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 #include <asm/mmu_context.h> 116 117 /* 118 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 119 * max_pfn_mapped: highest direct mapped pfn over 4GB 120 * 121 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are 122 * represented by pfn_mapped 123 */ 124 unsigned long max_low_pfn_mapped; 125 unsigned long max_pfn_mapped; 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 #ifdef CONFIG_X86_32 150 /* cpu data as detected by the assembly code in head.S */ 151 struct cpuinfo_x86 new_cpu_data = { 152 .wp_works_ok = -1, 153 }; 154 /* common cpu data for all cpus */ 155 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 156 .wp_works_ok = -1, 157 }; 158 EXPORT_SYMBOL(boot_cpu_data); 159 160 unsigned int def_to_bigsmp; 161 162 /* for MCA, but anyone else can use it if they want */ 163 unsigned int machine_id; 164 unsigned int machine_submodel_id; 165 unsigned int BIOS_revision; 166 167 struct apm_info apm_info; 168 EXPORT_SYMBOL(apm_info); 169 170 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 171 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 172 struct ist_info ist_info; 173 EXPORT_SYMBOL(ist_info); 174 #else 175 struct ist_info ist_info; 176 #endif 177 178 #else 179 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 180 .x86_phys_bits = MAX_PHYSMEM_BITS, 181 }; 182 EXPORT_SYMBOL(boot_cpu_data); 183 #endif 184 185 186 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 187 __visible unsigned long mmu_cr4_features; 188 #else 189 __visible unsigned long mmu_cr4_features = X86_CR4_PAE; 190 #endif 191 192 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 193 int bootloader_type, bootloader_version; 194 195 /* 196 * Setup options 197 */ 198 struct screen_info screen_info; 199 EXPORT_SYMBOL(screen_info); 200 struct edid_info edid_info; 201 EXPORT_SYMBOL_GPL(edid_info); 202 203 extern int root_mountflags; 204 205 unsigned long saved_video_mode; 206 207 #define RAMDISK_IMAGE_START_MASK 0x07FF 208 #define RAMDISK_PROMPT_FLAG 0x8000 209 #define RAMDISK_LOAD_FLAG 0x4000 210 211 static char __initdata command_line[COMMAND_LINE_SIZE]; 212 #ifdef CONFIG_CMDLINE_BOOL 213 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 214 #endif 215 216 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 217 struct edd edd; 218 #ifdef CONFIG_EDD_MODULE 219 EXPORT_SYMBOL(edd); 220 #endif 221 /** 222 * copy_edd() - Copy the BIOS EDD information 223 * from boot_params into a safe place. 224 * 225 */ 226 static inline void __init copy_edd(void) 227 { 228 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 229 sizeof(edd.mbr_signature)); 230 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 231 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 232 edd.edd_info_nr = boot_params.eddbuf_entries; 233 } 234 #else 235 static inline void __init copy_edd(void) 236 { 237 } 238 #endif 239 240 void * __init extend_brk(size_t size, size_t align) 241 { 242 size_t mask = align - 1; 243 void *ret; 244 245 BUG_ON(_brk_start == 0); 246 BUG_ON(align & mask); 247 248 _brk_end = (_brk_end + mask) & ~mask; 249 BUG_ON((char *)(_brk_end + size) > __brk_limit); 250 251 ret = (void *)_brk_end; 252 _brk_end += size; 253 254 memset(ret, 0, size); 255 256 return ret; 257 } 258 259 #ifdef CONFIG_X86_32 260 static void __init cleanup_highmap(void) 261 { 262 } 263 #endif 264 265 static void __init reserve_brk(void) 266 { 267 if (_brk_end > _brk_start) 268 memblock_reserve(__pa_symbol(_brk_start), 269 _brk_end - _brk_start); 270 271 /* Mark brk area as locked down and no longer taking any 272 new allocations */ 273 _brk_start = 0; 274 } 275 276 u64 relocated_ramdisk; 277 278 #ifdef CONFIG_BLK_DEV_INITRD 279 280 static u64 __init get_ramdisk_image(void) 281 { 282 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 283 284 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 