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/export.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 #include <linux/mem_encrypt.h> 73 74 #include <linux/usb/xhci-dbgp.h> 75 #include <video/edid.h> 76 77 #include <asm/mtrr.h> 78 #include <asm/apic.h> 79 #include <asm/realmode.h> 80 #include <asm/e820/api.h> 81 #include <asm/mpspec.h> 82 #include <asm/setup.h> 83 #include <asm/efi.h> 84 #include <asm/timer.h> 85 #include <asm/i8259.h> 86 #include <asm/sections.h> 87 #include <asm/io_apic.h> 88 #include <asm/ist.h> 89 #include <asm/setup_arch.h> 90 #include <asm/bios_ebda.h> 91 #include <asm/cacheflush.h> 92 #include <asm/processor.h> 93 #include <asm/bugs.h> 94 #include <asm/kasan.h> 95 96 #include <asm/vsyscall.h> 97 #include <asm/cpu.h> 98 #include <asm/desc.h> 99 #include <asm/dma.h> 100 #include <asm/iommu.h> 101 #include <asm/gart.h> 102 #include <asm/mmu_context.h> 103 #include <asm/proto.h> 104 105 #include <asm/paravirt.h> 106 #include <asm/hypervisor.h> 107 #include <asm/olpc_ofw.h> 108 109 #include <asm/percpu.h> 110 #include <asm/topology.h> 111 #include <asm/apicdef.h> 112 #include <asm/amd_nb.h> 113 #include <asm/mce.h> 114 #include <asm/alternative.h> 115 #include <asm/prom.h> 116 #include <asm/microcode.h> 117 #include <asm/mmu_context.h> 118 #include <asm/kaslr.h> 119 #include <asm/unwind.h> 120 121 /* 122 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 123 * max_pfn_mapped: highest direct mapped pfn over 4GB 124 * 125 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 126 * represented by pfn_mapped 127 */ 128 unsigned long max_low_pfn_mapped; 129 unsigned long max_pfn_mapped; 130 131 #ifdef CONFIG_DMI 132 RESERVE_BRK(dmi_alloc, 65536); 133 #endif 134 135 136 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 137 unsigned long _brk_end = (unsigned long)__brk_base; 138 139 #ifdef CONFIG_X86_64 140 int default_cpu_present_to_apicid(int mps_cpu) 141 { 142 return __default_cpu_present_to_apicid(mps_cpu); 143 } 144 145 int default_check_phys_apicid_present(int phys_apicid) 146 { 147 return __default_check_phys_apicid_present(phys_apicid); 148 } 149 #endif 150 151 struct boot_params boot_params; 152 153 /* 154 * Machine setup.. 155 */ 156 static struct resource data_resource = { 157 .name = "Kernel data", 158 .start = 0, 159 .end = 0, 160 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 161 }; 162 163 static struct resource code_resource = { 164 .name = "Kernel code", 165 .start = 0, 166 .end = 0, 167 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 168 }; 169 170 static struct resource bss_resource = { 171 .name = "Kernel bss", 172 .start = 0, 173 .end = 0, 174 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 175 }; 176 177 178 #ifdef CONFIG_X86_32 179 /* cpu data as detected by the assembly code in head_32.S */ 180 struct cpuinfo_x86 new_cpu_data; 181 182 /* common cpu data for all cpus */ 183 struct cpuinfo_x86 boot_cpu_data __read_mostly; 184 EXPORT_SYMBOL(boot_cpu_data); 185 186 unsigned int def_to_bigsmp; 187 188 /* for MCA, but anyone else can use it if they want */ 189 unsigned int machine_id; 190 unsigned int machine_submodel_id; 191 unsigned int BIOS_revision; 192 193 struct apm_info apm_info; 194 EXPORT_SYMBOL(apm_info); 195 196 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 197 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 198 struct ist_info ist_info; 199 EXPORT_SYMBOL(ist_info); 200 #else 201 struct ist_info ist_info; 202 #endif 203 204 #else 205 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 206 .x86_phys_bits = MAX_PHYSMEM_BITS, 207 }; 208 EXPORT_SYMBOL(boot_cpu_data); 209 #endif 210 211 212 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 213 __visible unsigned long mmu_cr4_features __ro_after_init; 214 #else 215 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 216 #endif 217 218 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 219 int bootloader_type, bootloader_version; 220 221 /* 222 * Setup options 223 */ 224 struct screen_info screen_info; 225 EXPORT_SYMBOL(screen_info); 226 struct edid_info edid_info; 227 EXPORT_SYMBOL_GPL(edid_info); 228 229 extern int root_mountflags; 230 231 unsigned long saved_video_mode; 232 233 #define RAMDISK_IMAGE_START_MASK 0x07FF 234 #define RAMDISK_PROMPT_FLAG 0x8000 235 #define RAMDISK_LOAD_FLAG 0x4000 236 237 static char __initdata command_line[COMMAND_LINE_SIZE]; 238 #ifdef CONFIG_CMDLINE_BOOL 239 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 240 #endif 241 242 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 243 struct edd edd; 244 #ifdef CONFIG_EDD_MODULE 245 EXPORT_SYMBOL(edd); 246 #endif 247 /** 248 * copy_edd() - Copy the BIOS EDD information 249 * from boot_params into a safe place. 