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 #ifdef CONFIG_DMI 126 RESERVE_BRK(dmi_alloc, 65536); 127 #endif 128 129 130 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 131 unsigned long _brk_end = (unsigned long)__brk_base; 132 133 #ifdef CONFIG_X86_64 134 int default_cpu_present_to_apicid(int mps_cpu) 135 { 136 return __default_cpu_present_to_apicid(mps_cpu); 137 } 138 139 int default_check_phys_apicid_present(int phys_apicid) 140 { 141 return __default_check_phys_apicid_present(phys_apicid); 142 } 143 #endif 144 145 struct boot_params boot_params; 146 147 /* 148 * Machine setup.. 149 */ 150 static struct resource data_resource = { 151 .name = "Kernel data", 152 .start = 0, 153 .end = 0, 154 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 155 }; 156 157 static struct resource code_resource = { 158 .name = "Kernel code", 159 .start = 0, 160 .end = 0, 161 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 162 }; 163 164 static 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 */ 174 struct cpuinfo_x86 new_cpu_data = { 175 .wp_works_ok = -1, 176 }; 177 /* common cpu data for all cpus */ 178 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 179 .wp_works_ok = -1, 180 }; 181 EXPORT_SYMBOL(boot_cpu_data); 182 183 unsigned int def_to_bigsmp; 184 185 /* for MCA, but anyone else can use it if they want */ 186 unsigned int machine_id; 187 unsigned int machine_submodel_id; 188 unsigned int BIOS_revision; 189 190 struct apm_info apm_info; 191 EXPORT_SYMBOL(apm_info); 192 193 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 194 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 195 struct ist_info ist_info; 196 EXPORT_SYMBOL(ist_info); 197 #else 198 struct ist_info ist_info; 199 #endif 200 201 #else 202 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 203 .x86_phys_bits = MAX_PHYSMEM_BITS, 204 }; 205 EXPORT_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 */ 216 int bootloader_type, bootloader_version; 217 218 /* 219 * Setup options 220 */ 221 struct screen_info screen_info; 222 EXPORT_SYMBOL(screen_info); 223 struct edid_info edid_info; 224 EXPORT_SYMBOL_GPL(edid_info); 225 226 extern int root_mountflags; 227 228 unsigned long saved_video_mode; 229 230 #define RAMDISK_IMAGE_START_MASK 0x07FF 231 #define RAMDISK_PROMPT_FLAG 0x8000 232 #define RAMDISK_LOAD_FLAG 0x4000 233 234 static char __initdata command_line[COMMAND_LINE_SIZE]; 235 #ifdef CONFIG_CMDLINE_BOOL 236 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 237 #endif 238 239 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 240 struct edd edd; 241 #ifdef CONFIG_EDD_MODULE 242 EXPORT_SYMBOL(edd); 243 #endif 244 /** 245 * copy_edd() - Copy the BIOS EDD information 246 * from boot_params into a safe place. 247 * 248 */ 249 static 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 258 static inline void __init copy_edd(void) 259 { 260 } 261 #endif 262 263 void * __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_32 283 static void __init cleanup_highmap(void) 284 { 285 } 286 #endif 287 288 static 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 } 298 299 u64 relocated_ramdisk; 300 301 #ifdef CONFIG_BLK_DEV_INITRD 302 303 static 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 } 311 static 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) 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 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 } 370 371 static 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 } 384 static 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 420 static void __init early_reserve_initrd(void) 421 { 422 } 423 static void __init reserve_initrd(void) 424 { 425 } 426 #endif /* CONFIG_BLK_DEV_INITRD */ 427 428 static 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 } 459 460 static void __init e820_reserve_setup_data(void) 461 { 462 struct setup_data *data; 463 u64 