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 struct boot_params boot_params; 140 141 /* 142 * Machine setup.. 143 */ 144 static struct resource data_resource = { 145 .name = "Kernel data", 146 .start = 0, 147 .end = 0, 148 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 149 }; 150 151 static struct resource code_resource = { 152 .name = "Kernel code", 153 .start = 0, 154 .end = 0, 155 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 156 }; 157 158 static struct resource bss_resource = { 159 .name = "Kernel bss", 160 .start = 0, 161 .end = 0, 162 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 163 }; 164 165 166 #ifdef CONFIG_X86_32 167 /* cpu data as detected by the assembly code in head_32.S */ 168 struct cpuinfo_x86 new_cpu_data; 169 170 /* common cpu data for all cpus */ 171 struct cpuinfo_x86 boot_cpu_data __read_mostly; 172 EXPORT_SYMBOL(boot_cpu_data); 173 174 unsigned int def_to_bigsmp; 175 176 /* for MCA, but anyone else can use it if they want */ 177 unsigned int machine_id; 178 unsigned int machine_submodel_id; 179 unsigned int BIOS_revision; 180 181 struct apm_info apm_info; 182 EXPORT_SYMBOL(apm_info); 183 184 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 185 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 186 struct ist_info ist_info; 187 EXPORT_SYMBOL(ist_info); 188 #else 189 struct ist_info ist_info; 190 #endif 191 192 #else 193 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 194 .x86_phys_bits = MAX_PHYSMEM_BITS, 195 }; 196 EXPORT_SYMBOL(boot_cpu_data); 197 #endif 198 199 200 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 201 __visible unsigned long mmu_cr4_features __ro_after_init; 202 #else 203 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 204 #endif 205 206 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 207 int bootloader_type, bootloader_version; 208 209 /* 210 * Setup options 211 */ 212 struct screen_info screen_info; 213 EXPORT_SYMBOL(screen_info); 214 struct edid_info edid_info; 215 EXPORT_SYMBOL_GPL(edid_info); 216 217 extern int root_mountflags; 218 219 unsigned long saved_video_mode; 220 221 #define RAMDISK_IMAGE_START_MASK 0x07FF 222 #define RAMDISK_PROMPT_FLAG 0x8000 223 #define RAMDISK_LOAD_FLAG 0x4000 224 225 static char __initdata command_line[COMMAND_LINE_SIZE]; 226 #ifdef CONFIG_CMDLINE_BOOL 227 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 228 #endif 229 230 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 231 struct edd edd; 232 #ifdef CONFIG_EDD_MODULE 233 EXPORT_SYMBOL(edd); 234 #endif 235 /** 236 * copy_edd() - Copy the BIOS EDD information 237 * from boot_params into a safe place. 238 * 239 */ 240 static inline void __init copy_edd(void) 241 { 242 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 243 sizeof(edd.mbr_signature)); 244 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 245 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 246 edd.edd_info_nr = boot_params.eddbuf_entries; 247 } 248 #else 249 static inline void __init copy_edd(void) 250 { 251 } 252 #endif 253 254 void * __init extend_brk(size_t size, size_t align) 255 { 256 size_t mask = align - 1; 257 void *ret; 258 259 BUG_ON(_brk_start == 0); 260 BUG_ON(align & mask); 261 262 _brk_end = (_brk_end + mask) & ~mask; 263 BUG_ON((char *)(_brk_end + size) > __brk_limit); 264 265 ret = (void *)_brk_end; 266 _brk_end += size; 267 268 memset(ret, 0, size); 269 270 return ret; 271 } 272 273 #ifdef CONFIG_X86_32 274 static void __init cleanup_highmap(void) 275 { 276 } 277 #endif 278 279 static void __init reserve_brk(void) 280 { 281 if (_brk_end > _brk_start) 282 memblock_reserve(__pa_symbol(_brk_start), 283 _brk_end - _brk_start); 284 285 /* Mark brk area as locked down and no longer taking any 286 new allocations */ 287 _brk_start = 0; 288 } 289 290 u64 relocated_ramdisk; 291 292 #ifdef CONFIG_BLK_DEV_INITRD 293 294 static u64 __init get_ramdisk_image(void) 295 { 296 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 297 298 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 299 300 return ramdisk_image; 301 } 302 static u64 __init get_ramdisk_size(void) 303 { 304 