250 * 251 */ 252 static inline void __init copy_edd(void) 253 { 254 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 255 sizeof(edd.mbr_signature)); 256 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 257 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 258 edd.edd_info_nr = boot_params.eddbuf_entries; 259 } 260 #else 261 static inline void __init copy_edd(void) 262 { 263 } 264 #endif 265 266 void * __init extend_brk(size_t size, size_t align) 267 { 268 size_t mask = align - 1; 269 void *ret; 270 271 BUG_ON(_brk_start == 0); 272 BUG_ON(align & mask); 273 274 _brk_end = (_brk_end + mask) & ~mask; 275 BUG_ON((char *)(_brk_end + size) > __brk_limit); 276 277 ret = (void *)_brk_end; 278 _brk_end += size; 279 280 memset(ret, 0, size); 281 282 return ret; 283 } 284 285 #ifdef CONFIG_X86_32 286 static void __init cleanup_highmap(void) 287 { 288 } 289 #endif 290 291 static void __init reserve_brk(void) 292 { 293 if (_brk_end > _brk_start) 294 memblock_reserve(__pa_symbol(_brk_start), 295 _brk_end - _brk_start); 296 297 /* Mark brk area as locked down and no longer taking any 298 new allocations */ 299 _brk_start = 0; 300 } 301 302 u64 relocated_ramdisk; 303 304 #ifdef CONFIG_BLK_DEV_INITRD 305 306 static u64 __init get_ramdisk_image(void) 307 { 308 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 309 310 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 311 312 return ramdisk_image; 313 } 314 static u64 __init get_ramdisk_size(void) 315 { 316 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 317 318 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 319 320 return ramdisk_size; 321 } 322 323 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); 329 330 /* 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 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 347 348 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 349 " [mem %#010llx-%#010llx]\n", 350 ramdisk_image, ramdisk_image + ramdisk_size - 1, 351 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 352 } 353 354 static void __init early_reserve_initrd(void) 355 { 356 /* Assume only end is not page aligned */ 357 u64 ramdisk_image = get_ramdisk_image(); 358 u64 ramdisk_size = get_ramdisk_size(); 359 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 360 361 if (!boot_params.hdr.type_of_loader || 362 !ramdisk_image || !ramdisk_size) 363 return; /* No initrd provided by bootloader */ 364 365 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 366 } 367 static void __init reserve_initrd(void) 368 { 369 /* Assume only end is not page aligned */ 370 u64 ramdisk_image = get_ramdisk_image(); 371 u64 ramdisk_size = get_ramdisk_size(); 372 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 373 u64 mapped_size; 374 375 if (!boot_params.hdr.type_of_loader || 376 !ramdisk_image || !ramdisk_size) 377 return; /* No initrd provided by bootloader */ 378 379 /* 380 * If SME is active, this memory will be marked encrypted by the 381 * kernel when it is accessed (including relocation). However, the 382 * ramdisk image was loaded decrypted by the bootloader, so make 383 * sure that it is encrypted before accessing it. 384 */ 385 sme_early_encrypt(ramdisk_image, ramdisk_end - ramdisk_image); 386 387 initrd_start = 0; 388 389 mapped_size = memblock_mem_size(max_pfn_mapped); 390 if (ramdisk_size >= (mapped_size>>1)) 391 panic("initrd too large to handle, " 392 "disabling initrd (%lld needed, %lld available)\n", 393 ramdisk_size, mapped_size>>1); 394 395 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 396 ramdisk_end - 1); 397 398 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 399 PFN_DOWN(ramdisk_end))) { 400 /* All are mapped, easy case */ 401 initrd_start = ramdisk_image + PAGE_OFFSET; 402 initrd_end = initrd_start + ramdisk_size; 403 return; 404 } 405 406 relocate_initrd(); 407 408 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 409 } 410 411 #else 412 static void __init early_reserve_initrd(void) 413 { 414 } 415 static void __init reserve_initrd(void) 416 { 417 } 418 #endif /* CONFIG_BLK_DEV_INITRD */ 419 420 