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 #ifdef CONFIG_X86_64
 118 int default_cpu_present_to_apicid(int mps_cpu)
 119 {
 120         return __default_cpu_present_to_apicid(mps_cpu);
 121 }
 122
 123 int default_check_phys_apicid_present(int boot_cpu_physical_apicid)
 124 {
 125         return __default_check_phys_apicid_present(boot_cpu_physical_apicid);
 126 }
 127 #endif
 128
 129 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 130 struct boot_params __initdata boot_params;
 131 #else
 132 struct boot_params boot_params;
 133 #endif
 134
 135 /*
 136  * Machine setup..
 137  */
 138 static struct resource data_resource = {
 139         .name   = "Kernel data",
 140         .start  = 0,
 141         .end    = 0,
 142         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 143 };
 144
 145 static struct resource code_resource = {
 146         .name   = "Kernel code",
 147         .start  = 0,
 148         .end    = 0,
 149         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 150 };
 151
 152 static struct resource bss_resource = {
 153         .name   = "Kernel bss",
 154         .start  = 0,
 155         .end    = 0,
 156         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 157 };
 158
 159
 160 #ifdef CONFIG_X86_32
 161 /* This value is set up by the early boot code to point to the value
 162    immediately after the boot time page tables.  It contains a *physical*
 163    address, and must not be in the .bss segment! */
 164 unsigned long init_pg_tables_start __initdata = ~0UL;
 165 unsigned long init_pg_tables_end __initdata = ~0UL;
 166
 167 static struct resource video_ram_resource = {
 168         .name   = "Video RAM area",
 169         .start  = 0xa0000,
 170         .end    = 0xbffff,
 171         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 172 };
 173
 174 /* cpu data as detected by the assembly code in head.S */
 175 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 176 /* common cpu data for all cpus */
 177 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 178 EXPORT_SYMBOL(boot_cpu_data);
 179 static void set_mca_bus(int x)
 180 {
 181 #ifdef CONFIG_MCA
 182         MCA_bus = x;
 183 #endif
 184 }
 185
 186 unsigned int def_to_bigsmp;
 187
 188 /* for MCA, but anyone else can use it if they want */
 189 unsigned int machine_id;
 190 unsigned int machine_submodel_id;
 191 unsigned int BIOS_revision;
 192
 193 struct apm_info apm_info;
 194 EXPORT_SYMBOL(apm_info);
 195
 196 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 197         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 198 struct ist_info ist_info;
 199 EXPORT_SYMBOL(ist_info);
 200 #else
 201 struct ist_info ist_info;
 202 #endif
 203
 204 #else
 205 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 206         .x86_phys_bits = MAX_PHYSMEM_BITS,
 207 };
 208 EXPORT_SYMBOL(boot_cpu_data);
 209 #endif
 210
 211
 212 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 213 unsigned long mmu_cr4_features;
 214 #else
 215 unsigned long mmu_cr4_features = X86_CR4_PAE;
 216 #endif
 217
 218 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 219 int bootloader_type;
 220
 221 /*
 222  * Early DMI memory
 223  */
 224 int dmi_alloc_index;
 225 char dmi_alloc_data[DMI_MAX_DATA];
 226
 227 /*
 228  * Setup options
 229  */
 230 struct screen_info screen_info;
 231 EXPORT_SYMBOL(screen_info);
 232 struct edid_info edid_info;
 233 EXPORT_SYMBOL_GPL(edid_info);
 234
 235 extern int root_mountflags;
 236
 237 unsigned long saved_video_mode;
 238
 239 #define RAMDISK_IMAGE_START_MASK        0x07FF
 240 #define RAMDISK_PROMPT_FLAG             0x8000
 241 #define RAMDISK_LOAD_FLAG               0x4000
 242
 243 static char __initdata command_line[COMMAND_LINE_SIZE];
 244 #ifdef CONFIG_CMDLINE_BOOL
 245 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 246 #endif
 247
 248 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 249 struct edd edd;
 250 #ifdef CONFIG_EDD_MODULE
 251 EXPORT_SYMBOL(edd);
 252 #endif
 253 /**
 254  * copy_edd() - Copy the BIOS EDD information
 255  *              from boot_params into a safe place.
