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 73 #include <linux/usb/xhci-dbgp.h> 74 #include <video/edid.h> 75 76 #include <asm/mtrr.h> 77 #include <asm/apic.h> 78 #include <asm/realmode.h> 79 #include <asm/e820/api.h> 80 #include <asm/mpspec.h> 81 #include <asm/setup.h> 82 #include <asm/efi.h> 83 #include <asm/timer.h> 84 #include <asm/i8259.h> 85 #include <asm/sections.h> 86 #include <asm/io_apic.h> 87 #include <asm/ist.h> 88 #include <asm/setup_arch.h> 89 #include <asm/bios_ebda.h> 90 #include <asm/cacheflush.h> 91 #include <asm/processor.h> 92 #include <asm/bugs.h> 93 #include <asm/kasan.h> 94 95 #include <asm/vsyscall.h> 96 #include <asm/cpu.h> 97 #include <asm/desc.h> 98 #include <asm/dma.h> 99 #include <asm/iommu.h> 100 #include <asm/gart.h> 101 #include <asm/mmu_context.h> 102 #include <asm/proto.h> 103 104 #include <asm/paravirt.h> 105 #include <asm/hypervisor.h> 106 #include <asm/olpc_ofw.h> 107 108 #include <asm/percpu.h> 109 #include <asm/topology.h> 110 #include <asm/apicdef.h> 111 #include <asm/amd_nb.h> 112 #include <asm/mce.h> 113 #include <asm/alternative.h> 114 #include <asm/prom.h> 115 #include <asm/microcode.h> 116 #include <asm/mmu_context.h> 117 #include <asm/kaslr.h> 118 119 /* 120 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 121 * max_pfn_mapped: highest direct mapped pfn over 4GB 122 * 123 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 124 * represented by pfn_mapped 125 */ 126 unsigned long max_low_pfn_mapped; 127 unsigned long max_pfn_mapped; 128 129 #ifdef CONFIG_DMI 130 RESERVE_BRK(dmi_alloc, 65536); 131 #endif 132 133 134 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 135 unsigned long _brk_end = (unsigned long)__brk_base; 136 137 #ifdef CONFIG_X86_64 138 int default_cpu_present_to_apicid(int mps_cpu) 139 { 140 return __default_cpu_present_to_apicid(mps_cpu); 141 } 142 143 int default_check_phys_apicid_present(int phys_apicid) 144 { 145 return __default_check_phys_apicid_present(phys_apicid); 146 } 147 #endif 148 149 struct boot_params boot_params; 150 151 /* 152 * Machine setup.. 153 */ 154 static struct resource data_resource = { 155 .name = "Kernel data", 156 .start = 0, 157 .end = 0, 158 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 159 }; 160 161 static struct resource code_resource = { 162 .name = "Kernel code", 163 .start = 0, 164 .end = 0, 165 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 166 }; 167 168 static struct resource bss_resource = { 169 .name = "Kernel bss", 170 .start = 0, 171 .end = 0, 172 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 173 }; 174 175 176 #ifdef CONFIG_X86_32 177 /* cpu data as detected by the assembly code in head_32.S */ 178 struct cpuinfo_x86 new_cpu_data; 179 180 /* common cpu data for all cpus */ 181 struct cpuinfo_x86 boot_cpu_data __read_mostly; 182 EXPORT_SYMBOL(boot_cpu_data); 183 184 unsigned int def_to_bigsmp; 185 186 /* for MCA, but anyone else can use it if they want */ 187 unsigned int machine_id; 188 unsigned int machine_submodel_id; 189 unsigned int BIOS_revision; 190 191 struct apm_info apm_info; 192 EXPORT_SYMBOL(apm_info); 193 194 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 195 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 196 struct ist_info ist_info; 197 EXPORT_SYMBOL(ist_info); 198 #else 199 struct ist_info ist_info; 200 #endif 201 202 #else 203 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 204 .x86_phys_bits = MAX_PHYSMEM_BITS, 205 }; 206 EXPORT_SYMBOL(boot_cpu_data); 207 #endif 208 209 210 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 211 __visible unsigned long mmu_cr4_features __ro_after_init; 212 #else 213 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 214 #endif 215 216 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 217 int bootloader_type, bootloader_version; 218 219 /* 220 * Setup options 221 */ 222 struct screen_info screen_info; 223 EXPORT_SYMBOL(screen_info); 224 struct edid_info edid_info; 225 EXPORT_SYMBOL_GPL(edid_info); 226 227 extern int root_mountflags; 228 229 unsigned long saved_video_mode; 230 231 #define RAMDISK_IMAGE_START_MASK 0x07FF 232 #define RAMDISK_PROMPT_FLAG 0x8000 233 #define RAMDISK_LOAD_FLAG 0x4000 234 235 static char __initdata command_line[COMMAND_LINE_SIZE]; 236 #ifdef CONFIG_CMDLINE_BOOL 237 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 238 #endif 239 240 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 241 struct edd edd; 242 #ifdef CONFIG_EDD_MODULE 243 EXPORT_SYMBOL(edd); 244 #endif 245 /** 246 * copy_edd() - Copy the BIOS EDD information 247 * from boot_params into a safe place. 