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/dmi.h>
  63 #include <asm/io_apic.h>
  64 #include <asm/ist.h>
  65 #include <asm/io.h>
  66 #include <asm/vmi.h>
  67 #include <setup_arch.h>
  68 #include <asm/bios_ebda.h>
  69 #include <asm/cacheflush.h>
  70 #include <asm/processor.h>
  71 #include <asm/efi.h>
  72 #include <asm/bugs.h>
  73
  74 /* This value is set up by the early boot code to point to the value
  75    immediately after the boot time page tables.  It contains a *physical*
  76    address, and must not be in the .bss segment! */
  77 unsigned long init_pg_tables_start __initdata = ~0UL;
  78 unsigned long init_pg_tables_end __initdata = ~0UL;
  79
  80 /*
  81  * Machine setup..
  82  */
  83 static struct resource data_resource = {
  84         .name   = "Kernel data",
  85         .start  = 0,
  86         .end    = 0,
  87         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  88 };
  89
  90 static struct resource code_resource = {
  91         .name   = "Kernel code",
  92         .start  = 0,
  93         .end    = 0,
  94         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
  95 };
  96
  97 static struct resource bss_resource = {
  98         .name   = "Kernel bss",
  99         .start  = 0,
 100         .end    = 0,
 101         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 102 };
 103
 104 static struct resource video_ram_resource = {
 105         .name   = "Video RAM area",
 106         .start  = 0xa0000,
 107         .end    = 0xbffff,
 108         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
 109 };
 110
 111 static struct resource standard_io_resources[] = { {
 112         .name   = "dma1",
 113         .start  = 0x0000,
 114         .end    = 0x001f,
 115         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 116 }, {
 117         .name   = "pic1",
 118         .start  = 0x0020,
 119         .end    = 0x0021,
 120         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 121 }, {
 122         .name   = "timer0",
 123         .start  = 0x0040,
 124         .end    = 0x0043,
 125         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 126 }, {
 127         .name   = "timer1",
 128         .start  = 0x0050,
 129         .end    = 0x0053,
 130         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 131 }, {
 132         .name   = "keyboard",
 133         .start  = 0x0060,
 134         .end    = 0x0060,
 135         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 136 }, {
 137         .name   = "keyboard",
 138         .start  = 0x0064,
 139         .end    = 0x0064,
 140         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 141 }, {
 142         .name   = "dma page reg",
 143         .start  = 0x0080,
 144         .end    = 0x008f,
 145         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 146 }, {
 147         .name   = "pic2",
 148         .start  = 0x00a0,
 149         .end    = 0x00a1,
 150         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 151 }, {
 152         .name   = "dma2",
 153         .start  = 0x00c0,
 154         .end    = 0x00df,
 155         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 156 }, {
 157         .name   = "fpu",
 158         .start  = 0x00f0,
 159         .end    = 0x00ff,
 160         .flags  = IORESOURCE_BUSY | IORESOURCE_IO
 161 } };
 162
 163 /* cpu data as detected by the assembly code in head.S */
 164 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 165 /* common cpu data for all cpus */
 166 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 167 EXPORT_SYMBOL(boot_cpu_data);
 168
 169 unsigned int def_to_bigsmp;
 170
 171 #ifndef CONFIG_X86_PAE
 172 unsigned long mmu_cr4_features;
 173 #else
 174 unsigned long mmu_cr4_features = X86_CR4_PAE;
 175 #endif
 176
 177 /* for MCA, but anyone else can use it if they want */
 178 unsigned int machine_id;
 179 unsigned int machine_submodel_id;
 180 unsigned int BIOS_revision;
 181
 182 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
 183 int bootloader_type;
 184
 185 /* user-defined highmem size */
 186 static unsigned int highmem_pages = -1;
 187
 188 /*
 189  * Early DMI memory
 190  */
 191 int dmi_alloc_index;
 192 char dmi_alloc_data[DMI_MAX_DATA];
 193
 194 /*
 195  * Setup options
 196  */
 197 struct screen_info screen_info;
 198 EXPORT_SYMBOL(screen_info);
 199 struct apm_info apm_info;
 200 EXPORT_SYMBOL(apm_info);
 201 struct edid_info edid_info;
 202 EXPORT_SYMBOL_GPL(edid_info);
 203 struct ist_info ist_info;
 204 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
 205         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
 206 EXPORT_SYMBOL(ist_info);
 207 #endif
 208
 209 extern void early_cpu_init(void);
 210 extern int root_mountflags;
 211
 212 unsigned long saved_video_mode;
 213
 214 #define RAMDISK_IMAGE_START_MASK        0x07FF
 215 #define RAMDISK_PROMPT_FLAG             0x8000
 216 #define RAMDISK_LOAD_FLAG               0x4000
 217
 218 static char __initdata command_line[COMMAND_LINE_SIZE];
 219
 220 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 221 struct boot_params __initdata boot_params;
 222 #else
 223 struct boot_params boot_params;
 224 #endif
 225
 226 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 227 struct edd edd;
 228 #ifdef CONFIG_EDD_MODULE
 229 EXPORT_SYMBOL(edd);
 230 #endif
 231 /**
 232  * copy_edd() - Copy the BIOS EDD information
 233  *              from boot_params into a safe place.
