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/sfi.h> 31#include <linux/apm_bios.h> 32#include <linux/initrd.h> 33#include <linux/bootmem.h> 34#include <linux/seq_file.h> 35#include <linux/console.h> 36#include <linux/mca.h> 37#include <linux/root_dev.h> 38#include <linux/highmem.h> 39#include <linux/module.h> 40#include <linux/efi.h> 41#include <linux/init.h> 42#include <linux/edd.h> 43#include <linux/iscsi_ibft.h> 44#include <linux/nodemask.h> 45#include <linux/kexec.h> 46#include <linux/dmi.h> 47#include <linux/pfn.h> 48#include <linux/pci.h> 49#include <asm/pci-direct.h> 50#include <linux/init_ohci1394_dma.h> 51#include <linux/kvm_para.h> 52 53#include <linux/errno.h> 54#include <linux/kernel.h> 55#include <linux/stddef.h> 56#include <linux/unistd.h> 57#include <linux/ptrace.h> 58#include <linux/user.h> 59#include <linux/delay.h> 60 61#include <linux/kallsyms.h> 62#include <linux/cpufreq.h> 63#include <linux/dma-mapping.h> 64#include <linux/ctype.h> 65#include <linux/uaccess.h> 66 67#include <linux/percpu.h> 68#include <linux/crash_dump.h> 69#include <linux/tboot.h> 70 71#include <video/edid.h> 72 73#include <asm/mtrr.h> 74#include <asm/apic.h> 75#include <asm/trampoline.h> 76#include <asm/e820.h> 77#include <asm/mpspec.h> 78#include <asm/setup.h> 79#include <asm/efi.h> 80#include <asm/timer.h> 81#include <asm/i8259.h> 82#include <asm/sections.h> 83#include <asm/dmi.h> 84#include <asm/io_apic.h> 85#include <asm/ist.h> 86#include <asm/vmi.h> 87#include <asm/setup_arch.h> 88#include <asm/bios_ebda.h> 89#include <asm/cacheflush.h> 90#include <asm/processor.h> 91#include <asm/bugs.h> 92 93#include <asm/system.h> 94#include <asm/vsyscall.h> 95#include <asm/cpu.h> 96#include <asm/desc.h> 97#include <asm/dma.h> 98#include <asm/iommu.h> 99#include <asm/gart.h> 100#include <asm/mmu_context.h> 101#include <asm/proto.h> 102 103#include <asm/paravirt.h> 104#include <asm/hypervisor.h> 105#include <asm/olpc_ofw.h> 106 107#include <asm/percpu.h> 108#include <asm/topology.h> 109#include <asm/apicdef.h> 110#include <asm/k8.h> 111#ifdef CONFIG_X86_64 112#include <asm/numa_64.h> 113#endif 114#include <asm/mce.h> 115 116/* 117 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries. 118 * The direct mapping extends to max_pfn_mapped, so that we can directly access 119 * apertures, ACPI and other tables without having to play with fixmaps. 120 */ 121unsigned long max_low_pfn_mapped; 122unsigned long max_pfn_mapped; 123 124#ifdef CONFIG_DMI 125RESERVE_BRK(dmi_alloc, 65536); 126#endif 127 128unsigned int boot_cpu_id __read_mostly; 129 130static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 131unsigned long _brk_end = (unsigned long)__brk_base; 132 133#ifdef CONFIG_X86_64 134int default_cpu_present_to_apicid(int mps_cpu) 135{ 136 return __default_cpu_present_to_apicid(mps_cpu); 137} 138 139int default_check_phys_apicid_present(int phys_apicid) 140{ 141 return __default_check_phys_apicid_present(phys_apicid); 142} 143#endif 144 145#ifndef CONFIG_DEBUG_BOOT_PARAMS 146struct boot_params __initdata boot_params; 147#else 148struct boot_params boot_params; 149#endif 150 151/* 152 * Machine setup.. 