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