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
  51 #include <video/edid.h>
  52
  53 #include <asm/mtrr.h>
  54 #include <asm/apic.h>
  55 #include <asm/e820.h>
  56 #include <asm/mpspec.h>
  57 #include <asm/mmzone.h>
  58 #include <asm/setup.h>
  59 #include <asm/arch_hooks.h>
  60 #include <asm/sections.h>
  61 #include <asm/io_apic.h>
  62 #include <asm/ist.h>
  63 #include <asm/io.h>
  64 #include <asm/vmi.h>
  65 #include <setup_arch.h>
  66 #include <asm/bios_ebda.h>
  67 #include <asm/cacheflush.h>
  68 #include <asm/processor.h>
  69
  70 /* This value is set up by the early boot code to point to the value
  71    immediately after the boot time page tables.  It contains a *physical*
  72    address, and must not be in the .bss segment! */
  73 unsigned long init_pg_tables_end __initdata = ~0UL;
  74
  75 /*
  76  * Machine setup..
  77  */
  78 static struct resource data_resource = {
  79         .name   = "Kernel data",
  80         .start  = 0,
  81         .end    = 0,
  82         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  83 };
  84
  85 static struct resource code_resource = {
  86         .name   = "Kernel code",
  87         .start  = 0,
  88         .end    = 0,
  89         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  90 };
  91
  92 static struct resource bss_resource = {
  93         .name   = "Kernel bss",
  94         .start  = 0,
  95         .end    = 0,
  96         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  97 };
  98
  99 static struct resource video_ram_resource = {
 100         .name   = "Video RAM area",
 101         .start  = 0xa0000,
 102         .end    = 0xbffff,
 103         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 104 };
 105
 106 static struct resource standard_io_resources[] = { {
 107         .name   = "dma1",
 108         .start  = 0x0000,
 109         .end    = 0x001f,
 110         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 111 }, {
 112         .name   = "pic1",
 113         .start  = 0x0020,
 114         .end    = 0x0021,
 115         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 116 }, {
 117         .name   = "timer0",
 118         .start  = 0x0040,
 119         .end    = 0x0043,
 120         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 121 }, {
 122         .name   = "timer1",
 123         .start  = 0x0050,
 124         .end    = 0x0053,
 125         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 126 }, {
 127         .name   = "keyboard",
 128         .start  = 0x0060,
 129         .end    = 0x006f,
 130         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 131 }, {
 132         .name   = "dma page reg",
 133         .start  = 0x0080,
 134         .end    = 0x008f,
 135         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 136 }, {
 137         .name   = "pic2",
 138         .start  = 0x00a0,
 139         .end    = 0x00a1,
 140         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 141 }, {
 142         .name   = "dma2",
 143         .start  = 0x00c0,
 144         .end    = 0x00df,
 145         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 146 }, {
 147         .name   = "fpu",
 148         .start  = 0x00f0,
 149         .end    = 0x00ff,
 150         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 151 } };
 152
 153 /* cpu data as detected by the assembly code in head.S */
 154 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 155 /* common cpu data for all cpus */
 156 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 157 EXPORT_SYMBOL(boot_cpu_data);
 158
 159 unsigned int def_to_bigsmp;
 160
 161 #ifndef CONFIG_X86_PAE
 162 unsigned long mmu_cr4_features;
 163 #else
 164 unsigned long mmu_cr4_features = X86_CR4_PAE;
 165 #endif
 166
 167 /* for MCA, but anyone else can use it if they want */
 168 unsigned int machine_id;
 169 unsigned int machine_submodel_id;
 170 unsigned int BIOS_revision;
 171
 172 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 173 int bootloader_type;
 174
 175 /* user-defined highmem size */
 176 static unsigned int highmem_pages = -1;
 177
 178 /*
 179  * Setup options
 180  */
 181 struct screen_info screen_info;
 182 EXPORT_SYMBOL(screen_info);
 183 struct apm_info apm_info;
 184 EXPORT_SYMBOL(apm_info);
 185 struct edid_info edid_info;
 186 EXPORT_SYMBOL_GPL(edid_info);
 187 struct ist_info ist_info;
 188 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 189         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 190 EXPORT_SYMBOL(ist_info);
 191 #endif
 192
 193 extern void early_cpu_init(void);
 194 extern int root_mountflags;
 195
 196 unsigned long saved_video_mode;
 197
 198 #define RAMDISK_IMAGE_START_MASK        0x07FF
 199 #define RAMDISK_PROMPT_FLAG             0x8000
 200 #define RAMDISK_LOAD_FLAG               0x4000
 201
 202 static char __initdata command_line[COMMAND_LINE_SIZE];
 203
 204 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 205 struct boot_params __initdata boot_params;
 206 #else
 207 struct boot_params boot_params;
 208 #endif
 209
 210 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 211 struct edd edd;
 212 #ifdef CONFIG_EDD_MODULE
 213 EXPORT_SYMBOL(edd);
 214 #endif
 215 /**
 216  * copy_edd() - Copy the BIOS EDD information
 217  *              from boot_params into a safe place.
