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/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#include <linux/dma-contiguous.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#include <linux/jiffies.h> 72 73#include <video/edid.h> 74 75#include <asm/mtrr.h> 76#include <asm/apic.h> 77#include <asm/realmode.h> 78#include <asm/e820.h> 79#include <asm/mpspec.h> 80#include <asm/setup.h> 81#include <asm/efi.h> 82#include <asm/timer.h> 83#include <asm/i8259.h> 84#include <asm/sections.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#include <asm/kasan.h> 93 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#include <asm/mce.h> 112#include <asm/alternative.h> 113#include <asm/prom.h> 114 115/* 116 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 117 * max_pfn_mapped: highest direct mapped pfn over 4GB 118 * 119 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are 120 * represented by pfn_mapped 121 */ 122unsigned long max_low_pfn_mapped; 123unsigned long max_pfn_mapped; 124 125#ifdef CONFIG_DMI 126RESERVE_BRK(dmi_alloc, 65536); 127#endif 128 129 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 145struct boot_params boot_params; 146 147/* 148 * Machine setup.. 149 */ 150static struct resource data_resource = { 151 .name = "Kernel data", 152 .start = 0, 153 .end = 0, 154 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 155}; 156 157static struct resource code_resource = { 158 .name = "Kernel code", 159 .start = 0, 160 .end = 0, 161 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 162}; 163 164static struct resource bss_resource = { 165 .name = "Kernel bss", 166 .start = 0, 167 .end = 0, 168 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 169}; 170 171 172#ifdef CONFIG_X86_32 173/* cpu data as detected by the assembly code in head.S */ 174struct cpuinfo_x86 new_cpu_data = { 175 .wp_works_ok = -1, 176}; 177/* common cpu data for all cpus */ 178struct cpuinfo_x86 boot_cpu_data __read_mostly = { 179 .wp_works_ok = -1, 180}; 181EXPORT_SYMBOL(boot_cpu_data); 182 183unsigned int def_to_bigsmp; 184 185/* for MCA, but anyone else can use it if they want */ 186unsigned int machine_id; 187unsigned int machine_submodel_id; 188unsigned int BIOS_revision; 189 190struct apm_info apm_info; 191EXPORT_SYMBOL(apm_info); 192 193#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 194 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 195struct ist_info ist_info; 196EXPORT_SYMBOL(ist_info); 197#else 198struct ist_info ist_info; 199#endif 200 201#else 202struct cpuinfo_x86 boot_cpu_data __read_mostly = { 203 .x86_phys_bits = MAX_PHYSMEM_BITS, 204}; 205EXPORT_SYMBOL(boot_cpu_data); 206#endif 207 208 209#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 210__visible unsigned long mmu_cr4_features; 211#else 212__visible unsigned long mmu_cr4_features = X86_CR4_PAE; 213#endif 214 215/* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 216int bootloader_type, bootloader_version; 217 218/* 219 * Setup options 220 */ 221struct screen_info screen_info; 222EXPORT_SYMBOL(screen_info); 223struct edid_info edid_info; 224EXPORT_SYMBOL_GPL(edid_info); 225 226extern int root_mountflags; 227 228unsigned long saved_video_mode; 229 230#define RAMDISK_IMAGE_START_MASK 0x07FF 231#define RAMDISK_PROMPT_FLAG 0x8000 232#define RAMDISK_LOAD_FLAG 0x4000 233 234static char __initdata command_line[COMMAND_LINE_SIZE]; 235#ifdef CONFIG_CMDLINE_BOOL 236static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 237#endif 238 239#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 240struct edd edd; 241#ifdef CONFIG_EDD_MODULE 242EXPORT_SYMBOL(edd); 243#endif 244/** 245 * copy_edd() - Copy the BIOS EDD information 246 * from boot_params into a safe place. 