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 */ 122 unsigned long max_low_pfn_mapped; 123 unsigned long max_pfn_mapped; 124 125 bool __read_mostly kaslr_enabled = false; 126 127 #ifdef CONFIG_DMI 128 RESERVE_BRK(dmi_alloc, 65536); 129 #endif 130 131 132 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 133 unsigned long _brk_end = (unsigned long)__brk_base; 134 135 #ifdef CONFIG_X86_64 136 int default_cpu_present_to_apicid(int mps_cpu) 137 { 138 return __default_cpu_present_to_apicid(mps_cpu); 139 } 140 141 int default_check_phys_apicid_present(int phys_apicid) 142 { 143 return __default_check_phys_apicid_present(phys_apicid); 144 } 145 #endif 146 147 struct boot_params boot_params; 148 149 /* 150 * Machine setup.. 151 */ 152 static struct resource data_resource = { 153 .name = "Kernel data", 154 .start = 0, 155 .end = 0, 156 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 157 }; 158 159 static struct resource code_resource = { 160 .name = "Kernel code", 161 .start = 0, 162 .end = 0, 163 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 164 }; 165 166 static struct resource bss_resource = { 167 .name = "Kernel bss", 168 .start = 0, 169 .end = 0, 170 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 171 }; 172 173 174 #ifdef CONFIG_X86_32 175 /* cpu data as detected by the assembly code in head.S */ 176 struct cpuinfo_x86 new_cpu_data = { 177 .wp_works_ok = -1, 178 }; 179 /* common cpu data for all cpus */ 180 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 181 .wp_works_ok = -1, 182 }; 183 EXPORT_SYMBOL(boot_cpu_data); 184 185 unsigned int def_to_bigsmp; 186 187 /* for MCA, but anyone else can use it if they want */ 188 unsigned int machine_id; 189 unsigned int machine_submodel_id; 190 unsigned int BIOS_revision; 191 192 struct apm_info apm_info; 193 EXPORT_SYMBOL(apm_info); 194 195 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 196 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 197 struct ist_info ist_info; 198 EXPORT_SYMBOL(ist_info); 199 #else 200 struct ist_info ist_info; 201 #endif 202 203 #else 204 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 205 .x86_phys_bits = MAX_PHYSMEM_BITS, 206 }; 207 EXPORT_SYMBOL(boot_cpu_data); 208 #endif 209 210 211 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 212 __visible unsigned long mmu_cr4_features; 213 #else 214 __visible unsigned long mmu_cr4_features = X86_CR4_PAE; 215 #endif 216 217 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 218 int bootloader_type, bootloader_version; 219 220 /* 221 * Setup options 222 */ 223 struct screen_info screen_info; 224 EXPORT_SYMBOL(screen_info); 225 struct edid_info edid_info; 226 EXPORT_SYMBOL_GPL(edid_info); 227 228 extern int root_mountflags; 229 230 unsigned long saved_video_mode; 231 232 #define RAMDISK_IMAGE_START_MASK 0x07FF 233 #define RAMDISK_PROMPT_FLAG 0x8000 234 #define RAMDISK_LOAD_FLAG 0x4000 235 236 static char __initdata command_line[COMMAND_LINE_SIZE]; 237 #ifdef CONFIG_CMDLINE_BOOL 238 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 239 #endif 240 241 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 242 struct edd edd; 243 #ifdef CONFIG_EDD_MODULE 244 EXPORT_SYMBOL(edd); 245 #endif 246 /** 247 * copy_edd() - Copy the BIOS EDD information 248 * from boot_params into a safe place. 249 * 250 */ 251 static inline void __init copy_edd(void) 252 { 253 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 254 sizeof(edd.mbr_signature)); 255 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 256 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 257 edd.edd_info_nr = boot_params.eddbuf_entries; 258 } 259 #else 260 static inline void __init copy_edd(void) 261 { 262 } 263 #endif 264 265 void * __init extend_brk(size_t size, size_t align) 266 { 267 size_t mask = align - 1; 268 void *ret; 269 270 BUG_ON(_brk_start == 0); 271 BUG_ON(align & mask); 272 273 _brk_end = (_brk_end + mask) & ~mask; 274 BUG_ON((char *)(_brk_end + size) > __brk_limit); 275 276 ret = (void *)_brk_end; 277 _brk_end += size; 278 279 memset(ret, 0, size); 280 281 return