static void __init parse_setup_data(void) 421 { 422 struct setup_data *data; 423 u64 pa_data, pa_next; 424 425 pa_data = boot_params.hdr.setup_data; 426 while (pa_data) { 427 u32 data_len, data_type; 428 429 data = early_memremap(pa_data, sizeof(*data)); 430 data_len = data->len + sizeof(struct setup_data); 431 data_type = data->type; 432 pa_next = data->next; 433 early_memunmap(data, sizeof(*data)); 434 435 switch (data_type) { 436 case SETUP_E820_EXT: 437 e820__memory_setup_extended(pa_data, data_len); 438 break; 439 case SETUP_DTB: 440 add_dtb(pa_data); 441 break; 442 case SETUP_EFI: 443 parse_efi_setup(pa_data, data_len); 444 break; 445 default: 446 break; 447 } 448 pa_data = pa_next; 449 } 450 } 451 452 static void __init memblock_x86_reserve_range_setup_data(void) 453 { 454 struct setup_data *data; 455 u64 pa_data; 456 457 pa_data = boot_params.hdr.setup_data; 458 while (pa_data) { 459 data = early_memremap(pa_data, sizeof(*data)); 460 memblock_reserve(pa_data, sizeof(*data) + data->len); 461 pa_data = data->next; 462 early_memunmap(data, sizeof(*data)); 463 } 464 } 465 466 /* 467 * --------- Crashkernel reservation ------------------------------ 468 */ 469 470 #ifdef CONFIG_KEXEC_CORE 471 472 /* 16M alignment for crash kernel regions */ 473 #define CRASH_ALIGN (16 << 20) 474 475 /* 476 * Keep the crash kernel below this limit. On 32 bits earlier kernels 477 * would limit the kernel to the low 512 MiB due to mapping restrictions. 478 * On 64bit, old kexec-tools need to under 896MiB. 479 */ 480 #ifdef CONFIG_X86_32 481 # define CRASH_ADDR_LOW_MAX (512 << 20) 482 # define CRASH_ADDR_HIGH_MAX (512 << 20) 483 #else 484 # define CRASH_ADDR_LOW_MAX (896UL << 20) 485 # define CRASH_ADDR_HIGH_MAX MAXMEM 486 #endif 487 488 static int __init reserve_crashkernel_low(void) 489 { 490 #ifdef CONFIG_X86_64 491 unsigned long long base, low_base = 0, low_size = 0; 492 unsigned long total_low_mem; 493 int ret; 494 495 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 496 497 /* crashkernel=Y,low */ 498 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 499 if (ret) { 500 /* 501 * two parts from lib/swiotlb.c: 502 * -swiotlb size: user-specified with swiotlb= or default. 503 * 504 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 505 * to 8M for other buffers that may need to stay low too. Also 506 * make sure we allocate enough extra low memory so that we 507 * don't run out of DMA buffers for 32-bit devices. 508 */ 509 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 510 } else { 511 /* passed with crashkernel=0,low ? */ 512 if (!low_size) 513 return 0; 514 } 515 516 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 517 if (!low_base) { 518 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 519 (unsigned long)(low_size >> 20)); 520 return -ENOMEM; 521 } 522 523 ret = memblock_reserve(low_base, low_size); 524 if (ret) { 525 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 526 return ret; 527 } 528 529 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 530 (unsigned long)(low_size >> 20), 531 (unsigned long)(low_base >> 20), 532 (unsigned long)(total_low_mem >> 20)); 533 534 crashk_low_res.start = low_base; 535 crashk_low_res.end = low_base + low_size - 1; 536 insert_resource(&iomem_resource, &crashk_low_res); 537 #endif 538 return 0; 539 } 540 541 static void __init reserve_crashkernel(void) 542 { 543 unsigned long long crash_size, crash_base, total_mem; 544 bool high = false; 545 int ret; 546 547 total_mem = memblock_phys_mem_size(); 548 549 /* crashkernel=XM */ 550 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 551 if (ret != 0 || crash_size <= 0) { 552 /* crashkernel=X,high */ 553 ret = parse_crashkernel_high(boot_command_line, total_mem, 554 &crash_size, &crash_base); 555 if (ret != 0 || crash_size <= 0) 556 return; 557 high = true; 558 } 559 560 /* 0 means: find the address automatically */ 561 if (crash_base <= 0) { 562 /* 563 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 564 * as old kexec-tools loads bzImage below that, unless 565 * "crashkernel=size[KMG],high" is specified. 566 */ 567 crash_base = memblock_find_in_range(CRASH_ALIGN, 568 high ? CRASH_ADDR_HIGH_MAX 569 : CRASH_ADDR_LOW_MAX, 570 crash_size, CRASH_ALIGN); 571 if (!crash_base) { 572 pr_info("crashkernel reservation failed - No suitable area found.