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 /* This value is set up by the early boot code to point to the value
 165    immediately after the boot time page tables.  It contains a *physical*
 166    address, and must not be in the .bss segment! */
 167 unsigned long init_pg_tables_start __initdata = ~0UL;
 168 unsigned long init_pg_tables_end __initdata = ~0UL;
 169
 170 static struct resource video_ram_resource = {
 171         .name   = "Video RAM area",
 172         .start  = 0xa0000,
 173         .end    = 0xbffff,
 174         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 175 };
 176
 177 /* cpu data as detected by the assembly code in head.S */
 178 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 179 /* common cpu data for all cpus */
 180 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 181 EXPORT_SYMBOL(boot_cpu_data);
 182 static void set_mca_bus(int x)
 183 {
 184 #ifdef CONFIG_MCA
 185         MCA_bus = x;
 186 #endif
 187 }
 188
 189 unsigned int def_to_bigsmp;
 190
 191 /* for MCA, but anyone else can use it if they want */
 192 unsigned int machine_id;
 193 unsigned int machine_submodel_id;
 194 unsigned int BIOS_revision;
 195
 196 struct apm_info apm_info;
 197 EXPORT_SYMBOL(apm_info);
 198
 199 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 200         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 201 struct ist_info ist_info;
 202 EXPORT_SYMBOL(ist_info);
 203 #else
 204 struct ist_info ist_info;
 205 #endif
 206
 207 #else
 208 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 209         .x86_phys_bits = MAX_PHYSMEM_BITS,
 210 };
 211 EXPORT_SYMBOL(boot_cpu_data);
 212 #endif
 213
 214
 215 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 216 unsigned long mmu_cr4_features;
 217 #else
 218 unsigned long mmu_cr4_features = X86_CR4_PAE;
 219 #endif
 220
 221 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 222 int bootloader_type;
 223
 224 /*
 225  * Early DMI memory
 226  */
 227 int dmi_alloc_index;
 228 char dmi_alloc_data[DMI_MAX_DATA];
 229
 230 /*
 231  * Setup options
 232  */
 233 struct screen_info screen_info;
 234 EXPORT_SYMBOL(screen_info);
 235 struct edid_info edid_info;
 236 EXPORT_SYMBOL_GPL(edid_info);
 237
 238 extern int root_mountflags;
 239
 240 unsigned long saved_video_mode;
 241
 242 #define RAMDISK_IMAGE_START_MASK        0x07FF
 243 #define RAMDISK_PROMPT_FLAG             0x8000
 244 #define RAMDISK_LOAD_FLAG               0x4000
 245
 246 static char __initdata command_line[COMMAND_LINE_SIZE];
 247 #ifdef CONFIG_CMDLINE_BOOL
 248 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 249 #endif
 250
 251 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 252 struct edd edd;
 253 #ifdef CONFIG_EDD_MODULE
 254 EXPORT_SYMBOL(edd);
 255 #endif
 256 /**
 257  * copy_edd() - Copy the BIOS EDD information
 258  *              from boot_params into a safe place.