 234  *
 235  */
 236 static inline void copy_edd(void)
 237 {
 238      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 239             sizeof(edd.mbr_signature));
 240      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 241      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 242      edd.edd_info_nr = boot_params.eddbuf_entries;
 243 }
 244 #else
 245 static inline void copy_edd(void)
 246 {
 247 }
 248 #endif
 249
 250 #ifdef CONFIG_PROC_VMCORE
 251 /* elfcorehdr= specifies the location of elf core header
 252  * stored by the crashed kernel.
 253  */
 254 static int __init parse_elfcorehdr(char *arg)
 255 {
 256         if (!arg)
 257                 return -EINVAL;
 258
 259         elfcorehdr_addr = memparse(arg, &arg);
 260         return 0;
 261 }
 262 early_param("elfcorehdr", parse_elfcorehdr);
 263 #endif /* CONFIG_PROC_VMCORE */
 264
 265 /*
 266  * highmem=size forces highmem to be exactly 'size' bytes.
 267  * This works even on boxes that have no highmem otherwise.
 268  * This also works to reduce highmem size on bigger boxes.
 269  */
 270 static int __init parse_highmem(char *arg)
 271 {
 272         if (!arg)
 273                 return -EINVAL;
 274
 275         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 276         return 0;
 277 }
 278 early_param("highmem", parse_highmem);
 279
 280 /*
 281  * vmalloc=size forces the vmalloc area to be exactly 'size'
 282  * bytes. This can be used to increase (or decrease) the
 283  * vmalloc area - the default is 128m.
 284  */
 285 static int __init parse_vmalloc(char *arg)
 286 {
 287         if (!arg)
 288                 return -EINVAL;
 289
 290         __VMALLOC_RESERVE = memparse(arg, &arg);
 291         return 0;
 292 }
 293 early_param("vmalloc", parse_vmalloc);
 294
 295 /*
 296  * reservetop=size reserves a hole at the top of the kernel address space which
 297  * a hypervisor can load into later.  Needed for dynamically loaded hypervisors,
 298  * so relocating the fixmap can be done before paging initialization.
 299  */
 300 static int __init parse_reservetop(char *arg)
 301 {
 302         unsigned long address;
 303
 304         if (!arg)
 305                 return -EINVAL;
 306
 307         address = memparse(arg, &arg);
 308         reserve_top_address(address);
 309         return 0;
 310 }
 311 early_param("reservetop", parse_reservetop);
 312
 313 /*
 314  * Determine low and high memory ranges:
 315  */
 316 unsigned long __init find_max_low_pfn(void)
 317 {
 318         unsigned long max_low_pfn;
 319
 320         max_low_pfn = max_pfn;
 321         if (max_low_pfn > MAXMEM_PFN) {
 322                 if (highmem_pages == -1)
 323                         highmem_pages = max_pfn - MAXMEM_PFN;
 324                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 325                         max_pfn = MAXMEM_PFN + highmem_pages;
 326                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 327                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
 328                         highmem_pages = 0;
 329                 }
 330                 max_low_pfn = MAXMEM_PFN;
 331 #ifndef CONFIG_HIGHMEM
 332                 /* Maximum memory usable is what is directly addressable */
 333                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 334                                         MAXMEM>>20);
 335                 if (max_pfn > MAX_NONPAE_PFN)
 336                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 337                 else
 338                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 339                 max_pfn = MAXMEM_PFN;
 340 #else /* !CONFIG_HIGHMEM */
 341 #ifndef CONFIG_HIGHMEM64G
 342                 if (max_pfn > MAX_NONPAE_PFN) {
 343                         max_pfn = MAX_NONPAE_PFN;
 344                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
 345                         printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 346                 }
 347 #endif /* !CONFIG_HIGHMEM64G */
 348 #endif /* !CONFIG_HIGHMEM */
 349         } else {
 350                 if (highmem_pages == -1)
 351                         highmem_pages = 0;
 352 #ifdef CONFIG_HIGHMEM
 353                 if (highmem_pages >= max_pfn) {
 354                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 355                         highmem_pages = 0;
 356                 }
 357                 if (highmem_pages) {
 358                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
 359                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
