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/mca.h>
  38 #include <linux/root_dev.h>
  39 #include <linux/highmem.h>
  40 #include <linux/module.h>
  41 #include <linux/efi.h>
  42 #include <linux/init.h>
  43 #include <linux/edd.h>
  44 #include <linux/iscsi_ibft.h>
  45 #include <linux/nodemask.h>
  46 #include <linux/kexec.h>
  47 #include <linux/dmi.h>
  48 #include <linux/pfn.h>
  49 #include <linux/pci.h>
  50 #include <asm/pci-direct.h>
  51 #include <linux/init_ohci1394_dma.h>
  52 #include <linux/kvm_para.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
  72 #include <video/edid.h>
  73
  74 #include <asm/mtrr.h>
  75 #include <asm/apic.h>
  76 #include <asm/trampoline.h>
  77 #include <asm/e820.h>
  78 #include <asm/mpspec.h>
  79 #include <asm/setup.h>
  80 #include <asm/efi.h>
  81 #include <asm/timer.h>
  82 #include <asm/i8259.h>
  83 #include <asm/sections.h>
  84 #include <asm/dmi.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/system.h>
  94 #include <asm/vsyscall.h>
  95 #include <asm/cpu.h>
  96 #include <asm/desc.h>
  97 #include <asm/dma.h>
  98 #include <asm/iommu.h>
  99 #include <asm/gart.h>
 100 #include <asm/mmu_context.h>
 101 #include <asm/proto.h>
 102
 103 #include <asm/paravirt.h>
 104 #include <asm/hypervisor.h>
 105 #include <asm/olpc_ofw.h>
 106
 107 #include <asm/percpu.h>
 108 #include <asm/topology.h>
 109 #include <asm/apicdef.h>
 110 #include <asm/amd_nb.h>
 111 #ifdef CONFIG_X86_64
 112 #include <asm/numa_64.h>
 113 #endif
 114 #include <asm/mce.h>
 115 #include <asm/alternative.h>
 116 #include <asm/prom.h>
 117
 118 /*
 119  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
 120  * The direct mapping extends to max_pfn_mapped, so that we can directly access
 121  * apertures, ACPI and other tables without having to play with fixmaps.
 122  */
 123 unsigned long max_low_pfn_mapped;
 124 unsigned long max_pfn_mapped;
 125
 126 #ifdef CONFIG_DMI
 127 RESERVE_BRK(dmi_alloc, 65536);
 128 #endif
 129
 130
 131 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
 132 unsigned long _brk_end = (unsigned long)__brk_base;
 133
 134 #ifdef CONFIG_X86_64
 135 int default_cpu_present_to_apicid(int mps_cpu)
 136 {
 137         return __default_cpu_present_to_apicid(mps_cpu);
 138 }
 139
 140 int default_check_phys_apicid_present(int phys_apicid)
 141 {
 142         return __default_check_phys_apicid_present(phys_apicid);
 143 }
 144 #endif
 145
 146 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 147 struct boot_params __initdata boot_params;
 148 #else
 149 struct boot_params boot_params;
 150 #endif
 151
 152 /*
 153  * Machine setup..
