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/sfi.h>
  31 #include <linux/apm_bios.h>
  32 #include <linux/initrd.h>
  33 #include <linux/bootmem.h>
  34 #include <linux/memblock.h>
  35 #include <linux/seq_file.h>
  36 #include <linux/console.h>
  37 #include <linux/root_dev.h>
  38 #include <linux/highmem.h>
  39 #include <linux/module.h>
  40 #include <linux/efi.h>
  41 #include <linux/init.h>
  42 #include <linux/edd.h>
  43 #include <linux/iscsi_ibft.h>
  44 #include <linux/nodemask.h>
  45 #include <linux/kexec.h>
  46 #include <linux/dmi.h>
  47 #include <linux/pfn.h>
  48 #include <linux/pci.h>
  49 #include <asm/pci-direct.h>
  50 #include <linux/init_ohci1394_dma.h>
  51 #include <linux/kvm_para.h>
  52 #include <linux/dma-contiguous.h>
  53
  54 #include <linux/errno.h>
  55 #include <linux/kernel.h>
  56 #include <linux/stddef.h>
  57 #include <linux/unistd.h>
  58 #include <linux/ptrace.h>
  59 #include <linux/user.h>
  60 #include <linux/delay.h>
  61
  62 #include <linux/kallsyms.h>
  63 #include <linux/cpufreq.h>
  64 #include <linux/dma-mapping.h>
  65 #include <linux/ctype.h>
  66 #include <linux/uaccess.h>
  67
  68 #include <linux/percpu.h>
  69 #include <linux/crash_dump.h>
  70 #include <linux/tboot.h>
  71 #include <linux/jiffies.h>
  72
  73 #include <video/edid.h>
  74
  75 #include <asm/mtrr.h>
  76 #include <asm/apic.h>
  77 #include <asm/realmode.h>
  78 #include <asm/e820.h>
  79 #include <asm/mpspec.h>
  80 #include <asm/setup.h>
  81 #include <asm/efi.h>
  82 #include <asm/timer.h>
  83 #include <asm/i8259.h>
  84 #include <asm/sections.h>
  85 #include <asm/io_apic.h>
  86 #include <asm/ist.h>
  87 #include <asm/setup_arch.h>
  88 #include <asm/bios_ebda.h>
  89 #include <asm/cacheflush.h>
  90 #include <asm/processor.h>
  91 #include <asm/bugs.h>
  92
  93 #include <asm/vsyscall.h>
  94 #include <asm/cpu.h>
  95 #include <asm/desc.h>
  96 #include <asm/dma.h>
  97 #include <asm/iommu.h>
  98 #include <asm/gart.h>
  99 #include <asm/mmu_context.h>
 100 #include <asm/proto.h>
 101
 102 #include <asm/paravirt.h>
 103 #include <asm/hypervisor.h>
 104 #include <asm/olpc_ofw.h>
 105
 106 #include <asm/percpu.h>
 107 #include <asm/topology.h>
 108 #include <asm/apicdef.h>
 109 #include <asm/amd_nb.h>
 110 #include <asm/mce.h>
 111 #include <asm/alternative.h>
 112 #include <asm/prom.h>
 113
 114 /*
 115  * max_low_pfn_mapped: highest direct mapped pfn under 4GB
 116  * max_pfn_mapped:     highest direct mapped pfn over 4GB
 117  *
 118  * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
 119  * represented by pfn_mapped
 120  */
 121 unsigned long max_low_pfn_mapped;
 122 unsigned long max_pfn_mapped;
 123
 124 #ifdef CONFIG_DMI
 125 RESERVE_BRK(dmi_alloc, 65536);
 126 #endif
 127
 128
 129 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 130 unsigned long _brk_end = (unsigned long)__brk_base;
 131
 132 #ifdef CONFIG_X86_64
 133 int default_cpu_present_to_apicid(int mps_cpu)
 134 {
 135         return __default_cpu_present_to_apicid(mps_cpu);
 136 }
 137
 138 int default_check_phys_apicid_present(int phys_apicid)
 139 {
 140         return __default_check_phys_apicid_present(phys_apicid);
 141 }
 142 #endif
 143
 144 struct boot_params boot_params;
 145
 146 /*
 147  * Machine setup..
