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/nodemask.h>
  43 #include <linux/kexec.h>
  44 #include <linux/crash_dump.h>
  45 #include <linux/dmi.h>
  46 #include <linux/pfn.h>
  47 #include <linux/pci.h>
  48 #include <linux/init_ohci1394_dma.h>
  49
  50 #include <video/edid.h>
  51
  52 #include <asm/mtrr.h>
  53 #include <asm/apic.h>
  54 #include <asm/e820.h>
  55 #include <asm/mpspec.h>
  56 #include <asm/mmzone.h>
  57 #include <asm/setup.h>
  58 #include <asm/arch_hooks.h>
  59 #include <asm/sections.h>
  60 #include <asm/io_apic.h>
  61 #include <asm/ist.h>
  62 #include <asm/io.h>
  63 #include <asm/vmi.h>
  64 #include <setup_arch.h>
  65 #include <asm/bios_ebda.h>
  66 #include <asm/cacheflush.h>
  67 #include <asm/processor.h>
  68
  69 /* This value is set up by the early boot code to point to the value
  70    immediately after the boot time page tables.  It contains a *physical*
  71    address, and must not be in the .bss segment! */
  72 unsigned long init_pg_tables_end __initdata = ~0UL;
  73
  74 /*
  75  * Machine setup..
  76  */
  77 static struct resource data_resource = {
  78         .name   = "Kernel data",
  79         .start  = 0,
  80         .end    = 0,
  81         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  82 };
  83
  84 static struct resource code_resource = {
  85         .name   = "Kernel code",
  86         .start  = 0,
  87         .end    = 0,
  88         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  89 };
  90
  91 static struct resource bss_resource = {
  92         .name   = "Kernel bss",
  93         .start  = 0,
  94         .end    = 0,
  95         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  96 };
  97
  98 static struct resource video_ram_resource = {
  99         .name   = "Video RAM area",
 100         .start  = 0xa0000,
 101         .end    = 0xbffff,
 102         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 103 };
 104
 105 static struct resource standard_io_resources[] = { {
 106         .name   = "dma1",
 107         .start  = 0x0000,
 108         .end    = 0x001f,
 109         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 110 }, {
 111         .name   = "pic1",
 112         .start  = 0x0020,
 113         .end    = 0x0021,
 114         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 115 }, {
 116         .name   = "timer0",
 117         .start  = 0x0040,
 118         .end    = 0x0043,
 119         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 120 }, {
 121         .name   = "timer1",
 122         .start  = 0x0050,
 123         .end    = 0x0053,
 124         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 125 }, {
 126         .name   = "keyboard",
 127         .start  = 0x0060,
 128         .end    = 0x006f,
 129         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 130 }, {
 131         .name   = "dma page reg",
 132         .start  = 0x0080,
 133         .end    = 0x008f,
 134         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 135 }, {
 136         .name   = "pic2",
 137         .start  = 0x00a0,
 138         .end    = 0x00a1,
 139         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 140 }, {
 141         .name   = "dma2",
 142         .start  = 0x00c0,
 143         .end    = 0x00df,
 144         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 145 }, {
 146         .name   = "fpu",
 147         .start  = 0x00f0,
 148         .end    = 0x00ff,
 149         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 150 } };
 151
 152 /* cpu data as detected by the assembly code in head.S */
 153 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 154 /* common cpu data for all cpus */
 155 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 156 EXPORT_SYMBOL(boot_cpu_data);
 157
 158 #ifndef CONFIG_X86_PAE
 159 unsigned long mmu_cr4_features;
 160 #else
 161 unsigned long mmu_cr4_features = X86_CR4_PAE;
 162 #endif
 163
 164 /* for MCA, but anyone else can use it if they want */
 165 unsigned int machine_id;
 166 unsigned int machine_submodel_id;
 167 unsigned int BIOS_revision;
 168
 169 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 170 int bootloader_type;
 171
 172 /* user-defined highmem size */
 173 static unsigned int highmem_pages = -1;
 174
 175 /*
 176  * Setup options
 177  */
 178 struct screen_info screen_info;
 179 EXPORT_SYMBOL(screen_info);
 180 struct apm_info apm_info;
 181 EXPORT_SYMBOL(apm_info);
 182 struct edid_info edid_info;
 183 EXPORT_SYMBOL_GPL(edid_info);
 184 struct ist_info ist_info;
 185 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 186         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 187 EXPORT_SYMBOL(ist_info);
 188 #endif
 189
 190 extern void early_cpu_init(void);
 191 extern int root_mountflags;
 192
 193 unsigned long saved_videomode;
 194
 195 #define RAMDISK_IMAGE_START_MASK        0x07FF
 196 #define RAMDISK_PROMPT_FLAG             0x8000
 197 #define RAMDISK_LOAD_FLAG               0x4000
 198
 199 static char __initdata command_line[COMMAND_LINE_SIZE];
 200
 201 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 202 struct boot_params __initdata boot_params;
 203 #else
 204 struct boot_params boot_params;
 205 #endif
 206
 207 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 208 struct edd edd;
 209 #ifdef CONFIG_EDD_MODULE
 210 EXPORT_SYMBOL(edd);
 211 #endif
 212 /**
 213  * copy_edd() - Copy the BIOS EDD information
 214  *              from boot_params into a safe place.
