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 /*
 116  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
 117  * The direct mapping extends to max_pfn_mapped, so that we can directly access
 118  * apertures, ACPI and other tables without having to play with fixmaps.
 119  */
 120 unsigned long max_low_pfn_mapped;
 121 unsigned long max_pfn_mapped;
 122
 123 RESERVE_BRK(dmi_alloc, 65536);
 124
 125 unsigned int boot_cpu_id __read_mostly;
 126
 127 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 128 unsigned long _brk_end = (unsigned long)__brk_base;
 129
 130 #ifdef CONFIG_X86_64
 131 int default_cpu_present_to_apicid(int mps_cpu)
 132 {
 133         return __default_cpu_present_to_apicid(mps_cpu);
 134 }
 135
 136 int default_check_phys_apicid_present(int boot_cpu_physical_apicid)
 137 {
 138         return __default_check_phys_apicid_present(boot_cpu_physical_apicid);
 139 }
 140 #endif
 141
 142 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 143 struct boot_params __initdata boot_params;
 144 #else
 145 struct boot_params boot_params;
 146 #endif
 147
 148 /*
 149  * Machine setup..
 150  */
 151 static struct resource data_resource = {
 152         .name   = "Kernel data",
 153         .start  = 0,
 154         .end    = 0,
 155         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 156 };
 157
 158 static struct resource code_resource = {
 159         .name   = "Kernel code",
 160         .start  = 0,
 161         .end    = 0,
 162         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 163 };
 164
 165 static struct resource bss_resource = {
 166         .name   = "Kernel bss",
 167         .start  = 0,
 168         .end    = 0,
 169         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 170 };
 171
 172
 173 #ifdef CONFIG_X86_32
 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 and version as integers, for the benefit of proc_dointvec */
 219 int bootloader_type, bootloader_version;
 220
 221 /*
 222  * Setup options
 223  */
 224 struct screen_info screen_info;
 225 EXPORT_SYMBOL(screen_info);
 226 struct edid_info edid_info;
 227 EXPORT_SYMBOL_GPL(edid_info);
 228
 229 extern int root_mountflags;
 230
 231 unsigned long saved_video_mode;
 232
 233 #define RAMDISK_IMAGE_START_MASK        0x07FF
 234 #define RAMDISK_PROMPT_FLAG             0x8000
 235 #define RAMDISK_LOAD_FLAG               0x4000
 236
 237 static char __initdata command_line[COMMAND_LINE_SIZE];
 238 #ifdef CONFIG_CMDLINE_BOOL
 239 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 240 #endif
 241
 242 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 243 struct edd edd;
 244 #ifdef CONFIG_EDD_MODULE
 245 EXPORT_SYMBOL(edd);
 246 #endif
 247 /**
 248  * copy_edd() - Copy the BIOS EDD information
 249  *              from boot_params into a safe place.
 250  *
 251  */
 252 static inline void copy_edd(void)
 253 {
 254      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 255             sizeof(edd.mbr_signature));
 256      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 257      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 258      edd.edd_info_nr = boot_params.eddbuf_entries;
 259 }
 260 #else
 261 static inline void copy_edd(void)
 262 {
 263 }
 264 #endif
 265
 266 void * __init extend_brk(size_t size, size_t align)
 267 {
 268         size_t mask = align - 1;
 269         void *ret;
 270
 271         BUG_ON(_brk_start == 0);
 272         BUG_ON(align & mask);
 273
 274         _brk_end = (_brk_end + mask) & ~mask;
 275         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 276
 277         ret = (void *)_brk_end;
 278         _brk_end += size;
 279
 280         memset(ret, 0, size);
 281
 282         return ret;
 283 }
 284
 285 #ifdef CONFIG_X86_64
 286 static void __init init_gbpages(void)
 287 {
 288         if (direct_gbpages && cpu_has_gbpages)
 289                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 290         else
 291                 direct_gbpages = 0;
 292 }
 293 #else
 294 static inline void init_gbpages(void)
 295 {
 296 }
 297 #endif
 298
 299 static void __init reserve_brk(void)
 300 {
 301         if (_brk_end > _brk_start)
 302                 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK");
 303
 304         /* Mark brk area as locked down and no longer taking any
 305            new allocations */
 306         _brk_start = 0;
 307 }
 308
 309 #ifdef CONFIG_BLK_DEV_INITRD
 310
 311 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 312 static void __init relocate_initrd(void)
 313 {
 314
