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 #include <asm/microcode.h> 115 116 /* 117 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 118 * max_pfn_mapped: highest direct mapped pfn over 4GB 119 * 120 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are 121 * represented by pfn_mapped 122 */ 123 unsigned long max_low_pfn_mapped; 124 unsigned long max_pfn_mapped; 125 126 #ifdef CONFIG_DMI 127 RESERVE_BRK(dmi_alloc, 65536); 128 #endif 129 130 131 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 132 unsigned long _brk_end = (unsigned long)__brk_base; 133 134 #ifdef CONFIG_X86_64 135 int default_cpu_present_to_apicid(int mps_cpu) 136 { 137 return __default_cpu_present_to_apicid(mps_cpu); 138 } 139 140 int default_check_phys_apicid_present(int phys_apicid) 141 { 142 return __default_check_phys_apicid_present(phys_apicid); 143 } 144 #endif 145 146 struct boot_params boot_params; 147 148 /* 149 * Machine setup.. 150 */ 151 static struct resource data_resource = { 152 .name = "Kernel data", 153 .start = 0, 154 .end = 0, 155 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 156 }; 157 158 static struct resource code_resource = { 159 .name = "Kernel code", 160 .start = 0, 161 .end = 0, 162 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 163 }; 164 165 static struct resource bss_resource = { 166 .name = "Kernel bss", 167 .start = 0, 168 .end = 0, 169 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 170 }; 171 172 173 #ifdef CONFIG_X86_32 174 /* cpu data as detected by the assembly code in head.S */ 175 struct cpuinfo_x86 new_cpu_data = { 176 .wp_works_ok = -1, 177 }; 178 /* common cpu data for all cpus */ 179 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 180 .wp_works_ok = -1, 181 }; 182 EXPORT_SYMBOL(boot_cpu_data); 183 184 unsigned int def_to_bigsmp; 185 186 /* for MCA, but anyone else can use it if they want */ 187 unsigned int machine_id; 188 unsigned int machine_submodel_id; 189 unsigned int BIOS_revision; 190 191 struct apm_info apm_info; 192 EXPORT_SYMBOL(apm_info); 193 194 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 195 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 196 struct ist_info ist_info; 197 EXPORT_SYMBOL(ist_info); 198 #else 199 struct ist_info ist_info; 200 #endif 201 202 #else 203 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 204 .x86_phys_bits = MAX_PHYSMEM_BITS, 205 }; 206 EXPORT_SYMBOL(boot_cpu_data); 207 #endif 208 209 210 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 211 __visible unsigned long mmu_cr4_features; 212 #else 213 __visible unsigned long mmu_cr4_features = X86_CR4_PAE; 214 #endif 215 216 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 217 int bootloader_type, bootloader_version; 218 219 /* 220 * Setup options 221 */ 222 struct screen_info screen_info; 223 EXPORT_SYMBOL(screen_info); 224 struct edid_info edid_info; 225 EXPORT_SYMBOL_GPL(edid_info); 226 227 extern int root_mountflags; 228 229 unsigned long saved_video_mode; 230 231 #define RAMDISK_IMAGE_START_MASK 0x07FF 232 #define RAMDISK_PROMPT_FLAG 0x8000 233 #define RAMDISK_LOAD_FLAG 0x4000 234 235 static char __initdata command_line[COMMAND_LINE_SIZE]; 236 #ifdef CONFIG_CMDLINE_BOOL 237 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 238 #endif 239 240 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 241 struct edd edd; 242 #ifdef CONFIG_EDD_MODULE 243 EXPORT_SYMBOL(edd); 244 #endif 245 /** 246 * copy_edd() - Copy the BIOS EDD information 247 * from boot_params into a safe place. 