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    = 0x006f,
 131         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 132 }, {
 133         .name   = "dma page reg",
 134         .start  = 0x0080,
 135         .end    = 0x008f,
 136         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 137 }, {
 138         .name   = "pic2",
 139         .start  = 0x00a0,
 140         .end    = 0x00a1,
 141         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 142 }, {
 143         .name   = "dma2",
 144         .start  = 0x00c0,
 145         .end    = 0x00df,
 146         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 147 }, {
 148         .name   = "fpu",
 149         .start  = 0x00f0,
 150         .end    = 0x00ff,
 151         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 152 } };
 153
 154 /* cpu data as detected by the assembly code in head.S */
 155 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 156 /* common cpu data for all cpus */
 157 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 158 EXPORT_SYMBOL(boot_cpu_data);
 159
 160 unsigned int def_to_bigsmp;
 161
 162 #ifndef CONFIG_X86_PAE
 163 unsigned long mmu_cr4_features;
 164 #else
 165 unsigned long mmu_cr4_features = X86_CR4_PAE;
 166 #endif
 167
 168 /* for MCA, but anyone else can use it if they want */
 169 unsigned int machine_id;
 170 unsigned int machine_submodel_id;
 171 unsigned int BIOS_revision;
 172
 173 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 174 int bootloader_type;
 175
 176 /* user-defined highmem size */
 177 static unsigned int highmem_pages = -1;
 178
 179 /*
 180  * Setup options
 181  */
 182 struct screen_info screen_info;
 183 EXPORT_SYMBOL(screen_info);
 184 struct apm_info apm_info;
 185 EXPORT_SYMBOL(apm_info);
 186 struct edid_info edid_info;
 187 EXPORT_SYMBOL_GPL(edid_info);
 188 struct ist_info ist_info;
 189 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 190         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 191 EXPORT_SYMBOL(ist_info);
 192 #endif
 193
 194 extern void early_cpu_init(void);
 195 extern int root_mountflags;
 196
 197 unsigned long saved_video_mode;
 198
 199 #define RAMDISK_IMAGE_START_MASK        0x07FF
 200 #define RAMDISK_PROMPT_FLAG             0x8000
 201 #define RAMDISK_LOAD_FLAG               0x4000
 202
 203 static char __initdata command_line[COMMAND_LINE_SIZE];
 204
 205 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 206 struct boot_params __initdata boot_params;
 207 #else
 208 struct boot_params boot_params;
 209 #endif
 210
 211 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 212 struct edd edd;
 213 #ifdef CONFIG_EDD_MODULE
 214 EXPORT_SYMBOL(edd);
 215 #endif
 216 /**
 217  * copy_edd() - Copy the BIOS EDD information
 218  *              from boot_params into a safe place.
 219  *
 220  */
 221 static inline void copy_edd(void)
 222 {
 223      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 224             sizeof(edd.mbr_signature));
 225      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 226      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 227      edd.edd_info_nr = boot_params.eddbuf_entries;
 228 }
 229 #else
 230 static inline void copy_edd(void)
 231 {
 232 }
 233 #endif
 234
 235 int __initdata user_defined_memmap;
 236
 237 /*
 238  * "mem=nopentium" disables the 4MB page tables.
 239  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
 240  * to <mem>, overriding the bios size.
 241  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
 242  * <start> to <start>+<mem>, overriding the bios size.
 243  *
 244  * HPA tells me bootloaders need to parse mem=, so no new
 245  * option should be mem=  [also see Documentation/i386/boot.txt]
 246  */
 247 static int __init parse_mem(char *arg)
 248 {
 249         if (!arg)
 250                 return -EINVAL;
 251
 252         if (strcmp(arg, "nopentium") == 0) {
 253                 setup_clear_cpu_cap(X86_FEATURE_PSE);
 254         } else {
 255                 /* If the user specifies memory size, we
 256                  * limit the BIOS-provided memory map to
 257                  * that size. exactmap can be used to specify
 258                  * the exact map. mem=number can be used to
 259                  * trim the existing memory map.