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 305 306 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 307 308 return ramdisk_size; 309 } 310 311 static void __init relocate_initrd(void) 312 { 313 /* Assume only end is not page aligned */ 314 u64 ramdisk_image = get_ramdisk_image(); 315 u64 ramdisk_size = get_ramdisk_size(); 316 u64 area_size = PAGE_ALIGN(ramdisk_size); 317 318 /* We need to move the initrd down into directly mapped mem */ 319 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 320 area_size, PAGE_SIZE); 321 322 if (!relocated_ramdisk) 323 panic("Cannot find place for new RAMDISK of size %lld\n", 324 ramdisk_size); 325 326 /* Note: this includes all the mem currently occupied by 327 the initrd, we rely on that fact to keep the data intact. */ 328 memblock_reserve(relocated_ramdisk, area_size); 329 initrd_start = relocated_ramdisk + PAGE_OFFSET; 330 initrd_end = initrd_start + ramdisk_size; 331 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 332 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 333 334 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 335 336 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 337 " [mem %#010llx-%#010llx]\n", 338 ramdisk_image, ramdisk_image + ramdisk_size - 1, 339 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 340 } 341 342 static void __init early_reserve_initrd(void) 343 { 344 /* Assume only end is not page aligned */ 345 u64 ramdisk_image = get_ramdisk_image(); 346 u64 ramdisk_size = get_ramdisk_size(); 347 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 348 349 if (!boot_params.hdr.type_of_loader || 350 !ramdisk_image || !ramdisk_size) 351 return; /* No initrd provided by bootloader */ 352 353 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 354 } 355 static void __init reserve_initrd(void) 356 { 357 /* Assume only end is not page aligned */ 358 u64 ramdisk_image = get_ramdisk_image(); 359 u64 ramdisk_size = get_ramdisk_size(); 360 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 361 u64 mapped_size; 362 363 if (!boot_params.hdr.type_of_loader || 364 !ramdisk_image || !ramdisk_size) 365 return; /* No initrd provided by bootloader */ 366 367 initrd_start = 0; 368 369 mapped_size = memblock_mem_size(max_pfn_mapped); 370 if (ramdisk_size >= (mapped_size>>1)) 371 panic("initrd too large to handle, " 372 "disabling initrd (%lld needed, %lld available)\n", 373 ramdisk_size, mapped_size>>1); 374 375 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 376 ramdisk_end - 1); 377 378 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 379 PFN_DOWN(ramdisk_end))) { 380 /* All are mapped, easy case */ 381 initrd_start = ramdisk_image + PAGE_OFFSET; 382 initrd_end = initrd_start + ramdisk_size; 383 return; 384 } 385 386 relocate_initrd(); 387 388 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 389 } 390 391 #else 392 static void __init early_reserve_initrd(void) 393 { 394 } 395 static void __init reserve_initrd(void) 396 { 397 } 398 #endif /* CONFIG_BLK_DEV_INITRD */ 399 400 static void __init parse_setup_data(void) 401 { 402 struct setup_data *data; 403 u64 pa_data, pa_next; 404 405 pa_data = boot_params.hdr.setup_data; 406 while (pa_data) { 407 u32 data_len, data_type; 408 409 data = early_memremap(pa_data, sizeof(*data)); 410 data_len = data->len + sizeof(struct setup_data); 411 data_type = data->type; 412 pa_next = data->next; 413 early_memunmap(data, sizeof(*data)); 414 415 switch (data_type) { 416 case SETUP_E820_EXT: 417 e820__memory_setup_extended(pa_data, data_len); 418 break; 419 case SETUP_DTB: 420 add_dtb(pa_data); 421 break; 422 case SETUP_EFI: 423 parse_efi_setup(pa_data, data_len); 424 break; 425 default: 426 break; 427 } 428 pa_data = pa_next; 429 } 430 } 431 432 static void __init memblock_x86_reserve_range_setup_data(void) 433 { 434 struct setup_data *data; 435 u64 pa_data; 436 437 pa_data = boot_params.hdr.setup_data; 438 while (pa_data) { 439 data = early_memremap(pa_data, sizeof(*data)); 440 memblock_reserve(pa_data, sizeof(*data) + data->len); 441 pa_data = data->next; 442 early_memunmap(data, sizeof(*data)); 443 } 444 } 445 446 /* 447 * --------- Crashkernel reservation ------------------------------ 448 */ 449 450 #ifdef CONFIG_KEXEC_CORE 451 452 /* 16M alignment for crash kernel regions */ 453 #define CRASH_ALIGN (16 << 20) 454 455 /* 456 * Keep the crash kernel below this limit. On 32 bits earlier kernels 457 * would limit the kernel to the low 512 MiB due to mapping restrictions. 