kernel/2.6.29.6-aldebaran-rt/arch/x86/kernel/setup.c

   1 /*
   2  *  Copyright (C) 1995  Linus Torvalds
   3  *
   4  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   5  *
   6  *  Memory region support
   7  *      David Parsons <orc@pell.chi.il.us>, July-August 1999
   8  *
   9  *  Added E820 sanitization routine (removes overlapping memory regions);
  10  *  Brian Moyle <bmoyle@mvista.com>, February 2001
  11  *
  12  * Moved CPU detection code to cpu/${cpu}.c
  13  *    Patrick Mochel <mochel@osdl.org>, March 2002
  14  *
  15  *  Provisions for empty E820 memory regions (reported by certain BIOSes).
  16  *  Alex Achenbach <xela@slit.de>, December 2002.
  17  *
  18  */
  19
  20 /*
  21  * This file handles the architecture-dependent parts of initialization
  22  */
  23
  24 #include <linux/sched.h>
  25 #include <linux/mm.h>
  26 #include <linux/mmzone.h>
  27 #include <linux/screen_info.h>
  28 #include <linux/ioport.h>
  29 #include <linux/acpi.h>
  30 #include <linux/apm_bios.h>
  31 #include <linux/initrd.h>
  32 #include <linux/bootmem.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/console.h>
  35 #include <linux/mca.h>
  36 #include <linux/root_dev.h>
  37 #include <linux/highmem.h>
  38 #include <linux/module.h>
  39 #include <linux/efi.h>
  40 #include <linux/init.h>
  41 #include <linux/edd.h>
  42 #include <linux/iscsi_ibft.h>
  43 #include <linux/nodemask.h>
  44 #include <linux/kexec.h>
  45 #include <linux/dmi.h>
  46 #include <linux/pfn.h>
  47 #include <linux/pci.h>
  48 #include <asm/pci-direct.h>
  49 #include <linux/init_ohci1394_dma.h>
  50 #include <linux/kvm_para.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/kernel.h>
  54 #include <linux/stddef.h>
  55 #include <linux/unistd.h>
  56 #include <linux/ptrace.h>
  57 #include <linux/slab.h>
  58 #include <linux/user.h>
  59 #include <linux/delay.h>
  60
  61 #include <linux/kallsyms.h>
  62 #include <linux/cpufreq.h>
  63 #include <linux/dma-mapping.h>
  64 #include <linux/ctype.h>
  65 #include <linux/uaccess.h>
  66
  67 #include <linux/percpu.h>
  68 #include <linux/crash_dump.h>
  69
  70 #include <video/edid.h>
  71
  72 #include <asm/mtrr.h>
  73 #include <asm/apic.h>
  74 #include <asm/e820.h>
  75 #include <asm/mpspec.h>
  76 #include <asm/setup.h>
  77 #include <asm/efi.h>
  78 #include <asm/timer.h>
  79 #include <asm/i8259.h>
  80 #include <asm/sections.h>
  81 #include <asm/dmi.h>
  82 #include <asm/io_apic.h>
  83 #include <asm/ist.h>
  84 #include <asm/vmi.h>
  85 #include <asm/setup_arch.h>
  86 #include <asm/bios_ebda.h>
  87 #include <asm/cacheflush.h>
  88 #include <asm/processor.h>
  89 #include <asm/bugs.h>
  90
  91 #include <asm/system.h>
  92 #include <asm/vsyscall.h>
  93 #include <asm/cpu.h>
  94 #include <asm/desc.h>
  95 #include <asm/dma.h>
  96 #include <asm/iommu.h>
  97 #include <asm/gart.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/proto.h>
 100
 101 #include <asm/paravirt.h>
 102 #include <asm/hypervisor.h>
 103
 104 #include <asm/percpu.h>
 105 #include <asm/topology.h>
 106 #include <asm/apicdef.h>
 107 #ifdef CONFIG_X86_64
 108 #include <asm/numa_64.h>
 109 #endif
 110
 111 #ifndef ARCH_SETUP
 112 #define ARCH_SETUP
 113 #endif
 114
 115 RESERVE_BRK(dmi_alloc, 65536);
 116
 117 unsigned int boot_cpu_id __read_mostly;
 118
 119 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 120 unsigned long _brk_end = (unsigned long)__brk_base;
 121
 122 #ifdef CONFIG_X86_64
 123 int default_cpu_present_to_apicid(int mps_cpu)
 124 {
 125         return __default_cpu_present_to_apicid(mps_cpu);
 126 }
 127
 128 int default_check_phys_apicid_present(int boot_cpu_physical_apicid)
 129 {
 130         return __default_check_phys_apicid_present(boot_cpu_physical_apicid);
 131 }
 132 #endif
 133
 134 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 135 struct boot_params __initdata boot_params;
 136 #else
 137 struct boot_params boot_params;
 138 #endif
 139
 140 /*
 141  * Machine setup..