 218  *
 219  */
 220 static inline void copy_edd(void)
 221 {
 222      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 223             sizeof(edd.mbr_signature));
 224      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 225      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 226      edd.edd_info_nr = boot_params.eddbuf_entries;
 227 }
 228 #else
 229 static inline void copy_edd(void)
 230 {
 231 }
 232 #endif
 233
 234 int __initdata user_defined_memmap;
 235
 236 /*
 237  * "mem=nopentium" disables the 4MB page tables.
 238  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
 239  * to <mem>, overriding the bios size.
 240  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
 241  * <start> to <start>+<mem>, overriding the bios size.
 242  *
 243  * HPA tells me bootloaders need to parse mem=, so no new
 244  * option should be mem=  [also see Documentation/i386/boot.txt]
 245  */
 246 static int __init parse_mem(char *arg)
 247 {
 248         if (!arg)
 249                 return -EINVAL;
 250
 251         if (strcmp(arg, "nopentium") == 0) {
 252                 setup_clear_cpu_cap(X86_FEATURE_PSE);
 253         } else {
 254                 /* If the user specifies memory size, we
 255                  * limit the BIOS-provided memory map to
 256                  * that size. exactmap can be used to specify
 257                  * the exact map. mem=number can be used to
 258                  * trim the existing memory map.
 259                  */
 260                 unsigned long long mem_size;
 261
 262                 mem_size = memparse(arg, &arg);
 263                 limit_regions(mem_size);
 264                 user_defined_memmap = 1;
 265         }
 266         return 0;
 267 }
 268 early_param("mem", parse_mem);
 269
 270 #ifdef CONFIG_PROC_VMCORE
 271 /* elfcorehdr= specifies the location of elf core header
 272  * stored by the crashed kernel.
 273  */
 274 static int __init parse_elfcorehdr(char *arg)
 275 {
 276         if (!arg)
 277                 return -EINVAL;
 278
 279         elfcorehdr_addr = memparse(arg, &arg);
 280         return 0;
 281 }
 282 early_param("elfcorehdr", parse_elfcorehdr);
 283 #endif /* CONFIG_PROC_VMCORE */
 284
 285 /*
 286  * highmem=size forces highmem to be exactly 'size' bytes.
 287  * This works even on boxes that have no highmem otherwise.
 288  * This also works to reduce highmem size on bigger boxes.
 289  */
 290 static int __init parse_highmem(char *arg)
 291 {
 292         if (!arg)
 293                 return -EINVAL;
 294
 295         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 296         return 0;
 297 }
 298 early_param("highmem", parse_highmem);
 299
 300 /*
 301  * vmalloc=size forces the vmalloc area to be exactly 'size'
 302  * bytes. This can be used to increase (or decrease) the
 303  * vmalloc area - the default is 128m.
 304  */
 305 static int __init parse_vmalloc(char *arg)
 306 {
 307         if (!arg)
 308                 return -EINVAL;
 309
 310         __VMALLOC_RESERVE = memparse(arg, &arg);
 311         return 0;
 312 }
 313 early_param("vmalloc", parse_vmalloc);
 314
 315 /*
 316  * reservetop=size reserves a hole at the top of the kernel address space which
 317  * a hypervisor can load into later.  Needed for dynamically loaded hypervisors,
 318  * so relocating the fixmap can be done before paging initialization.