 154  */
 155 static struct resource data_resource = {
 156         .name   = "Kernel data",
 157         .start  = 0,
 158         .end    = 0,
 159         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 160 };
 161
 162 static struct resource code_resource = {
 163         .name   = "Kernel code",
 164         .start  = 0,
 165         .end    = 0,
 166         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 167 };
 168
 169 static struct resource bss_resource = {
 170         .name   = "Kernel bss",
 171         .start  = 0,
 172         .end    = 0,
 173         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 174 };
 175
 176
 177 #ifdef CONFIG_X86_32
 178 /* cpu data as detected by the assembly code in head.S */
 179 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 180 /* common cpu data for all cpus */
 181 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
 182 EXPORT_SYMBOL(boot_cpu_data);
 183 static void set_mca_bus(int x)
 184 {
 185 #ifdef CONFIG_MCA
 186         MCA_bus = x;
 187 #endif
 188 }
 189
 190 unsigned int def_to_bigsmp;
 191
 192 /* for MCA, but anyone else can use it if they want */
 193 unsigned int machine_id;
 194 unsigned int machine_submodel_id;
 195 unsigned int BIOS_revision;
 196
 197 struct apm_info apm_info;
 198 EXPORT_SYMBOL(apm_info);
 199
 200 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 201         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 202 struct ist_info ist_info;
 203 EXPORT_SYMBOL(ist_info);
 204 #else
 205 struct ist_info ist_info;
 206 #endif
 207
 208 #else
 209 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
 210         .x86_phys_bits = MAX_PHYSMEM_BITS,
 211 };
 212 EXPORT_SYMBOL(boot_cpu_data);
 213 #endif
 214
 215
 216 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
 217 unsigned long mmu_cr4_features;
 218 #else
 219 unsigned long mmu_cr4_features = X86_CR4_PAE;
 220 #endif
 221
 222 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
 223 int bootloader_type, bootloader_version;
 224
 225 /*
 226  * Setup options
 227  */
 228 struct screen_info screen_info;
 229 EXPORT_SYMBOL(screen_info);
 230 struct edid_info edid_info;
 231 EXPORT_SYMBOL_GPL(edid_info);
 232
 233 extern int root_mountflags;
 234
 235 unsigned long saved_video_mode;
 236
 237 #define RAMDISK_IMAGE_START_MASK        0x07FF
 238 #define RAMDISK_PROMPT_FLAG             0x8000
 239 #define RAMDISK_LOAD_FLAG               0x4000
 240
 241 static char __initdata command_line[COMMAND_LINE_SIZE];
 242 #ifdef CONFIG_CMDLINE_BOOL
 243 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
 244 #endif
 245
 246 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 247 struct edd edd;
 248 #ifdef CONFIG_EDD_MODULE
 249 EXPORT_SYMBOL(edd);
 250 #endif
 251 /**
 252  * copy_edd() - Copy the BIOS EDD information
 253  *              from boot_params into a safe place.
 254  *
 255  */
 256 static inline void __init copy_edd(void)
 257 {
 258      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 259             sizeof(edd.mbr_signature));
 260      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 261      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 262      edd.edd_info_nr = boot_params.eddbuf_entries;
 263 }
 264 #else
 265 static inline void __init copy_edd(void)
 266 {
 267 }
 268 #endif
 269
 270 void * __init extend_brk(size_t size, size_t align)
 271 {
 272         size_t mask = align - 1;
 273         void *ret;
 274
 275         BUG_ON(_brk_start == 0);
 276         BUG_ON(align & mask);
 277
 278         _brk_end = (_brk_end + mask) & ~mask;
 279         BUG_ON((char *)(_brk_end + size) > __brk_limit);
 280
 281         ret = (void *)_brk_end;
 282         _brk_end += size;
 283
 284         memset(ret, 0, size);
 285
 286         return ret;
 287 }
 288
 289 #ifdef CONFIG_X86_64
 290 static void __init init_gbpages(void)
 291 {
 292         if (direct_gbpages && cpu_has_gbpages)
 293                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 294         else
 295                 direct_gbpages = 0;
 296 }
 297 #else
 298 static inline void init_gbpages(void)
 299 {
 300 }
 301 static void __init cleanup_highmap(void)
 302 {
 303 }
 304 #endif
 305
 306 static void __init reserve_brk(void)
 307 {