 315         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 316         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 317         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 318         u64 ramdisk_here;
 319         unsigned long slop, clen, mapaddr;
 320         char *p, *q;
 321
 322         /* We need to move the initrd down into lowmem */
 323         ramdisk_here = find_e820_area(0, end_of_lowmem, ramdisk_size,
 324                                          PAGE_SIZE);
 325
 326         if (ramdisk_here == -1ULL)
 327                 panic("Cannot find place for new RAMDISK of size %lld\n",
 328                          ramdisk_size);
 329
 330         /* Note: this includes all the lowmem currently occupied by
 331            the initrd, we rely on that fact to keep the data intact. */
 332         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 333                          "NEW RAMDISK");
 334         initrd_start = ramdisk_here + PAGE_OFFSET;
 335         initrd_end   = initrd_start + ramdisk_size;
 336         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 337                          ramdisk_here, ramdisk_here + ramdisk_size);
 338
 339         q = (char *)initrd_start;
 340
 341         /* Copy any lowmem portion of the initrd */
 342         if (ramdisk_image < end_of_lowmem) {
 343                 clen = end_of_lowmem - ramdisk_image;
 344                 p = (char *)__va(ramdisk_image);
 345                 memcpy(q, p, clen);
 346                 q += clen;
 347                 ramdisk_image += clen;
 348                 ramdisk_size  -= clen;
 349         }
 350
 351         /* Copy the highmem portion of the initrd */
 352         while (ramdisk_size) {
 353                 slop = ramdisk_image & ~PAGE_MASK;
 354                 clen = ramdisk_size;
 355                 if (clen > MAX_MAP_CHUNK-slop)
 356                         clen = MAX_MAP_CHUNK-slop;
 357                 mapaddr = ramdisk_image & PAGE_MASK;
 358                 p = early_memremap(mapaddr, clen+slop);
 359                 memcpy(q, p+slop, clen);
 360                 early_iounmap(p, clen+slop);
 361                 q += clen;
 362                 ramdisk_image += clen;
 363                 ramdisk_size  -= clen;
 364         }
 365         /* high pages is not converted by early_res_to_bootmem */
 366         ramdisk_image = boot_params.hdr.ramdisk_image;
 367         ramdisk_size  = boot_params.hdr.ramdisk_size;
 368         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 369                 " %08llx - %08llx\n",
 370                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 371                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 372 }
 373
 374 static void __init reserve_initrd(void)
 375 {
 376         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 377         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 378         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 379         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 380
 381         if (!boot_params.hdr.type_of_loader ||
 382             !ramdisk_image || !ramdisk_size)
 383                 return;         /* No initrd provided by bootloader */
 384
 385         initrd_start = 0;
 386
 387         if (ramdisk_size >= (end_of_lowmem>>1)) {
 388                 free_early(ramdisk_image, ramdisk_end);
 389                 printk(KERN_ERR "initrd too large to handle, "
 390                        "disabling initrd\n");
 391                 return;
 392         }
 393
 394         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 395                         ramdisk_end);
 396
 397
 398         if (ramdisk_end <= end_of_lowmem) {
 399                 /* All in lowmem, easy case */
 400                 /*
 401                  * don't need to reserve again, already reserved early
 402                  * in i386_start_kernel
 403                  */
 404                 initrd_start = ramdisk_image + PAGE_OFFSET;
 405                 initrd_end = initrd_start + ramdisk_size;
 406                 return;
 407         }
 408
 409         relocate_initrd();
 410
 411         free_early(ramdisk_image, ramdisk_end);
 412 }
 413 #else
 414 static void __init reserve_initrd(void)
 415 {
 416 }
 417 #endif /* CONFIG_BLK_DEV_INITRD */
 418
 419 static void __init parse_setup_data(void)
 420 {
 421         struct setup_data *data;
 422         u64 pa_data;
 423
 424         if (boot_params.hdr.version < 0x0209)
 425                 return;
 426         pa_data = boot_params.hdr.setup_data;
 427         while (pa_data) {
 428                 data = early_memremap(pa_data, PAGE_SIZE);
 429                 switch (data->type) {
 430                 case SETUP_E820_EXT:
 431                         parse_e820_ext(data, pa_data);
 432                         break;
 433                 default:
 434                         break;
 435                 }
 436                 pa_data = data->next;
 437                 early_iounmap(data, PAGE_SIZE);
 438         }
 439 }
 440
 441 static void __init e820_reserve_setup_data(void)
 442 {
 443         struct setup_data *data;
 444         u64 pa_data;
 445         int found = 0;
 446
 447         if (boot_params.hdr.version < 0x0209)
 448                 return;
 449         pa_data = boot_params.hdr.setup_data;
 450         while (pa_data) {
 451                 data = early_memremap(pa_data, sizeof(*data));
 452                 e820_update_range(pa_data, sizeof(*data)+data->len,
 453                          E820_RAM, E820_RESERVED_KERN);
 454                 found = 1;
 455                 pa_data = data->next;
 456                 early_iounmap(data, sizeof(*data));
 457         }
 458         if (!found)
 459                 return;
 460
 461         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 462         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 463         printk(KERN_INFO "extended physical RAM map:\n");
 464         e820_print_map("reserve setup_data");
 465 }
 466
 467 static void __init reserve_early_setup_data(void)
 468 {
 469         struct setup_data *data;
 470         u64 pa_data;
 471         char buf[32];
 472
 473         if (boot_params.hdr.version < 0x0209)
 474                 return;
 475         pa_data = boot_params.hdr.setup_data;
 476         while (pa_data) {
 477                 data = early_memremap(pa_data, sizeof(*data));
 478                 sprintf(buf, "setup data %x", data->type);
 479                 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf);
 480                 pa_data = data->next;
 481                 early_iounmap(data, sizeof(*data));
 482         }
 483 }
 484
 485 /*
 486  * --------- Crashkernel reservation ------------------------------
 487  */
 488
 489 #ifdef CONFIG_KEXEC
 490
 491 /**
 492  * Reserve @size bytes of crashkernel memory at any suitable offset.
 493  *
 494  * @size: Size of the crashkernel memory to reserve.
 495  * Returns the base address on success, and -1ULL on failure.
 496  */
 497 static
 498 unsigned long long __init find_and_reserve_crashkernel(unsigned long long size)
 499 {
 500         const unsigned long long alignment = 16<<20;    /* 16M */
 501         unsigned long long start = 0LL;
 502
 503         while (1) {
 504                 int ret;
 505
 506                 start = find_e820_area(start, ULONG_MAX, size, alignment);
 507                 if (start == -1ULL)
 508                         return start;
 509
 510                 /* try to reserve it */
 511                 ret = reserve_bootmem_generic(start, size, BOOTMEM_EXCLUSIVE);
 512                 if (ret >= 0)
 513                         return start;
 514
 515                 start += alignment;
 516         }
 517 }
 518
 519 static inline unsigned long long get_total_mem(void)
 520 {
 521         unsigned long long total;
 522
 523         total = max_low_pfn - min_low_pfn;
 524 #ifdef CONFIG_HIGHMEM
 525         total += highend_pfn - highstart_pfn;
 526 #endif
 527
 528         return total << PAGE_SHIFT;
 529 }
 530
 531 static void __init reserve_crashkernel(void)
 532 {
 533         unsigned long long total_mem;
 534         unsigned long long crash_size, crash_base;
 535         int ret;
 536
 537         total_mem = get_total_mem();
 538
 539         ret = parse_crashkernel(boot_command_line, total_mem,
 540                         &crash_size, &crash_base);
 541         if (ret != 0 || crash_size <= 0)
 542                 return;
 543
 544         /* 0 means: find the address automatically */
 545         if (crash_base <= 0) {
 546                 crash_base = find_and_reserve_crashkernel(crash_size);
 547                 if (crash_base == -1ULL) {
 548                         pr_info("crashkernel reservation failed. "
 549                                 "No suitable area found.\n");
 550                         return;
 551                 }
 552         } else {
 553                 ret = reserve_bootmem_generic(crash_base, crash_size,
 554                                         BOOTMEM_EXCLUSIVE);
 555                 if (ret < 0) {
 556                         pr_info("crashkernel reservation failed - "
 557                                 "memory is in use\n");
 558                         return;
 559                 }
 560         }
 561
 562         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 563                         "for crashkernel (System RAM: %ldMB)\n",