pa_data; 464 465 pa_data = boot_params.hdr.setup_data; 466 if (!pa_data) 467 return; 468 469 while (pa_data) { 470 data = early_memremap(pa_data, sizeof(*data)); 471 e820_update_range(pa_data, sizeof(*data)+data->len, 472 E820_RAM, E820_RESERVED_KERN); 473 pa_data = data->next; 474 early_memunmap(data, sizeof(*data)); 475 } 476 477 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 478 memcpy(&e820_saved, &e820, sizeof(struct e820map)); 479 printk(KERN_INFO "extended physical RAM map:\n"); 480 e820_print_map("reserve setup_data"); 481 } 482 483 static void __init memblock_x86_reserve_range_setup_data(void) 484 { 485 struct setup_data *data; 486 u64 pa_data; 487 488 pa_data = boot_params.hdr.setup_data; 489 while (pa_data) { 490 data = early_memremap(pa_data, sizeof(*data)); 491 memblock_reserve(pa_data, sizeof(*data) + data->len); 492 pa_data = data->next; 493 early_memunmap(data, sizeof(*data)); 494 } 495 } 496 497 /* 498 * --------- Crashkernel reservation ------------------------------ 499 */ 500 501 #ifdef CONFIG_KEXEC 502 503 /* 504 * Keep the crash kernel below this limit. On 32 bits earlier kernels 505 * would limit the kernel to the low 512 MiB due to mapping restrictions. 506 * On 64bit, old kexec-tools need to under 896MiB. 507 */ 508 #ifdef CONFIG_X86_32 509 # define CRASH_KERNEL_ADDR_LOW_MAX (512 << 20) 510 # define CRASH_KERNEL_ADDR_HIGH_MAX (512 << 20) 511 #else 512 # define CRASH_KERNEL_ADDR_LOW_MAX (896UL<<20) 513 # define CRASH_KERNEL_ADDR_HIGH_MAX MAXMEM 514 #endif 515 516 static void __init reserve_crashkernel_low(void) 517 { 518 #ifdef CONFIG_X86_64 519 const unsigned long long alignment = 16<<20; /* 16M */ 520 unsigned long long low_base = 0, low_size = 0; 521 unsigned long total_low_mem; 522 unsigned long long base; 523 bool auto_set = false; 524 int ret; 525 526 total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT)); 527 /* crashkernel=Y,low */ 528 ret = parse_crashkernel_low(boot_command_line, total_low_mem, 529 &low_size, &base); 530 if (ret != 0) { 531 /* 532 * two parts from lib/swiotlb.c: 533 * -swiotlb size: user-specified with swiotlb= or default. 534 * 535 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 536 * to 8M for other buffers that may need to stay low too. Also 537 * make sure we allocate enough extra low memory so that we 538 * don't run out of DMA buffers for 32-bit devices. 539 */ 540 low_size = max(swiotlb_size_or_default() + (8UL<<20), 256UL<<20); 541 auto_set = true; 542 } else { 543 /* passed with crashkernel=0,low ? */ 544 if (!low_size) 545 return; 546 } 547 548 low_base = memblock_find_in_range(low_size, (1ULL<<32), 549 low_size, alignment); 550 551 if (!low_base) { 552 if (!auto_set) 553 pr_info("crashkernel low reservation failed - No suitable area found.\n"); 554 555 return; 556 } 557 558 memblock_reserve(low_base, low_size); 559 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 560 (unsigned long)(low_size >> 20), 561 (unsigned long)(low_base >> 20), 562 (unsigned long)(total_low_mem >> 20)); 563 crashk_low_res.start = low_base; 564 crashk_low_res.end = low_base + low_size - 1; 565 insert_resource(&iomem_resource, &crashk_low_res); 566 #endif 567 } 568 569 static void __init reserve_crashkernel(void) 570 { 571 const unsigned long long alignment = 16<<20; /* 16M */ 572 unsigned long long total_mem; 573 unsigned long long crash_size, crash_base; 574 bool high = false; 575 int ret; 576 577 total_mem = memblock_phys_mem_size(); 578 579 /* crashkernel=XM */ 580 ret = parse_crashkernel(boot_command_line, total_mem, 581 &crash_size, &crash_base); 582 if (ret != 0 || crash_size <= 0) { 583 /* crashkernel=X,high */ 584 ret = parse_crashkernel_high(boot_command_line, total_mem, 585 &crash_size, &crash_base); 586 if (ret != 0 || crash_size <= 0) 587 return; 588 high = true; 589 } 590 591 /* 0 means: find the address automatically */ 592 if (crash_base <= 0) { 593 /* 594 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 595 */ 596 crash_base = memblock_find_in_range(alignment, 597 high ? CRASH_KERNEL_ADDR_HIGH_MAX : 598 CRASH_KERNEL_ADDR_LOW_MAX, 599 crash_size, alignment); 600 601 if (!crash_base) { 602 pr_info("crashkernel reservation failed - No suitable area found.