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  * 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 void * __init extend_brk(size_t size, size_t align)
 270 {
 271         size_t mask = align - 1;
 272         void *ret;
 273
 274         BUG_ON(_brk_start == 0);
 275         BUG_ON(align & mask);
 276
 277         _brk_end = (_brk_end + mask) & ~mask;
 278         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 279
 280         ret = (void *)_brk_end;
 281         _brk_end += size;
 282
 283         memset(ret, 0, size);
 284
 285         return ret;
 286 }
 287
 288 static void __init reserve_brk(void)
 289 {
 290         if (_brk_end > _brk_start)
 291                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 292
 293         /* Mark brk area as locked down and no longer taking any
 294            new allocations */
 295         _brk_start = 0;
 296 }
 297
 298 #ifdef CONFIG_BLK_DEV_INITRD
 299
 300 #ifdef CONFIG_X86_32
 301
 302 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 303 static void __init relocate_initrd(void)
 304 {
 305
 306         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 307         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 308         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 309         u64 ramdisk_here;
 310         unsigned long slop, clen, mapaddr;
 311         char *p, *q;
 312
 313         /* We need to move the initrd down into lowmem */
 314         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 315                                          PAGE_SIZE);
 316
 317         if (ramdisk_here == -1ULL)
 318                 panic("Cannot find place for new RAMDISK of size %lld\n",
 319                          ramdisk_size);
 320
 321         /* Note: this includes all the lowmem currently occupied by
 322            the initrd, we rely on that fact to keep the data intact. */
 323         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 324                          "NEW RAMDISK");
 325         initrd_start = ramdisk_here + PAGE_OFFSET;
 326         initrd_end   = initrd_start + ramdisk_size;
 327         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 328                          ramdisk_here, ramdisk_here + ramdisk_size);
 329
 330         q = (char *)initrd_start;
 331
 332         /* Copy any lowmem portion of the initrd */
 333         if (ramdisk_image < end_of_lowmem) {
 334                 clen = end_of_lowmem - ramdisk_image;
 335                 p = (char *)__va(ramdisk_image);
 336                 memcpy(q, p, clen);
 337                 q += clen;
 338                 ramdisk_image += clen;
 339                 ramdisk_size  -= clen;
 340         }
 341
 342         /* Copy the highmem portion of the initrd */
 343         while (ramdisk_size) {
 344                 slop = ramdisk_image & ~PAGE_MASK;
 345                 clen = ramdisk_size;
 346                 if (clen > MAX_MAP_CHUNK-slop)
 347                         clen = MAX_MAP_CHUNK-slop;
 348                 mapaddr = ramdisk_image & PAGE_MASK;
 349                 p = early_memremap(mapaddr, clen+slop);
 350                 memcpy(q, p+slop, clen);
 351                 early_iounmap(p, clen+slop);
 352                 q += clen;
 353                 ramdisk_image += clen;
 354                 ramdisk_size  -= clen;
 355         }
 356         /* high pages is not converted by early_res_to_bootmem */
 357         ramdisk_image = boot_params.hdr.ramdisk_image;
 358         ramdisk_size  = boot_params.hdr.ramdisk_size;
 359         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 360                 " %08llx - %08llx\n",
 361                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 362                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 363 }
 364 #endif
 365
 366 static void __init reserve_initrd(void)
 367 {
 368         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 369         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 370         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 371         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 372
 373         if (!boot_params.hdr.type_of_loader ||