 256  *
 257  */
 258 static inline void copy_edd(void)
 259 {
 260      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 261             sizeof(edd.mbr_signature));
 262      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 263      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 264      edd.edd_info_nr = boot_params.eddbuf_entries;
 265 }
 266 #else
 267 static inline void copy_edd(void)
 268 {
 269 }
 270 #endif
 271
 272 #ifdef CONFIG_BLK_DEV_INITRD
 273
 274 #ifdef CONFIG_X86_32
 275
 276 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 277 static void __init relocate_initrd(void)
 278 {
 279
 280         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 281         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 282         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 283         u64 ramdisk_here;
 284         unsigned long slop, clen, mapaddr;
 285         char *p, *q;
 286
 287         /* We need to move the initrd down into lowmem */
 288         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 289                                          PAGE_SIZE);
 290
 291         if (ramdisk_here == -1ULL)
 292                 panic("Cannot find place for new RAMDISK of size %lld\n",
 293                          ramdisk_size);
 294
 295         /* Note: this includes all the lowmem currently occupied by
 296            the initrd, we rely on that fact to keep the data intact. */
 297         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 298                          "NEW RAMDISK");
 299         initrd_start = ramdisk_here + PAGE_OFFSET;
 300         initrd_end   = initrd_start + ramdisk_size;
 301         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 302                          ramdisk_here, ramdisk_here + ramdisk_size);
 303
 304         q = (char *)initrd_start;
 305
 306         /* Copy any lowmem portion of the initrd */
 307         if (ramdisk_image < end_of_lowmem) {
 308                 clen = end_of_lowmem - ramdisk_image;
 309                 p = (char *)__va(ramdisk_image);
 310                 memcpy(q, p, clen);
 311                 q += clen;
 312                 ramdisk_image += clen;
 313                 ramdisk_size  -= clen;
 314         }
 315
 316         /* Copy the highmem portion of the initrd */
 317         while (ramdisk_size) {
 318                 slop = ramdisk_image & ~PAGE_MASK;
 319                 clen = ramdisk_size;
 320                 if (clen > MAX_MAP_CHUNK-slop)
 321                         clen = MAX_MAP_CHUNK-slop;
 322                 mapaddr = ramdisk_image & PAGE_MASK;
 323                 p = early_memremap(mapaddr, clen+slop);
 324                 memcpy(q, p+slop, clen);
 325                 early_iounmap(p, clen+slop);
 326                 q += clen;
 327                 ramdisk_image += clen;
 328                 ramdisk_size  -= clen;
 329         }
 330         /* high pages is not converted by early_res_to_bootmem */
 331         ramdisk_image = boot_params.hdr.ramdisk_image;
 332         ramdisk_size  = boot_params.hdr.ramdisk_size;
 333         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 334                 " %08llx - %08llx\n",
 335                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 336                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 337 }
 338 #endif
 339
 340 static void __init reserve_initrd(void)
 341 {
 342         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 343         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 344         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 345         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 346
 347         if (!boot_params.hdr.type_of_loader ||
 348             !ramdisk_image || !ramdisk_size)
 349                 return;         /* No initrd provided by bootloader */
 350
 351         initrd_start = 0;
 352
 353         if (ramdisk_size >= (end_of_lowmem>>1)) {
 354                 free_early(ramdisk_image, ramdisk_end);
 355                 printk(KERN_ERR "initrd too large to handle, "
 356                        "disabling initrd\n");
 357                 return;
 358         }
 359
 360         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 361                         ramdisk_end);
 362
 363
 364         if (ramdisk_end <= end_of_lowmem) {
 365                 /* All in lowmem, easy case */
 366                 /*
 367                  * don't need to reserve again, already reserved early
 368                  * in i386_start_kernel
 369                  */
 370                 initrd_start = ramdisk_image + PAGE_OFFSET;
 371                 initrd_end = initrd_start + ramdisk_size;
 372                 return;
 373         }
 374
 375 #ifdef CONFIG_X86_32
 376         relocate_initrd();
 377 #else
 378         printk(KERN_ERR "initrd extends beyond end of memory "
 379                "(0x%08llx > 0x%08llx)\ndisabling initrd\n",