ret; 282 } 283 284 #ifdef CONFIG_X86_32 285 static void __init cleanup_highmap(void) 286 { 287 } 288 #endif 289 290 static void __init reserve_brk(void) 291 { 292 if (_brk_end > _brk_start) 293 memblock_reserve(__pa_symbol(_brk_start), 294 _brk_end - _brk_start); 295 296 /* Mark brk area as locked down and no longer taking any 297 new allocations */ 298 _brk_start = 0; 299 } 300 301 u64 relocated_ramdisk; 302 303 #ifdef CONFIG_BLK_DEV_INITRD 304 305 static u64 __init get_ramdisk_image(void) 306 { 307 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 308 309 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 310 311 return ramdisk_image; 312 } 313 static u64 __init get_ramdisk_size(void) 314 { 315 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 316 317 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 318 319 return ramdisk_size; 320 } 321 322 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT) 323 static void __init relocate_initrd(void) 324 { 325 /* Assume only end is not page aligned */ 326 u64 ramdisk_image = get_ramdisk_image(); 327 u64 ramdisk_size = get_ramdisk_size(); 328 u64 area_size = PAGE_ALIGN(ramdisk_size); 329 unsigned long slop, clen, mapaddr; 330 char *p, *q; 331 332 /* We need to move the initrd down into directly mapped mem */ 333 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 334 area_size, PAGE_SIZE); 335 336 if (!relocated_ramdisk) 337 panic("Cannot find place for new RAMDISK of size %lld\n", 338 ramdisk_size); 339 340 /* Note: this includes all the mem currently occupied by 341 the initrd, we rely on that fact to keep the data intact. */ 342 memblock_reserve(relocated_ramdisk, area_size); 343 initrd_start = relocated_ramdisk + PAGE_OFFSET; 344 initrd_end = initrd_start + ramdisk_size; 345 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 346 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 347 348 q = (char *)initrd_start; 349 350 /* Copy the initrd */ 351 while (ramdisk_size) { 352 slop = ramdisk_image & ~PAGE_MASK; 353 clen = ramdisk_size; 354 if (clen > MAX_MAP_CHUNK-slop) 355 clen = MAX_MAP_CHUNK-slop; 356 mapaddr = ramdisk_image & PAGE_MASK; 357 p = early_memremap(mapaddr, clen+slop); 358 memcpy(q, p+slop, clen); 359 early_iounmap(p, clen+slop); 360 q += clen; 361 ramdisk_image += clen; 362 ramdisk_size -= clen; 363 } 364 365 ramdisk_image = get_ramdisk_image(); 366 ramdisk_size = get_ramdisk_size(); 367 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 368 " [mem %#010llx-%#010llx]\n", 369 ramdisk_image, ramdisk_image + ramdisk_size - 1, 370 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 371 } 372 373 static void __init early_reserve_initrd(void) 374 { 375 /* Assume only end is not page aligned */ 376 u64 ramdisk_image = get_ramdisk_image(); 377 u64 ramdisk_size = get_ramdisk_size(); 378 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 379 380 if (!boot_params.hdr.type_of_loader || 381 !ramdisk_image || !ramdisk_size) 382 return; /* No initrd provided by bootloader */ 383 384 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 385 } 386 static void __init reserve_initrd(void) 387 { 388 /* Assume only end is not page aligned */ 389 u64 ramdisk_image = get_ramdisk_image(); 390 u64 ramdisk_size = get_ramdisk_size(); 391 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 392 u64 mapped_size; 393 394 if (!boot_params.hdr.type_of_loader || 395 !ramdisk_image || !ramdisk_size) 396 return; /* No initrd provided by bootloader */ 397 398 initrd_start = 0; 399 400 mapped_size = memblock_mem_size(max_pfn_mapped); 401 if (ramdisk_size >= (mapped_size>>1)) 402 panic("initrd too large to handle, " 403 "disabling initrd (%lld needed, %lld available)\n", 404 ramdisk_size, mapped_size>>1); 405 406 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 407 ramdisk_end - 1); 408 409 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 410 PFN_DOWN(ramdisk_end))) { 411 /* All are mapped, easy case */ 412 initrd_start = ramdisk_image + PAGE_OFFSET; 413 initrd_end = initrd_start + ramdisk_size; 414 return; 415 } 416 417 relocate_initrd(); 418 419 