 374             !ramdisk_image || !ramdisk_size)
 375                 return;         /* No initrd provided by bootloader */
 376
 377         initrd_start = 0;
 378
 379         if (ramdisk_size >= (end_of_lowmem>>1)) {
 380                 free_early(ramdisk_image, ramdisk_end);
 381                 printk(KERN_ERR "initrd too large to handle, "
 382                        "disabling initrd\n");
 383                 return;
 384         }
 385
 386         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 387                         ramdisk_end);
 388
 389
 390         if (ramdisk_end <= end_of_lowmem) {
 391                 /* All in lowmem, easy case */
 392                 /*
 393                  * don't need to reserve again, already reserved early
 394                  * in i386_start_kernel
 395                  */
 396                 initrd_start = ramdisk_image + PAGE_OFFSET;
 397                 initrd_end = initrd_start + ramdisk_size;
 398                 return;
 399         }
 400
 401 #ifdef CONFIG_X86_32
 402         relocate_initrd();
 403 #else
 404         printk(KERN_ERR "initrd extends beyond end of memory "
 405                "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
 406                ramdisk_end, end_of_lowmem);
 407         initrd_start = 0;
 408 #endif
 409         free_early(ramdisk_image, ramdisk_end);
 410 }
 411 #else
 412 static void __init reserve_initrd(void)
 413 {
 414 }
 415 #endif /* CONFIG_BLK_DEV_INITRD */
 416
 417 static void __init parse_setup_data(void)
 418 {
 419         struct setup_data *data;
 420         u64 pa_data;
 421
 422         if (boot_params.hdr.version < 0x0209)
 423                 return;
 424         pa_data = boot_params.hdr.setup_data;
 425         while (pa_data) {
 426                 data = early_memremap(pa_data, PAGE_SIZE);
 427                 switch (data->type) {
 428                 case SETUP_E820_EXT:
 429                         parse_e820_ext(data, pa_data);
 430                         break;
 431                 default:
 432                         break;
 433                 }
 434                 pa_data = data->next;
 435                 early_iounmap(data, PAGE_SIZE);
 436         }
 437 }
 438
 439 static void __init e820_reserve_setup_data(void)
 440 {
 441         struct setup_data *data;
 442         u64 pa_data;
 443         int found = 0;
 444
 445         if (boot_params.hdr.version < 0x0209)
 446                 return;
 447         pa_data = boot_params.hdr.setup_data;
 448         while (pa_data) {
 449                 data = early_memremap(pa_data, sizeof(*data));
 450                 e820_update_range(pa_data, sizeof(*data)+data->len,
 451                          E820_RAM, E820_RESERVED_KERN);
 452                 found = 1;
 453                 pa_data = data->next;
 454                 early_iounmap(data, sizeof(*data));
 455         }
 456         if (!found)
 457                 return;
 458
 459         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 460         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 461         printk(KERN_INFO "extended physical RAM map:\n");
 462         e820_print_map("reserve setup_data");
 463 }
 464
 465 static void __init reserve_early_setup_data(void)
 466 {
 467         struct setup_data *data;
 468         u64 pa_data;
 469         char buf[32];
 470
 471         if (boot_params.hdr.version < 0x0209)
 472                 return;
 473         pa_data = boot_params.hdr.setup_data;
 474         while (pa_data) {
 475                 data = early_memremap(pa_data, sizeof(*data));
 476                 sprintf(buf, "setup data %x", data->type);
 477                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 478                 pa_data = data->next;
 479                 early_iounmap(data, sizeof(*data));
 480         }
 481 }
 482
 483 /*
 484  * --------- Crashkernel reservation ------------------------------
 485  */
 486
 487 #ifdef CONFIG_KEXEC
 488
 489 /**
 490  * Reserve @size bytes of crashkernel memory at any suitable offset.
 491  *
 492  * @size: Size of the crashkernel memory to reserve.
 493  * Returns the base address on success, and -1ULL on failure.