 308         if (_brk_end > _brk_start)
 309                 memblock_x86_reserve_range(__pa(_brk_start), __pa(_brk_end), "BRK");
 310
 311         /* Mark brk area as locked down and no longer taking any
 312            new allocations */
 313         _brk_start = 0;
 314 }
 315
 316 #ifdef CONFIG_BLK_DEV_INITRD
 317
 318 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 319 static void __init relocate_initrd(void)
 320 {
 321         /* Assume only end is not page aligned */
 322         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 323         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 324         u64 area_size     = PAGE_ALIGN(ramdisk_size);
 325         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 326         u64 ramdisk_here;
 327         unsigned long slop, clen, mapaddr;
 328         char *p, *q;
 329
 330         /* We need to move the initrd down into lowmem */
 331         ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size,
 332                                          PAGE_SIZE);
 333
 334         if (ramdisk_here == MEMBLOCK_ERROR)
 335                 panic("Cannot find place for new RAMDISK of size %lld\n",
 336                          ramdisk_size);
 337
 338         /* Note: this includes all the lowmem currently occupied by
 339            the initrd, we rely on that fact to keep the data intact. */
 340         memblock_x86_reserve_range(ramdisk_here, ramdisk_here + area_size, "NEW RAMDISK");
 341         initrd_start = ramdisk_here + PAGE_OFFSET;
 342         initrd_end   = initrd_start + ramdisk_size;
 343         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 344                          ramdisk_here, ramdisk_here + ramdisk_size);
 345
 346         q = (char *)initrd_start;
 347
 348         /* Copy any lowmem portion of the initrd */
 349         if (ramdisk_image < end_of_lowmem) {
 350                 clen = end_of_lowmem - ramdisk_image;
 351                 p = (char *)__va(ramdisk_image);
 352                 memcpy(q, p, clen);
 353                 q += clen;
 354                 ramdisk_image += clen;
 355                 ramdisk_size  -= clen;
 356         }
 357
 358         /* Copy the highmem portion of the initrd */
 359         while (ramdisk_size) {
 360                 slop = ramdisk_image & ~PAGE_MASK;
 361                 clen = ramdisk_size;
 362                 if (clen > MAX_MAP_CHUNK-slop)
 363                         clen = MAX_MAP_CHUNK-slop;
 364                 mapaddr = ramdisk_image & PAGE_MASK;
 365                 p = early_memremap(mapaddr, clen+slop);
 366                 memcpy(q, p+slop, clen);
 367                 early_iounmap(p, clen+slop);
 368                 q += clen;
 369                 ramdisk_image += clen;
 370                 ramdisk_size  -= clen;
 371         }
 372         /* high pages is not converted by early_res_to_bootmem */
 373         ramdisk_image = boot_params.hdr.ramdisk_image;
 374         ramdisk_size  = boot_params.hdr.ramdisk_size;
 375         printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
 376                 " %08llx - %08llx\n",
 377                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 378                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 379 }
 380
 381 static void __init reserve_initrd(void)
 382 {
 383         /* Assume only end is not page aligned */
 384         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 385         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 386         u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
 387         u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
 388
 389         if (!boot_params.hdr.type_of_loader ||
 390             !ramdisk_image || !ramdisk_size)
 391                 return;         /* No initrd provided by bootloader */
 392
 393         initrd_start = 0;
 394
 395         if (ramdisk_size >= (end_of_lowmem>>1)) {
 396                 memblock_x86_free_range(ramdisk_image, ramdisk_end);
 397                 printk(KERN_ERR "initrd too large to handle, "
 398                        "disabling initrd\n");
 399                 return;
 400         }
 401
 402         printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
 403                         ramdisk_end);
 404
 405