 215  *
 216  */
 217 static inline void copy_edd(void)
 218 {
 219      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 220             sizeof(edd.mbr_signature));
 221      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 222      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 223      edd.edd_info_nr = boot_params.eddbuf_entries;
 224 }
 225 #else
 226 static inline void copy_edd(void)
 227 {
 228 }
 229 #endif
 230
 231 int __initdata user_defined_memmap;
 232
 233 /*
 234  * "mem=nopentium" disables the 4MB page tables.
 235  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
 236  * to <mem>, overriding the bios size.
 237  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
 238  * <start> to <start>+<mem>, overriding the bios size.
 239  *
 240  * HPA tells me bootloaders need to parse mem=, so no new
 241  * option should be mem=  [also see Documentation/i386/boot.txt]
 242  */
 243 static int __init parse_mem(char *arg)
 244 {
 245         if (!arg)
 246                 return -EINVAL;
 247
 248         if (strcmp(arg, "nopentium") == 0) {
 249                 setup_clear_cpu_cap(X86_FEATURE_PSE);
 250         } else {
 251                 /* If the user specifies memory size, we
 252                  * limit the BIOS-provided memory map to
 253                  * that size. exactmap can be used to specify
 254                  * the exact map. mem=number can be used to
 255                  * trim the existing memory map.
 256                  */
 257                 unsigned long long mem_size;
 258
 259                 mem_size = memparse(arg, &arg);
 260                 limit_regions(mem_size);
 261                 user_defined_memmap = 1;
 262         }
 263         return 0;
 264 }
 265 early_param("mem", parse_mem);
 266
 267 #ifdef CONFIG_PROC_VMCORE
 268 /* elfcorehdr= specifies the location of elf core header
 269  * stored by the crashed kernel.
 270  */
 271 static int __init parse_elfcorehdr(char *arg)
 272 {
 273         if (!arg)
 274                 return -EINVAL;
 275
 276         elfcorehdr_addr = memparse(arg, &arg);
 277         return 0;
 278 }
 279 early_param("elfcorehdr", parse_elfcorehdr);
 280 #endif /* CONFIG_PROC_VMCORE */
 281
 282 /*
 283  * highmem=size forces highmem to be exactly 'size' bytes.
 284  * This works even on boxes that have no highmem otherwise.
 285  * This also works to reduce highmem size on bigger boxes.
 286  */
 287 static int __init parse_highmem(char *arg)
 288 {
 289         if (!arg)
 290                 return -EINVAL;
 291
 292         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 293         return 0;
 294 }
 295 early_param("highmem", parse_highmem);
 296
 297 /*
 298  * vmalloc=size forces the vmalloc area to be exactly 'size'
 299  * bytes. This can be used to increase (or decrease) the
 300  * vmalloc area - the default is 128m.
 301  */
 302 static int __init parse_vmalloc(char *arg)
 303 {
 304         if (!arg)
 305                 return -EINVAL;
 306
 307         __VMALLOC_RESERVE = memparse(arg, &arg);
 308         return 0;
 309 }
 310 early_param("vmalloc", parse_vmalloc);
 311
 312 /*
 313  * reservetop=size reserves a hole at the top of the kernel address space which
 314  * a hypervisor can load into later.  Needed for dynamically loaded hypervisors,
 315  * so relocating the fixmap can be done before paging initialization.