248 * 249 */ 250 static inline void __init copy_edd(void) 251 { 252 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 253 sizeof(edd.mbr_signature)); 254 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 255 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 256 edd.edd_info_nr = boot_params.eddbuf_entries; 257 } 258 #else 259 static inline void __init copy_edd(void) 260 { 261 } 262 #endif 263 264 void * __init extend_brk(size_t size, size_t align) 265 { 266 size_t mask = align - 1; 267 void *ret; 268 269 BUG_ON(_brk_start == 0); 270 BUG_ON(align & mask); 271 272 _brk_end = (_brk_end + mask) & ~mask; 273 BUG_ON((char *)(_brk_end + size) > __brk_limit); 274 275 ret = (void *)_brk_end; 276 _brk_end += size; 277 278 memset(ret, 0, size); 279 280 return ret; 281 } 282 283 #ifdef CONFIG_X86_32 284 static void __init cleanup_highmap(void) 285 { 286 } 287 #endif 288 289 static void __init reserve_brk(void) 290 { 291 if (_brk_end > _brk_start) 292 memblock_reserve(__pa_symbol(_brk_start), 293 _brk_end - _brk_start); 294 295 /* Mark brk area as locked down and no longer taking any 296 new allocations */ 297 _brk_start = 0; 298 } 299 300 u64 relocated_ramdisk; 301 302 #ifdef CONFIG_BLK_DEV_INITRD 303 304 static u64 __init get_ramdisk_image(void) 305 { 306 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 307 308 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 309 310 return ramdisk_image; 311 } 312 static u64 __init get_ramdisk_size(void) 313 { 314 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 315 316 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 317 318 return ramdisk_size; 319 } 320 321 static 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); 327 328 /* We need to move the initrd down into directly mapped mem */ 329 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 330 area_size, PAGE_SIZE); 331 332 if (!relocated_ramdisk) 333 panic("Cannot find place for new RAMDISK of size %lld\n", 334 ramdisk_size); 335 336 /* Note: this includes all the mem currently occupied by 337 the initrd, we rely on that fact to keep the data intact. */ 338 memblock_reserve(relocated_ramdisk, area_size); 339 initrd_start = relocated_ramdisk + PAGE_OFFSET; 340 initrd_end = initrd_start + ramdisk_size; 341 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 342 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 343 344 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 345 346 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 347 " [mem %#010llx-%#010llx]\n", 348 ramdisk_image, ramdisk_image + ramdisk_size - 1, 349 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 350 } 351 352 static void __init early_reserve_initrd(void) 353 { 354 /* Assume only end is not page aligned */ 355 u64 ramdisk_image = get_ramdisk_image(); 356 u64 ramdisk_size = get_ramdisk_size(); 357 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 358 359 if (!boot_params.hdr.type_of_loader || 360 !ramdisk_image || !ramdisk_size) 361 return; /* No initrd provided by bootloader */ 362 363 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 364 } 365 static void __init reserve_initrd(void) 366 { 367 /* Assume only end is not page aligned */ 368 u64 ramdisk_image = get_ramdisk_image(); 369 u64 ramdisk_size = get_ramdisk_size(); 370 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 371 u64 mapped_size; 372 373 if (!boot_params.hdr.type_of_loader || 374 !ramdisk_image || !ramdisk_size) 375 return; /* No initrd provided by bootloader */ 376 377 initrd_start = 0; 378 379 mapped_size = memblock_mem_size(max_pfn_mapped); 380 if (ramdisk_size >= (mapped_size>>1)) 381 panic("initrd too large to handle, " 382 "disabling initrd (%lld needed, %lld available)\n", 383 ramdisk_size, mapped_size>>1); 384 385 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 386 ramdisk_end - 1); 387 388 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 389 PFN_DOWN(ramdisk_end))) { 390 /* All are mapped, easy case */ 391 initrd_start = ramdisk_image + PAGE_OFFSET; 392 initrd_end = initrd_start + ramdisk_size; 393 return; 394 } 395 396 relocate_initrd(); 397 398 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 399 } 400 #else 401 static void __init early_reserve_initrd(void) 402 { 403 } 404 static void __init reserve_initrd(void) 405 { 406 } 407 #endif /* CONFIG_BLK_DEV_INITRD */ 408 409 static void __init parse_setup_data(void) 410 { 411 struct setup_data *data; 412 u64 pa_data, pa_next; 413 414 pa_data = boot_params.hdr.setup_data; 415 while (pa_data) { 416 u32 data_len, data_type; 417 418 data = early_memremap(pa_data, sizeof(*data)); 419 data_len = data->len + sizeof(struct setup_data); 420 data_type = data->type; 421 pa_next = data->next; 422 early_memunmap(data, sizeof(*data)); 423 424 switch (data_type) { 425 case SETUP_E820_EXT: 426 parse_e820_ext(pa_data, data_len); 427 break; 428 case SETUP_DTB: 429 add_dtb(pa_data); 430 break; 431 case SETUP_EFI: 432 parse_efi_setup(pa_data, data_len); 433 break; 434 default: 435 break; 436 } 437 pa_data = pa_next; 438 } 439 } 440 441 static void __init e820_reserve_setup_data(void) 442 { 443 struct setup_data *data; 444 u64 pa_data; 445 446 pa_data = boot_params.hdr.setup_data; 447 if (!pa_data) 448 return; 449 450 while (pa_data) { 451 data = early_memremap(pa_data, sizeof(*data)); 452 e820_update_range(pa_data, sizeof(*data)+data->len, 453 E820_RAM, E820_RESERVED_KERN); 454 pa_data = data->next; 455 early_memunmap(data, sizeof(*data)); 456 } 457 458 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 459 memcpy(&e820_saved, &e820, sizeof(struct e820map)); 460 printk(KERN_INFO "extended physical RAM map:\n"); 461 e820_print_map("reserve setup_data"); 462 } 463 464 static void __init memblock_x86_reserve_range_setup_data(void) 465 { 466 struct setup_data *data; 467 u64 pa_data; 468 469 pa_data = boot_params.hdr.setup_data; 470 while (pa_data) { 471 data = early_memremap(pa_data, sizeof(*data)); 472 memblock_reserve(pa_data, sizeof(*data) + data->len); 473 pa_data = data->next; 474 early_memunmap(data, sizeof(*data)); 475 } 476 } 477 478 /* 479 * --------- Crashkernel reservation ------------------------------ 480 */ 481 482 #ifdef CONFIG_KEXEC_CORE 483 484 /* 16M alignment for crash kernel regions */ 485 #define CRASH_ALIGN (16 << 20) 486 487 /* 488 * Keep the crash kernel below this limit. On 32 bits earlier kernels 489 * would limit the kernel to the low 512 MiB due to mapping restrictions. 490 * On 64bit, old kexec-tools need to under 896MiB. 491 */ 492 #ifdef CONFIG_X86_32 493 # define CRASH_ADDR_LOW_MAX (512 << 20) 494 # define CRASH_ADDR_HIGH_MAX (512 << 20) 495 #else 496 # define CRASH_ADDR_LOW_MAX (896UL << 20) 497 # define CRASH_ADDR_HIGH_MAX MAXMEM 498 #endif 499 500 static int __init reserve_crashkernel_low(void) 501 { 502 #ifdef CONFIG_X86_64 503 unsigned long long base, low_base = 0, low_size = 0; 504 unsigned long total_low_mem; 505 int ret; 506 507 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 508 509 /* crashkernel=Y,low */ 510 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 511 if (ret) { 512 /* 513 * two parts from lib/swiotlb.c: 514 * -swiotlb size: user-specified with swiotlb= or default. 515 * 516 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 517 * to 8M for other buffers that may need to stay low too. Also 518 * make sure we allocate enough extra low memory so that we 519 * don't run out of DMA buffers for 32-bit devices. 520 */ 521 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 522 } else { 523 /* passed with crashkernel=0,low ? */ 524 if (!low_size) 525 return 0; 526 } 527 528 low_base = memblock_find_in_range(low_size, 1ULL << 32, low_size, CRASH_ALIGN); 529 if (!low_base) { 530 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 531 (unsigned long)(low_size >> 20)); 532 return -ENOMEM; 533 } 534 535 ret = memblock_reserve(low_base, low_size); 536 if (ret) { 537 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 538 return ret; 539 } 540 541 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 542 (unsigned long)(low_size >> 20), 543 (unsigned long)(low_base >> 20), 544 (unsigned long)(total_low_mem >> 20)); 545 546 crashk_low_res.start = low_base; 547 crashk_low_res.end = low_base + low_size - 1; 548 insert_resource(&iomem_resource, &crashk_low_res); 549 #endif 550 return 0; 551 } 552 553 static void __init reserve_crashkernel(void) 554 { 555 unsigned long long crash_size, crash_base, total_mem; 556 bool high = false; 557 int ret; 558 559 total_mem = memblock_phys_mem_size(); 