248 * 249 */ 250 static inline void __init copy_edd(void) 251 { 252 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 253 sizeof(edd.mbr_signature)); 254 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 255 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 256 edd.edd_info_nr = boot_params.eddbuf_entries; 257 } 258 #else 259 static inline void __init copy_edd(void) 260 { 261 } 262 #endif 263 264 void * __init extend_brk(size_t size, size_t align) 265 { 266 size_t mask = align - 1; 267 void *ret; 268 269 BUG_ON(_brk_start == 0); 270 BUG_ON(align & mask); 271 272 _brk_end = (_brk_end + mask) & ~mask; 273 BUG_ON((char *)(_brk_end + size) > __brk_limit); 274 275 ret = (void *)_brk_end; 276 _brk_end += size; 277 278 memset(ret, 0, size); 279 280 return ret; 281 } 282 283 #ifdef CONFIG_X86_32 284 static void __init cleanup_highmap(void) 285 { 286 } 287 #endif 288 289 static void __init reserve_brk(void) 290 { 291 if (_brk_end > _brk_start) 292 memblock_reserve(__pa_symbol(_brk_start), 293 _brk_end - _brk_start); 294 295 /* Mark brk area as locked down and no longer taking any 296 new allocations */ 297 _brk_start = 0; 298 } 299 300 u64 relocated_ramdisk; 301 302 #ifdef CONFIG_BLK_DEV_INITRD 303 304 static u64 __init get_ramdisk_image(void) 305 { 306 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 307 308 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 309 310 return ramdisk_image; 311 } 312 static u64 __init get_ramdisk_size(void) 313 { 314 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 315 316 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 317 318 return ramdisk_size; 319 } 320 321 static void __init relocate_initrd(void) 322 { 323 /* Assume only end is not page aligned */ 324 u64 ramdisk_image = get_ramdisk_image(); 325 u64 ramdisk_size = get_ramdisk_size(); 326 u64 area_size = PAGE_ALIGN(ramdisk_size); 327 328 /* We need to move the initrd down into directly mapped mem */ 329 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 330 area_size, PAGE_SIZE); 331 332 if (!relocated_ramdisk) 333 panic("Cannot find place for new RAMDISK of size %lld\n", 334 ramdisk_size); 335 336 /* Note: this includes all the mem currently occupied by 337 the initrd, we rely on that fact to keep the data intact. */ 338 memblock_reserve(relocated_ramdisk, area_size); 339 initrd_start = relocated_ramdisk + PAGE_OFFSET; 340 initrd_end = initrd_start + ramdisk_size; 341 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 342 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 343 344 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 345 346 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 347 " [mem %#010llx-%#010llx]\n", 348 ramdisk_image, ramdisk_image + ramdisk_size - 1, 349 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 350 } 351 352 static void __init early_reserve_initrd(void) 353 { 354 /* Assume only end is not page aligned */ 355 u64 ramdisk_image = get_ramdisk_image(); 356 u64 ramdisk_size = get_ramdisk_size(); 357 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 358 359 if (!boot_params.hdr.type_of_loader || 360 !ramdisk_image || !ramdisk_size) 361 return; /* No initrd provided by bootloader */ 362 363 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 364 } 365 static void __init reserve_initrd(void) 366 { 367 /* Assume only end is not page aligned */ 368 u64 ramdisk_image = get_ramdisk_image(); 369 u64 ramdisk_size = get_ramdisk_size(); 370 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 371 u64 mapped_size; 372 373 if (!boot_params.hdr.type_of_loader || 374 !ramdisk_image || !ramdisk_size) 375 return; /* No initrd provided by bootloader */ 376 377 initrd_start = 0; 378 379 mapped_size = memblock_mem_size(max_pfn_mapped); 380 if (ramdisk_size >= (mapped_size>>1)) 381 panic("initrd too large to handle, " 382 "disabling initrd (%lld needed, %lld available)\n", 383 ramdisk_size, mapped_size>>1); 384 385 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 386 ramdisk_end - 1); 387 388 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 389 PFN_DOWN(ramdisk_end))) { 390 /* All are mapped, easy case */ 391 initrd_start = ramdisk_image + PAGE_OFFSET; 392 initrd_end = initrd_start + ramdisk_size; 393 return; 