 316  */
 317 static int __init parse_reservetop(char *arg)
 318 {
 319         unsigned long address;
 320
 321         if (!arg)
 322                 return -EINVAL;
 323
 324         address = memparse(arg, &arg);
 325         reserve_top_address(address);
 326         return 0;
 327 }
 328 early_param("reservetop", parse_reservetop);
 329
 330 /*
 331  * Determine low and high memory ranges:
 332  */
 333 unsigned long __init find_max_low_pfn(void)
 334 {
 335         unsigned long max_low_pfn;
 336
 337         max_low_pfn = max_pfn;
 338         if (max_low_pfn > MAXMEM_PFN) {
 339                 if (highmem_pages == -1)
 340                         highmem_pages = max_pfn - MAXMEM_PFN;
 341                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 342                         max_pfn = MAXMEM_PFN + highmem_pages;
 343                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 344                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
 345                         highmem_pages = 0;
 346                 }
 347                 max_low_pfn = MAXMEM_PFN;
 348 #ifndef CONFIG_HIGHMEM
 349                 /* Maximum memory usable is what is directly addressable */
 350                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 351                                         MAXMEM>>20);
 352                 if (max_pfn > MAX_NONPAE_PFN)
 353                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 354                 else
 355                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 356                 max_pfn = MAXMEM_PFN;
 357 #else /* !CONFIG_HIGHMEM */
 358 #ifndef CONFIG_HIGHMEM64G
 359                 if (max_pfn > MAX_NONPAE_PFN) {
 360                         max_pfn = MAX_NONPAE_PFN;
 361                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
 362                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 363                 }
 364 #endif /* !CONFIG_HIGHMEM64G */
 365 #endif /* !CONFIG_HIGHMEM */
 366         } else {
 367                 if (highmem_pages == -1)
 368                         highmem_pages = 0;
 369 #ifdef CONFIG_HIGHMEM
 370                 if (highmem_pages >= max_pfn) {
 371                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 372                         highmem_pages = 0;
 373                 }
 374                 if (highmem_pages) {
 375                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
 376                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
 377                                 highmem_pages = 0;
 378                         }
 379                         max_low_pfn -= highmem_pages;
 380                 }
 381 #else
 382                 if (highmem_pages)
 383                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 384 #endif
 385         }
 386         return max_low_pfn;
 387 }
 388
 389 #define BIOS_EBDA_SEGMENT 0x40E
 390 #define BIOS_LOWMEM_KILOBYTES 0x413
 391
 392 /*
 393  * The BIOS places the EBDA/XBDA at the top of conventional
 394  * memory, and usually decreases the reported amount of
 395  * conventional memory (int 0x12) too. This also contains a
 396  * workaround for Dell systems that neglect to reserve EBDA.
 397  * The same workaround also avoids a problem with the AMD768MPX
 398  * chipset: reserve a page before VGA to prevent PCI prefetch
 399  * into it (errata #56). Usually the page is reserved anyways,
 400  * unless you have no PS/2 mouse plugged in.
 401  */
 402 static void __init reserve_ebda_region(void)
 403 {
 404         unsigned int lowmem, ebda_addr;
 405
 406         /* To determine the position of the EBDA and the */
 407         /* end of conventional memory, we need to look at */
 408         /* the BIOS data area. In a paravirtual environment */
 409         /* that area is absent. We'll just have to assume */
 410         /* that the paravirt case can handle memory setup */
 411         /* correctly, without our help. */
 412         if (paravirt_enabled())
 413                 return;
 414
 415         /* end of low (conventional) memory */
 416         lowmem = *(unsigned short *)__va(BIOS_LOWMEM_KILOBYTES);
 417         lowmem <<= 10;
 418
 419         /* start of EBDA area */
 420         ebda_addr = *(unsigned short *)__va(BIOS_EBDA_SEGMENT);
 421         ebda_addr <<= 4;
 422
 423         /* Fixup: bios puts an EBDA in the top 64K segment */
 424         /* of conventional memory, but does not adjust lowmem. */
 425         if ((lowmem - ebda_addr) <= 0x10000)
 426                 lowmem = ebda_addr;
 427
 428         /* Fixup: bios does not report an EBDA at all. */
 429         /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */
 430         if ((ebda_addr == 0) && (lowmem >= 0x9f000))
 431                 lowmem = 0x9f000;
 432
 433         /* Paranoia: should never happen, but... */
 434         if ((lowmem == 0) || (lowmem >= 0x100000))
 435                 lowmem = 0x9f000;
 436
 437         /* reserve all memory between lowmem and the 1MB mark */
 438         reserve_bootmem(lowmem, 0x100000 - lowmem, BOOTMEM_DEFAULT);
 439 }
 440
 441 #ifndef CONFIG_NEED_MULTIPLE_NODES
 442 void __init setup_bootmem_allocator(void);
 443 static unsigned long __init setup_memory(void)
 444 {
 445         /*
 446          * partially used pages are not usable - thus
 447          * we are rounding upwards:
 448          */
 449         min_low_pfn = PFN_UP(init_pg_tables_end);
 450
 451         max_low_pfn = find_max_low_pfn();
 452
 453 #ifdef CONFIG_HIGHMEM
 454         highstart_pfn = highend_pfn = max_pfn;
 455         if (max_pfn > max_low_pfn) {
 456                 highstart_pfn = max_low_pfn;
 457         }
 458         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 459                 pages_to_mb(highend_pfn - highstart_pfn));
 460         num_physpages = highend_pfn;
 461         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 462 #else
 463         num_physpages = max_low_pfn;
 464         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 465 #endif
 466 #ifdef CONFIG_FLATMEM
 467         max_mapnr = num_physpages;
 468 #endif
 469         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 470                         pages_to_mb(max_low_pfn));
 471
 472         setup_bootmem_allocator();
 473
 474         return max_low_pfn;
 475 }
 476
 477 void __init zone_sizes_init(void)
 478 {
 479         unsigned long max_zone_pfns[MAX_NR_ZONES];
 480         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 481         max_zone_pfns[ZONE_DMA] =
 482                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 483         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 484 #ifdef CONFIG_HIGHMEM
 485         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 486         add_active_range(0, 0, highend_pfn);
 487 #else
 488         add_active_range(0, 0, max_low_pfn);
 489 #endif
 490
 491         free_area_init_nodes(max_zone_pfns);
 492 }
 493 #else
 494 extern unsigned long __init setup_memory(void);
 495 extern void zone_sizes_init(void);
 496 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 497
 498 static inline unsigned long long get_total_mem(void)
 499 {
 500         unsigned long long total;
 501
 502         total = max_low_pfn - min_low_pfn;
 503 #ifdef CONFIG_HIGHMEM
 504         total += highend_pfn - highstart_pfn;
 505 #endif
 506
 507         return total << PAGE_SHIFT;
 508 }
 509
 510 #ifdef CONFIG_KEXEC
 511 static void __init reserve_crashkernel(void)
 512 {
 513         unsigned long long total_mem;
 514         unsigned long long crash_size, crash_base;
 515         int ret;
 516
 517         total_mem = get_total_mem();
 518
 519         ret = parse_crashkernel(boot_command_line, total_mem,
 520                         &crash_size, &crash_base);
 521         if (ret == 0 && crash_size > 0) {
 522                 if (crash_base > 0) {
 523                         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 524                                         "for crashkernel (System RAM: %ldMB)\n",
 525                                         (unsigned long)(crash_size >> 20),
 526                                         (unsigned long)(crash_base >> 20),
 527                                         (unsigned long)(total_mem >> 20));
 528                         crashk_res.start = crash_base;
 529                         crashk_res.end   = crash_base + crash_size - 1;
 530                         reserve_bootmem(crash_base, crash_size,
 531                                         BOOTMEM_DEFAULT);
 532                 } else
 533                         printk(KERN_INFO "crashkernel reservation failed - "
 534                                         "you have to specify a base address\n");
 535         }
 536 }
 537 #else
 538 static inline void __init reserve_crashkernel(void)
 539 {}
 540 #endif
 541
 542 #ifdef CONFIG_BLK_DEV_INITRD
 543
 544 static bool do_relocate_initrd = false;
 545
 546 static void __init reserve_initrd(void)
 547 {
 548         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 549         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 550         unsigned long ramdisk_end   = ramdisk_image + ramdisk_size;
 551         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 552         unsigned long ramdisk_here;
 553
 554         initrd_start = 0;
 555
 556         if (!boot_params.hdr.type_of_loader ||
 557             !ramdisk_image || !ramdisk_size)