560 561 /* crashkernel=XM */ 562 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 563 if (ret != 0 || crash_size <= 0) { 564 /* crashkernel=X,high */ 565 ret = parse_crashkernel_high(boot_command_line, total_mem, 566 &crash_size, &crash_base); 567 if (ret != 0 || crash_size <= 0) 568 return; 569 high = true; 570 } 571 572 /* 0 means: find the address automatically */ 573 if (crash_base <= 0) { 574 /* 575 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 576 */ 577 crash_base = memblock_find_in_range(CRASH_ALIGN, 578 high ? CRASH_ADDR_HIGH_MAX 579 : CRASH_ADDR_LOW_MAX, 580 crash_size, CRASH_ALIGN); 581 if (!crash_base) { 582 pr_info("crashkernel reservation failed - No suitable area found.\n"); 583 return; 584 } 585 586 } else { 587 unsigned long long start; 588 589 start = memblock_find_in_range(crash_base, 590 crash_base + crash_size, 591 crash_size, 1 << 20); 592 if (start != crash_base) { 593 pr_info("crashkernel reservation failed - memory is in use.\n"); 594 return; 595 } 596 } 597 ret = memblock_reserve(crash_base, crash_size); 598 if (ret) { 599 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 600 return; 601 } 602 603 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 604 memblock_free(crash_base, crash_size); 605 return; 606 } 607 608 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 609 (unsigned long)(crash_size >> 20), 610 (unsigned long)(crash_base >> 20), 611 (unsigned long)(total_mem >> 20)); 612 613 crashk_res.start = crash_base; 614 crashk_res.end = crash_base + crash_size - 1; 615 insert_resource(&iomem_resource, &crashk_res); 616 } 617 #else 618 static void __init reserve_crashkernel(void) 619 { 620 } 621 #endif 622 623 static struct resource standard_io_resources[] = { 624 { .name = "dma1", .start = 0x00, .end = 0x1f, 625 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 626 { .name = "pic1", .start = 0x20, .end = 0x21, 627 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 628 { .name = "timer0", .start = 0x40, .end = 0x43, 629 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 630 { .name = "timer1", .start = 0x50, .end = 0x53, 631 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 632 { .name = "keyboard", .start = 0x60, .end = 0x60, 633 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 634 { .name = "keyboard", .start = 0x64, .end = 0x64, 635 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 636 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 637 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 638 { .name = "pic2", .start = 0xa0, .end = 0xa1, 639 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 640 { .name = "dma2", .start = 0xc0, .end = 0xdf, 641 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 642 { .name = "fpu", .start = 0xf0, .end = 0xff, 643 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 644 }; 645 646 void __init reserve_standard_io_resources(void) 647 { 648 int i; 649 650 /* request I/O space for devices used on all i[345]86 PCs */ 651 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 652 request_resource(&ioport_resource, &standard_io_resources[i]); 653 654 } 655 656 static __init void reserve_ibft_region(void) 657 { 658 unsigned long addr, size = 0; 659 660 addr = find_ibft_region(&size); 661 662 if (size) 663 memblock_reserve(addr, size); 664 } 665 666 static bool __init snb_gfx_workaround_needed(void) 667 { 668 #ifdef CONFIG_PCI 669 int i; 670 u16 vendor, devid; 671 static const __initconst u16 snb_ids[] = { 672 0x0102, 673 0x0112, 674 0x0122, 675 0x0106, 676 0x0116, 677 0x0126, 678 0x010a, 679 }; 680 681 /* Assume no if something weird is going on with PCI */ 682 if (!early_pci_allowed()) 683 return false; 684 685 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 686 if (vendor != 0x8086) 687 return false; 688 689 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 690 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 691 if (devid == snb_ids[i]) 692 return true; 693 #endif 694 695 return false; 696 } 697 698 /* 699 * Sandy Bridge graphics has trouble with certain ranges, exclude 700 * them from allocation. 