458 * On 64bit, old kexec-tools need to under 896MiB. 459 */ 460 #ifdef CONFIG_X86_32 461 # define CRASH_ADDR_LOW_MAX (512 << 20) 462 # define CRASH_ADDR_HIGH_MAX (512 << 20) 463 #else 464 # define CRASH_ADDR_LOW_MAX (896UL << 20) 465 # define CRASH_ADDR_HIGH_MAX MAXMEM 466 #endif 467 468 static int __init reserve_crashkernel_low(void) 469 { 470 #ifdef CONFIG_X86_64 471 unsigned long long base, low_base = 0, low_size = 0; 472 unsigned long total_low_mem; 473 int ret; 474 475 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 476 477 /* crashkernel=Y,low */ 478 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 479 if (ret) { 480 /* 481 * two parts from lib/swiotlb.c: 482 * -swiotlb size: user-specified with swiotlb= or default. 483 * 484 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 485 * to 8M for other buffers that may need to stay low too. Also 486 * make sure we allocate enough extra low memory so that we 487 * don't run out of DMA buffers for 32-bit devices. 488 */ 489 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 490 } else { 491 /* passed with crashkernel=0,low ? */ 492 if (!low_size) 493 return 0; 494 } 495 496 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 497 if (!low_base) { 498 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 499 (unsigned long)(low_size >> 20)); 500 return -ENOMEM; 501 } 502 503 ret = memblock_reserve(low_base, low_size); 504 if (ret) { 505 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 506 return ret; 507 } 508 509 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 510 (unsigned long)(low_size >> 20), 511 (unsigned long)(low_base >> 20), 512 (unsigned long)(total_low_mem >> 20)); 513 514 crashk_low_res.start = low_base; 515 crashk_low_res.end = low_base + low_size - 1; 516 insert_resource(&iomem_resource, &crashk_low_res); 517 #endif 518 return 0; 519 } 520 521 static void __init reserve_crashkernel(void) 522 { 523 unsigned long long crash_size, crash_base, total_mem; 524 bool high = false; 525 int ret; 526 527 total_mem = memblock_phys_mem_size(); 528 529 /* crashkernel=XM */ 530 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 531 if (ret != 0 || crash_size <= 0) { 532 /* crashkernel=X,high */ 533 ret = parse_crashkernel_high(boot_command_line, total_mem, 534 &crash_size, &crash_base); 535 if (ret != 0 || crash_size <= 0) 536 return; 537 high = true; 538 } 539 540 /* 0 means: find the address automatically */ 541 if (crash_base <= 0) { 542 /* 543 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 544 * as old kexec-tools loads bzImage below that, unless 545 * "crashkernel=size[KMG],high" is specified. 546 */ 547 crash_base = memblock_find_in_range(CRASH_ALIGN, 548 high ? CRASH_ADDR_HIGH_MAX 549 : CRASH_ADDR_LOW_MAX, 550 crash_size, CRASH_ALIGN); 551 if (!crash_base) { 552 pr_info("crashkernel reservation failed - No suitable area found.\n"); 553 return; 554 } 555 556 } else { 557 unsigned long long start; 558 559 start = memblock_find_in_range(crash_base, 560 crash_base + crash_size, 561 crash_size, 1 << 20); 562 if (start != crash_base) { 563 pr_info("crashkernel reservation failed - memory is in use.\n"); 564 return; 565 } 566 } 567 ret = memblock_reserve(crash_base, crash_size); 568 if (ret) { 569 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 570 return; 571 } 572 573 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 574 memblock_free(crash_base, crash_size); 575 return; 576 } 577 578 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 579 (unsigned long)(crash_size >> 20), 580 (unsigned long)(crash_base >> 20), 581 (unsigned long)(total_mem >> 20)); 582 583 crashk_res.start = crash_base; 584 crashk_res.end = crash_base + crash_size - 1; 585 insert_resource(&iomem_resource, &crashk_res); 586 } 587 #else 588 static void __init