\n"); 573 return; 574 } 575 576 } else { 577 unsigned long long start; 578 579 start = memblock_find_in_range(crash_base, 580 crash_base + crash_size, 581 crash_size, 1 << 20); 582 if (start != crash_base) { 583 pr_info("crashkernel reservation failed - memory is in use.\n"); 584 return; 585 } 586 } 587 ret = memblock_reserve(crash_base, crash_size); 588 if (ret) { 589 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 590 return; 591 } 592 593 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 594 memblock_free(crash_base, crash_size); 595 return; 596 } 597 598 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 599 (unsigned long)(crash_size >> 20), 600 (unsigned long)(crash_base >> 20), 601 (unsigned long)(total_mem >> 20)); 602 603 crashk_res.start = crash_base; 604 crashk_res.end = crash_base + crash_size - 1; 605 insert_resource(&iomem_resource, &crashk_res); 606 } 607 #else 608 static void __init reserve_crashkernel(void) 609 { 610 } 611 #endif 612 613 static struct resource standard_io_resources[] = { 614 { .name = "dma1", .start = 0x00, .end = 0x1f, 615 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 616 { .name = "pic1", .start = 0x20, .end = 0x21, 617 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 618 { .name = "timer0", .start = 0x40, .end = 0x43, 619 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 620 { .name = "timer1", .start = 0x50, .end = 0x53, 621 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 622 { .name = "keyboard", .start = 0x60, .end = 0x60, 623 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 624 { .name = "keyboard", .start = 0x64, .end = 0x64, 625 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 626 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 627 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 628 { .name = "pic2", .start = 0xa0, .end = 0xa1, 629 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 630 { .name = "dma2", .start = 0xc0, .end = 0xdf, 631 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 632 { .name = "fpu", .start = 0xf0, .end = 0xff, 633 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 634 }; 635 636 void __init reserve_standard_io_resources(void) 637 { 638 int i; 639 640 /* request I/O space for devices used on all i[345]86 PCs */ 641 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 642 request_resource(&ioport_resource, &standard_io_resources[i]); 643 644 } 645 646 static __init void reserve_ibft_region(void) 647 { 648 unsigned long addr, size = 0; 649 650 addr = find_ibft_region(&size); 651 652 if (size) 653 memblock_reserve(addr, size); 654 } 655 656 static bool __init snb_gfx_workaround_needed(void) 657 { 658 #ifdef CONFIG_PCI 659 int i; 660 u16 vendor, devid; 661 static const __initconst u16 snb_ids[] = { 662 0x0102, 663 0x0112, 664 0x0122, 665 0x0106, 666 0x0116, 667 0x0126, 668 0x010a, 669 }; 670 671 /* Assume no if something weird is going on with PCI */ 672 if (!early_pci_allowed()) 673 return false; 674 675 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 676 if (vendor != 0x8086) 677 return false; 678 679 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 680 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 681 if (devid == snb_ids[i]) 682 return true; 683 #endif 684 685 return false; 686 } 687 688 /* 689 * Sandy Bridge graphics has trouble with certain ranges, exclude 690 * them from allocation. 691 */ 692 static void __init trim_snb_memory(void) 693 { 694 static const __initconst unsigned long bad_pages[] = { 695 0x20050000, 696 0x20110000, 697 0x20130000, 698 0x20138000, 699 0x40004000, 700 }; 701 int i; 702 703 if (!snb_gfx_workaround_needed()) 704 return; 705 706 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 707 708 /* 709 * Reserve all memory below the 1 MB mark that has not 710 * already been reserved. 711 */ 712 memblock_reserve(0, 1<<20); 713 714 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 715 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 716 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 717 bad_pages[i]); 718 } 719 } 720 721 /* 722 * Here we put platform-specific memory range workarounds, i.e. 723 * memory known to be corrupt or otherwise in need to be reserved on 724 * specific platforms. 725 * 726 * If this gets used more widely it could use a real dispatch mechanism. 