 564                         (unsigned long)(crash_size >> 20),
 565                         (unsigned long)(crash_base >> 20),
 566                         (unsigned long)(total_mem >> 20));
 567
 568         crashk_res.start = crash_base;
 569         crashk_res.end   = crash_base + crash_size - 1;
 570         insert_resource(&iomem_resource, &crashk_res);
 571 }
 572 #else
 573 static void __init reserve_crashkernel(void)
 574 {
 575 }
 576 #endif
 577
 578 static struct resource standard_io_resources[] = {
 579         { .name = "dma1", .start = 0x00, .end = 0x1f,
 580                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 581         { .name = "pic1", .start = 0x20, .end = 0x21,
 582                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 583         { .name = "timer0", .start = 0x40, .end = 0x43,
 584                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 585         { .name = "timer1", .start = 0x50, .end = 0x53,
 586                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 587         { .name = "keyboard", .start = 0x60, .end = 0x60,
 588                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 589         { .name = "keyboard", .start = 0x64, .end = 0x64,
 590                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 591         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 592                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 593         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 594                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 595         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 596                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 597         { .name = "fpu", .start = 0xf0, .end = 0xff,
 598                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 599 };
 600
 601 void __init reserve_standard_io_resources(void)
 602 {
 603         int i;
 604
 605         /* request I/O space for devices used on all i[345]86 PCs */
 606         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 607                 request_resource(&ioport_resource, &standard_io_resources[i]);
 608
 609 }
 610
 611 /*
 612  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
 613  * is_kdump_kernel() to determine if we are booting after a panic. Hence
 614  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
 615  */
 616
 617 #ifdef CONFIG_CRASH_DUMP
 618 /* elfcorehdr= specifies the location of elf core header
 619  * stored by the crashed kernel. This option will be passed
 620  * by kexec loader to the capture kernel.
 621  */
 622 static int __init setup_elfcorehdr(char *arg)
 623 {
 624         char *end;
 625         if (!arg)
 626                 return -EINVAL;
 627         elfcorehdr_addr = memparse(arg, &end);
 628         return end > arg ? 0 : -EINVAL;
 629 }
 630 early_param("elfcorehdr", setup_elfcorehdr);
 631 #endif
 632
 633 static struct x86_quirks default_x86_quirks __initdata;
 634
 635 struct x86_quirks *x86_quirks __initdata = &default_x86_quirks;
 636
 637 #ifdef CONFIG_X86_RESERVE_LOW_64K
 638 static int __init dmi_low_memory_corruption(const struct dmi_system_id *d)
 639 {
 640         printk(KERN_NOTICE
 641                 "%s detected: BIOS may corrupt low RAM, working around it.\n",
 642                 d->ident);
 643
 644         e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED);
 645         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 646
 647         return 0;
 648 }
 649 #endif
 650
 651 /* List of systems that have known low memory corruption BIOS problems */
 652 static struct dmi_system_id __initdata bad_bios_dmi_table[] = {
 653 #ifdef CONFIG_X86_RESERVE_LOW_64K
 654         {
 655                 .callback = dmi_low_memory_corruption,
 656                 .ident = "AMI BIOS",
 657                 .matches = {
 658                         DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."),
 659                 },
 660         },
 661         {
 662                 .callback = dmi_low_memory_corruption,
 663                 .ident = "Phoenix BIOS",
 664                 .matches = {
 665                         DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"),
 666                 },
 667         },
 668         {
 669         /*
 670          * AMI BIOS with low memory corruption was found on Intel DG45ID board.
 671          * It hase different DMI_BIOS_VENDOR = "Intel Corp.", for now we will
 672          * match only DMI_BOARD_NAME and see if there is more bad products
 673          * with this vendor.
