arch/x86/kernel/setup_32.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/crash_dump.h>
  46 #include <linux/dmi.h>
  47 #include <linux/pfn.h>
  48 #include <linux/pci.h>
  49 #include <linux/init_ohci1394_dma.h>
  50 #include <linux/kvm_para.h>
  51
  52 #include <video/edid.h>
  53
  54 #include <asm/mtrr.h>
  55 #include <asm/apic.h>
  56 #include <asm/e820.h>
  57 #include <asm/mpspec.h>
  58 #include <asm/mmzone.h>
  59 #include <asm/setup.h>
  60 #include <asm/arch_hooks.h>
  61 #include <asm/sections.h>
  62 #include <asm/io_apic.h>
  63 #include <asm/ist.h>
  64 #include <asm/io.h>
  65 #include <asm/vmi.h>
  66 #include <setup_arch.h>
  67 #include <asm/bios_ebda.h>
  68 #include <asm/cacheflush.h>
  69 #include <asm/processor.h>
  70
  71 /* This value is set up by the early boot code to point to the value
  72    immediately after the boot time page tables.  It contains a *physical*
  73    address, and must not be in the .bss segment! */
  74 unsigned long init_pg_tables_end __initdata = ~0UL;
  75
  76 /*
  77  * Machine setup..
  78  */
  79 static struct resource data_resource = {
  80         .name   = "Kernel data",
  81         .start  = 0,
  82         .end    = 0,
  83         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  84 };
  85
  86 static struct resource code_resource = {
  87         .name   = "Kernel code",
  88         .start  = 0,
  89         .end    = 0,
  90         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  91 };
  92
  93 static struct resource bss_resource = {
  94         .name   = "Kernel bss",
  95         .start  = 0,
  96         .end    = 0,
  97         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  98 };
  99
 100 static struct resource video_ram_resource = {
 101         .name   = "Video RAM area",
 102         .start  = 0xa0000,
 103         .end    = 0xbffff,
 104         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 105 };
 106
 107 static struct resource standard_io_resources[] = { {
 108         .name   = "dma1",
 109         .start  = 0x0000,
 110         .end    = 0x001f,
 111         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 112 }, {
 113         .name   = "pic1",
 114         .start  = 0x0020,
 115         .end    = 0x0021,
 116         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 117 }, {
 118         .name   = "timer0",
 119         .start  = 0x0040,
 120         .end    = 0x0043,
 121         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 122 }, {
 123         .name   = "timer1",
 124         .start  = 0x0050,
 125         .end    = 0x0053,
 126         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 127 }, {
 128         .name   = "keyboard",
 129         .start  = 0x0060,
 130         .end    = 0x0060,
 131         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 132 }, {
 133         .name   = "keyboard",
 134         .start  = 0x0064,
 135         .end    = 0x0064,
 136         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 137 }, {
 138         .name   = "dma page reg",
 139         .start  = 0x0080,
 140         .end    = 0x008f,
 141         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 142 }, {
 143         .name   = "pic2",
 144         .start  = 0x00a0,
 145         .end    = 0x00a1,
 146         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 147 }, {
 148         .name   = "dma2",
 149         .start  = 0x00c0,
 150         .end    = 0x00df,
 151         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 152 }, {
 153         .name   = "fpu",
 154         .start  = 0x00f0,
 155         .end    = 0x00ff,
 156         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 157 } };
 158
 159 /* cpu data as detected by the assembly code in head.S */
 160 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 161 /* common cpu data for all cpus */
 162 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 163 EXPORT_SYMBOL(boot_cpu_data);
 164
 165 unsigned int def_to_bigsmp;
 166
 167 #ifndef CONFIG_X86_PAE
 168 unsigned long mmu_cr4_features;
 169 #else
 170 unsigned long mmu_cr4_features = X86_CR4_PAE;
 171 #endif
 172
 173 /* for MCA, but anyone else can use it if they want */
 174 unsigned int machine_id;
 175 unsigned int machine_submodel_id;
 176 unsigned int BIOS_revision;
 177
 178 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 179 int bootloader_type;
 180
 181 /* user-defined highmem size */
 182 static unsigned int highmem_pages = -1;
 183
 184 /*
 185  * Setup options
 186  */
 187 struct screen_info screen_info;
 188 EXPORT_SYMBOL(screen_info);
 189 struct apm_info apm_info;
 190 EXPORT_SYMBOL(apm_info);
 191 struct edid_info edid_info;
 192 EXPORT_SYMBOL_GPL(edid_info);
 193 struct ist_info ist_info;
 194 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 195         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 196 EXPORT_SYMBOL(ist_info);
 197 #endif
 198
 199 extern void early_cpu_init(void);
 200 extern int root_mountflags;
 201
 202 unsigned long saved_video_mode;
 203
 204 #define RAMDISK_IMAGE_START_MASK        0x07FF
 205 #define RAMDISK_PROMPT_FLAG             0x8000
 206 #define RAMDISK_LOAD_FLAG               0x4000
 207
 208 static char __initdata command_line[COMMAND_LINE_SIZE];
 209
 210 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 211 struct boot_params __initdata boot_params;
 212 #else
 213 struct boot_params boot_params;
 214 #endif
 215
 216 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 217 struct edd edd;
 218 #ifdef CONFIG_EDD_MODULE
 219 EXPORT_SYMBOL(edd);
 220 #endif
 221 /**
 222  * copy_edd() - Copy the BIOS EDD information
 223  *              from boot_params into a safe place.