 148  */
 149 static struct resource data_resource = {
 150         .name   = "Kernel data",
 151         .start  = 0,
 152         .end    = 0,
 153         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 154 };
 155
 156 static struct resource code_resource = {
 157         .name   = "Kernel code",
 158         .start  = 0,
 159         .end    = 0,
 160         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 161 };
 162
 163 static struct resource bss_resource = {
 164         .name   = "Kernel bss",
 165         .start  = 0,
 166         .end    = 0,
 167         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 168 };
 169
 170
 171 #ifdef CONFIG_X86_32
 172 /* cpu data as detected by the assembly code in head.S */
 173 struct cpuinfo_x86 new_cpu_data = {
 174         .wp_works_ok = -1,
 175 };
 176 /* common cpu data for all cpus */
 177 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 178         .wp_works_ok = -1,
 179 };
 180 EXPORT_SYMBOL(boot_cpu_data);
 181
 182 unsigned int def_to_bigsmp;
 183
 184 /* for MCA, but anyone else can use it if they want */
 185 unsigned int machine_id;
 186 unsigned int machine_submodel_id;
 187 unsigned int BIOS_revision;
 188
 189 struct apm_info apm_info;
 190 EXPORT_SYMBOL(apm_info);
 191
 192 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 193         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 194 struct ist_info ist_info;
 195 EXPORT_SYMBOL(ist_info);
 196 #else
 197 struct ist_info ist_info;
 198 #endif
 199
 200 #else
 201 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 202         .x86_phys_bits = MAX_PHYSMEM_BITS,
 203 };
 204 EXPORT_SYMBOL(boot_cpu_data);
 205 #endif
 206
 207
 208 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 209 __visible unsigned long mmu_cr4_features;
 210 #else
 211 __visible unsigned long mmu_cr4_features = X86_CR4_PAE;
 212 #endif
 213
 214 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
 215 int bootloader_type, bootloader_version;
 216
 217 /*
 218  * Setup options
 219  */
 220 struct screen_info screen_info;
 221 EXPORT_SYMBOL(screen_info);
 222 struct edid_info edid_info;
 223 EXPORT_SYMBOL_GPL(edid_info);
 224
 225 extern int root_mountflags;
 226
 227 unsigned long saved_video_mode;
 228
 229 #define RAMDISK_IMAGE_START_MASK        0x07FF
 230 #define RAMDISK_PROMPT_FLAG             0x8000
 231 #define RAMDISK_LOAD_FLAG               0x4000
 232
 233 static char __initdata command_line[COMMAND_LINE_SIZE];
 234 #ifdef CONFIG_CMDLINE_BOOL
 235 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 236 #endif
 237
 238 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 239 struct edd edd;
 240 #ifdef CONFIG_EDD_MODULE
 241 EXPORT_SYMBOL(edd);
 242 #endif
 243 /**
 244  * copy_edd() - Copy the BIOS EDD information
 245  *              from boot_params into a safe place.
 246  *
 247  */
 248 static inline void __init copy_edd(void)
 249 {
 250      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 251             sizeof(edd.mbr_signature));
 252      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 253      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 254      edd.edd_info_nr = boot_params.eddbuf_entries;
 255 }
 256 #else
 257 static inline void __init copy_edd(void)
 258 {
 259 }
 260 #endif
 261
 262 void * __init extend_brk(size_t size, size_t align)
 263 {
 264         size_t mask = align - 1;
 265         void *ret;
 266
 267         BUG_ON(_brk_start == 0);
 268         BUG_ON(align & mask);
 269
 270         _brk_end = (_brk_end + mask) & ~mask;
 271         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 272
 273         ret = (void *)_brk_end;
 274         _brk_end += size;
 275
 276         memset(ret, 0, size);
 277
 278         return ret;
 279 }
 280
 281 #ifdef CONFIG_X86_32
 282 static void __init cleanup_highmap(void)
 283 {
 284 }
 285 #endif
 286
 287 static void __init reserve_brk(void)
 288 {
 289         if (_brk_end > _brk_start)
 290                 memblock_reserve(__pa_symbol(_brk_start),
 291                                  _brk_end - _brk_start);
 292
 293         /* Mark brk area as locked down and no longer taking any
 294            new allocations */
 295         _brk_start = 0;
 296 }
 297
 298 #ifdef CONFIG_BLK_DEV_INITRD
 299
 300 static u64 __init get_ramdisk_image(void)
 301 {
 302         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 303
 304         ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
 305
 306         return ramdisk_image;
 307 }
 308 static u64 __init get_ramdisk_size(void)
 309 {
 310         u64 ramdisk_size = boot_params.hdr.ramdisk_size;
 311
 312         ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
 313
 314         return ramdisk_size;
 315 }
 316
 317 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 318 static void __init relocate_initrd(void)
 319 {
 320         /* Assume only end is not page aligned */
 321         u64 ramdisk_image = get_ramdisk_image();
 322         u64 ramdisk_size  = get_ramdisk_size();
 323         u64 area_size     = PAGE_ALIGN(ramdisk_size);
 324         u64 ramdisk_here;
 325         unsigned long slop, clen, mapaddr;
 326         char *p, *q;
 327
 328         /* We need to move the initrd down into directly mapped mem */
 329         ramdisk_here = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
 330                                                  area_size, PAGE_SIZE);
 331
 332         if (!ramdisk_here)
 333                 panic("Cannot find place for new RAMDISK of size %lld\n",
 334                          ramdisk_size);
 335
 336         /* Note: this includes all the mem currently occupied by
 337            the initrd, we rely on that fact to keep the data intact. */
 338         memblock_reserve(ramdisk_here, area_size);
 339         initrd_start = ramdisk_here + PAGE_OFFSET;
 340         initrd_end   = initrd_start + ramdisk_size;
 341         printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
 342                          ramdisk_here, ramdisk_here + ramdisk_size - 1);