394 } 395 396 relocate_initrd(); 397 398 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 399 } 400 401 #else 402 static void __init early_reserve_initrd(void) 403 { 404 } 405 static void __init reserve_initrd(void) 406 { 407 } 408 #endif /* CONFIG_BLK_DEV_INITRD */ 409 410 static void __init parse_setup_data(void) 411 { 412 struct setup_data *data; 413 u64 pa_data, pa_next; 414 415 pa_data = boot_params.hdr.setup_data; 416 while (pa_data) { 417 u32 data_len, data_type; 418 419 data = early_memremap(pa_data, sizeof(*data)); 420 data_len = data->len + sizeof(struct setup_data); 421 data_type = data->type; 422 pa_next = data->next; 423 early_memunmap(data, sizeof(*data)); 424 425 switch (data_type) { 426 case SETUP_E820_EXT: 427 e820__memory_setup_extended(pa_data, data_len); 428 break; 429 case SETUP_DTB: 430 add_dtb(pa_data); 431 break; 432 case SETUP_EFI: 433 parse_efi_setup(pa_data, data_len); 434 break; 435 default: 436 break; 437 } 438 pa_data = pa_next; 439 } 440 } 441 442 static void __init memblock_x86_reserve_range_setup_data(void) 443 { 444 struct setup_data *data; 445 u64 pa_data; 446 447 pa_data = boot_params.hdr.setup_data; 448 while (pa_data) { 449 data = early_memremap(pa_data, sizeof(*data)); 450 memblock_reserve(pa_data, sizeof(*data) + data->len); 451 pa_data = data->next; 452 early_memunmap(data, sizeof(*data)); 453 } 454 } 455 456 /* 457 * --------- Crashkernel reservation ------------------------------ 458 */ 459 460 #ifdef CONFIG_KEXEC_CORE 461 462 /* 16M alignment for crash kernel regions */ 463 #define CRASH_ALIGN (16 << 20) 464 465 /* 466 * Keep the crash kernel below this limit. On 32 bits earlier kernels 467 * would limit the kernel to the low 512 MiB due to mapping restrictions. 468 * On 64bit, old kexec-tools need to under 896MiB. 469 */ 470 #ifdef CONFIG_X86_32 471 # define CRASH_ADDR_LOW_MAX (512 << 20) 472 # define CRASH_ADDR_HIGH_MAX (512 << 20) 473 #else 474 # define CRASH_ADDR_LOW_MAX (896UL << 20) 475 # define CRASH_ADDR_HIGH_MAX MAXMEM 476 #endif 477 478 static int __init reserve_crashkernel_low(void) 479 { 480 #ifdef CONFIG_X86_64 481 unsigned long long base, low_base = 0, low_size = 0; 482 unsigned long total_low_mem; 483 int ret; 484 485 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 486 487 /* crashkernel=Y,low */ 488 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 489 if (ret) { 490 /* 491 * two parts from lib/swiotlb.c: 492 * -swiotlb size: user-specified with swiotlb= or default. 493 * 494 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 495 * to 8M for other buffers that may need to stay low too. Also 496 * make sure we allocate enough extra low memory so that we 497 * don't run out of DMA buffers for 32-bit devices. 498 */ 499 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 500 } else { 501 /* passed with crashkernel=0,low ? */ 502 if (!low_size) 503 return 0; 504 } 505 506 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 507 if (!low_base) { 508 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 509 (unsigned long)(low_size >> 20)); 510 return -ENOMEM; 511 } 512 513 ret = memblock_reserve(low_base, low_size); 514 if (ret) { 515 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 516 return ret; 517 } 518 519 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 520 (unsigned long)(low_size >> 20), 521 (unsigned long)(low_base >> 20), 522 (unsigned long)(total_low_mem >> 20)); 523 524 crashk_low_res.start = low_base; 525 crashk_low_res.end = low_base + low_size - 1; 526 insert_resource(&iomem_resource, &crashk_low_res); 527 #endif 528 return 0; 529 } 530 531 static void __init reserve_crashkernel(void) 532 { 533 unsigned long long crash_size, crash_base, total_mem; 534 bool high = false; 535 int ret; 536 537 total_mem = memblock_phys_mem_size(); 538 539 /* crashkernel=XM */ 540 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 541 if (ret != 0 || crash_size <= 0) { 542 /* crashkernel=X,high */ 543 ret = parse_crashkernel_high(boot_command_line, total_mem, 544 &crash_size, &crash_base); 545 if (ret != 0 || crash_size <= 0) 546 return; 547 high = true; 548 } 549 550 /* 0 means: find the address automatically */ 551 if (crash_base <= 0) { 552 /* 553 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 554 * as old kexec-tools loads bzImage below that, unless 555 * "crashkernel=size[KMG],high" is specified. 