 380                ramdisk_end, end_of_lowmem);
 381         initrd_start = 0;
 382 #endif
 383         free_early(ramdisk_image, ramdisk_end);
 384 }
 385 #else
 386 static void __init reserve_initrd(void)
 387 {
 388 }
 389 #endif /* CONFIG_BLK_DEV_INITRD */
 390
 391 static void __init parse_setup_data(void)
 392 {
 393         struct setup_data *data;
 394         u64 pa_data;
 395
 396         if (boot_params.hdr.version < 0x0209)
 397                 return;
 398         pa_data = boot_params.hdr.setup_data;
 399         while (pa_data) {
 400                 data = early_memremap(pa_data, PAGE_SIZE);
 401                 switch (data->type) {
 402                 case SETUP_E820_EXT:
 403                         parse_e820_ext(data, pa_data);
 404                         break;
 405                 default:
 406                         break;
 407                 }
 408                 pa_data = data->next;
 409                 early_iounmap(data, PAGE_SIZE);
 410         }
 411 }
 412
 413 static void __init e820_reserve_setup_data(void)
 414 {
 415         struct setup_data *data;
 416         u64 pa_data;
 417         int found = 0;
 418
 419         if (boot_params.hdr.version < 0x0209)
 420                 return;
 421         pa_data = boot_params.hdr.setup_data;
 422         while (pa_data) {
 423                 data = early_memremap(pa_data, sizeof(*data));
 424                 e820_update_range(pa_data, sizeof(*data)+data->len,
 425                          E820_RAM, E820_RESERVED_KERN);
 426                 found = 1;
 427                 pa_data = data->next;
 428                 early_iounmap(data, sizeof(*data));
 429         }
 430         if (!found)
 431                 return;
 432
 433         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 434         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 435         printk(KERN_INFO "extended physical RAM map:\n");
 436         e820_print_map("reserve setup_data");
 437 }
 438
 439 static void __init reserve_early_setup_data(void)
 440 {
 441         struct setup_data *data;
 442         u64 pa_data;
 443         char buf[32];
 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                 sprintf(buf, "setup data %x", data->type);
 451                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 452                 pa_data = data->next;
 453                 early_iounmap(data, sizeof(*data));
 454         }
 455 }
 456
 457 /*
 458  * --------- Crashkernel reservation ------------------------------
 459  */
 460
 461 #ifdef CONFIG_KEXEC
 462
 463 /**
 464  * Reserve @size bytes of crashkernel memory at any suitable offset.
 465  *
 466  * @size: Size of the crashkernel memory to reserve.
 467  * Returns the base address on success, and -1ULL on failure.
 468  */
 469 static
 470 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 471 {
 472         const unsigned long long alignment = 16<<20;    /* 16M */
 473         unsigned long long start = 0LL;
 474
 475         while (1) {
 476                 int ret;
 477
 478                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 479                 if (start == -1ULL)
 480                         return start;
 481
 482                 /* try to reserve it */
 483                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 484                 if (ret >= 0)
 485                         return start;
 486
 487                 start += alignment;
 488         }
 489 }
 490
 491 static inline unsigned long long get_total_mem(void)
 492 {
 493         unsigned long long total;
 494
 495         total = max_low_pfn - min_low_pfn;
 496 #ifdef CONFIG_HIGHMEM
 497         total += highend_pfn - highstart_pfn;
 498 #endif
 499
 500         return total << PAGE_SHIFT;
 501 }
 502
 503 static void __init reserve_crashkernel(void)
 504 {
 505         unsigned long long total_mem;
 506         unsigned long long crash_size, crash_base;
 507         int ret;
 508
 509         total_mem = get_total_mem();
 510
 511         ret = parse_crashkernel(boot_command_line, total_mem,
 512                         &crash_size, &crash_base);
 513         if (ret != 0 || crash_size <= 0)
 514                 return;
 515
 516         /* 0 means: find the address automatically */
 517         if (crash_base <= 0) {
 518                 crash_base = find_and_reserve_crashkernel(crash_size);
 519                 if (crash_base == -1ULL) {
 520                         pr_info("crashkernel reservation failed. "
 521                                 "No suitable area found.\n");
 522                         return;
 523                 }
 524         } else {
 525                 ret = reserve_bootmem_generic(crash_base, crash_size,
 526                                         BOOTMEM_EXCLUSIVE);
 527                 if (ret < 0) {
 528                         pr_info("crashkernel reservation failed - "
 529                                 "memory is in use\n");
 530                         return;