 494  */
 495 static
 496 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 497 {
 498         const unsigned long long alignment = 16<<20;    /* 16M */
 499         unsigned long long start = 0LL;
 500
 501         while (1) {
 502                 int ret;
 503
 504                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 505                 if (start == -1ULL)
 506                         return start;
 507
 508                 /* try to reserve it */
 509                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 510                 if (ret >= 0)
 511                         return start;
 512
 513                 start += alignment;
 514         }
 515 }
 516
 517 static inline unsigned long long get_total_mem(void)
 518 {
 519         unsigned long long total;
 520
 521         total = max_low_pfn - min_low_pfn;
 522 #ifdef CONFIG_HIGHMEM
 523         total += highend_pfn - highstart_pfn;
 524 #endif
 525
 526         return total << PAGE_SHIFT;
 527 }
 528
 529 static void __init reserve_crashkernel(void)
 530 {
 531         unsigned long long total_mem;
 532         unsigned long long crash_size, crash_base;
 533         int ret;
 534
 535         total_mem = get_total_mem();
 536
 537         ret = parse_crashkernel(boot_command_line, total_mem,
 538                         &crash_size, &crash_base);
 539         if (ret != 0 || crash_size <= 0)
 540                 return;
 541
 542         /* 0 means: find the address automatically */
 543         if (crash_base <= 0) {
 544                 crash_base = find_and_reserve_crashkernel(crash_size);
 545                 if (crash_base == -1ULL) {
 546                         pr_info("crashkernel reservation failed. "
 547                                 "No suitable area found.\n");
 548                         return;
 549                 }
 550         } else {
 551                 ret = reserve_bootmem_generic(crash_base, crash_size,
 552                                         BOOTMEM_EXCLUSIVE);
 553                 if (ret < 0) {
 554                         pr_info("crashkernel reservation failed - "
 555                                 "memory is in use\n");
 556                         return;
 557                 }
 558         }
 559
 560         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 561                         "for crashkernel (System RAM: %ldMB)\n",
 562                         (unsigned long)(crash_size >> 20),
 563                         (unsigned long)(crash_base >> 20),
 564                         (unsigned long)(total_mem >> 20));
 565
 566         crashk_res.start = crash_base;
 567         crashk_res.end   = crash_base + crash_size - 1;
 568         insert_resource(&iomem_resource, &crashk_res);
 569 }
 570 #else
 571 static void __init reserve_crashkernel(void)
 572 {
 573 }
 574 #endif
 575
 576 static struct resource standard_io_resources[] = {
 577         { .name = "dma1", .start = 0x00, .end = 0x1f,
 578                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 579         { .name = "pic1", .start = 0x20, .end = 0x21,
 580                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 581         { .name = "timer0", .start = 0x40, .end = 0x43,
 582                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 583         { .name = "timer1", .start = 0x50, .end = 0x53,
 584                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 585         { .name = "keyboard", .start = 0x60, .end = 0x60,
 586                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 587         { .name = "keyboard", .start = 0x64, .end = 0x64,
 588                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 589         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 590                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 591         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 592                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 593         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 594                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 595         { .name = "fpu", .start = 0xf0, .end = 0xff,
 596                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 597 };
 598
 599 static void __init reserve_standard_io_resources(void)
 600 {
 601         int i;
 602
 603         /* request I/O space for devices used on all i[345]86 PCs */
 604         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 605                 request_resource(&ioport_resource, &standard_io_resources[i]);
 606
 607 }
 608
 609 /*
 610  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 611  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 612  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 613  */
 614
 615 #ifdef CONFIG_CRASH_DUMP
 616 /* elfcorehdr= specifies the location of elf core header
 617  * stored by the crashed kernel. This option will be passed
 618  * by kexec loader to the capture kernel.
 619  */
 620 static int __init setup_elfcorehdr(char *arg)
 621 {
 622         char *end;
 623         if (!arg)
 624                 return -EINVAL;
 625         elfcorehdr_addr = memparse(arg, &end);
 626         return end > arg ? 0 : -EINVAL;
 627 }
 628 early_param("elfcorehdr", setup_elfcorehdr);
 629 #endif
 630
 631 static struct x86_quirks default_x86_quirks __initdata;
 632
 633 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 634
 635 #ifdef CONFIG_X86_RESERVE_LOW_64K
 636 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 637 {
 638         printk(KERN_NOTICE
 639                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 640                 d->ident);
 641
 642         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 643         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 644
 645         return 0;
 646 }
 647 #endif
 648
 649 /* List of systems that have known low memory corruption BIOS problems */
 650 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 651 #ifdef CONFIG_X86_RESERVE_LOW_64K
 652         {
 653                 .callback = dmi_low_memory_corruption,
 654                 .ident = "AMI BIOS",
 655                 .matches = {
 656                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 657                 },
 658         },
 659         {
 660                 .callback = dmi_low_memory_corruption,
 661                 .ident = "Phoenix BIOS",
 662                 .matches = {
 663                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 664                 },
 665         },
 666 #endif
 667         {}
 668 };
 669
 670 /*
 671  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 672  * passed the efi memmap, systab, etc., so we should use these data structures
 673  * for initialization.  Note, the efi init code path is determined by the
 674  * global efi_enabled. This allows the same kernel image to be used on existing
 675  * systems (with a traditional BIOS) as well as on EFI systems.