556 */ 557 crash_base = memblock_find_in_range(CRASH_ALIGN, 558 high ? CRASH_ADDR_HIGH_MAX 559 : CRASH_ADDR_LOW_MAX, 560 crash_size, CRASH_ALIGN); 561 if (!crash_base) { 562 pr_info("crashkernel reservation failed - No suitable area found.\n"); 563 return; 564 } 565 566 } else { 567 unsigned long long start; 568 569 start = memblock_find_in_range(crash_base, 570 crash_base + crash_size, 571 crash_size, 1 << 20); 572 if (start != crash_base) { 573 pr_info("crashkernel reservation failed - memory is in use.\n"); 574 return; 575 } 576 } 577 ret = memblock_reserve(crash_base, crash_size); 578 if (ret) { 579 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 580 return; 581 } 582 583 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 584 memblock_free(crash_base, crash_size); 585 return; 586 } 587 588 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 589 (unsigned long)(crash_size >> 20), 590 (unsigned long)(crash_base >> 20), 591 (unsigned long)(total_mem >> 20)); 592 593 crashk_res.start = crash_base; 594 crashk_res.end = crash_base + crash_size - 1; 595 insert_resource(&iomem_resource, &crashk_res); 596 } 597 #else 598 static void __init reserve_crashkernel(void) 599 { 600 } 601 #endif 602 603 static struct resource standard_io_resources[] = { 604 { .name = "dma1", .start = 0x00, .end = 0x1f, 605 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 606 { .name = "pic1", .start = 0x20, .end = 0x21, 607 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 608 { .name = "timer0", .start = 0x40, .end = 0x43, 609 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 610 { .name = "timer1", .start = 0x50, .end = 0x53, 611 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 612 { .name = "keyboard", .start = 0x60, .end = 0x60, 613 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 614 { .name = "keyboard", .start = 0x64, .end = 0x64, 615 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 616 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 617 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 618 { .name = "pic2", .start = 0xa0, .end = 0xa1, 619 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 620 { .name = "dma2", .start = 0xc0, .end = 0xdf, 621 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 622 { .name = "fpu", .start = 0xf0, .end = 0xff, 623 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 624 }; 625 626 void __init reserve_standard_io_resources(void) 627 { 628 int i; 629 630 /* request I/O space for devices used on all i[345]86 PCs */ 631 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 632 request_resource(&ioport_resource, &standard_io_resources[i]); 633 634 } 635 636 static __init void reserve_ibft_region(void) 637 { 638 unsigned long addr, size = 0; 639 640 addr = find_ibft_region(&size); 641 642 if (size) 643 memblock_reserve(addr, size); 644 } 645 646 static bool __init snb_gfx_workaround_needed(void) 647 { 648 #ifdef CONFIG_PCI 649 int i; 650 u16 vendor, devid; 651 static const __initconst u16 snb_ids[] = { 652 0x0102, 653 0x0112, 654 0x0122, 655 0x0106, 656 0x0116, 657 0x0126, 658 0x010a, 659 }; 660 661 /* Assume no if something weird is going on with PCI */ 662 if (!early_pci_allowed()) 663 return false; 664 665 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 666 if (vendor != 0x8086) 667 return false; 668 669 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 670 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 671 if (devid == snb_ids[i]) 672 return true; 673 #endif 674 675 return false; 676 } 677 678 /* 679 * Sandy Bridge graphics has trouble with certain ranges, exclude 680 * them from allocation. 681 */ 682 static void __init trim_snb_memory(void) 683 { 684 static const __initconst unsigned long bad_pages[] = { 685 0x20050000, 686 0x20110000, 687 0x20130000, 688 0x20138000, 689 0x40004000, 690 }; 691 int i; 692 693 if (!snb_gfx_workaround_needed()) 694 return; 695 696 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 697 698 /* 699 * Reserve all memory below the 1 MB mark that has not 700 * already been reserved. 701 */ 702 memblock_reserve(0, 1<<20); 703 704 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 705 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 706 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 707 bad_pages[i]); 708 } 709 } 710 711 /* 712 * Here we put platform-specific memory range workarounds, i.e. 713 * memory known to be corrupt or otherwise in need to be reserved on 714 * specific platforms. 