701 */ 702 static void __init trim_snb_memory(void) 703 { 704 static const __initconst unsigned long bad_pages[] = { 705 0x20050000, 706 0x20110000, 707 0x20130000, 708 0x20138000, 709 0x40004000, 710 }; 711 int i; 712 713 if (!snb_gfx_workaround_needed()) 714 return; 715 716 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 717 718 /* 719 * Reserve all memory below the 1 MB mark that has not 720 * already been reserved. 721 */ 722 memblock_reserve(0, 1<<20); 723 724 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 725 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 726 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 727 bad_pages[i]); 728 } 729 } 730 731 /* 732 * Here we put platform-specific memory range workarounds, i.e. 733 * memory known to be corrupt or otherwise in need to be reserved on 734 * specific platforms. 735 * 736 * If this gets used more widely it could use a real dispatch mechanism. 737 */ 738 static void __init trim_platform_memory_ranges(void) 739 { 740 trim_snb_memory(); 741 } 742 743 static void __init trim_bios_range(void) 744 { 745 /* 746 * A special case is the first 4Kb of memory; 747 * This is a BIOS owned area, not kernel ram, but generally 748 * not listed as such in the E820 table. 749 * 750 * This typically reserves additional memory (64KiB by default) 751 * since some BIOSes are known to corrupt low memory. See the 752 * Kconfig help text for X86_RESERVE_LOW. 753 */ 754 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 755 756 /* 757 * special case: Some BIOSen report the PC BIOS 758 * area (640->1Mb) as ram even though it is not. 759 * take them out. 760 */ 761 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 762 763 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 764 } 765 766 /* called before trim_bios_range() to spare extra sanitize */ 767 static void __init e820_add_kernel_range(void) 768 { 769 u64 start = __pa_symbol(_text); 770 u64 size = __pa_symbol(_end) - start; 771 772 /* 773 * Complain if .text .data and .bss are not marked as E820_RAM and 774 * attempt to fix it by adding the range. We may have a confused BIOS, 775 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 776 * exclude kernel range. If we really are running on top non-RAM, 777 * we will crash later anyways. 778 */ 779 if (e820_all_mapped(start, start + size, E820_RAM)) 780 return; 781 782 pr_warn(".text .data .bss are not marked as E820_RAM!\n"); 783 e820_remove_range(start, size, E820_RAM, 0); 784 e820_add_region(start, size, E820_RAM); 785 } 786 787 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 788 789 static int __init parse_reservelow(char *p) 790 { 791 unsigned long long size; 792 793 if (!p) 794 return -EINVAL; 795 796 size = memparse(p, &p); 797 798 if (size < 4096) 799 size = 4096; 800 801 if (size > 640*1024) 802 size = 640*1024; 803 804 reserve_low = size; 805 806 return 0; 807 } 808 809 early_param("reservelow", parse_reservelow); 810 811 static void __init trim_low_memory_range(void) 812 { 813 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 814 } 815 816 /* 817 * Dump out kernel offset information on panic. 818 */ 819 static int 820 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 821 { 822 if (kaslr_enabled()) { 823 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 824 kaslr_offset(), 825 __START_KERNEL, 826 __START_KERNEL_map, 827 MODULES_VADDR-1); 828 } else { 829 pr_emerg("Kernel Offset: disabled\n"); 830 } 831 832 return 0; 833 } 834 835 /* 836 * Determine if we were loaded by an EFI loader. If so, then we have also been 837 * passed the efi memmap, systab, etc., so we should use these data structures 838 * for initialization. Note, the efi init code path is determined by the 839 * global efi_enabled. This allows the same kernel image to be used on existing 840 * systems (with a traditional BIOS) as well as on EFI systems. 