153 */ 154static struct resource data_resource = { 155 .name = "Kernel data", 156 .start = 0, 157 .end = 0, 158 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 159}; 160 161static struct resource code_resource = { 162 .name = "Kernel code", 163 .start = 0, 164 .end = 0, 165 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 166}; 167 168static struct resource bss_resource = { 169 .name = "Kernel bss", 170 .start = 0, 171 .end = 0, 172 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 173}; 174 175 176#ifdef CONFIG_X86_32 177/* cpu data as detected by the assembly code in head.S */ 178struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1}; 179/* common cpu data for all cpus */ 180struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1}; 181EXPORT_SYMBOL(boot_cpu_data); 182static void set_mca_bus(int x) 183{ 184#ifdef CONFIG_MCA 185 MCA_bus = x; 186#endif 187} 188 189unsigned int def_to_bigsmp; 190 191/* for MCA, but anyone else can use it if they want */ 192unsigned int machine_id; 193unsigned int machine_submodel_id; 194unsigned int BIOS_revision; 195 196struct apm_info apm_info; 197EXPORT_SYMBOL(apm_info); 198 199#if defined(CONFIG_X86_SPEEDSTEP_SMI) || defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 200struct ist_info ist_info; 201EXPORT_SYMBOL(ist_info); 202#else 203struct ist_info ist_info; 204#endif 205 206#else 207struct cpuinfo_x86 boot_cpu_data __read_mostly = { 208 .x86_phys_bits = MAX_PHYSMEM_BITS, 209}; 210EXPORT_SYMBOL(boot_cpu_data); 211#endif 212 213 214#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 215unsigned long mmu_cr4_features; 216#else 217unsigned long mmu_cr4_features = X86_CR4_PAE; 218#endif 219 220/* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 221int bootloader_type, bootloader_version; 222 223/* 224 * Setup options 225 */ 226struct screen_info screen_info; 227EXPORT_SYMBOL(screen_info); 228struct edid_info edid_info; 229EXPORT_SYMBOL_GPL(edid_info); 230 231extern int root_mountflags; 232 233unsigned long saved_video_mode; 234 235#define RAMDISK_IMAGE_START_MASK 0x07FF 236#define RAMDISK_PROMPT_FLAG 0x8000 237#define RAMDISK_LOAD_FLAG 0x4000 238 239static char __initdata command_line[COMMAND_LINE_SIZE]; 240#ifdef CONFIG_CMDLINE_BOOL 241static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 242#endif 243 244#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 245struct edd edd; 246#ifdef CONFIG_EDD_MODULE 247EXPORT_SYMBOL(edd); 248#endif 249/** 250 * copy_edd() - Copy the BIOS EDD information 251 * from boot_params into a safe place. 252 * 253 */ 254static inline void __init copy_edd(void) 255{ 256 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 257 sizeof(edd.mbr_signature)); 258 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 259 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 260 edd.edd_info_nr = boot_params.eddbuf_entries; 261} 262#else 263static inline void __init copy_edd(void) 264{ 265} 266#endif 267 268void * __init extend_brk(size_t size, size_t align) 269{ 270 size_t mask = align - 1; 271 void *ret; 272 273 BUG_ON(_brk_start == 0); 274 BUG_ON(align & mask); 275 276 _brk_end = (_brk_end + mask) & ~mask; 277 BUG_ON((char *)(_brk_end + size) > __brk_limit); 278 279 ret = (void *)_brk_end; 280 _brk_end += size; 281 282 memset(ret, 0, size); 283 284 return ret; 285} 286 287#ifdef CONFIG_X86_64 288static void __init init_gbpages(void) 289{ 290 if (direct_gbpages && cpu_has_gbpages) 291 printk(KERN_INFO "Using GB pages for direct mapping\n"); 292 else 293 direct_gbpages = 0; 294} 295#else 296static inline void init_gbpages(void) 297{ 298} 299#endif 300 301static void __init reserve_brk(void) 302{ 303 if (_brk_end > _brk_start) 304 reserve_early(__pa(_brk_start), __pa(_brk_end), "BRK"); 305 306 /* Mark brk area as locked down and no longer taking any 307 new allocations */ 308 _brk_start = 0; 309} 310 311#ifdef CONFIG_BLK_DEV_INITRD 312 313#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT) 314static void __init relocate_initrd(void) 315{ 316 /* Assume only end