 259  *
 260  */
 261 static inline void copy_edd(void)
 262 {
 263      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 264             sizeof(edd.mbr_signature));
 265      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 266      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 267      edd.edd_info_nr = boot_params.eddbuf_entries;
 268 }
 269 #else
 270 static inline void copy_edd(void)
 271 {
 272 }
 273 #endif
 274
 275 void * __init extend_brk(size_t size, size_t align)
 276 {
 277         size_t mask = align - 1;
 278         void *ret;
 279
 280         BUG_ON(_brk_start == 0);
 281         BUG_ON(align & mask);
 282
 283         _brk_end = (_brk_end + mask) & ~mask;
 284         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 285
 286         ret = (void *)_brk_end;
 287         _brk_end += size;
 288
 289         memset(ret, 0, size);
 290
 291         return ret;
 292 }
 293
 294 static void __init reserve_brk(void)
 295 {
 296         if (_brk_end > _brk_start)
 297                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 298
 299         /* Mark brk area as locked down and no longer taking any
 300            new allocations */
 301         _brk_start = 0;
 302 }
 303
 304 #ifdef CONFIG_BLK_DEV_INITRD
 305
 306 #ifdef CONFIG_X86_32
 307
 308 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 309 static void __init relocate_initrd(void)
 310 {
 311
 312         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 313         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 314         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 315         u64 ramdisk_here;
 316         unsigned long slop, clen, mapaddr;
 317         char *p, *q;
 318
 319         /* We need to move the initrd down into lowmem */
 320         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 321                                          PAGE_SIZE);
 322
 323         if (ramdisk_here == -1ULL)
 324                 panic("Cannot find place for new RAMDISK of size %lld\n",
 325                          ramdisk_size);
 326
 327         /* Note: this includes all the lowmem currently occupied by
 328            the initrd, we rely on that fact to keep the data intact. */
 329         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 330                          "NEW RAMDISK");
 331         initrd_start = ramdisk_here + PAGE_OFFSET;
 332         initrd_end   = initrd_start + ramdisk_size;
 333         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 334                          ramdisk_here, ramdisk_here + ramdisk_size);
 335
 336         q = (char *)initrd_start;
 337
 338         /* Copy any lowmem portion of the initrd */
 339         if (ramdisk_image < end_of_lowmem) {
 340                 clen = end_of_lowmem - ramdisk_image;
 341                 p = (char *)__va(ramdisk_image);
 342                 memcpy(q, p, clen);
 343                 q += clen;
 344                 ramdisk_image += clen;
 345                 ramdisk_size  -= clen;
 346         }
 347
 348         /* Copy the highmem portion of the initrd */
 349         while (ramdisk_size) {
 350                 slop = ramdisk_image & ~PAGE_MASK;
 351                 clen = ramdisk_size;
 352                 if (clen > MAX_MAP_CHUNK-slop)
 353                         clen = MAX_MAP_CHUNK-slop;
 354                 mapaddr = ramdisk_image & PAGE_MASK;
 355                 p = early_memremap(mapaddr, clen+slop);
 356                 memcpy(q, p+slop, clen);
 357                 early_iounmap(p, clen+slop);
 358                 q += clen;
 359                 ramdisk_image += clen;
 360                 ramdisk_size  -= clen;
 361         }
 362         /* high pages is not converted by early_res_to_bootmem */
 363         ramdisk_image = boot_params.hdr.ramdisk_image;
 364         ramdisk_size  = boot_params.hdr.ramdisk_size;
 365         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 366                 " %08llx - %08llx\n",
 367                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 368                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 369 }
 370 #endif
 371
 372 static void __init reserve_initrd(void)
 373 {
 374         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 375         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 376         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 377         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 378
 379         if (!boot_params.hdr.type_of_loader ||
 380             !ramdisk_image || !ramdisk_size)
 381                 return;         /* No initrd provided by bootloader */
 382
 383         initrd_start = 0;
 384
 385         if (ramdisk_size >= (end_of_lowmem>>1)) {
 386                 free_early(ramdisk_image, ramdisk_end);
 387                 printk(KERN_ERR "initrd too large to handle, "
 388                        "disabling initrd\n");
 389                 return;
 390         }
 391
 392         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 393                         ramdisk_end);
 394
 395
 396         if (ramdisk_end <= end_of_lowmem) {
 397                 /* All in lowmem, easy case */
 398                 /*
 399                  * don't need to reserve again, already reserved early