 319  */
 320 static int __init parse_reservetop(char *arg)
 321 {
 322         unsigned long address;
 323
 324         if (!arg)
 325                 return -EINVAL;
 326
 327         address = memparse(arg, &arg);
 328         reserve_top_address(address);
 329         return 0;
 330 }
 331 early_param("reservetop", parse_reservetop);
 332
 333 /*
 334  * Determine low and high memory ranges:
 335  */
 336 unsigned long __init find_max_low_pfn(void)
 337 {
 338         unsigned long max_low_pfn;
 339
 340         max_low_pfn = max_pfn;
 341         if (max_low_pfn > MAXMEM_PFN) {
 342                 if (highmem_pages == -1)
 343                         highmem_pages = max_pfn - MAXMEM_PFN;
 344                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 345                         max_pfn = MAXMEM_PFN + highmem_pages;
 346                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 347                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
 348                         highmem_pages = 0;
 349                 }
 350                 max_low_pfn = MAXMEM_PFN;
 351 #ifndef CONFIG_HIGHMEM
 352                 /* Maximum memory usable is what is directly addressable */
 353                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 354                                         MAXMEM>>20);
 355                 if (max_pfn > MAX_NONPAE_PFN)
 356                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 357                 else
 358                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 359                 max_pfn = MAXMEM_PFN;
 360 #else /* !CONFIG_HIGHMEM */
 361 #ifndef CONFIG_HIGHMEM64G
 362                 if (max_pfn > MAX_NONPAE_PFN) {
 363                         max_pfn = MAX_NONPAE_PFN;
 364                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
 365                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 366                 }
 367 #endif /* !CONFIG_HIGHMEM64G */
 368 #endif /* !CONFIG_HIGHMEM */
 369         } else {
 370                 if (highmem_pages == -1)
 371                         highmem_pages = 0;
 372 #ifdef CONFIG_HIGHMEM
 373                 if (highmem_pages >= max_pfn) {
 374                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 375                         highmem_pages = 0;
 376                 }
 377                 if (highmem_pages) {
 378                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
 379                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
 380                                 highmem_pages = 0;
 381                         }
 382                         max_low_pfn -= highmem_pages;
 383                 }
 384 #else
 385                 if (highmem_pages)
 386                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 387 #endif
 388         }
 389         return max_low_pfn;
 390 }
 391
 392 #define BIOS_EBDA_SEGMENT 0x40E
 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 = *(unsigned short *)__va(BIOS_EBDA_SEGMENT);
 424         ebda_addr <<= 4;
 425
 426         /* Fixup: bios puts an EBDA in the top 64K segment */
 427         /* of conventional memory, but does not adjust lowmem. */
 428         if ((lowmem - ebda_addr) <= 0x10000)
 429                 lowmem = ebda_addr;
 430
 431         /* Fixup: bios does not report an EBDA at all. */
 432         /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */
 433         if ((ebda_addr == 0) && (lowmem >= 0x9f000))
 434                 lowmem = 0x9f000;
 435
 436         /* Paranoia: should never happen, but... */
 437         if ((lowmem == 0) || (lowmem >= 0x100000))
 438                 lowmem = 0x9f000;
 439
 440         /* reserve all memory between lowmem and the 1MB mark */
 441         reserve_bootmem(lowmem, 0x100000 - lowmem, BOOTMEM_DEFAULT);
 442 }
 443
 444 #ifndef CONFIG_NEED_MULTIPLE_NODES
 445 static void __init setup_bootmem_allocator(void);
 446 static unsigned long __init setup_memory(void)
 447 {
 448         /*
 449          * partially used pages are not usable - thus
 450          * we are rounding upwards:
 451          */
 452         min_low_pfn = PFN_UP(init_pg_tables_end);
 453
 454         max_low_pfn = find_max_low_pfn();
 455
 456 #ifdef CONFIG_HIGHMEM
 457         highstart_pfn = highend_pfn = max_pfn;