 676  */
 677 /*
 678  * setup_arch - architecture-specific boot-time initializations
 679  *
 680  * Note: On x86_64, fixmaps are ready for use even before this is called.
 681  */
 682
 683 void __init setup_arch(char **cmdline_p)
 684 {
 685 #ifdef CONFIG_X86_32
 686         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 687         visws_early_detect();
 688 #else
 689         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 690 #endif
 691
 692         /* VMI may relocate the fixmap; do this before touching ioremap area */
 693         vmi_init();
 694
 695         early_cpu_init();
 696         early_ioremap_init();
 697
 698         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 699         screen_info = boot_params.screen_info;
 700         edid_info = boot_params.edid_info;
 701 #ifdef CONFIG_X86_32
 702         apm_info.bios = boot_params.apm_bios_info;
 703         ist_info = boot_params.ist_info;
 704         if (boot_params.sys_desc_table.length != 0) {
 705                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 706                 machine_id = boot_params.sys_desc_table.table[0];
 707                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 708                 BIOS_revision = boot_params.sys_desc_table.table[2];
 709         }
 710 #endif
 711         saved_video_mode = boot_params.hdr.vid_mode;
 712         bootloader_type = boot_params.hdr.type_of_loader;
 713
 714 #ifdef CONFIG_BLK_DEV_RAM
 715         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 716         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 717         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 718 #endif
 719 #ifdef CONFIG_EFI
 720         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 721 #ifdef CONFIG_X86_32
 722                      "EL32",
 723 #else
 724                      "EL64",
 725 #endif
 726          4)) {
 727                 efi_enabled = 1;
 728                 efi_reserve_early();
 729         }
 730 #endif
 731
 732         ARCH_SETUP
 733
 734         setup_memory_map();
 735         parse_setup_data();
 736         /* update the e820_saved too */
 737         e820_reserve_setup_data();
 738
 739         copy_edd();
 740
 741         if (!boot_params.hdr.root_flags)
 742                 root_mountflags &= ~MS_RDONLY;
 743         init_mm.start_code = (unsigned long) _text;
 744         init_mm.end_code = (unsigned long) _etext;
 745         init_mm.end_data = (unsigned long) _edata;
 746         init_mm.brk = _brk_end;
 747
 748         code_resource.start = virt_to_phys(_text);
 749         code_resource.end = virt_to_phys(_etext)-1;
 750         data_resource.start = virt_to_phys(_etext);
 751         data_resource.end = virt_to_phys(_edata)-1;
 752         bss_resource.start = virt_to_phys(&__bss_start);
 753         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 754
 755 #ifdef CONFIG_CMDLINE_BOOL
 756 #ifdef CONFIG_CMDLINE_OVERRIDE
 757         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 758 #else
 759         if (builtin_cmdline[0]) {
 760                 /* append boot loader cmdline to builtin */
 761                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 762                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 763                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 764         }
 765 #endif
 766 #endif
 767
 768         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 769         *cmdline_p = command_line;
 770
 771         parse_early_param();
 772
 773 #ifdef CONFIG_X86_64
 774         check_efer();
 775 #endif
 776
 777         /* Must be before kernel pagetables are setup */
 778         vmi_activate();
 779
 780         /* after early param, so could get panic from serial */
 781         reserve_early_setup_data();
 782
 783         if (acpi_mps_check()) {
 784 #ifdef CONFIG_X86_LOCAL_APIC
 785                 disable_apic = 1;
 786 #endif
 787                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 788         }
 789
 790 #ifdef CONFIG_PCI
 791         if (pci_early_dump_regs)
 792                 early_dump_pci_devices();
 793 #endif
 794
 795         finish_e820_parsing();
 796
 797         if (efi_enabled)
 798                 efi_init();
 799
 800         dmi_scan_machine();
 801
 802         dmi_check_system(bad_bios_dmi_table);
 803
 804         /*
 805          * VMware detection requires dmi to be available, so this
 806          * needs to be done after dmi_scan_machine, for the BP.