 260                  */
 261                 unsigned long long mem_size;
 262
 263                 mem_size = memparse(arg, &arg);
 264                 limit_regions(mem_size);
 265                 user_defined_memmap = 1;
 266         }
 267         return 0;
 268 }
 269 early_param("mem", parse_mem);
 270
 271 #ifdef CONFIG_PROC_VMCORE
 272 /* elfcorehdr= specifies the location of elf core header
 273  * stored by the crashed kernel.
 274  */
 275 static int __init parse_elfcorehdr(char *arg)
 276 {
 277         if (!arg)
 278                 return -EINVAL;
 279
 280         elfcorehdr_addr = memparse(arg, &arg);
 281         return 0;
 282 }
 283 early_param("elfcorehdr", parse_elfcorehdr);
 284 #endif /* CONFIG_PROC_VMCORE */
 285
 286 /*
 287  * highmem=size forces highmem to be exactly 'size' bytes.
 288  * This works even on boxes that have no highmem otherwise.
 289  * This also works to reduce highmem size on bigger boxes.
 290  */
 291 static int __init parse_highmem(char *arg)
 292 {
 293         if (!arg)
 294                 return -EINVAL;
 295
 296         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 297         return 0;
 298 }
 299 early_param("highmem", parse_highmem);
 300
 301 /*
 302  * vmalloc=size forces the vmalloc area to be exactly 'size'
 303  * bytes. This can be used to increase (or decrease) the
 304  * vmalloc area - the default is 128m.
 305  */
 306 static int __init parse_vmalloc(char *arg)
 307 {
 308         if (!arg)
 309                 return -EINVAL;
 310
 311         __VMALLOC_RESERVE = memparse(arg, &arg);
 312         return 0;
 313 }
 314 early_param("vmalloc", parse_vmalloc);
 315
 316 /*
 317  * reservetop=size reserves a hole at the top of the kernel address space which
 318  * a hypervisor can load into later.  Needed for dynamically loaded hypervisors,
 319  * so relocating the fixmap can be done before paging initialization.
 320  */
 321 static int __init parse_reservetop(char *arg)
 322 {
 323         unsigned long address;
 324
 325         if (!arg)
 326                 return -EINVAL;
 327
 328         address = memparse(arg, &arg);
 329         reserve_top_address(address);
 330         return 0;
 331 }
 332 early_param("reservetop", parse_reservetop);
 333
 334 /*
 335  * Determine low and high memory ranges:
 336  */
 337 unsigned long __init find_max_low_pfn(void)
 338 {
 339         unsigned long max_low_pfn;
 340
 341         max_low_pfn = max_pfn;
 342         if (max_low_pfn > MAXMEM_PFN) {
 343                 if (highmem_pages == -1)
 344                         highmem_pages = max_pfn - MAXMEM_PFN;
 345                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 346                         max_pfn = MAXMEM_PFN + highmem_pages;
 347                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 348                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
 349                         highmem_pages = 0;
 350                 }
 351                 max_low_pfn = MAXMEM_PFN;
 352 #ifndef CONFIG_HIGHMEM
 353                 /* Maximum memory usable is what is directly addressable */
 354                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 355                                         MAXMEM>>20);
 356                 if (max_pfn > MAX_NONPAE_PFN)
 357                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 358                 else
 359                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 360                 max_pfn = MAXMEM_PFN;
 361 #else /* !CONFIG_HIGHMEM */
 362 #ifndef CONFIG_HIGHMEM64G
 363                 if (max_pfn > MAX_NONPAE_PFN) {
 364                         max_pfn = MAX_NONPAE_PFN;
 365                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
 366                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 367                 }
 368 #endif /* !CONFIG_HIGHMEM64G */
 369 #endif /* !CONFIG_HIGHMEM */
 370         } else {
 371                 if (highmem_pages == -1)
 372                         highmem_pages = 0;
 373 #ifdef CONFIG_HIGHMEM
 374                 if (highmem_pages >= max_pfn) {
 375                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 376                         highmem_pages = 0;
 377                 }
 378                 if (highmem_pages) {
 379                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
 380                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
 381                                 highmem_pages = 0;
 382                         }
 383                         max_low_pfn -= highmem_pages;
 384                 }
 385 #else
 386                 if (highmem_pages)
 387                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 388 #endif
 389         }
 390         return max_low_pfn;
 391 }
 392
 393 #define BIOS_LOWMEM_KILOBYTES 0x413
 394
 395 /*
 396  * The BIOS places the EBDA/XBDA at the top of conventional
 397  * memory, and usually decreases the reported amount of
 398  * conventional memory (int 0x12) too. This also contains a
 399  * workaround for Dell systems that neglect to reserve EBDA.