 343
 344         q = (char *)initrd_start;
 345
 346         /* Copy the initrd */
 347         while (ramdisk_size) {
 348                 slop = ramdisk_image & ~PAGE_MASK;
 349                 clen = ramdisk_size;
 350                 if (clen > MAX_MAP_CHUNK-slop)
 351                         clen = MAX_MAP_CHUNK-slop;
 352                 mapaddr = ramdisk_image & PAGE_MASK;
 353                 p = early_memremap(mapaddr, clen+slop);
 354                 memcpy(q, p+slop, clen);
 355                 early_iounmap(p, clen+slop);
 356                 q += clen;
 357                 ramdisk_image += clen;
 358                 ramdisk_size  -= clen;
 359         }
 360
 361         ramdisk_image = get_ramdisk_image();
 362         ramdisk_size  = get_ramdisk_size();
 363         printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
 364                 " [mem %#010llx-%#010llx]\n",
 365                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 366                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 367 }
 368
 369 static void __init early_reserve_initrd(void)
 370 {
 371         /* Assume only end is not page aligned */
 372         u64 ramdisk_image = get_ramdisk_image();
 373         u64 ramdisk_size  = get_ramdisk_size();
 374         u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
 375
 376         if (!boot_params.hdr.type_of_loader ||
 377             !ramdisk_image || !ramdisk_size)
 378                 return;         /* No initrd provided by bootloader */
 379
 380         memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
 381 }
 382 static void __init reserve_initrd(void)
 383 {
 384         /* Assume only end is not page aligned */
 385         u64 ramdisk_image = get_ramdisk_image();
 386         u64 ramdisk_size  = get_ramdisk_size();
 387         u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
 388         u64 mapped_size;
 389
 390         if (!boot_params.hdr.type_of_loader ||
 391             !ramdisk_image || !ramdisk_size)
 392                 return;         /* No initrd provided by bootloader */
 393
 394         initrd_start = 0;
 395
 396         mapped_size = memblock_mem_size(max_pfn_mapped);
 397         if (ramdisk_size >= (mapped_size>>1))
 398                 panic("initrd too large to handle, "
 399                        "disabling initrd (%lld needed, %lld available)\n",
 400                        ramdisk_size, mapped_size>>1);
 401
 402         printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
 403                         ramdisk_end - 1);
 404
 405         if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
 406                                 PFN_DOWN(ramdisk_end))) {
 407                 /* All are mapped, easy case */
 408                 initrd_start = ramdisk_image + PAGE_OFFSET;
 409                 initrd_end = initrd_start + ramdisk_size;
 410                 return;
 411         }
 412
 413         relocate_initrd();
 414
 415         memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
 416 }
 417 #else
 418 static void __init early_reserve_initrd(void)
 419 {
 420 }
 421 static void __init reserve_initrd(void)
 422 {
 423 }
 424 #endif /* CONFIG_BLK_DEV_INITRD */
 425
 426 static void __init parse_setup_data(void)
 427 {
 428         struct setup_data *data;
 429         u64 pa_data, pa_next;
 430
 431         pa_data = boot_params.hdr.setup_data;
 432         while (pa_data) {
 433                 u32 data_len, map_len, data_type;
 434
 435                 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
 436                               (u64)sizeof(struct setup_data));
 437                 data = early_memremap(pa_data, map_len);
 438                 data_len = data->len + sizeof(struct setup_data);
 439                 data_type = data->type;
 440                 pa_next = data->next;
 441                 early_iounmap(data, map_len);
 442
 443                 switch (data_type) {
 444                 case SETUP_E820_EXT:
 445                         parse_e820_ext(pa_data, data_len);
 446                         break;
 447                 case SETUP_DTB:
 448                         add_dtb(pa_data);
 449                         break;
 450                 default:
 451                         break;
 452                 }
 453                 pa_data = pa_next;
 454         }
 455 }
 456
 457 static void __init e820_reserve_setup_data(void)
 458 {
 459         struct setup_data *data;
 460         u64 pa_data;
 461         int found = 0;
 462
 463         pa_data = boot_params.hdr.setup_data;
 464         while (pa_data) {
 465                 data = early_memremap(pa_data, sizeof(*data));
 466                 e820_update_range(pa_data, sizeof(*data)+data->len,
 467                          E820_RAM, E820_RESERVED_KERN);
 468                 found = 1;
 469                 pa_data = data->next;
 470                 early_iounmap(data, sizeof(*data));
 471         }
 472         if (!found)
 473                 return;
 474
 475         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 476         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 477         printk(KERN_INFO "extended physical RAM map:\n");
 478         e820_print_map("reserve setup_data");
 479 }
 480
 481 static void __init memblock_x86_reserve_range_setup_data(void)
 482 {
 483         struct setup_data *data;
 484         u64 pa_data;
 485
 486         pa_data = boot_params.hdr.setup_data;
 487         while (pa_data) {
 488                 data = early_memremap(pa_data, sizeof(*data));
 489                 memblock_reserve(pa_data, sizeof(*data) + data->len);
 490                 pa_data = data->next;
 491                 early_iounmap(data, sizeof(*data));
 492         }
 493 }
 494
 495 /*
 496  * --------- Crashkernel reservation ------------------------------
 497  */
 498
 499 #ifdef CONFIG_KEXEC
 500
 501 /*
 502  * Keep the crash kernel below this limit.  On 32 bits earlier kernels
 503  * would limit the kernel to the low 512 MiB due to mapping restrictions.