285 286 return ramdisk_image; 287 } 288 static u64 __init get_ramdisk_size(void) 289 { 290 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 291 292 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 293 294 return ramdisk_size; 295 } 296 297 static void __init relocate_initrd(void) 298 { 299 /* Assume only end is not page aligned */ 300 u64 ramdisk_image = get_ramdisk_image(); 301 u64 ramdisk_size = get_ramdisk_size(); 302 u64 area_size = PAGE_ALIGN(ramdisk_size); 303 304 /* We need to move the initrd down into directly mapped mem */ 305 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 306 area_size, PAGE_SIZE); 307 308 if (!relocated_ramdisk) 309 panic("Cannot find place for new RAMDISK of size %lld\n", 310 ramdisk_size); 311 312 /* Note: this includes all the mem currently occupied by 313 the initrd, we rely on that fact to keep the data intact. */ 314 memblock_reserve(relocated_ramdisk, area_size); 315 initrd_start = relocated_ramdisk + PAGE_OFFSET; 316 initrd_end = initrd_start + ramdisk_size; 317 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 318 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 319 320 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 321 322 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 323 " [mem %#010llx-%#010llx]\n", 324 ramdisk_image, ramdisk_image + ramdisk_size - 1, 325 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 326 } 327 328 static void __init early_reserve_initrd(void) 329 { 330 /* Assume only end is not page aligned */ 331 u64 ramdisk_image = get_ramdisk_image(); 332 u64 ramdisk_size = get_ramdisk_size(); 333 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 334 335 if (!boot_params.hdr.type_of_loader || 336 !ramdisk_image || !ramdisk_size) 337 return; /* No initrd provided by bootloader */ 338 339 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 340 } 341 static void __init reserve_initrd(void) 342 { 343 /* Assume only end is not page aligned */ 344 u64 ramdisk_image = get_ramdisk_image(); 345 u64 ramdisk_size = get_ramdisk_size(); 346 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 347 u64 mapped_size; 348 349 if (!boot_params.hdr.type_of_loader || 350 !ramdisk_image || !ramdisk_size) 351 return; /* No initrd provided by bootloader */ 352 353 initrd_start = 0; 354 355 mapped_size = memblock_mem_size(max_pfn_mapped); 356 if (ramdisk_size >= (mapped_size>>1)) 357 panic("initrd too large to handle, " 358 "disabling initrd (%lld needed, %lld available)\n", 359 ramdisk_size, mapped_size>>1); 360 361 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 362 ramdisk_end - 1); 363 364 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 365 PFN_DOWN(ramdisk_end))) { 366 /* All are mapped, easy case */ 367 initrd_start = ramdisk_image + PAGE_OFFSET; 368 initrd_end = initrd_start + ramdisk_size; 369 return; 370 } 371 372 relocate_initrd(); 373 374 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 375 } 376 #else 377 static void __init early_reserve_initrd(void) 378 { 379 } 380 static void __init reserve_initrd(void) 381 { 382 } 383 #endif /* CONFIG_BLK_DEV_INITRD */ 384 385 static void __init parse_setup_data(void) 386 { 387 struct setup_data *data; 388 u64 pa_data, pa_next; 389 390 pa_data = boot_params.hdr.setup_data; 391 while (pa_data) { 392 u32 data_len, data_type; 393 394 data = early_memremap(pa_data, sizeof(*data)); 395 data_len = data->len + sizeof(struct setup_data); 396 data_type = data->type; 397 pa_next = data->next; 398 early_memunmap(data, sizeof(*data)); 399 400 switch (data_type) { 401 case SETUP_E820_EXT: 402 parse_e820_ext(pa_data, data_len); 403 break; 404 case SETUP_DTB: 405 add_dtb(pa_data); 406 break; 407 case SETUP_EFI: 408 parse_efi_setup(pa_data, data_len); 409 break; 410 default: 411 break; 412 } 413 pa_data = pa_next; 414 } 415 } 416 417 static void __init