 558                 return;         /* No initrd provided by bootloader */
 559
 560         if (ramdisk_end < ramdisk_image) {
 561                 printk(KERN_ERR "initrd wraps around end of memory, "
 562                        "disabling initrd\n");
 563                 return;
 564         }
 565         if (ramdisk_size >= end_of_lowmem/2) {
 566                 printk(KERN_ERR "initrd too large to handle, "
 567                        "disabling initrd\n");
 568                 return;
 569         }
 570         if (ramdisk_end <= end_of_lowmem) {
 571                 /* All in lowmem, easy case */
 572                 reserve_bootmem(ramdisk_image, ramdisk_size, BOOTMEM_DEFAULT);
 573                 initrd_start = ramdisk_image + PAGE_OFFSET;
 574                 initrd_end = initrd_start+ramdisk_size;
 575                 return;
 576         }
 577
 578         /* We need to move the initrd down into lowmem */
 579         ramdisk_here = (end_of_lowmem - ramdisk_size) & PAGE_MASK;
 580
 581         /* Note: this includes all the lowmem currently occupied by
 582            the initrd, we rely on that fact to keep the data intact. */
 583         reserve_bootmem(ramdisk_here, ramdisk_size, BOOTMEM_DEFAULT);
 584         initrd_start = ramdisk_here + PAGE_OFFSET;
 585         initrd_end   = initrd_start + ramdisk_size;
 586
 587         do_relocate_initrd = true;
 588 }
 589
 590 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 591
 592 static void __init relocate_initrd(void)
 593 {
 594         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 595         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 596         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 597         unsigned long ramdisk_here;
 598         unsigned long slop, clen, mapaddr;
 599         char *p, *q;
 600
 601         if (!do_relocate_initrd)
 602                 return;
 603
 604         ramdisk_here = initrd_start - PAGE_OFFSET;
 605
 606         q = (char *)initrd_start;
 607
 608         /* Copy any lowmem portion of the initrd */
 609         if (ramdisk_image < end_of_lowmem) {
 610                 clen = end_of_lowmem - ramdisk_image;
 611                 p = (char *)__va(ramdisk_image);
 612                 memcpy(q, p, clen);
 613                 q += clen;
 614                 ramdisk_image += clen;
 615                 ramdisk_size  -= clen;
 616         }
 617
 618         /* Copy the highmem portion of the initrd */
 619         while (ramdisk_size) {
 620                 slop = ramdisk_image & ~PAGE_MASK;
 621                 clen = ramdisk_size;
 622                 if (clen > MAX_MAP_CHUNK-slop)
 623                         clen = MAX_MAP_CHUNK-slop;
 624                 mapaddr = ramdisk_image & PAGE_MASK;
 625                 p = early_ioremap(mapaddr, clen+slop);
 626                 memcpy(q, p+slop, clen);
 627                 early_iounmap(p, clen+slop);
 628                 q += clen;
 629                 ramdisk_image += clen;
 630                 ramdisk_size  -= clen;
 631         }
 632 }
 633
 634 #endif /* CONFIG_BLK_DEV_INITRD */
 635
 636 void __init setup_bootmem_allocator(void)
 637 {
 638         unsigned long bootmap_size;
 639         /*
 640          * Initialize the boot-time allocator (with low memory only):
 641          */
 642         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
 643
 644         register_bootmem_low_pages(max_low_pfn);
 645
 646         /*
 647          * Reserve the bootmem bitmap itself as well. We do this in two
 648          * steps (first step was init_bootmem()) because this catches
 649          * the (very unlikely) case of us accidentally initializing the
 650          * bootmem allocator with an invalid RAM area.
 651          */
 652         reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
 653                          bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text),
 654                          BOOTMEM_DEFAULT);
 655
 656         /*
 657          * reserve physical page 0 - it's a special BIOS page on many boxes,
 658          * enabling clean reboots, SMP operation, laptop functions.
 659          */
 660         reserve_bootmem(0, PAGE_SIZE, BOOTMEM_DEFAULT);
 661
 662         /* reserve EBDA region */
 663         reserve_ebda_region();
 664
 665 #ifdef CONFIG_SMP
 666         /*
 667          * But first pinch a few for the stack/trampoline stuff
 668          * FIXME: Don't need the extra page at 4K, but need to fix
 669          * trampoline before removing it. (see the GDT stuff)
 670          */
 671         reserve_bootmem(PAGE_SIZE, PAGE_SIZE, BOOTMEM_DEFAULT);
 672 #endif
 673 #ifdef CONFIG_ACPI_SLEEP
 674         /*
 675          * Reserve low memory region for sleep support.