247 * 248 */ 249static inline void __init copy_edd(void) 250{ 251 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 252 sizeof(edd.mbr_signature)); 253 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 254 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 255 edd.edd_info_nr = boot_params.eddbuf_entries; 256} 257#else 258static inline void __init copy_edd(void) 259{ 260} 261#endif 262 263void * __init extend_brk(size_t size, size_t align) 264{ 265 size_t mask = align - 1; 266 void *ret; 267 268 BUG_ON(_brk_start == 0); 269 BUG_ON(align & mask); 270 271 _brk_end = (_brk_end + mask) & ~mask; 272 BUG_ON((char *)(_brk_end + size) > __brk_limit); 273 274 ret = (void *)_brk_end; 275 _brk_end += size; 276 277 memset(ret, 0, size); 278 279 return ret; 280} 281 282#ifdef CONFIG_X86_32 283static void __init cleanup_highmap(void) 284{ 285} 286#endif 287 288static void __init reserve_brk(void) 289{ 290 if (_brk_end > _brk_start) 291 memblock_reserve(__pa_symbol(_brk_start), 292 _brk_end - _brk_start); 293 294 /* Mark brk area as locked down and no longer taking any 295 new allocations */ 296 _brk_start = 0; 297} 298 299u64 relocated_ramdisk; 300 301#ifdef CONFIG_BLK_DEV_INITRD 302 303static u64 __init get_ramdisk_image(void) 304{ 305 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 306 307 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 308 309 return ramdisk_image; 310} 311static u64 __init get_ramdisk_size(void) 312{ 313 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 314 315 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 316 317 return ramdisk_size; 318} 319 320#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT) 321static void __init relocate_initrd(void) 322{ 323 /* Assume only end is not page aligned */ 324 u64 ramdisk_image = get_ramdisk_image(); 325 u64 ramdisk_size = get_ramdisk_size(); 326 u64 area_size = PAGE_ALIGN(ramdisk_size); 327 unsigned long slop, clen, mapaddr; 328 char *p, *q; 329 330 /* We need to move the initrd down into directly mapped mem */ 331 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 332 area_size, PAGE_SIZE); 333 334 if (!relocated_ramdisk) 335 panic("Cannot find place for new RAMDISK of size %lld\n", 336 ramdisk_size); 337 338 /* Note: this includes all the mem currently occupied by 339 the initrd, we rely on that fact to keep the data intact. */ 340 memblock_reserve(relocated_ramdisk, area_size); 341 initrd_start = relocated_ramdisk + PAGE_OFFSET; 342 initrd_end = initrd_start + ramdisk_size; 343 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 344 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 345 346 q = (char *)initrd_start; 347 348 /* Copy the initrd */ 349 while (ramdisk_size) { 350 slop = ramdisk_image & ~PAGE_MASK; 351 clen = ramdisk_size; 352 if (clen > MAX_MAP_CHUNK-slop) 353 clen = MAX_MAP_CHUNK-slop; 354 mapaddr = ramdisk_image & PAGE_MASK; 355 p = early_memremap(mapaddr, clen+slop); 356 memcpy(q, p+slop, clen); 357 early_memunmap(p, clen+slop); 358 q += clen; 359 ramdisk_image += clen; 360 ramdisk_size -= clen; 361 } 362 363 ramdisk_image = get_ramdisk_image(); 364 ramdisk_size = get_ramdisk_size(); 365 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 366 " [mem %#010llx-%#010llx]\n", 367 ramdisk_image, ramdisk_image + ramdisk_size - 1, 368 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 369} 370 371static void __init early_reserve_initrd(void) 372{ 373 /* Assume only end is not page aligned */ 374 u64 ramdisk_image = get_ramdisk_image(); 375 u64 ramdisk_size = get_ramdisk_size(); 376 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 377 378 if (!boot_params.hdr.type_of_loader || 379 !ramdisk_image || !ramdisk_size) 380 return; /* No initrd provided by bootloader */ 381 382 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 383} 384static void __init reserve_initrd(void) 385{ 386 /* Assume only end is not page aligned */ 387 u64 ramdisk_image = get_ramdisk_image(); 388 u64 ramdisk_size = get_ramdisk_size(); 389 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 390 u64 mapped_size; 391 392 if (!boot_params.hdr.type_of_loader || 393 !ramdisk_image || !ramdisk_size) 394 return; /* No initrd provided by bootloader */ 395 396 initrd_start = 0; 397 398 mapped_size = memblock_mem_size(max_pfn_mapped); 399 if (ramdisk_size >= (mapped_size>>1)) 400 panic("initrd too large to handle, " 401 "disabling initrd (%lld needed, %lld available)\n", 402 ramdisk_size, mapped_size>>1); 403 404 