e820_reserve_setup_data(void) 418 { 419 struct setup_data *data; 420 u64 pa_data; 421 422 pa_data = boot_params.hdr.setup_data; 423 if (!pa_data) 424 return; 425 426 while (pa_data) { 427 data = early_memremap(pa_data, sizeof(*data)); 428 e820_update_range(pa_data, sizeof(*data)+data->len, 429 E820_RAM, E820_RESERVED_KERN); 430 pa_data = data->next; 431 early_memunmap(data, sizeof(*data)); 432 } 433 434 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 435 memcpy(&e820_saved, &e820, sizeof(struct e820map)); 436 printk(KERN_INFO "extended physical RAM map:\n"); 437 e820_print_map("reserve setup_data"); 438 } 439 440 static void __init memblock_x86_reserve_range_setup_data(void) 441 { 442 struct setup_data *data; 443 u64 pa_data; 444 445 pa_data = boot_params.hdr.setup_data; 446 while (pa_data) { 447 data = early_memremap(pa_data, sizeof(*data)); 448 memblock_reserve(pa_data, sizeof(*data) + data->len); 449 pa_data = data->next; 450 early_memunmap(data, sizeof(*data)); 451 } 452 } 453 454 /* 455 * --------- Crashkernel reservation ------------------------------ 456 */ 457 458 #ifdef CONFIG_KEXEC_CORE 459 460 /* 16M alignment for crash kernel regions */ 461 #define CRASH_ALIGN (16 << 20) 462 463 /* 464 * Keep the crash kernel below this limit. On 32 bits earlier kernels 465 * would limit the kernel to the low 512 MiB due to mapping restrictions. 466 * On 64bit, old kexec-tools need to under 896MiB. 467 */ 468 #ifdef CONFIG_X86_32 469 # define CRASH_ADDR_LOW_MAX (512 << 20) 470 # define CRASH_ADDR_HIGH_MAX (512 << 20) 471 #else 472 # define CRASH_ADDR_LOW_MAX (896UL << 20) 473 # define CRASH_ADDR_HIGH_MAX MAXMEM 474 #endif 475 476 static int __init reserve_crashkernel_low(void) 477 { 478 #ifdef CONFIG_X86_64 479 unsigned long long base, low_base = 0, low_size = 0; 480 unsigned long total_low_mem; 481 int ret; 482 483 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 484 485 /* crashkernel=Y,low */ 486 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 487 if (ret) { 488 /* 489 * two parts from lib/swiotlb.c: 490 * -swiotlb size: user-specified with swiotlb= or default. 491 * 492 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 493 * to 8M for other buffers that may need to stay low too. Also 494 * make sure we allocate enough extra low memory so that we 495 * don't run out of DMA buffers for 32-bit devices. 496 */ 497 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 498 } else { 499 /* passed with crashkernel=0,low ? */ 500 if (!low_size) 501 return 0; 502 } 503 504 low_base = memblock_find_in_range(low_size, 1ULL << 32, low_size, CRASH_ALIGN); 505 if (!low_base) { 506 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 507 (unsigned long)(low_size >> 20)); 508 return -ENOMEM; 509 } 510 511 ret = memblock_reserve(low_base, low_size); 512 if (ret) { 513 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 514 return ret; 515 } 516 517 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 518 (unsigned long)(low_size >> 20), 519 (unsigned long)(low_base >> 20), 520 (unsigned long)(total_low_mem >> 20)); 521 522 crashk_low_res.start = low_base; 523 crashk_low_res.end = low_base + low_size - 1; 524 insert_resource(&iomem_resource, &crashk_low_res); 525 #endif 526 return 0; 527 } 528 529 static void __init reserve_crashkernel(void) 530 { 531 unsigned long long crash_size, crash_base, total_mem; 532 bool high = false; 533 int ret; 534 535 total_mem = memblock_phys_mem_size(); 536 537 /* crashkernel=XM */ 538 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 539 if (ret != 0 || crash_size <= 0) { 540 /* crashkernel=X,high */ 541 ret = parse_crashkernel_high(boot_command_line, total_mem, 542 &crash_size, &crash_base); 543 if (ret != 0 || crash_size <= 0) 544 return; 545 high = true; 546 } 547 548 /* 0 means: find the address automatically */ 549 if (crash_base <= 0) { 550 /* 551 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 552 */ 553 crash_base = memblock_find_in_range(CRASH_ALIGN, 554 high ? CRASH_ADDR_HIGH_MAX 555 : CRASH_ADDR_LOW_MAX, 556 crash_size, CRASH_ALIGN); 557 if (!crash_base) { 558 pr_info("crashkernel reservation failed - No suitable area found.