 400  * The same workaround also avoids a problem with the AMD768MPX
 401  * chipset: reserve a page before VGA to prevent PCI prefetch
 402  * into it (errata #56). Usually the page is reserved anyways,
 403  * unless you have no PS/2 mouse plugged in.
 404  */
 405 static void __init reserve_ebda_region(void)
 406 {
 407         unsigned int lowmem, ebda_addr;
 408
 409         /* To determine the position of the EBDA and the */
 410         /* end of conventional memory, we need to look at */
 411         /* the BIOS data area. In a paravirtual environment */
 412         /* that area is absent. We'll just have to assume */
 413         /* that the paravirt case can handle memory setup */
 414         /* correctly, without our help. */
 415         if (paravirt_enabled())
 416                 return;
 417
 418         /* end of low (conventional) memory */
 419         lowmem = *(unsigned short *)__va(BIOS_LOWMEM_KILOBYTES);
 420         lowmem <<= 10;
 421
 422         /* start of EBDA area */
 423         ebda_addr = get_bios_ebda();
 424
 425         /* Fixup: bios puts an EBDA in the top 64K segment */
 426         /* of conventional memory, but does not adjust lowmem. */
 427         if ((lowmem - ebda_addr) <= 0x10000)
 428                 lowmem = ebda_addr;
 429
 430         /* Fixup: bios does not report an EBDA at all. */
 431         /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */
 432         if ((ebda_addr == 0) && (lowmem >= 0x9f000))
 433                 lowmem = 0x9f000;
 434
 435         /* Paranoia: should never happen, but... */
 436         if ((lowmem == 0) || (lowmem >= 0x100000))
 437                 lowmem = 0x9f000;
 438
 439         /* reserve all memory between lowmem and the 1MB mark */
 440         reserve_bootmem(lowmem, 0x100000 - lowmem, BOOTMEM_DEFAULT);
 441 }
 442
 443 #ifndef CONFIG_NEED_MULTIPLE_NODES
 444 static void __init setup_bootmem_allocator(void);
 445 static unsigned long __init setup_memory(void)
 446 {
 447         /*
 448          * partially used pages are not usable - thus
 449          * we are rounding upwards:
 450          */
 451         min_low_pfn = PFN_UP(init_pg_tables_end);
 452
 453         max_low_pfn = find_max_low_pfn();
 454
 455 #ifdef CONFIG_HIGHMEM
 456         highstart_pfn = highend_pfn = max_pfn;
 457         if (max_pfn > max_low_pfn) {
 458                 highstart_pfn = max_low_pfn;
 459         }
 460         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 461                 pages_to_mb(highend_pfn - highstart_pfn));
 462         num_physpages = highend_pfn;
 463         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 464 #else
 465         num_physpages = max_low_pfn;
 466         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 467 #endif
 468 #ifdef CONFIG_FLATMEM
 469         max_mapnr = num_physpages;
 470 #endif
 471         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 472                         pages_to_mb(max_low_pfn));
 473
 474         setup_bootmem_allocator();
 475
 476         return max_low_pfn;
 477 }
 478
 479 static void __init zone_sizes_init(void)
 480 {
 481         unsigned long max_zone_pfns[MAX_NR_ZONES];
 482         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 483         max_zone_pfns[ZONE_DMA] =
 484                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 485         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 486 #ifdef CONFIG_HIGHMEM
 487         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 488         add_active_range(0, 0, highend_pfn);
 489 #else
 490         add_active_range(0, 0, max_low_pfn);
 491 #endif
 492
 493         free_area_init_nodes(max_zone_pfns);
 494 }
 495 #else
 496 extern unsigned long __init setup_memory(void);
 497 extern void zone_sizes_init(void);
 498 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 499