 807          */
 808         init_hypervisor(&boot_cpu_data);
 809
 810 #ifdef CONFIG_X86_32
 811         probe_roms();
 812 #endif
 813
 814         /* after parse_early_param, so could debug it */
 815         insert_resource(&iomem_resource, &code_resource);
 816         insert_resource(&iomem_resource, &data_resource);
 817         insert_resource(&iomem_resource, &bss_resource);
 818
 819
 820 #ifdef CONFIG_X86_32
 821         if (ppro_with_ram_bug()) {
 822                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 823                                   E820_RESERVED);
 824                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 825                 printk(KERN_INFO "fixed physical RAM map:\n");
 826                 e820_print_map("bad_ppro");
 827         }
 828 #else
 829         early_gart_iommu_check();
 830 #endif
 831
 832         /*
 833          * partially used pages are not usable - thus
 834          * we are rounding upwards:
 835          */
 836         max_pfn = e820_end_of_ram_pfn();
 837
 838         /* preallocate 4k for mptable mpc */
 839         early_reserve_e820_mpc_new();
 840         /* update e820 for memory not covered by WB MTRRs */
 841         mtrr_bp_init();
 842         if (mtrr_trim_uncached_memory(max_pfn))
 843                 max_pfn = e820_end_of_ram_pfn();
 844
 845 #ifdef CONFIG_X86_32
 846         /* max_low_pfn get updated here */
 847         find_low_pfn_range();
 848 #else
 849         num_physpages = max_pfn;
 850
 851         check_x2apic();
 852
 853         /* How many end-of-memory variables you have, grandma! */
 854         /* need this before calling reserve_initrd */
 855         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 856                 max_low_pfn = e820_end_of_low_ram_pfn();
 857         else
 858                 max_low_pfn = max_pfn;
 859
 860         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 861 #endif
 862
 863 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 864         setup_bios_corruption_check();
 865 #endif
 866
 867         reserve_brk();
 868
 869         /* max_pfn_mapped is updated here */
 870         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 871         max_pfn_mapped = max_low_pfn_mapped;
 872
 873 #ifdef CONFIG_X86_64
 874         if (max_pfn > max_low_pfn) {
 875                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 876                                                      max_pfn<<PAGE_SHIFT);
 877                 /* can we preseve max_low_pfn ?*/
 878                 max_low_pfn = max_pfn;
 879         }
 880 #endif
 881
 882         /*
 883          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 884          */
 885
 886 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 887         if (init_ohci1394_dma_early)
 888                 init_ohci1394_dma_on_all_controllers();
 889 #endif
 890
 891         reserve_initrd();
 892
 893         vsmp_init();
 894
 895         io_delay_init();
 896
 897         /*
 898          * Parse the ACPI tables for possible boot-time SMP configuration.
 899          */
 900         acpi_boot_table_init();
 901
 902         early_acpi_boot_init();
 903
 904 #ifdef CONFIG_ACPI_NUMA
 905         /*
 906          * Parse SRAT to discover nodes.
 907          */
 908         acpi_numa_init();
 909 #endif
 910
 911         initmem_init(0, max_pfn);
 912
 913 #ifdef CONFIG_ACPI_SLEEP
 914         /*
 915          * Reserve low memory region for sleep support.
 916          */
 917         acpi_reserve_bootmem();
 918 #endif
 919         /*
 920          * Find and reserve possible boot-time SMP configuration:
 921          */
 922         find_smp_config();
 923
 924         reserve_crashkernel();
 925
 926 #ifdef CONFIG_X86_64
 927         /*
 928          * dma32_reserve_bootmem() allocates bootmem which may conflict
 929          * with the crashkernel command line, so do that after
 930          * reserve_crashkernel()
 931          */
 932         dma32_reserve_bootmem();
 933 #endif
 934
 935         reserve_ibft_region();
 936
 937 #ifdef CONFIG_KVM_CLOCK
 938         kvmclock_init();
 939 #endif
 940
 941         paravirt_pagetable_setup_start(swapper_pg_dir);
 942         paging_init();
 943         paravirt_pagetable_setup_done(swapper_pg_dir);
 944         paravirt_post_allocator_init();
 945
 946 #ifdef CONFIG_X86_64
 947         map_vsyscall();
 948 #endif
 949
 950         generic_apic_probe();
 951
 952         early_quirks();
 953
 954         /*
 955          * Read APIC and some other early information from ACPI tables.