841 */ 842 /* 843 * setup_arch - architecture-specific boot-time initializations 844 * 845 * Note: On x86_64, fixmaps are ready for use even before this is called. 846 */ 847 848 void __init setup_arch(char **cmdline_p) 849 { 850 memblock_reserve(__pa_symbol(_text), 851 (unsigned long)__bss_stop - (unsigned long)_text); 852 853 early_reserve_initrd(); 854 855 /* 856 * At this point everything still needed from the boot loader 857 * or BIOS or kernel text should be early reserved or marked not 858 * RAM in e820. All other memory is free game. 859 */ 860 861 #ifdef CONFIG_X86_32 862 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 863 864 /* 865 * copy kernel address range established so far and switch 866 * to the proper swapper page table 867 */ 868 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 869 initial_page_table + KERNEL_PGD_BOUNDARY, 870 KERNEL_PGD_PTRS); 871 872 load_cr3(swapper_pg_dir); 873 /* 874 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 875 * a cr3 based tlb flush, so the following __flush_tlb_all() 876 * will not flush anything because the cpu quirk which clears 877 * X86_FEATURE_PGE has not been invoked yet. Though due to the 878 * load_cr3() above the TLB has been flushed already. The 879 * quirk is invoked before subsequent calls to __flush_tlb_all() 880 * so proper operation is guaranteed. 881 */ 882 __flush_tlb_all(); 883 #else 884 printk(KERN_INFO "Command line: %s\n", boot_command_line); 885 #endif 886 887 /* 888 * If we have OLPC OFW, we might end up relocating the fixmap due to 889 * reserve_top(), so do this before touching the ioremap area. 890 */ 891 olpc_ofw_detect(); 892 893 early_trap_init(); 894 early_cpu_init(); 895 early_ioremap_init(); 896 897 setup_olpc_ofw_pgd(); 898 899 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 900 screen_info = boot_params.screen_info; 901 edid_info = boot_params.edid_info; 902 #ifdef CONFIG_X86_32 903 apm_info.bios = boot_params.apm_bios_info; 904 ist_info = boot_params.ist_info; 905 #endif 906 saved_video_mode = boot_params.hdr.vid_mode; 907 bootloader_type = boot_params.hdr.type_of_loader; 908 if ((bootloader_type >> 4) == 0xe) { 909 bootloader_type &= 0xf; 910 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 911 } 912 bootloader_version = bootloader_type & 0xf; 913 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 914 915 #ifdef CONFIG_BLK_DEV_RAM 916 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 917 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 918 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 919 #endif 920 #ifdef CONFIG_EFI 921 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 922 EFI32_LOADER_SIGNATURE, 4)) { 923 set_bit(EFI_BOOT, &efi.flags); 924 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 925 EFI64_LOADER_SIGNATURE, 4)) { 926 set_bit(EFI_BOOT, &efi.flags); 927 set_bit(EFI_64BIT, &efi.flags); 928 } 929 930 if (efi_enabled(EFI_BOOT)) 931 efi_memblock_x86_reserve_range(); 932 #endif 933 934 x86_init.oem.arch_setup(); 935 936 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 937 setup_memory_map(); 938 parse_setup_data(); 939 940 copy_edd(); 941 942 if (!boot_params.hdr.root_flags) 943 root_mountflags &= ~MS_RDONLY; 944 init_mm.start_code = (unsigned long) _text; 945 init_mm.end_code = (unsigned long) _etext; 946 init_mm.end_data = (unsigned long) _edata; 947 init_mm.brk = _brk_end; 948 949 mpx_mm_init(&init_mm); 950 951 code_resource.start = __pa_symbol(_text); 952 code_resource.end = __pa_symbol(_etext)-1; 953 data_resource.start = __pa_symbol(_etext); 954 data_resource.end = __pa_symbol(_edata)-1; 955 bss_resource.start = __pa_symbol(__bss_start); 956 bss_resource.end = __pa_symbol(__bss_stop)-1; 957 958 #ifdef CONFIG_CMDLINE_BOOL 959 #ifdef CONFIG_CMDLINE_OVERRIDE 960 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 961 #else 962 if (builtin_cmdline[0]) { 963 /* append boot loader cmdline to builtin */ 964 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 965 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 966 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 967 } 968 #endif 969 #endif 970 971 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 972 *cmdline_p = command_line; 973 974 /* 975 * x86_configure_nx() is called before parse_early_param() to detect 976 * whether hardware doesn't support NX (so that the early EHCI debug 977 * console setup can safely call set_fixmap()). It may then be called 978 * again from within noexec_setup() during parsing early parameters 979 * to honor the respective command line option. 