 406         if (ramdisk_end <= end_of_lowmem) {
 407                 /* All in lowmem, easy case */
 408                 /*
 409                  * don't need to reserve again, already reserved early
 410                  * in i386_start_kernel
 411                  */
 412                 initrd_start = ramdisk_image + PAGE_OFFSET;
 413                 initrd_end = initrd_start + ramdisk_size;
 414                 return;
 415         }
 416
 417         relocate_initrd();
 418
 419         memblock_x86_free_range(ramdisk_image, ramdisk_end);
 420 }
 421 #else
 422 static void __init reserve_initrd(void)
 423 {
 424 }
 425 #endif /* CONFIG_BLK_DEV_INITRD */
 426
 427 static void __init parse_setup_data(void)
 428 {
 429         struct setup_data *data;
 430         u64 pa_data;
 431
 432         if (boot_params.hdr.version < 0x0209)
 433                 return;
 434         pa_data = boot_params.hdr.setup_data;
 435         while (pa_data) {
 436                 u32 data_len, map_len;
 437
 438                 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
 439                               (u64)sizeof(struct setup_data));
 440                 data = early_memremap(pa_data, map_len);
 441                 data_len = data->len + sizeof(struct setup_data);
 442                 if (data_len > map_len) {
 443                         early_iounmap(data, map_len);
 444                         data = early_memremap(pa_data, data_len);
 445                         map_len = data_len;
 446                 }
 447
 448                 switch (data->type) {
 449                 case SETUP_E820_EXT:
 450                         parse_e820_ext(data);
 451                         break;
 452                 case SETUP_DTB:
 453                         add_dtb(pa_data);
 454                         break;
 455                 default:
 456                         break;
 457                 }
 458                 pa_data = data->next;
 459                 early_iounmap(data, map_len);
 460         }
 461 }
 462
 463 static void __init e820_reserve_setup_data(void)
 464 {
 465         struct setup_data *data;
 466         u64 pa_data;
 467         int found = 0;
 468
 469         if (boot_params.hdr.version < 0x0209)
 470                 return;
 471         pa_data = boot_params.hdr.setup_data;
 472         while (pa_data) {
 473                 data = early_memremap(pa_data, sizeof(*data));
 474                 e820_update_range(pa_data, sizeof(*data)+data->len,
 475                          E820_RAM, E820_RESERVED_KERN);
 476                 found = 1;
 477                 pa_data = data->next;
 478                 early_iounmap(data, sizeof(*data));
 479         }
 480         if (!found)
 481                 return;
 482
 483         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 484         memcpy(&e820_saved, &e820, sizeof(struct e820map));
 485         printk(KERN_INFO "extended physical RAM map:\n");
 486         e820_print_map("reserve setup_data");
 487 }
 488
 489 static void __init memblock_x86_reserve_range_setup_data(void)
 490 {
 491         struct setup_data *data;
 492         u64 pa_data;
 493         char buf[32];
 494
 495         if (boot_params.hdr.version < 0x0209)
 496                 return;
 497         pa_data = boot_params.hdr.setup_data;
 498         while (pa_data) {
 499                 data = early_memremap(pa_data, sizeof(*data));
 500                 sprintf(buf, "setup data %x", data->type);
 501                 memblock_x86_reserve_range(pa_data, pa_data+sizeof(*data)+data->len, buf);
 502                 pa_data = data->next;
 503                 early_iounmap(data, sizeof(*data));
 504         }
 505 }
 506
 507 /*
 508  * --------- Crashkernel reservation ------------------------------
 509  */
 510
 511 #ifdef CONFIG_KEXEC
 512
 513 static inline unsigned long long get_total_mem(void)
 514 {
 515         unsigned long long total;
 516
 517         total = max_pfn - min_low_pfn;
 518
 519         return total << PAGE_SHIFT;
 520 }
 521
 522 /*
 523  * Keep the crash kernel below this limit.  On 32 bits earlier kernels
 524  * would limit the kernel to the low 512 MiB due to mapping restrictions.
 525  * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this
 526  * limit once kexec-tools are fixed.