 400                  * in i386_start_kernel
 401                  */
 402                 initrd_start = ramdisk_image + PAGE_OFFSET;
 403                 initrd_end = initrd_start + ramdisk_size;
 404                 return;
 405         }
 406
 407 #ifdef CONFIG_X86_32
 408         relocate_initrd();
 409 #else
 410         printk(KERN_ERR "initrd extends beyond end of memory "
 411                "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
 412                ramdisk_end, end_of_lowmem);
 413         initrd_start = 0;
 414 #endif
 415         free_early(ramdisk_image, ramdisk_end);
 416 }
 417 #else
 418 static void __init reserve_initrd(void)
 419 {
 420 }
 421 #endif /* CONFIG_BLK_DEV_INITRD */
 422
 423 static void __init parse_setup_data(void)
 424 {
 425         struct setup_data *data;
 426         u64 pa_data;
 427
 428         if (boot_params.hdr.version < 0x0209)
 429                 return;
 430         pa_data = boot_params.hdr.setup_data;
 431         while (pa_data) {
 432                 data = early_memremap(pa_data, PAGE_SIZE);
 433                 switch (data->type) {
 434                 case SETUP_E820_EXT:
 435                         parse_e820_ext(data, pa_data);
 436                         break;
 437                 default:
 438                         break;
 439                 }
 440                 pa_data = data->next;
 441                 early_iounmap(data, PAGE_SIZE);
 442         }
 443 }
 444
 445 static void __init e820_reserve_setup_data(void)
 446 {
 447         struct setup_data *data;
 448         u64 pa_data;
 449         int found = 0;
 450
 451         if (boot_params.hdr.version < 0x0209)
 452                 return;
 453         pa_data = boot_params.hdr.setup_data;
 454         while (pa_data) {
 455                 data = early_memremap(pa_data, sizeof(*data));
 456                 e820_update_range(pa_data, sizeof(*data)+data->len,
 457                          E820_RAM, E820_RESERVED_KERN);
 458                 found = 1;
 459                 pa_data = data->next;
 460                 early_iounmap(data, sizeof(*data));
 461         }
 462         if (!found)
 463                 return;
 464
 465         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 466         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 467         printk(KERN_INFO "extended physical RAM map:\n");
 468         e820_print_map("reserve setup_data");
 469 }
 470
 471 static void __init reserve_early_setup_data(void)
 472 {
 473         struct setup_data *data;
 474         u64 pa_data;
 475         char buf[32];
 476
 477         if (boot_params.hdr.version < 0x0209)
 478                 return;
 479         pa_data = boot_params.hdr.setup_data;
 480         while (pa_data) {
 481                 data = early_memremap(pa_data, sizeof(*data));
 482                 sprintf(buf, "setup data %x", data->type);
 483                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 484                 pa_data = data->next;
 485                 early_iounmap(data, sizeof(*data));
 486         }
 487 }
 488
 489 /*
 490  * --------- Crashkernel reservation ------------------------------
 491  */
 492
 493 #ifdef CONFIG_KEXEC
 494
 495 /**
 496  * Reserve @size bytes of crashkernel memory at any suitable offset.
 497  *
 498  * @size: Size of the crashkernel memory to reserve.
 499  * Returns the base address on success, and -1ULL on failure.
 500  */
 501 static
 502 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 503 {
 504         const unsigned long long alignment = 16<<20;    /* 16M */
 505         unsigned long long start = 0LL;
 506
 507         while (1) {
 508                 int ret;
 509
 510                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 511                 if (start == -1ULL)
 512                         return start;
 513
 514                 /* try to reserve it */
 515                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 516                 if (ret >= 0)
 517                         return start;
 518
 519                 start += alignment;
 520         }
 521 }
 522
 523 static inline unsigned long long get_total_mem(void)
 524 {
 525         unsigned long long total;
 526
 527         total = max_low_pfn - min_low_pfn;
 528 #ifdef CONFIG_HIGHMEM
 529         total += highend_pfn - highstart_pfn;
 530 #endif
 531
 532         return total << PAGE_SHIFT;
 533 }
 534
 535 static void __init reserve_crashkernel(void)
 536 {
 537         unsigned long long total_mem;
 538         unsigned long long crash_size, crash_base;
 539         int ret;
 540
 541         total_mem = get_total_mem();
 542
 543         ret = parse_crashkernel(boot_command_line, total_mem,
 544                         &crash_size, &crash_base);
 545         if (ret != 0 || crash_size <= 0)
 546                 return;
 547
 548         /* 0 means: find the address automatically */
 549         if (crash_base <= 0) {
 550                 crash_base = find_and_reserve_crashkernel(crash_size);
 551                 if (crash_base == -1ULL) {
 552                         pr_info("crashkernel reservation failed. "
 553                                 "No suitable area found.\n");
 554                         return;
 555                 }
 556         } else {