 360                                 highmem_pages = 0;
 361                         }
 362                         max_low_pfn -= highmem_pages;
 363                 }
 364 #else
 365                 if (highmem_pages)
 366                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 367 #endif
 368         }
 369         return max_low_pfn;
 370 }
 371
 372 #ifndef CONFIG_NEED_MULTIPLE_NODES
 373 static void __init setup_bootmem_allocator(void);
 374 static unsigned long __init setup_memory(void)
 375 {
 376         /*
 377          * partially used pages are not usable - thus
 378          * we are rounding upwards:
 379          */
 380         min_low_pfn = PFN_UP(init_pg_tables_end);
 381
 382         max_low_pfn = find_max_low_pfn();
 383
 384 #ifdef CONFIG_HIGHMEM
 385         highstart_pfn = highend_pfn = max_pfn;
 386         if (max_pfn > max_low_pfn) {
 387                 highstart_pfn = max_low_pfn;
 388         }
 389         memory_present(0, 0, highend_pfn);
 390         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 391                 pages_to_mb(highend_pfn - highstart_pfn));
 392         num_physpages = highend_pfn;
 393         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 394 #else
 395         memory_present(0, 0, max_low_pfn);
 396         num_physpages = max_low_pfn;
 397         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 398 #endif
 399 #ifdef CONFIG_FLATMEM
 400         max_mapnr = num_physpages;
 401 #endif
 402         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 403                         pages_to_mb(max_low_pfn));
 404
 405         setup_bootmem_allocator();
 406
 407         return max_low_pfn;
 408 }
 409
 410 static void __init zone_sizes_init(void)
 411 {
 412         unsigned long max_zone_pfns[MAX_NR_ZONES];
 413         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 414         max_zone_pfns[ZONE_DMA] =
 415                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 416         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 417         remove_all_active_ranges();
 418 #ifdef CONFIG_HIGHMEM
 419         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 420         e820_register_active_regions(0, 0, highend_pfn);
 421 #else
 422         e820_register_active_regions(0, 0, max_low_pfn);
 423 #endif
 424
 425         free_area_init_nodes(max_zone_pfns);
 426 }
 427 #else
 428 extern unsigned long __init setup_memory(void);
 429 extern void zone_sizes_init(void);
 430 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 431
 432 #ifdef CONFIG_BLK_DEV_INITRD
 433
 434 static bool do_relocate_initrd = false;
 435
 436 static void __init reserve_initrd(void)
 437 {
 438         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 439         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 440         u64 ramdisk_end   = ramdisk_image + ramdisk_size;
 441         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 442         u64 ramdisk_here;
 443
 444         if (!boot_params.hdr.type_of_loader ||
 445             !ramdisk_image || !ramdisk_size)
 446                 return;         /* No initrd provided by bootloader */
 447
 448         initrd_start = 0;
 449
 450         if (ramdisk_size >= end_of_lowmem/2) {
 451                 free_early(ramdisk_image, ramdisk_end);
 452                 printk(KERN_ERR "initrd too large to handle, "
 453                        "disabling initrd\n");
 454                 return;
 455         }
 456
 457         printk(KERN_INFO "old RAMDISK: %08llx - %08llx\n", ramdisk_image,
 458                         ramdisk_end);
 459
 460
 461         if (ramdisk_end <= end_of_lowmem) {
 462                 /* All in lowmem, easy case */
 463                 /*
 464                  * don't need to reserve again, already reserved early
 465                  * in i386_start_kernel
 466                  */
 467                 initrd_start = ramdisk_image + PAGE_OFFSET;
 468                 initrd_end = initrd_start+ramdisk_size;
 469                 return;
 470         }
 471
 472         /* We need to move the initrd down into lowmem */
 473         ramdisk_here = find_e820_area(min_low_pfn<<PAGE_SHIFT,
 474                                  end_of_lowmem, ramdisk_size,
 475                                  PAGE_SIZE);
 476
 477         if (ramdisk_here == -1ULL)
 478                 panic("Cannot find place for new RAMDISK of size %lld\n",
 479                          ramdisk_size);
 480
 481         /* Note: this includes all the lowmem currently occupied by
 482            the initrd, we rely on that fact to keep the data intact. */