 500 static inline unsigned long long get_total_mem(void)
 501 {
 502         unsigned long long total;
 503
 504         total = max_low_pfn - min_low_pfn;
 505 #ifdef CONFIG_HIGHMEM
 506         total += highend_pfn - highstart_pfn;
 507 #endif
 508
 509         return total << PAGE_SHIFT;
 510 }
 511
 512 #ifdef CONFIG_KEXEC
 513 static void __init reserve_crashkernel(void)
 514 {
 515         unsigned long long total_mem;
 516         unsigned long long crash_size, crash_base;
 517         int ret;
 518
 519         total_mem = get_total_mem();
 520
 521         ret = parse_crashkernel(boot_command_line, total_mem,
 522                         &crash_size, &crash_base);
 523         if (ret == 0 && crash_size > 0) {
 524                 if (crash_base > 0) {
 525                         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 526                                         "for crashkernel (System RAM: %ldMB)\n",
 527                                         (unsigned long)(crash_size >> 20),
 528                                         (unsigned long)(crash_base >> 20),
 529                                         (unsigned long)(total_mem >> 20));
 530                         crashk_res.start = crash_base;
 531                         crashk_res.end   = crash_base + crash_size - 1;
 532                         reserve_bootmem(crash_base, crash_size,
 533                                         BOOTMEM_DEFAULT);
 534                 } else
 535                         printk(KERN_INFO "crashkernel reservation failed - "
 536                                         "you have to specify a base address\n");
 537         }
 538 }
 539 #else
 540 static inline void __init reserve_crashkernel(void)
 541 {}
 542 #endif
 543
 544 #ifdef CONFIG_BLK_DEV_INITRD
 545
 546 static bool do_relocate_initrd = false;
 547
 548 static void __init reserve_initrd(void)
 549 {
 550         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 551         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 552         unsigned long ramdisk_end   = ramdisk_image + ramdisk_size;
 553         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 554         unsigned long ramdisk_here;
 555
 556         initrd_start = 0;
 557
 558         if (!boot_params.hdr.type_of_loader ||
 559             !ramdisk_image || !ramdisk_size)
 560                 return;         /* No initrd provided by bootloader */
 561
 562         if (ramdisk_end < ramdisk_image) {
 563                 printk(KERN_ERR "initrd wraps around end of memory, "
 564                        "disabling initrd\n");
 565                 return;
 566         }
 567         if (ramdisk_size >= end_of_lowmem/2) {
 568                 printk(KERN_ERR "initrd too large to handle, "
 569                        "disabling initrd\n");
 570                 return;
 571         }
 572         if (ramdisk_end <= end_of_lowmem) {
 573                 /* All in lowmem, easy case */
 574                 reserve_bootmem(ramdisk_image, ramdisk_size, BOOTMEM_DEFAULT);
 575                 initrd_start = ramdisk_image + PAGE_OFFSET;
 576                 initrd_end = initrd_start+ramdisk_size;
 577                 return;
 578         }
 579
 580         /* We need to move the initrd down into lowmem */
 581         ramdisk_here = (end_of_lowmem - ramdisk_size) & PAGE_MASK;
 582
 583         /* Note: this includes all the lowmem currently occupied by
 584            the initrd, we rely on that fact to keep the data intact. */
 585         reserve_bootmem(ramdisk_here, ramdisk_size, BOOTMEM_DEFAULT);
 586         initrd_start = ramdisk_here + PAGE_OFFSET;
 587         initrd_end   = initrd_start + ramdisk_size;
 588
 589         do_relocate_initrd = true;
 590 }
 591
 592 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 593
 594 static void __init relocate_initrd(void)
 595 {