is not page aligned */ 317 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 318 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 319 u64 area_size = PAGE_ALIGN(ramdisk_size); 320 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT; 321 u64 ramdisk_here; 322 unsigned long slop, clen, mapaddr; 323 char *p, *q; 324 325 /* We need to move the initrd down into lowmem */ 326 ramdisk_here = find_e820_area(0, end_of_lowmem, area_size, 327 PAGE_SIZE); 328 329 if (ramdisk_here == -1ULL) 330 panic("Cannot find place for new RAMDISK of size %lld\n", 331 ramdisk_size); 332 333 /* Note: this includes all the lowmem currently occupied by 334 the initrd, we rely on that fact to keep the data intact. */ 335 reserve_early(ramdisk_here, ramdisk_here + area_size, 336 "NEW RAMDISK"); 337 initrd_start = ramdisk_here + PAGE_OFFSET; 338 initrd_end = initrd_start + ramdisk_size; 339 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n", 340 ramdisk_here, ramdisk_here + ramdisk_size); 341 342 q = (char *)initrd_start; 343 344 /* Copy any lowmem portion of the initrd */ 345 if (ramdisk_image < end_of_lowmem) { 346 clen = end_of_lowmem - ramdisk_image; 347 p = (char *)__va(ramdisk_image); 348 memcpy(q, p, clen); 349 q += clen; 350 ramdisk_image += clen; 351 ramdisk_size -= clen; 352 } 353 354 /* Copy the highmem portion of the initrd */ 355 while (ramdisk_size) { 356 slop = ramdisk_image & ~PAGE_MASK; 357 clen = ramdisk_size; 358 if (clen > MAX_MAP_CHUNK-slop) 359 clen = MAX_MAP_CHUNK-slop; 360 mapaddr = ramdisk_image & PAGE_MASK; 361 p = early_memremap(mapaddr, clen+slop); 362 memcpy(q, p+slop, clen); 363 early_iounmap(p, clen+slop); 364 q += clen; 365 ramdisk_image += clen; 366 ramdisk_size -= clen; 367 } 368 /* high pages is not converted by early_res_to_bootmem */ 369 ramdisk_image = boot_params.hdr.ramdisk_image; 370 ramdisk_size = boot_params.hdr.ramdisk_size; 371 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to" 372 " %08llx - %08llx\n", 373 ramdisk_image, ramdisk_image + ramdisk_size - 1, 374 ramdisk_here, ramdisk_here + ramdisk_size - 1); 375} 376 377static void __init reserve_initrd(void) 378{ 379 /* Assume only end is not page aligned */ 380 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 381 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 382 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 383 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT; 384 385 if (!boot_params.hdr.type_of_loader || 386 !ramdisk_image || !ramdisk_size) 387 return; /* No initrd provided by bootloader */ 388 389 initrd_start = 0; 390 391 if (ramdisk_size >= (end_of_lowmem>>1)) { 392 free_early(ramdisk_image, ramdisk_end); 393 printk(KERN_ERR "initrd too large to handle, " 394 "disabling initrd\n"); 395 return; 396 } 397 398 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image, 399 ramdisk_end); 400 401 402 if (ramdisk_end <= end_of_lowmem) { 403 /* All in lowmem, easy case */ 404 /* 405 * don't need to reserve again, already reserved early 406 * in i386_start_kernel 407 */ 408 initrd_start = ramdisk_image + PAGE_OFFSET; 409 initrd_end = initrd_start + ramdisk_size; 410 return; 411 } 412 413 relocate_initrd(); 414 415 free_early(ramdisk_image, ramdisk_end); 416} 417#else 418static void __init reserve_initrd(void) 419{ 420} 421#endif /* CONFIG_BLK_DEV_INITRD */ 422 423static void __init parse_setup_data(void) 424{ 425 struct setup_data *data; 426 u64 pa_data; 427 428 if (boot_params.hdr.version < 0x0209) 429 return; 430 pa_data = boot_params.hdr.setup_data; 431 while (pa_data) { 432 data = early_memremap(pa_data, PAGE_SIZE); 