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 420 } 421 #else 422 static void __init early_reserve_initrd(void) 423 { 424 } 425 static void __init reserve_initrd(void) 426 { 427 } 428 #endif /* CONFIG_BLK_DEV_INITRD */ 429 430 static void __init parse_kaslr_setup(u64 pa_data, u32 data_len) 431 { 432 kaslr_enabled = (bool)(pa_data + sizeof(struct setup_data)); 433 } 434 435 static void __init parse_setup_data(void) 436 { 437 struct setup_data *data; 438 u64 pa_data, pa_next; 439 440 pa_data = boot_params.hdr.setup_data; 441 while (pa_data) { 442 u32 data_len, data_type; 443 444 data = early_memremap(pa_data, sizeof(*data)); 445 data_len = data->len + sizeof(struct setup_data); 446 data_type = data->type; 447 pa_next = data->next; 448 early_iounmap(data, sizeof(*data)); 449 450 switch (data_type) { 451 case SETUP_E820_EXT: 452 parse_e820_ext(pa_data, data_len); 453 break; 454 case SETUP_DTB: 455 add_dtb(pa_data); 456 break; 457 case SETUP_EFI: 458 parse_efi_setup(pa_data, data_len); 459 break; 460 case SETUP_KASLR: 461 parse_kaslr_setup(pa_data, data_len); 462 break; 463 default: 464 break; 465 } 466 pa_data = pa_next; 467 } 468 } 469 470 static void __init e820_reserve_setup_data(void) 471 { 472 struct setup_data *data; 473 u64 pa_data; 474 int found = 0; 475 476 pa_data = boot_params.hdr.setup_data; 477 while (pa_data) { 478 data = early_memremap(pa_data, sizeof(*data)); 479 e820_update_range(pa_data, sizeof(*data)+data->len, 480 E820_RAM, E820_RESERVED_KERN); 481 found = 1; 482 pa_data = data->next; 483 early_iounmap(data, sizeof(*data)); 484 } 485 if (!found) 486 return; 487 488 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 489 memcpy(&e820_saved, &e820, sizeof(struct e820map)); 490 printk(KERN_INFO "extended physical RAM map:\n"); 491 e820_print_map("reserve setup_data"); 492 } 493 494 static void __init memblock_x86_reserve_range_setup_data(void) 495 { 496 struct setup_data *data; 497 u64 pa_data; 498 499 pa_data = boot_params.hdr.setup_data; 500 while (pa_data) { 501 data = early_memremap(pa_data, sizeof(*data)); 502 memblock_reserve(pa_data, sizeof(*data) + data->len); 503 pa_data = data->next; 504 early_iounmap(data, sizeof(*data)); 505 } 506 } 507 508 /* 509 * --------- Crashkernel reservation ------------------------------ 510 */ 511 512 #ifdef CONFIG_KEXEC 513 514 /* 515 * Keep the crash kernel below this limit. On 32 bits earlier kernels 516 * would limit the kernel to the low 512 MiB due to mapping restrictions. 517 * On 64bit, old kexec-tools need to under 896MiB. 518 */ 519 #ifdef CONFIG_X86_32 520 # define CRASH_KERNEL_ADDR_LOW_MAX (512 << 20) 521 # define CRASH_KERNEL_ADDR_HIGH_MAX (512 << 20) 522 #else 523 # define CRASH_KERNEL_ADDR_LOW_MAX (896UL<<20) 524 # define CRASH_KERNEL_ADDR_HIGH_MAX MAXMEM 525 #endif 526 527 static void __init reserve_crashkernel_low(void) 528 { 529 #ifdef CONFIG_X86_64 530 const unsigned long long alignment = 16<<20; /* 16M */ 531 unsigned long long low_base = 0, low_size = 0; 532 unsigned long total_low_mem; 533 unsigned long long base; 534 bool auto_set = false; 535 int ret; 536 537 total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT)); 538 /* crashkernel=Y,low */ 539 ret = parse_crashkernel_low(boot_command_line, total_low_mem, 540 &low_size, &base); 541 if (ret != 0) { 542 /* 543 * two parts from lib/swiotlb.c: 544 * swiotlb size: user specified with swiotlb= or default. 545 * swiotlb overflow buffer: now is hardcoded to 32k. 546 * We round it to 8M for other buffers that 547 * may need to stay low too. 548 */ 549 low_size = swiotlb_size_or_default() + (8UL<<20); 550 auto_set = true; 551 } else { 552 /* passed with crashkernel=0,low ? */ 553 if (!low_size) 554 return; 555 } 556 557 low_base = memblock_find_in_range(low_size, (1ULL<<32), 558 low_size, alignment); 559 560 if (!low_base) { 561 if (!auto_set) 562 pr_info("crashkernel low reservation failed - No suitable area found.