 224  *
 225  */
 226 static inline void copy_edd(void)
 227 {
 228      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 229             sizeof(edd.mbr_signature));
 230      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 231      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 232      edd.edd_info_nr = boot_params.eddbuf_entries;
 233 }
 234 #else
 235 static inline void copy_edd(void)
 236 {
 237 }
 238 #endif
 239
 240 int __initdata user_defined_memmap;
 241
 242 /*
 243  * "mem=nopentium" disables the 4MB page tables.
 244  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
 245  * to <mem>, overriding the bios size.
 246  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
 247  * <start> to <start>+<mem>, overriding the bios size.
 248  *
 249  * HPA tells me bootloaders need to parse mem=, so no new
 250  * option should be mem=  [also see Documentation/i386/boot.txt]
 251  */
 252 static int __init parse_mem(char *arg)
 253 {
 254         if (!arg)
 255                 return -EINVAL;
 256
 257         if (strcmp(arg, "nopentium") == 0) {
 258                 setup_clear_cpu_cap(X86_FEATURE_PSE);
 259         } else {
 260                 /* If the user specifies memory size, we
 261                  * limit the BIOS-provided memory map to
 262                  * that size. exactmap can be used to specify
 263                  * the exact map. mem=number can be used to
 264                  * trim the existing memory map.
 265                  */
 266                 unsigned long long mem_size;
 267
 268                 mem_size = memparse(arg, &arg);
 269                 limit_regions(mem_size);
 270                 user_defined_memmap = 1;
 271         }
 272         return 0;
 273 }
 274 early_param("mem", parse_mem);
 275
 276 #ifdef CONFIG_PROC_VMCORE
 277 /* elfcorehdr= specifies the location of elf core header
 278  * stored by the crashed kernel.
 279  */
 280 static int __init parse_elfcorehdr(char *arg)
 281 {
 282         if (!arg)
 283                 return -EINVAL;
 284
 285         elfcorehdr_addr = memparse(arg, &arg);
 286         return 0;
 287 }
 288 early_param("elfcorehdr", parse_elfcorehdr);
 289 #endif /* CONFIG_PROC_VMCORE */
 290
 291 /*
 292  * highmem=size forces highmem to be exactly 'size' bytes.
 293  * This works even on boxes that have no highmem otherwise.
 294  * This also works to reduce highmem size on bigger boxes.
 295  */
 296 static int __init parse_highmem(char *arg)
 297 {
 298         if (!arg)
 299                 return -EINVAL;
 300
 301         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 302         return 0;
 303 }
 304 early_param("highmem", parse_highmem);
 305
 306 /*
 307  * vmalloc=size forces the vmalloc area to be exactly 'size'
 308  * bytes. This can be used to increase (or decrease) the
 309  * vmalloc area - the default is 128m.
 310  */
 311 static int __init parse_vmalloc(char *arg)
 312 {
 313         if (!arg)
 314                 return -EINVAL;
 315
 316         __VMALLOC_RESERVE = memparse(arg, &arg);
 317         return 0;
 318 }
 319 early_param("vmalloc", parse_vmalloc);
 320
 321 /*
 322  * reservetop=size reserves a hole at the top of the kernel address space which
 323  * a hypervisor can load into later.  Needed for dynamically loaded hypervisors,
 324  * so relocating the fixmap can be done before paging initialization.
