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/memblock.h> 35 #include <linux/seq_file.h> 36 #include <linux/console.h> 37 #include <linux/mca.h> 38 #include <linux/root_dev.h> 39 #include <linux/highmem.h> 40 #include <linux/module.h> 41 #include <linux/efi.h> 42 #include <linux/init.h> 43 #include <linux/edd.h> 44 #include <linux/iscsi_ibft.h> 45 #include <linux/nodemask.h> 46 #include <linux/kexec.h> 47 #include <linux/dmi.h> 48 #include <linux/pfn.h> 49 #include <linux/pci.h> 50 #include <asm/pci-direct.h> 51 #include <linux/init_ohci1394_dma.h> 52 #include <linux/kvm_para.h> 53 54 #include <linux/errno.h> 55 #include <linux/kernel.h> 56 #include <linux/stddef.h> 57 #include <linux/unistd.h> 58 #include <linux/ptrace.h> 59 #include <linux/user.h> 60 #include <linux/delay.h> 61 62 #include <linux/kallsyms.h> 63 #include <linux/cpufreq.h> 64 #include <linux/dma-mapping.h> 65 #include <linux/ctype.h> 66 #include <linux/uaccess.h> 67 68 #include <linux/percpu.h> 69 #include <linux/crash_dump.h> 70 #include <linux/tboot.h> 71 72 #include <video/edid.h> 73 74 #include <asm/mtrr.h> 75 #include <asm/apic.h> 76 #include <asm/trampoline.h> 77 #include <asm/e820.h> 78 #include <asm/mpspec.h> 79 #include <asm/setup.h> 80 #include <asm/efi.h> 81 #include <asm/timer.h> 82 #include <asm/i8259.h> 83 #include <asm/sections.h> 84 #include <asm/dmi.h> 85 #include <asm/io_apic.h> 86 #include <asm/ist.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/amd_nb.h> 111 #ifdef CONFIG_X86_64 112 #include <asm/numa_64.h> 113 #endif 114 #include <asm/mce.h> 115 #include <asm/alternative.h> 116 117 /* 118 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries. 119 * The direct mapping extends to max_pfn_mapped, so that we can directly access 120 * apertures, ACPI and other tables without having to play with fixmaps. 121 */ 122 unsigned long max_low_pfn_mapped; 123 unsigned long max_pfn_mapped; 124 125 #ifdef CONFIG_DMI 126 RESERVE_BRK(dmi_alloc, 65536); 127 #endif 128 129 130 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 131 unsigned long _brk_end = (unsigned long)__brk_base; 132 133 #ifdef CONFIG_X86_64 134 int default_cpu_present_to_apicid(int mps_cpu) 135 { 136 return __default_cpu_present_to_apicid(mps_cpu); 137 } 138 139 int 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 146 struct boot_params __initdata boot_params; 147 #else 148 struct boot_params boot_params; 149 #endif 150 151 /* 152 * Machine setup.. 153 */ 154 static struct resource data_resource = { 155 .name = "Kernel data", 156 .start = 0, 157 .end = 0, 158 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 159 }; 160 161 static struct resource code_resource = { 162 .name = "Kernel code", 163 .start = 0, 164 .end = 0, 165 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 166 }; 167 168 static 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 */ 178 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1}; 179 /* common cpu data for all cpus */ 180 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1}; 181 EXPORT_SYMBOL(boot_cpu_data); 182 static void set_mca_bus(int x) 183 { 184 #ifdef CONFIG_MCA 185 MCA_bus = x; 186 #endif 187 } 188 189 unsigned int def_to_bigsmp; 190 191 /* for MCA, but anyone else can use it if they want */ 192 unsigned int machine_id; 193 unsigned int machine_submodel_id; 194 unsigned int BIOS_revision; 195 196 struct apm_info apm_info; 197 EXPORT_SYMBOL(apm_info); 198 199 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 200 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 201 struct ist_info ist_info; 202 EXPORT_SYMBOL(ist_info); 203 #else 204 struct ist_info ist_info; 205 #endif 206 207 #else 208 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 209 .x86_phys_bits = MAX_PHYSMEM_BITS, 210 }; 211 EXPORT_SYMBOL(boot_cpu_data); 212 #endif 213 214 215 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 216 unsigned long mmu_cr4_features; 217 #else 218 unsigned long mmu_cr4_features = X86_CR4_PAE; 219 #endif 220 221 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 222 int bootloader_type, bootloader_version; 223 224 /* 225 * Setup options 226 */ 227 struct screen_info screen_info; 228 EXPORT_SYMBOL(screen_info); 229 struct edid_info edid_info; 230 EXPORT_SYMBOL_GPL(edid_info); 231 232 extern int root_mountflags; 233 234 unsigned long saved_video_mode; 235 236 #define RAMDISK_IMAGE_START_MASK 0x07FF 237 #define RAMDISK_PROMPT_FLAG 0x8000 238 #define RAMDISK_LOAD_FLAG 0x4000 239 240 static char __initdata command_line[COMMAND_LINE_SIZE]; 241 #ifdef CONFIG_CMDLINE_BOOL 242 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 243 #endif 244 245 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 246 struct edd edd; 247 #ifdef CONFIG_EDD_MODULE 248 EXPORT_SYMBOL(edd); 249 #endif 250 /** 251 * copy_edd() - Copy the BIOS EDD information 252 * from boot_params into a safe place. 253 * 254 */ 255 static inline void __init copy_edd(void) 256 { 257 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 258 sizeof(edd.mbr_signature)); 259 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 260 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 261 edd.edd_info_nr = boot_params.eddbuf_entries; 262 } 263 #else 264 static inline void __init copy_edd(void) 265 { 266 } 267 #endif 268 269 void * __init extend_brk(size_t size, size_t align) 270 { 271 size_t mask = align - 1; 272 void *ret; 273 274 BUG_ON(_brk_start == 0); 275 BUG_ON(align & mask); 276 277 _brk_end = (_brk_end + mask) & ~mask; 278 BUG_ON((char *)(_brk_end + size) > __brk_limit); 279 280 ret = (void *)_brk_end; 281 _brk_end += size; 282 283 memset(ret, 0, size); 284 285 return ret; 286 } 287 288 #ifdef CONFIG_X86_64 289 static void __init init_gbpages(void) 290 { 291 if (direct_gbpages && cpu_has_gbpages) 292 printk(KERN_INFO "Using GB pages for direct mapping\n"); 293 else 294 direct_gbpages = 0; 295 } 296 #else 297 static inline void init_gbpages(void) 298 { 299 } 300 #endif 301 302 static void __init reserve_brk(void) 303 { 304 if (_brk_end > _brk_start) 305 memblock_x86_reserve_range(__pa(_brk_start), __pa(_brk_end), "BRK"); 306 307 /* Mark brk area as locked down and no longer taking any 308 new allocations */ 309 _brk_start = 0; 310 } 311 312 #ifdef CONFIG_BLK_DEV_INITRD 313 314 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT) 315 static void __init relocate_initrd(void) 316 { 317 /* Assume only end is not page aligned */ 318 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 319 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 320 u64 area_size = PAGE_ALIGN(ramdisk_size); 321 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT; 322 u64 ramdisk_here; 323 unsigned long slop, clen, mapaddr; 324 char *p, *q; 325 326 /* We need to move the initrd down into lowmem */ 327 ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size, 328 PAGE_SIZE); 329 330 if (ramdisk_here == MEMBLOCK_ERROR) 331 panic("Cannot find place for new RAMDISK of size %lld\n", 332 ramdisk_size); 333 334 /* Note: this includes all the lowmem currently occupied by 335 the initrd, we rely on that fact to keep the data intact. */ 336 memblock_x86_reserve_range(ramdisk_here, ramdisk_here + area_size, "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 } 376 377 static 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 memblock_x86_free_range(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 memblock_x86_free_range(ramdisk_image, ramdisk_end); 416 } 417 #else 418 static void __init reserve_initrd(void) 419 { 420 } 421 #endif /* CONFIG_BLK_DEV_INITRD */ 422 423 static 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 } 444 445 static 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 } 470 471 static void __init memblock_x86_reserve_range_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 memblock_x86_reserve_range(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 494 495 static 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 } 503 504 /* 505 * Keep the crash kernel below this limit. On 32 bits earlier kernels 506 * would limit the kernel to the low 512 MiB due to mapping restrictions. 