715 * 716 * If this gets used more widely it could use a real dispatch mechanism. 717 */ 718 static void __init trim_platform_memory_ranges(void) 719 { 720 trim_snb_memory(); 721 } 722 723 static void __init trim_bios_range(void) 724 { 725 /* 726 * A special case is the first 4Kb of memory; 727 * This is a BIOS owned area, not kernel ram, but generally 728 * not listed as such in the E820 table. 729 * 730 * This typically reserves additional memory (64KiB by default) 731 * since some BIOSes are known to corrupt low memory. See the 732 * Kconfig help text for X86_RESERVE_LOW. 733 */ 734 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 735 736 /* 737 * special case: Some BIOSen report the PC BIOS 738 * area (640->1Mb) as ram even though it is not. 739 * take them out. 740 */ 741 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 742 743 e820__update_table(e820_table); 744 } 745 746 /* called before trim_bios_range() to spare extra sanitize */ 747 static void __init e820_add_kernel_range(void) 748 { 749 u64 start = __pa_symbol(_text); 750 u64 size = __pa_symbol(_end) - start; 751 752 /* 753 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 754 * attempt to fix it by adding the range. We may have a confused BIOS, 755 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 756 * exclude kernel range. If we really are running on top non-RAM, 757 * we will crash later anyways. 758 */ 759 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 760 return; 761 762 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 763 e820__range_remove(start, size, E820_TYPE_RAM, 0); 764 e820__range_add(start, size, E820_TYPE_RAM); 765 } 766 767 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 768 769 static int __init parse_reservelow(char *p) 770 { 771 unsigned long long size; 772 773 if (!p) 774 return -EINVAL; 775 776 size = memparse(p, &p); 777 778 if (size < 4096) 779 size = 4096; 780 781 if (size > 640*1024) 782 size = 640*1024; 783 784 reserve_low = size; 785 786 return 0; 787 } 788 789 early_param("reservelow", parse_reservelow); 790 791 static void __init trim_low_memory_range(void) 792 { 793 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 794 } 795 796 /* 797 * Dump out kernel offset information on panic. 798 */ 799 static int 800 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 801 { 802 if (kaslr_enabled()) { 803 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 804 kaslr_offset(), 805 __START_KERNEL, 806 __START_KERNEL_map, 807 MODULES_VADDR-1); 808 } else { 809 pr_emerg("Kernel Offset: disabled\n"); 810 } 811 812 return 0; 813 } 814 815 static void __init simple_udelay_calibration(void) 816 { 817 unsigned int tsc_khz, cpu_khz; 818 unsigned long lpj; 819 820 if (!boot_cpu_has(X86_FEATURE_TSC)) 821 return; 822 823 cpu_khz = x86_platform.calibrate_cpu(); 824 tsc_khz = x86_platform.calibrate_tsc(); 825 826 tsc_khz = tsc_khz ? : cpu_khz; 827 if (!tsc_khz) 828 return; 829 830 lpj = tsc_khz * 1000; 831 do_div(lpj, HZ); 832 loops_per_jiffy = lpj; 833 } 834 835 /* 836 * Determine if we were loaded by an EFI loader. If so, then we have also been 837 * passed the efi memmap, systab, etc., so we should use these data structures 838 * for initialization. Note, the efi init code path is determined by the 839 * global efi_enabled. This allows the same kernel image to be used on existing 840 * systems (with a traditional BIOS) as well as on EFI systems. 841 */ 842 /* 843 * setup_arch - architecture-specific boot-time initializations 844 * 845 * Note: On x86_64, fixmaps are ready for use even before this is called. 846 */ 847 848 void __init setup_arch(char **cmdline_p) 849 { 850 memblock_reserve(__pa_symbol(_text), 851 (unsigned long)__bss_stop - (unsigned long)_text); 852 853 early_reserve_initrd(); 854 855 /* 856 * At this point everything still needed from the boot loader 857 * or BIOS or kernel text should be early reserved or marked not 858 * RAM in e820. All other memory is free game. 859 */ 860 861 #ifdef CONFIG_X86_32 862 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 863 864 /* 865 * copy kernel address range established so far and switch 866 * to the proper swapper page table 867 */ 868 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 869 initial_page_table + KERNEL_PGD_BOUNDARY, 870 KERNEL_PGD_PTRS); 871 872 load_cr3(swapper_pg_dir); 873 /* 874 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 875 * a cr3 based tlb flush, so the following __flush_tlb_all() 876 * will not flush anything because the cpu quirk which clears 877 * X86_FEATURE_PGE has not been invoked yet. Though due to the 878 * load_cr3() above the TLB has been flushed already. The 879 * quirk is invoked before subsequent calls to __flush_tlb_all() 880 * so proper operation is guaranteed. 