 674          */
 675                 .callback = dmi_low_memory_corruption,
 676                 .ident = "AMI BIOS",
 677                 .matches = {
 678                         DMI_MATCH(DMI_BOARD_NAME, "DG45ID"),
 679                 },
 680         },
 681 #endif
 682         {}
 683 };
 684
 685 /*
 686  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 687  * passed the efi memmap, systab, etc., so we should use these data structures
 688  * for initialization.  Note, the efi init code path is determined by the
 689  * global efi_enabled. This allows the same kernel image to be used on existing
 690  * systems (with a traditional BIOS) as well as on EFI systems.
 691  */
 692 /*
 693  * setup_arch - architecture-specific boot-time initializations
 694  *
 695  * Note: On x86_64, fixmaps are ready for use even before this is called.
 696  */
 697
 698 void __init setup_arch(char **cmdline_p)
 699 {
 700 #ifdef CONFIG_X86_32
 701         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 702         visws_early_detect();
 703 #else
 704         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 705 #endif
 706
 707         /* VMI may relocate the fixmap; do this before touching ioremap area */
 708         vmi_init();
 709
 710         early_cpu_init();
 711         early_ioremap_init();
 712
 713         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 714         screen_info = boot_params.screen_info;
 715         edid_info = boot_params.edid_info;
 716 #ifdef CONFIG_X86_32
 717         apm_info.bios = boot_params.apm_bios_info;
 718         ist_info = boot_params.ist_info;
 719         if (boot_params.sys_desc_table.length != 0) {
 720                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 721                 machine_id = boot_params.sys_desc_table.table[0];
 722                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 723                 BIOS_revision = boot_params.sys_desc_table.table[2];
 724         }
 725 #endif
 726         saved_video_mode = boot_params.hdr.vid_mode;
 727         bootloader_type = boot_params.hdr.type_of_loader;
 728         if ((bootloader_type >> 4) == 0xe) {
 729                 bootloader_type &= 0xf;
 730                 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
 731         }
 732         bootloader_version  = bootloader_type & 0xf;
 733         bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
 734
 735 #ifdef CONFIG_BLK_DEV_RAM
 736         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 737         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 738         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 739 #endif
 740 #ifdef CONFIG_EFI
 741         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 742 #ifdef CONFIG_X86_32
 743                      "EL32",
 744 #else
 745                      "EL64",
 746 #endif
 747          4)) {
 748                 efi_enabled = 1;
 749                 efi_reserve_early();
 750         }
 751 #endif
 752
 753         ARCH_SETUP
 754
 755         setup_memory_map();
 756         parse_setup_data();
 757         /* update the e820_saved too */
 758         e820_reserve_setup_data();
 759
 760         copy_edd();
 761
 762         if (!boot_params.hdr.root_flags)
 763                 root_mountflags &= ~MS_RDONLY;
 764         init_mm.start_code = (unsigned long) _text;
 765         init_mm.end_code = (unsigned long) _etext;
 766         init_mm.end_data = (unsigned long) _edata;
 767         init_mm.brk = _brk_end;
 768
 769         code_resource.start = virt_to_phys(_text);
 770         code_resource.end = virt_to_phys(_etext)-1;
 771         data_resource.start = virt_to_phys(_etext);
 772         data_resource.end = virt_to_phys(_edata)-1;
 773         bss_resource.start = virt_to_phys(&__bss_start);
 774         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 775
 776 #ifdef CONFIG_CMDLINE_BOOL
 777 #ifdef CONFIG_CMDLINE_OVERRIDE
 778         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 779 #else
 780         if (builtin_cmdline[0]) {
 781                 /* append boot loader cmdline to builtin */
 782                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 783                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 784                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 785         }
 786 #endif
 787 #endif
 788
 789         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 790         *cmdline_p = command_line;
 791
 792         parse_early_param();
 793
 794 #ifdef CONFIG_X86_64
 795         check_efer();
 796 #endif
 797
 798         /* Must be before kernel pagetables are setup */
 799         vmi_activate();
 800
 801         /* after early param, so could get panic from serial */
 802         reserve_early_setup_data();
 803
 804         if (acpi_mps_check()) {
 805 #ifdef CONFIG_X86_LOCAL_APIC
 806                 disable_apic = 1;
 807 #endif
 808                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 809         }
 810
 811 #ifdef CONFIG_PCI
 812         if (pci_early_dump_regs)
 813                 early_dump_pci_devices();
 814 #endif
 815
 816         finish_e820_parsing();
 817
 818         if (efi_enabled)
 819                 efi_init();
 820
 821         dmi_scan_machine();
 822
 823         dmi_check_system(bad_bios_dmi_table);
 824
 825         /*
 826          * VMware detection requires dmi to be available, so this
 827          * needs to be done after dmi_scan_machine, for the BP.