507 * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this 508 * limit once kexec-tools are fixed. 509 */ 510 #ifdef CONFIG_X86_32 511 # define CRASH_KERNEL_ADDR_MAX (512 << 20) 512 #else 513 # define CRASH_KERNEL_ADDR_MAX (896 << 20) 514 #endif 515 516 static void __init reserve_crashkernel(void) 517 { 518 unsigned long long total_mem; 519 unsigned long long crash_size, crash_base; 520 int ret; 521 522 total_mem = get_total_mem(); 523 524 ret = parse_crashkernel(boot_command_line, total_mem, 525 &crash_size, &crash_base); 526 if (ret != 0 || crash_size <= 0) 527 return; 528 529 /* 0 means: find the address automatically */ 530 if (crash_base <= 0) { 531 const unsigned long long alignment = 16<<20; /* 16M */ 532 533 /* 534 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 535 */ 536 crash_base = memblock_find_in_range(alignment, 537 CRASH_KERNEL_ADDR_MAX, crash_size, alignment); 538 539 if (crash_base == MEMBLOCK_ERROR) { 540 pr_info("crashkernel reservation failed - No suitable area found.\n"); 541 return; 542 } 543 } else { 544 unsigned long long start; 545 546 start = memblock_find_in_range(crash_base, 547 crash_base + crash_size, crash_size, 1<<20); 548 if (start != crash_base) { 549 pr_info("crashkernel reservation failed - memory is in use.\n"); 550 return; 551 } 552 } 553 memblock_x86_reserve_range(crash_base, crash_base + crash_size, "CRASH KERNEL"); 554 555 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " 556 "for crashkernel (System RAM: %ldMB)\n", 557 (unsigned long)(crash_size >> 20), 558 (unsigned long)(crash_base >> 20), 559 (unsigned long)(total_mem >> 20)); 560 561 crashk_res.start = crash_base; 562 crashk_res.end = crash_base + crash_size - 1; 563 insert_resource(&iomem_resource, &crashk_res); 564 } 565 #else 566 static void __init reserve_crashkernel(void) 567 { 568 } 569 #endif 570 571 static struct resource standard_io_resources[] = { 572 { .name = "dma1", .start = 0x00, .end = 0x1f, 573 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 574 { .name = "pic1", .start = 0x20, .end = 0x21, 575 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 576 { .name = "timer0", .start = 0x40, .end = 0x43, 577 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 578 { .name = "timer1", .start = 0x50, .end = 0x53, 579 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 580 { .name = "keyboard", .start = 0x60, .end = 0x60, 581 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 582 { .name = "keyboard", .start = 0x64, .end = 0x64, 583 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 584 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 585 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 586 { .name = "pic2", .start = 0xa0, .end = 0xa1, 587 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 588 { .name = "dma2", .start = 0xc0, .end = 0xdf, 589 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 590 { .name = "fpu", .start = 0xf0, .end = 0xff, 591 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 592 }; 593 594 void __init reserve_standard_io_resources(void) 595 { 596 int i; 597 598 /* request I/O space for devices used on all i[345]86 PCs */ 599 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 600 request_resource(&ioport_resource, &standard_io_resources[i]); 601 602 } 603 604 /* 605 * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by 606 * is_kdump_kernel() to determine if we are booting after a panic. Hence 607 * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE. 608 */ 609 610 #ifdef CONFIG_CRASH_DUMP 611 /* elfcorehdr= specifies the location of elf core header 612 * stored by the crashed kernel. This option will be passed 613 * by kexec loader to the capture kernel. 614 */ 615 static int __init setup_elfcorehdr(char *arg) 616 { 617 char *end; 618 if (!arg) 619 return -EINVAL; 620 elfcorehdr_addr = memparse(arg, &end); 621 return end > arg ? 