881 */ 882 __flush_tlb_all(); 883 #else 884 printk(KERN_INFO "Command line: %s\n", boot_command_line); 885 #endif 886 887 /* 888 * If we have OLPC OFW, we might end up relocating the fixmap due to 889 * reserve_top(), so do this before touching the ioremap area. 890 */ 891 olpc_ofw_detect(); 892 893 early_trap_init(); 894 early_cpu_init(); 895 early_ioremap_init(); 896 897 setup_olpc_ofw_pgd(); 898 899 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 900 screen_info = boot_params.screen_info; 901 edid_info = boot_params.edid_info; 902 #ifdef CONFIG_X86_32 903 apm_info.bios = boot_params.apm_bios_info; 904 ist_info = boot_params.ist_info; 905 #endif 906 saved_video_mode = boot_params.hdr.vid_mode; 907 bootloader_type = boot_params.hdr.type_of_loader; 908 if ((bootloader_type >> 4) == 0xe) { 909 bootloader_type &= 0xf; 910 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 911 } 912 bootloader_version = bootloader_type & 0xf; 913 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 914 915 #ifdef CONFIG_BLK_DEV_RAM 916 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 917 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 918 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 919 #endif 920 #ifdef CONFIG_EFI 921 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 922 EFI32_LOADER_SIGNATURE, 4)) { 923 set_bit(EFI_BOOT, &efi.flags); 924 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 925 EFI64_LOADER_SIGNATURE, 4)) { 926 set_bit(EFI_BOOT, &efi.flags); 927 set_bit(EFI_64BIT, &efi.flags); 928 } 929 930 if (efi_enabled(EFI_BOOT)) 931 efi_memblock_x86_reserve_range(); 932 #endif 933 934 x86_init.oem.arch_setup(); 935 936 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 937 e820__memory_setup(); 938 parse_setup_data(); 939 940 copy_edd(); 941 942 if (!boot_params.hdr.root_flags) 943 root_mountflags &= ~MS_RDONLY; 944 init_mm.start_code = (unsigned long) _text; 945 init_mm.end_code = (unsigned long) _etext; 946 init_mm.end_data = (unsigned long) _edata; 947 init_mm.brk = _brk_end; 948 949 mpx_mm_init(&init_mm); 950 951 code_resource.start = __pa_symbol(_text); 952 code_resource.end = __pa_symbol(_etext)-1; 953 data_resource.start = __pa_symbol(_etext); 954 data_resource.end = __pa_symbol(_edata)-1; 955 bss_resource.start = __pa_symbol(__bss_start); 956 bss_resource.end = __pa_symbol(__bss_stop)-1; 957 958 #ifdef CONFIG_CMDLINE_BOOL 959 #ifdef CONFIG_CMDLINE_OVERRIDE 960 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 961 #else 962 if (builtin_cmdline[0]) { 963 /* append boot loader cmdline to builtin */ 964 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 965 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 966 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 967 } 968 #endif 969 #endif 970 971 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 972 *cmdline_p = command_line; 973 974 /* 975 * x86_configure_nx() is called before parse_early_param() to detect 976 * whether hardware doesn't support NX (so that the early EHCI debug 977 * console setup can safely call set_fixmap()). It may then be called 978 * again from within noexec_setup() during parsing early parameters 979 * to honor the respective command line option. 980 */ 981 x86_configure_nx(); 982 983 parse_early_param(); 984 985 #ifdef CONFIG_MEMORY_HOTPLUG 986 /* 987 * Memory used by the kernel cannot be hot-removed because Linux 988 * cannot migrate the kernel pages. When memory hotplug is 989 * enabled, we should prevent memblock from allocating memory 990 * for the kernel. 991 * 992 * ACPI SRAT records all hotpluggable memory ranges. But before 993 * SRAT is parsed, we don't know about it. 994 * 995 * The kernel image is loaded into memory at very early time. We 996 * cannot prevent this anyway. So on NUMA system, we set any 997 * node the kernel resides in as un-hotpluggable. 998 * 999 * Since on modern servers, one node could have double-digit 1000 * gigabytes memory, we can assume the memory around the kernel 1001 * image is also un-hotpluggable. So before SRAT is parsed, just 1002 * allocate memory near the kernel image to try the best to keep 1003 * the kernel away from hotpluggable memory. 