x2apic();
 839
 840         /* How many end-of-memory variables you have, grandma! */
 841         /* need this before calling reserve_initrd */
 842         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 843                 max_low_pfn = e820_end_of_low_ram_pfn();
 844         else
 845                 max_low_pfn = max_pfn;
 846
 847         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 848 #endif
 849
 850 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 851         setup_bios_corruption_check();
 852 #endif
 853
 854         /* max_pfn_mapped is updated here */
 855         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 856         max_pfn_mapped = max_low_pfn_mapped;
 857
 858 #ifdef CONFIG_X86_64
 859         if (max_pfn > max_low_pfn) {
 860                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 861                                                      max_pfn<<PAGE_SHIFT);
 862                 /* can we preseve max_low_pfn ?*/
 863                 max_low_pfn = max_pfn;
 864         }
 865 #endif
 866
 867         /*
 868          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 869          */
 870
 871 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 872         if (init_ohci1394_dma_early)
 873                 init_ohci1394_dma_on_all_controllers();
 874 #endif
 875
 876         reserve_initrd();
 877
 878 #ifdef CONFIG_X86_64
 879         vsmp_init();
 880 #endif
 881
 882         io_delay_init();
 883
 884         /*
 885          * Parse the ACPI tables for possible boot-time SMP configuration.
 886          */
 887         acpi_boot_table_init();
 888
 889         early_acpi_boot_init();
 890
 891 #ifdef CONFIG_ACPI_NUMA
 892         /*
 893          * Parse SRAT to discover nodes.
 894          */
 895         acpi_numa_init();
 896 #endif
 897
 898         initmem_init(0, max_pfn);
 899
 900 #ifdef CONFIG_ACPI_SLEEP
 901         /*
 902          * Reserve low memory region for sleep support.
 903          */
 904         acpi_reserve_bootmem();
 905 #endif
 906         /*
 907          * Find and reserve possible boot-time SMP configuration:
 908          */
 909         find_smp_config();
 910
 911         reserve_crashkernel();
 912
 913 #ifdef CONFIG_X86_64
 914         /*
 915          * dma32_reserve_bootmem() allocates bootmem which may conflict
 916          * with the crashkernel command line, so do that after
 917          * reserve_crashkernel()
 918          */
 919         dma32_reserve_bootmem();
 920 #endif
 921
 922         reserve_ibft_region();
 923
 924 #ifdef CONFIG_KVM_CLOCK
 925         kvmclock_init();
 926 #endif
 927
 928         paravirt_pagetable_setup_start(swapper_pg_dir);
 929         paging_init();
 930         paravirt_pagetable_setup_done(swapper_pg_dir);
 931         paravirt_post_allocator_init();
 932
 933 #ifdef CONFIG_X86_64
 934         map_vsyscall();
 935 #endif
 936
 937         generic_apic_probe();
 938
 939         early_quirks();
 940
 941         /*
 942          * Read APIC and some other early information from ACPI tables.
 943          */
 944         acpi_boot_init();
 945
 946 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 947         /*
 948          * get boot-time SMP configuration:
 949          */
 950         if (smp_found_config)
 951                 get_smp_config();
 952 #endif
 953
 954         prefill_possible_map();
 955
 956 #ifdef CONFIG_X86_64
 957         init_cpu_to_node();
 958 #endif
 959
 960         init_apic_mappings();
 961         ioapic_init_mappings();
 962
 963         /* need to wait for io_apic is mapped */
 964         probe_nr_irqs_gsi();
 965
 966         kvm_guest_init();
 967
 968         e820_reserve_resources();
 969         e820_mark_nosave_regions(max_low_pfn);
 970
 971 #ifdef CONFIG_X86_32
 972         request_resource(&iomem_resource, &video_ram_resource);
 973 #endif
 974         reserve_standard_io_resources();
 975
 976         e820_setup_gap();
 977
 978 #ifdef CONFIG_VT
 979 #if defined(CONFIG_VGA_CONSOLE)
 980         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 981                 conswitchp = &vga_con;
 982 #elif defined(CONFIG_DUMMY_CONSOLE)
 983         conswitchp = &dummy_con;
 984 #endif
 985 #endif
 986 }
 987
 988 #ifdef CONFIG_X86_32
 989
 990 /**
 991  * pre_intr_init_hook - initialisation prior to setting up interrupt vectors
 992  *
 993  * Description:
 994  *      Perform any necessary interrupt initialisation prior to setting up
 995  *      the "ordinary" interrupt call gates.  For legacy reasons, the ISA
 996  *      interrupts should be initialised here if the machine emulates a PC
 997  *      in any way.