0 : -EINVAL; 622 } 623 early_param("elfcorehdr", setup_elfcorehdr); 624 #endif 625 626 static __init void reserve_ibft_region(void) 627 { 628 unsigned long addr, size = 0; 629 630 addr = find_ibft_region(&size); 631 632 if (size) 633 memblock_x86_reserve_range(addr, addr + size, "* ibft"); 634 } 635 636 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 637 638 static void __init trim_bios_range(void) 639 { 640 /* 641 * A special case is the first 4Kb of memory; 642 * This is a BIOS owned area, not kernel ram, but generally 643 * not listed as such in the E820 table. 644 * 645 * This typically reserves additional memory (64KiB by default) 646 * since some BIOSes are known to corrupt low memory. See the 647 * Kconfig help text for X86_RESERVE_LOW. 648 */ 649 e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE), 650 E820_RAM, E820_RESERVED); 651 652 /* 653 * special case: Some BIOSen report the PC BIOS 654 * area (640->1Mb) as ram even though it is not. 655 * take them out. 656 */ 657 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 658 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 659 } 660 661 static int __init parse_reservelow(char *p) 662 { 663 unsigned long long size; 664 665 if (!p) 666 return -EINVAL; 667 668 size = memparse(p, &p); 669 670 if (size < 4096) 671 size = 4096; 672 673 if (size > 640*1024) 674 size = 640*1024; 675 676 reserve_low = size; 677 678 return 0; 679 } 680 681 early_param("reservelow", parse_reservelow); 682 683 static u64 __init get_max_mapped(void) 684 { 685 u64 end = max_pfn_mapped; 686 687 end <<= PAGE_SHIFT; 688 689 return end; 690 } 691 692 /* 693 * Determine if we were loaded by an EFI loader. If so, then we have also been 694 * passed the efi memmap, systab, etc., so we should use these data structures 695 * for initialization. Note, the efi init code path is determined by the 696 * global efi_enabled. This allows the same kernel image to be used on existing 697 * systems (with a traditional BIOS) as well as on EFI systems. 698 */ 699 /* 700 * setup_arch - architecture-specific boot-time initializations 701 * 702 * Note: On x86_64, fixmaps are ready for use even before this is called. 703 */ 704 705 void __init setup_arch(char **cmdline_p) 706 { 707 int acpi = 0; 708 int amd = 0; 709 unsigned long flags; 710 711 #ifdef CONFIG_X86_32 712 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 713 visws_early_detect(); 714 715 /* 716 * copy kernel address range established so far and switch 717 * to the proper swapper page table 718 */ 719 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 720 initial_page_table + KERNEL_PGD_BOUNDARY, 721 KERNEL_PGD_PTRS); 722 723 load_cr3(swapper_pg_dir); 724 __flush_tlb_all(); 725 #else 726 printk(KERN_INFO "Command line: %s\n", boot_command_line); 727 #endif 728 729 /* 730 * If we have OLPC OFW, we might end up relocating the fixmap due to 731 * reserve_top(), so do this before touching the ioremap area. 732 */ 733 olpc_ofw_detect(); 734 735 early_trap_init(); 736 early_cpu_init(); 737 early_ioremap_init(); 738 739 setup_olpc_ofw_pgd(); 740 741 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 742 screen_info = boot_params.screen_info; 743 edid_info = boot_params.edid_info; 744 #ifdef CONFIG_X86_32 745 apm_info.bios = boot_params.apm_bios_info; 746 ist_info = boot_params.ist_info; 747 if (boot_params.sys_desc_table.length != 0) { 748 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2); 749 machine_id = boot_params.sys_desc_table.table[0]; 750 machine_submodel_id = boot_params.sys_desc_table.table[1]; 751 BIOS_revision = boot_params.sys_desc_table.table[2]; 752 } 753 #endif 754 saved_video_mode = boot_params.hdr.vid_mode; 755 bootloader_type = boot_params.hdr.type_of_loader; 756 if ((bootloader_type >> 4) == 0xe) { 757 bootloader_type &= 0xf; 758 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 759 } 760 