 676          */
 677         acpi_reserve_bootmem();
 678 #endif
 679 #ifdef CONFIG_X86_FIND_SMP_CONFIG
 680         /*
 681          * Find and reserve possible boot-time SMP configuration:
 682          */
 683         find_smp_config();
 684 #endif
 685 #ifdef CONFIG_BLK_DEV_INITRD
 686         reserve_initrd();
 687 #endif
 688         numa_kva_reserve();
 689         reserve_crashkernel();
 690 }
 691
 692 /*
 693  * The node 0 pgdat is initialized before all of these because
 694  * it's needed for bootmem.  node>0 pgdats have their virtual
 695  * space allocated before the pagetables are in place to access
 696  * them, so they can't be cleared then.
 697  *
 698  * This should all compile down to nothing when NUMA is off.
 699  */
 700 static void __init remapped_pgdat_init(void)
 701 {
 702         int nid;
 703
 704         for_each_online_node(nid) {
 705                 if (nid != 0)
 706                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
 707         }
 708 }
 709
 710 #ifdef CONFIG_MCA
 711 static void set_mca_bus(int x)
 712 {
 713         MCA_bus = x;
 714 }
 715 #else
 716 static void set_mca_bus(int x) { }
 717 #endif
 718
 719 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
 720 char * __init __attribute__((weak)) memory_setup(void)
 721 {
 722         return machine_specific_memory_setup();
 723 }
 724
 725 #ifdef CONFIG_NUMA
 726 /*
 727  * In the golden day, when everything among i386 and x86_64 will be
 728  * integrated, this will not live here
 729  */
 730 void *x86_cpu_to_node_map_early_ptr;
 731 int x86_cpu_to_node_map_init[NR_CPUS] = {
 732         [0 ... NR_CPUS-1] = NUMA_NO_NODE
 733 };
 734 DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE;
 735 #endif
 736
 737 /*
 738  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 739  * passed the efi memmap, systab, etc., so we should use these data structures
 740  * for initialization.  Note, the efi init code path is determined by the
 741  * global efi_enabled. This allows the same kernel image to be used on existing
 742  * systems (with a traditional BIOS) as well as on EFI systems.
 743  */
 744 void __init setup_arch(char **cmdline_p)
 745 {
 746         unsigned long max_low_pfn;
 747
 748         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 749         pre_setup_arch_hook();
 750         early_cpu_init();
 751         early_ioremap_init();
 752
 753 #ifdef CONFIG_EFI
 754         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 755                      "EL32", 4))
 756                 efi_enabled = 1;
 757 #endif
 758
 759         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 760         screen_info = boot_params.screen_info;
 761         edid_info = boot_params.edid_info;
 762         apm_info.bios = boot_params.apm_bios_info;
 763         ist_info = boot_params.ist_info;
 764         saved_videomode = boot_params.hdr.vid_mode;
 765         if( boot_params.sys_desc_table.length != 0 ) {
 766                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 767                 machine_id = boot_params.sys_desc_table.table[0];
 768                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 769                 BIOS_revision = boot_params.sys_desc_table.table[2];
 770         }
 771         bootloader_type = boot_params.hdr.type_of_loader;
 772
 773 #ifdef CONFIG_BLK_DEV_RAM
 774         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 775         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 776         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 777 #endif
 778         ARCH_SETUP
 779
 780         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 781         print_memory_map(memory_setup());
 782
 783         copy_edd();
 784
 785         if (!boot_params.hdr.root_flags)
 786                 root_mountflags &= ~MS_RDONLY;
 787         init_mm.start_code = (unsigned long) _text;
 788         init_mm.end_code = (unsigned long) _etext;
 789         init_mm.end_data = (unsigned long) _edata;
 790         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 791
 792         code_resource.start = virt_to_phys(_text);
 793         code_resource.end = virt_to_phys(_etext)-1;
 794         data_resource.start = virt_to_phys(_etext);
 795         data_resource.end = virt_to_phys(_edata)-1;
 796         bss_resource.start = virt_to_phys(&__bss_start);
 797         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 798
 799         parse_early_param();
 800
 801         if (user_defined_memmap) {
 802                 printk(KERN_INFO "user-defined physical RAM map:\n");
 803                 print_memory_map("user");
 804         }
 805
 806         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 807         *cmdline_p = command_line;
 808
 809         if (efi_enabled)
 810                 efi_init();
 811
 812         /* update e820 for memory not covered by WB MTRRs */
 813         find_max_pfn();
 814         mtrr_bp_init();
 815         if (mtrr_trim_uncached_memory(max_pfn))
 816                 find_max_pfn();
 817
 818         max_low_pfn = setup_memory();
 819
 820 #ifdef CONFIG_VMI
 821         /*
 822          * Must be after max_low_pfn is determined, and before kernel
 823          * pagetables are setup.