 458         if (max_pfn > max_low_pfn) {
 459                 highstart_pfn = max_low_pfn;
 460         }
 461         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 462                 pages_to_mb(highend_pfn - highstart_pfn));
 463         num_physpages = highend_pfn;
 464         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 465 #else
 466         num_physpages = max_low_pfn;
 467         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 468 #endif
 469 #ifdef CONFIG_FLATMEM
 470         max_mapnr = num_physpages;
 471 #endif
 472         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 473                         pages_to_mb(max_low_pfn));
 474
 475         setup_bootmem_allocator();
 476
 477         return max_low_pfn;
 478 }
 479
 480 static void __init zone_sizes_init(void)
 481 {
 482         unsigned long max_zone_pfns[MAX_NR_ZONES];
 483         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 484         max_zone_pfns[ZONE_DMA] =
 485                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 486         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 487 #ifdef CONFIG_HIGHMEM
 488         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 489         add_active_range(0, 0, highend_pfn);
 490 #else
 491         add_active_range(0, 0, max_low_pfn);
 492 #endif
 493
 494         free_area_init_nodes(max_zone_pfns);
 495 }
 496 #else
 497 extern unsigned long __init setup_memory(void);
 498 extern void zone_sizes_init(void);
 499 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 500
 501 static inline unsigned long long get_total_mem(void)
 502 {
 503         unsigned long long total;
 504
 505         total = max_low_pfn - min_low_pfn;
 506 #ifdef CONFIG_HIGHMEM
 507         total += highend_pfn - highstart_pfn;
 508 #endif
 509
 510         return total << PAGE_SHIFT;
 511 }
 512
 513 #ifdef CONFIG_KEXEC
 514 static void __init reserve_crashkernel(void)
 515 {
 516         unsigned long long total_mem;
 517         unsigned long long crash_size, crash_base;
 518         int ret;
 519
 520         total_mem = get_total_mem();
 521
 522         ret = parse_crashkernel(boot_command_line, total_mem,
 523                         &crash_size, &crash_base);
 524         if (ret == 0 && crash_size > 0) {
 525                 if (crash_base > 0) {
 526                         printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 527                                         "for crashkernel (System RAM: %ldMB)\n",
 528                                         (unsigned long)(crash_size >> 20),
 529                                         (unsigned long)(crash_base >> 20),
 530                                         (unsigned long)(total_mem >> 20));
 531                         crashk_res.start = crash_base;
 532                         crashk_res.end   = crash_base + crash_size - 1;
 533                         reserve_bootmem(crash_base, crash_size,
 534                                         BOOTMEM_DEFAULT);
 535                 } else
 536                         printk(KERN_INFO "crashkernel reservation failed - "
 537                                         "you have to specify a base address\n");
 538         }
 539 }
 540 #else
 541 static inline void __init reserve_crashkernel(void)
 542 {}
 543 #endif
 544
 545 #ifdef CONFIG_BLK_DEV_INITRD
 546
 547 static bool do_relocate_initrd = false;
 548
 549 static void __init reserve_initrd(void)
 550 {
 551         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 552         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 553         unsigned long ramdisk_end   = ramdisk_image + ramdisk_size;
 554         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 555         unsigned long ramdisk_here;
 556
 557         initrd_start = 0;
 558
 559         if (!boot_params.hdr.type_of_loader ||
 560             !ramdisk_image || !ramdisk_size)
 561                 return;         /* No initrd provided by bootloader */
 562
 563         if (ramdisk_end < ramdisk_image) {
 564                 printk(KERN_ERR "initrd wraps around end of memory, "
 565                        "disabling initrd\n");
 566                 return;
 567         }
 568         if (ramdisk_size >= end_of_lowmem/2) {
 569                 printk(KERN_ERR "initrd too large to handle, "