reserve_crashkernel(void) 589 { 590 } 591 #endif 592 593 static struct resource standard_io_resources[] = { 594 { .name = "dma1", .start = 0x00, .end = 0x1f, 595 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 596 { .name = "pic1", .start = 0x20, .end = 0x21, 597 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 598 { .name = "timer0", .start = 0x40, .end = 0x43, 599 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 600 { .name = "timer1", .start = 0x50, .end = 0x53, 601 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 602 { .name = "keyboard", .start = 0x60, .end = 0x60, 603 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 604 { .name = "keyboard", .start = 0x64, .end = 0x64, 605 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 606 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 607 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 608 { .name = "pic2", .start = 0xa0, .end = 0xa1, 609 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 610 { .name = "dma2", .start = 0xc0, .end = 0xdf, 611 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 612 { .name = "fpu", .start = 0xf0, .end = 0xff, 613 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 614 }; 615 616 void __init reserve_standard_io_resources(void) 617 { 618 int i; 619 620 /* request I/O space for devices used on all i[345]86 PCs */ 621 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 622 request_resource(&ioport_resource, &standard_io_resources[i]); 623 624 } 625 626 static __init void reserve_ibft_region(void) 627 { 628 unsigned long addr, size = 0; 629 630 addr = find_ibft_region(&size); 631 632 if (size) 633 memblock_reserve(addr, size); 634 } 635 636 static bool __init snb_gfx_workaround_needed(void) 637 { 638 #ifdef CONFIG_PCI 639 int i; 640 u16 vendor, devid; 641 static const __initconst u16 snb_ids[] = { 642 0x0102, 643 0x0112, 644 0x0122, 645 0x0106, 646 0x0116, 647 0x0126, 648 0x010a, 649 }; 650 651 /* Assume no if something weird is going on with PCI */ 652 if (!early_pci_allowed()) 653 return false; 654 655 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 656 if (vendor != 0x8086) 657 return false; 658 659 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 660 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 661 if (devid == snb_ids[i]) 662 return true; 663 #endif 664 665 return false; 666 } 667 668 /* 669 * Sandy Bridge graphics has trouble with certain ranges, exclude 670 * them from allocation. 671 */ 672 static void __init trim_snb_memory(void) 673 { 674 static const __initconst unsigned long bad_pages[] = { 675 0x20050000, 676 0x20110000, 677 0x20130000, 678 0x20138000, 679 0x40004000, 680 }; 681 int i; 682 683 if (!snb_gfx_workaround_needed()) 684 return; 685 686 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 687 688 /* 689 * Reserve all memory below the 1 MB mark that has not 690 * already been reserved. 691 */ 692 memblock_reserve(0, 1<<20); 693 694 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 695 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 696 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 697 bad_pages[i]); 698 } 699 } 700 701 /* 702 * Here we put platform-specific memory range workarounds, i.e. 703 * memory known to be corrupt or otherwise in need to be reserved on 704 * specific platforms. 705 * 706 * If this gets used more widely it could use a real dispatch mechanism. 707 */ 708 static void __init trim_platform_memory_ranges(void) 709 { 710 trim_snb_memory(); 711 } 712 713 static void __init trim_bios_range(void) 714 { 715 /* 716 * A special case is the first 4Kb of memory; 717 * This is a BIOS owned area, not kernel ram, but generally 718 * not listed as such in the E820 table. 