bootloader_version = bootloader_type & 0xf; 761 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 762 763 #ifdef CONFIG_BLK_DEV_RAM 764 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 765 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 766 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 767 #endif 768 #ifdef CONFIG_EFI 769 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 770 #ifdef CONFIG_X86_32 771 "EL32", 772 #else 773 "EL64", 774 #endif 775 4)) { 776 efi_enabled = 1; 777 efi_memblock_x86_reserve_range(); 778 } 779 #endif 780 781 x86_init.oem.arch_setup(); 782 783 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 784 setup_memory_map(); 785 parse_setup_data(); 786 /* update the e820_saved too */ 787 e820_reserve_setup_data(); 788 789 copy_edd(); 790 791 if (!boot_params.hdr.root_flags) 792 root_mountflags &= ~MS_RDONLY; 793 init_mm.start_code = (unsigned long) _text; 794 init_mm.end_code = (unsigned long) _etext; 795 init_mm.end_data = (unsigned long) _edata; 796 init_mm.brk = _brk_end; 797 798 code_resource.start = virt_to_phys(_text); 799 code_resource.end = virt_to_phys(_etext)-1; 800 data_resource.start = virt_to_phys(_etext); 801 data_resource.end = virt_to_phys(_edata)-1; 802 bss_resource.start = virt_to_phys(&__bss_start); 803 bss_resource.end = virt_to_phys(&__bss_stop)-1; 804 805 #ifdef CONFIG_CMDLINE_BOOL 806 #ifdef CONFIG_CMDLINE_OVERRIDE 807 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 808 #else 809 if (builtin_cmdline[0]) { 810 /* append boot loader cmdline to builtin */ 811 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 812 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 813 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 814 } 815 #endif 816 #endif 817 818 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 819 *cmdline_p = command_line; 820 821 /* 822 * x86_configure_nx() is called before parse_early_param() to detect 823 * whether hardware doesn't support NX (so that the early EHCI debug 824 * console setup can safely call set_fixmap()). It may then be called 825 * again from within noexec_setup() during parsing early parameters 826 * to honor the respective command line option. 827 */ 828 x86_configure_nx(); 829 830 parse_early_param(); 831 832 x86_report_nx(); 833 834 /* after early param, so could get panic from serial */ 835 memblock_x86_reserve_range_setup_data(); 836 837 if (acpi_mps_check()) { 838 #ifdef CONFIG_X86_LOCAL_APIC 839 disable_apic = 1; 840 #endif 841 setup_clear_cpu_cap(X86_FEATURE_APIC); 842 } 843 844 #ifdef CONFIG_PCI 845 if (pci_early_dump_regs) 846 early_dump_pci_devices(); 847 #endif 848 849 finish_e820_parsing(); 850 851 if (efi_enabled) 852 efi_init(); 853 854 dmi_scan_machine(); 855 856 /* 857 * VMware detection requires dmi to be available, so this 858 * needs to be done after dmi_scan_machine, for the BP. 859 */ 860 init_hypervisor_platform(); 861 862 x86_init.resources.probe_roms(); 863 864 /* after parse_early_param, so could debug it */ 865 insert_resource(&iomem_resource, &code_resource); 866 insert_resource(&iomem_resource, &data_resource); 867 insert_resource(&iomem_resource, &bss_resource); 868 869 trim_bios_range(); 870 #ifdef CONFIG_X86_32 871 if (ppro_with_ram_bug()) { 872 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 873 E820_RESERVED); 874 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 875 printk(KERN_INFO "fixed physical RAM map:\n"); 876 e820_print_map("bad_ppro"); 877 } 878 #else 879 early_gart_iommu_check(); 880 #endif 881 882 /* 883 * partially used pages are not usable - thus 884 * we are rounding upwards: 885 */ 886 max_pfn = e820_end_of_ram_pfn(); 887 888 /* update e820 for memory not covered by WB MTRRs */ 889 mtrr_bp_init(); 890 if (mtrr_trim_uncached_memory(max_pfn)) 891 max_pfn = e820_end_of_ram_pfn(); 892 893 #ifdef CONFIG_X86_32 894 /* max_low_pfn get updated here */ 895 find_low_pfn_range(); 896 #else 897 num_physpages = max_pfn; 898 899 check_x2apic(); 900 