 527  */
 528 #ifdef CONFIG_X86_32
 529 # define CRASH_KERNEL_ADDR_MAX  (512 << 20)
 530 #else
 531 # define CRASH_KERNEL_ADDR_MAX  (896 << 20)
 532 #endif
 533
 534 static void __init reserve_crashkernel(void)
 535 {
 536         unsigned long long total_mem;
 537         unsigned long long crash_size, crash_base;
 538         int ret;
 539
 540         total_mem = get_total_mem();
 541
 542         ret = parse_crashkernel(boot_command_line, total_mem,
 543                         &crash_size, &crash_base);
 544         if (ret != 0 || crash_size <= 0)
 545                 return;
 546
 547         /* 0 means: find the address automatically */
 548         if (crash_base <= 0) {
 549                 const unsigned long long alignment = 16<<20;    /* 16M */
 550
 551                 /*
 552                  *  kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
 553                  */
 554                 crash_base = memblock_find_in_range(alignment,
 555                                CRASH_KERNEL_ADDR_MAX, crash_size, alignment);
 556
 557                 if (crash_base == MEMBLOCK_ERROR) {
 558                         pr_info("crashkernel reservation failed - No suitable area found.\n");
 559                         return;
 560                 }
 561         } else {
 562                 unsigned long long start;
 563
 564                 start = memblock_find_in_range(crash_base,
 565                                  crash_base + crash_size, crash_size, 1<<20);
 566                 if (start != crash_base) {
 567                         pr_info("crashkernel reservation failed - memory is in use.\n");
 568                         return;
 569                 }
 570         }
 571         memblock_x86_reserve_range(crash_base, crash_base + crash_size, "CRASH KERNEL");
 572
 573         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 574                         "for crashkernel (System RAM: %ldMB)\n",
 575                         (unsigned long)(crash_size >> 20),
 576                         (unsigned long)(crash_base >> 20),
 577                         (unsigned long)(total_mem >> 20));
 578
 579         crashk_res.start = crash_base;
 580         crashk_res.end   = crash_base + crash_size - 1;
 581         insert_resource(&iomem_resource, &crashk_res);
 582 }
 583 #else
 584 static void __init reserve_crashkernel(void)
 585 {
 586 }
 587 #endif
 588
 589 static struct resource standard_io_resources[] = {
 590         { .name = "dma1", .start = 0x00, .end = 0x1f,
 591                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 592         { .name = "pic1", .start = 0x20, .end = 0x21,
 593                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 594         { .name = "timer0", .start = 0x40, .end = 0x43,
 595                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 596         { .name = "timer1", .start = 0x50, .end = 0x53,
 597                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 598         { .name = "keyboard", .start = 0x60, .end = 0x60,
 599                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 600         { .name = "keyboard", .start = 0x64, .end = 0x64,
 601                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 602         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 603                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 604         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 605                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 606         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 607                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 608         { .name = "fpu", .start = 0xf0, .end = 0xff,
 609                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 610 };
 611
 612 void __init reserve_standard_io_resources(void)
 613 {
 614         int i;
 615
 616         /* request I/O space for devices used on all i[345]86 PCs */
 617         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 618                 request_resource(&ioport_resource, &standard_io_resources[i]);
 619
 620 }
 621
 622 static __init void reserve_ibft_region(void)
 623 {
 624         unsigned long addr, size = 0;
 625
 626         addr = find_ibft_region(&size);
 627
 628         if (size)
 629                 memblock_x86_reserve_range(addr, addr + size, "* ibft");
 630 }
 631
 632 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
 633
 634 static void __init trim_bios_range(void)
 635 {
 636         /*
 637          * A special case is the first 4Kb of memory;
 638          * This is a BIOS owned area, not kernel ram, but generally
 639          * not listed as such in the E820 table.
 640          *
 641          * This typically reserves additional memory (64KiB by default)
 642          * since some BIOSes are known to corrupt low memory.  See the
 643          * Kconfig help text for X86_RESERVE_LOW.
 644          */
 645         e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE),
 646                           E820_RAM, E820_RESERVED);
 647
 648         /*
 649          * special case: Some BIOSen report the PC BIOS
 650          * area (640->1Mb) as ram even though it is not.
 651          * take them out.
 652          */
 653         e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
 654         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 655 }
 656
 657 static int __init parse_reservelow(char *p)
 658 {
 659         unsigned long long size;
 660
 661         if (!p)
 662                 return -EINVAL;
 663
 664         size = memparse(p, &p);
 665
 666         if (size < 4096)
 667                 size = 4096;
 668
 669         if (size > 640*1024)
 670                 size = 640*1024;
 671
 672         reserve_low = size;
 673
 674         return 0;
 675 }
 676
 677 early_param("reservelow", parse_reservelow);
 678
 679 /*
 680  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 681  * passed the efi memmap, systab, etc., so we should use these data structures
 682  * for initialization.  Note, the efi init code path is determined by the
 683  * global efi_enabled. This allows the same kernel image to be used on existing
 684  * systems (with a traditional BIOS) as well as on EFI systems.