719 * 720 * This typically reserves additional memory (64KiB by default) 721 * since some BIOSes are known to corrupt low memory. See the 722 * Kconfig help text for X86_RESERVE_LOW. 723 */ 724 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 725 726 /* 727 * special case: Some BIOSen report the PC BIOS 728 * area (640->1Mb) as ram even though it is not. 729 * take them out. 730 */ 731 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 732 733 e820__update_table(e820_table); 734 } 735 736 /* called before trim_bios_range() to spare extra sanitize */ 737 static void __init e820_add_kernel_range(void) 738 { 739 u64 start = __pa_symbol(_text); 740 u64 size = __pa_symbol(_end) - start; 741 742 /* 743 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 744 * attempt to fix it by adding the range. We may have a confused BIOS, 745 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 746 * exclude kernel range. If we really are running on top non-RAM, 747 * we will crash later anyways. 748 */ 749 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 750 return; 751 752 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 753 e820__range_remove(start, size, E820_TYPE_RAM, 0); 754 e820__range_add(start, size, E820_TYPE_RAM); 755 } 756 757 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 758 759 static int __init parse_reservelow(char *p) 760 { 761 unsigned long long size; 762 763 if (!p) 764 return -EINVAL; 765 766 size = memparse(p, &p); 767 768 if (size < 4096) 769 size = 4096; 770 771 if (size > 640*1024) 772 size = 640*1024; 773 774 reserve_low = size; 775 776 return 0; 777 } 778 779 early_param("reservelow", parse_reservelow); 780 781 static void __init trim_low_memory_range(void) 782 { 783 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 784 } 785 786 /* 787 * Dump out kernel offset information on panic. 788 */ 789 static int 790 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 791 { 792 if (kaslr_enabled()) { 793 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 794 kaslr_offset(), 795 __START_KERNEL, 796 __START_KERNEL_map, 797 MODULES_VADDR-1); 798 } else { 799 pr_emerg("Kernel Offset: disabled\n"); 800 } 801 802 return 0; 803 } 804 805 /* 806 * Determine if we were loaded by an EFI loader. If so, then we have also been 807 * passed the efi memmap, systab, etc., so we should use these data structures 808 * for initialization. Note, the efi init code path is determined by the 809 * global efi_enabled. This allows the same kernel image to be used on existing 810 * systems (with a traditional BIOS) as well as on EFI systems. 811 */ 812 /* 813 * setup_arch - architecture-specific boot-time initializations 814 * 815 * Note: On x86_64, fixmaps are ready for use even before this is called. 816 */ 817 818 void __init setup_arch(char **cmdline_p) 819 { 820 memblock_reserve(__pa_symbol(_text), 821 (unsigned long)__bss_stop - (unsigned long)_text); 822 823 early_reserve_initrd(); 824 825 /* 826 * At this point everything still needed from the boot loader 827 * or BIOS or kernel text should be early reserved or marked not 828 * RAM in e820. All other memory is free game. 829 */ 830 831 #ifdef CONFIG_X86_32 832 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 833 834 /* 835 * copy kernel address range established so far and switch 836 * to the proper swapper page table 837 */ 838 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 839 initial_page_table + KERNEL_PGD_BOUNDARY, 840 KERNEL_PGD_PTRS); 841 842 load_cr3(swapper_pg_dir); 843 /* 844 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 845 * a cr3 based tlb flush, so the following __flush_tlb_all() 846 * will not flush anything because the cpu quirk which clears 847 * X86_FEATURE_PGE has not been invoked yet. Though due to the 848 * load_cr3() above the TLB has been flushed already. The 849 * quirk is invoked before subsequent calls to __flush_tlb_all() 850 * so proper operation is guaranteed. 