\n"); 603 return; 604 } 605 606 } else { 607 unsigned long long start; 608 609 start = memblock_find_in_range(crash_base, 610 crash_base + crash_size, crash_size, 1<<20); 611 if (start != crash_base) { 612 pr_info("crashkernel reservation failed - memory is in use.\n"); 613 return; 614 } 615 } 616 memblock_reserve(crash_base, crash_size); 617 618 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " 619 "for crashkernel (System RAM: %ldMB)\n", 620 (unsigned long)(crash_size >> 20), 621 (unsigned long)(crash_base >> 20), 622 (unsigned long)(total_mem >> 20)); 623 624 crashk_res.start = crash_base; 625 crashk_res.end = crash_base + crash_size - 1; 626 insert_resource(&iomem_resource, &crashk_res); 627 628 if (crash_base >= (1ULL<<32)) 629 reserve_crashkernel_low(); 630 } 631 #else 632 static void __init reserve_crashkernel(void) 633 { 634 } 635 #endif 636 637 static struct resource standard_io_resources[] = { 638 { .name = "dma1", .start = 0x00, .end = 0x1f, 639 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 640 { .name = "pic1", .start = 0x20, .end = 0x21, 641 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 642 { .name = "timer0", .start = 0x40, .end = 0x43, 643 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 644 { .name = "timer1", .start = 0x50, .end = 0x53, 645 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 646 { .name = "keyboard", .start = 0x60, .end = 0x60, 647 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 648 { .name = "keyboard", .start = 0x64, .end = 0x64, 649 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 650 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 651 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 652 { .name = "pic2", .start = 0xa0, .end = 0xa1, 653 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 654 { .name = "dma2", .start = 0xc0, .end = 0xdf, 655 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 656 { .name = "fpu", .start = 0xf0, .end = 0xff, 657 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 658 }; 659 660 void __init reserve_standard_io_resources(void) 661 { 662 int i; 663 664 /* request I/O space for devices used on all i[345]86 PCs */ 665 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 666 request_resource(&ioport_resource, &standard_io_resources[i]); 667 668 } 669 670 static __init void reserve_ibft_region(void) 671 { 672 unsigned long addr, size = 0; 673 674 addr = find_ibft_region(&size); 675 676 if (size) 677 memblock_reserve(addr, size); 678 } 679 680 static bool __init snb_gfx_workaround_needed(void) 681 { 682 #ifdef CONFIG_PCI 683 int i; 684 u16 vendor, devid; 685 static const __initconst u16 snb_ids[] = { 686 0x0102, 687 0x0112, 688 0x0122, 689 0x0106, 690 0x0116, 691 0x0126, 692 0x010a, 693 }; 694 695 /* Assume no if something weird is going on with PCI */ 696 if (!early_pci_allowed()) 697 return false; 698 699 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 700 if (vendor != 0x8086) 701 return false; 702 703 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 704 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 705 if (devid == snb_ids[i]) 706 return true; 707 #endif 708 709 return false; 710 } 711 712 /* 713 * Sandy Bridge graphics has trouble with certain ranges, exclude 714 * them from allocation. 715 */ 716 static void __init trim_snb_memory(void) 717 { 718 static const __initconst unsigned long bad_pages[] = { 719 0x20050000, 720 0x20110000, 721 0x20130000, 722 0x20138000, 723 0x40004000, 724 }; 725 int i; 726 727 if (!snb_gfx_workaround_needed()) 728 return; 729 730 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 731 732 /* 733 * Reserve all memory below the 1 MB mark that has not 734 * already been reserved. 