\n"); 559 return; 560 } 561 562 } else { 563 unsigned long long start; 564 565 start = memblock_find_in_range(crash_base, 566 crash_base + crash_size, 567 crash_size, 1 << 20); 568 if (start != crash_base) { 569 pr_info("crashkernel reservation failed - memory is in use.\n"); 570 return; 571 } 572 } 573 ret = memblock_reserve(crash_base, crash_size); 574 if (ret) { 575 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 576 return; 577 } 578 579 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 580 memblock_free(crash_base, crash_size); 581 return; 582 } 583 584 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 585 (unsigned long)(crash_size >> 20), 586 (unsigned long)(crash_base >> 20), 587 (unsigned long)(total_mem >> 20)); 588 589 crashk_res.start = crash_base; 590 crashk_res.end = crash_base + crash_size - 1; 591 insert_resource(&iomem_resource, &crashk_res); 592 } 593 #else 594 static void __init reserve_crashkernel(void) 595 { 596 } 597 #endif 598 599 static struct resource standard_io_resources[] = { 600 { .name = "dma1", .start = 0x00, .end = 0x1f, 601 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 602 { .name = "pic1", .start = 0x20, .end = 0x21, 603 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 604 { .name = "timer0", .start = 0x40, .end = 0x43, 605 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 606 { .name = "timer1", .start = 0x50, .end = 0x53, 607 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 608 { .name = "keyboard", .start = 0x60, .end = 0x60, 609 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 610 { .name = "keyboard", .start = 0x64, .end = 0x64, 611 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 612 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 613 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 614 { .name = "pic2", .start = 0xa0, .end = 0xa1, 615 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 616 { .name = "dma2", .start = 0xc0, .end = 0xdf, 617 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 618 { .name = "fpu", .start = 0xf0, .end = 0xff, 619 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 620 }; 621 622 void __init reserve_standard_io_resources(void) 623 { 624 int i; 625 626 /* request I/O space for devices used on all i[345]86 PCs */ 627 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 628 request_resource(&ioport_resource, &standard_io_resources[i]); 629 630 } 631 632 static __init void reserve_ibft_region(void) 633 { 634 unsigned long addr, size = 0; 635 636 addr = find_ibft_region(&size); 637 638 if (size) 639 memblock_reserve(addr, size); 640 } 641 642 static bool __init snb_gfx_workaround_needed(void) 643 { 644 #ifdef CONFIG_PCI 645 int i; 646 u16 vendor, devid; 647 static const __initconst u16 snb_ids[] = { 648 0x0102, 649 0x0112, 650 0x0122, 651 0x0106, 652 0x0116, 653 0x0126, 654 0x010a, 655 }; 656 657 /* Assume no if something weird is going on with PCI */ 658 if (!early_pci_allowed()) 659 return false; 660 661 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 662 if (vendor != 0x8086) 663 return false; 664 665 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 666 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 667 if (devid == snb_ids[i]) 668 return true; 669 #endif 670 671 return false; 672 } 673 674 /* 675 * Sandy Bridge graphics has trouble with certain ranges, exclude 676 * them from allocation. 