1004 */ 1005 if (movable_node_is_enabled()) 1006 memblock_set_bottom_up(true); 1007 #endif 1008 1009 x86_report_nx(); 1010 1011 /* after early param, so could get panic from serial */ 1012 memblock_x86_reserve_range_setup_data(); 1013 1014 if (acpi_mps_check()) { 1015 #ifdef CONFIG_X86_LOCAL_APIC 1016 disable_apic = 1; 1017 #endif 1018 setup_clear_cpu_cap(X86_FEATURE_APIC); 1019 } 1020 1021 #ifdef CONFIG_PCI 1022 if (pci_early_dump_regs) 1023 early_dump_pci_devices(); 1024 #endif 1025 1026 e820__reserve_setup_data(); 1027 e820__finish_early_params(); 1028 1029 if (efi_enabled(EFI_BOOT)) 1030 efi_init(); 1031 1032 dmi_scan_machine(); 1033 dmi_memdev_walk(); 1034 dmi_set_dump_stack_arch_desc(); 1035 1036 /* 1037 * VMware detection requires dmi to be available, so this 1038 * needs to be done after dmi_scan_machine, for the BP. 1039 */ 1040 init_hypervisor_platform(); 1041 1042 simple_udelay_calibration(); 1043 1044 x86_init.resources.probe_roms(); 1045 1046 /* after parse_early_param, so could debug it */ 1047 insert_resource(&iomem_resource, &code_resource); 1048 insert_resource(&iomem_resource, &data_resource); 1049 insert_resource(&iomem_resource, &bss_resource); 1050 1051 e820_add_kernel_range(); 1052 trim_bios_range(); 1053 #ifdef CONFIG_X86_32 1054 if (ppro_with_ram_bug()) { 1055 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1056 E820_TYPE_RESERVED); 1057 e820__update_table(e820_table); 1058 printk(KERN_INFO "fixed physical RAM map:\n"); 1059 e820__print_table("bad_ppro"); 1060 } 1061 #else 1062 early_gart_iommu_check(); 1063 #endif 1064 1065 /* 1066 * partially used pages are not usable - thus 1067 * we are rounding upwards: 1068 */ 1069 max_pfn = e820__end_of_ram_pfn(); 1070 1071 /* update e820 for memory not covered by WB MTRRs */ 1072 mtrr_bp_init(); 1073 if (mtrr_trim_uncached_memory(max_pfn)) 1074 max_pfn = e820__end_of_ram_pfn(); 1075 1076 max_possible_pfn = max_pfn; 1077 1078 /* 1079 * This call is required when the CPU does not support PAT. If 1080 * mtrr_bp_init() invoked it already via pat_init() the call has no 1081 * effect. 1082 */ 1083 init_cache_modes(); 1084 1085 /* 1086 * Define random base addresses for memory sections after max_pfn is 1087 * defined and before each memory section base is used. 1088 */ 1089 kernel_randomize_memory(); 1090 1091 #ifdef CONFIG_X86_32 1092 /* max_low_pfn get updated here */ 1093 find_low_pfn_range(); 1094 #else 1095 check_x2apic(); 1096 1097 /* How many end-of-memory variables you have, grandma! */ 1098 /* need this before calling reserve_initrd */ 1099 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1100 max_low_pfn = e820__end_of_low_ram_pfn(); 1101 else 1102 max_low_pfn = max_pfn; 1103 1104 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1105 #endif 1106 1107 /* 1108 * Find and reserve possible boot-time SMP configuration: 1109 */ 1110 find_smp_config(); 1111 1112 reserve_ibft_region(); 1113 1114 early_alloc_pgt_buf(); 1115 1116 /* 1117 * Need to conclude brk, before e820__memblock_setup() 1118 * it could use memblock_find_in_range, could overlap with 1119 * brk area. 1120 */ 1121 reserve_brk(); 1122 1123 cleanup_highmap(); 1124 1125 memblock_set_current_limit(ISA_END_ADDRESS); 1126 e820__memblock_setup(); 1127 1128 if (!early_xdbc_setup_hardware()) 1129 early_xdbc_register_console(); 1130 1131 reserve_bios_regions(); 1132 1133 if (efi_enabled(EFI_MEMMAP)) { 1134 efi_fake_memmap(); 1135 efi_find_mirror(); 1136 efi_esrt_init(); 1137 1138 /* 1139 * The EFI specification says that boot service code won't be 1140 * called after ExitBootServices(). This is, in fact, a lie. 1141 */ 1142 efi_reserve_boot_services(); 1143 } 1144 1145 /* preallocate 4k for mptable mpc */ 1146 e820__memblock_alloc_reserved_mpc_new(); 1147 1148 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1149 setup_bios_corruption_check(); 1150 #endif 1151 1152 #ifdef CONFIG_X86_32 1153 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1154 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1155 #endif 1156 1157 reserve_real_mode(); 1158 1159 trim_platform_memory_ranges(); 1160 trim_low_memory_range(); 1161 1162 init_mem_mapping(); 1163 1164 early_trap_pf_init(); 1165 1166 /* 1167 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1168 * with the current CR4 