980 */ 981 x86_configure_nx(); 982 983 parse_early_param(); 984 985 x86_report_nx(); 986 987 /* after early param, so could get panic from serial */ 988 memblock_x86_reserve_range_setup_data(); 989 990 if (acpi_mps_check()) { 991 #ifdef CONFIG_X86_LOCAL_APIC 992 disable_apic = 1; 993 #endif 994 setup_clear_cpu_cap(X86_FEATURE_APIC); 995 } 996 997 #ifdef CONFIG_PCI 998 if (pci_early_dump_regs) 999 early_dump_pci_devices(); 1000 #endif 1001 1002 /* update the e820_saved too */ 1003 e820_reserve_setup_data(); 1004 finish_e820_parsing(); 1005 1006 if (efi_enabled(EFI_BOOT)) 1007 efi_init(); 1008 1009 dmi_scan_machine(); 1010 dmi_memdev_walk(); 1011 dmi_set_dump_stack_arch_desc(); 1012 1013 /* 1014 * VMware detection requires dmi to be available, so this 1015 * needs to be done after dmi_scan_machine, for the BP. 1016 */ 1017 init_hypervisor_platform(); 1018 1019 x86_init.resources.probe_roms(); 1020 1021 /* after parse_early_param, so could debug it */ 1022 insert_resource(&iomem_resource, &code_resource); 1023 insert_resource(&iomem_resource, &data_resource); 1024 insert_resource(&iomem_resource, &bss_resource); 1025 1026 e820_add_kernel_range(); 1027 trim_bios_range(); 1028 #ifdef CONFIG_X86_32 1029 if (ppro_with_ram_bug()) { 1030 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 1031 E820_RESERVED); 1032 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 1033 printk(KERN_INFO "fixed physical RAM map:\n"); 1034 e820_print_map("bad_ppro"); 1035 } 1036 #else 1037 early_gart_iommu_check(); 1038 #endif 1039 1040 /* 1041 * partially used pages are not usable - thus 1042 * we are rounding upwards: 1043 */ 1044 max_pfn = e820_end_of_ram_pfn(); 1045 1046 /* update e820 for memory not covered by WB MTRRs */ 1047 mtrr_bp_init(); 1048 if (mtrr_trim_uncached_memory(max_pfn)) 1049 max_pfn = e820_end_of_ram_pfn(); 1050 1051 #ifdef CONFIG_X86_32 1052 /* max_low_pfn get updated here */ 1053 find_low_pfn_range(); 1054 #else 1055 check_x2apic(); 1056 1057 /* How many end-of-memory variables you have, grandma! */ 1058 /* need this before calling reserve_initrd */ 1059 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1060 max_low_pfn = e820_end_of_low_ram_pfn(); 1061 else 1062 max_low_pfn = max_pfn; 1063 1064 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1065 #endif 1066 1067 /* 1068 * Find and reserve possible boot-time SMP configuration: 1069 */ 1070 find_smp_config(); 1071 1072 reserve_ibft_region(); 1073 1074 early_alloc_pgt_buf(); 1075 1076 /* 1077 * Need to conclude brk, before memblock_x86_fill() 1078 * it could use memblock_find_in_range, could overlap with 1079 * brk area. 1080 */ 1081 reserve_brk(); 1082 1083 cleanup_highmap(); 1084 1085 memblock_set_current_limit(ISA_END_ADDRESS); 1086 memblock_x86_fill(); 1087 1088 if (efi_enabled(EFI_BOOT)) { 1089 efi_fake_memmap(); 1090 efi_find_mirror(); 1091 } 1092 1093 /* 1094 * The EFI specification says that boot service code won't be called 1095 * after ExitBootServices(). This is, in fact, a lie. 1096 */ 1097 if (efi_enabled(EFI_MEMMAP)) 1098 efi_reserve_boot_services(); 1099 1100 /* preallocate 4k for mptable mpc */ 1101 early_reserve_e820_mpc_new(); 1102 1103 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1104 setup_bios_corruption_check(); 1105 #endif 1106 1107 #ifdef CONFIG_X86_32 1108 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1109 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1110 #endif 1111 1112 reserve_real_mode(); 1113 1114 trim_platform_memory_ranges(); 1115 