 998  **/
 999 void __init pre_intr_init_hook(void)
1000 {
1001         if (x86_quirks->arch_pre_intr_init) {
1002                 if (x86_quirks->arch_pre_intr_init())
1003                         return;
1004         }
1005         init_ISA_irqs();
1006 }
1007
1008 /**
1009  * intr_init_hook - post gate setup interrupt initialisation
1010  *
1011  * Description:
1012  *      Fill in any interrupts that may have been left out by the general
1013  *      init_IRQ() routine.  interrupts having to do with the machine rather
1014  *      than the devices on the I/O bus (like APIC interrupts in intel MP
1015  *      systems) are started here.
1016  **/
1017 void __init intr_init_hook(void)
1018 {
1019         if (x86_quirks->arch_intr_init) {
1020                 if (x86_quirks->arch_intr_init())
1021                         return;
1022         }
1023 }
1024
1025 /**
1026  * pre_setup_arch_hook - hook called prior to any setup_arch() execution
1027  *
1028  * Description:
1029  *      generally used to activate any machine specific identification
1030  *      routines that may be needed before setup_arch() runs.  On Voyager
1031  *      this is used to get the board revision and type.
1032  **/
1033 void __init pre_setup_arch_hook(void)
1034 {
1035 }
1036
1037 /**
1038  * trap_init_hook - initialise system specific traps
1039  *
1040  * Description:
1041  *      Called as the final act of trap_init().  Used in VISWS to initialise
1042  *      the various board specific APIC traps.
1043  **/
1044 void __init trap_init_hook(void)
1045 {
1046         if (x86_quirks->arch_trap_init) {
1047                 if (x86_quirks->arch_trap_init())
1048                         return;
1049         }
1050 }
1051
1052 static struct irqaction irq0  = {
1053         .handler = timer_interrupt,
1054         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL,
1055         .mask = CPU_MASK_NONE,
1056         .name = "timer"
1057 };
1058
1059 /**
1060  * pre_time_init_hook - do any specific initialisations before.
1061  *
1062  **/
1063 void __init pre_time_init_hook(void)
1064 {
1065         if (x86_quirks->arch_pre_time_init)
1066                 x86_quirks->arch_pre_time_init();
1067 }
1068
1069 /**
1070  * time_init_hook - do any specific initialisations for the system timer.
1071  *
1072  * Description:
1073  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1074  *      in irq_vectors.h:TIMER_IRQ
1075  **/
1076 void __init time_init_hook(void)
1077 {
1078         if (x86_quirks->arch_time_init) {
1079                 /*
1080                  * A nonzero return code does not mean failure, it means
1081                  * that the architecture quirk does not want any
1082                  * generic (timer) setup to be performed after this:
1083                  */
1084                 if (x86_quirks->arch_time_init())
1085                         return;
1086         }
1087
1088         irq0.mask = cpumask_of_cpu(0);
1089         setup_irq(0, &irq0);
1090 }
1091
1092 #ifdef CONFIG_MCA
1093 /**
1094  * mca_nmi_hook - hook into MCA specific NMI chain
1095  *
1096  * Description:
1097  *      The MCA (Microchannel Architecture) has an NMI chain for NMI sources
1098  *      along the MCA bus.  Use this to hook into that chain if you will need
1099  *      it.
1100  **/
1101 void mca_nmi_hook(void)
1102 {
1103         /*
1104          * If I recall correctly, there's a whole bunch of other things that
1105          * we can do to check for NMI problems, but that's all I know about
1106          * at the moment.
1107          */
1108         pr_warning("NMI generated from unknown source!\n");
1109 }
1110 #endif /* CONFIG_MCA */
1111
1112 #endif /* CONFIG_X86_32 */