 531                 }
 532         }
 533
 534         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 535                         "for crashkernel (System RAM: %ldMB)\n",
 536                         (unsigned long)(crash_size >> 20),
 537                         (unsigned long)(crash_base >> 20),
 538                         (unsigned long)(total_mem >> 20));
 539
 540         crashk_res.start = crash_base;
 541         crashk_res.end   = crash_base + crash_size - 1;
 542         insert_resource(&iomem_resource, &crashk_res);
 543 }
 544 #else
 545 static void __init reserve_crashkernel(void)
 546 {
 547 }
 548 #endif
 549
 550 static struct resource standard_io_resources[] = {
 551         { .name = "dma1", .start = 0x00, .end = 0x1f,
 552                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 553         { .name = "pic1", .start = 0x20, .end = 0x21,
 554                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 555         { .name = "timer0", .start = 0x40, .end = 0x43,
 556                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 557         { .name = "timer1", .start = 0x50, .end = 0x53,
 558                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 559         { .name = "keyboard", .start = 0x60, .end = 0x60,
 560                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 561         { .name = "keyboard", .start = 0x64, .end = 0x64,
 562                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 563         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 564                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 565         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 566                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 567         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 568                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 569         { .name = "fpu", .start = 0xf0, .end = 0xff,
 570                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 571 };
 572
 573 static void __init reserve_standard_io_resources(void)
 574 {
 575         int i;
 576
 577         /* request I/O space for devices used on all i[345]86 PCs */
 578         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 579                 request_resource(&ioport_resource, &standard_io_resources[i]);
 580
 581 }
 582
 583 /*
 584  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 585  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 586  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 587  */
 588
 589 #ifdef CONFIG_CRASH_DUMP
 590 /* elfcorehdr= specifies the location of elf core header
 591  * stored by the crashed kernel. This option will be passed
 592  * by kexec loader to the capture kernel.
 593  */
 594 static int __init setup_elfcorehdr(char *arg)
 595 {
 596         char *end;
 597         if (!arg)
 598                 return -EINVAL;
 599         elfcorehdr_addr = memparse(arg, &end);
 600         return end > arg ? 0 : -EINVAL;
 601 }
 602 early_param("elfcorehdr", setup_elfcorehdr);
 603 #endif
 604
 605 static struct x86_quirks default_x86_quirks __initdata;
 606
 607 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 608
 609 #ifdef CONFIG_X86_RESERVE_LOW_64K
 610 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 611 {
 612         printk(KERN_NOTICE
 613                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 614                 d->ident);
 615
 616         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 617         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 618
 619         return 0;
 620 }
 621 #endif
 622
 623 /* List of systems that have known low memory corruption BIOS problems */
 624 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 625 #ifdef CONFIG_X86_RESERVE_LOW_64K
 626         {
 627                 .callback = dmi_low_memory_corruption,
 628                 .ident = "AMI BIOS",
 629                 .matches = {
 630                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 631                 },
 632         },
 633         {
 634                 .callback = dmi_low_memory_corruption,
 635                 .ident = "Phoenix BIOS",
 636                 .matches = {
 637                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 638                 },
 639         },
 640 #endif
 641         {}
 642 };
 643
 644 /*
 645  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 646  * passed the efi memmap, systab, etc., so we should use these data structures
 647  * for initialization.  Note, the efi init code path is determined by the
 648  * global efi_enabled. This allows the same kernel image to be used on existing
 649  * systems (with a traditional BIOS) as well as on EFI systems.
 650  */
 651 /*
 652  * setup_arch - architecture-specific boot-time initializations
 653  *
 654  * Note: On x86_64, fixmaps are ready for use even before this is called.