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 405 ramdisk_end - 1); 406 407 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 408 PFN_DOWN(ramdisk_end))) { 409 /* All are mapped, easy case */ 410 initrd_start = ramdisk_image + PAGE_OFFSET; 411 initrd_end = initrd_start + ramdisk_size; 412 return; 413 } 414 415 relocate_initrd(); 416 417 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 418} 419#else 420static void __init early_reserve_initrd(void) 421{ 422} 423static void __init reserve_initrd(void) 424{ 425} 426#endif /* CONFIG_BLK_DEV_INITRD */ 427 428static void __init parse_setup_data(void) 429{ 430 struct setup_data *data; 431 u64 pa_data, pa_next; 432 433 pa_data = boot_params.hdr.setup_data; 434 while (pa_data) { 435 u32 data_len, data_type; 436 437 data = early_memremap(pa_data, sizeof(*data)); 438 data_len = data->len + sizeof(struct setup_data); 439 data_type = data->type; 440 pa_next = data->next; 441 early_memunmap(data, sizeof(*data)); 442 443 switch (data_type) { 444 case SETUP_E820_EXT: 445 parse_e820_ext(pa_data, data_len); 446 break; 447 case SETUP_DTB: 448 add_dtb(pa_data); 449 break; 450 case SETUP_EFI: 451 parse_efi_setup(pa_data, data_len); 452 break; 453 default: 454 break; 455 } 456 pa_data = pa_next; 457 } 458} 459 460static void __init e820_reserve_setup_data(void) 461{ 462 struct setup_data *data; 463 u64 pa_data; 464 int found = 0; 465 466 pa_data = boot_params.hdr.setup_data; 467 while (pa_data) { 468 data = early_memremap(pa_data, sizeof(*data)); 469 e820_update_range(pa_data, sizeof(*data)+data->len, 470 E820_RAM, E820_RESERVED_KERN); 471 found = 1; 472 pa_data = data->next; 473 early_memunmap(data, sizeof(*data)); 474 } 475 if (!found) 476 return; 477 478 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 479 memcpy(&e820_saved, &e820, sizeof(struct e820map)); 480 printk(KERN_INFO "extended physical RAM map:\n"); 481 e820_print_map("reserve setup_data"); 482} 483 484static void __init memblock_x86_reserve_range_setup_data(void) 485{ 486 struct setup_data *data; 487 u64 pa_data; 488 489 pa_data = boot_params.hdr.setup_data; 490 while (pa_data) { 491 data = early_memremap(pa_data, sizeof(*data)); 492 memblock_reserve(pa_data, sizeof(*data) + data->len); 493 pa_data = data->next; 494 early_memunmap(data, sizeof(*data)); 495 } 496} 497 498/* 499 * --------- Crashkernel reservation ------------------------------ 500 */ 501 502#ifdef CONFIG_KEXEC 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 64bit, old kexec-tools need to under 896MiB. 508 */ 509#ifdef CONFIG_X86_32 510# define CRASH_KERNEL_ADDR_LOW_MAX (512 << 20) 511# define CRASH_KERNEL_ADDR_HIGH_MAX (512 << 20) 512#else 513# define CRASH_KERNEL_ADDR_LOW_MAX (896UL<<20) 514# define CRASH_KERNEL_ADDR_HIGH_MAX MAXMEM 515#endif 516 517static void __init reserve_crashkernel_low(void) 518{ 519#ifdef CONFIG_X86_64 520 const unsigned long long alignment = 16<<20; /* 16M */ 521 unsigned long long low_base = 0, low_size = 0; 522 unsigned long total_low_mem; 523 unsigned long long base; 524 bool auto_set = false; 525 int ret; 526 527 total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT)); 528 /* crashkernel=Y,low */ 529 ret = parse_crashkernel_low(boot_command_line, total_low_mem, 530 &low_size, &base); 531 if (ret != 0) { 532 /* 533 * two parts from lib/swiotlb.c: 534 * swiotlb size: user specified with swiotlb= or default. 535 * swiotlb overflow buffer: now is hardcoded to 32k. 536 * We round it to 8M for other buffers that 537 * may need to stay low too. 538 */ 539 low_size = swiotlb_size_or_default() + (8UL<<20); 540 auto_set = true; 541 } else { 542 /* passed with crashkernel=0,low ? */ 543 if (!low_size) 544 return; 545 } 546 547 low_base = memblock_find_in_range(low_size, (1ULL<<32), 548 low_size, alignment); 549 550 if (!low_base) { 551 if (!auto_set) 552 pr_info("crashkernel low reservation failed - No suitable area found.