 142  */
 143 static struct resource data_resource = {
 144         .name   = "Kernel data",
 145         .start  = 0,
 146         .end    = 0,
 147         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 148 };
 149
 150 static struct resource code_resource = {
 151         .name   = "Kernel code",
 152         .start  = 0,
 153         .end    = 0,
 154         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 155 };
 156
 157 static struct resource bss_resource = {
 158         .name   = "Kernel bss",
 159         .start  = 0,
 160         .end    = 0,
 161         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 162 };
 163
 164
 165 #ifdef CONFIG_X86_32
 166 static struct resource video_ram_resource = {
 167         .name   = "Video RAM area",
 168         .start  = 0xa0000,
 169         .end    = 0xbffff,
 170         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 171 };
 172
 173 /* cpu data as detected by the assembly code in head.S */
 174 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 175 /* common cpu data for all cpus */
 176 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 177 EXPORT_SYMBOL(boot_cpu_data);
 178 static void set_mca_bus(int x)
 179 {
 180 #ifdef CONFIG_MCA
 181         MCA_bus = x;
 182 #endif
 183 }
 184
 185 unsigned int def_to_bigsmp;
 186
 187 /* for MCA, but anyone else can use it if they want */
 188 unsigned int machine_id;
 189 unsigned int machine_submodel_id;
 190 unsigned int BIOS_revision;
 191
 192 struct apm_info apm_info;
 193 EXPORT_SYMBOL(apm_info);
 194
 195 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 196         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 197 struct ist_info ist_info;
 198 EXPORT_SYMBOL(ist_info);
 199 #else
 200 struct ist_info ist_info;
 201 #endif
 202
 203 #else
 204 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 205         .x86_phys_bits = MAX_PHYSMEM_BITS,
 206 };
 207 EXPORT_SYMBOL(boot_cpu_data);
 208 #endif
 209
 210
 211 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 212 unsigned long mmu_cr4_features;
 213 #else
 214 unsigned long mmu_cr4_features = X86_CR4_PAE;
 215 #endif
 216
 217 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 218 int bootloader_type;
 219
 220 /*
 221  * Setup options
 222  */
 223 struct screen_info screen_info;
 224 EXPORT_SYMBOL(screen_info);
 225 struct edid_info edid_info;
 226 EXPORT_SYMBOL_GPL(edid_info);
 227
 228 extern int root_mountflags;
 229
 230 unsigned long saved_video_mode;
 231
 232 #define RAMDISK_IMAGE_START_MASK        0x07FF
 233 #define RAMDISK_PROMPT_FLAG             0x8000
 234 #define RAMDISK_LOAD_FLAG               0x4000
 235
 236 static char __initdata command_line[COMMAND_LINE_SIZE];
 237 #ifdef CONFIG_CMDLINE_BOOL
 238 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 239 #endif
 240
 241 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 242 struct edd edd;
 243 #ifdef CONFIG_EDD_MODULE
 244 EXPORT_SYMBOL(edd);
 245 #endif
 246 /**
 247  * copy_edd() - Copy the BIOS EDD information
 248  *              from boot_params into a safe place.