851 */ 852 __flush_tlb_all(); 853 #else 854 printk(KERN_INFO "Command line: %s\n", boot_command_line); 855 #endif 856 857 /* 858 * If we have OLPC OFW, we might end up relocating the fixmap due to 859 * reserve_top(), so do this before touching the ioremap area. 860 */ 861 olpc_ofw_detect(); 862 863 idt_setup_early_traps(); 864 early_cpu_init(); 865 early_ioremap_init(); 866 867 setup_olpc_ofw_pgd(); 868 869 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 870 screen_info = boot_params.screen_info; 871 edid_info = boot_params.edid_info; 872 #ifdef CONFIG_X86_32 873 apm_info.bios = boot_params.apm_bios_info; 874 ist_info = boot_params.ist_info; 875 #endif 876 saved_video_mode = boot_params.hdr.vid_mode; 877 bootloader_type = boot_params.hdr.type_of_loader; 878 if ((bootloader_type >> 4) == 0xe) { 879 bootloader_type &= 0xf; 880 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 881 } 882 bootloader_version = bootloader_type & 0xf; 883 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 884 885 #ifdef CONFIG_BLK_DEV_RAM 886 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 887 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 888 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 889 #endif 890 #ifdef CONFIG_EFI 891 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 892 EFI32_LOADER_SIGNATURE, 4)) { 893 set_bit(EFI_BOOT, &efi.flags); 894 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 895 EFI64_LOADER_SIGNATURE, 4)) { 896 set_bit(EFI_BOOT, &efi.flags); 897 set_bit(EFI_64BIT, &efi.flags); 898 } 899 #endif 900 901 x86_init.oem.arch_setup(); 902 903 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 904 e820__memory_setup(); 905 parse_setup_data(); 906 907 copy_edd(); 908 909 if (!boot_params.hdr.root_flags) 910 root_mountflags &= ~MS_RDONLY; 911 init_mm.start_code = (unsigned long) _text; 912 init_mm.end_code = (unsigned long) _etext; 913 init_mm.end_data = (unsigned long) _edata; 914 init_mm.brk = _brk_end; 915 916 mpx_mm_init(&init_mm); 917 918 code_resource.start = __pa_symbol(_text); 919 code_resource.end = __pa_symbol(_etext)-1; 920 data_resource.start = __pa_symbol(_etext); 921 data_resource.end = __pa_symbol(_edata)-1; 922 bss_resource.start = __pa_symbol(__bss_start); 923 bss_resource.end = __pa_symbol(__bss_stop)-1; 924 925 #ifdef CONFIG_CMDLINE_BOOL 926 #ifdef CONFIG_CMDLINE_OVERRIDE 927 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 928 #else 929 if (builtin_cmdline[0]) { 930 /* append boot loader cmdline to builtin */ 931 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 932 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 933 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 934 } 935 #endif 936 #endif 937 938 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 939 *cmdline_p = command_line; 940 941 /* 942 * x86_configure_nx() is called before parse_early_param() to detect 943 * whether hardware doesn't support NX (so that the early EHCI debug 944 * console setup can safely call set_fixmap()). It may then be called 945 * again from within noexec_setup() during parsing early parameters 946 * to honor the respective command line option. 947 */ 948 x86_configure_nx(); 949 950 parse_early_param(); 951 952 if (efi_enabled(EFI_BOOT)) 953 efi_memblock_x86_reserve_range(); 954 #ifdef CONFIG_MEMORY_HOTPLUG 955 /* 956 * Memory used by the kernel cannot be hot-removed because Linux 957 * cannot migrate the kernel pages. When memory hotplug is 958 * enabled, we should prevent memblock from allocating memory 959 * for the kernel. 960 * 961 * ACPI SRAT records all hotpluggable memory ranges. But before 962 * SRAT is parsed, we don't know about it. 963 * 964 * The kernel image is loaded into memory at very early time. We 965 * cannot prevent this anyway. So on NUMA system, we set any 966 * node the kernel resides in as un-hotpluggable. 967 * 968 * Since on modern servers, one node could have double-digit 969 * gigabytes memory, we can assume the memory around the kernel 970 * image is also un-hotpluggable. So before SRAT is parsed, just 971 * allocate memory near the kernel image to try the best to keep 972 * the kernel away from hotpluggable memory. 