\n"); 553 554 return; 555 } 556 557 memblock_reserve(low_base, low_size); 558 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 559 (unsigned long)(low_size >> 20), 560 (unsigned long)(low_base >> 20), 561 (unsigned long)(total_low_mem >> 20)); 562 crashk_low_res.start = low_base; 563 crashk_low_res.end = low_base + low_size - 1; 564 insert_resource(&iomem_resource, &crashk_low_res); 565#endif 566} 567 568static void __init reserve_crashkernel(void) 569{ 570 const unsigned long long alignment = 16<<20; /* 16M */ 571 unsigned long long total_mem; 572 unsigned long long crash_size, crash_base; 573 bool high = false; 574 int ret; 575 576 total_mem = memblock_phys_mem_size(); 577 578 /* crashkernel=XM */ 579 ret = parse_crashkernel(boot_command_line, total_mem, 580 &crash_size, &crash_base); 581 if (ret != 0 || crash_size <= 0) { 582 /* crashkernel=X,high */ 583 ret = parse_crashkernel_high(boot_command_line, total_mem, 584 &crash_size, &crash_base); 585 if (ret != 0 || crash_size <= 0) 586 return; 587 high = true; 588 } 589 590 /* 0 means: find the address automatically */ 591 if (crash_base <= 0) { 592 /* 593 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 594 */ 595 crash_base = memblock_find_in_range(alignment, 596 high ? CRASH_KERNEL_ADDR_HIGH_MAX : 597 CRASH_KERNEL_ADDR_LOW_MAX, 598 crash_size, alignment); 599 600 if (!crash_base) { 601 pr_info("crashkernel reservation failed - No suitable area found.\n"); 602 return; 603 } 604 605 } else { 606 unsigned long long start; 607 608 start = memblock_find_in_range(crash_base, 609 crash_base + crash_size, crash_size, 1<<20); 610 if (start != crash_base) { 611 pr_info("crashkernel reservation failed - memory is in use.\n"); 612 return; 613 } 614 } 615 memblock_reserve(crash_base, crash_size); 616 617 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " 618 "for crashkernel (System RAM: %ldMB)\n", 619 (unsigned long)(crash_size >> 20), 620 (unsigned long)(crash_base >> 20), 621 (unsigned long)(total_mem >> 20)); 622 623 crashk_res.start = crash_base; 624 crashk_res.end = crash_base + crash_size - 1; 625 insert_resource(&iomem_resource, &crashk_res); 626 627 if (crash_base >= (1ULL<<32)) 628 reserve_crashkernel_low(); 629} 630#else 631static void __init reserve_crashkernel(void) 632{ 633} 634#endif 635 636static struct resource standard_io_resources[] = { 637 { .name = "dma1", .start = 0x00, .end = 0x1f, 638 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 639 { .name = "pic1", .start = 0x20, .end = 0x21, 640 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 641 { .name = "timer0", .start = 0x40, .end = 0x43, 642 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 643 { .name = "timer1", .start = 0x50, .end = 0x53, 644 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 645 { .name = "keyboard", .start = 0x60, .end = 0x60, 646 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 647 { .name = "keyboard", .start = 0x64, .end = 0x64, 648 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 649 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 650 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 651 { .name = "pic2", .start = 0xa0, .end = 0xa1, 652 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 653 { .name = "dma2", .start = 0xc0, .end = 0xdf, 654 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 655 { .name = "fpu", .start = 0xf0, .end = 0xff, 656 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 657}; 658 659void __init reserve_standard_io_resources(void) 660{ 661 int i; 662 663 /* request I/O space for devices used on all i[345]86 PCs */ 664 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 665 request_resource(&ioport_resource, &standard_io_resources[i]); 666 667} 668 669static __init void reserve_ibft_region(void) 670{ 671 unsigned long addr, size = 0; 672 673 addr = find_ibft_region(&size); 674 675 if (size) 676 memblock_reserve(addr, size); 677} 678 679static bool __init snb_gfx_workaround_needed(void) 680{ 681#ifdef CONFIG_PCI 682 int i; 683 u16 vendor, devid; 684 static const __initconst u16 snb_ids[] = { 685 0x0102, 686 0x0112, 687 0x0122, 688 0x0106, 689 0x0116, 690 0x0126, 691 0x010a, 692 }; 693 694 /* Assume no if something weird is going on with PCI */ 695 if (!early_pci_allowed()) 696 return false; 697 698 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 699 if (vendor != 0x8086) 700 return false; 701 702 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 703 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 704 if (devid == snb_ids[i]) 705 return true; 706#endif 707 708 return false; 709} 710 711/* 712 * Sandy Bridge graphics has trouble with certain ranges, exclude 713 * them from allocation. 