 570                        "disabling initrd\n");
 571                 return;
 572         }
 573         if (ramdisk_end <= end_of_lowmem) {
 574                 /* All in lowmem, easy case */
 575                 reserve_bootmem(ramdisk_image, ramdisk_size, BOOTMEM_DEFAULT);
 576                 initrd_start = ramdisk_image + PAGE_OFFSET;
 577                 initrd_end = initrd_start+ramdisk_size;
 578                 return;
 579         }
 580
 581         /* We need to move the initrd down into lowmem */
 582         ramdisk_here = (end_of_lowmem - ramdisk_size) & PAGE_MASK;
 583
 584         /* Note: this includes all the lowmem currently occupied by
 585            the initrd, we rely on that fact to keep the data intact. */
 586         reserve_bootmem(ramdisk_here, ramdisk_size, BOOTMEM_DEFAULT);
 587         initrd_start = ramdisk_here + PAGE_OFFSET;
 588         initrd_end   = initrd_start + ramdisk_size;
 589
 590         do_relocate_initrd = true;
 591 }
 592
 593 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 594
 595 static void __init relocate_initrd(void)
 596 {
 597         unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 598         unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 599         unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 600         unsigned long ramdisk_here;
 601         unsigned long slop, clen, mapaddr;
 602         char *p, *q;
 603
 604         if (!do_relocate_initrd)
 605                 return;
 606
 607         ramdisk_here = initrd_start - PAGE_OFFSET;
 608
 609         q = (char *)initrd_start;
 610
 611         /* Copy any lowmem portion of the initrd */
 612         if (ramdisk_image < end_of_lowmem) {
 613                 clen = end_of_lowmem - ramdisk_image;
 614                 p = (char *)__va(ramdisk_image);
 615                 memcpy(q, p, clen);
 616                 q += clen;
 617                 ramdisk_image += clen;
 618                 ramdisk_size  -= clen;
 619         }
 620
 621         /* Copy the highmem portion of the initrd */
 622         while (ramdisk_size) {
 623                 slop = ramdisk_image & ~PAGE_MASK;
 624                 clen = ramdisk_size;
 625                 if (clen > MAX_MAP_CHUNK-slop)
 626                         clen = MAX_MAP_CHUNK-slop;
 627                 mapaddr = ramdisk_image & PAGE_MASK;
 628                 p = early_ioremap(mapaddr, clen+slop);
 629                 memcpy(q, p+slop, clen);
 630                 early_iounmap(p, clen+slop);
 631                 q += clen;
 632                 ramdisk_image += clen;
 633                 ramdisk_size  -= clen;
 634         }
 635 }
 636
 637 #endif /* CONFIG_BLK_DEV_INITRD */
 638
 639 void __init setup_bootmem_allocator(void)
 640 {
 641         unsigned long bootmap_size;
 642         /*
 643          * Initialize the boot-time allocator (with low memory only):
 644          */
 645         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
 646
 647         register_bootmem_low_pages(max_low_pfn);
 648
 649         /*
 650          * Reserve the bootmem bitmap itself as well. We do this in two
 651          * steps (first step was init_bootmem()) because this catches
 652          * the (very unlikely) case of us accidentally initializing the
 653          * bootmem allocator with an invalid RAM area.
 654          */
 655         reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
 656                          bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text),
 657                          BOOTMEM_DEFAULT);
 658
 659         /*
 660          * reserve physical page 0 - it's a special BIOS page on many boxes,
 661          * enabling clean reboots, SMP operation, laptop functions.
 662          */
 663         reserve_bootmem(0, PAGE_SIZE, BOOTMEM_DEFAULT);
 664
 665         /* reserve EBDA region */
 666         reserve_ebda_region();
 667
 668 #ifdef CONFIG_SMP
 669         /*
 670          * But first pinch a few for the stack/trampoline stuff
 671          * FIXME: Don't need the extra page at 4K, but need to fix
 672          * trampoline before removing it. (see the GDT stuff)
 673          */
 674         reserve_bootmem(PAGE_SIZE, PAGE_SIZE, BOOTMEM_DEFAULT);
 675 #endif
 676 #ifdef CONFIG_ACPI_SLEEP
 677         /*
 678          * Reserve low memory region for sleep support.