 504  * On 64bit, old kexec-tools need to under 896MiB.
 505  */
 506 #ifdef CONFIG_X86_32
 507 # define CRASH_KERNEL_ADDR_LOW_MAX      (512 << 20)
 508 # define CRASH_KERNEL_ADDR_HIGH_MAX     (512 << 20)
 509 #else
 510 # define CRASH_KERNEL_ADDR_LOW_MAX      (896UL<<20)
 511 # define CRASH_KERNEL_ADDR_HIGH_MAX     MAXMEM
 512 #endif
 513
 514 static int __init reserve_crashkernel_low(void)
 515 {
 516 #ifdef CONFIG_X86_64
 517         const unsigned long long alignment = 16<<20;    /* 16M */
 518         unsigned long long low_base = 0, low_size = 0;
 519         unsigned long total_low_mem;
 520         unsigned long long base;
 521         bool auto_set = false;
 522         int ret;
 523
 524         total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT));
 525         /* crashkernel=Y,low */
 526         ret = parse_crashkernel_low(boot_command_line, total_low_mem,
 527                                                 &low_size, &base);
 528         if (ret != 0) {
 529                 /*
 530                  * two parts from lib/swiotlb.c:
 531                  *      swiotlb size: user specified with swiotlb= or default.
 532                  *      swiotlb overflow buffer: now is hardcoded to 32k.
 533                  *              We round it to 8M for other buffers that
 534                  *              may need to stay low too.
 535                  */
 536                 low_size = swiotlb_size_or_default() + (8UL<<20);
 537                 auto_set = true;
 538         } else {
 539                 /* passed with crashkernel=0,low ? */
 540                 if (!low_size)
 541                         return 0;
 542         }
 543
 544         low_base = memblock_find_in_range(low_size, (1ULL<<32),
 545                                         low_size, alignment);
 546
 547         if (!low_base) {
 548                 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
 549                        (unsigned long)(low_size >> 20));
 550                 return -ENOMEM;
 551         }
 552
 553         memblock_reserve(low_base, low_size);
 554         pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
 555                         (unsigned long)(low_size >> 20),
 556                         (unsigned long)(low_base >> 20),
 557                         (unsigned long)(total_low_mem >> 20));
 558         crashk_low_res.start = low_base;
 559         crashk_low_res.end   = low_base + low_size - 1;
 560         insert_resource(&iomem_resource, &crashk_low_res);
 561 #endif
 562         return 0;
 563 }
 564
 565 static void __init reserve_crashkernel(void)
 566 {
 567         const unsigned long long alignment = 16<<20;    /* 16M */
 568         unsigned long long total_mem;
 569         unsigned long long crash_size, crash_base;
 570         bool high = false;
 571         int ret;
 572
 573         total_mem = memblock_phys_mem_size();
 574
 575         /* crashkernel=XM */
 576         ret = parse_crashkernel(boot_command_line, total_mem,
 577                         &crash_size, &crash_base);
 578         if (ret != 0 || crash_size <= 0) {
 579                 /* crashkernel=X,high */
 580                 ret = parse_crashkernel_high(boot_command_line, total_mem,
 581                                 &crash_size, &crash_base);
 582                 if (ret != 0 || crash_size <= 0)
 583                         return;
 584                 high = true;
 585         }
 586
 587         /* 0 means: find the address automatically */
 588         if (crash_base <= 0) {
 589                 /*
 590                  *  kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
 591                  */
 592                 crash_base = memblock_find_in_range(alignment,
 593                                         high ? CRASH_KERNEL_ADDR_HIGH_MAX :
 594                                                CRASH_KERNEL_ADDR_LOW_MAX,
 595                                         crash_size, alignment);
 596
 597                 if (!crash_base) {
 598                         pr_info("crashkernel reservation failed - No suitable area found.\n");
 599                         return;
 600                 }
 601
 602         } else {
 603                 unsigned long long start;
 604
 605                 start = memblock_find_in_range(crash_base,
 606                                  crash_base + crash_size, crash_size, 1<<20);
 607                 if (start != crash_base) {
 608                         pr_info("crashkernel reservation failed - memory is in use.\n");
 609                         return;
 610                 }
 611         }
 612         memblock_reserve(crash_base, crash_size);
 613
 614         if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
 615                 memblock_free(crash_base, crash_size);
 616                 return;
 617         }
 618
 619         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 620                         "for crashkernel (System RAM: %ldMB)\n",
 621                         (unsigned long)(crash_size >> 20),
 622                         (unsigned long)(crash_base >> 20),
 623                         (unsigned long)(total_mem >> 20));
 624
 625         crashk_res.start = crash_base;
 626         crashk_res.end   = crash_base + crash_size - 1;