 325  */
 326 static int __init parse_reservetop(char *arg)
 327 {
 328         unsigned long address;
 329
 330         if (!arg)
 331                 return -EINVAL;
 332
 333         address = memparse(arg, &arg);
 334         reserve_top_address(address);
 335         return 0;
 336 }
 337 early_param("reservetop", parse_reservetop);
 338
 339 /*
 340  * Determine low and high memory ranges:
 341  */
 342 unsigned long __init find_max_low_pfn(void)
 343 {
 344         unsigned long max_low_pfn;
 345
 346         max_low_pfn = max_pfn;
 347         if (max_low_pfn > MAXMEM_PFN) {
 348                 if (highmem_pages == -1)
 349                         highmem_pages = max_pfn - MAXMEM_PFN;
 350                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 351                         max_pfn = MAXMEM_PFN + highmem_pages;
 352                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 353                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
 354                         highmem_pages = 0;
 355                 }
 356                 max_low_pfn = MAXMEM_PFN;
 357 #ifndef CONFIG_HIGHMEM
 358                 /* Maximum memory usable is what is directly addressable */
 359                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 360                                         MAXMEM>>20);
 361                 if (max_pfn > MAX_NONPAE_PFN)
 362                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 363                 else
 364                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 365                 max_pfn = MAXMEM_PFN;
 366 #else /* !CONFIG_HIGHMEM */
 367 #ifndef CONFIG_HIGHMEM64G
 368                 if (max_pfn > MAX_NONPAE_PFN) {
 369                         max_pfn = MAX_NONPAE_PFN;
 370                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
 371                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 372                 }
 373 #endif /* !CONFIG_HIGHMEM64G */
 374 #endif /* !CONFIG_HIGHMEM */
 375         } else {
 376                 if (highmem_pages == -1)
 377                         highmem_pages = 0;
 378 #ifdef CONFIG_HIGHMEM
 379                 if (highmem_pages >= max_pfn) {
 380                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 381                         highmem_pages = 0;
 382                 }
 383                 if (highmem_pages) {
 384                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
 385                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
 386                                 highmem_pages = 0;
 387                         }
 388                         max_low_pfn -= highmem_pages;
 389                 }
 390 #else
 391                 if (highmem_pages)
 392                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 393 #endif
 394         }
 395         return max_low_pfn;
 396 }
 397
 398 #define BIOS_LOWMEM_KILOBYTES 0x413
 399
 400 /*
 401  * The BIOS places the EBDA/XBDA at the top of conventional
 402  * memory, and usually decreases the reported amount of
 403  * conventional memory (int 0x12) too. This also contains a
 404  * workaround for Dell systems that neglect to reserve EBDA.
 405  * The same workaround also avoids a problem with the AMD768MPX
 406  * chipset: reserve a page before VGA to prevent PCI prefetch
 407  * into it (errata #56). Usually the page is reserved anyways,
 408  * unless you have no PS/2 mouse plugged in.
 409  */
 410 static void __init reserve_ebda_region(void)
 411 {
 412         unsigned int lowmem, ebda_addr;
 413
 414         /* To determine the position of the EBDA and the */
 415         /* end of conventional memory, we need to look at */
 416         /* the BIOS data area. In a paravirtual environment */
 417         /* that area is absent. We'll just have to assume */
 418         /* that the paravirt case can handle memory setup */
 419         /* correctly, without our help. */
 420         if (paravirt_enabled())
 421                 return;
 422
 423         /* end of low (conventional) memory */
 424         lowmem = *(unsigned short *)__va(BIOS_LOWMEM_KILOBYTES);
 425         lowmem <<= 10;
 426
 427         /* start of EBDA area */
 428         ebda_addr = get_bios_ebda();
 429
 430         /* Fixup: bios puts an EBDA in the top 64K segment */
 431         /* of conventional memory, but does not adjust lowmem. */
 432         if ((lowmem - ebda_addr) <= 0x10000)
 433                 lowmem = ebda_addr;
 434
 435         /* Fixup: bios does not report an EBDA at all. */
 436         /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */
 437         if ((ebda_addr == 0) && (lowmem >= 0x9f000))
 438                 lowmem = 0x9f000;