433 switch (data->type) { 434 case SETUP_E820_EXT: 435 parse_e820_ext(data, pa_data); 436 break; 437 default: 438 break; 439 } 440 pa_data = data->next; 441 early_iounmap(data, PAGE_SIZE); 442 } 443} 444 445static void __init e820_reserve_setup_data(void) 446{ 447 struct setup_data *data; 448 u64 pa_data; 449 int found = 0; 450 451 if (boot_params.hdr.version < 0x0209) 452 return; 453 pa_data = boot_params.hdr.setup_data; 454 while (pa_data) { 455 data = early_memremap(pa_data, sizeof(*data)); 456 e820_update_range(pa_data, sizeof(*data)+data->len, 457 E820_RAM, E820_RESERVED_KERN); 458 found = 1; 459 pa_data = data->next; 460 early_iounmap(data, sizeof(*data)); 461 } 462 if (!found) 463 return; 464 465 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 466 memcpy(&e820_saved, &e820, sizeof(struct e820map)); 467 printk(KERN_INFO "extended physical RAM map:\n"); 468 e820_print_map("reserve setup_data"); 469} 470 471static void __init reserve_early_setup_data(void) 472{ 473 struct setup_data *data; 474 u64 pa_data; 475 char buf[32]; 476 477 if (boot_params.hdr.version < 0x0209) 478 return; 479 pa_data = boot_params.hdr.setup_data; 480 while (pa_data) { 481 data = early_memremap(pa_data, sizeof(*data)); 482 sprintf(buf, "setup data %x", data->type); 483 reserve_early(pa_data, pa_data+sizeof(*data)+data->len, buf); 484 pa_data = data->next; 485 early_iounmap(data, sizeof(*data)); 486 } 487} 488 489/* 490 * --------- Crashkernel reservation ------------------------------ 491 */ 492 493#ifdef CONFIG_KEXEC 494 495static inline unsigned long long get_total_mem(void) 496{ 497 unsigned long long total; 498 499 total = max_pfn - min_low_pfn; 500 501 return total << PAGE_SHIFT; 502} 503 504static void __init reserve_crashkernel(void) 505{ 506 unsigned long long total_mem; 507 unsigned long long crash_size, crash_base; 508 int ret; 509 510 total_mem = get_total_mem(); 511 512 ret = parse_crashkernel(boot_command_line, total_mem, 513 &crash_size, &crash_base); 514 if (ret != 0 || crash_size <= 0) 515 return; 516 517 /* 0 means: find the address automatically */ 518 if (crash_base <= 0) { 519 const unsigned long long alignment = 16<<20; /* 16M */ 520 521 crash_base = find_e820_area(alignment, ULONG_MAX, crash_size, 522 alignment); 523 if (crash_base == -1ULL) { 524 pr_info("crashkernel reservation failed - No suitable area found.\n"); 525 return; 526 } 527 } else { 528 unsigned long long start; 529 530 start = find_e820_area(crash_base, ULONG_MAX, crash_size, 531 1<<20); 532 if (start != crash_base) { 533 pr_info("crashkernel reservation failed - memory is in use.\n"); 534 return; 535 } 536 } 537 reserve_early(crash_base, crash_base + crash_size, "CRASH KERNEL"); 538 539 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " 540 "for crashkernel (System RAM: %ldMB)\n", 541 (unsigned long)(crash_size >> 20), 542 (unsigned long)(crash_base >> 20), 543 (unsigned long)(total_mem >> 20)); 544 545 crashk_res.start = crash_base; 546 crashk_res.end = crash_base + crash_size - 1; 547 insert_resource(&iomem_resource, &crashk_res); 548} 549#else 550static void __init reserve_crashkernel(void) 551{ 552} 553#endif 554 555static struct resource standard_io_resources[] = { 556 { .name = "dma1", .start = 0x00, .end = 0x1f, 557 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 558 { .name = "pic1", .start = 0x20, .end = 0x21, 559 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 560 { .name = "timer0", .start = 0x40, .end = 0x43, 561 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 562 { .name = "timer1", .start = 0x50, .end = 0x53, 