\n"); 563 564 return; 565 } 566 567 memblock_reserve(low_base, low_size); 568 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 569 (unsigned long)(low_size >> 20), 570 (unsigned long)(low_base >> 20), 571 (unsigned long)(total_low_mem >> 20)); 572 crashk_low_res.start = low_base; 573 crashk_low_res.end = low_base + low_size - 1; 574 insert_resource(&iomem_resource, &crashk_low_res); 575 #endif 576 } 577 578 static void __init reserve_crashkernel(void) 579 { 580 const unsigned long long alignment = 16<<20; /* 16M */ 581 unsigned long long total_mem; 582 unsigned long long crash_size, crash_base; 583 bool high = false; 584 int ret; 585 586 total_mem = memblock_phys_mem_size(); 587 588 /* crashkernel=XM */ 589 ret = parse_crashkernel(boot_command_line, total_mem, 590 &crash_size, &crash_base); 591 if (ret != 0 || crash_size <= 0) { 592 /* crashkernel=X,high */ 593 ret = parse_crashkernel_high(boot_command_line, total_mem, 594 &crash_size, &crash_base); 595 if (ret != 0 || crash_size <= 0) 596 return; 597 high = true; 598 } 599 600 /* 0 means: find the address automatically */ 601 if (crash_base <= 0) { 602 /* 603 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 604 */ 605 crash_base = memblock_find_in_range(alignment, 606 high ? CRASH_KERNEL_ADDR_HIGH_MAX : 607 CRASH_KERNEL_ADDR_LOW_MAX, 608 crash_size, alignment); 609 610 if (!crash_base) { 611 pr_info("crashkernel reservation failed - No suitable area found.\n"); 612 return; 613 } 614 615 } else { 616 unsigned long long start; 617 618 start = memblock_find_in_range(crash_base, 619 crash_base + crash_size, crash_size, 1<<20); 620 if (start != crash_base) { 621 pr_info("crashkernel reservation failed - memory is in use.\n"); 622 return; 623 } 624 } 625 memblock_reserve(crash_base, crash_size); 626 627 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " 628 "for crashkernel (System RAM: %ldMB)\n", 629 (unsigned long)(crash_size >> 20), 630 (unsigned long)(crash_base >> 20), 631 (unsigned long)(total_mem >> 20)); 632 633 crashk_res.start = crash_base; 634 crashk_res.end = crash_base + crash_size - 1; 635 insert_resource(&iomem_resource, &crashk_res); 636 637 if (crash_base >= (1ULL<<32)) 638 reserve_crashkernel_low(); 639 } 640 #else 641 static void __init reserve_crashkernel(void) 642 { 643 } 644 #endif 645 646 static struct resource standard_io_resources[] = { 647 { .name = "dma1", .start = 0x00, .end = 0x1f, 648 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 649 { .name = "pic1", .start = 0x20, .end = 0x21, 650 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 651 { .name = "timer0", .start = 0x40, .end = 0x43, 652 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 653 { .name = "timer1", .start = 0x50, .end = 0x53, 654 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 655 { .name = "keyboard", .start = 0x60, .end = 0x60, 656 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 657 { .name = "keyboard", .start = 0x64, .end = 0x64, 658 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 659 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 660 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 661 { .name = "pic2", .start = 0xa0, .end = 0xa1, 662 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 663 { .name = "dma2", .start = 0xc0, .end = 0xdf, 664 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 665 { .name = "fpu", .start = 0xf0, .end = 0xff, 666 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 667 }; 668 669 void __init reserve_standard_io_resources(void) 670 { 671 int i; 672 673 /* request I/O space for devices used on all i[345]86 PCs */ 674 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 675 request_resource(&ioport_resource, &standard_io_resources[i]); 676 677 } 678 679 static __init void reserve_ibft_region(void) 680 { 681 unsigned long addr, size = 0; 682 683 addr = find_ibft_region(&size); 684 685 if (size) 686 memblock_reserve(addr, size); 687 } 688 689 static bool __init snb_gfx_workaround_needed(void) 690 { 691 #ifdef CONFIG_PCI 692 int i; 693 u16 vendor, devid; 694 static const __initconst u16 snb_ids[] = { 695 0x0102, 696 0x0112, 697 0x0122, 698 0x0106, 699 0x0116, 700 0x0126, 701 0x010a, 702 }; 703 704 /* Assume no if something weird is going on with PCI */ 705 if (!early_pci_allowed()) 706 return false; 707 708 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 709 if (vendor != 0x8086) 710 return false; 711 712 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 713 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 714 if (devid == snb_ids[i]) 715 return true; 716 #endif 717 718 return false; 719 } 720 721 /* 722 * Sandy Bridge graphics has trouble with certain ranges, exclude 723 * them from allocation. 