714 */ 715static void __init trim_snb_memory(void) 716{ 717 static const __initconst unsigned long bad_pages[] = { 718 0x20050000, 719 0x20110000, 720 0x20130000, 721 0x20138000, 722 0x40004000, 723 }; 724 int i; 725 726 if (!snb_gfx_workaround_needed()) 727 return; 728 729 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 730 731 /* 732 * Reserve all memory below the 1 MB mark that has not 733 * already been reserved. 734 */ 735 memblock_reserve(0, 1<<20); 736 737 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 738 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 739 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 740 bad_pages[i]); 741 } 742} 743 744/* 745 * Here we put platform-specific memory range workarounds, i.e. 746 * memory known to be corrupt or otherwise in need to be reserved on 747 * specific platforms. 748 * 749 * If this gets used more widely it could use a real dispatch mechanism. 750 */ 751static void __init trim_platform_memory_ranges(void) 752{ 753 trim_snb_memory(); 754} 755 756static void __init trim_bios_range(void) 757{ 758 /* 759 * A special case is the first 4Kb of memory; 760 * This is a BIOS owned area, not kernel ram, but generally 761 * not listed as such in the E820 table. 762 * 763 * This typically reserves additional memory (64KiB by default) 764 * since some BIOSes are known to corrupt low memory. See the 765 * Kconfig help text for X86_RESERVE_LOW. 766 */ 767 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 768 769 /* 770 * special case: Some BIOSen report the PC BIOS 771 * area (640->1Mb) as ram even though it is not. 772 * take them out. 773 */ 774 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 775 776 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 777} 778 779/* called before trim_bios_range() to spare extra sanitize */ 780static void __init e820_add_kernel_range(void) 781{ 782 u64 start = __pa_symbol(_text); 783 u64 size = __pa_symbol(_end) - start; 784 785 /* 786 * Complain if .text .data and .bss are not marked as E820_RAM and 787 * attempt to fix it by adding the range. We may have a confused BIOS, 788 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 789 * exclude kernel range. If we really are running on top non-RAM, 790 * we will crash later anyways. 791 */ 792 if (e820_all_mapped(start, start + size, E820_RAM)) 793 return; 794 795 pr_warn(".text .data .bss are not marked as E820_RAM!\n"); 796 e820_remove_range(start, size, E820_RAM, 0); 797 e820_add_region(start, size, E820_RAM); 798} 799 800static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 801 802static int __init parse_reservelow(char *p) 803{ 804 unsigned long long size; 805 806 if (!p) 807 return -EINVAL; 808 809 size = memparse(p, &p); 810 811 if (size < 4096) 812 size = 4096; 813 814 if (size > 640*1024) 815 size = 640*1024; 816 817 reserve_low = size; 818 819 return 0; 820} 821 822early_param("reservelow", parse_reservelow); 823 824static void __init trim_low_memory_range(void) 825{ 826 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 827} 828 829/* 830 * Dump out kernel offset information on panic. 831 */ 832static int 833dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 834{ 835 if (kaslr_enabled()) { 836 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 837 (unsigned long)&_text - __START_KERNEL, 838 __START_KERNEL, 839 __START_KERNEL_map, 840 MODULES_VADDR-1); 841 } else { 842 pr_emerg("Kernel Offset: disabled\n"); 843 } 844 845 return 0; 846} 847 848/* 849 * Determine if we were loaded by an EFI loader. If so, then we have also been 850 * passed the efi memmap, systab, etc., so we should use these data structures 851 * for initialization. Note, the efi init code path is determined by the 852 * global efi_enabled. This allows the same kernel image to be used on existing 853 * systems (with a traditional BIOS) as well as on EFI systems. 