 557                 ret = reserve_bootmem_generic(crash_base, crash_size,
 558                                         BOOTMEM_EXCLUSIVE);
 559                 if (ret < 0) {
 560                         pr_info("crashkernel reservation failed - "
 561                                 "memory is in use\n");
 562                         return;
 563                 }
 564         }
 565
 566         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 567                         "for crashkernel (System RAM: %ldMB)\n",
 568                         (unsigned long)(crash_size >> 20),
 569                         (unsigned long)(crash_base >> 20),
 570                         (unsigned long)(total_mem >> 20));
 571
 572         crashk_res.start = crash_base;
 573         crashk_res.end   = crash_base + crash_size - 1;
 574         insert_resource(&iomem_resource, &crashk_res);
 575 }
 576 #else
 577 static void __init reserve_crashkernel(void)
 578 {
 579 }
 580 #endif
 581
 582 static struct resource standard_io_resources[] = {
 583         { .name = "dma1", .start = 0x00, .end = 0x1f,
 584                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 585         { .name = "pic1", .start = 0x20, .end = 0x21,
 586                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 587         { .name = "timer0", .start = 0x40, .end = 0x43,
 588                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 589         { .name = "timer1", .start = 0x50, .end = 0x53,
 590                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 591         { .name = "keyboard", .start = 0x60, .end = 0x60,
 592                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 593         { .name = "keyboard", .start = 0x64, .end = 0x64,
 594                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 595         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 596                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 597         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 598                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 599         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 600                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 601         { .name = "fpu", .start = 0xf0, .end = 0xff,
 602                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 603 };
 604
 605 static void __init reserve_standard_io_resources(void)
 606 {
 607         int i;
 608
 609         /* request I/O space for devices used on all i[345]86 PCs */
 610         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 611                 request_resource(&ioport_resource, &standard_io_resources[i]);
 612
 613 }
 614
 615 /*
 616  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 617  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 618  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 619  */
 620
 621 #ifdef CONFIG_CRASH_DUMP
 622 /* elfcorehdr= specifies the location of elf core header
 623  * stored by the crashed kernel. This option will be passed
 624  * by kexec loader to the capture kernel.
 625  */
 626 static int __init setup_elfcorehdr(char *arg)
 627 {
 628         char *end;
 629         if (!arg)
 630                 return -EINVAL;
 631         elfcorehdr_addr = memparse(arg, &end);
 632         return end > arg ? 0 : -EINVAL;
 633 }
 634 early_param("elfcorehdr", setup_elfcorehdr);
 635 #endif
 636
 637 static struct x86_quirks default_x86_quirks __initdata;
 638
 639 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 640
 641 #ifdef CONFIG_X86_RESERVE_LOW_64K
 642 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 643 {
 644         printk(KERN_NOTICE
 645                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 646                 d->ident);
 647
 648         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 649         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 650
 651         return 0;
 652 }
 653 #endif
 654
 655 /* List of systems that have known low memory corruption BIOS problems */
 656 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 657 #ifdef CONFIG_X86_RESERVE_LOW_64K
 658         {
 659                 .callback = dmi_low_memory_corruption,
 660                 .ident = "AMI BIOS",
 661                 .matches = {
 662                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 663                 },
 664         },
 665         {
 666                 .callback = dmi_low_memory_corruption,
 667                 .ident = "Phoenix BIOS",
 668                 .matches = {
 669                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 670                 },
 671         },
 672 #endif
 673         {}
 674 };
 675
 676 /*
 677  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 678  * passed the efi memmap, systab, etc., so we should use these data structures
 679  * for initialization.  Note, the efi init code path is determined by the
 680  * global efi_enabled. This allows the same kernel image to be used on existing
 681  * systems (with a traditional BIOS) as well as on EFI systems.
 682  */
 683 /*
 684  * setup_arch - architecture-specific boot-time initializations
 685  *
 686  * Note: On x86_64, fixmaps are ready for use even before this is called.