 627         insert_resource(&iomem_resource, &crashk_res);
 628 }
 629 #else
 630 static void __init reserve_crashkernel(void)
 631 {
 632 }
 633 #endif
 634
 635 static struct resource standard_io_resources[] = {
 636         { .name = "dma1", .start = 0x00, .end = 0x1f,
 637                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 638         { .name = "pic1", .start = 0x20, .end = 0x21,
 639                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 640         { .name = "timer0", .start = 0x40, .end = 0x43,
 641                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 642         { .name = "timer1", .start = 0x50, .end = 0x53,
 643                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 644         { .name = "keyboard", .start = 0x60, .end = 0x60,
 645                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 646         { .name = "keyboard", .start = 0x64, .end = 0x64,
 647                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 648         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 649                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 650         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 651                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 652         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 653                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 654         { .name = "fpu", .start = 0xf0, .end = 0xff,
 655                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 656 };
 657
 658 void __init reserve_standard_io_resources(void)
 659 {
 660         int i;
 661
 662         /* request I/O space for devices used on all i[345]86 PCs */
 663         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 664                 request_resource(&ioport_resource, &standard_io_resources[i]);
 665
 666 }
 667
 668 static __init void reserve_ibft_region(void)
 669 {
 670         unsigned long addr, size = 0;
 671
 672         addr = find_ibft_region(&size);
 673
 674         if (size)
 675                 memblock_reserve(addr, size);
 676 }
 677
 678 static bool __init snb_gfx_workaround_needed(void)
 679 {
 680 #ifdef CONFIG_PCI
 681         int i;
 682         u16 vendor, devid;
 683         static const __initconst u16 snb_ids[] = {
 684                 0x0102,
 685                 0x0112,
 686                 0x0122,
 687                 0x0106,
 688                 0x0116,
 689                 0x0126,
 690                 0x010a,
 691         };
 692
 693         /* Assume no if something weird is going on with PCI */
 694         if (!early_pci_allowed())
 695                 return false;
 696
 697         vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
 698         if (vendor != 0x8086)
 699                 return false;
 700
 701         devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
 702         for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
 703                 if (devid == snb_ids[i])
 704                         return true;
 705 #endif
 706
 707         return false;
 708 }
 709
 710 /*
 711  * Sandy Bridge graphics has trouble with certain ranges, exclude
 712  * them from allocation.
 713  */
 714 static void __init trim_snb_memory(void)
 715 {
 716         static const __initconst unsigned long bad_pages[] = {
 717                 0x20050000,
 718                 0x20110000,
 719                 0x20130000,
 720                 0x20138000,
 721                 0x40004000,
 722         };
 723         int i;
 724
 725         if (!snb_gfx_workaround_needed())
 726                 return;
 727
 728         printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
 729
 730         /*
 731          * Reserve all memory below the 1 MB mark that has not
 732          * already been reserved.
 733          */
 734         memblock_reserve(0, 1<<20);
 735
 736         for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
 737                 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
 738                         printk(KERN_WARNING "failed to reserve 0x%08lx\n",
 739                                bad_pages[i]);
 740         }
 741 }
 742
 743 /*
 744  * Here we put platform-specific memory range workarounds, i.e.
 745  * memory known to be corrupt or otherwise in need to be reserved on
 746  * specific platforms.
 747  *
 748  * If this gets used more widely it could use a real dispatch mechanism.
 749  */
 750 static void __init trim_platform_memory_ranges(void)
 751 {
 752         trim_snb_memory();
 753 }
 754
 755 static void __init trim_bios_range(void)
 756 {
 757         /*
 758          * A special case is the first 4Kb of memory;
 759          * This is a BIOS owned area, not kernel ram, but generally
 760          * not listed as such in the E820 table.
 761          *
 762          * This typically reserves additional memory (64KiB by default)
 763          * since some BIOSes are known to corrupt low memory.  See the
 764          * Kconfig help text for X86_RESERVE_LOW.
 765          */
 766         e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
 767
 768         /*
 769          * special case: Some BIOSen report the PC BIOS
 770          * area (640->1Mb) as ram even though it is not.
 771          * take them out.