901 /* How many end-of-memory variables you have, grandma! */ 902 /* need this before calling reserve_initrd */ 903 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 904 max_low_pfn = e820_end_of_low_ram_pfn(); 905 else 906 max_low_pfn = max_pfn; 907 908 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 909 #endif 910 911 /* 912 * Find and reserve possible boot-time SMP configuration: 913 */ 914 find_smp_config(); 915 916 reserve_ibft_region(); 917 918 /* 919 * Need to conclude brk, before memblock_x86_fill() 920 * it could use memblock_find_in_range, could overlap with 921 * brk area. 922 */ 923 reserve_brk(); 924 925 memblock.current_limit = get_max_mapped(); 926 memblock_x86_fill(); 927 928 /* preallocate 4k for mptable mpc */ 929 early_reserve_e820_mpc_new(); 930 931 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 932 setup_bios_corruption_check(); 933 #endif 934 935 printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n", 936 max_pfn_mapped<<PAGE_SHIFT); 937 938 reserve_trampoline_memory(); 939 940 #ifdef CONFIG_ACPI_SLEEP 941 /* 942 * Reserve low memory region for sleep support. 943 * even before init_memory_mapping 944 */ 945 acpi_reserve_wakeup_memory(); 946 #endif 947 init_gbpages(); 948 949 /* max_pfn_mapped is updated here */ 950 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT); 951 max_pfn_mapped = max_low_pfn_mapped; 952 953 #ifdef CONFIG_X86_64 954 if (max_pfn > max_low_pfn) { 955 max_pfn_mapped = init_memory_mapping(1UL<<32, 956 max_pfn<<PAGE_SHIFT); 957 /* can we preseve max_low_pfn ?*/ 958 max_low_pfn = max_pfn; 959 } 960 #endif 961 memblock.current_limit = get_max_mapped(); 962 963 /* 964 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 965 */ 966 967 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 968 if (init_ohci1394_dma_early) 969 init_ohci1394_dma_on_all_controllers(); 970 #endif 971 972 reserve_initrd(); 973 974 reserve_crashkernel(); 975 976 vsmp_init(); 977 978 io_delay_init(); 979 980 /* 981 * Parse the ACPI tables for possible boot-time SMP configuration. 982 */ 983 acpi_boot_table_init(); 984 985 early_acpi_boot_init(); 986 987 #ifdef CONFIG_ACPI_NUMA 988 /* 989 * Parse SRAT to discover nodes. 990 */ 991 acpi = acpi_numa_init(); 992 #endif 993 994 #ifdef CONFIG_AMD_NUMA 995 if (!acpi) 996 amd = !amd_numa_init(0, max_pfn); 997 #endif 998 999 initmem_init(0, max_pfn, acpi, amd); 1000 memblock_find_dma_reserve(); 1001 dma32_reserve_bootmem(); 1002 1003 #ifdef CONFIG_KVM_CLOCK 1004 kvmclock_init(); 1005 #endif 1006 1007 x86_init.paging.pagetable_setup_start(swapper_pg_dir); 1008 paging_init(); 1009 x86_init.paging.pagetable_setup_done(swapper_pg_dir); 1010 1011 #ifdef CONFIG_X86_32 1012 /* sync back kernel address range */ 1013 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1014 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1015 KERNEL_PGD_PTRS); 1016 #endif 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_and_gsi_init(); 1049 1050 kvm_guest_init(); 1051 1052 e820_reserve_resources(); 1053 e820_mark_nosave_regions(max_low_pfn); 1054 1055 x86_init.resources.reserve_resources(); 1056 1057 e820_setup_gap(); 1058 1059 #ifdef CONFIG_VT 1060 #if defined(CONFIG_VGA_CONSOLE) 1061 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1062 conswitchp = &vga_con; 1063 #elif defined(CONFIG_DUMMY_CONSOLE) 1064 conswitchp = &dummy_con; 1065 #endif 1066 #endif 1067 x86_init.oem.banner(); 1068 1069 mcheck_init(); 1070 1071 local_irq_save(flags); 1072 arch_init_ideal_nop5(); 1073 local_irq_restore(flags); 1074 } 1075 1076 #ifdef CONFIG_X86_32 1077 1078 static struct resource video_ram_resource = { 1079 .name = "Video RAM area", 1080 .start = 0xa0000, 1081 .end = 0xbffff, 1082 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1083 }; 1084 1085 void __init i386_reserve_resources(void) 1086 { 1087 request_resource(&iomem_resource, &video_ram_resource); 1088 reserve_standard_io_resources(); 1089 } 1090 1091 #endif /* CONFIG_X86_32 */ 1092