735 */ 736 memblock_reserve(0, 1<<20); 737 738 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 739 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 740 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 741 bad_pages[i]); 742 } 743 } 744 745 /* 746 * Here we put platform-specific memory range workarounds, i.e. 747 * memory known to be corrupt or otherwise in need to be reserved on 748 * specific platforms. 749 * 750 * If this gets used more widely it could use a real dispatch mechanism. 751 */ 752 static void __init trim_platform_memory_ranges(void) 753 { 754 trim_snb_memory(); 755 } 756 757 static void __init trim_bios_range(void) 758 { 759 /* 760 * A special case is the first 4Kb of memory; 761 * This is a BIOS owned area, not kernel ram, but generally 762 * not listed as such in the E820 table. 763 * 764 * This typically reserves additional memory (64KiB by default) 765 * since some BIOSes are known to corrupt low memory. See the 766 * Kconfig help text for X86_RESERVE_LOW. 767 */ 768 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 769 770 /* 771 * special case: Some BIOSen report the PC BIOS 772 * area (640->1Mb) as ram even though it is not. 773 * take them out. 774 */ 775 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 776 777 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 778 } 779 780 /* called before trim_bios_range() to spare extra sanitize */ 781 static void __init e820_add_kernel_range(void) 782 { 783 u64 start = __pa_symbol(_text); 784 u64 size = __pa_symbol(_end) - start; 785 786 /* 787 * Complain if .text .data and .bss are not marked as E820_RAM and 788 * attempt to fix it by adding the range. We may have a confused BIOS, 789 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 790 * exclude kernel range. If we really are running on top non-RAM, 791 * we will crash later anyways. 792 */ 793 if (e820_all_mapped(start, start + size, E820_RAM)) 794 return; 795 796 pr_warn(".text .data .bss are not marked as E820_RAM!\n"); 797 e820_remove_range(start, size, E820_RAM, 0); 798 e820_add_region(start, size, E820_RAM); 799 } 800 801 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 802 803 static int __init parse_reservelow(char *p) 804 { 805 unsigned long long size; 806 807 if (!p) 808 return -EINVAL; 809 810 size = memparse(p, &p); 811 812 if (size < 4096) 813 size = 4096; 814 815 if (size > 640*1024) 816 size = 640*1024; 817 818 reserve_low = size; 819 820 return 0; 821 } 822 823 early_param("reservelow", parse_reservelow); 824 825 static void __init trim_low_memory_range(void) 826 { 827 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 828 } 829 830 /* 831 * Dump out kernel offset information on panic. 832 */ 833 static int 834 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 835 { 836 if (kaslr_enabled()) { 837 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 838 kaslr_offset(), 839 __START_KERNEL, 840 __START_KERNEL_map, 841 MODULES_VADDR-1); 842 } else { 843 pr_emerg("Kernel Offset: disabled\n"); 844 } 845 846 return 0; 847 } 848 849 /* 850 * Determine if we were loaded by an EFI loader. If so, then we have also been 851 * passed the efi memmap, systab, etc., so we should use these data structures 852 * for initialization. Note, the efi init code path is determined by the 853 * global efi_enabled. This allows the same kernel image to be used on existing 854 * systems (with a traditional BIOS) as well as on EFI systems. 855 */ 856 /* 857 * setup_arch - architecture-specific boot-time initializations 858 * 859 * Note: On x86_64, fixmaps are ready for use even before this is called. 