 685  */
 686 /*
 687  * setup_arch - architecture-specific boot-time initializations
 688  *
 689  * Note: On x86_64, fixmaps are ready for use even before this is called.
 690  */
 691
 692 void __init setup_arch(char **cmdline_p)
 693 {
 694 #ifdef CONFIG_X86_32
 695         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 696         visws_early_detect();
 697
 698         /*
 699          * copy kernel address range established so far and switch
 700          * to the proper swapper page table
 701          */
 702         clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
 703                         initial_page_table + KERNEL_PGD_BOUNDARY,
 704                         KERNEL_PGD_PTRS);
 705
 706         load_cr3(swapper_pg_dir);
 707         __flush_tlb_all();
 708 #else
 709         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 710 #endif
 711
 712         /*
 713          * If we have OLPC OFW, we might end up relocating the fixmap due to
 714          * reserve_top(), so do this before touching the ioremap area.
 715          */
 716         olpc_ofw_detect();
 717
 718         early_trap_init();
 719         early_cpu_init();
 720         early_ioremap_init();
 721
 722         setup_olpc_ofw_pgd();
 723
 724         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 725         screen_info = boot_params.screen_info;
 726         edid_info = boot_params.edid_info;
 727 #ifdef CONFIG_X86_32
 728         apm_info.bios = boot_params.apm_bios_info;
 729         ist_info = boot_params.ist_info;
 730         if (boot_params.sys_desc_table.length != 0) {
 731                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 732                 machine_id = boot_params.sys_desc_table.table[0];
 733                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 734                 BIOS_revision = boot_params.sys_desc_table.table[2];
 735         }
 736 #endif
 737         saved_video_mode = boot_params.hdr.vid_mode;
 738         bootloader_type = boot_params.hdr.type_of_loader;
 739         if ((bootloader_type >> 4) == 0xe) {
 740                 bootloader_type &= 0xf;
 741                 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
 742         }
 743         bootloader_version  = bootloader_type & 0xf;
 744         bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
 745
 746 #ifdef CONFIG_BLK_DEV_RAM
 747         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 748         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 749         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 750 #endif
 751 #ifdef CONFIG_EFI
 752         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 753 #ifdef CONFIG_X86_32
 754                      "EL32",
 755 #else
 756                      "EL64",
 757 #endif
 758          4)) {
 759                 efi_enabled = 1;
 760                 efi_memblock_x86_reserve_range();
 761         }
 762 #endif
 763
 764         x86_init.oem.arch_setup();
 765
 766         iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
 767         setup_memory_map();
 768         parse_setup_data();
 769         /* update the e820_saved too */
 770         e820_reserve_setup_data();
 771
 772         copy_edd();
 773
 774         if (!boot_params.hdr.root_flags)
 775                 root_mountflags &= ~MS_RDONLY;
 776         init_mm.start_code = (unsigned long) _text;
 777         init_mm.end_code = (unsigned long) _etext;
 778         init_mm.end_data = (unsigned long) _edata;
 779         init_mm.brk = _brk_end;
 780
 781         code_resource.start = virt_to_phys(_text);
 782         code_resource.end = virt_to_phys(_etext)-1;
 783         data_resource.start = virt_to_phys(_etext);
 784         data_resource.end = virt_to_phys(_edata)-1;
 785         bss_resource.start = virt_to_phys(&__bss_start);
 786         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 787
 788 #ifdef CONFIG_CMDLINE_BOOL
 789 #ifdef CONFIG_CMDLINE_OVERRIDE
 790         strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 791 #else
 792         if (builtin_cmdline[0]) {
 793                 /* append boot loader cmdline to builtin */
 794                 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
 795                 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 796                 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
 797         }
 798 #endif
 799 #endif
 800
 801         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 802         *cmdline_p = command_line;
 803
 804         /*
 805          * x86_configure_nx() is called before parse_early_param() to detect
 806          * whether hardware doesn't support NX (so that the early EHCI debug
 807          * console setup can safely call set_fixmap()). It may then be called
 808          * again from within noexec_setup() during parsing early parameters
 809          * to honor the respective command line option.