 596         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 597         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 598         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 599         unsigned long ramdisk_here;
 600         unsigned long slop, clen, mapaddr;
 601         char *p, *q;
 602
 603         if (!do_relocate_initrd)
 604                 return;
 605
 606         ramdisk_here = initrd_start - PAGE_OFFSET;
 607
 608         q = (char *)initrd_start;
 609
 610         /* Copy any lowmem portion of the initrd */
 611         if (ramdisk_image < end_of_lowmem) {
 612                 clen = end_of_lowmem - ramdisk_image;
 613                 p = (char *)__va(ramdisk_image);
 614                 memcpy(q, p, clen);
 615                 q += clen;
 616                 ramdisk_image += clen;
 617                 ramdisk_size  -= clen;
 618         }
 619
 620         /* Copy the highmem portion of the initrd */
 621         while (ramdisk_size) {
 622                 slop = ramdisk_image & ~PAGE_MASK;
 623                 clen = ramdisk_size;
 624                 if (clen > MAX_MAP_CHUNK-slop)
 625                         clen = MAX_MAP_CHUNK-slop;
 626                 mapaddr = ramdisk_image & PAGE_MASK;
 627                 p = early_ioremap(mapaddr, clen+slop);
 628                 memcpy(q, p+slop, clen);
 629                 early_iounmap(p, clen+slop);
 630                 q += clen;
 631                 ramdisk_image += clen;
 632                 ramdisk_size  -= clen;
 633         }
 634 }
 635
 636 #endif /* CONFIG_BLK_DEV_INITRD */
 637
 638 void __init setup_bootmem_allocator(void)
 639 {
 640         unsigned long bootmap_size;
 641         /*
 642          * Initialize the boot-time allocator (with low memory only):
 643          */
 644         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
 645
 646         register_bootmem_low_pages(max_low_pfn);
 647
 648         /*
 649          * Reserve the bootmem bitmap itself as well. We do this in two
 650          * steps (first step was init_bootmem()) because this catches
 651          * the (very unlikely) case of us accidentally initializing the
 652          * bootmem allocator with an invalid RAM area.
 653          */
 654         reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
 655                          bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text),
 656                          BOOTMEM_DEFAULT);
 657
 658         /*
 659          * reserve physical page 0 - it's a special BIOS page on many boxes,
 660          * enabling clean reboots, SMP operation, laptop functions.
 661          */
 662         reserve_bootmem(0, PAGE_SIZE, BOOTMEM_DEFAULT);
 663
 664         /* reserve EBDA region */
 665         reserve_ebda_region();
 666
 667 #ifdef CONFIG_SMP
 668         /*
 669          * But first pinch a few for the stack/trampoline stuff
 670          * FIXME: Don't need the extra page at 4K, but need to fix
 671          * trampoline before removing it. (see the GDT stuff)
 672          */
 673         reserve_bootmem(PAGE_SIZE, PAGE_SIZE, BOOTMEM_DEFAULT);
 674 #endif
 675 #ifdef CONFIG_ACPI_SLEEP
 676         /*
 677          * Reserve low memory region for sleep support.
 678          */
 679         acpi_reserve_bootmem();
 680 #endif
 681 #ifdef CONFIG_X86_FIND_SMP_CONFIG
 682         /*
 683          * Find and reserve possible boot-time SMP configuration:
 684          */
 685         find_smp_config();
 686 #endif
 687 #ifdef CONFIG_BLK_DEV_INITRD
 688         reserve_initrd();
 689 #endif
 690         numa_kva_reserve();
 691         reserve_crashkernel();
 692
 693         reserve_ibft_region();
 694 }
 695
 696 /*
 697  * The node 0 pgdat is initialized before all of these because
 698  * it's needed for bootmem.  node>0 pgdats have their virtual
 699  * space allocated before the pagetables are in place to access
 700  * them, so they can't be cleared then.