860 */ 861 862 void __init setup_arch(char **cmdline_p) 863 { 864 memblock_reserve(__pa_symbol(_text), 865 (unsigned long)__bss_stop - (unsigned long)_text); 866 867 early_reserve_initrd(); 868 869 /* 870 * At this point everything still needed from the boot loader 871 * or BIOS or kernel text should be early reserved or marked not 872 * RAM in e820. All other memory is free game. 873 */ 874 875 #ifdef CONFIG_X86_32 876 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 877 878 /* 879 * copy kernel address range established so far and switch 880 * to the proper swapper page table 881 */ 882 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 883 initial_page_table + KERNEL_PGD_BOUNDARY, 884 KERNEL_PGD_PTRS); 885 886 load_cr3(swapper_pg_dir); 887 /* 888 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 889 * a cr3 based tlb flush, so the following __flush_tlb_all() 890 * will not flush anything because the cpu quirk which clears 891 * X86_FEATURE_PGE has not been invoked yet. Though due to the 892 * load_cr3() above the TLB has been flushed already. The 893 * quirk is invoked before subsequent calls to __flush_tlb_all() 894 * so proper operation is guaranteed. 895 */ 896 __flush_tlb_all(); 897 #else 898 printk(KERN_INFO "Command line: %s\n", boot_command_line); 899 #endif 900 901 /* 902 * If we have OLPC OFW, we might end up relocating the fixmap due to 903 * reserve_top(), so do this before touching the ioremap area. 904 */ 905 olpc_ofw_detect(); 906 907 early_trap_init(); 908 early_cpu_init(); 909 early_ioremap_init(); 910 911 setup_olpc_ofw_pgd(); 912 913 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 914 screen_info = boot_params.screen_info; 915 edid_info = boot_params.edid_info; 916 #ifdef CONFIG_X86_32 917 apm_info.bios = boot_params.apm_bios_info; 918 ist_info = boot_params.ist_info; 919 if (boot_params.sys_desc_table.length != 0) { 920 machine_id = boot_params.sys_desc_table.table[0]; 921 machine_submodel_id = boot_params.sys_desc_table.table[1]; 922 BIOS_revision = boot_params.sys_desc_table.table[2]; 923 } 924 #endif 925 saved_video_mode = boot_params.hdr.vid_mode; 926 bootloader_type = boot_params.hdr.type_of_loader; 927 if ((bootloader_type >> 4) == 0xe) { 928 bootloader_type &= 0xf; 929 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 930 } 931 bootloader_version = bootloader_type & 0xf; 932 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 933 934 #ifdef CONFIG_BLK_DEV_RAM 935 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 936 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 937 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 938 #endif 939 #ifdef CONFIG_EFI 940 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 941 EFI32_LOADER_SIGNATURE, 4)) { 942 set_bit(EFI_BOOT, &efi.flags); 943 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 944 EFI64_LOADER_SIGNATURE, 4)) { 945 set_bit(EFI_BOOT, &efi.flags); 946 set_bit(EFI_64BIT, &efi.flags); 947 } 948 949 if (efi_enabled(EFI_BOOT)) 950 efi_memblock_x86_reserve_range(); 951 #endif 952 953 x86_init.oem.arch_setup(); 954 955 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 956 setup_memory_map(); 957 parse_setup_data(); 958 959 copy_edd(); 960 961 if (!boot_params.hdr.root_flags) 962 root_mountflags &= ~MS_RDONLY; 963 init_mm.start_code = (unsigned long) _text; 964 init_mm.end_code = (unsigned long) _etext; 965 init_mm.end_data = (unsigned long) _edata; 966 init_mm.brk = _brk_end; 967 968 mpx_mm_init(&init_mm); 969 970 code_resource.start = __pa_symbol(_text); 971 code_resource.end = __pa_symbol(_etext)-1; 972 data_resource.start = __pa_symbol(_etext); 973 data_resource.end = __pa_symbol(_edata)-1; 974 bss_resource.start = __pa_symbol(__bss_start); 975 bss_resource.end = __pa_symbol(__bss_stop)-1; 976 977 #ifdef CONFIG_CMDLINE_BOOL 978 #ifdef CONFIG_CMDLINE_OVERRIDE 979 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 980 #else 981 if (builtin_cmdline[0]) { 982 /* append boot loader cmdline to builtin */ 983 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 984 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 985 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 986 } 987 #endif 988 #endif 989 990 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 991 *cmdline_p = command_line; 992 993 /* 994 * x86_configure_nx() is called before parse_early_param() to detect 995 * whether hardware doesn't support NX (so that the early EHCI debug 996 * console setup can safely call set_fixmap()). It may then be called 997 * again from within noexec_setup() during parsing early parameters 998 * to honor the respective command line option. 