 249  *
 250  */
 251 static inline void copy_edd(void)
 252 {
 253      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 254             sizeof(edd.mbr_signature));
 255      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 256      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 257      edd.edd_info_nr = boot_params.eddbuf_entries;
 258 }
 259 #else
 260 static inline void copy_edd(void)
 261 {
 262 }
 263 #endif
 264
 265 void * __init extend_brk(size_t size, size_t align)
 266 {
 267         size_t mask = align - 1;
 268         void *ret;
 269
 270         BUG_ON(_brk_start == 0);
 271         BUG_ON(align & mask);
 272
 273         _brk_end = (_brk_end + mask) & ~mask;
 274         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 275
 276         ret = (void *)_brk_end;
 277         _brk_end += size;
 278
 279         memset(ret, 0, size);
 280
 281         return ret;
 282 }
 283
 284 static void __init reserve_brk(void)
 285 {
 286         if (_brk_end > _brk_start)
 287                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 288
 289         /* Mark brk area as locked down and no longer taking any
 290            new allocations */
 291         _brk_start = 0;
 292 }
 293
 294 #ifdef CONFIG_BLK_DEV_INITRD
 295
 296 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 297 static void __init relocate_initrd(void)
 298 {
 299
 300         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 301         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 302         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 303         u64 ramdisk_here;
 304         unsigned long slop, clen, mapaddr;
 305         char *p, *q;
 306
 307         /* We need to move the initrd down into lowmem */
 308         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 309                                          PAGE_SIZE);
 310
 311         if (ramdisk_here == -1ULL)
 312                 panic("Cannot find place for new RAMDISK of size %lld\n",
 313                          ramdisk_size);
 314
 315         /* Note: this includes all the lowmem currently occupied by
 316            the initrd, we rely on that fact to keep the data intact. */
 317         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 318                          "NEW RAMDISK");
 319         initrd_start = ramdisk_here + PAGE_OFFSET;
 320         initrd_end   = initrd_start + ramdisk_size;
 321         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 322                          ramdisk_here, ramdisk_here + ramdisk_size);
 323
 324         q = (char *)initrd_start;
 325
 326         /* Copy any lowmem portion of the initrd */
 327         if (ramdisk_image < end_of_lowmem) {
 328                 clen = end_of_lowmem - ramdisk_image;
 329                 p = (char *)__va(ramdisk_image);
 330                 memcpy(q, p, clen);
 331                 q += clen;
 332                 ramdisk_image += clen;
 333                 ramdisk_size  -= clen;
 334         }
 335
 336         /* Copy the highmem portion of the initrd */
 337         while (ramdisk_size) {
 338                 slop = ramdisk_image & ~PAGE_MASK;
 339                 clen = ramdisk_size;
 340                 if (clen > MAX_MAP_CHUNK-slop)
 341                         clen = MAX_MAP_CHUNK-slop;
 342                 mapaddr = ramdisk_image & PAGE_MASK;
 343                 p = early_memremap(mapaddr, clen+slop);
 344                 memcpy(q, p+slop, clen);
 345                 early_iounmap(p, clen+slop);
 346                 q += clen;
 347                 ramdisk_image += clen;
 348                 ramdisk_size  -= clen;
 349         }
 350         /* high pages is not converted by early_res_to_bootmem */
 351         ramdisk_image = boot_params.hdr.ramdisk_image;
 352         ramdisk_size  = boot_params.hdr.ramdisk_size;
 353         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 354                 " %08llx - %08llx\n",
 355                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 356                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 357 }
 358
 359 static void __init reserve_initrd(void)
 360 {