 828          */
 829         init_hypervisor(&boot_cpu_data);
 830
 831         x86_init.resources.probe_roms();
 832
 833         /* after parse_early_param, so could debug it */
 834         insert_resource(&iomem_resource, &code_resource);
 835         insert_resource(&iomem_resource, &data_resource);
 836         insert_resource(&iomem_resource, &bss_resource);
 837
 838
 839 #ifdef CONFIG_X86_32
 840         if (ppro_with_ram_bug()) {
 841                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 842                                   E820_RESERVED);
 843                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 844                 printk(KERN_INFO "fixed physical RAM map:\n");
 845                 e820_print_map("bad_ppro");
 846         }
 847 #else
 848         early_gart_iommu_check();
 849 #endif
 850
 851         /*
 852          * partially used pages are not usable - thus
 853          * we are rounding upwards:
 854          */
 855         max_pfn = e820_end_of_ram_pfn();
 856
 857         /* preallocate 4k for mptable mpc */
 858         early_reserve_e820_mpc_new();
 859         /* update e820 for memory not covered by WB MTRRs */
 860         mtrr_bp_init();
 861         if (mtrr_trim_uncached_memory(max_pfn))
 862                 max_pfn = e820_end_of_ram_pfn();
 863
 864 #ifdef CONFIG_X86_32
 865         /* max_low_pfn get updated here */
 866         find_low_pfn_range();
 867 #else
 868         num_physpages = max_pfn;
 869
 870         check_x2apic();
 871
 872         /* How many end-of-memory variables you have, grandma! */
 873         /* need this before calling reserve_initrd */
 874         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 875                 max_low_pfn = e820_end_of_low_ram_pfn();
 876         else
 877                 max_low_pfn = max_pfn;
 878
 879         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 880         max_pfn_mapped = KERNEL_IMAGE_SIZE >> PAGE_SHIFT;
 881 #endif
 882
 883 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 884         setup_bios_corruption_check();
 885 #endif
 886
 887         printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
 888                         max_pfn_mapped<<PAGE_SHIFT);
 889
 890         reserve_brk();
 891
 892         init_gbpages();
 893
 894         /* max_pfn_mapped is updated here */
 895         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 896         max_pfn_mapped = max_low_pfn_mapped;
 897
 898 #ifdef CONFIG_X86_64
 899         if (max_pfn > max_low_pfn) {
 900                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 901                                                      max_pfn<<PAGE_SHIFT);
 902                 /* can we preseve max_low_pfn ?*/
 903                 max_low_pfn = max_pfn;
 904         }
 905 #endif
 906
 907         /*
 908          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 909          */
 910
 911 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 912         if (init_ohci1394_dma_early)
 913                 init_ohci1394_dma_on_all_controllers();
 914 #endif
 915
 916         reserve_initrd();
 917
 918         vsmp_init();
 919
 920         io_delay_init();
 921
 922         /*
 923          * Parse the ACPI tables for possible boot-time SMP configuration.
 924          */
 925         acpi_boot_table_init();
 926
 927         early_acpi_boot_init();
 928
 929 #ifdef CONFIG_ACPI_NUMA
 930         /*
 931          * Parse SRAT to discover nodes.
 932          */
 933         acpi_numa_init();
 934 #endif
 935
 936         initmem_init(0, max_pfn);
 937
 938 #ifdef CONFIG_ACPI_SLEEP
 939         /*
 940          * Reserve low memory region for sleep support.