 772          */
 773         e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
 774
 775         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 776 }
 777
 778 /* called before trim_bios_range() to spare extra sanitize */
 779 static void __init e820_add_kernel_range(void)
 780 {
 781         u64 start = __pa_symbol(_text);
 782         u64 size = __pa_symbol(_end) - start;
 783
 784         /*
 785          * Complain if .text .data and .bss are not marked as E820_RAM and
 786          * attempt to fix it by adding the range. We may have a confused BIOS,
 787          * or the user may have used memmap=exactmap or memmap=xxM$yyM to
 788          * exclude kernel range. If we really are running on top non-RAM,
 789          * we will crash later anyways.
 790          */
 791         if (e820_all_mapped(start, start + size, E820_RAM))
 792                 return;
 793
 794         pr_warn(".text .data .bss are not marked as E820_RAM!\n");
 795         e820_remove_range(start, size, E820_RAM, 0);
 796         e820_add_region(start, size, E820_RAM);
 797 }
 798
 799 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
 800
 801 static int __init parse_reservelow(char *p)
 802 {
 803         unsigned long long size;
 804
 805         if (!p)
 806                 return -EINVAL;
 807
 808         size = memparse(p, &p);
 809
 810         if (size < 4096)
 811                 size = 4096;
 812
 813         if (size > 640*1024)
 814                 size = 640*1024;
 815
 816         reserve_low = size;
 817
 818         return 0;
 819 }
 820
 821 early_param("reservelow", parse_reservelow);
 822
 823 static void __init trim_low_memory_range(void)
 824 {
 825         memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
 826 }
 827
 828 /*
 829  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 830  * passed the efi memmap, systab, etc., so we should use these data structures
 831  * for initialization.  Note, the efi init code path is determined by the
 832  * global efi_enabled. This allows the same kernel image to be used on existing
 833  * systems (with a traditional BIOS) as well as on EFI systems.
 834  */
 835 /*
 836  * setup_arch - architecture-specific boot-time initializations
 837  *
 838  * Note: On x86_64, fixmaps are ready for use even before this is called.
 839  */
 840
 841 void __init setup_arch(char **cmdline_p)
 842 {
 843         memblock_reserve(__pa_symbol(_text),
 844                          (unsigned long)__bss_stop - (unsigned long)_text);
 845
 846         early_reserve_initrd();
 847
 848         /*
 849          * At this point everything still needed from the boot loader
 850          * or BIOS or kernel text should be early reserved or marked not
 851          * RAM in e820. All other memory is free game.
 852          */
 853
 854 #ifdef CONFIG_X86_32
 855         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 856         visws_early_detect();
 857
 858         /*
 859          * copy kernel address range established so far and switch
 860          * to the proper swapper page table
 861          */
 862         clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
 863                         initial_page_table + KERNEL_PGD_BOUNDARY,
 864                         KERNEL_PGD_PTRS);
 865
 866         load_cr3(swapper_pg_dir);
 867         __flush_tlb_all();
 868 #else
 869         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 870 #endif
 871
 872         /*
 873          * If we have OLPC OFW, we might end up relocating the fixmap due to
 874          * reserve_top(), so do this before touching the ioremap area.
 875          */
 876         olpc_ofw_detect();
 877
 878         early_trap_init();
 879         early_cpu_init();
 880         early_ioremap_init();
 881
 882         setup_olpc_ofw_pgd();
 883
 884         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 885         screen_info = boot_params.screen_info;
 886         edid_info = boot_params.edid_info;
 887 #ifdef CONFIG_X86_32
 888         apm_info.bios = boot_params.apm_bios_info;
 889         ist_info = boot_params.ist_info;
 890         if (boot_params.sys_desc_table.length != 0) {
 891                 machine_id = boot_params.sys_desc_table.table[0];
 892                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 893                 BIOS_revision = boot_params.sys_desc_table.table[2];
 894         }
 895 #endif
 896         saved_video_mode = boot_params.hdr.vid_mode;
 897         bootloader_type = boot_params.hdr.type_of_loader;
 898         if ((bootloader_type >> 4) == 0xe) {
 899                 bootloader_type &= 0xf;
 900                 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
 901         }
 902         bootloader_version  = bootloader_type & 0xf;
 903         bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
 904
 905 #ifdef CONFIG_BLK_DEV_RAM
 906         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 907         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 908         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 909 #endif
 910 #ifdef CONFIG_EFI
 911         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 912                      "EL32", 4)) {
 913                 set_bit(EFI_BOOT, &x86_efi_facility);
 914         } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 915                      "EL64", 4)) {
 916                 set_bit(EFI_BOOT, &x86_efi_facility);
 917                 set_bit(EFI_64BIT, &x86_efi_facility);
 918         }
 919
 920         if (efi_enabled(EFI_BOOT))