 439
 440         /* Paranoia: should never happen, but... */
 441         if ((lowmem == 0) || (lowmem >= 0x100000))
 442                 lowmem = 0x9f000;
 443
 444         /* reserve all memory between lowmem and the 1MB mark */
 445         reserve_bootmem(lowmem, 0x100000 - lowmem, BOOTMEM_DEFAULT);
 446 }
 447
 448 #ifndef CONFIG_NEED_MULTIPLE_NODES
 449 static void __init setup_bootmem_allocator(void);
 450 static unsigned long __init setup_memory(void)
 451 {
 452         /*
 453          * partially used pages are not usable - thus
 454          * we are rounding upwards:
 455          */
 456         min_low_pfn = PFN_UP(init_pg_tables_end);
 457
 458         max_low_pfn = find_max_low_pfn();
 459
 460 #ifdef CONFIG_HIGHMEM
 461         highstart_pfn = highend_pfn = max_pfn;
 462         if (max_pfn > max_low_pfn) {
 463                 highstart_pfn = max_low_pfn;
 464         }
 465         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 466                 pages_to_mb(highend_pfn - highstart_pfn));
 467         num_physpages = highend_pfn;
 468         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 469 #else
 470         num_physpages = max_low_pfn;
 471         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 472 #endif
 473 #ifdef CONFIG_FLATMEM
 474         max_mapnr = num_physpages;
 475 #endif
 476         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 477                         pages_to_mb(max_low_pfn));
 478
 479         setup_bootmem_allocator();
 480
 481         return max_low_pfn;
 482 }
 483
 484 static void __init zone_sizes_init(void)
 485 {
 486         unsigned long max_zone_pfns[MAX_NR_ZONES];
 487         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 488         max_zone_pfns[ZONE_DMA] =
 489                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 490         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 491 #ifdef CONFIG_HIGHMEM
 492         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 493         add_active_range(0, 0, highend_pfn);
 494 #else
 495         add_active_range(0, 0, max_low_pfn);
 496 #endif
 497
 498         free_area_init_nodes(max_zone_pfns);
 499 }
 500 #else
 501 extern unsigned long __init setup_memory(void);
 502 extern void zone_sizes_init(void);
 503 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 504
 505 static inline unsigned long long get_total_mem(void)
 506 {
 507         unsigned long long total;
 508
 509         total = max_low_pfn - min_low_pfn;
 510 #ifdef CONFIG_HIGHMEM
 511         total += highend_pfn - highstart_pfn;
 512 #endif
 513
 514         return total << PAGE_SHIFT;
 515 }
 516
 517 #ifdef CONFIG_KEXEC
 518 static void __init reserve_crashkernel(void)
 519 {
 520         unsigned long long total_mem;
 521         unsigned long long crash_size, crash_base;
 522         int ret;
 523
 524         total_mem = get_total_mem();
 525
 526         ret = parse_crashkernel(boot_command_line, total_mem,
 527                         &crash_size, &crash_base);
 528         if (ret == 0 && crash_size > 0) {
 529                 if (crash_base > 0) {
 530                         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 531                                         "for crashkernel (System RAM: %ldMB)\n",
 532                                         (unsigned long)(crash_size >> 20),
 533                                         (unsigned long)(crash_base >> 20),
 534                                         (unsigned long)(total_mem >> 20));
 535                         crashk_res.start = crash_base;
 536                         crashk_res.end   = crash_base + crash_size - 1;
 537                         reserve_bootmem(crash_base, crash_size,
 538                                         BOOTMEM_DEFAULT);
 539                 } else
 540                         printk(KERN_INFO "crashkernel reservation failed - "
 541                                         "you have to specify a base address\n");
 542         }
 543 }
 544 #else
 545 static inline void __init reserve_crashkernel(void)
 546 {}
 547 #endif
 548
 549 #ifdef CONFIG_BLK_DEV_INITRD
 550
 551 static bool do_relocate_initrd = false;
 552
 553 static void __init reserve_initrd(void)
 554 {
 555         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 556         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 557         unsigned long ramdisk_end   = ramdisk_image + ramdisk_size;
 558         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 559         unsigned long ramdisk_here;
 560
 561         initrd_start = 0;
 562
 563         if (!boot_params.hdr.type_of_loader ||
 564             !ramdisk_image || !ramdisk_size)