727 */ 728 static void __init trim_platform_memory_ranges(void) 729 { 730 trim_snb_memory(); 731 } 732 733 static void __init trim_bios_range(void) 734 { 735 /* 736 * A special case is the first 4Kb of memory; 737 * This is a BIOS owned area, not kernel ram, but generally 738 * not listed as such in the E820 table. 739 * 740 * This typically reserves additional memory (64KiB by default) 741 * since some BIOSes are known to corrupt low memory. See the 742 * Kconfig help text for X86_RESERVE_LOW. 743 */ 744 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 745 746 /* 747 * special case: Some BIOSen report the PC BIOS 748 * area (640->1Mb) as ram even though it is not. 749 * take them out. 750 */ 751 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 752 753 e820__update_table(e820_table); 754 } 755 756 /* called before trim_bios_range() to spare extra sanitize */ 757 static void __init e820_add_kernel_range(void) 758 { 759 u64 start = __pa_symbol(_text); 760 u64 size = __pa_symbol(_end) - start; 761 762 /* 763 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 764 * attempt to fix it by adding the range. We may have a confused BIOS, 765 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 766 * exclude kernel range. If we really are running on top non-RAM, 767 * we will crash later anyways. 768 */ 769 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 770 return; 771 772 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 773 e820__range_remove(start, size, E820_TYPE_RAM, 0); 774 e820__range_add(start, size, E820_TYPE_RAM); 775 } 776 777 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 778 779 static int __init parse_reservelow(char *p) 780 { 781 unsigned long long size; 782 783 if (!p) 784 return -EINVAL; 785 786 size = memparse(p, &p); 787 788 if (size < 4096) 789 size = 4096; 790 791 if (size > 640*1024) 792 size = 640*1024; 793 794 reserve_low = size; 795 796 return 0; 797 } 798 799 early_param("reservelow", parse_reservelow); 800 801 static void __init trim_low_memory_range(void) 802 { 803 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 804 } 805 806 /* 807 * Dump out kernel offset information on panic. 808 */ 809 static int 810 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 811 { 812 if (kaslr_enabled()) { 813 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 814 kaslr_offset(), 815 __START_KERNEL, 816 __START_KERNEL_map, 817 MODULES_VADDR-1); 818 } else { 819 pr_emerg("Kernel Offset: disabled\n"); 820 } 821 822 return 0; 823 } 824 825 static void __init simple_udelay_calibration(void) 826 { 827 unsigned int tsc_khz, cpu_khz; 828 unsigned long lpj; 829 830 if (!boot_cpu_has(X86_FEATURE_TSC)) 831 return; 832 833 cpu_khz = x86_platform.calibrate_cpu(); 834 tsc_khz = x86_platform.calibrate_tsc(); 835 836 tsc_khz = tsc_khz ? : cpu_khz; 837 if (!tsc_khz) 838 return; 839 840 lpj = tsc_khz * 1000; 841 do_div(lpj, HZ); 842 loops_per_jiffy = lpj; 843 } 844 845 /* 846 * Determine if we were loaded by an EFI loader. If so, then we have also been 847 * passed the efi memmap, systab, etc., so we should use these data structures 848 * for initialization. Note, the efi init code path is determined by the 849 * global efi_enabled. This allows the same kernel image to be used on existing 850 * systems (with a traditional BIOS) as well as on EFI systems. 851 */ 852 /* 853 * setup_arch - architecture-specific boot-time initializations 854 * 855 * Note: On x86_64, fixmaps are ready for use even before this is called. 856 */ 857 858 void __init setup_arch(char **cmdline_p) 859 { 860 memblock_reserve(__pa_symbol(_text), 861 (unsigned long)__bss_stop - (unsigned long)_text); 862 863 early_reserve_initrd(); 864 865 /* 866 * At this point everything still needed from the boot loader 867 * or BIOS or kernel text should be early reserved or marked not 868 * RAM in e820. All other memory is free game. 869 */ 870 871 #ifdef CONFIG_X86_32 872 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 873 874 /* 875 * copy kernel address range established so far and switch 876 * to the proper swapper page table 877 */ 878 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 879 initial_page_table + KERNEL_PGD_BOUNDARY, 880 KERNEL_PGD_PTRS); 881 882 load_cr3(swapper_pg_dir); 883 /* 884 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 885 * a cr3 based tlb flush, so the following __flush_tlb_all() 886 * will not flush anything because the cpu quirk which clears 887 * X86_FEATURE_PGE has not been invoked yet. Though due to the 888 * load_cr3() above the TLB has been flushed already. The 889 * quirk is invoked before subsequent calls to __flush_tlb_all() 890 * so proper operation is guaranteed. 