854 */ 855/* 856 * setup_arch - architecture-specific boot-time initializations 857 * 858 * Note: On x86_64, fixmaps are ready for use even before this is called. 859 */ 860 861void __init setup_arch(char **cmdline_p) 862{ 863 memblock_reserve(__pa_symbol(_text), 864 (unsigned long)__bss_stop - (unsigned long)_text); 865 866 early_reserve_initrd(); 867 868 /* 869 * At this point everything still needed from the boot loader 870 * or BIOS or kernel text should be early reserved or marked not 871 * RAM in e820. All other memory is free game. 872 */ 873 874#ifdef CONFIG_X86_32 875 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 876 877 /* 878 * copy kernel address range established so far and switch 879 * to the proper swapper page table 880 */ 881 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 882 initial_page_table + KERNEL_PGD_BOUNDARY, 883 KERNEL_PGD_PTRS); 884 885 load_cr3(swapper_pg_dir); 886 /* 887 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 888 * a cr3 based tlb flush, so the following __flush_tlb_all() 889 * will not flush anything because the cpu quirk which clears 890 * X86_FEATURE_PGE has not been invoked yet. Though due to the 891 * load_cr3() above the TLB has been flushed already. The 892 * quirk is invoked before subsequent calls to __flush_tlb_all() 893 * so proper operation is guaranteed. 894 */ 895 __flush_tlb_all(); 896#else 897 printk(KERN_INFO "Command line: %s\n", boot_command_line); 898#endif 899 900 /* 901 * If we have OLPC OFW, we might end up relocating the fixmap due to 902 * reserve_top(), so do this before touching the ioremap area. 903 */ 904 olpc_ofw_detect(); 905 906 early_trap_init(); 907 early_cpu_init(); 908 early_ioremap_init(); 909 910 setup_olpc_ofw_pgd(); 911 912 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 913 screen_info = boot_params.screen_info; 914 edid_info = boot_params.edid_info; 915#ifdef CONFIG_X86_32 916 apm_info.bios = boot_params.apm_bios_info; 917 ist_info = boot_params.ist_info; 918 if (boot_params.sys_desc_table.length != 0) { 919 machine_id = boot_params.sys_desc_table.table[0]; 920 machine_submodel_id = boot_params.sys_desc_table.table[1]; 921 BIOS_revision = boot_params.sys_desc_table.table[2]; 922 } 923#endif 924 saved_video_mode = boot_params.hdr.vid_mode; 925 bootloader_type = boot_params.hdr.type_of_loader; 926 if ((bootloader_type >> 4) == 0xe) { 927 bootloader_type &= 0xf; 928 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 929 } 930 bootloader_version = bootloader_type & 0xf; 931 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 932 933#ifdef CONFIG_BLK_DEV_RAM 934 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 935 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 936 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 937#endif 938#ifdef CONFIG_EFI 939 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 940 EFI32_LOADER_SIGNATURE, 4)) { 941 set_bit(EFI_BOOT, &efi.flags); 942 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 943 EFI64_LOADER_SIGNATURE, 4)) { 944 set_bit(EFI_BOOT, &efi.flags); 945 set_bit(EFI_64BIT, &efi.flags); 946 } 947 948 if (efi_enabled(EFI_BOOT)) 949 efi_memblock_x86_reserve_range(); 950#endif 951 952 x86_init.oem.arch_setup(); 953 954 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 955 setup_memory_map(); 956 parse_setup_data(); 957 958 copy_edd(); 959 960 if (!boot_params.hdr.root_flags) 961 root_mountflags &= ~MS_RDONLY; 962 init_mm.start_code = (unsigned long) _text; 963 init_mm.end_code = (unsigned long) _etext; 964 init_mm.end_data = (unsigned long) _edata; 965 init_mm.brk = _brk_end; 966 967 mpx_mm_init(&init_mm); 968 969 code_resource.start = __pa_symbol(_text); 970 code_resource.end = __pa_symbol(_etext)-1; 971 data_resource.start = __pa_symbol(_etext); 972 data_resource.end = __pa_symbol(_edata)-1; 973 bss_resource.start = __pa_symbol(__bss_start); 974 bss_resource.end = __pa_symbol(__bss_stop)-1; 975 976#ifdef CONFIG_CMDLINE_BOOL 977#ifdef