563 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 564 { .name = "keyboard", .start = 0x60, .end = 0x60, 565 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 566 { .name = "keyboard", .start = 0x64, .end = 0x64, 567 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 568 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 569 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 570 { .name = "pic2", .start = 0xa0, .end = 0xa1, 571 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 572 { .name = "dma2", .start = 0xc0, .end = 0xdf, 573 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 574 { .name = "fpu", .start = 0xf0, .end = 0xff, 575 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 576}; 577 578void __init reserve_standard_io_resources(void) 579{ 580 int i; 581 582 /* request I/O space for devices used on all i[345]86 PCs */ 583 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 584 request_resource(&ioport_resource, &standard_io_resources[i]); 585 586} 587 588/* 589 * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by 590 * is_kdump_kernel() to determine if we are booting after a panic. Hence 591 * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE. 592 */ 593 594#ifdef CONFIG_CRASH_DUMP 595/* elfcorehdr= specifies the location of elf core header 596 * stored by the crashed kernel. This option will be passed 597 * by kexec loader to the capture kernel. 598 */ 599static int __init setup_elfcorehdr(char *arg) 600{ 601 char *end; 602 if (!arg) 603 return -EINVAL; 604 elfcorehdr_addr = memparse(arg, &end); 605 return end > arg ? 0 : -EINVAL; 606} 607early_param("elfcorehdr", setup_elfcorehdr); 608#endif 609 610static __init void reserve_ibft_region(void) 611{ 612 unsigned long addr, size = 0; 613 614 addr = find_ibft_region(&size); 615 616 if (size) 617 reserve_early_overlap_ok(addr, addr + size, "ibft"); 618} 619 620#ifdef CONFIG_X86_RESERVE_LOW_64K 621static int __init dmi_low_memory_corruption(const struct dmi_system_id *d) 622{ 623 printk(KERN_NOTICE 624 "%s detected: BIOS may corrupt low RAM, working around it.\n", 625 d->ident); 626 627 e820_update_range(0, 0x10000, E820_RAM, E820_RESERVED); 628 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 629 630 return 0; 631} 632#endif 633 634/* List of systems that have known low memory corruption BIOS problems */ 635static struct dmi_system_id __initdata bad_bios_dmi_table[] = { 636#ifdef CONFIG_X86_RESERVE_LOW_64K 637 { 638 .callback = dmi_low_memory_corruption, 639 .ident = "AMI BIOS", 640 .matches = { 641 DMI_MATCH(DMI_BIOS_VENDOR, "American Megatrends Inc."), 642 }, 643 }, 644 { 645 .callback = dmi_low_memory_corruption, 646 .ident = "Phoenix BIOS", 647 .matches = { 648 DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix Technologies"), 649 }, 650 }, 651 { 652 .callback = dmi_low_memory_corruption, 653 .ident = "Phoenix/MSC BIOS", 654 .matches = { 655 DMI_MATCH(DMI_BIOS_VENDOR, "Phoenix/MSC"), 656 }, 657 }, 658 /* 659 * AMI BIOS with low memory corruption was found on Intel DG45ID and 660 * DG45FC boards. 661 * It has a different DMI_BIOS_VENDOR = "Intel Corp.", for now we will 662 * match only DMI_BOARD_NAME and see if there is more bad products 663 * with this vendor. 