 687  */
 688
 689 void __init setup_arch(char **cmdline_p)
 690 {
 691 #ifdef CONFIG_X86_32
 692         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 693         visws_early_detect();
 694 #else
 695         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 696 #endif
 697
 698         /* VMI may relocate the fixmap; do this before touching ioremap area */
 699         vmi_init();
 700
 701         early_cpu_init();
 702         early_ioremap_init();
 703
 704         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 705         screen_info = boot_params.screen_info;
 706         edid_info = boot_params.edid_info;
 707 #ifdef CONFIG_X86_32
 708         apm_info.bios = boot_params.apm_bios_info;
 709         ist_info = boot_params.ist_info;
 710         if (boot_params.sys_desc_table.length != 0) {
 711                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 712                 machine_id = boot_params.sys_desc_table.table[0];
 713                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 714                 BIOS_revision = boot_params.sys_desc_table.table[2];
 715         }
 716 #endif
 717         saved_video_mode = boot_params.hdr.vid_mode;
 718         bootloader_type = boot_params.hdr.type_of_loader;
 719
 720 #ifdef CONFIG_BLK_DEV_RAM
 721         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 722         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 723         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 724 #endif
 725 #ifdef CONFIG_EFI
 726         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 727 #ifdef CONFIG_X86_32
 728                      "EL32",
 729 #else
 730                      "EL64",
 731 #endif
 732          4)) {
 733                 efi_enabled = 1;
 734                 efi_reserve_early();
 735         }
 736 #endif
 737
 738         ARCH_SETUP
 739
 740         setup_memory_map();
 741         parse_setup_data();
 742         /* update the e820_saved too */
 743         e820_reserve_setup_data();
 744
 745         copy_edd();
 746
 747         if (!boot_params.hdr.root_flags)
 748                 root_mountflags &= ~MS_RDONLY;
 749         init_mm.start_code = (unsigned long) _text;
 750         init_mm.end_code = (unsigned long) _etext;
 751         init_mm.end_data = (unsigned long) _edata;
 752         init_mm.brk = _brk_end;
 753
 754         code_resource.start = virt_to_phys(_text);
 755         code_resource.end = virt_to_phys(_etext)-1;
 756         data_resource.start = virt_to_phys(_etext);
 757         data_resource.end = virt_to_phys(_edata)-1;
 758         bss_resource.start = virt_to_phys(&__bss_start);
 759         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 760
 761 #ifdef CONFIG_CMDLINE_BOOL
 762 #ifdef CONFIG_CMDLINE_OVERRIDE
 763         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 764 #else
 765         if (builtin_cmdline[0]) {
 766                 /* append boot loader cmdline to builtin */
 767                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 768                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 769                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 770         }
 771 #endif
 772 #endif
 773
 774         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 775         *cmdline_p = command_line;
 776
 777         parse_early_param();
 778
 779 #ifdef CONFIG_X86_64
 780         check_efer();
 781 #endif
 782
 783         /* Must be before kernel pagetables are setup */
 784         vmi_activate();
 785
 786         /* after early param, so could get panic from serial */
 787         reserve_early_setup_data();
 788
 789         if (acpi_mps_check()) {
 790 #ifdef CONFIG_X86_LOCAL_APIC
 791                 disable_apic = 1;
 792 #endif
 793                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 794         }
 795
 796 #ifdef CONFIG_PCI
 797         if (pci_early_dump_regs)
 798                 early_dump_pci_devices();
 799 #endif
 800
 801         finish_e820_parsing();
 802
 803         if (efi_enabled)
 804                 efi_init();
 805
 806         dmi_scan_machine();
 807
 808         dmi_check_system(bad_bios_dmi_table);
 809
 810         /*
 811          * VMware detection requires dmi to be available, so this
 812          * needs to be done after dmi_scan_machine, for the BP.