 565                 return;         /* No initrd provided by bootloader */
 566
 567         if (ramdisk_end < ramdisk_image) {
 568                 printk(KERN_ERR "initrd wraps around end of memory, "
 569                        "disabling initrd\n");
 570                 return;
 571         }
 572         if (ramdisk_size >= end_of_lowmem/2) {
 573                 printk(KERN_ERR "initrd too large to handle, "
 574                        "disabling initrd\n");
 575                 return;
 576         }
 577         if (ramdisk_end <= end_of_lowmem) {
 578                 /* All in lowmem, easy case */
 579                 reserve_bootmem(ramdisk_image, ramdisk_size, BOOTMEM_DEFAULT);
 580                 initrd_start = ramdisk_image + PAGE_OFFSET;
 581                 initrd_end = initrd_start+ramdisk_size;
 582                 return;
 583         }
 584
 585         /* We need to move the initrd down into lowmem */
 586         ramdisk_here = (end_of_lowmem - ramdisk_size) & PAGE_MASK;
 587
 588         /* Note: this includes all the lowmem currently occupied by
 589            the initrd, we rely on that fact to keep the data intact. */
 590         reserve_bootmem(ramdisk_here, ramdisk_size, BOOTMEM_DEFAULT);
 591         initrd_start = ramdisk_here + PAGE_OFFSET;
 592         initrd_end   = initrd_start + ramdisk_size;
 593
 594         do_relocate_initrd = true;
 595 }
 596
 597 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 598
 599 static void __init relocate_initrd(void)
 600 {
 601         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 602         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 603         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 604         unsigned long ramdisk_here;
 605         unsigned long slop, clen, mapaddr;
 606         char *p, *q;
 607
 608         if (!do_relocate_initrd)
 609                 return;
 610
 611         ramdisk_here = initrd_start - PAGE_OFFSET;
 612
 613         q = (char *)initrd_start;
 614
 615         /* Copy any lowmem portion of the initrd */
 616         if (ramdisk_image < end_of_lowmem) {
 617                 clen = end_of_lowmem - ramdisk_image;
 618                 p = (char *)__va(ramdisk_image);
 619                 memcpy(q, p, clen);
 620                 q += clen;
 621                 ramdisk_image += clen;
 622                 ramdisk_size  -= clen;
 623         }
 624
 625         /* Copy the highmem portion of the initrd */
 626         while (ramdisk_size) {
 627                 slop = ramdisk_image & ~PAGE_MASK;
 628                 clen = ramdisk_size;
 629                 if (clen > MAX_MAP_CHUNK-slop)
 630                         clen = MAX_MAP_CHUNK-slop;
 631                 mapaddr = ramdisk_image & PAGE_MASK;
 632                 p = early_ioremap(mapaddr, clen+slop);
 633                 memcpy(q, p+slop, clen);
 634                 early_iounmap(p, clen+slop);
 635                 q += clen;
 636                 ramdisk_image += clen;
 637                 ramdisk_size  -= clen;
 638         }
 639 }
 640
 641 #endif /* CONFIG_BLK_DEV_INITRD */
 642
 643 void __init setup_bootmem_allocator(void)
 644 {
 645         unsigned long bootmap_size;
 646         /*
 647          * Initialize the boot-time allocator (with low memory only):
 648          */
 649         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
 650
 651         register_bootmem_low_pages(max_low_pfn);
 652
 653         /*
 654          * Reserve the bootmem bitmap itself as well. We do this in two
 655          * steps (first step was init_bootmem()) because this catches
 656          * the (very unlikely) case of us accidentally initializing the
 657          * bootmem allocator with an invalid RAM area.
 658          */
 659         reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
 660                          bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text),
 661                          BOOTMEM_DEFAULT);
 662
 663         /*
 664          * reserve physical page 0 - it's a special BIOS page on many boxes,
 665          * enabling clean reboots, SMP operation, laptop functions.
 666          */
 667         reserve_bootmem(0, PAGE_SIZE, BOOTMEM_DEFAULT);
 668
 669         /* reserve EBDA region */
 670         reserve_ebda_region();
 671
 672 #ifdef CONFIG_SMP
 673         /*
 674          * But first pinch a few for the stack/trampoline stuff
 675          * FIXME: Don't need the extra page at 4K, but need to fix
 676          * trampoline before removing it. (see the GDT stuff)
 677          */
 678         reserve_bootmem(PAGE_SIZE, PAGE_SIZE, BOOTMEM_DEFAULT);
 679 #endif
 680 #ifdef CONFIG_ACPI_SLEEP
 681         /*
 682          * Reserve low memory region for sleep support.