CONFIG_CMDLINE_OVERRIDE 978 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 979#else 980 if (builtin_cmdline[0]) { 981 /* append boot loader cmdline to builtin */ 982 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 983 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 984 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 985 } 986#endif 987#endif 988 989 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 990 *cmdline_p = command_line; 991 992 /* 993 * x86_configure_nx() is called before parse_early_param() to detect 994 * whether hardware doesn't support NX (so that the early EHCI debug 995 * console setup can safely call set_fixmap()). It may then be called 996 * again from within noexec_setup() during parsing early parameters 997 * to honor the respective command line option. 998 */ 999 x86_configure_nx(); 1000 1001 parse_early_param(); 1002 1003 x86_report_nx(); 1004 1005 /* after early param, so could get panic from serial */ 1006 memblock_x86_reserve_range_setup_data(); 1007 1008 if (acpi_mps_check()) { 1009#ifdef CONFIG_X86_LOCAL_APIC 1010 disable_apic = 1; 1011#endif 1012 setup_clear_cpu_cap(X86_FEATURE_APIC); 1013 } 1014 1015#ifdef CONFIG_PCI 1016 if (pci_early_dump_regs) 1017 early_dump_pci_devices(); 1018#endif 1019 1020 /* update the e820_saved too */ 1021 e820_reserve_setup_data(); 1022 finish_e820_parsing(); 1023 1024 if (efi_enabled(EFI_BOOT)) 1025 efi_init(); 1026 1027 dmi_scan_machine(); 1028 dmi_memdev_walk(); 1029 dmi_set_dump_stack_arch_desc(); 1030 1031 /* 1032 * VMware detection requires dmi to be available, so this 1033 * needs to be done after dmi_scan_machine, for the BP. 1034 */ 1035 init_hypervisor_platform(); 1036 1037 x86_init.resources.probe_roms(); 1038 1039 /* after parse_early_param, so could debug it */ 1040 insert_resource(&iomem_resource, &code_resource); 1041 insert_resource(&iomem_resource, &data_resource); 1042 insert_resource(&iomem_resource, &bss_resource); 1043 1044 e820_add_kernel_range(); 1045 trim_bios_range(); 1046#ifdef CONFIG_X86_32 1047 if (ppro_with_ram_bug()) { 1048 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 1049 E820_RESERVED); 1050 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 1051 printk(KERN_INFO "fixed physical RAM map:\n"); 1052 e820_print_map("bad_ppro"); 1053 } 1054#else 1055 early_gart_iommu_check(); 1056#endif 1057 1058 /* 1059 * partially used pages are not usable - thus 1060 * we are rounding upwards: 1061 */ 1062 max_pfn = e820_end_of_ram_pfn(); 1063 1064 /* update e820 for memory not covered by WB MTRRs */ 1065 mtrr_bp_init(); 1066 if (mtrr_trim_uncached_memory(max_pfn)) 1067 max_pfn = e820_end_of_ram_pfn(); 1068 1069#ifdef CONFIG_X86_32 1070 /* max_low_pfn get updated here */ 1071 find_low_pfn_range(); 1072#else 1073 check_x2apic(); 1074 1075 /* How many end-of-memory variables you have, grandma! */ 1076 /* need this before calling reserve_initrd */ 1077 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1078 max_low_pfn = e820_end_of_low_ram_pfn(); 1079 else 1080 max_low_pfn = max_pfn; 1081 1082 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1083#endif 1084 1085 /* 1086 * Find and reserve possible boot-time SMP configuration: 1087 */ 1088 find_smp_config(); 1089 1090 reserve_ibft_region(); 1091 1092 early_alloc_pgt_buf(); 1093 1094 /* 1095 * Need to conclude brk, before memblock_x86_fill() 1096 * it could use memblock_find_in_range, could overlap with 1097 * brk area. 1098 */ 1099 reserve_brk(); 1100 1101 cleanup_highmap(); 1102 1103 memblock_set_current_limit(ISA_END_ADDRESS); 1104 memblock_x86_fill(); 1105 1106 /* 1107 * The EFI specification says that boot service code won't be called 1108 * after ExitBootServices(). This is, in fact, a lie. 1109 */ 1110 if (efi_enabled(EFI_MEMMAP)) 1111 efi_reserve_boot_services(); 1112 1113 /* preallocate 4k for mptable mpc */ 1114 early_reserve_e820_mpc_new(); 1115 1116#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1117 setup_bios_corruption_check(); 1118#endif 1119 1120#ifdef CONFIG_X86_32 1121 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1122 