trim_low_memory_range(); 1116 1117 init_mem_mapping(); 1118 1119 early_trap_pf_init(); 1120 1121 setup_real_mode(); 1122 1123 memblock_set_current_limit(get_max_mapped()); 1124 1125 /* 1126 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1127 */ 1128 1129 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1130 if (init_ohci1394_dma_early) 1131 init_ohci1394_dma_on_all_controllers(); 1132 #endif 1133 /* Allocate bigger log buffer */ 1134 setup_log_buf(1); 1135 1136 reserve_initrd(); 1137 1138 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD) 1139 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start); 1140 #endif 1141 1142 vsmp_init(); 1143 1144 io_delay_init(); 1145 1146 /* 1147 * Parse the ACPI tables for possible boot-time SMP configuration. 1148 */ 1149 acpi_boot_table_init(); 1150 1151 early_acpi_boot_init(); 1152 1153 initmem_init(); 1154 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1155 1156 /* 1157 * Reserve memory for crash kernel after SRAT is parsed so that it 1158 * won't consume hotpluggable memory. 1159 */ 1160 reserve_crashkernel(); 1161 1162 memblock_find_dma_reserve(); 1163 1164 #ifdef CONFIG_KVM_GUEST 1165 kvmclock_init(); 1166 #endif 1167 1168 x86_init.paging.pagetable_init(); 1169 1170 kasan_init(); 1171 1172 if (boot_cpu_data.cpuid_level >= 0) { 1173 /* A CPU has %cr4 if and only if it has CPUID */ 1174 mmu_cr4_features = __read_cr4(); 1175 if (trampoline_cr4_features) 1176 *trampoline_cr4_features = mmu_cr4_features; 1177 } 1178 1179 #ifdef CONFIG_X86_32 1180 /* sync back kernel address range */ 1181 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1182 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1183 KERNEL_PGD_PTRS); 1184 1185 /* 1186 * sync back low identity map too. It is used for example 1187 * in the 32-bit EFI stub. 1188 */ 1189 clone_pgd_range(initial_page_table, 1190 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1191 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 1192 #endif 1193 1194 tboot_probe(); 1195 1196 map_vsyscall(); 1197 1198 generic_apic_probe(); 1199 1200 early_quirks(); 1201 1202 /* 1203 * Read APIC and some other early information from ACPI tables. 1204 */ 1205 acpi_boot_init(); 1206 sfi_init(); 1207 x86_dtb_init(); 1208 1209 /* 1210 * get boot-time SMP configuration: 1211 */ 1212 if (smp_found_config) 1213 get_smp_config(); 1214 1215 prefill_possible_map(); 1216 1217 init_cpu_to_node(); 1218 1219 init_apic_mappings(); 1220 io_apic_init_mappings(); 1221 1222 kvm_guest_init(); 1223 1224 e820_reserve_resources(); 1225 e820_mark_nosave_regions(max_low_pfn); 1226 1227 x86_init.resources.reserve_resources(); 1228 1229 e820_setup_gap(); 1230 1231 #ifdef CONFIG_VT 1232 #if defined(CONFIG_VGA_CONSOLE) 1233 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1234 conswitchp = &vga_con; 1235 #elif defined(CONFIG_DUMMY_CONSOLE) 1236 conswitchp = &dummy_con; 1237 #endif 1238 #endif 1239 x86_init.oem.banner(); 1240 1241 x86_init.timers.wallclock_init(); 1242 1243 mcheck_init(); 1244 1245 arch_init_ideal_nops(); 1246 1247 register_refined_jiffies(CLOCK_TICK_RATE); 1248 1249 #ifdef CONFIG_EFI 1250 if (efi_enabled(EFI_BOOT)) 1251 efi_apply_memmap_quirks(); 1252 #endif 1253 1254 microcode_init(); 1255 } 1256 1257 #ifdef CONFIG_X86_32 1258 1259 static struct resource video_ram_resource = { 1260 .name = "Video RAM area", 1261 .start = 0xa0000, 1262 .end = 0xbffff, 1263 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1264 }; 1265 1266 void __init i386_reserve_resources(void) 1267 { 1268 request_resource(&iomem_resource, &video_ram_resource); 1269 reserve_standard_io_resources(); 1270 } 1271 1272 #endif /* CONFIG_X86_32 */ 1273 1274 static struct notifier_block kernel_offset_notifier = { 1275 .notifier_call = dump_kernel_offset 1276 }; 1277 1278 static int __init register_kernel_offset_dumper(void) 1279 { 1280 atomic_notifier_chain_register(&panic_notifier_list, 1281 &kernel_offset_notifier); 1282 return 0; 1283 } 1284 __initcall(register_kernel_offset_dumper); 1285