 813          */
 814         init_hypervisor(&boot_cpu_data);
 815
 816 #ifdef CONFIG_X86_32
 817         probe_roms();
 818 #endif
 819
 820         /* after parse_early_param, so could debug it */
 821         insert_resource(&iomem_resource, &code_resource);
 822         insert_resource(&iomem_resource, &data_resource);
 823         insert_resource(&iomem_resource, &bss_resource);
 824
 825
 826 #ifdef CONFIG_X86_32
 827         if (ppro_with_ram_bug()) {
 828                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 829                                   E820_RESERVED);
 830                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 831                 printk(KERN_INFO "fixed physical RAM map:\n");
 832                 e820_print_map("bad_ppro");
 833         }
 834 #else
 835         early_gart_iommu_check();
 836 #endif
 837
 838         /*
 839          * partially used pages are not usable - thus
 840          * we are rounding upwards:
 841          */
 842         max_pfn = e820_end_of_ram_pfn();
 843
 844         /* preallocate 4k for mptable mpc */
 845         early_reserve_e820_mpc_new();
 846         /* update e820 for memory not covered by WB MTRRs */
 847         mtrr_bp_init();
 848         if (mtrr_trim_uncached_memory(max_pfn))
 849                 max_pfn = e820_end_of_ram_pfn();
 850
 851 #ifdef CONFIG_X86_32
 852         /* max_low_pfn get updated here */
 853         find_low_pfn_range();
 854 #else
 855         num_physpages = max_pfn;
 856
 857         check_x2apic();
 858
 859         /* How many end-of-memory variables you have, grandma! */
 860         /* need this before calling reserve_initrd */
 861         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 862                 max_low_pfn = e820_end_of_low_ram_pfn();
 863         else
 864                 max_low_pfn = max_pfn;
 865
 866         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 867 #endif
 868
 869 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 870         setup_bios_corruption_check();
 871 #endif
 872
 873         reserve_brk();
 874
 875         /* max_pfn_mapped is updated here */
 876         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 877         max_pfn_mapped = max_low_pfn_mapped;
 878
 879 #ifdef CONFIG_X86_64
 880         if (max_pfn > max_low_pfn) {
 881                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 882                                                      max_pfn<<PAGE_SHIFT);
 883                 /* can we preseve max_low_pfn ?*/
 884                 max_low_pfn = max_pfn;
 885         }
 886 #endif
 887
 888         /*
 889          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 890          */
 891
 892 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 893         if (init_ohci1394_dma_early)
 894                 init_ohci1394_dma_on_all_controllers();
 895 #endif
 896
 897         reserve_initrd();
 898
 899         vsmp_init();
 900
 901         io_delay_init();
 902
 903         /*
 904          * Parse the ACPI tables for possible boot-time SMP configuration.
 905          */
 906         acpi_boot_table_init();
 907
 908         early_acpi_boot_init();
 909
 910 #ifdef CONFIG_ACPI_NUMA
 911         /*
 912          * Parse SRAT to discover nodes.
 913          */
 914         acpi_numa_init();
 915 #endif
 916
 917         initmem_init(0, max_pfn);
 918
 919 #ifdef CONFIG_ACPI_SLEEP
 920         /*
 921          * Reserve low memory region for sleep support.
 922          */
 923         acpi_reserve_bootmem();
 924 #endif
 925         /*
 926          * Find and reserve possible boot-time SMP configuration:
 927          */
 928         find_smp_config();
 929
 930         reserve_crashkernel();
 931
 932 #ifdef CONFIG_X86_64
 933         /*
 934          * dma32_reserve_bootmem() allocates bootmem which may conflict
 935          * with the crashkernel command line, so do that after
 936          * reserve_crashkernel()
 937          */
 938         dma32_reserve_bootmem();
 939 #endif
 940
 941         reserve_ibft_region();
 942
 943 #ifdef CONFIG_KVM_CLOCK
 944         kvmclock_init();
 945 #endif
 946
 947         paravirt_pagetable_setup_start(swapper_pg_dir);
 948         paging_init();
 949         paravirt_pagetable_setup_done(swapper_pg_dir);
 950         paravirt_post_allocator_init();
 951
 952 #ifdef CONFIG_X86_64
 953         map_vsyscall();
 954 #endif
 955
 956         generic_apic_probe();
 957
 958         early_quirks();
 959
 960         /*
 961          * Read APIC and some other early information from ACPI tables.