 701  *
 702  * This should all compile down to nothing when NUMA is off.
 703  */
 704 static void __init remapped_pgdat_init(void)
 705 {
 706         int nid;
 707
 708         for_each_online_node(nid) {
 709                 if (nid != 0)
 710                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
 711         }
 712 }
 713
 714 #ifdef CONFIG_MCA
 715 static void set_mca_bus(int x)
 716 {
 717         MCA_bus = x;
 718 }
 719 #else
 720 static void set_mca_bus(int x) { }
 721 #endif
 722
 723 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
 724 char * __init __attribute__((weak)) memory_setup(void)
 725 {
 726         return machine_specific_memory_setup();
 727 }
 728
 729 #ifdef CONFIG_NUMA
 730 /*
 731  * In the golden day, when everything among i386 and x86_64 will be
 732  * integrated, this will not live here
 733  */
 734 void *x86_cpu_to_node_map_early_ptr;
 735 int x86_cpu_to_node_map_init[NR_CPUS] = {
 736         [0 ... NR_CPUS-1] = NUMA_NO_NODE
 737 };
 738 DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE;
 739 #endif
 740
 741 /*
 742  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 743  * passed the efi memmap, systab, etc., so we should use these data structures
 744  * for initialization.  Note, the efi init code path is determined by the
 745  * global efi_enabled. This allows the same kernel image to be used on existing
 746  * systems (with a traditional BIOS) as well as on EFI systems.
 747  */
 748 void __init setup_arch(char **cmdline_p)
 749 {
 750         unsigned long max_low_pfn;
 751
 752         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 753         pre_setup_arch_hook();
 754         early_cpu_init();
 755         early_ioremap_init();
 756
 757 #ifdef CONFIG_EFI
 758         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 759                      "EL32", 4))
 760                 efi_enabled = 1;
 761 #endif
 762
 763         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 764         screen_info = boot_params.screen_info;
 765         edid_info = boot_params.edid_info;
 766         apm_info.bios = boot_params.apm_bios_info;
 767         ist_info = boot_params.ist_info;
 768         saved_video_mode = boot_params.hdr.vid_mode;
 769         if( boot_params.sys_desc_table.length != 0 ) {
 770                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 771                 machine_id = boot_params.sys_desc_table.table[0];
 772                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 773                 BIOS_revision = boot_params.sys_desc_table.table[2];
 774         }
 775         bootloader_type = boot_params.hdr.type_of_loader;
 776
 777 #ifdef CONFIG_BLK_DEV_RAM
 778         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 779         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 780         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 781 #endif
 782         ARCH_SETUP
 783
 784         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 785         print_memory_map(memory_setup());
 786
 787         copy_edd();
 788
 789         if (!boot_params.hdr.root_flags)
 790                 root_mountflags &= ~MS_RDONLY;
 791         init_mm.start_code = (unsigned long) _text;
 792         init_mm.end_code = (unsigned long) _etext;
 793         init_mm.end_data = (unsigned long) _edata;
 794         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 795
 796         code_resource.start = virt_to_phys(_text);
 797         code_resource.end = virt_to_phys(_etext)-1;
 798         data_resource.start = virt_to_phys(_etext);
 799         data_resource.end = virt_to_phys(_edata)-1;
 800         bss_resource.start = virt_to_phys(&__bss_start);
 801         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 802
 803         parse_early_param();
 804
 805         if (user_defined_memmap) {
 806                 printk(KERN_INFO "user-defined physical RAM map:\n");
 807                 print_memory_map("user");
 808         }
 809
 810         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 811         *cmdline_p = command_line;
 812
 813         if (efi_enabled)
 814                 efi_init();
 815
 816         /* update e820 for memory not covered by WB MTRRs */
 817         propagate_e820_map();
 818         mtrr_bp_init();
 819         if (mtrr_trim_uncached_memory(max_pfn))
 820                 propagate_e820_map();
 821
 822         max_low_pfn = setup_memory();
 823
 824 #ifdef CONFIG_KVM_CLOCK
 825         kvmclock_init();
 826 #endif
 827
 828 #ifdef CONFIG_VMI
 829         /*
 830          * Must be after max_low_pfn is determined, and before kernel
 831          * pagetables are setup.