 810          */
 811         x86_configure_nx();
 812
 813         parse_early_param();
 814
 815         x86_report_nx();
 816
 817         /* after early param, so could get panic from serial */
 818         memblock_x86_reserve_range_setup_data();
 819
 820         if (acpi_mps_check()) {
 821 #ifdef CONFIG_X86_LOCAL_APIC
 822                 disable_apic = 1;
 823 #endif
 824                 setup_clear_cpu_cap(X86_FEATURE_APIC);
 825         }
 826
 827 #ifdef CONFIG_PCI
 828         if (pci_early_dump_regs)
 829                 early_dump_pci_devices();
 830 #endif
 831
 832         finish_e820_parsing();
 833
 834         if (efi_enabled)
 835                 efi_init();
 836
 837         dmi_scan_machine();
 838
 839         /*
 840          * VMware detection requires dmi to be available, so this
 841          * needs to be done after dmi_scan_machine, for the BP.
 842          */
 843         init_hypervisor_platform();
 844
 845         x86_init.resources.probe_roms();
 846
 847         /* after parse_early_param, so could debug it */
 848         insert_resource(&iomem_resource, &code_resource);
 849         insert_resource(&iomem_resource, &data_resource);
 850         insert_resource(&iomem_resource, &bss_resource);
 851
 852         trim_bios_range();
 853 #ifdef CONFIG_X86_32
 854         if (ppro_with_ram_bug()) {
 855                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 856                                   E820_RESERVED);
 857                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 858                 printk(KERN_INFO "fixed physical RAM map:\n");
 859                 e820_print_map("bad_ppro");
 860         }
 861 #else
 862         early_gart_iommu_check();
 863 #endif
 864
 865         /*
 866          * partially used pages are not usable - thus
 867          * we are rounding upwards:
 868          */
 869         max_pfn = e820_end_of_ram_pfn();
 870
 871         /* update e820 for memory not covered by WB MTRRs */
 872         mtrr_bp_init();
 873         if (mtrr_trim_uncached_memory(max_pfn))
 874                 max_pfn = e820_end_of_ram_pfn();
 875
 876 #ifdef CONFIG_X86_32
 877         /* max_low_pfn get updated here */
 878         find_low_pfn_range();
 879 #else
 880         num_physpages = max_pfn;
 881
 882         check_x2apic();
 883
 884         /* How many end-of-memory variables you have, grandma! */
 885         /* need this before calling reserve_initrd */
 886         if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
 887                 max_low_pfn = e820_end_of_low_ram_pfn();
 888         else
 889                 max_low_pfn = max_pfn;
 890
 891         high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 892 #endif
 893
 894         /*
 895          * Find and reserve possible boot-time SMP configuration:
 896          */
 897         find_smp_config();
 898
 899         reserve_ibft_region();
 900
 901         /*
 902          * Need to conclude brk, before memblock_x86_fill()
 903          *  it could use memblock_find_in_range, could overlap with
 904          *  brk area.
 905          */
 906         reserve_brk();
 907
 908         cleanup_highmap();
 909
 910         memblock.current_limit = get_max_mapped();
 911         memblock_x86_fill();
 912
 913         /*
 914          * The EFI specification says that boot service code won't be called
 915          * after ExitBootServices(). This is, in fact, a lie.