 483         reserve_early(ramdisk_here, ramdisk_here + ramdisk_size,
 484                          "NEW RAMDISK");
 485         initrd_start = ramdisk_here + PAGE_OFFSET;
 486         initrd_end   = initrd_start + ramdisk_size;
 487         printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
 488                          ramdisk_here, ramdisk_here + ramdisk_size);
 489
 490         do_relocate_initrd = true;
 491 }
 492
 493 #define MAX_MAP_CHUNK   (NR_FIX_BTMAPS << PAGE_SHIFT)
 494
 495 static void __init relocate_initrd(void)
 496 {
 497         u64 ramdisk_image = boot_params.hdr.ramdisk_image;
 498         u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
 499         u64 end_of_lowmem = max_low_pfn << PAGE_SHIFT;
 500         u64 ramdisk_here;
 501         unsigned long slop, clen, mapaddr;
 502         char *p, *q;
 503
 504         if (!do_relocate_initrd)
 505                 return;
 506
 507         ramdisk_here = initrd_start - PAGE_OFFSET;
 508
 509         q = (char *)initrd_start;
 510
 511         /* Copy any lowmem portion of the initrd */
 512         if (ramdisk_image < end_of_lowmem) {
 513                 clen = end_of_lowmem - ramdisk_image;
 514                 p = (char *)__va(ramdisk_image);
 515                 memcpy(q, p, clen);
 516                 q += clen;
 517                 /* need to free these low pages...*/
 518                 printk(KERN_INFO "Freeing old partial RAMDISK %08llx-%08llx\n",
 519                          ramdisk_image, ramdisk_image + clen - 1);
 520                 free_bootmem(ramdisk_image, clen);
 521                 ramdisk_image += clen;
 522                 ramdisk_size  -= clen;
 523         }
 524
 525         /* Copy the highmem portion of the initrd */
 526         while (ramdisk_size) {
 527                 slop = ramdisk_image & ~PAGE_MASK;
 528                 clen = ramdisk_size;
 529                 if (clen > MAX_MAP_CHUNK-slop)
 530                         clen = MAX_MAP_CHUNK-slop;
 531                 mapaddr = ramdisk_image & PAGE_MASK;
 532                 p = early_ioremap(mapaddr, clen+slop);
 533                 memcpy(q, p+slop, clen);
 534                 early_iounmap(p, clen+slop);
 535                 q += clen;
 536                 ramdisk_image += clen;
 537                 ramdisk_size  -= clen;
 538         }
 539         /* high pages is not converted by early_res_to_bootmem */
 540         ramdisk_image = boot_params.hdr.ramdisk_image;
 541         ramdisk_size  = boot_params.hdr.ramdisk_size;
 542         printk(KERN_INFO "Copied RAMDISK from %016llx - %016llx to %08llx - %08llx\n",
 543                 ramdisk_image, ramdisk_image + ramdisk_size - 1,
 544                 ramdisk_here, ramdisk_here + ramdisk_size - 1);
 545
 546         /* need to free that, otherwise init highmem will reserve it again */
 547         free_early(ramdisk_image, ramdisk_image+ramdisk_size);
 548 }
 549
 550 #endif /* CONFIG_BLK_DEV_INITRD */
 551
 552 void __init setup_bootmem_allocator(void)
 553 {
 554         int i;
 555         unsigned long bootmap_size, bootmap;
 556         /*
 557          * Initialize the boot-time allocator (with low memory only):
 558          */
 559         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 560         bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
 561                                  max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 562                                  PAGE_SIZE);
 563         if (bootmap == -1L)
 564                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 565         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 566 #ifdef CONFIG_BLK_DEV_INITRD
 567         reserve_initrd();
 568 #endif
 569         bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, max_low_pfn);
 570         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 571                  max_pfn_mapped<<PAGE_SHIFT);
 572         printk(KERN_INFO "  low ram: %08lx - %08lx\n",
 573                  min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
 574         printk(KERN_INFO "  bootmap %08lx - %08lx\n",
 575                  bootmap, bootmap + bootmap_size);
 576         for_each_online_node(i)
 577                 free_bootmem_with_active_regions(i, max_low_pfn);
 578         early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
 579
 580 }
 581
 582 /*
 583  * The node 0 pgdat is initialized before all of these because
 584  * it's needed for bootmem.  node>0 pgdats have their virtual
 585  * space allocated before the pagetables are in place to access
 586  * them, so they can't be cleared then.