891 */ 892 __flush_tlb_all(); 893 #else 894 printk(KERN_INFO "Command line: %s\n", boot_command_line); 895 #endif 896 897 /* 898 * If we have OLPC OFW, we might end up relocating the fixmap due to 899 * reserve_top(), so do this before touching the ioremap area. 900 */ 901 olpc_ofw_detect(); 902 903 idt_setup_early_traps(); 904 early_cpu_init(); 905 early_ioremap_init(); 906 907 setup_olpc_ofw_pgd(); 908 909 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 910 screen_info = boot_params.screen_info; 911 edid_info = boot_params.edid_info; 912 #ifdef CONFIG_X86_32 913 apm_info.bios = boot_params.apm_bios_info; 914 ist_info = boot_params.ist_info; 915 #endif 916 saved_video_mode = boot_params.hdr.vid_mode; 917 bootloader_type = boot_params.hdr.type_of_loader; 918 if ((bootloader_type >> 4) == 0xe) { 919 bootloader_type &= 0xf; 920 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 921 } 922 bootloader_version = bootloader_type & 0xf; 923 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 924 925 #ifdef CONFIG_BLK_DEV_RAM 926 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 927 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 928 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 929 #endif 930 #ifdef CONFIG_EFI 931 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 932 EFI32_LOADER_SIGNATURE, 4)) { 933 set_bit(EFI_BOOT, &efi.flags); 934 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 935 EFI64_LOADER_SIGNATURE, 4)) { 936 set_bit(EFI_BOOT, &efi.flags); 937 set_bit(EFI_64BIT, &efi.flags); 938 } 939 940 if (efi_enabled(EFI_BOOT)) 941 efi_memblock_x86_reserve_range(); 942 #endif 943 944 x86_init.oem.arch_setup(); 945 946 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 947 e820__memory_setup(); 948 parse_setup_data(); 949 950 copy_edd(); 951 952 if (!boot_params.hdr.root_flags) 953 root_mountflags &= ~MS_RDONLY; 954 init_mm.start_code = (unsigned long) _text; 955 init_mm.end_code = (unsigned long) _etext; 956 init_mm.end_data = (unsigned long) _edata; 957 init_mm.brk = _brk_end; 958 959 mpx_mm_init(&init_mm); 960 961 code_resource.start = __pa_symbol(_text); 962 code_resource.end = __pa_symbol(_etext)-1; 963 data_resource.start = __pa_symbol(_etext); 964 data_resource.end = __pa_symbol(_edata)-1; 965 bss_resource.start = __pa_symbol(__bss_start); 966 bss_resource.end = __pa_symbol(__bss_stop)-1; 967 968 #ifdef CONFIG_CMDLINE_BOOL 969 #ifdef CONFIG_CMDLINE_OVERRIDE 970 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 971 #else 972 if (builtin_cmdline[0]) { 973 /* append boot loader cmdline to builtin */ 974 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 975 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 976 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 977 } 978 #endif 979 #endif 980 981 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 982 *cmdline_p = command_line; 983 984 /* 985 * x86_configure_nx() is called before parse_early_param() to detect 986 * whether hardware doesn't support NX (so that the early EHCI debug 987 * console setup can safely call set_fixmap()). It may then be called 988 * again from within noexec_setup() during parsing early parameters 989 * to honor the respective command line option. 990 */ 991 x86_configure_nx(); 992 993 parse_early_param(); 994 995 #ifdef CONFIG_MEMORY_HOTPLUG 996 /* 997 * Memory used by the kernel cannot be hot-removed because Linux 998 * cannot migrate the kernel pages. When memory hotplug is 999 * enabled, we should prevent memblock from allocating memory 1000 * for the kernel. 1001 * 1002 * ACPI SRAT records all hotpluggable memory ranges. But before 1003 * SRAT is parsed, we don't know about it. 1004 * 1005 * The kernel image is loaded into memory at very early time. We 1006 * cannot prevent this anyway. So on NUMA system, we set any 1007 * node the kernel resides in as un-hotpluggable. 1008 * 1009 * Since on modern servers, one node could have double-digit 1010 * gigabytes memory, we can assume the memory around the kernel 1011 * image is also un-hotpluggable. So before SRAT is parsed, just 1012 * allocate memory near the kernel image to try the best to keep 1013 * the kernel away from hotpluggable memory. 