724 */ 725 static void __init trim_snb_memory(void) 726 { 727 static const __initconst unsigned long bad_pages[] = { 728 0x20050000, 729 0x20110000, 730 0x20130000, 731 0x20138000, 732 0x40004000, 733 }; 734 int i; 735 736 if (!snb_gfx_workaround_needed()) 737 return; 738 739 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 740 741 /* 742 * Reserve all memory below the 1 MB mark that has not 743 * already been reserved. 744 */ 745 memblock_reserve(0, 1<<20); 746 747 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 748 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 749 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 750 bad_pages[i]); 751 } 752 } 753 754 /* 755 * Here we put platform-specific memory range workarounds, i.e. 756 * memory known to be corrupt or otherwise in need to be reserved on 757 * specific platforms. 758 * 759 * If this gets used more widely it could use a real dispatch mechanism. 760 */ 761 static void __init trim_platform_memory_ranges(void) 762 { 763 trim_snb_memory(); 764 } 765 766 static void __init trim_bios_range(void) 767 { 768 /* 769 * A special case is the first 4Kb of memory; 770 * This is a BIOS owned area, not kernel ram, but generally 771 * not listed as such in the E820 table. 772 * 773 * This typically reserves additional memory (64KiB by default) 774 * since some BIOSes are known to corrupt low memory. See the 775 * Kconfig help text for X86_RESERVE_LOW. 776 */ 777 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 778 779 /* 780 * special case: Some BIOSen report the PC BIOS 781 * area (640->1Mb) as ram even though it is not. 782 * take them out. 783 */ 784 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 785 786 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 787 } 788 789 /* called before trim_bios_range() to spare extra sanitize */ 790 static void __init e820_add_kernel_range(void) 791 { 792 u64 start = __pa_symbol(_text); 793 u64 size = __pa_symbol(_end) - start; 794 795 /* 796 * Complain if .text .data and .bss are not marked as E820_RAM and 797 * attempt to fix it by adding the range. We may have a confused BIOS, 798 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 799 * exclude kernel range. If we really are running on top non-RAM, 800 * we will crash later anyways. 801 */ 802 if (e820_all_mapped(start, start + size, E820_RAM)) 803 return; 804 805 pr_warn(".text .data .bss are not marked as E820_RAM!\n"); 806 e820_remove_range(start, size, E820_RAM, 0); 807 e820_add_region(start, size, E820_RAM); 808 } 809 810 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 811 812 static int __init parse_reservelow(char *p) 813 { 814 unsigned long long size; 815 816 if (!p) 817 return -EINVAL; 818 819 size = memparse(p, &p); 820 821 if (size < 4096) 822 size = 4096; 823 824 if (size > 640*1024) 825 size = 640*1024; 826 827 reserve_low = size; 828 829 return 0; 830 } 831 832 early_param("reservelow", parse_reservelow); 833 834 static void __init trim_low_memory_range(void) 835 { 836 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 837 } 838 839 /* 840 * Dump out kernel offset information on panic. 841 */ 842 static int 843 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 844 { 845 if (kaslr_enabled) 846 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 847 (unsigned long)&_text - __START_KERNEL, 848 __START_KERNEL, 849 __START_KERNEL_map, 850 MODULES_VADDR-1); 851 else 852 pr_emerg("Kernel Offset: disabled\n"); 853 854 return 0; 855 } 856 857 /* 858 * Determine if we were loaded by an EFI loader. If so, then we have also been 859 * passed the efi memmap, systab, etc., so we should use these data structures 860 * for initialization. Note, the efi init code path is determined by the 861 * global efi_enabled. This allows the same kernel image to be used on existing 862 * systems (with a traditional BIOS) as well as on EFI systems. 