677 */ 678 static void __init trim_snb_memory(void) 679 { 680 static const __initconst unsigned long bad_pages[] = { 681 0x20050000, 682 0x20110000, 683 0x20130000, 684 0x20138000, 685 0x40004000, 686 }; 687 int i; 688 689 if (!snb_gfx_workaround_needed()) 690 return; 691 692 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 693 694 /* 695 * Reserve all memory below the 1 MB mark that has not 696 * already been reserved. 697 */ 698 memblock_reserve(0, 1<<20); 699 700 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 701 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 702 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 703 bad_pages[i]); 704 } 705 } 706 707 /* 708 * Here we put platform-specific memory range workarounds, i.e. 709 * memory known to be corrupt or otherwise in need to be reserved on 710 * specific platforms. 711 * 712 * If this gets used more widely it could use a real dispatch mechanism. 713 */ 714 static void __init trim_platform_memory_ranges(void) 715 { 716 trim_snb_memory(); 717 } 718 719 static void __init trim_bios_range(void) 720 { 721 /* 722 * A special case is the first 4Kb of memory; 723 * This is a BIOS owned area, not kernel ram, but generally 724 * not listed as such in the E820 table. 725 * 726 * This typically reserves additional memory (64KiB by default) 727 * since some BIOSes are known to corrupt low memory. See the 728 * Kconfig help text for X86_RESERVE_LOW. 729 */ 730 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 731 732 /* 733 * special case: Some BIOSen report the PC BIOS 734 * area (640->1Mb) as ram even though it is not. 735 * take them out. 736 */ 737 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 738 739 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 740 } 741 742 /* called before trim_bios_range() to spare extra sanitize */ 743 static void __init e820_add_kernel_range(void) 744 { 745 u64 start = __pa_symbol(_text); 746 u64 size = __pa_symbol(_end) - start; 747 748 /* 749 * Complain if .text .data and .bss are not marked as E820_RAM and 750 * attempt to fix it by adding the range. We may have a confused BIOS, 751 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 752 * exclude kernel range. If we really are running on top non-RAM, 753 * we will crash later anyways. 754 */ 755 if (e820_all_mapped(start, start + size, E820_RAM)) 756 return; 757 758 pr_warn(".text .data .bss are not marked as E820_RAM!\n"); 759 e820_remove_range(start, size, E820_RAM, 0); 760 e820_add_region(start, size, E820_RAM); 761 } 762 763 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 764 765 static int __init parse_reservelow(char *p) 766 { 767 unsigned long long size; 768 769 if (!p) 770 return -EINVAL; 771 772 size = memparse(p, &p); 773 774 if (size < 4096) 775 size = 4096; 776 777 if (size > 640*1024) 778 size = 640*1024; 779 780 reserve_low = size; 781 782 return 0; 783 } 784 785 early_param("reservelow", parse_reservelow); 786 787 static void __init trim_low_memory_range(void) 788 { 789 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 790 } 791 792 /* 793 * Dump out kernel offset information on panic. 794 */ 795 static int 796 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 797 { 798 if (kaslr_enabled()) { 799 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 800 kaslr_offset(), 801 __START_KERNEL, 802 __START_KERNEL_map, 803 MODULES_VADDR-1); 804 } else { 805 pr_emerg("Kernel Offset: disabled\n"); 806 } 807 808 return 0; 809 } 810 811 /* 812 * Determine if we were loaded by an EFI loader. If so, then we have also been 813 * passed the efi memmap, systab, etc., so we should use these data structures 814 * for initialization. Note, the efi init code path is determined by the 815 * global efi_enabled. This allows the same kernel image to be used on existing 816 * systems (with a traditional BIOS) as well as on EFI systems. 817 */ 818 /* 819 * setup_arch - architecture-specific boot-time initializations 820 * 821 * Note: On x86_64, fixmaps are ready for use even before this is called. 