 683          */
 684         acpi_reserve_bootmem();
 685 #endif
 686 #ifdef CONFIG_X86_FIND_SMP_CONFIG
 687         /*
 688          * Find and reserve possible boot-time SMP configuration:
 689          */
 690         find_smp_config();
 691 #endif
 692 #ifdef CONFIG_BLK_DEV_INITRD
 693         reserve_initrd();
 694 #endif
 695         numa_kva_reserve();
 696         reserve_crashkernel();
 697
 698         reserve_ibft_region();
 699 }
 700
 701 /*
 702  * The node 0 pgdat is initialized before all of these because
 703  * it's needed for bootmem.  node>0 pgdats have their virtual
 704  * space allocated before the pagetables are in place to access
 705  * them, so they can't be cleared then.
 706  *
 707  * This should all compile down to nothing when NUMA is off.
 708  */
 709 static void __init remapped_pgdat_init(void)
 710 {
 711         int nid;
 712
 713         for_each_online_node(nid) {
 714                 if (nid != 0)
 715                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
 716         }
 717 }
 718
 719 #ifdef CONFIG_MCA
 720 static void set_mca_bus(int x)
 721 {
 722         MCA_bus = x;
 723 }
 724 #else
 725 static void set_mca_bus(int x) { }
 726 #endif
 727
 728 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
 729 char * __init __attribute__((weak)) memory_setup(void)
 730 {
 731         return machine_specific_memory_setup();
 732 }
 733
 734 #ifdef CONFIG_NUMA
 735 /*
 736  * In the golden day, when everything among i386 and x86_64 will be
 737  * integrated, this will not live here
 738  */
 739 void *x86_cpu_to_node_map_early_ptr;
 740 int x86_cpu_to_node_map_init[NR_CPUS] = {
 741         [0 ... NR_CPUS-1] = NUMA_NO_NODE
 742 };
 743 DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE;
 744 #endif
 745
 746 /*
 747  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 748  * passed the efi memmap, systab, etc., so we should use these data structures
 749  * for initialization.  Note, the efi init code path is determined by the
 750  * global efi_enabled. This allows the same kernel image to be used on existing
 751  * systems (with a traditional BIOS) as well as on EFI systems.
 752  */
 753 void __init setup_arch(char **cmdline_p)
 754 {
 755         unsigned long max_low_pfn;
 756
 757         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 758         pre_setup_arch_hook();
 759         early_cpu_init();
 760         early_ioremap_init();
 761
 762 #ifdef CONFIG_EFI
 763         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 764                      "EL32", 4))
 765                 efi_enabled = 1;
 766 #endif
 767
 768         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 769         screen_info = boot_params.screen_info;
 770         edid_info = boot_params.edid_info;
 771         apm_info.bios = boot_params.apm_bios_info;
 772         ist_info = boot_params.ist_info;
 773         saved_video_mode = boot_params.hdr.vid_mode;
 774         if( boot_params.sys_desc_table.length != 0 ) {
 775                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 776                 machine_id = boot_params.sys_desc_table.table[0];
 777                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 778                 BIOS_revision = boot_params.sys_desc_table.table[2];
 779         }
 780         bootloader_type = boot_params.hdr.type_of_loader;
 781
 782 #ifdef CONFIG_BLK_DEV_RAM
 783         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 784         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 785         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 786 #endif
 787         ARCH_SETUP
 788
 789         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 790         print_memory_map(memory_setup());
 791
 792         copy_edd();
 793
 794         if (!boot_params.hdr.root_flags)
 795                 root_mountflags &= ~MS_RDONLY;
 796         init_mm.start_code = (unsigned long) _text;
 797         init_mm.end_code = (unsigned long) _etext;
 798         init_mm.end_data = (unsigned long) _edata;
 799         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 800
 801         code_resource.start = virt_to_phys(_text);
 802         code_resource.end = virt_to_phys(_etext)-1;
 803         data_resource.start = virt_to_phys(_etext);
 804         data_resource.end = virt_to_phys(_edata)-1;
 805         bss_resource.start = virt_to_phys(&__bss_start);
 806         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 807
 808         parse_early_param();
 809
 810         if (user_defined_memmap) {
 811                 printk(KERN_INFO "user-defined physical RAM map:\n");
 812                 print_memory_map("user");
 813         }
 814
 815         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 816         *cmdline_p = command_line;
 817
 818         if (efi_enabled)
 819                 efi_init();
 820
 821         /* update e820 for memory not covered by WB MTRRs */
 822         propagate_e820_map();
 823         mtrr_bp_init();
 824         if (mtrr_trim_uncached_memory(max_pfn))
 825                 propagate_e820_map();
 826
 827         max_low_pfn = setup_memory();
 828
 829 #ifdef CONFIG_KVM_CLOCK
 830         kvmclock_init();
 831 #endif
 832
 833 #ifdef CONFIG_VMI
 834         /*
 835          * Must be after max_low_pfn is determined, and before kernel
 836          * pagetables are setup.