863 */ 864 /* 865 * setup_arch - architecture-specific boot-time initializations 866 * 867 * Note: On x86_64, fixmaps are ready for use even before this is called. 868 */ 869 870 void __init setup_arch(char **cmdline_p) 871 { 872 memblock_reserve(__pa_symbol(_text), 873 (unsigned long)__bss_stop - (unsigned long)_text); 874 875 early_reserve_initrd(); 876 877 /* 878 * At this point everything still needed from the boot loader 879 * or BIOS or kernel text should be early reserved or marked not 880 * RAM in e820. All other memory is free game. 881 */ 882 883 #ifdef CONFIG_X86_32 884 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 885 886 /* 887 * copy kernel address range established so far and switch 888 * to the proper swapper page table 889 */ 890 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 891 initial_page_table + KERNEL_PGD_BOUNDARY, 892 KERNEL_PGD_PTRS); 893 894 load_cr3(swapper_pg_dir); 895 /* 896 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 897 * a cr3 based tlb flush, so the following __flush_tlb_all() 898 * will not flush anything because the cpu quirk which clears 899 * X86_FEATURE_PGE has not been invoked yet. Though due to the 900 * load_cr3() above the TLB has been flushed already. The 901 * quirk is invoked before subsequent calls to __flush_tlb_all() 902 * so proper operation is guaranteed. 903 */ 904 __flush_tlb_all(); 905 #else 906 printk(KERN_INFO "Command line: %s\n", boot_command_line); 907 #endif 908 909 /* 910 * If we have OLPC OFW, we might end up relocating the fixmap due to 911 * reserve_top(), so do this before touching the ioremap area. 912 */ 913 olpc_ofw_detect(); 914 915 early_trap_init(); 916 early_cpu_init(); 917 early_ioremap_init(); 918 919 setup_olpc_ofw_pgd(); 920 921 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 922 screen_info = boot_params.screen_info; 923 edid_info = boot_params.edid_info; 924 #ifdef CONFIG_X86_32 925 apm_info.bios = boot_params.apm_bios_info; 926 ist_info = boot_params.ist_info; 927 if (boot_params.sys_desc_table.length != 0) { 928 machine_id = boot_params.sys_desc_table.table[0]; 929 machine_submodel_id = boot_params.sys_desc_table.table[1]; 930 BIOS_revision = boot_params.sys_desc_table.table[2]; 931 } 932 #endif 933 saved_video_mode = boot_params.hdr.vid_mode; 934 bootloader_type = boot_params.hdr.type_of_loader; 935 if ((bootloader_type >> 4) == 0xe) { 936 bootloader_type &= 0xf; 937 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 938 } 939 bootloader_version = bootloader_type & 0xf; 940 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 941 942 #ifdef CONFIG_BLK_DEV_RAM 943 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 944 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 945 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 946 #endif 947 #ifdef CONFIG_EFI 948 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 949 EFI32_LOADER_SIGNATURE, 4)) { 950 set_bit(EFI_BOOT, &efi.flags); 951 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 952 EFI64_LOADER_SIGNATURE, 4)) { 953 set_bit(EFI_BOOT, &efi.flags); 954 set_bit(EFI_64BIT, &efi.flags); 955 } 956 957 if (efi_enabled(EFI_BOOT)) 958 efi_memblock_x86_reserve_range(); 959 #endif 960 961 x86_init.oem.arch_setup(); 962 963 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 964 setup_memory_map(); 965 parse_setup_data(); 966 967 copy_edd(); 968 969 if (!boot_params.hdr.root_flags) 970 root_mountflags &= ~MS_RDONLY; 971 init_mm.start_code = (unsigned long) _text; 972 init_mm.end_code = (unsigned long) _etext; 973 init_mm.end_data = (unsigned long) _edata; 974 init_mm.brk = _brk_end; 975 976 mpx_mm_init(&init_mm); 977 978 code_resource.start = __pa_symbol(_text); 979 code_resource.end = __pa_symbol(_etext)-1; 980 data_resource.start = __pa_symbol(_etext); 981 data_resource.end = __pa_symbol(_edata)-1; 982 bss_resource.start = __pa_symbol(__bss_start); 983 bss_resource.end = __pa_symbol(__bss_stop)-1; 984 985 #ifdef CONFIG_CMDLINE_BOOL 986 #ifdef CONFIG_CMDLINE_OVERRIDE 987 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 988 #else 989 if (builtin_cmdline[0]) { 990 /* append boot loader cmdline to builtin */ 991 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 992 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 993 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 994 } 995 #endif 996 #endif 997 998 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 999 *cmdline_p = command_line; 1000 1001 /* 1002 * x86_configure_nx() is called before parse_early_param() to detect 1003 * whether hardware doesn't support NX (so that the early EHCI debug 1004 * console setup can safely call set_fixmap()). It may then be called 1005 * again from within noexec_setup() during parsing early parameters 1006 * to honor the respective command line option. 1007 */ 1008 x86_configure_nx(); 1009 1010 parse_early_param(); 1011 1012 x86_report_nx(); 1013 1014 /* after early param, so could get panic from serial */ 1015 memblock_x86_reserve_range_setup_data(); 1016 1017 if (acpi_mps_check()) { 1018 #ifdef CONFIG_X86_LOCAL_APIC 1019 disable_apic = 1; 1020 #endif 1021 setup_clear_cpu_cap(X86_FEATURE_APIC); 1022 } 1023 1024 #ifdef CONFIG_PCI 1025 if (pci_early_dump_regs) 1026 early_dump_pci_devices(); 1027 #endif 1028 1029 /* update the e820_saved too */ 1030 e820_reserve_setup_data(); 1031 finish_e820_parsing(); 1032 1033 if (efi_enabled(EFI_BOOT)) 1034 efi_init(); 1035 1036 dmi_scan_machine(); 1037 dmi_memdev_walk(); 1038 dmi_set_dump_stack_arch_desc(); 1039 1040 /* 1041 * VMware detection requires dmi to be available, so this 1042 * needs to be done after dmi_scan_machine, for the BP. 1043 */ 1044 init_hypervisor_platform(); 1045 1046 x86_init.resources.probe_roms(); 1047 1048 /* after parse_early_param, so could debug it */ 1049 insert_resource(&iomem_resource, &code_resource); 1050 insert_resource(&iomem_resource, &data_resource); 1051 insert_resource(&iomem_resource, &bss_resource); 1052 1053 e820_add_kernel_range(); 1054 trim_bios_range(); 1055 #ifdef CONFIG_X86_32 1056 if (ppro_with_ram_bug()) { 1057 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 1058 E820_RESERVED); 1059 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 1060 printk(KERN_INFO "fixed physical RAM map:\n"); 1061 e820_print_map("bad_ppro"); 1062 } 1063 #else 1064 early_gart_iommu_check(); 1065 #endif 1066 1067 /* 1068 * partially used pages are not usable - thus 1069 * we are rounding upwards: 1070 */ 1071 max_pfn = e820_end_of_ram_pfn(); 1072 1073 /* update e820 for memory not covered by WB MTRRs */ 1074 mtrr_bp_init(); 1075 if (mtrr_trim_uncached_memory(max_pfn)) 1076 max_pfn = e820_end_of_ram_pfn(); 1077 1078 #ifdef CONFIG_X86_32 1079 /* max_low_pfn get updated here */ 1080 find_low_pfn_range(); 1081 #else 1082 check_x2apic(); 1083 1084 /* How many end-of-memory variables you have, grandma! */ 1085 /* need this before calling reserve_initrd */ 1086 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1087 max_low_pfn = e820_end_of_low_ram_pfn(); 1088 else 1089 max_low_pfn = max_pfn; 1090 1091 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1092 #endif 1093 1094 /* 1095 * Find and reserve possible boot-time SMP configuration: 1096 */ 1097 find_smp_config(); 1098 1099 reserve_ibft_region(); 1100 1101 early_alloc_pgt_buf(); 1102 1103 /* 1104 * Need to conclude brk, before memblock_x86_fill() 1105 * it could use memblock_find_in_range, could overlap with 1106 * brk area. 1107 */ 1108 reserve_brk(); 1109 1110 cleanup_highmap(); 1111 1112 memblock_set_current_limit(ISA_END_ADDRESS); 1113 memblock_x86_fill(); 1114 1115 /* 1116 * The EFI specification says that boot service code won't be called 1117 * after ExitBootServices(). This is, in fact, a lie. 1118 */ 1119 if (efi_enabled(EFI_MEMMAP)) 1120 efi_reserve_boot_services(); 1121 1122 /* preallocate 4k for mptable mpc */ 1123 early_reserve_e820_mpc_new(); 1124 1125 