1014 */ 1015 if (movable_node_is_enabled()) 1016 memblock_set_bottom_up(true); 1017 #endif 1018 1019 x86_report_nx(); 1020 1021 /* after early param, so could get panic from serial */ 1022 memblock_x86_reserve_range_setup_data(); 1023 1024 if (acpi_mps_check()) { 1025 #ifdef CONFIG_X86_LOCAL_APIC 1026 disable_apic = 1; 1027 #endif 1028 setup_clear_cpu_cap(X86_FEATURE_APIC); 1029 } 1030 1031 #ifdef CONFIG_PCI 1032 if (pci_early_dump_regs) 1033 early_dump_pci_devices(); 1034 #endif 1035 1036 e820__reserve_setup_data(); 1037 e820__finish_early_params(); 1038 1039 if (efi_enabled(EFI_BOOT)) 1040 efi_init(); 1041 1042 dmi_scan_machine(); 1043 dmi_memdev_walk(); 1044 dmi_set_dump_stack_arch_desc(); 1045 1046 /* 1047 * VMware detection requires dmi to be available, so this 1048 * needs to be done after dmi_scan_machine, for the BP. 1049 */ 1050 init_hypervisor_platform(); 1051 1052 simple_udelay_calibration(); 1053 1054 x86_init.resources.probe_roms(); 1055 1056 /* after parse_early_param, so could debug it */ 1057 insert_resource(&iomem_resource, &code_resource); 1058 insert_resource(&iomem_resource, &data_resource); 1059 insert_resource(&iomem_resource, &bss_resource); 1060 1061 e820_add_kernel_range(); 1062 trim_bios_range(); 1063 #ifdef CONFIG_X86_32 1064 if (ppro_with_ram_bug()) { 1065 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1066 E820_TYPE_RESERVED); 1067 e820__update_table(e820_table); 1068 printk(KERN_INFO "fixed physical RAM map:\n"); 1069 e820__print_table("bad_ppro"); 1070 } 1071 #else 1072 early_gart_iommu_check(); 1073 #endif 1074 1075 /* 1076 * partially used pages are not usable - thus 1077 * we are rounding upwards: 1078 */ 1079 max_pfn = e820__end_of_ram_pfn(); 1080 1081 /* update e820 for memory not covered by WB MTRRs */ 1082 mtrr_bp_init(); 1083 if (mtrr_trim_uncached_memory(max_pfn)) 1084 max_pfn = e820__end_of_ram_pfn(); 1085 1086 max_possible_pfn = max_pfn; 1087 1088 /* 1089 * This call is required when the CPU does not support PAT. If 1090 * mtrr_bp_init() invoked it already via pat_init() the call has no 1091 * effect. 1092 */ 1093 init_cache_modes(); 1094 1095 /* 1096 * Define random base addresses for memory sections after max_pfn is 1097 * defined and before each memory section base is used. 1098 */ 1099 kernel_randomize_memory(); 1100 1101 #ifdef CONFIG_X86_32 1102 /* max_low_pfn get updated here */ 1103 find_low_pfn_range(); 1104 #else 1105 check_x2apic(); 1106 1107 /* How many end-of-memory variables you have, grandma! */ 1108 /* need this before calling reserve_initrd */ 1109 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1110 max_low_pfn = e820__end_of_low_ram_pfn(); 1111 else 1112 max_low_pfn = max_pfn; 1113 1114 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1115 #endif 1116 1117 /* 1118 * Find and reserve possible boot-time SMP configuration: 1119 */ 1120 find_smp_config(); 1121 1122 reserve_ibft_region(); 1123 1124 early_alloc_pgt_buf(); 1125 1126 /* 1127 * Need to conclude brk, before e820__memblock_setup() 1128 * it could use memblock_find_in_range, could overlap with 1129 * brk area. 1130 */ 1131 reserve_brk(); 1132 1133 cleanup_highmap(); 1134 1135 memblock_set_current_limit(ISA_END_ADDRESS); 1136 e820__memblock_setup(); 1137 1138 if (!early_xdbc_setup_hardware()) 1139 early_xdbc_register_console(); 1140 1141 reserve_bios_regions(); 1142 1143 if (efi_enabled(EFI_MEMMAP)) { 1144 efi_fake_memmap(); 1145 efi_find_mirror(); 1146 efi_esrt_init(); 1147 1148 /* 1149 * The EFI specification says that boot service code won't be 1150 * called after ExitBootServices(). This is, in fact, a lie. 1151 */ 1152 efi_reserve_boot_services(); 1153 } 1154 1155 /* preallocate 4k for mptable mpc */ 1156 e820__memblock_alloc_reserved_mpc_new(); 1157 1158 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1159 setup_bios_corruption_check(); 1160 #endif 1161 1162 #ifdef CONFIG_X86_32 1163 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1164 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1165 #endif 1166 1167 reserve_real_mode(); 1168 1169 trim_platform_memory_ranges(); 1170 trim_low_memory_range(); 1171 1172 init_mem_mapping(); 1173 1174 idt_setup_early_pf(); 1175 1176 /* 1177 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1178 * with the current CR4 value. This may not be necessary, but 1179 * auditing all the early-boot CR4 manipulation would be needed to 1180 * rule it out. 1181 * 1182 * Mask off features that don't work outside long mode (just 1183 * PCIDE for now). 