664 */ 665 { 666 .callback = dmi_low_memory_corruption, 667 .ident = "AMI BIOS", 668 .matches = { 669 DMI_MATCH(DMI_BOARD_NAME, "DG45ID"), 670 }, 671 }, 672 { 673 .callback = dmi_low_memory_corruption, 674 .ident = "AMI BIOS", 675 .matches = { 676 DMI_MATCH(DMI_BOARD_NAME, "DG45FC"), 677 }, 678 }, 679 /* 680 * The Dell Inspiron Mini 1012 has DMI_BIOS_VENDOR = "Dell Inc.", so 681 * match on the product name. 682 */ 683 { 684 .callback = dmi_low_memory_corruption, 685 .ident = "Phoenix BIOS", 686 .matches = { 687 DMI_MATCH(DMI_PRODUCT_NAME, "Inspiron 1012"), 688 }, 689 }, 690#endif 691 {} 692}; 693 694static void __init trim_bios_range(void) 695{ 696 /* 697 * A special case is the first 4Kb of memory; 698 * This is a BIOS owned area, not kernel ram, but generally 699 * not listed as such in the E820 table. 700 */ 701 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 702 /* 703 * special case: Some BIOSen report the PC BIOS 704 * area (640->1Mb) as ram even though it is not. 705 * take them out. 706 */ 707 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 708 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 709} 710 711/* 712 * Determine if we were loaded by an EFI loader. If so, then we have also been 713 * passed the efi memmap, systab, etc., so we should use these data structures 714 * for initialization. Note, the efi init code path is determined by the 715 * global efi_enabled. This allows the same kernel image to be used on existing 716 * systems (with a traditional BIOS) as well as on EFI systems. 717 */ 718/* 719 * setup_arch - architecture-specific boot-time initializations 720 * 721 * Note: On x86_64, fixmaps are ready for use even before this is called. 722 */ 723 724void __init setup_arch(char **cmdline_p) 725{ 726 int acpi = 0; 727 int k8 = 0; 728 729#ifdef CONFIG_X86_32 730 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 731 visws_early_detect(); 732#else 733 printk(KERN_INFO "Command line: %s\n", boot_command_line); 734#endif 735 736 /* VMI may relocate the fixmap; do this before touching ioremap area */ 737 vmi_init(); 738 739 /* OFW also may relocate the fixmap */ 740 olpc_ofw_detect(); 741 742 early_trap_init(); 743 early_cpu_init(); 744 early_ioremap_init(); 745 746 setup_olpc_ofw_pgd(); 747 748 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 749 screen_info = boot_params.screen_info; 750 edid_info = boot_params.edid_info; 751#ifdef CONFIG_X86_32 752 apm_info.bios = boot_params.apm_bios_info; 753 ist_info = boot_params.ist_info; 754 if (boot_params.sys_desc_table.length != 0) { 755 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2); 756 machine_id = boot_params.sys_desc_table.table[0]; 757 machine_submodel_id = boot_params.sys_desc_table.table[1]; 758 BIOS_revision = boot_params.sys_desc_table.table[2]; 759 } 760#endif 761 saved_video_mode = boot_params.hdr.vid_mode; 762 bootloader_type = boot_params.hdr.type_of_loader; 763 if ((bootloader_type >> 4) == 0xe) { 764 bootloader_type &= 0xf; 765 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 766 } 767 bootloader_version = bootloader_type & 0xf; 768 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 769 770#ifdef CONFIG_BLK_DEV_RAM 771 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 772 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 773 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 774#endif 775#ifdef CONFIG_EFI 776 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 777#ifdef CONFIG_X86_32 778 "EL32", 779#else 780 "EL64", 781#endif 782 4)) { 783 efi_enabled = 1; 784 efi_reserve_early(); 785 } 786#endif 787 788 x86_init.oem.arch_setup(); 789 790 setup_memory_map(); 791 parse_setup_data(); 792 /* update the e820_saved too */ 793 e820_reserve_setup_data(); 794 795 copy_edd(); 796 797 if (!boot_params.hdr.root_flags) 798 root_mountflags &= ~MS_RDONLY; 799 init_mm.start_code = (unsigned long) _text; 800 init_mm.end_code = (unsigned long) _etext; 801 init_mm.end_data = (unsigned long) _edata; 802 init_mm.brk = _brk_end; 803 804 code_resource.start = virt_to_phys(_text); 805 code_resource.end = virt_to_phys(_etext)-1; 806 data_resource.start = virt_to_phys(_etext); 807 data_resource.end = virt_to_phys(_edata)-1; 808 bss_resource.start = virt_to_phys(&__bss_start); 809 bss_resource.end = virt_to_phys(&__bss_stop)-1; 810 811#ifdef CONFIG_CMDLINE_BOOL 812#ifdef CONFIG_CMDLINE_OVERRIDE 813 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 814#else 815 if (builtin_cmdline[0]) { 816 /* append boot loader cmdline to builtin */ 817 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 818 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 819 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 820 } 821#endif 822#endif 823 824 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 825 *cmdline_p = command_line; 826 827 /* 828 * x86_configure_nx() is called before parse_early_param() to detect 829 * whether hardware doesn't support NX (so that the early EHCI debug 830 * console setup can safely call set_fixmap()). It may then be called 831 * again from within noexec_setup() during parsing early parameters 832 * to honor the respective command line option. 833 */ 834 x86_configure_nx(); 835 836 parse_early_param(); 837 838 x86_report_nx(); 839 840 /* Must be before kernel pagetables are setup */ 841 vmi_activate(); 842 843 /* after early param, so could get panic from serial */ 844 reserve_early_setup_data(); 845 846 if (acpi_mps_check()) { 847#ifdef CONFIG_X86_LOCAL_APIC 848 disable_apic = 1; 849#endif 850 setup_clear_cpu_cap(X86_FEATURE_APIC); 851 } 852 853#ifdef CONFIG_PCI 854 if (pci_early_dump_regs) 855 early_dump_pci_devices(); 856#endif 857 858 finish_e820_parsing(); 859 860 if (efi_enabled) 861 efi_init(); 862 863 dmi_scan_machine(); 864 865 dmi_check_system(bad_bios_dmi_table); 866 867 /* 868 * VMware detection requires dmi to be available, so this 869 * needs to be done after dmi_scan_machine, for the BP. 870 */ 871 init_hypervisor_platform(); 872 873 x86_init.resources.probe_roms(); 874 875 /* after parse_early_param, so could debug it */ 876 insert_resource(&iomem_resource, &code_resource); 877 insert_resource(&iomem_resource, &data_resource); 878 insert_resource(&iomem_resource, &bss_resource); 879 880 trim_bios_range(); 881#ifdef CONFIG_X86_32 882 if (ppro_with_ram_bug()) { 883 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 884 E820_RESERVED); 885 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 886 printk(KERN_INFO "fixed physical RAM map:\n"); 887 e820_print_map("bad_ppro"); 888 } 889#else 890 early_gart_iommu_check(); 891#endif 892 893 /* 894 * partially used pages are not usable - thus 895 * we are rounding upwards: 896 */ 897 max_pfn = e820_end_of_ram_pfn(); 898 899 /* preallocate 4k for mptable mpc */ 900 early_reserve_e820_mpc_new(); 901 /* update e820 for memory not covered by WB MTRRs */ 902 mtrr_bp_init(); 903 if (mtrr_trim_uncached_memory(max_pfn)) 904 max_pfn = e820_end_of_ram_pfn(); 905 906#ifdef CONFIG_X86_32 907 /* max_low_pfn get updated here */ 908 find_low_pfn_range(); 909#else 910 num_physpages = max_pfn; 911 912 check_x2apic(); 913 914 /* How many end-of-memory variables you have, grandma! */ 915 /* need this before calling reserve_initrd */ 916 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 917 max_low_pfn = e820_end_of_low_ram_pfn(); 918 else 919 max_low_pfn = max_pfn; 920 921 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 922 max_pfn_mapped = KERNEL_IMAGE_SIZE >> PAGE_SHIFT; 923#endif 924 925#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 926 setup_bios_corruption_check(); 927#endif 928 929 printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n", 930 max_pfn_mapped<<PAGE_SHIFT); 931 932 reserve_brk(); 933 934 /* 935 * Find and reserve possible boot-time SMP configuration: 936 */ 937 find_smp_config(); 938 939 reserve_ibft_region(); 940 941 reserve_trampoline_memory(); 942 943#ifdef CONFIG_ACPI_SLEEP 944 /* 945 * Reserve low memory region for sleep support. 