 941          */
 942         acpi_reserve_bootmem();
 943 #endif
 944         /*
 945          * Find and reserve possible boot-time SMP configuration:
 946          */
 947         find_smp_config();
 948
 949         reserve_crashkernel();
 950
 951 #ifdef CONFIG_X86_64
 952         /*
 953          * dma32_reserve_bootmem() allocates bootmem which may conflict
 954          * with the crashkernel command line, so do that after
 955          * reserve_crashkernel()
 956          */
 957         dma32_reserve_bootmem();
 958 #endif
 959
 960         reserve_ibft_region();
 961
 962 #ifdef CONFIG_KVM_CLOCK
 963         kvmclock_init();
 964 #endif
 965
 966         paravirt_pagetable_setup_start(swapper_pg_dir);
 967         paging_init();
 968         paravirt_pagetable_setup_done(swapper_pg_dir);
 969         paravirt_post_allocator_init();
 970
 971 #ifdef CONFIG_X86_64
 972         map_vsyscall();
 973 #endif
 974
 975         generic_apic_probe();
 976
 977         early_quirks();
 978
 979         /*
 980          * Read APIC and some other early information from ACPI tables.
 981          */
 982         acpi_boot_init();
 983
 984         /*
 985          * get boot-time SMP configuration:
 986          */
 987         if (smp_found_config)
 988                 get_smp_config();
 989
 990         prefill_possible_map();
 991
 992 #ifdef CONFIG_X86_64
 993         init_cpu_to_node();
 994 #endif
 995
 996         init_apic_mappings();
 997         ioapic_init_mappings();
 998
 999         /* need to wait for io_apic is mapped */
1000         probe_nr_irqs_gsi();
1001
1002         kvm_guest_init();
1003
1004         e820_reserve_resources();
1005         e820_mark_nosave_regions(max_low_pfn);
1006
1007         x86_init.resources.reserve_resources();
1008
1009         e820_setup_gap();
1010
1011 #ifdef CONFIG_VT
1012 #if defined(CONFIG_VGA_CONSOLE)
1013         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1014                 conswitchp = &vga_con;
1015 #elif defined(CONFIG_DUMMY_CONSOLE)
1016         conswitchp = &dummy_con;
1017 #endif
1018 #endif
1019 }
1020
1021 #ifdef CONFIG_X86_32
1022
1023 /**
1024  * x86_quirk_trap_init - initialise system specific traps
1025  *
1026  * Description:
1027  *      Called as the final act of trap_init().  Used in VISWS to initialise
1028  *      the various board specific APIC traps.
1029  **/
1030 void __init x86_quirk_trap_init(void)
1031 {
1032         if (x86_quirks->arch_trap_init) {
1033                 if (x86_quirks->arch_trap_init())
1034                         return;
1035         }
1036 }
1037
1038 static struct irqaction irq0  = {
1039         .handler = timer_interrupt,
1040         .flags = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_IRQPOLL | IRQF_TIMER,
1041         .name = "timer"
1042 };
1043
1044 /**
1045  * x86_quirk_pre_time_init - do any specific initialisations before.
1046  *
1047  **/
1048 void __init x86_quirk_pre_time_init(void)
1049 {
1050         if (x86_quirks->arch_pre_time_init)
1051                 x86_quirks->arch_pre_time_init();
1052 }
1053
1054 /**
1055  * x86_quirk_time_init - do any specific initialisations for the system timer.
1056  *
1057  * Description:
1058  *      Must plug the system timer interrupt source at HZ into the IRQ listed
1059  *      in irq_vectors.h:TIMER_IRQ
1060  **/
1061 void __init x86_quirk_time_init(void)
1062 {
1063         if (x86_quirks->arch_time_init) {
1064                 /*
1065                  * A nonzero return code does not mean failure, it means
1066                  * that the architecture quirk does not want any
1067                  * generic (timer) setup to be performed after this:
1068                  */
1069                 if (x86_quirks->arch_time_init())
1070                         return;
1071         }
1072
1073         irq0.mask = cpumask_of_cpu(0);
1074         setup_irq(0, &irq0);
1075 }
1076
1077 static struct resource video_ram_resource = {
1078         .name   = "Video RAM area",
1079         .start  = 0xa0000,
1080         .end    = 0xbffff,
1081         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
1082 };
1083
1084 void __init i386_reserve_resources(void)
1085 {
1086         request_resource(&iomem_resource, &video_ram_resource);
1087         reserve_standard_io_resources();
1088 }
1089
1090 #endif /* CONFIG_X86_32 */