822 */ 823 824 void __init setup_arch(char **cmdline_p) 825 { 826 memblock_reserve(__pa_symbol(_text), 827 (unsigned long)__bss_stop - (unsigned long)_text); 828 829 early_reserve_initrd(); 830 831 /* 832 * At this point everything still needed from the boot loader 833 * or BIOS or kernel text should be early reserved or marked not 834 * RAM in e820. All other memory is free game. 835 */ 836 837 #ifdef CONFIG_X86_32 838 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 839 840 /* 841 * copy kernel address range established so far and switch 842 * to the proper swapper page table 843 */ 844 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 845 initial_page_table + KERNEL_PGD_BOUNDARY, 846 KERNEL_PGD_PTRS); 847 848 load_cr3(swapper_pg_dir); 849 /* 850 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 851 * a cr3 based tlb flush, so the following __flush_tlb_all() 852 * will not flush anything because the cpu quirk which clears 853 * X86_FEATURE_PGE has not been invoked yet. Though due to the 854 * load_cr3() above the TLB has been flushed already. The 855 * quirk is invoked before subsequent calls to __flush_tlb_all() 856 * so proper operation is guaranteed. 857 */ 858 __flush_tlb_all(); 859 #else 860 printk(KERN_INFO "Command line: %s\n", boot_command_line); 861 #endif 862 863 /* 864 * If we have OLPC OFW, we might end up relocating the fixmap due to 865 * reserve_top(), so do this before touching the ioremap area. 866 */ 867 olpc_ofw_detect(); 868 869 early_trap_init(); 870 early_cpu_init(); 871 early_ioremap_init(); 872 873 setup_olpc_ofw_pgd(); 874 875 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 876 screen_info = boot_params.screen_info; 877 edid_info = boot_params.edid_info; 878 #ifdef CONFIG_X86_32 879 apm_info.bios = boot_params.apm_bios_info; 880 ist_info = boot_params.ist_info; 881 #endif 882 saved_video_mode = boot_params.hdr.vid_mode; 883 bootloader_type = boot_params.hdr.type_of_loader; 884 if ((bootloader_type >> 4) == 0xe) { 885 bootloader_type &= 0xf; 886 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 887 } 888 bootloader_version = bootloader_type & 0xf; 889 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 890 891 #ifdef CONFIG_BLK_DEV_RAM 892 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 893 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 894 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 895 #endif 896 #ifdef CONFIG_EFI 897 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 898 EFI32_LOADER_SIGNATURE, 4)) { 899 set_bit(EFI_BOOT, &efi.flags); 900 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 901 EFI64_LOADER_SIGNATURE, 4)) { 902 set_bit(EFI_BOOT, &efi.flags); 903 set_bit(EFI_64BIT, &efi.flags); 904 } 905 906 if (efi_enabled(EFI_BOOT)) 907 efi_memblock_x86_reserve_range(); 908 #endif 909 910 x86_init.oem.arch_setup(); 911 912 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 913 setup_memory_map(); 914 parse_setup_data(); 915 916 copy_edd(); 917 918 if (!boot_params.hdr.root_flags) 919 root_mountflags &= ~MS_RDONLY; 920 init_mm.start_code = (unsigned long) _text; 921 init_mm.end_code = (unsigned long) _etext; 922 init_mm.end_data = (unsigned long) _edata; 923 init_mm.brk = _brk_end; 924 925 mpx_mm_init(&init_mm); 926 927 #ifdef CONFIG_CMDLINE_BOOL 928 #ifdef CONFIG_CMDLINE_OVERRIDE 929 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 930 #else 931 if (builtin_cmdline[0]) { 932 /* append boot loader cmdline to builtin */ 933 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 934 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 935 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 936 } 937 #endif 938 #endif 939 940 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 941 *cmdline_p = command_line; 942 943 /* 944 * x86_configure_nx() is called before parse_early_param() to detect 945 * whether hardware doesn't support NX (so that the early EHCI debug 946 * console setup can safely call set_fixmap()). It may then be called 947 * again from within noexec_setup() during parsing early parameters 948 * to honor the respective command line option. 