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1123#endif 1124 1125 reserve_real_mode(); 1126 1127 trim_platform_memory_ranges(); 1128 trim_low_memory_range(); 1129 1130 init_mem_mapping(); 1131 1132 early_trap_pf_init(); 1133 1134 setup_real_mode(); 1135 1136 memblock_set_current_limit(get_max_mapped()); 1137 1138 /* 1139 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1140 */ 1141 1142#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1143 if (init_ohci1394_dma_early) 1144 init_ohci1394_dma_on_all_controllers(); 1145#endif 1146 /* Allocate bigger log buffer */ 1147 setup_log_buf(1); 1148 1149 reserve_initrd(); 1150 1151#if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD) 1152 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start); 1153#endif 1154 1155 vsmp_init(); 1156 1157 io_delay_init(); 1158 1159 /* 1160 * Parse the ACPI tables for possible boot-time SMP configuration. 1161 */ 1162 acpi_boot_table_init(); 1163 1164 early_acpi_boot_init(); 1165 1166 initmem_init(); 1167 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1168 1169 /* 1170 * Reserve memory for crash kernel after SRAT is parsed so that it 1171 * won't consume hotpluggable memory. 1172 */ 1173 reserve_crashkernel(); 1174 1175 memblock_find_dma_reserve(); 1176 1177#ifdef CONFIG_KVM_GUEST 1178 kvmclock_init(); 1179#endif 1180 1181 x86_init.paging.pagetable_init(); 1182 1183 kasan_init(); 1184 1185 if (boot_cpu_data.cpuid_level >= 0) { 1186 /* A CPU has %cr4 if and only if it has CPUID */ 1187 mmu_cr4_features = __read_cr4(); 1188 if (trampoline_cr4_features) 1189 *trampoline_cr4_features = mmu_cr4_features; 1190 } 1191 1192#ifdef CONFIG_X86_32 1193 /* sync back kernel address range */ 1194 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1195 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1196 KERNEL_PGD_PTRS); 1197 1198 /* 1199 * sync back low identity map too. It is used for example 1200 * in the 32-bit EFI stub. 1201 */ 1202 clone_pgd_range(initial_page_table, 1203 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1204 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 1205#endif 1206 1207 tboot_probe(); 1208 1209 map_vsyscall(); 1210 1211 generic_apic_probe(); 1212 1213 early_quirks(); 1214 1215 /* 1216 * Read APIC and some other early information from ACPI tables. 1217 */ 1218 acpi_boot_init(); 1219 sfi_init(); 1220 x86_dtb_init(); 1221 1222 /* 1223 * get boot-time SMP configuration: 1224 */ 1225 if (smp_found_config) 1226 get_smp_config(); 1227 1228 prefill_possible_map(); 1229 1230 init_cpu_to_node(); 1231 1232 init_apic_mappings(); 1233 if (x86_io_apic_ops.init) 1234 x86_io_apic_ops.init(); 1235 1236 kvm_guest_init(); 1237 1238 e820_reserve_resources(); 1239 e820_mark_nosave_regions(max_low_pfn); 1240 1241 x86_init.resources.reserve_resources(); 1242 1243 e820_setup_gap(); 1244 1245#ifdef CONFIG_VT 1246#if defined(CONFIG_VGA_CONSOLE) 1247 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1248 conswitchp = &vga_con; 1249#elif defined(CONFIG_DUMMY_CONSOLE) 1250 conswitchp = &dummy_con; 1251#endif 1252#endif 1253 x86_init.oem.banner(); 1254 1255 x86_init.timers.wallclock_init(); 1256 1257 mcheck_init(); 1258 1259 arch_init_ideal_nops(); 1260 1261 register_refined_jiffies(CLOCK_TICK_RATE); 1262 1263#ifdef CONFIG_EFI 1264 if (efi_enabled(EFI_BOOT)) 1265 efi_apply_memmap_quirks(); 1266#endif 1267} 1268 1269#ifdef CONFIG_X86_32 1270 1271static struct resource video_ram_resource = { 1272 .name = "Video RAM area", 1273 .start = 0xa0000, 1274 .end = 0xbffff, 1275 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1276}; 1277 1278void __init i386_reserve_resources(void) 1279{ 1280 request_resource(&iomem_resource, &video_ram_resource); 1281 reserve_standard_io_resources(); 1282} 1283 1284#endif /* CONFIG_X86_32 */ 1285 1286static struct notifier_block kernel_offset_notifier = { 1287 .notifier_call = dump_kernel_offset 1288}; 1289 1290static int __init register_kernel_offset_dumper(void) 1291{ 1292 atomic_notifier_chain_register(&panic_notifier_list, 1293 &kernel_offset_notifier); 1294 return 0; 1295} 1296__initcall(register_kernel_offset_dumper); 1297