973 */ 974 if (movable_node_is_enabled()) 975 memblock_set_bottom_up(true); 976 #endif 977 978 x86_report_nx(); 979 980 /* after early param, so could get panic from serial */ 981 memblock_x86_reserve_range_setup_data(); 982 983 if (acpi_mps_check()) { 984 #ifdef CONFIG_X86_LOCAL_APIC 985 disable_apic = 1; 986 #endif 987 setup_clear_cpu_cap(X86_FEATURE_APIC); 988 } 989 990 #ifdef CONFIG_PCI 991 if (pci_early_dump_regs) 992 early_dump_pci_devices(); 993 #endif 994 995 e820__reserve_setup_data(); 996 e820__finish_early_params(); 997 998 if (efi_enabled(EFI_BOOT)) 999 efi_init(); 1000 1001 dmi_scan_machine(); 1002 dmi_memdev_walk(); 1003 dmi_set_dump_stack_arch_desc(); 1004 1005 /* 1006 * VMware detection requires dmi to be available, so this 1007 * needs to be done after dmi_scan_machine(), for the boot CPU. 1008 */ 1009 init_hypervisor_platform(); 1010 1011 x86_init.resources.probe_roms(); 1012 1013 /* after parse_early_param, so could debug it */ 1014 insert_resource(&iomem_resource, &code_resource); 1015 insert_resource(&iomem_resource, &data_resource); 1016 insert_resource(&iomem_resource, &bss_resource); 1017 1018 e820_add_kernel_range(); 1019 trim_bios_range(); 1020 #ifdef CONFIG_X86_32 1021 if (ppro_with_ram_bug()) { 1022 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1023 E820_TYPE_RESERVED); 1024 e820__update_table(e820_table); 1025 printk(KERN_INFO "fixed physical RAM map:\n"); 1026 e820__print_table("bad_ppro"); 1027 } 1028 #else 1029 early_gart_iommu_check(); 1030 #endif 1031 1032 /* 1033 * partially used pages are not usable - thus 1034 * we are rounding upwards: 1035 */ 1036 max_pfn = e820__end_of_ram_pfn(); 1037 1038 /* update e820 for memory not covered by WB MTRRs */ 1039 mtrr_bp_init(); 1040 if (mtrr_trim_uncached_memory(max_pfn)) 1041 max_pfn = e820__end_of_ram_pfn(); 1042 1043 max_possible_pfn = max_pfn; 1044 1045 /* 1046 * This call is required when the CPU does not support PAT. If 1047 * mtrr_bp_init() invoked it already via pat_init() the call has no 1048 * effect. 1049 */ 1050 init_cache_modes(); 1051 1052 /* 1053 * Define random base addresses for memory sections after max_pfn is 1054 * defined and before each memory section base is used. 1055 */ 1056 kernel_randomize_memory(); 1057 1058 #ifdef CONFIG_X86_32 1059 /* max_low_pfn get updated here */ 1060 find_low_pfn_range(); 1061 #else 1062 check_x2apic(); 1063 1064 /* How many end-of-memory variables you have, grandma! */ 1065 /* need this before calling reserve_initrd */ 1066 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1067 max_low_pfn = e820__end_of_low_ram_pfn(); 1068 else 1069 max_low_pfn = max_pfn; 1070 1071 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1072 #endif 1073 1074 /* 1075 * Find and reserve possible boot-time SMP configuration: 1076 */ 1077 find_smp_config(); 1078 1079 reserve_ibft_region(); 1080 1081 early_alloc_pgt_buf(); 1082 1083 /* 1084 * Need to conclude brk, before e820__memblock_setup() 1085 * it could use memblock_find_in_range, could overlap with 1086 * brk area. 1087 */ 1088 reserve_brk(); 1089 1090 cleanup_highmap(); 1091 1092 memblock_set_current_limit(ISA_END_ADDRESS); 1093 e820__memblock_setup(); 1094 1095 reserve_bios_regions(); 1096 1097 if (efi_enabled(EFI_MEMMAP)) { 1098 efi_fake_memmap(); 1099 efi_find_mirror(); 1100 efi_esrt_init(); 1101 1102 /* 1103 * The EFI specification says that boot service code won't be 1104 * called after ExitBootServices(). This is, in fact, a lie. 1105 */ 1106 efi_reserve_boot_services(); 1107 } 1108 1109 /* preallocate 4k for mptable mpc */ 1110 e820__memblock_alloc_reserved_mpc_new(); 1111 1112 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1113 setup_bios_corruption_check(); 1114 #endif 1115 1116 #ifdef CONFIG_X86_32 1117 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1118 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1119 #endif 1120 1121 reserve_real_mode(); 1122 1123 trim_platform_memory_ranges(); 1124 trim_low_memory_range(); 1125 1126 init_mem_mapping(); 1127 1128 idt_setup_early_pf(); 1129 1130 /* 1131 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1132 * with the current CR4 value. This may not be necessary, but 1133 * auditing all the early-boot CR4 manipulation would be needed to 1134 * rule it out. 1135 * 1136 * Mask off features that don't work outside long mode (just 1137 * PCIDE for now). 