 587  *
 588  * This should all compile down to nothing when NUMA is off.
 589  */
 590 static void __init remapped_pgdat_init(void)
 591 {
 592         int nid;
 593
 594         for_each_online_node(nid) {
 595                 if (nid != 0)
 596                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
 597         }
 598 }
 599
 600 #ifdef CONFIG_MCA
 601 static void set_mca_bus(int x)
 602 {
 603         MCA_bus = x;
 604 }
 605 #else
 606 static void set_mca_bus(int x) { }
 607 #endif
 608
 609 static void probe_roms(void);
 610
 611 /*
 612  * Determine if we were loaded by an EFI loader.  If so, then we have also been
 613  * passed the efi memmap, systab, etc., so we should use these data structures
 614  * for initialization.  Note, the efi init code path is determined by the
 615  * global efi_enabled. This allows the same kernel image to be used on existing
 616  * systems (with a traditional BIOS) as well as on EFI systems.
 617  */
 618 void __init setup_arch(char **cmdline_p)
 619 {
 620         int i;
 621         unsigned long max_low_pfn;
 622
 623         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
 624         pre_setup_arch_hook();
 625         early_cpu_init();
 626         early_ioremap_init();
 627         reserve_setup_data();
 628
 629 #ifdef CONFIG_EFI
 630         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 631                      "EL32", 4)) {
 632                 efi_enabled = 1;
 633                 efi_reserve_early();
 634         }
 635 #endif
 636
 637         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 638         screen_info = boot_params.screen_info;
 639         edid_info = boot_params.edid_info;
 640         apm_info.bios = boot_params.apm_bios_info;
 641         ist_info = boot_params.ist_info;
 642         saved_video_mode = boot_params.hdr.vid_mode;
 643         if( boot_params.sys_desc_table.length != 0 ) {
 644                 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
 645                 machine_id = boot_params.sys_desc_table.table[0];
 646                 machine_submodel_id = boot_params.sys_desc_table.table[1];
 647                 BIOS_revision = boot_params.sys_desc_table.table[2];
 648         }
 649         bootloader_type = boot_params.hdr.type_of_loader;
 650
 651 #ifdef CONFIG_BLK_DEV_RAM
 652         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 653         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 654         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 655 #endif
 656         ARCH_SETUP
 657
 658         setup_memory_map();
 659
 660         copy_edd();
 661
 662         if (!boot_params.hdr.root_flags)
 663                 root_mountflags &= ~MS_RDONLY;
 664         init_mm.start_code = (unsigned long) _text;
 665         init_mm.end_code = (unsigned long) _etext;
 666         init_mm.end_data = (unsigned long) _edata;
 667         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
 668
 669         code_resource.start = virt_to_phys(_text);
 670         code_resource.end = virt_to_phys(_etext)-1;
 671         data_resource.start = virt_to_phys(_etext);
 672         data_resource.end = virt_to_phys(_edata)-1;
 673         bss_resource.start = virt_to_phys(&__bss_start);
 674         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 675
 676         parse_setup_data();
 677
 678         parse_early_param();
 679
 680         if (acpi_mps_check()){
 681                 enable_local_apic = -1;
 682                 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
 683         }
 684
 685         finish_e820_parsing();
 686
 687         probe_roms();
 688
 689         /* after parse_early_param, so could debug it */
 690         insert_resource(&iomem_resource, &code_resource);
 691         insert_resource(&iomem_resource, &data_resource);
 692         insert_resource(&iomem_resource, &bss_resource);
 693
 694         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 695         *cmdline_p = command_line;
 696
 697         if (efi_enabled)
 698                 efi_init();
 699
 700         if (ppro_with_ram_bug()) {
 701                 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
 702                                   E820_RESERVED);
 703                 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
 704                 printk(KERN_INFO "fixed physical RAM map:\n");
 705                 e820_print_map("bad_ppro");
 706         }
 707
 708         e820_register_active_regions(0, 0, -1UL);
 709         /*
 710          * partially used pages are not usable - thus
 711          * we are rounding upwards:
 712          */
 713         max_pfn = e820_end_of_ram();
 714
 715         /* preallocate 4k for mptable mpc */
 716         early_reserve_e820_mpc_new();
 717         /* update e820 for memory not covered by WB MTRRs */
 718         mtrr_bp_init();
 719         if (mtrr_trim_uncached_memory(max_pfn)) {
 720                 remove_all_active_ranges();
 721                 e820_register_active_regions(0, 0, -1UL);
 722                 max_pfn = e820_end_of_ram();
 723         }
 724
 725         dmi_scan_machine();
 726
 727         io_delay_init();
 728
 729         /*
 730          * Parse the ACPI tables for possible boot-time SMP configuration.