949 */ 950 x86_configure_nx(); 951 952 parse_early_param(); 953 954 x86_report_nx(); 955 956 /* after early param, so could get panic from serial */ 957 memblock_x86_reserve_range_setup_data(); 958 959 if (acpi_mps_check()) { 960 #ifdef CONFIG_X86_LOCAL_APIC 961 disable_apic = 1; 962 #endif 963 setup_clear_cpu_cap(X86_FEATURE_APIC); 964 } 965 966 #ifdef CONFIG_PCI 967 if (pci_early_dump_regs) 968 early_dump_pci_devices(); 969 #endif 970 971 /* update the e820_saved too */ 972 e820_reserve_setup_data(); 973 finish_e820_parsing(); 974 975 if (efi_enabled(EFI_BOOT)) 976 efi_init(); 977 978 dmi_scan_machine(); 979 dmi_memdev_walk(); 980 dmi_set_dump_stack_arch_desc(); 981 982 /* 983 * VMware detection requires dmi to be available, so this 984 * needs to be done after dmi_scan_machine, for the BP. 985 */ 986 init_hypervisor_platform(); 987 988 x86_init.resources.probe_roms(); 989 990 e820_add_kernel_range(); 991 trim_bios_range(); 992 #ifdef CONFIG_X86_32 993 if (ppro_with_ram_bug()) { 994 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 995 E820_RESERVED); 996 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 997 printk(KERN_INFO "fixed physical RAM map:\n"); 998 e820_print_map("bad_ppro"); 999 } 1000 #else 1001 early_gart_iommu_check(); 1002 #endif 1003 1004 /* 1005 * partially used pages are not usable - thus 1006 * we are rounding upwards: 1007 */ 1008 max_pfn = e820_end_of_ram_pfn(); 1009 1010 /* update e820 for memory not covered by WB MTRRs */ 1011 mtrr_bp_init(); 1012 if (mtrr_trim_uncached_memory(max_pfn)) 1013 max_pfn = e820_end_of_ram_pfn(); 1014 1015 max_possible_pfn = max_pfn; 1016 1017 #ifdef CONFIG_X86_32 1018 /* max_low_pfn get updated here */ 1019 find_low_pfn_range(); 1020 #else 1021 check_x2apic(); 1022 1023 /* How many end-of-memory variables you have, grandma! */ 1024 /* need this before calling reserve_initrd */ 1025 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1026 max_low_pfn = e820_end_of_low_ram_pfn(); 1027 else 1028 max_low_pfn = max_pfn; 1029 1030 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1031 #endif 1032 1033 /* 1034 * Find and reserve possible boot-time SMP configuration: 1035 */ 1036 find_smp_config(); 1037 1038 reserve_ibft_region(); 1039 1040 early_alloc_pgt_buf(); 1041 1042 /* 1043 * Need to conclude brk, before memblock_x86_fill() 1044 * it could use memblock_find_in_range, could overlap with 1045 * brk area. 1046 */ 1047 reserve_brk(); 1048 1049 cleanup_highmap(); 1050 1051 memblock_set_current_limit(ISA_END_ADDRESS); 1052 memblock_x86_fill(); 1053 1054 if (efi_enabled(EFI_BOOT)) { 1055 efi_fake_memmap(); 1056 efi_find_mirror(); 1057 } 1058 1059 /* 1060 * The EFI specification says that boot service code won't be called 1061 * after ExitBootServices(). This is, in fact, a lie. 1062 */ 1063 if (efi_enabled(EFI_MEMMAP)) 1064 efi_reserve_boot_services(); 1065 1066 /* preallocate 4k for mptable mpc */ 1067 early_reserve_e820_mpc_new(); 1068 1069 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1070 setup_bios_corruption_check(); 1071 #endif 1072 1073 #ifdef CONFIG_X86_32 1074 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1075 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1076 #endif 1077 1078 reserve_real_mode(); 1079 1080 trim_platform_memory_ranges(); 1081 trim_low_memory_range(); 1082 1083 init_mem_mapping(); 1084 1085 early_trap_pf_init(); 1086 1087 setup_real_mode(); 1088 1089 memblock_set_current_limit(get_max_mapped()); 1090 1091 /* 1092 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1093 */ 1094 1095 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1096 if (init_ohci1394_dma_early) 1097 init_ohci1394_dma_on_all_controllers(); 1098 #endif 1099 /* Allocate bigger log buffer */ 1100 setup_log_buf(1); 1101 1102 reserve_initrd(); 1103 1104 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD) 1105 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start); 1106 #endif 1107 1108 vsmp_init(); 1109 1110 io_delay_init(); 1111 1112 /* 1113 * Parse the ACPI tables for possible boot-time SMP configuration. 