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1126 setup_bios_corruption_check(); 1127 #endif 1128 1129 #ifdef CONFIG_X86_32 1130 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1131 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1132 #endif 1133 1134 reserve_real_mode(); 1135 1136 trim_platform_memory_ranges(); 1137 trim_low_memory_range(); 1138 1139 init_mem_mapping(); 1140 1141 early_trap_pf_init(); 1142 1143 setup_real_mode(); 1144 1145 memblock_set_current_limit(get_max_mapped()); 1146 1147 /* 1148 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1149 */ 1150 1151 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1152 if (init_ohci1394_dma_early) 1153 init_ohci1394_dma_on_all_controllers(); 1154 #endif 1155 /* Allocate bigger log buffer */ 1156 setup_log_buf(1); 1157 1158 reserve_initrd(); 1159 1160 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD) 1161 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start); 1162 #endif 1163 1164 vsmp_init(); 1165 1166 io_delay_init(); 1167 1168 /* 1169 * Parse the ACPI tables for possible boot-time SMP configuration. 1170 */ 1171 acpi_boot_table_init(); 1172 1173 early_acpi_boot_init(); 1174 1175 initmem_init(); 1176 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1177 1178 /* 1179 * Reserve memory for crash kernel after SRAT is parsed so that it 1180 * won't consume hotpluggable memory. 1181 */ 1182 reserve_crashkernel(); 1183 1184 memblock_find_dma_reserve(); 1185 1186 #ifdef CONFIG_KVM_GUEST 1187 kvmclock_init(); 1188 #endif 1189 1190 x86_init.paging.pagetable_init(); 1191 1192 kasan_init(); 1193 1194 if (boot_cpu_data.cpuid_level >= 0) { 1195 /* A CPU has %cr4 if and only if it has CPUID */ 1196 mmu_cr4_features = __read_cr4(); 1197 if (trampoline_cr4_features) 1198 *trampoline_cr4_features = mmu_cr4_features; 1199 } 1200 1201 #ifdef CONFIG_X86_32 1202 /* sync back kernel address range */ 1203 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1204 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1205 KERNEL_PGD_PTRS); 1206 #endif 1207 1208 tboot_probe(); 1209 1210 map_vsyscall(); 1211 1212 generic_apic_probe(); 1213 1214 early_quirks(); 1215 1216 /* 1217 * Read APIC and some other early information from ACPI tables. 1218 */ 1219 acpi_boot_init(); 1220 sfi_init(); 1221 x86_dtb_init(); 1222 1223 /* 1224 * get boot-time SMP configuration: 1225 */ 1226 if (smp_found_config) 1227 get_smp_config(); 1228 1229 prefill_possible_map(); 1230 1231 init_cpu_to_node(); 1232 1233 init_apic_mappings(); 1234 if (x86_io_apic_ops.init) 1235 x86_io_apic_ops.init(); 1236 1237 kvm_guest_init(); 1238 1239 e820_reserve_resources(); 1240 e820_mark_nosave_regions(max_low_pfn); 1241 1242 x86_init.resources.reserve_resources(); 1243 1244 e820_setup_gap(); 1245 1246 #ifdef CONFIG_VT 1247 #if defined(CONFIG_VGA_CONSOLE) 1248 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1249 conswitchp = &vga_con; 1250 #elif defined(CONFIG_DUMMY_CONSOLE) 1251 conswitchp = &dummy_con; 1252 #endif 1253 #endif 1254 x86_init.oem.banner(); 1255 1256 x86_init.timers.wallclock_init(); 1257 1258 mcheck_init(); 1259 1260 arch_init_ideal_nops(); 1261 1262 register_refined_jiffies(CLOCK_TICK_RATE); 1263 1264 #ifdef CONFIG_EFI 1265 if (efi_enabled(EFI_BOOT)) 1266 efi_apply_memmap_quirks(); 1267 #endif 1268 } 1269 1270 #ifdef CONFIG_X86_32 1271 1272 static struct resource video_ram_resource = { 1273 .name = "Video RAM area", 1274 .start = 0xa0000, 1275 .end = 0xbffff, 1276 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1277 }; 1278 1279 void __init i386_reserve_resources(void) 1280 { 1281 request_resource(&iomem_resource, &video_ram_resource); 1282 reserve_standard_io_resources(); 1283 } 1284 1285 #endif /* CONFIG_X86_32 */ 1286 1287 static struct notifier_block kernel_offset_notifier = { 1288 .notifier_call = dump_kernel_offset 1289 }; 1290 1291 static int __init register_kernel_offset_dumper(void) 1292 { 1293 atomic_notifier_chain_register(&panic_notifier_list, 1294 &kernel_offset_notifier); 1295 return 0; 1296 } 1297 __initcall(register_kernel_offset_dumper); 1298