 361         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 362         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 363         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 364         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 365
 366         if (!boot_params.hdr.type_of_loader ||
 367             !ramdisk_image || !ramdisk_size)
 368                 return;         /* No initrd provided by bootloader */
 369
 370         initrd_start = 0;
 371
 372         if (ramdisk_size >= (end_of_lowmem>>1)) {
 373                 free_early(ramdisk_image, ramdisk_end);
 374                 printk(KERN_ERR "initrd too large to handle, "
 375                        "disabling initrd\n");
 376                 return;
 377         }
 378
 379         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 380                         ramdisk_end);
 381
 382
 383         if (ramdisk_end <= end_of_lowmem) {
 384                 /* All in lowmem, easy case */
 385                 /*
 386                  * don't need to reserve again, already reserved early
 387                  * in i386_start_kernel
 388                  */
 389                 initrd_start = ramdisk_image + PAGE_OFFSET;
 390                 initrd_end = initrd_start + ramdisk_size;
 391                 return;
 392         }
 393
 394         relocate_initrd();
 395
 396         free_early(ramdisk_image, ramdisk_end);
 397 }
 398 #else
 399 static void __init reserve_initrd(void)
 400 {
 401 }
 402 #endif /* CONFIG_BLK_DEV_INITRD */
 403
 404 static void __init parse_setup_data(void)
 405 {
 406         struct setup_data *data;
 407         u64 pa_data;
 408
 409         if (boot_params.hdr.version < 0x0209)
 410                 return;
 411         pa_data = boot_params.hdr.setup_data;
 412         while (pa_data) {
 413                 data = early_memremap(pa_data, PAGE_SIZE);
 414                 switch (data->type) {
 415                 case SETUP_E820_EXT:
 416                         parse_e820_ext(data, pa_data);
 417                         break;
 418                 default:
 419                         break;
 420                 }
 421                 pa_data = data->next;
 422                 early_iounmap(data, PAGE_SIZE);
 423         }
 424 }
 425
 426 static void __init e820_reserve_setup_data(void)
 427 {
 428         struct setup_data *data;
 429         u64 pa_data;
 430         int found = 0;
 431
 432         if (boot_params.hdr.version < 0x0209)
 433                 return;
 434         pa_data = boot_params.hdr.setup_data;
 435         while (pa_data) {
 436                 data = early_memremap(pa_data, sizeof(*data));
 437                 e820_update_range(pa_data, sizeof(*data)+data->len,
 438                          E820_RAM, E820_RESERVED_KERN);
 439                 found = 1;
 440                 pa_data = data->next;
 441                 early_iounmap(data, sizeof(*data));
 442         }
 443         if (!found)
 444                 return;
 445
 446         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 447         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 448         printk(KERN_INFO "extended physical RAM map:\n");
 449         e820_print_map("reserve setup_data");
 450 }
 451
 452 static void __init reserve_early_setup_data(void)
 453 {
 454         struct setup_data *data;
 455         u64 pa_data;
 456         char buf[32];
 457
 458         if (boot_params.hdr.version < 0x0209)
 459                 return;
 460         pa_data = boot_params.hdr.setup_data;
 461         while (pa_data) {
 462                 data = early_memremap(pa_data, sizeof(*data));
 463                 sprintf(buf, "setup data %x", data->type);
 464                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 465                 pa_data = data->next;
 466                 early_iounmap(data, sizeof(*data));
 467         }
 468 }
 469
 470 /*
 471  * --------- Crashkernel reservation ------------------------------
 472  */
 473
 474 #ifdef CONFIG_KEXEC
 475
 476 /**
 477  * Reserve @size bytes of crashkernel memory at any suitable offset.
 478  *
 479  * @size: Size of the crashkernel memory to reserve.
 480  * Returns the base address on success, and -1ULL on failure.