value. This may not be necessary, but 1169 * auditing all the early-boot CR4 manipulation would be needed to 1170 * rule it out. 1171 */ 1172 mmu_cr4_features = __read_cr4(); 1173 1174 memblock_set_current_limit(get_max_mapped()); 1175 1176 /* 1177 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1178 */ 1179 1180 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1181 if (init_ohci1394_dma_early) 1182 init_ohci1394_dma_on_all_controllers(); 1183 #endif 1184 /* Allocate bigger log buffer */ 1185 setup_log_buf(1); 1186 1187 if (efi_enabled(EFI_BOOT)) { 1188 switch (boot_params.secure_boot) { 1189 case efi_secureboot_mode_disabled: 1190 pr_info("Secure boot disabled\n"); 1191 break; 1192 case efi_secureboot_mode_enabled: 1193 pr_info("Secure boot enabled\n"); 1194 break; 1195 default: 1196 pr_info("Secure boot could not be determined\n"); 1197 break; 1198 } 1199 } 1200 1201 reserve_initrd(); 1202 1203 acpi_table_upgrade(); 1204 1205 vsmp_init(); 1206 1207 io_delay_init(); 1208 1209 /* 1210 * Parse the ACPI tables for possible boot-time SMP configuration. 1211 */ 1212 acpi_boot_table_init(); 1213 1214 early_acpi_boot_init(); 1215 1216 initmem_init(); 1217 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1218 1219 /* 1220 * Reserve memory for crash kernel after SRAT is parsed so that it 1221 * won't consume hotpluggable memory. 1222 */ 1223 reserve_crashkernel(); 1224 1225 memblock_find_dma_reserve(); 1226 1227 #ifdef CONFIG_KVM_GUEST 1228 kvmclock_init(); 1229 #endif 1230 1231 x86_init.paging.pagetable_init(); 1232 1233 kasan_init(); 1234 1235 #ifdef CONFIG_X86_32 1236 /* sync back kernel address range */ 1237 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1238 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1239 KERNEL_PGD_PTRS); 1240 1241 /* 1242 * sync back low identity map too. It is used for example 1243 * in the 32-bit EFI stub. 1244 */ 1245 clone_pgd_range(initial_page_table, 1246 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1247 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 1248 #endif 1249 1250 tboot_probe(); 1251 1252 map_vsyscall(); 1253 1254 generic_apic_probe(); 1255 1256 early_quirks(); 1257 1258 /* 1259 * Read APIC and some other early information from ACPI tables. 1260 */ 1261 acpi_boot_init(); 1262 sfi_init(); 1263 x86_dtb_init(); 1264 1265 /* 1266 * get boot-time SMP configuration: 1267 */ 1268 get_smp_config(); 1269 1270 /* 1271 * Systems w/o ACPI and mptables might not have it mapped the local 1272 * APIC yet, but prefill_possible_map() might need to access it. 1273 */ 1274 init_apic_mappings(); 1275 1276 prefill_possible_map(); 1277 1278 init_cpu_to_node(); 1279 1280 io_apic_init_mappings(); 1281 1282 kvm_guest_init(); 1283 1284 e820__reserve_resources(); 1285 e820__register_nosave_regions(max_low_pfn); 1286 1287 x86_init.resources.reserve_resources(); 1288 1289 e820__setup_pci_gap(); 1290 1291 #ifdef CONFIG_VT 1292 #if defined(CONFIG_VGA_CONSOLE) 1293 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1294 conswitchp = &vga_con; 1295 #elif defined(CONFIG_DUMMY_CONSOLE) 1296 conswitchp = &dummy_con; 1297 #endif 1298 #endif 1299 x86_init.oem.banner(); 1300 1301 x86_init.timers.wallclock_init(); 1302 1303 mcheck_init(); 1304 1305 arch_init_ideal_nops(); 1306 1307 register_refined_jiffies(CLOCK_TICK_RATE); 1308 1309 #ifdef CONFIG_EFI 1310 if (efi_enabled(EFI_BOOT)) 1311 efi_apply_memmap_quirks(); 1312 #endif 1313 } 1314 1315 #ifdef CONFIG_X86_32 1316 1317 static struct resource video_ram_resource = { 1318 .name = "Video RAM area", 1319 .start = 0xa0000, 1320 .end = 0xbffff, 1321 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1322 }; 1323 1324 void __init i386_reserve_resources(void) 1325 { 1326 request_resource(&iomem_resource, &video_ram_resource); 1327 reserve_standard_io_resources(); 1328 } 1329 1330 #endif /* CONFIG_X86_32 */ 1331 1332 static struct notifier_block kernel_offset_notifier = { 1333 .notifier_call = dump_kernel_offset 1334 }; 1335 1336 static int __init register_kernel_offset_dumper(void) 1337 { 1338 atomic_notifier_chain_register(&panic_notifier_list, 1339 &kernel_offset_notifier); 1340 return 0; 1341 } 1342 __initcall(register_kernel_offset_dumper); 1343 1344 void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 1345 { 1346 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 1347 return; 1348 1349 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1350 } 1351