 916          */
 917         if (efi_enabled)
 918                 efi_reserve_boot_services();
 919
 920         /* preallocate 4k for mptable mpc */
 921         early_reserve_e820_mpc_new();
 922
 923 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
 924         setup_bios_corruption_check();
 925 #endif
 926
 927         printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
 928                         max_pfn_mapped<<PAGE_SHIFT);
 929
 930         setup_trampolines();
 931
 932         init_gbpages();
 933
 934         /* max_pfn_mapped is updated here */
 935         max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
 936         max_pfn_mapped = max_low_pfn_mapped;
 937
 938 #ifdef CONFIG_X86_64
 939         if (max_pfn > max_low_pfn) {
 940                 max_pfn_mapped = init_memory_mapping(1UL<<32,
 941                                                      max_pfn<<PAGE_SHIFT);
 942                 /* can we preseve max_low_pfn ?*/
 943                 max_low_pfn = max_pfn;
 944         }
 945 #endif
 946         memblock.current_limit = get_max_mapped();
 947
 948         /*
 949          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 950          */
 951
 952 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 953         if (init_ohci1394_dma_early)
 954                 init_ohci1394_dma_on_all_controllers();
 955 #endif
 956         /* Allocate bigger log buffer */
 957         setup_log_buf(1);
 958
 959         reserve_initrd();
 960
 961         reserve_crashkernel();
 962
 963         vsmp_init();
 964
 965         io_delay_init();
 966
 967         /*
 968          * Parse the ACPI tables for possible boot-time SMP configuration.
 969          */
 970         acpi_boot_table_init();
 971
 972         early_acpi_boot_init();
 973
 974         initmem_init();
 975         memblock_find_dma_reserve();
 976
 977 #ifdef CONFIG_KVM_CLOCK
 978         kvmclock_init();
 979 #endif
 980
 981         x86_init.paging.pagetable_setup_start(swapper_pg_dir);
 982         paging_init();
 983         x86_init.paging.pagetable_setup_done(swapper_pg_dir);
 984
 985         if (boot_cpu_data.cpuid_level >= 0) {
 986                 /* A CPU has %cr4 if and only if it has CPUID */
 987                 mmu_cr4_features = read_cr4();
 988         }
 989
 990 #ifdef CONFIG_X86_32
 991         /* sync back kernel address range */
 992         clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
 993                         swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
 994                         KERNEL_PGD_PTRS);
 995 #endif
 996
 997         tboot_probe();
 998
 999 #ifdef CONFIG_X86_64
1000         map_vsyscall();
1001 #endif
1002
1003         generic_apic_probe();
1004
1005         early_quirks();
1006
1007         /*
1008          * Read APIC and some other early information from ACPI tables.
1009          */
1010         acpi_boot_init();
1011         sfi_init();
1012         x86_dtb_init();
1013
1014         /*
1015          * get boot-time SMP configuration:
1016          */
1017         if (smp_found_config)
1018                 get_smp_config();
1019
1020         prefill_possible_map();
1021
1022         init_cpu_to_node();
1023
1024         init_apic_mappings();
1025         ioapic_and_gsi_init();
1026
1027         kvm_guest_init();
1028
1029         e820_reserve_resources();
1030         e820_mark_nosave_regions(max_low_pfn);
1031
1032         x86_init.resources.reserve_resources();
1033
1034         e820_setup_gap();
1035
1036 #ifdef CONFIG_VT
1037 #if defined(CONFIG_VGA_CONSOLE)
1038         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1039                 conswitchp = &vga_con;
1040 #elif defined(CONFIG_DUMMY_CONSOLE)
1041         conswitchp = &dummy_con;
1042 #endif
1043 #endif
1044         x86_init.oem.banner();
1045
1046         x86_init.timers.wallclock_init();
1047
1048         mcheck_init();
1049
1050         arch_init_ideal_nops();
1051 }
1052
1053 #ifdef CONFIG_X86_32
1054
1055 static struct resource video_ram_resource = {
1056         .name   = "Video RAM area",
1057         .start  = 0xa0000,
1058         .end    = 0xbffff,
1059         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
1060 };
1061
1062 void __init i386_reserve_resources(void)
1063 {
1064         request_resource(&iomem_resource, &video_ram_resource);
1065         reserve_standard_io_resources();
1066 }
1067
1068 #endif /* CONFIG_X86_32 */