 962          */
 963         acpi_boot_init();
 964
 965 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 966         /*
 967          * get boot-time SMP configuration:
 968          */
 969         if (smp_found_config)
 970                 get_smp_config();
 971 #endif
 972
 973         prefill_possible_map();
 974
 975 #ifdef CONFIG_X86_64
 976         init_cpu_to_node();
 977 #endif
 978
 979         init_apic_mappings();
 980         ioapic_init_mappings();
 981
 982         /* need to wait for io_apic is mapped */
 983         probe_nr_irqs_gsi();
 984
 985         kvm_guest_init();
 986
 987         e820_reserve_resources();
 988         e820_mark_nosave_regions(max_low_pfn);
 989
 990 #ifdef CONFIG_X86_32
 991         request_resource(&iomem_resource, &video_ram_resource);
 992 #endif
 993         reserve_standard_io_resources();
 994
 995         e820_setup_gap();
 996
 997 #ifdef CONFIG_VT
 998 #if defined(CONFIG_VGA_CONSOLE)
 999         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1000                 conswitchp = &vga_con;
1001 #elif defined(CONFIG_DUMMY_CONSOLE)
1002         conswitchp = &dummy_con;
1003 #endif
1004 #endif
1005 }
1006
1007 #ifdef CONFIG_X86_32
1008
1009 /**
1010  * x86_quirk_pre_intr_init - initialisation prior to setting up interrupt vectors
1011  *
1012  * Description:
1013  *      Perform any necessary interrupt initialisation prior to setting up
1014  *      the "ordinary" interrupt call gates.  For legacy reasons, the ISA
1015  *      interrupts should be initialised here if the machine emulates a PC
1016  *      in any way.
1017  **/
1018 void __init x86_quirk_pre_intr_init(void)
1019 {
1020         if (x86_quirks->arch_pre_intr_init) {
1021                 if (x86_quirks->arch_pre_intr_init())
1022                         return;
1023         }
1024         init_ISA_irqs();
1025 }
1026
1027 /**
1028  * x86_quirk_intr_init - post gate setup interrupt initialisation
1029  *
1030  * Description:
1031  *      Fill in any interrupts that may have been left out by the general
1032  *      init_IRQ() routine.  interrupts having to do with the machine rather
1033  *      than the devices on the I/O bus (like APIC interrupts in intel MP
1034  *      systems) are started here.
1035  **/
1036 void __init x86_quirk_intr_init(void)
1037 {
1038         if (x86_quirks->arch_intr_init) {
1039                 if (x86_quirks->arch_intr_init())
1040                         return;
1041         }
1042 }
1043
1044 /**
1045  * x86_quirk_trap_init - initialise system specific traps
1046  *
1047  * Description:
1048  *      Called as the final act of trap_init().  Used in VISWS to initialise
1049  *      the various board specific APIC traps.
1050  **/
1051 void __init x86_quirk_trap_init(void)
1052 {
1053         if (x86_quirks->arch_trap_init) {
1054                 if (x86_quirks->arch_trap_init())
1055                         return;
1056         }
1057 }
1058
1059 static struct irqaction irq0  = {
1060         .handler = timer_interrupt,
1061         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1062         .mask = CPU_MASK_NONE,
1063         .name = "timer"
1064 };
1065
1066 /**
1067  * x86_quirk_pre_time_init - do any specific initialisations before.
1068  *
1069  **/
1070 void __init x86_quirk_pre_time_init(void)
1071 {
1072         if (x86_quirks->arch_pre_time_init)
1073                 x86_quirks->arch_pre_time_init();
1074 }
1075
1076 /**
1077  * x86_quirk_time_init - do any specific initialisations for the system timer.
1078  *
1079  * Description:
1080  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1081  *      in irq_vectors.h:TIMER_IRQ
1082  **/
1083 void __init x86_quirk_time_init(void)
1084 {
1085         if (x86_quirks->arch_time_init) {
1086                 /*
1087                  * A nonzero return code does not mean failure, it means
1088                  * that the architecture quirk does not want any
1089                  * generic (timer) setup to be performed after this:
1090                  */
1091                 if (x86_quirks->arch_time_init())
1092                         return;
1093         }
1094
1095         irq0.mask = cpumask_of_cpu(0);
1096         setup_irq(0, &irq0);
1097 }
1098 #endif /* CONFIG_X86_32 */