1138 */ 1139 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1140 1141 memblock_set_current_limit(get_max_mapped()); 1142 1143 /* 1144 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1145 */ 1146 1147 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1148 if (init_ohci1394_dma_early) 1149 init_ohci1394_dma_on_all_controllers(); 1150 #endif 1151 /* Allocate bigger log buffer */ 1152 setup_log_buf(1); 1153 1154 if (efi_enabled(EFI_BOOT)) { 1155 switch (boot_params.secure_boot) { 1156 case efi_secureboot_mode_disabled: 1157 pr_info("Secure boot disabled\n"); 1158 break; 1159 case efi_secureboot_mode_enabled: 1160 pr_info("Secure boot enabled\n"); 1161 break; 1162 default: 1163 pr_info("Secure boot could not be determined\n"); 1164 break; 1165 } 1166 } 1167 1168 reserve_initrd(); 1169 1170 acpi_table_upgrade(); 1171 1172 vsmp_init(); 1173 1174 io_delay_init(); 1175 1176 early_platform_quirks(); 1177 1178 /* 1179 * Parse the ACPI tables for possible boot-time SMP configuration. 1180 */ 1181 acpi_boot_table_init(); 1182 1183 early_acpi_boot_init(); 1184 1185 initmem_init(); 1186 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1187 1188 /* 1189 * Reserve memory for crash kernel after SRAT is parsed so that it 1190 * won't consume hotpluggable memory. 1191 */ 1192 reserve_crashkernel(); 1193 1194 memblock_find_dma_reserve(); 1195 1196 #ifdef CONFIG_KVM_GUEST 1197 kvmclock_init(); 1198 #endif 1199 1200 tsc_early_delay_calibrate(); 1201 if (!early_xdbc_setup_hardware()) 1202 early_xdbc_register_console(); 1203 1204 x86_init.paging.pagetable_init(); 1205 1206 kasan_init(); 1207 1208 #ifdef CONFIG_X86_32 1209 /* sync back kernel address range */ 1210 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1211 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1212 KERNEL_PGD_PTRS); 1213 1214 /* 1215 * sync back low identity map too. It is used for example 1216 * in the 32-bit EFI stub. 1217 */ 1218 clone_pgd_range(initial_page_table, 1219 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1220 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 1221 #endif 1222 1223 tboot_probe(); 1224 1225 map_vsyscall(); 1226 1227 generic_apic_probe(); 1228 1229 early_quirks(); 1230 1231 /* 1232 * Read APIC and some other early information from ACPI tables. 1233 */ 1234 acpi_boot_init(); 1235 sfi_init(); 1236 x86_dtb_init(); 1237 1238 /* 1239 * get boot-time SMP configuration: 1240 */ 1241 get_smp_config(); 1242 1243 /* 1244 * Systems w/o ACPI and mptables might not have it mapped the local 1245 * APIC yet, but prefill_possible_map() might need to access it. 1246 */ 1247 init_apic_mappings(); 1248 1249 prefill_possible_map(); 1250 1251 init_cpu_to_node(); 1252 1253 io_apic_init_mappings(); 1254 1255 x86_init.hyper.guest_late_init(); 1256 1257 e820__reserve_resources(); 1258 e820__register_nosave_regions(max_low_pfn); 1259 1260 x86_init.resources.reserve_resources(); 1261 1262 e820__setup_pci_gap(); 1263 1264 #ifdef CONFIG_VT 1265 #if defined(CONFIG_VGA_CONSOLE) 1266 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1267 conswitchp = &vga_con; 1268 #elif defined(CONFIG_DUMMY_CONSOLE) 1269 conswitchp = &dummy_con; 1270 #endif 1271 #endif 1272 x86_init.oem.banner(); 1273 1274 x86_init.timers.wallclock_init(); 1275 1276 mcheck_init(); 1277 1278 arch_init_ideal_nops(); 1279 1280 register_refined_jiffies(CLOCK_TICK_RATE); 1281 1282 #ifdef CONFIG_EFI 1283 if (efi_enabled(EFI_BOOT)) 1284 efi_apply_memmap_quirks(); 1285 #endif 1286 1287 unwind_init(); 1288 } 1289 1290 #ifdef CONFIG_X86_32 1291 1292 static struct resource video_ram_resource = { 1293 .name = "Video RAM area", 1294 .start = 0xa0000, 1295 .end = 0xbffff, 1296 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1297 }; 1298 1299 void __init i386_reserve_resources(void) 1300 { 1301 request_resource(&iomem_resource, &video_ram_resource); 1302 reserve_standard_io_resources(); 1303 } 1304 1305 #endif /* CONFIG_X86_32 */ 1306 1307 static struct notifier_block kernel_offset_notifier = { 1308 .notifier_call = dump_kernel_offset 1309 }; 1310 1311 static int __init register_kernel_offset_dumper(void) 1312 { 1313 atomic_notifier_chain_register(&panic_notifier_list, 1314 &kernel_offset_notifier); 1315 return 0; 1316 } 1317 __initcall(register_kernel_offset_dumper); 1318 1319 void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 1320 { 1321 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 1322 return; 1323 1324 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1325 } 1326