 481  */
 482 static
 483 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 484 {
 485         const unsigned long long alignment = 16<<20;    /* 16M */
 486         unsigned long long start = 0LL;
 487
 488         while (1) {
 489                 int ret;
 490
 491                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 492                 if (start == -1ULL)
 493                         return start;
 494
 495                 /* try to reserve it */
 496                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 497                 if (ret >= 0)
 498                         return start;
 499
 500                 start += alignment;
 501         }
 502 }
 503
 504 static inline unsigned long long get_total_mem(void)
 505 {
 506         unsigned long long total;
 507
 508         total = max_low_pfn - min_low_pfn;
 509 #ifdef CONFIG_HIGHMEM
 510         total += highend_pfn - highstart_pfn;
 511 #endif
 512
 513         return total << PAGE_SHIFT;
 514 }
 515
 516 static void __init reserve_crashkernel(void)
 517 {
 518         unsigned long long total_mem;
 519         unsigned long long crash_size, crash_base;
 520         int ret;
 521
 522         total_mem = get_total_mem();
 523
 524         ret = parse_crashkernel(boot_command_line, total_mem,
 525                         &crash_size, &crash_base);
 526         if (ret != 0 || crash_size <= 0)
 527                 return;
 528
 529         /* 0 means: find the address automatically */
 530         if (crash_base <= 0) {
 531                 crash_base = find_and_reserve_crashkernel(crash_size);
 532                 if (crash_base == -1ULL) {
 533                         pr_info("crashkernel reservation failed. "
 534                                 "No suitable area found.\n");
 535                         return;
 536                 }
 537         } else {
 538                 ret = reserve_bootmem_generic(crash_base, crash_size,
 539                                         BOOTMEM_EXCLUSIVE);
 540                 if (ret < 0) {
 541                         pr_info("crashkernel reservation failed - "
 542                                 "memory is in use\n");
 543                         return;
 544                 }
 545         }
 546
 547         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 548                         "for crashkernel (System RAM: %ldMB)\n",
 549                         (unsigned long)(crash_size >> 20),
 550                         (unsigned long)(crash_base >> 20),
 551                         (unsigned long)(total_mem >> 20));
 552
 553         crashk_res.start = crash_base;
 554         crashk_res.end   = crash_base + crash_size - 1;
 555         insert_resource(&iomem_resource, &crashk_res);
 556 }
 557 #else
 558 static void __init reserve_crashkernel(void)
 559 {
 560 }
 561 #endif
 562
 563 static struct resource standard_io_resources[] = {
 564         { .name = "dma1", .start = 0x00, .end = 0x1f,
 565                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 566         { .name = "pic1", .start = 0x20, .end = 0x21,
 567                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 568         { .name = "timer0", .start = 0x40, .end = 0x43,
 569                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 570         { .name = "timer1", .start = 0x50, .end = 0x53,
 571                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 572         { .name = "keyboard", .start = 0x60, .end = 0x60,
 573                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 574         { .name = "keyboard", .start = 0x64, .end = 0x64,
 575                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 576         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 577                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 578         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 579                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 580         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 581                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 582         { .name = "fpu", .start = 0xf0, .end = 0xff,
 583                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 584 };
 585
 586 static void __init reserve_standard_io_resources(void)
 587 {
 588         int i;
 589
 590         /* request I/O space for devices used on all i[345]86 PCs */
 591         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 592                 request_resource(&ioport_resource, &standard_io_resources[i]);
 593
 594 }
 595
 596 /*
 597  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 598  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 599  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 600  */
 601
 602 #ifdef CONFIG_CRASH_DUMP
 603 /* elfcorehdr= specifies the location of elf core header
 604  * stored by the crashed kernel. This option will be passed
 605  * by kexec loader to the capture kernel.