999 */ 1000 x86_configure_nx(); 1001 1002 parse_early_param(); 1003 1004 x86_report_nx(); 1005 1006 /* after early param, so could get panic from serial */ 1007 memblock_x86_reserve_range_setup_data(); 1008 1009 if (acpi_mps_check()) { 1010 #ifdef CONFIG_X86_LOCAL_APIC 1011 disable_apic = 1; 1012 #endif 1013 setup_clear_cpu_cap(X86_FEATURE_APIC); 1014 } 1015 1016 #ifdef CONFIG_PCI 1017 if (pci_early_dump_regs) 1018 early_dump_pci_devices(); 1019 #endif 1020 1021 /* update the e820_saved too */ 1022 e820_reserve_setup_data(); 1023 finish_e820_parsing(); 1024 1025 if (efi_enabled(EFI_BOOT)) 1026 efi_init(); 1027 1028 dmi_scan_machine(); 1029 dmi_memdev_walk(); 1030 dmi_set_dump_stack_arch_desc(); 1031 1032 /* 1033 * VMware detection requires dmi to be available, so this 1034 * needs to be done after dmi_scan_machine, for the BP. 1035 */ 1036 init_hypervisor_platform(); 1037 1038 x86_init.resources.probe_roms(); 1039 1040 /* after parse_early_param, so could debug it */ 1041 insert_resource(&iomem_resource, &code_resource); 1042 insert_resource(&iomem_resource, &data_resource); 1043 insert_resource(&iomem_resource, &bss_resource); 1044 1045 e820_add_kernel_range(); 1046 trim_bios_range(); 1047 #ifdef CONFIG_X86_32 1048 if (ppro_with_ram_bug()) { 1049 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 1050 E820_RESERVED); 1051 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 1052 printk(KERN_INFO "fixed physical RAM map:\n"); 1053 e820_print_map("bad_ppro"); 1054 } 1055 #else 1056 early_gart_iommu_check(); 1057 #endif 1058 1059 /* 1060 * partially used pages are not usable - thus 1061 * we are rounding upwards: 1062 */ 1063 max_pfn = e820_end_of_ram_pfn(); 1064 1065 /* update e820 for memory not covered by WB MTRRs */ 1066 mtrr_bp_init(); 1067 if (mtrr_trim_uncached_memory(max_pfn)) 1068 max_pfn = e820_end_of_ram_pfn(); 1069 1070 #ifdef CONFIG_X86_32 1071 /* max_low_pfn get updated here */ 1072 find_low_pfn_range(); 1073 #else 1074 check_x2apic(); 1075 1076 /* How many end-of-memory variables you have, grandma! */ 1077 /* need this before calling reserve_initrd */ 1078 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1079 max_low_pfn = e820_end_of_low_ram_pfn(); 1080 else 1081 max_low_pfn = max_pfn; 1082 1083 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1084 #endif 1085 1086 /* 1087 * Find and reserve possible boot-time SMP configuration: 1088 */ 1089 find_smp_config(); 1090 1091 reserve_ibft_region(); 1092 1093 early_alloc_pgt_buf(); 1094 1095 /* 1096 * Need to conclude brk, before memblock_x86_fill() 1097 * it could use memblock_find_in_range, could overlap with 1098 * brk area. 1099 */ 1100 reserve_brk(); 1101 1102 cleanup_highmap(); 1103 1104 memblock_set_current_limit(ISA_END_ADDRESS); 1105 memblock_x86_fill(); 1106 1107 if (efi_enabled(EFI_BOOT)) 1108 efi_find_mirror(); 1109 1110 /* 1111 * The EFI specification says that boot service code won't be called 1112 * after ExitBootServices(). This is, in fact, a lie. 1113 */ 1114 if (efi_enabled(EFI_MEMMAP)) 1115 efi_reserve_boot_services(); 1116 1117 /* preallocate 4k for mptable mpc */ 1118 early_reserve_e820_mpc_new(); 1119 1120 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1121 setup_bios_corruption_check(); 1122 #endif 1123 1124 #ifdef CONFIG_X86_32 1125 