 956          */
 957         acpi_boot_init();
 958
 959 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 960         /*
 961          * get boot-time SMP configuration:
 962          */
 963         if (smp_found_config)
 964                 get_smp_config();
 965 #endif
 966
 967         prefill_possible_map();
 968
 969 #ifdef CONFIG_X86_64
 970         init_cpu_to_node();
 971 #endif
 972
 973         init_apic_mappings();
 974         ioapic_init_mappings();
 975
 976         /* need to wait for io_apic is mapped */
 977         probe_nr_irqs_gsi();
 978
 979         kvm_guest_init();
 980
 981         e820_reserve_resources();
 982         e820_mark_nosave_regions(max_low_pfn);
 983
 984 #ifdef CONFIG_X86_32
 985         request_resource(&iomem_resource, &video_ram_resource);
 986 #endif
 987         reserve_standard_io_resources();
 988
 989         e820_setup_gap();
 990
 991 #ifdef CONFIG_VT
 992 #if defined(CONFIG_VGA_CONSOLE)
 993         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 994                 conswitchp = &vga_con;
 995 #elif defined(CONFIG_DUMMY_CONSOLE)
 996         conswitchp = &dummy_con;
 997 #endif
 998 #endif
 999 }
1000
1001 #ifdef CONFIG_X86_32
1002
1003 /**
1004  * x86_quirk_pre_intr_init - initialisation prior to setting up interrupt vectors
1005  *
1006  * Description:
1007  *      Perform any necessary interrupt initialisation prior to setting up
1008  *      the "ordinary" interrupt call gates.  For legacy reasons, the ISA
1009  *      interrupts should be initialised here if the machine emulates a PC
1010  *      in any way.
1011  **/
1012 void __init x86_quirk_pre_intr_init(void)
1013 {
1014         if (x86_quirks->arch_pre_intr_init) {
1015                 if (x86_quirks->arch_pre_intr_init())
1016                         return;
1017         }
1018         init_ISA_irqs();
1019 }
1020
1021 /**
1022  * x86_quirk_intr_init - post gate setup interrupt initialisation
1023  *
1024  * Description:
1025  *      Fill in any interrupts that may have been left out by the general
1026  *      init_IRQ() routine.  interrupts having to do with the machine rather
1027  *      than the devices on the I/O bus (like APIC interrupts in intel MP
1028  *      systems) are started here.
1029  **/
1030 void __init x86_quirk_intr_init(void)
1031 {
1032         if (x86_quirks->arch_intr_init) {
1033                 if (x86_quirks->arch_intr_init())
1034                         return;
1035         }
1036 }
1037
1038 /**
1039  * x86_quirk_trap_init - initialise system specific traps
1040  *
1041  * Description:
1042  *      Called as the final act of trap_init().  Used in VISWS to initialise
1043  *      the various board specific APIC traps.
1044  **/
1045 void __init x86_quirk_trap_init(void)
1046 {
1047         if (x86_quirks->arch_trap_init) {
1048                 if (x86_quirks->arch_trap_init())
1049                         return;
1050         }
1051 }
1052
1053 static struct irqaction irq0  = {
1054         .handler = timer_interrupt,
1055         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1056         .mask = CPU_MASK_NONE,
1057         .name = "timer"
1058 };
1059
1060 /**
1061  * x86_quirk_pre_time_init - do any specific initialisations before.
1062  *
1063  **/
1064 void __init x86_quirk_pre_time_init(void)
1065 {
1066         if (x86_quirks->arch_pre_time_init)
1067                 x86_quirks->arch_pre_time_init();
1068 }
1069
1070 /**
1071  * x86_quirk_time_init - do any specific initialisations for the system timer.
1072  *
1073  * Description:
1074  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1075  *      in irq_vectors.h:TIMER_IRQ
1076  **/
1077 void __init x86_quirk_time_init(void)
1078 {
1079         if (x86_quirks->arch_time_init) {
1080                 /*
1081                  * A nonzero return code does not mean failure, it means
1082                  * that the architecture quirk does not want any
1083                  * generic (timer) setup to be performed after this:
1084                  */
1085                 if (x86_quirks->arch_time_init())
1086                         return;
1087         }
1088
1089         irq0.mask = cpumask_of_cpu(0);
1090         setup_irq(0, &irq0);
1091 }
1092 #endif /* CONFIG_X86_32 */