 832          */
 833         vmi_init();
 834 #endif
 835         kvm_guest_init();
 836
 837         /*
 838          * NOTE: before this point _nobody_ is allowed to allocate
 839          * any memory using the bootmem allocator.  Although the
 840          * allocator is now initialised only the first 8Mb of the kernel
 841          * virtual address space has been mapped.  All allocations before
 842          * paging_init() has completed must use the alloc_bootmem_low_pages()
 843          * variant (which allocates DMA'able memory) and care must be taken
 844          * not to exceed the 8Mb limit.
 845          */
 846
 847 #ifdef CONFIG_SMP
 848         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
 849 #endif
 850         paging_init();
 851
 852         /*
 853          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 854          */
 855
 856 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 857         if (init_ohci1394_dma_early)
 858                 init_ohci1394_dma_on_all_controllers();
 859 #endif
 860
 861         remapped_pgdat_init();
 862         sparse_init();
 863         zone_sizes_init();
 864
 865         /*
 866          * NOTE: at this point the bootmem allocator is fully available.
 867          */
 868
 869 #ifdef CONFIG_BLK_DEV_INITRD
 870         relocate_initrd();
 871 #endif
 872
 873         paravirt_post_allocator_init();
 874
 875         dmi_scan_machine();
 876
 877         io_delay_init();
 878
 879 #ifdef CONFIG_X86_SMP
 880         /*
 881          * setup to use the early static init tables during kernel startup
 882          * X86_SMP will exclude sub-arches that don't deal well with it.
 883          */
 884         x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
 885         x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
 886 #ifdef CONFIG_NUMA
 887         x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
 888 #endif
 889 #endif
 890
 891 #ifdef CONFIG_X86_GENERICARCH
 892         generic_apic_probe();
 893 #endif
 894
 895 #ifdef CONFIG_ACPI
 896         /*
 897          * Parse the ACPI tables for possible boot-time SMP configuration.
 898          */
 899         acpi_boot_table_init();
 900 #endif
 901
 902         early_quirks();
 903
 904 #ifdef CONFIG_ACPI
 905         acpi_boot_init();
 906
 907 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
 908         if (def_to_bigsmp)
 909                 printk(KERN_WARNING "More than 8 CPUs detected and "
 910                         "CONFIG_X86_PC cannot handle it.\nUse "
 911                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
 912 #endif
 913 #endif
 914 #ifdef CONFIG_X86_LOCAL_APIC
 915         if (smp_found_config)
 916                 get_smp_config();
 917 #endif
 918
 919         e820_register_memory();
 920         e820_mark_nosave_regions();
 921
 922 #ifdef CONFIG_VT
 923 #if defined(CONFIG_VGA_CONSOLE)
 924         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 925                 conswitchp = &vga_con;
 926 #elif defined(CONFIG_DUMMY_CONSOLE)
 927         conswitchp = &dummy_con;
 928 #endif
 929 #endif
 930 }
 931
 932 /*
 933  * Request address space for all standard resources
 934  *
 935  * This is called just before pcibios_init(), which is also a
 936  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
 937  */
 938 static int __init request_standard_resources(void)
 939 {
 940         int i;
 941
 942         printk(KERN_INFO "Setting up standard PCI resources\n");
 943         init_iomem_resources(&code_resource, &data_resource, &bss_resource);
 944
 945         request_resource(&iomem_resource, &video_ram_resource);
 946
 947         /* request I/O space for devices used on all i[345]86 PCs */
 948         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 949                 request_resource(&ioport_resource, &standard_io_resources[i]);
 950         return 0;
 951 }
 952
 953 subsys_initcall(request_standard_resources);