 655  */
 656
 657 void __init setup_arch(char **cmdline_p)
 658 {
 659 #ifdef CONFIG_X86_32
 660         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 661         visws_early_detect();
 662 #else
 663         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 664 #endif
 665
 666         /* VMI may relocate the fixmap; do this before touching ioremap area */
 667         vmi_init();
 668
 669         early_cpu_init();
 670         early_ioremap_init();
 671
 672         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 673         screen_info = boot_params.screen_info;
 674         edid_info = boot_params.edid_info;
 675 #ifdef CONFIG_X86_32
 676         apm_info.bios = boot_params.apm_bios_info;
 677         ist_info = boot_params.ist_info;
 678         if (boot_params.sys_desc_table.length != 0) {
 679                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 680                 machine_id = boot_params.sys_desc_table.table[0];
 681                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 682                 BIOS_revision = boot_params.sys_desc_table.table[2];
 683         }
 684 #endif
 685         saved_video_mode = boot_params.hdr.vid_mode;
 686         bootloader_type = boot_params.hdr.type_of_loader;
 687
 688 #ifdef CONFIG_BLK_DEV_RAM
 689         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 690         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 691         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 692 #endif
 693 #ifdef CONFIG_EFI
 694         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 695 #ifdef CONFIG_X86_32
 696                      "EL32",
 697 #else
 698                      "EL64",
 699 #endif
 700          4)) {
 701                 efi_enabled = 1;
 702                 efi_reserve_early();
 703         }
 704 #endif
 705
 706         ARCH_SETUP
 707
 708         setup_memory_map();
 709         parse_setup_data();
 710         /* update the e820_saved too */
 711         e820_reserve_setup_data();
 712
 713         copy_edd();
 714
 715         if (!boot_params.hdr.root_flags)
 716                 root_mountflags &= ~MS_RDONLY;
 717         init_mm.start_code = (unsigned long) _text;
 718         init_mm.end_code = (unsigned long) _etext;
 719         init_mm.end_data = (unsigned long) _edata;
 720 #ifdef CONFIG_X86_32
 721         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 722 #else
 723         init_mm.brk = (unsigned long) &_end;
 724 #endif
 725
 726         code_resource.start = virt_to_phys(_text);
 727         code_resource.end = virt_to_phys(_etext)-1;
 728         data_resource.start = virt_to_phys(_etext);
 729         data_resource.end = virt_to_phys(_edata)-1;
 730         bss_resource.start = virt_to_phys(&__bss_start);
 731         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 732
 733 #ifdef CONFIG_CMDLINE_BOOL
 734 #ifdef CONFIG_CMDLINE_OVERRIDE
 735         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 736 #else
 737         if (builtin_cmdline[0]) {
 738                 /* append boot loader cmdline to builtin */
 739                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 740                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 741                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 742         }
 743 #endif
 744 #endif
 745
 746         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 747         *cmdline_p = command_line;
 748
 749         parse_early_param();
 750
 751 #ifdef CONFIG_X86_64
 752         check_efer();
 753 #endif
 754
 755         /* Must be before kernel pagetables are setup */
 756         vmi_activate();
 757
 758         /* after early param, so could get panic from serial */
 759         reserve_early_setup_data();
 760
 761         if (acpi_mps_check()) {
 762 #ifdef CONFIG_X86_LOCAL_APIC
 763                 disable_apic = 1;
 764 #endif
 765                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 766         }
 767
 768 #ifdef CONFIG_PCI
 769         if (pci_early_dump_regs)
 770                 early_dump_pci_devices();
 771 #endif
 772
 773         finish_e820_parsing();
 774
 775         if (efi_enabled)
 776                 efi_init();
 777
 778         dmi_scan_machine();
 779
 780         dmi_check_system(bad_bios_dmi_table);
 781
 782         /*
 783          * VMware detection requires dmi to be available, so this
 784          * needs to be done after dmi_scan_machine, for the BP.