 731          */
 732         acpi_boot_table_init();
 733
 734 #ifdef CONFIG_ACPI_NUMA
 735         /*
 736          * Parse SRAT to discover nodes.
 737          */
 738         acpi_numa_init();
 739 #endif
 740
 741         max_low_pfn = setup_memory();
 742
 743 #ifdef CONFIG_ACPI_SLEEP
 744         /*
 745          * Reserve low memory region for sleep support.
 746          */
 747         acpi_reserve_bootmem();
 748 #endif
 749 #ifdef CONFIG_X86_FIND_SMP_CONFIG
 750         /*
 751          * Find and reserve possible boot-time SMP configuration:
 752          */
 753         find_smp_config();
 754 #endif
 755         reserve_crashkernel();
 756
 757         reserve_ibft_region();
 758
 759 #ifdef CONFIG_KVM_CLOCK
 760         kvmclock_init();
 761 #endif
 762
 763 #ifdef CONFIG_VMI
 764         /*
 765          * Must be after max_low_pfn is determined, and before kernel
 766          * pagetables are setup.
 767          */
 768         vmi_init();
 769 #endif
 770         kvm_guest_init();
 771
 772         /*
 773          * NOTE: before this point _nobody_ is allowed to allocate
 774          * any memory using the bootmem allocator.  Although the
 775          * allocator is now initialised only the first 8Mb of the kernel
 776          * virtual address space has been mapped.  All allocations before
 777          * paging_init() has completed must use the alloc_bootmem_low_pages()
 778          * variant (which allocates DMA'able memory) and care must be taken
 779          * not to exceed the 8Mb limit.
 780          */
 781
 782         paging_init();
 783
 784         /*
 785          * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
 786          */
 787
 788 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 789         if (init_ohci1394_dma_early)
 790                 init_ohci1394_dma_on_all_controllers();
 791 #endif
 792
 793         /*
 794          * NOTE: at this point the bootmem allocator is fully available.
 795          */
 796
 797 #ifdef CONFIG_BLK_DEV_INITRD
 798         relocate_initrd();
 799 #endif
 800
 801         remapped_pgdat_init();
 802         sparse_init();
 803         zone_sizes_init();
 804
 805         paravirt_post_allocator_init();
 806
 807 #ifdef CONFIG_X86_GENERICARCH
 808         generic_apic_probe();
 809 #endif
 810
 811         early_quirks();
 812
 813         acpi_boot_init();
 814
 815 #if defined(CONFIG_X86_MPPARSE) || defined(CONFIG_X86_VISWS)
 816         if (smp_found_config)
 817                 get_smp_config();
 818 #endif
 819 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
 820         if (def_to_bigsmp)
 821                 printk(KERN_WARNING "More than 8 CPUs detected and "
 822                         "CONFIG_X86_PC cannot handle it.\nUse "
 823                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
 824 #endif
 825
 826         e820_reserve_resources();
 827         e820_mark_nosave_regions(max_low_pfn);
 828
 829         request_resource(&iomem_resource, &video_ram_resource);
 830         /* request I/O space for devices used on all i[345]86 PCs */
 831         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 832                 request_resource(&ioport_resource, &standard_io_resources[i]);
 833
 834         e820_setup_gap();
 835
 836 #ifdef CONFIG_VT
 837 #if defined(CONFIG_VGA_CONSOLE)
 838         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 839                 conswitchp = &vga_con;
 840 #elif defined(CONFIG_DUMMY_CONSOLE)
 841         conswitchp = &dummy_con;
 842 #endif
 843 #endif
 844 }
 845
 846 static struct resource system_rom_resource = {
 847         .name   = "System ROM",
 848         .start  = 0xf0000,
 849         .end    = 0xfffff,
 850         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 851 };
 852
 853 static struct resource extension_rom_resource = {
 854         .name   = "Extension ROM",
 855         .start  = 0xe0000,
 856         .end    = 0xeffff,
 857         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 858 };
 859
 860 static struct resource adapter_rom_resources[] = { {
 861         .name   = "Adapter ROM",
 862         .start  = 0xc8000,
 863         .end    = 0,
 864         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 865 }, {
 866         .name   = "Adapter ROM",
 867         .start  = 0,
 868         .end    = 0,
 869         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 870 }, {