 606  */
 607 static int __init setup_elfcorehdr(char *arg)
 608 {
 609         char *end;
 610         if (!arg)
 611                 return -EINVAL;
 612         elfcorehdr_addr = memparse(arg, &end);
 613         return end > arg ? 0 : -EINVAL;
 614 }
 615 early_param("elfcorehdr", setup_elfcorehdr);
 616 #endif
 617
 618 static struct x86_quirks default_x86_quirks __initdata;
 619
 620 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 621
 622 #ifdef CONFIG_X86_RESERVE_LOW_64K
 623 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 624 {
 625         printk(KERN_NOTICE
 626                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 627                 d->ident);
 628
 629         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 630         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 631
 632         return 0;
 633 }
 634 #endif
 635
 636 /* List of systems that have known low memory corruption BIOS problems */
 637 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 638 #ifdef CONFIG_X86_RESERVE_LOW_64K
 639         {
 640                 .callback = dmi_low_memory_corruption,
 641                 .ident = "AMI BIOS",
 642                 .matches = {
 643                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 644                 },
 645         },
 646         {
 647                 .callback = dmi_low_memory_corruption,
 648                 .ident = "Phoenix BIOS",
 649                 .matches = {
 650                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 651                 },
 652         },
 653 #endif
 654         {}
 655 };
 656
 657 /*
 658  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 659  * passed the efi memmap, systab, etc., so we should use these data structures
 660  * for initialization.  Note, the efi init code path is determined by the
 661  * global efi_enabled. This allows the same kernel image to be used on existing
 662  * systems (with a traditional BIOS) as well as on EFI systems.
 663  */
 664 /*
 665  * setup_arch - architecture-specific boot-time initializations
 666  *
 667  * Note: On x86_64, fixmaps are ready for use even before this is called.
 668  */
 669
 670 void __init setup_arch(char **cmdline_p)
 671 {
 672 #ifdef CONFIG_X86_32
 673         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 674         visws_early_detect();
 675 #else
 676         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 677 #endif
 678
 679         /* VMI may relocate the fixmap; do this before touching ioremap area */
 680         vmi_init();
 681
 682         early_cpu_init();
 683         early_ioremap_init();
 684
 685         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 686         screen_info = boot_params.screen_info;
 687         edid_info = boot_params.edid_info;
 688 #ifdef CONFIG_X86_32
 689         apm_info.bios = boot_params.apm_bios_info;
 690         ist_info = boot_params.ist_info;
 691         if (boot_params.sys_desc_table.length != 0) {
 692                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 693                 machine_id = boot_params.sys_desc_table.table[0];
 694                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 695                 BIOS_revision = boot_params.sys_desc_table.table[2];
 696         }
 697 #endif
 698         saved_video_mode = boot_params.hdr.vid_mode;
 699         bootloader_type = boot_params.hdr.type_of_loader;
 700
 701 #ifdef CONFIG_BLK_DEV_RAM
 702         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 703         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 704         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 705 #endif
 706 #ifdef CONFIG_EFI
 707         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 708 #ifdef CONFIG_X86_32
 709                      "EL32",
 710 #else
 711                      "EL64",
 712 #endif
 713          4)) {
 714                 efi_enabled = 1;
 715                 efi_reserve_early();
 716         }
 717 #endif
 718
 719         ARCH_SETUP
 720
 721         setup_memory_map();
 722         parse_setup_data();
 723         /* update the e820_saved too */
 724         e820_reserve_setup_data();
 725
 726         copy_edd();
 727
 728         if (!boot_params.hdr.root_flags)
 729                 root_mountflags &= ~MS_RDONLY;
 730         init_mm.start_code = (unsigned long) _text;
 731         init_mm.end_code = (unsigned long) _etext;
 732         init_mm.end_data = (unsigned long) _edata;
 733         init_mm.brk = _brk_end;
 734
 735         code_resource.start = virt_to_phys(_text);
 736         code_resource.end = virt_to_phys(_etext)-1;
 737         data_resource.start = virt_to_phys(_etext);
 738         data_resource.end = virt_to_phys(_edata)-1;
 739         bss_resource.start = virt_to_phys(&__bss_start);
 740         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 741
 742 #ifdef CONFIG_CMDLINE_BOOL
 743 #ifdef CONFIG_CMDLINE_OVERRIDE
 744         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 745 #else
 746         if (builtin_cmdline[0]) {
 747                 /* append boot loader cmdline to builtin */
 748                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 749                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 750                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 751         }
 752 #endif
 753 #endif
 754
 755         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 756         *cmdline_p = command_line;
 757
 758         parse_early_param();
 759
 760 #ifdef CONFIG_X86_64
 761         check_efer();
 762 #endif
 763
 764         /* Must be before kernel pagetables are setup */
 765         vmi_activate();
 766
 767         /* after early param, so could get panic from serial */
 768         reserve_early_setup_data();
 769
 770         if (acpi_mps_check()) {
 771 #ifdef CONFIG_X86_LOCAL_APIC
 772                 disable_apic = 1;
 773 #endif
 774                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 775         }
 776
 777 #ifdef CONFIG_PCI
 778         if (pci_early_dump_regs)
 779                 early_dump_pci_devices();
 780 #endif
 781
 782         finish_e820_parsing();
 783
 784         if (efi_enabled)
 785                 efi_init();
 786
 787         dmi_scan_machine();
 788
 789         dmi_check_system(bad_bios_dmi_table);
 790
 791         /*
 792          * VMware detection requires dmi to be available, so this
 793          * needs to be done after dmi_scan_machine, for the BP.