 785          */
 786         init_hypervisor(&boot_cpu_data);
 787
 788 #ifdef CONFIG_X86_32
 789         probe_roms();
 790 #endif
 791
 792         /* after parse_early_param, so could debug it */
 793         insert_resource(&iomem_resource, &code_resource);
 794         insert_resource(&iomem_resource, &data_resource);
 795         insert_resource(&iomem_resource, &bss_resource);
 796
 797
 798 #ifdef CONFIG_X86_32
 799         if (ppro_with_ram_bug()) {
 800                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 801                                   E820_RESERVED);
 802                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 803                 printk(KERN_INFO "fixed physical RAM map:\n");
 804                 e820_print_map("bad_ppro");
 805         }
 806 #else
 807         early_gart_iommu_check();
 808 #endif
 809
 810         /*
 811          * partially used pages are not usable - thus
 812          * we are rounding upwards:
 813          */
 814         max_pfn = e820_end_of_ram_pfn();
 815
 816         /* preallocate 4k for mptable mpc */
 817         early_reserve_e820_mpc_new();
 818         /* update e820 for memory not covered by WB MTRRs */
 819         mtrr_bp_init();
 820         if (mtrr_trim_uncached_memory(max_pfn))
 821                 max_pfn = e820_end_of_ram_pfn();
 822
 823 #ifdef CONFIG_X86_32
 824         /* max_low_pfn get updated here */
 825         find_low_pfn_range();
 826 #else
 827         num_physpages = max_pfn;
 828
 829         check_x2apic();
 830
 831         /* How many end-of-memory variables you have, grandma! */
 832         /* need this before calling reserve_initrd */
 833         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 834                 max_low_pfn = e820_end_of_low_ram_pfn();
 835         else
 836                 max_low_pfn = max_pfn;
 837
 838         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 839 #endif
 840
 841 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 842         setup_bios_corruption_check();
 843 #endif
 844
 845         /* max_pfn_mapped is updated here */
 846         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 847         max_pfn_mapped = max_low_pfn_mapped;
 848
 849 #ifdef CONFIG_X86_64
 850         if (max_pfn > max_low_pfn) {
 851                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 852                                                      max_pfn<<PAGE_SHIFT);
 853                 /* can we preseve max_low_pfn ?*/
 854                 max_low_pfn = max_pfn;
 855         }
 856 #endif
 857
 858         /*
 859          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 860          */
 861
 862 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 863         if (init_ohci1394_dma_early)
 864                 init_ohci1394_dma_on_all_controllers();
 865 #endif
 866
 867         reserve_initrd();
 868
 869         vsmp_init();
 870
 871         io_delay_init();
 872
 873         /*
 874          * Parse the ACPI tables for possible boot-time SMP configuration.
 875          */
 876         acpi_boot_table_init();
 877
 878         early_acpi_boot_init();
 879
 880 #ifdef CONFIG_ACPI_NUMA
 881         /*
 882          * Parse SRAT to discover nodes.
 883          */
 884         acpi_numa_init();
 885 #endif
 886
 887         initmem_init(0, max_pfn);
 888
 889 #ifdef CONFIG_ACPI_SLEEP
 890         /*
 891          * Reserve low memory region for sleep support.
 892          */
 893         acpi_reserve_bootmem();
 894 #endif
 895         /*
 896          * Find and reserve possible boot-time SMP configuration:
 897          */
 898         find_smp_config();
 899
 900         reserve_crashkernel();
 901
 902 #ifdef CONFIG_X86_64
 903         /*
 904          * dma32_reserve_bootmem() allocates bootmem which may conflict
 905          * with the crashkernel command line, so do that after
 906          * reserve_crashkernel()
 907          */
 908         dma32_reserve_bootmem();
 909 #endif
 910
 911         reserve_ibft_region();
 912
 913 #ifdef CONFIG_KVM_CLOCK
 914         kvmclock_init();
 915 #endif
 916
 917         paravirt_pagetable_setup_start(swapper_pg_dir);
 918         paging_init();
 919         paravirt_pagetable_setup_done(swapper_pg_dir);
 920         paravirt_post_allocator_init();
 921
 922 #ifdef CONFIG_X86_64
 923         map_vsyscall();
 924 #endif
 925
 926         generic_apic_probe();
 927
 928         early_quirks();
 929
 930         /*
 931          * Read APIC and some other early information from ACPI tables.