1184 */ 1185 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1186 1187 memblock_set_current_limit(get_max_mapped()); 1188 1189 /* 1190 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1191 */ 1192 1193 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1194 if (init_ohci1394_dma_early) 1195 init_ohci1394_dma_on_all_controllers(); 1196 #endif 1197 /* Allocate bigger log buffer */ 1198 setup_log_buf(1); 1199 1200 if (efi_enabled(EFI_BOOT)) { 1201 switch (boot_params.secure_boot) { 1202 case efi_secureboot_mode_disabled: 1203 pr_info("Secure boot disabled\n"); 1204 break; 1205 case efi_secureboot_mode_enabled: 1206 pr_info("Secure boot enabled\n"); 1207 break; 1208 default: 1209 pr_info("Secure boot could not be determined\n"); 1210 break; 1211 } 1212 } 1213 1214 reserve_initrd(); 1215 1216 acpi_table_upgrade(); 1217 1218 vsmp_init(); 1219 1220 io_delay_init(); 1221 1222 early_platform_quirks(); 1223 1224 /* 1225 * Parse the ACPI tables for possible boot-time SMP configuration. 1226 */ 1227 acpi_boot_table_init(); 1228 1229 early_acpi_boot_init(); 1230 1231 initmem_init(); 1232 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1233 1234 /* 1235 * Reserve memory for crash kernel after SRAT is parsed so that it 1236 * won't consume hotpluggable memory. 1237 */ 1238 reserve_crashkernel(); 1239 1240 memblock_find_dma_reserve(); 1241 1242 #ifdef CONFIG_KVM_GUEST 1243 kvmclock_init(); 1244 #endif 1245 1246 x86_init.paging.pagetable_init(); 1247 1248 kasan_init(); 1249 1250 #ifdef CONFIG_X86_32 1251 /* sync back kernel address range */ 1252 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1253 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1254 KERNEL_PGD_PTRS); 1255 1256 /* 1257 * sync back low identity map too. It is used for example 1258 * in the 32-bit EFI stub. 1259 */ 1260 clone_pgd_range(initial_page_table, 1261 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1262 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 1263 #endif 1264 1265 tboot_probe(); 1266 1267 map_vsyscall(); 1268 1269 generic_apic_probe(); 1270 1271 early_quirks(); 1272 1273 /* 1274 * Read APIC and some other early information from ACPI tables. 1275 */ 1276 acpi_boot_init(); 1277 sfi_init(); 1278 x86_dtb_init(); 1279 1280 /* 1281 * get boot-time SMP configuration: 1282 */ 1283 get_smp_config(); 1284 1285 /* 1286 * Systems w/o ACPI and mptables might not have it mapped the local 1287 * APIC yet, but prefill_possible_map() might need to access it. 1288 */ 1289 init_apic_mappings(); 1290 1291 prefill_possible_map(); 1292 1293 init_cpu_to_node(); 1294 1295 io_apic_init_mappings(); 1296 1297 kvm_guest_init(); 1298 1299 e820__reserve_resources(); 1300 e820__register_nosave_regions(max_low_pfn); 1301 1302 x86_init.resources.reserve_resources(); 1303 1304 e820__setup_pci_gap(); 1305 1306 #ifdef CONFIG_VT 1307 #if defined(CONFIG_VGA_CONSOLE) 1308 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1309 conswitchp = &vga_con; 1310 #elif defined(CONFIG_DUMMY_CONSOLE) 1311 conswitchp = &dummy_con; 1312 #endif 1313 #endif 1314 x86_init.oem.banner(); 1315 1316 x86_init.timers.wallclock_init(); 1317 1318 mcheck_init(); 1319 1320 arch_init_ideal_nops(); 1321 1322 register_refined_jiffies(CLOCK_TICK_RATE); 1323 1324 #ifdef CONFIG_EFI 1325 if (efi_enabled(EFI_BOOT)) 1326 efi_apply_memmap_quirks(); 1327 #endif 1328 1329 unwind_init(); 1330 } 1331 1332 #ifdef CONFIG_X86_32 1333 1334 static struct resource video_ram_resource = { 1335 .name = "Video RAM area", 1336 .start = 0xa0000, 1337 .end = 0xbffff, 1338 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1339 }; 1340 1341 void __init i386_reserve_resources(void) 1342 { 1343 request_resource(&iomem_resource, &video_ram_resource); 1344 reserve_standard_io_resources(); 1345 } 1346 1347 #endif /* CONFIG_X86_32 */ 1348 1349 static struct notifier_block kernel_offset_notifier = { 1350 .notifier_call = dump_kernel_offset 1351 }; 1352 1353 static int __init register_kernel_offset_dumper(void) 1354 { 1355 atomic_notifier_chain_register(&panic_notifier_list, 1356 &kernel_offset_notifier); 1357 return 0; 1358 } 1359 __initcall(register_kernel_offset_dumper); 1360 1361 void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 1362 { 1363 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 1364 return; 1365 1366 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1367 } 1368