1114 */ 1115 acpi_boot_table_init(); 1116 1117 early_acpi_boot_init(); 1118 1119 initmem_init(); 1120 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1121 1122 /* 1123 * Reserve memory for crash kernel after SRAT is parsed so that it 1124 * won't consume hotpluggable memory. 1125 */ 1126 reserve_crashkernel(); 1127 1128 memblock_find_dma_reserve(); 1129 1130 #ifdef CONFIG_KVM_GUEST 1131 kvmclock_init(); 1132 #endif 1133 1134 x86_init.paging.pagetable_init(); 1135 1136 kasan_init(); 1137 1138 if (boot_cpu_data.cpuid_level >= 0) { 1139 /* A CPU has %cr4 if and only if it has CPUID */ 1140 mmu_cr4_features = __read_cr4(); 1141 if (trampoline_cr4_features) 1142 *trampoline_cr4_features = mmu_cr4_features; 1143 } 1144 1145 #ifdef CONFIG_X86_32 1146 /* sync back kernel address range */ 1147 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1148 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1149 KERNEL_PGD_PTRS); 1150 1151 /* 1152 * sync back low identity map too. It is used for example 1153 * in the 32-bit EFI stub. 1154 */ 1155 clone_pgd_range(initial_page_table, 1156 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1157 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 1158 #endif 1159 1160 tboot_probe(); 1161 1162 map_vsyscall(); 1163 1164 generic_apic_probe(); 1165 1166 early_quirks(); 1167 1168 /* 1169 * Read APIC and some other early information from ACPI tables. 1170 */ 1171 acpi_boot_init(); 1172 sfi_init(); 1173 x86_dtb_init(); 1174 1175 /* 1176 * get boot-time SMP configuration: 1177 */ 1178 if (smp_found_config) 1179 get_smp_config(); 1180 1181 prefill_possible_map(); 1182 1183 init_cpu_to_node(); 1184 1185 init_apic_mappings(); 1186 io_apic_init_mappings(); 1187 1188 kvm_guest_init(); 1189 1190 e820_reserve_resources(); 1191 e820_mark_nosave_regions(max_low_pfn); 1192 1193 x86_init.resources.reserve_resources(); 1194 1195 e820_setup_gap(); 1196 1197 #ifdef CONFIG_VT 1198 #if defined(CONFIG_VGA_CONSOLE) 1199 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1200 conswitchp = &vga_con; 1201 #elif defined(CONFIG_DUMMY_CONSOLE) 1202 conswitchp = &dummy_con; 1203 #endif 1204 #endif 1205 x86_init.oem.banner(); 1206 1207 x86_init.timers.wallclock_init(); 1208 1209 mcheck_init(); 1210 1211 arch_init_ideal_nops(); 1212 1213 register_refined_jiffies(CLOCK_TICK_RATE); 1214 1215 #ifdef CONFIG_EFI 1216 if (efi_enabled(EFI_BOOT)) 1217 efi_apply_memmap_quirks(); 1218 #endif 1219 } 1220 1221 #ifdef CONFIG_X86_32 1222 1223 static struct resource video_ram_resource = { 1224 .name = "Video RAM area", 1225 .start = 0xa0000, 1226 .end = 0xbffff, 1227 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1228 }; 1229 1230 void __init i386_reserve_resources(void) 1231 { 1232 request_resource(&iomem_resource, &video_ram_resource); 1233 reserve_standard_io_resources(); 1234 } 1235 1236 #endif /* CONFIG_X86_32 */ 1237 1238 static struct notifier_block kernel_offset_notifier = { 1239 .notifier_call = dump_kernel_offset 1240 }; 1241 1242 static int __init register_kernel_offset_dumper(void) 1243 { 1244 atomic_notifier_chain_register(&panic_notifier_list, 1245 &kernel_offset_notifier); 1246 return 0; 1247 } 1248 __initcall(register_kernel_offset_dumper); 1249 1250 void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 1251 { 1252 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 1253 return; 1254 1255 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1256 } 1257