 932          */
 933         acpi_boot_init();
 934
 935 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 936         /*
 937          * get boot-time SMP configuration:
 938          */
 939         if (smp_found_config)
 940                 get_smp_config();
 941 #endif
 942
 943         prefill_possible_map();
 944
 945 #ifdef CONFIG_X86_64
 946         init_cpu_to_node();
 947 #endif
 948
 949         init_apic_mappings();
 950         ioapic_init_mappings();
 951
 952         /* need to wait for io_apic is mapped */
 953         probe_nr_irqs_gsi();
 954
 955         kvm_guest_init();
 956
 957         e820_reserve_resources();
 958         e820_mark_nosave_regions(max_low_pfn);
 959
 960 #ifdef CONFIG_X86_32
 961         request_resource(&iomem_resource, &video_ram_resource);
 962 #endif
 963         reserve_standard_io_resources();
 964
 965         e820_setup_gap();
 966
 967 #ifdef CONFIG_VT
 968 #if defined(CONFIG_VGA_CONSOLE)
 969         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 970                 conswitchp = &vga_con;
 971 #elif defined(CONFIG_DUMMY_CONSOLE)
 972         conswitchp = &dummy_con;
 973 #endif
 974 #endif
 975 }
 976
 977 #ifdef CONFIG_X86_32
 978
 979 /**
 980  * x86_quirk_pre_intr_init - initialisation prior to setting up interrupt vectors
 981  *
 982  * Description:
 983  *      Perform any necessary interrupt initialisation prior to setting up
 984  *      the "ordinary" interrupt call gates.  For legacy reasons, the ISA
 985  *      interrupts should be initialised here if the machine emulates a PC
 986  *      in any way.
 987  **/
 988 void __init x86_quirk_pre_intr_init(void)
 989 {
 990         if (x86_quirks->arch_pre_intr_init) {
 991                 if (x86_quirks->arch_pre_intr_init())
 992                         return;
 993         }
 994         init_ISA_irqs();
 995 }
 996
 997 /**
 998  * x86_quirk_intr_init - post gate setup interrupt initialisation
 999  *
1000  * Description:
1001  *      Fill in any interrupts that may have been left out by the general
1002  *      init_IRQ() routine.  interrupts having to do with the machine rather
1003  *      than the devices on the I/O bus (like APIC interrupts in intel MP
1004  *      systems) are started here.
1005  **/
1006 void __init x86_quirk_intr_init(void)
1007 {
1008         if (x86_quirks->arch_intr_init) {
1009                 if (x86_quirks->arch_intr_init())
1010                         return;
1011         }
1012 }
1013
1014 /**
1015  * x86_quirk_trap_init - initialise system specific traps
1016  *
1017  * Description:
1018  *      Called as the final act of trap_init().  Used in VISWS to initialise
1019  *      the various board specific APIC traps.
1020  **/
1021 void __init x86_quirk_trap_init(void)
1022 {
1023         if (x86_quirks->arch_trap_init) {
1024                 if (x86_quirks->arch_trap_init())
1025                         return;
1026         }
1027 }
1028
1029 static struct irqaction irq0  = {
1030         .handler = timer_interrupt,
1031         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1032         .mask = CPU_MASK_NONE,
1033         .name = "timer"
1034 };
1035
1036 /**
1037  * x86_quirk_pre_time_init - do any specific initialisations before.
1038  *
1039  **/
1040 void __init x86_quirk_pre_time_init(void)
1041 {
1042         if (x86_quirks->arch_pre_time_init)
1043                 x86_quirks->arch_pre_time_init();
1044 }
1045
1046 /**
1047  * x86_quirk_time_init - do any specific initialisations for the system timer.
1048  *
1049  * Description:
1050  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1051  *      in irq_vectors.h:TIMER_IRQ
1052  **/
1053 void __init x86_quirk_time_init(void)
1054 {
1055         if (x86_quirks->arch_time_init) {
1056                 /*
1057                  * A nonzero return code does not mean failure, it means
1058                  * that the architecture quirk does not want any
1059                  * generic (timer) setup to be performed after this:
1060                  */
1061                 if (x86_quirks->arch_time_init())
1062                         return;
1063         }
1064
1065         irq0.mask = cpumask_of_cpu(0);
1066         setup_irq(0, &irq0);
1067 }
1068 #endif /* CONFIG_X86_32 */