 837          */
 838         vmi_init();
 839 #endif
 840         kvm_guest_init();
 841
 842         /*
 843          * NOTE: before this point _nobody_ is allowed to allocate
 844          * any memory using the bootmem allocator.  Although the
 845          * allocator is now initialised only the first 8Mb of the kernel
 846          * virtual address space has been mapped.  All allocations before
 847          * paging_init() has completed must use the alloc_bootmem_low_pages()
 848          * variant (which allocates DMA'able memory) and care must be taken
 849          * not to exceed the 8Mb limit.
 850          */
 851
 852 #ifdef CONFIG_SMP
 853         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
 854 #endif
 855         paging_init();
 856
 857         /*
 858          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 859          */
 860
 861 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 862         if (init_ohci1394_dma_early)
 863                 init_ohci1394_dma_on_all_controllers();
 864 #endif
 865
 866         remapped_pgdat_init();
 867         sparse_init();
 868         zone_sizes_init();
 869
 870         /*
 871          * NOTE: at this point the bootmem allocator is fully available.
 872          */
 873
 874 #ifdef CONFIG_BLK_DEV_INITRD
 875         relocate_initrd();
 876 #endif
 877
 878         paravirt_post_allocator_init();
 879
 880         dmi_scan_machine();
 881
 882         io_delay_init();
 883
 884 #ifdef CONFIG_X86_SMP
 885         /*
 886          * setup to use the early static init tables during kernel startup
 887          * X86_SMP will exclude sub-arches that don't deal well with it.
 888          */
 889         x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
 890         x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
 891 #ifdef CONFIG_NUMA
 892         x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
 893 #endif
 894 #endif
 895
 896 #ifdef CONFIG_X86_GENERICARCH
 897         generic_apic_probe();
 898 #endif
 899
 900 #ifdef CONFIG_ACPI
 901         /*
 902          * Parse the ACPI tables for possible boot-time SMP configuration.
 903          */
 904         acpi_boot_table_init();
 905 #endif
 906
 907         early_quirks();
 908
 909 #ifdef CONFIG_ACPI
 910         acpi_boot_init();
 911
 912 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
 913         if (def_to_bigsmp)
 914                 printk(KERN_WARNING "More than 8 CPUs detected and "
 915                         "CONFIG_X86_PC cannot handle it.\nUse "
 916                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
 917 #endif
 918 #endif
 919 #ifdef CONFIG_X86_LOCAL_APIC
 920         if (smp_found_config)
 921                 get_smp_config();
 922 #endif
 923
 924         e820_register_memory();
 925         e820_mark_nosave_regions();
 926
 927 #ifdef CONFIG_VT
 928 #if defined(CONFIG_VGA_CONSOLE)
 929         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 930                 conswitchp = &vga_con;
 931 #elif defined(CONFIG_DUMMY_CONSOLE)
 932         conswitchp = &dummy_con;
 933 #endif
 934 #endif
 935 }
 936
 937 /*
 938  * Request address space for all standard resources
 939  *
 940  * This is called just before pcibios_init(), which is also a
 941  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
 942  */
 943 static int __init request_standard_resources(void)
 944 {
 945         int i;
 946
 947         printk(KERN_INFO "Setting up standard PCI resources\n");
 948         init_iomem_resources(&code_resource, &data_resource, &bss_resource);
 949
 950         request_resource(&iomem_resource, &video_ram_resource);
 951
 952         /* request I/O space for devices used on all i[345]86 PCs */
 953         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 954                 request_resource(&ioport_resource, &standard_io_resources[i]);
 955         return 0;
 956 }
 957
 958 subsys_initcall(request_standard_resources);