 794          */
 795         init_hypervisor(&boot_cpu_data);
 796
 797 #ifdef CONFIG_X86_32
 798         probe_roms();
 799 #endif
 800
 801         /* after parse_early_param, so could debug it */
 802         insert_resource(&iomem_resource, &code_resource);
 803         insert_resource(&iomem_resource, &data_resource);
 804         insert_resource(&iomem_resource, &bss_resource);
 805
 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         reserve_brk();
 855
 856         /* max_pfn_mapped is updated here */
 857         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 858         max_pfn_mapped = max_low_pfn_mapped;
 859
 860 #ifdef CONFIG_X86_64
 861         if (max_pfn > max_low_pfn) {
 862                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 863                                                      max_pfn<<PAGE_SHIFT);
 864                 /* can we preseve max_low_pfn ?*/
 865                 max_low_pfn = max_pfn;
 866         }
 867 #endif
 868
 869         /*
 870          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 871          */
 872
 873 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 874         if (init_ohci1394_dma_early)
 875                 init_ohci1394_dma_on_all_controllers();
 876 #endif
 877
 878         reserve_initrd();
 879
 880         vsmp_init();
 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         bootmem_state = DURING_BOOTMEM;
 900
 901 #ifdef CONFIG_ACPI_SLEEP
 902         /*
 903          * Reserve low memory region for sleep support.
 904          */
 905         acpi_reserve_bootmem();
 906 #endif
 907         /*
 908          * Find and reserve possible boot-time SMP configuration:
 909          */
 910         find_smp_config();
 911
 912         reserve_crashkernel();
 913
 914 #ifdef CONFIG_X86_64
 915         /*
 916          * dma32_reserve_bootmem() allocates bootmem which may conflict
 917          * with the crashkernel command line, so do that after
 918          * reserve_crashkernel()
 919          */
 920         dma32_reserve_bootmem();
 921 #endif
 922
 923         reserve_ibft_region();
 924
 925 #ifdef CONFIG_KVM_CLOCK
 926         kvmclock_init();
 927 #endif
 928
 929         paravirt_pagetable_setup_start(swapper_pg_dir);
 930         paging_init();
 931         paravirt_pagetable_setup_done(swapper_pg_dir);
 932         paravirt_post_allocator_init();
 933
 934 #ifdef CONFIG_X86_64
 935         map_vsyscall();
 936 #endif
 937
 938         generic_apic_probe();
 939
 940         early_quirks();
 941
 942         /*
 943          * Read APIC and some other early information from ACPI tables.