 921                 efi_memblock_x86_reserve_range();
 922 #endif
 923
 924         x86_init.oem.arch_setup();
 925
 926         iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
 927         setup_memory_map();
 928         parse_setup_data();
 929         /* update the e820_saved too */
 930         e820_reserve_setup_data();
 931
 932         copy_edd();
 933
 934         if (!boot_params.hdr.root_flags)
 935                 root_mountflags &= ~MS_RDONLY;
 936         init_mm.start_code = (unsigned long) _text;
 937         init_mm.end_code = (unsigned long) _etext;
 938         init_mm.end_data = (unsigned long) _edata;
 939         init_mm.brk = _brk_end;
 940
 941         code_resource.start = __pa_symbol(_text);
 942         code_resource.end = __pa_symbol(_etext)-1;
 943         data_resource.start = __pa_symbol(_etext);
 944         data_resource.end = __pa_symbol(_edata)-1;
 945         bss_resource.start = __pa_symbol(__bss_start);
 946         bss_resource.end = __pa_symbol(__bss_stop)-1;
 947
 948 #ifdef CONFIG_CMDLINE_BOOL
 949 #ifdef CONFIG_CMDLINE_OVERRIDE
 950         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 951 #else
 952         if (builtin_cmdline[0]) {
 953                 /* append boot loader cmdline to builtin */
 954                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 955                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 956                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 957         }
 958 #endif
 959 #endif
 960
 961         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 962         *cmdline_p = command_line;
 963
 964         /*
 965          * x86_configure_nx() is called before parse_early_param() to detect
 966          * whether hardware doesn't support NX (so that the early EHCI debug
 967          * console setup can safely call set_fixmap()). It may then be called
 968          * again from within noexec_setup() during parsing early parameters
 969          * to honor the respective command line option.
 970          */
 971         x86_configure_nx();
 972
 973         parse_early_param();
 974
 975         x86_report_nx();
 976
 977         /* after early param, so could get panic from serial */
 978         memblock_x86_reserve_range_setup_data();
 979
 980         if (acpi_mps_check()) {
 981 #ifdef CONFIG_X86_LOCAL_APIC
 982                 disable_apic = 1;
 983 #endif
 984                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 985         }
 986
 987 #ifdef CONFIG_PCI
 988         if (pci_early_dump_regs)
 989                 early_dump_pci_devices();
 990 #endif
 991
 992         finish_e820_parsing();
 993
 994         if (efi_enabled(EFI_BOOT))
 995                 efi_init();
 996
 997         dmi_scan_machine();
 998         dmi_set_dump_stack_arch_desc();
 999
1000         /*
1001          * VMware detection requires dmi to be available, so this
1002          * needs to be done after dmi_scan_machine, for the BP.
1003          */
1004         init_hypervisor_platform();
1005
1006         x86_init.resources.probe_roms();
1007
1008         /* after parse_early_param, so could debug it */
1009         insert_resource(&iomem_resource, &code_resource);
1010         insert_resource(&iomem_resource, &data_resource);
1011         insert_resource(&iomem_resource, &bss_resource);
1012
1013         e820_add_kernel_range();
1014         trim_bios_range();
1015 #ifdef CONFIG_X86_32
1016         if (ppro_with_ram_bug()) {
1017                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1018                                   E820_RESERVED);
1019                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1020                 printk(KERN_INFO "fixed physical RAM map:\n");
1021                 e820_print_map("bad_ppro");
1022         }
1023 #else
1024         early_gart_iommu_check();
1025 #endif
1026
1027         /*
1028          * partially used pages are not usable - thus
1029          * we are rounding upwards:
1030          */
1031         max_pfn = e820_end_of_ram_pfn();
1032
1033         /* update e820 for memory not covered by WB MTRRs */
1034         mtrr_bp_init();
1035         if (mtrr_trim_uncached_memory(max_pfn))
1036                 max_pfn = e820_end_of_ram_pfn();
1037
1038 #ifdef CONFIG_X86_32
1039         /* max_low_pfn get updated here */
1040         find_low_pfn_range();
1041 #else
1042         check_x2apic();
1043
1044         /* How many end-of-memory variables you have, grandma! */
1045         /* need this before calling reserve_initrd */
1046         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1047                 max_low_pfn = e820_end_of_low_ram_pfn();
1048         else
1049                 max_low_pfn = max_pfn;
1050
1051         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1052 #endif
1053
1054         /*
1055          * Find and reserve possible boot-time SMP configuration:
1056          */
1057         find_smp_config();
1058
1059         reserve_ibft_region();
1060
1061         early_alloc_pgt_buf();
1062
1063         /*
1064          * Need to conclude brk, before memblock_x86_fill()
1065          *  it could use memblock_find_in_range, could overlap with
1066          *  brk area.
1067          */
1068         reserve_brk();
1069
1070         cleanup_highmap();
1071
1072         memblock_set_current_limit(ISA_END_ADDRESS);
1073         memblock_x86_fill();
1074
1075         /*
1076          * The EFI specification says that boot service code won't be called
1077          * after ExitBootServices(). This is, in fact, a lie.