 679          */
 680         acpi_reserve_bootmem();
 681 #endif
 682 #ifdef CONFIG_X86_FIND_SMP_CONFIG
 683         /*
 684          * Find and reserve possible boot-time SMP configuration:
 685          */
 686         find_smp_config();
 687 #endif
 688 #ifdef CONFIG_BLK_DEV_INITRD
 689         reserve_initrd();
 690 #endif
 691         numa_kva_reserve();
 692         reserve_crashkernel();
 693
 694         reserve_ibft_region();
 695 }
 696
 697 /*
 698  * The node 0 pgdat is initialized before all of these because
 699  * it's needed for bootmem.  node>0 pgdats have their virtual
 700  * space allocated before the pagetables are in place to access
 701  * them, so they can't be cleared then.
 702  *
 703  * This should all compile down to nothing when NUMA is off.
 704  */
 705 static void __init remapped_pgdat_init(void)
 706 {
 707         int nid;
 708
 709         for_each_online_node(nid) {
 710                 if (nid != 0)
 711                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
 712         }
 713 }
 714
 715 #ifdef CONFIG_MCA
 716 static void set_mca_bus(int x)
 717 {
 718         MCA_bus = x;
 719 }
 720 #else
 721 static void set_mca_bus(int x) { }
 722 #endif
 723
 724 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
 725 char * __init __attribute__((weak)) memory_setup(void)
 726 {
 727         return machine_specific_memory_setup();
 728 }
 729
 730 #ifdef CONFIG_NUMA
 731 /*
 732  * In the golden day, when everything among i386 and x86_64 will be
 733  * integrated, this will not live here
 734  */
 735 void *x86_cpu_to_node_map_early_ptr;
 736 int x86_cpu_to_node_map_init[NR_CPUS] = {
 737         [0 ... NR_CPUS-1] = NUMA_NO_NODE
 738 };
 739 DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE;
 740 #endif
 741
 742 /*
 743  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 744  * passed the efi memmap, systab, etc., so we should use these data structures
 745  * for initialization.  Note, the efi init code path is determined by the
 746  * global efi_enabled. This allows the same kernel image to be used on existing
 747  * systems (with a traditional BIOS) as well as on EFI systems.
 748  */
 749 void __init setup_arch(char **cmdline_p)
 750 {
 751         unsigned long max_low_pfn;
 752
 753         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 754         pre_setup_arch_hook();
 755         early_cpu_init();
 756         early_ioremap_init();
 757
 758 #ifdef CONFIG_EFI
 759         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 760                      "EL32", 4))
 761                 efi_enabled = 1;
 762 #endif
 763
 764         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 765         screen_info = boot_params.screen_info;
 766         edid_info = boot_params.edid_info;
 767         apm_info.bios = boot_params.apm_bios_info;
 768         ist_info = boot_params.ist_info;
 769         saved_video_mode = boot_params.hdr.vid_mode;
 770         if( boot_params.sys_desc_table.length != 0 ) {
 771                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 772                 machine_id = boot_params.sys_desc_table.table[0];
 773                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 774                 BIOS_revision = boot_params.sys_desc_table.table[2];
 775         }
 776         bootloader_type = boot_params.hdr.type_of_loader;
 777
 778 #ifdef CONFIG_BLK_DEV_RAM
 779         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 780         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 781         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 782 #endif
 783         ARCH_SETUP
 784
 785         printk(KERN_INFO "BIOS-provided physical RAM map:\n");
 786         print_memory_map(memory_setup());
 787
 788         copy_edd();
 789
 790         if (!boot_params.hdr.root_flags)
 791                 root_mountflags &= ~MS_RDONLY;
 792         init_mm.start_code = (unsigned long) _text;
 793         init_mm.end_code = (unsigned long) _etext;
 794         init_mm.end_data = (unsigned long) _edata;
 795         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 796
 797         code_resource.start = virt_to_phys(_text);
 798         code_resource.end = virt_to_phys(_etext)-1;
 799         data_resource.start = virt_to_phys(_etext);
 800         data_resource.end = virt_to_phys(_edata)-1;
 801         bss_resource.start = virt_to_phys(&__bss_start);
 802         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 803
 804         parse_early_param();
 805
 806         if (user_defined_memmap) {
 807                 printk(KERN_INFO "user-defined physical RAM map:\n");
 808                 print_memory_map("user");
 809         }
 810
 811         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 812         *cmdline_p = command_line;
 813
 814         if (efi_enabled)
 815                 efi_init();
 816
 817         /* update e820 for memory not covered by WB MTRRs */
 818         propagate_e820_map();
 819         mtrr_bp_init();
 820         if (mtrr_trim_uncached_memory(max_pfn))
 821                 propagate_e820_map();
 822
 823         max_low_pfn = setup_memory();
 824
 825 #ifdef CONFIG_VMI
 826         /*
 827          * Must be after max_low_pfn is determined, and before kernel
 828          * pagetables are setup.