 871         .name   = "Adapter ROM",
 872         .start  = 0,
 873         .end    = 0,
 874         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 875 }, {
 876         .name   = "Adapter ROM",
 877         .start  = 0,
 878         .end    = 0,
 879         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 880 }, {
 881         .name   = "Adapter ROM",
 882         .start  = 0,
 883         .end    = 0,
 884         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 885 }, {
 886         .name   = "Adapter ROM",
 887         .start  = 0,
 888         .end    = 0,
 889         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 890 } };
 891
 892 static struct resource video_rom_resource = {
 893         .name   = "Video ROM",
 894         .start  = 0xc0000,
 895         .end    = 0xc7fff,
 896         .flags  = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
 897 };
 898
 899 #define ROMSIGNATURE 0xaa55
 900
 901 static int __init romsignature(const unsigned char *rom)
 902 {
 903         const unsigned short * const ptr = (const unsigned short *)rom;
 904         unsigned short sig;
 905
 906         return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
 907 }
 908
 909 static int __init romchecksum(const unsigned char *rom, unsigned long length)
 910 {
 911         unsigned char sum, c;
 912
 913         for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
 914                 sum += c;
 915         return !length && !sum;
 916 }
 917
 918 static void __init probe_roms(void)
 919 {
 920         const unsigned char *rom;
 921         unsigned long start, length, upper;
 922         unsigned char c;
 923         int i;
 924
 925         /* video rom */
 926         upper = adapter_rom_resources[0].start;
 927         for (start = video_rom_resource.start; start < upper; start += 2048) {
 928                 rom = isa_bus_to_virt(start);
 929                 if (!romsignature(rom))
 930                         continue;
 931
 932                 video_rom_resource.start = start;
 933
 934                 if (probe_kernel_address(rom + 2, c) != 0)
 935                         continue;
 936
 937                 /* 0 < length <= 0x7f * 512, historically */
 938                 length = c * 512;
 939
 940                 /* if checksum okay, trust length byte */
 941                 if (length && romchecksum(rom, length))
 942                         video_rom_resource.end = start + length - 1;
 943
 944                 request_resource(&iomem_resource, &video_rom_resource);
 945                 break;
 946         }
 947
 948         start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
 949         if (start < upper)
 950                 start = upper;
 951
 952         /* system rom */
 953         request_resource(&iomem_resource, &system_rom_resource);
 954         upper = system_rom_resource.start;
 955
 956         /* check for extension rom (ignore length byte!) */
 957         rom = isa_bus_to_virt(extension_rom_resource.start);
 958         if (romsignature(rom)) {
 959                 length = extension_rom_resource.end - extension_rom_resource.start + 1;
 960                 if (romchecksum(rom, length)) {
 961                         request_resource(&iomem_resource, &extension_rom_resource);
 962                         upper = extension_rom_resource.start;
 963                 }
 964         }
 965
 966         /* check for adapter roms on 2k boundaries */
 967         for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
 968                 rom = isa_bus_to_virt(start);
 969                 if (!romsignature(rom))
 970                         continue;
 971
 972                 if (probe_kernel_address(rom + 2, c) != 0)
 973                         continue;
 974
 975                 /* 0 < length <= 0x7f * 512, historically */
 976                 length = c * 512;
 977
 978                 /* but accept any length that fits if checksum okay */
 979                 if (!length || start + length > upper || !romchecksum(rom, length))
 980                         continue;
 981
 982                 adapter_rom_resources[i].start = start;
 983                 adapter_rom_resources[i].end = start + length - 1;
 984                 request_resource(&iomem_resource, &adapter_rom_resources[i]);
 985
 986                 start = adapter_rom_resources[i++].end & ~2047UL;
 987         }
 988 }
 989