1078          */
1079         if (efi_enabled(EFI_MEMMAP))
1080                 efi_reserve_boot_services();
1081
1082         /* preallocate 4k for mptable mpc */
1083         early_reserve_e820_mpc_new();
1084
1085 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1086         setup_bios_corruption_check();
1087 #endif
1088
1089 #ifdef CONFIG_X86_32
1090         printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1091                         (max_pfn_mapped<<PAGE_SHIFT) - 1);
1092 #endif
1093
1094         reserve_real_mode();
1095
1096         trim_platform_memory_ranges();
1097         trim_low_memory_range();
1098
1099         init_mem_mapping();
1100
1101         early_trap_pf_init();
1102
1103         setup_real_mode();
1104
1105         memblock_set_current_limit(get_max_mapped());
1106         dma_contiguous_reserve(0);
1107
1108         /*
1109          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1110          */
1111
1112 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1113         if (init_ohci1394_dma_early)
1114                 init_ohci1394_dma_on_all_controllers();
1115 #endif
1116         /* Allocate bigger log buffer */
1117         setup_log_buf(1);
1118
1119         reserve_initrd();
1120
1121 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1122         acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1123 #endif
1124
1125         reserve_crashkernel();
1126
1127         vsmp_init();
1128
1129         io_delay_init();
1130
1131         /*
1132          * Parse the ACPI tables for possible boot-time SMP configuration.
1133          */
1134         acpi_boot_table_init();
1135
1136         early_acpi_boot_init();
1137
1138         initmem_init();
1139         memblock_find_dma_reserve();
1140
1141 #ifdef CONFIG_KVM_GUEST
1142         kvmclock_init();
1143 #endif
1144
1145         x86_init.paging.pagetable_init();
1146
1147         if (boot_cpu_data.cpuid_level >= 0) {
1148                 /* A CPU has %cr4 if and only if it has CPUID */
1149                 mmu_cr4_features = read_cr4();
1150                 if (trampoline_cr4_features)
1151                         *trampoline_cr4_features = mmu_cr4_features;
1152         }
1153
1154 #ifdef CONFIG_X86_32
1155         /* sync back kernel address range */
1156         clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1157                         swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
1158                         KERNEL_PGD_PTRS);
1159
1160         /*
1161          * sync back low identity map too.  It is used for example
1162          * in the 32-bit EFI stub.
1163          */
1164         clone_pgd_range(initial_page_table,
1165                         swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
1166                         min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1167 #endif
1168
1169         tboot_probe();
1170
1171 #ifdef CONFIG_X86_64
1172         map_vsyscall();
1173 #endif
1174
1175         generic_apic_probe();
1176
1177         early_quirks();
1178
1179         /*
1180          * Read APIC and some other early information from ACPI tables.
1181          */
1182         acpi_boot_init();
1183         sfi_init();
1184         x86_dtb_init();
1185
1186         /*
1187          * get boot-time SMP configuration:
1188          */
1189         if (smp_found_config)
1190                 get_smp_config();
1191
1192         prefill_possible_map();
1193
1194         init_cpu_to_node();
1195
1196         init_apic_mappings();
1197         if (x86_io_apic_ops.init)
1198                 x86_io_apic_ops.init();
1199
1200         kvm_guest_init();
1201
1202         e820_reserve_resources();
1203         e820_mark_nosave_regions(max_low_pfn);
1204
1205         x86_init.resources.reserve_resources();
1206
1207         e820_setup_gap();
1208
1209 #ifdef CONFIG_VT
1210 #if defined(CONFIG_VGA_CONSOLE)
1211         if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1212                 conswitchp = &vga_con;
1213 #elif defined(CONFIG_DUMMY_CONSOLE)
1214         conswitchp = &dummy_con;
1215 #endif
1216 #endif
1217         x86_init.oem.banner();
1218
1219         x86_init.timers.wallclock_init();
1220
1221         mcheck_init();
1222
1223         arch_init_ideal_nops();
1224
1225         register_refined_jiffies(CLOCK_TICK_RATE);
1226
1227 #ifdef CONFIG_EFI
1228         /* Once setup is done above, unmap the EFI memory map on
1229          * mismatched firmware/kernel archtectures since there is no
1230          * support for runtime services.
1231          */
1232         if (efi_enabled(EFI_BOOT) && !efi_is_native()) {
1233                 pr_info("efi: Setup done, disabling due to 32/64-bit mismatch\n");
1234                 efi_unmap_memmap();
1235         }
1236 #endif
1237 }
1238
1239 #ifdef CONFIG_X86_32
1240
1241 static struct resource video_ram_resource = {
1242         .name   = "Video RAM area",
1243         .start  = 0xa0000,
1244         .end    = 0xbffff,
1245         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
1246 };
1247
1248 void __init i386_reserve_resources(void)
1249 {
1250         request_resource(&iomem_resource, &video_ram_resource);
1251         reserve_standard_io_resources();
1252 }
1253
1254 #endif /* CONFIG_X86_32 */