 824          */
 825         vmi_init();
 826 #endif
 827
 828         /*
 829          * NOTE: before this point _nobody_ is allowed to allocate
 830          * any memory using the bootmem allocator.  Although the
 831          * allocator is now initialised only the first 8Mb of the kernel
 832          * virtual address space has been mapped.  All allocations before
 833          * paging_init() has completed must use the alloc_bootmem_low_pages()
 834          * variant (which allocates DMA'able memory) and care must be taken
 835          * not to exceed the 8Mb limit.
 836          */
 837
 838 #ifdef CONFIG_SMP
 839         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
 840 #endif
 841         paging_init();
 842
 843         /*
 844          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 845          */
 846
 847 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 848         if (init_ohci1394_dma_early)
 849                 init_ohci1394_dma_on_all_controllers();
 850 #endif
 851
 852         remapped_pgdat_init();
 853         sparse_init();
 854         zone_sizes_init();
 855
 856         /*
 857          * NOTE: at this point the bootmem allocator is fully available.
 858          */
 859
 860 #ifdef CONFIG_BLK_DEV_INITRD
 861         relocate_initrd();
 862 #endif
 863
 864         paravirt_post_allocator_init();
 865
 866         dmi_scan_machine();
 867
 868         io_delay_init();
 869
 870 #ifdef CONFIG_X86_SMP
 871         /*
 872          * setup to use the early static init tables during kernel startup
 873          * X86_SMP will exclude sub-arches that don't deal well with it.
 874          */
 875         x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
 876         x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
 877 #ifdef CONFIG_NUMA
 878         x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
 879 #endif
 880 #endif
 881
 882 #ifdef CONFIG_X86_GENERICARCH
 883         generic_apic_probe();
 884 #endif
 885
 886 #ifdef CONFIG_ACPI
 887         /*
 888          * Parse the ACPI tables for possible boot-time SMP configuration.
 889          */
 890         acpi_boot_table_init();
 891 #endif
 892
 893         early_quirks();
 894
 895 #ifdef CONFIG_ACPI
 896         acpi_boot_init();
 897
 898 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
 899         if (def_to_bigsmp)
 900                 printk(KERN_WARNING "More than 8 CPUs detected and "
 901                         "CONFIG_X86_PC cannot handle it.\nUse "
 902                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
 903 #endif
 904 #endif
 905 #ifdef CONFIG_X86_LOCAL_APIC
 906         if (smp_found_config)
 907                 get_smp_config();
 908 #endif
 909
 910         e820_register_memory();
 911         e820_mark_nosave_regions();
 912
 913 #ifdef CONFIG_VT
 914 #if defined(CONFIG_VGA_CONSOLE)
 915         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 916                 conswitchp = &vga_con;
 917 #elif defined(CONFIG_DUMMY_CONSOLE)
 918         conswitchp = &dummy_con;
 919 #endif
 920 #endif
 921 }
 922
 923 /*
 924  * Request address space for all standard resources
 925  *
 926  * This is called just before pcibios_init(), which is also a
 927  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
 928  */
 929 static int __init request_standard_resources(void)
 930 {
 931         int i;
 932
 933         printk(KERN_INFO "Setting up standard PCI resources\n");
 934         init_iomem_resources(&code_resource, &data_resource, &bss_resource);
 935
 936         request_resource(&iomem_resource, &video_ram_resource);
 937
 938         /* request I/O space for devices used on all i[345]86 PCs */
 939         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 940                 request_resource(&ioport_resource, &standard_io_resources[i]);
 941         return 0;
 942 }
 943
 944 subsys_initcall(request_standard_resources);