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1126 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1127 #endif 1128 1129 reserve_real_mode(); 1130 1131 trim_platform_memory_ranges(); 1132 trim_low_memory_range(); 1133 1134 init_mem_mapping(); 1135 1136 early_trap_pf_init(); 1137 1138 setup_real_mode(); 1139 1140 memblock_set_current_limit(get_max_mapped()); 1141 1142 /* 1143 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1144 */ 1145 1146 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1147 if (init_ohci1394_dma_early) 1148 init_ohci1394_dma_on_all_controllers(); 1149 #endif 1150 /* Allocate bigger log buffer */ 1151 setup_log_buf(1); 1152 1153 reserve_initrd(); 1154 1155 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD) 1156 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start); 1157 #endif 1158 1159 vsmp_init(); 1160 1161 io_delay_init(); 1162 1163 /* 1164 * Parse the ACPI tables for possible boot-time SMP configuration. 1165 */ 1166 acpi_boot_table_init(); 1167 1168 early_acpi_boot_init(); 1169 1170 initmem_init(); 1171 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1172 1173 /* 1174 * Reserve memory for crash kernel after SRAT is parsed so that it 1175 * won't consume hotpluggable memory. 1176 */ 1177 reserve_crashkernel(); 1178 1179 memblock_find_dma_reserve(); 1180 1181 #ifdef CONFIG_KVM_GUEST 1182 kvmclock_init(); 1183 #endif 1184 1185 x86_init.paging.pagetable_init(); 1186 1187 kasan_init(); 1188 1189 if (boot_cpu_data.cpuid_level >= 0) { 1190 /* A CPU has %cr4 if and only if it has CPUID */ 1191 mmu_cr4_features = __read_cr4(); 1192 if (trampoline_cr4_features) 1193 *trampoline_cr4_features = mmu_cr4_features; 1194 } 1195 1196 #ifdef CONFIG_X86_32 1197 /* sync back kernel address range */ 1198 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1199 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1200 KERNEL_PGD_PTRS); 1201 #endif 1202 1203 tboot_probe(); 1204 1205 map_vsyscall(); 1206 1207 generic_apic_probe(); 1208 1209 early_quirks(); 1210 1211 /* 1212 * Read APIC and some other early information from ACPI tables. 1213 */ 1214 acpi_boot_init(); 1215 sfi_init(); 1216 x86_dtb_init(); 1217 1218 /* 1219 * get boot-time SMP configuration: 1220 */ 1221 if (smp_found_config) 1222 get_smp_config(); 1223 1224 prefill_possible_map(); 1225 1226 init_cpu_to_node(); 1227 1228 init_apic_mappings(); 1229 io_apic_init_mappings(); 1230 1231 kvm_guest_init(); 1232 1233 e820_reserve_resources(); 1234 e820_mark_nosave_regions(max_low_pfn); 1235 1236 x86_init.resources.reserve_resources(); 1237 1238 e820_setup_gap(); 1239 1240 #ifdef CONFIG_VT 1241 #if defined(CONFIG_VGA_CONSOLE) 1242 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1243 conswitchp = &vga_con; 1244 #elif defined(CONFIG_DUMMY_CONSOLE) 1245 conswitchp = &dummy_con; 1246 #endif 1247 #endif 1248 x86_init.oem.banner(); 1249 1250 x86_init.timers.wallclock_init(); 1251 1252 mcheck_init(); 1253 1254 arch_init_ideal_nops(); 1255 1256 register_refined_jiffies(CLOCK_TICK_RATE); 1257 1258 #ifdef CONFIG_EFI 1259 if (efi_enabled(EFI_BOOT)) 1260 efi_apply_memmap_quirks(); 1261 #endif 1262 } 1263 1264 #ifdef CONFIG_X86_32 1265 1266 static struct resource video_ram_resource = { 1267 .name = "Video RAM area", 1268 .start = 0xa0000, 1269 .end = 0xbffff, 1270 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1271 }; 1272 1273 void __init i386_reserve_resources(void) 1274 { 1275 request_resource(&iomem_resource, &video_ram_resource); 1276 reserve_standard_io_resources(); 1277 } 1278 1279 #endif /* CONFIG_X86_32 */ 1280 1281 static struct notifier_block kernel_offset_notifier = { 1282 .notifier_call = dump_kernel_offset 1283 }; 1284 1285 static int __init register_kernel_offset_dumper(void) 1286 { 1287 atomic_notifier_chain_register(&panic_notifier_list, 1288 &kernel_offset_notifier); 1289 return 0; 1290 } 1291 __initcall(register_kernel_offset_dumper); 1292