 944          */
 945         acpi_boot_init();
 946
 947 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 948         /*
 949          * get boot-time SMP configuration:
 950          */
 951         if (smp_found_config)
 952                 get_smp_config();
 953 #endif
 954
 955         prefill_possible_map();
 956
 957 #ifdef CONFIG_X86_64
 958         init_cpu_to_node();
 959 #endif
 960
 961         init_apic_mappings();
 962         ioapic_init_mappings();
 963
 964         /* need to wait for io_apic is mapped */
 965         probe_nr_irqs_gsi();
 966
 967         kvm_guest_init();
 968
 969         e820_reserve_resources();
 970         e820_mark_nosave_regions(max_low_pfn);
 971
 972 #ifdef CONFIG_X86_32
 973         request_resource(&iomem_resource, &video_ram_resource);
 974 #endif
 975         reserve_standard_io_resources();
 976
 977         e820_setup_gap();
 978
 979 #ifdef CONFIG_VT
 980 #if defined(CONFIG_VGA_CONSOLE)
 981         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 982                 conswitchp = &vga_con;
 983 #elif defined(CONFIG_DUMMY_CONSOLE)
 984         conswitchp = &dummy_con;
 985 #endif
 986 #endif
 987 }
 988
 989 #ifdef CONFIG_X86_32
 990
 991 /**
 992  * x86_quirk_pre_intr_init - initialisation prior to setting up interrupt vectors
 993  *
 994  * Description:
 995  *      Perform any necessary interrupt initialisation prior to setting up
 996  *      the "ordinary" interrupt call gates.  For legacy reasons, the ISA
 997  *      interrupts should be initialised here if the machine emulates a PC
 998  *      in any way.
 999  **/
1000 void __init x86_quirk_pre_intr_init(void)
1001 {
1002         if (x86_quirks->arch_pre_intr_init) {
1003                 if (x86_quirks->arch_pre_intr_init())
1004                         return;
1005         }
1006         init_ISA_irqs();
1007 }
1008
1009 /**
1010  * x86_quirk_intr_init - post gate setup interrupt initialisation
1011  *
1012  * Description:
1013  *      Fill in any interrupts that may have been left out by the general
1014  *      init_IRQ() routine.  interrupts having to do with the machine rather
1015  *      than the devices on the I/O bus (like APIC interrupts in intel MP
1016  *      systems) are started here.
1017  **/
1018 void __init x86_quirk_intr_init(void)
1019 {
1020         if (x86_quirks->arch_intr_init) {
1021                 if (x86_quirks->arch_intr_init())
1022                         return;
1023         }
1024 }
1025
1026 /**
1027  * x86_quirk_trap_init - initialise system specific traps
1028  *
1029  * Description:
1030  *      Called as the final act of trap_init().  Used in VISWS to initialise
1031  *      the various board specific APIC traps.
1032  **/
1033 void __init x86_quirk_trap_init(void)
1034 {
1035         if (x86_quirks->arch_trap_init) {
1036                 if (x86_quirks->arch_trap_init())
1037                         return;
1038         }
1039 }
1040
1041 static struct irqaction irq0  = {
1042         .handler = timer_interrupt,
1043         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1044         .mask = CPU_MASK_NONE,
1045         .name = "timer"
1046 };
1047
1048 /**
1049  * x86_quirk_pre_time_init - do any specific initialisations before.
1050  *
1051  **/
1052 void __init x86_quirk_pre_time_init(void)
1053 {
1054         if (x86_quirks->arch_pre_time_init)
1055                 x86_quirks->arch_pre_time_init();
1056 }
1057
1058 /**
1059  * x86_quirk_time_init - do any specific initialisations for the system timer.
1060  *
1061  * Description:
1062  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1063  *      in irq_vectors.h:TIMER_IRQ
1064  **/
1065 void __init x86_quirk_time_init(void)
1066 {
1067         if (x86_quirks->arch_time_init) {
1068                 /*
1069                  * A nonzero return code does not mean failure, it means
1070                  * that the architecture quirk does not want any
1071                  * generic (timer) setup to be performed after this:
1072                  */
1073                 if (x86_quirks->arch_time_init())
1074                         return;
1075         }
1076
1077         irq0.mask = cpumask_of_cpu(0);
1078         setup_irq(0, &irq0);
1079 }
1080 #endif /* CONFIG_X86_32 */