946 * even before init_memory_mapping 947 */ 948 acpi_reserve_wakeup_memory(); 949#endif 950 init_gbpages(); 951 952 /* max_pfn_mapped is updated here */ 953 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT); 954 max_pfn_mapped = max_low_pfn_mapped; 955 956#ifdef CONFIG_X86_64 957 if (max_pfn > max_low_pfn) { 958 max_pfn_mapped = init_memory_mapping(1UL<<32, 959 max_pfn<<PAGE_SHIFT); 960 /* can we preseve max_low_pfn ?*/ 961 max_low_pfn = max_pfn; 962 } 963#endif 964 965 /* 966 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 967 */ 968 969#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 970 if (init_ohci1394_dma_early) 971 init_ohci1394_dma_on_all_controllers(); 972#endif 973 974 reserve_initrd(); 975 976 reserve_crashkernel(); 977 978 vsmp_init(); 979 980 io_delay_init(); 981 982 /* 983 * Parse the ACPI tables for possible boot-time SMP configuration. 984 */ 985 acpi_boot_table_init(); 986 987 early_acpi_boot_init(); 988 989#ifdef CONFIG_ACPI_NUMA 990 /* 991 * Parse SRAT to discover nodes. 992 */ 993 acpi = acpi_numa_init(); 994#endif 995 996#ifdef CONFIG_K8_NUMA 997 if (!acpi) 998 k8 = !k8_numa_init(0, max_pfn); 999#endif 1000 1001 initmem_init(0, max_pfn, acpi, k8); 1002#ifndef CONFIG_NO_BOOTMEM 1003 early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT); 1004#endif 1005 1006 dma32_reserve_bootmem(); 1007 1008#ifdef CONFIG_KVM_CLOCK 1009 kvmclock_init(); 1010#endif 1011 1012 x86_init.paging.pagetable_setup_start(swapper_pg_dir); 1013 paging_init(); 1014 x86_init.paging.pagetable_setup_done(swapper_pg_dir); 1015 1016 setup_trampoline_page_table(); 1017 1018 tboot_probe(); 1019 1020#ifdef CONFIG_X86_64 1021 map_vsyscall(); 1022#endif 1023 1024 generic_apic_probe(); 1025 1026 early_quirks(); 1027 1028 /* 1029 * Read APIC and some other early information from ACPI tables. 1030 */ 1031 acpi_boot_init(); 1032 1033 sfi_init(); 1034 1035 /* 1036 * get boot-time SMP configuration: 1037 */ 1038 if (smp_found_config) 1039 get_smp_config(); 1040 1041 prefill_possible_map(); 1042 1043#ifdef CONFIG_X86_64 1044 init_cpu_to_node(); 1045#endif 1046 1047 init_apic_mappings(); 1048 ioapic_init_mappings(); 1049 1050 /* need to wait for io_apic is mapped */ 1051 probe_nr_irqs_gsi(); 1052 1053 kvm_guest_init(); 1054 1055 e820_reserve_resources(); 1056 e820_mark_nosave_regions(max_low_pfn); 1057 1058 x86_init.resources.reserve_resources(); 1059 1060 e820_setup_gap(); 1061 1062#ifdef CONFIG_VT 1063#if defined(CONFIG_VGA_CONSOLE) 1064 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1065 conswitchp = &vga_con; 1066#elif defined(CONFIG_DUMMY_CONSOLE) 1067 conswitchp = &dummy_con; 1068#endif 1069#endif 1070 x86_init.oem.banner(); 1071 1072 mcheck_init(); 1073} 1074 1075#ifdef CONFIG_X86_32 1076 1077static struct resource video_ram_resource = { 1078 .name = "Video RAM area", 1079 .start = 0xa0000, 1080 .end = 0xbffff, 1081 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1082}; 1083 1084void __init i386_reserve_resources(void) 1085{ 1086 request_resource(&iomem_resource, &video_ram_resource); 1087 reserve_standard_io_resources(); 1088} 1089 1090#endif /* CONFIG_X86_32 */ 1091