 829          */
 830         vmi_init();
 831 #endif
 832
 833         /*
 834          * NOTE: before this point _nobody_ is allowed to allocate
 835          * any memory using the bootmem allocator.  Although the
 836          * allocator is now initialised only the first 8Mb of the kernel
 837          * virtual address space has been mapped.  All allocations before
 838          * paging_init() has completed must use the alloc_bootmem_low_pages()
 839          * variant (which allocates DMA'able memory) and care must be taken
 840          * not to exceed the 8Mb limit.
 841          */
 842
 843 #ifdef CONFIG_SMP
 844         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
 845 #endif
 846         paging_init();
 847
 848         /*
 849          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 850          */
 851
 852 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 853         if (init_ohci1394_dma_early)
 854                 init_ohci1394_dma_on_all_controllers();
 855 #endif
 856
 857         remapped_pgdat_init();
 858         sparse_init();
 859         zone_sizes_init();
 860
 861         /*
 862          * NOTE: at this point the bootmem allocator is fully available.
 863          */
 864
 865 #ifdef CONFIG_BLK_DEV_INITRD
 866         relocate_initrd();
 867 #endif
 868
 869         paravirt_post_allocator_init();
 870
 871         dmi_scan_machine();
 872
 873         io_delay_init();
 874
 875 #ifdef CONFIG_X86_SMP
 876         /*
 877          * setup to use the early static init tables during kernel startup
 878          * X86_SMP will exclude sub-arches that don't deal well with it.
 879          */
 880         x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
 881         x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
 882 #ifdef CONFIG_NUMA
 883         x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
 884 #endif
 885 #endif
 886
 887 #ifdef CONFIG_X86_GENERICARCH
 888         generic_apic_probe();
 889 #endif
 890
 891 #ifdef CONFIG_ACPI
 892         /*
 893          * Parse the ACPI tables for possible boot-time SMP configuration.
 894          */
 895         acpi_boot_table_init();
 896 #endif
 897
 898         early_quirks();
 899
 900 #ifdef CONFIG_ACPI
 901         acpi_boot_init();
 902
 903 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
 904         if (def_to_bigsmp)
 905                 printk(KERN_WARNING "More than 8 CPUs detected and "
 906                         "CONFIG_X86_PC cannot handle it.\nUse "
 907                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
 908 #endif
 909 #endif
 910 #ifdef CONFIG_X86_LOCAL_APIC
 911         if (smp_found_config)
 912                 get_smp_config();
 913 #endif
 914
 915         e820_register_memory();
 916         e820_mark_nosave_regions();
 917
 918 #ifdef CONFIG_VT
 919 #if defined(CONFIG_VGA_CONSOLE)
 920         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 921                 conswitchp = &vga_con;
 922 #elif defined(CONFIG_DUMMY_CONSOLE)
 923         conswitchp = &dummy_con;
 924 #endif
 925 #endif
 926 }
 927
 928 /*
 929  * Request address space for all standard resources
 930  *
 931  * This is called just before pcibios_init(), which is also a
 932  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
 933  */
 934 static int __init request_standard_resources(void)
 935 {
 936         int i;
 937
 938         printk(KERN_INFO "Setting up standard PCI resources\n");
 939         init_iomem_resources(&code_resource, &data_resource, &bss_resource);
 940
 941         request_resource(&iomem_resource, &video_ram_resource);
 942
 943         /* request I/O space for devices used on all i[345]86 PCs */
 944         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 945                 request_resource(&ioport_resource, &standard_io_resources[i]);
 946         return 0;
 947 }
 948
 949 subsys_initcall(request_standard_resources);