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 #include <xen/xen.h> 54 55 #include <linux/errno.h> 56 #include <linux/kernel.h> 57 #include <linux/stddef.h> 58 #include <linux/unistd.h> 59 #include <linux/ptrace.h> 60 #include <linux/user.h> 61 #include <linux/delay.h> 62 63 #include <linux/kallsyms.h> 64 #include <linux/cpufreq.h> 65 #include <linux/dma-mapping.h> 66 #include <linux/ctype.h> 67 #include <linux/uaccess.h> 68 69 #include <linux/percpu.h> 70 #include <linux/crash_dump.h> 71 #include <linux/tboot.h> 72 #include <linux/jiffies.h> 73 #include <linux/mem_encrypt.h> 74 75 #include <linux/usb/xhci-dbgp.h> 76 #include <video/edid.h> 77 78 #include <asm/mtrr.h> 79 #include <asm/apic.h> 80 #include <asm/realmode.h> 81 #include <asm/e820/api.h> 82 #include <asm/mpspec.h> 83 #include <asm/setup.h> 84 #include <asm/efi.h> 85 #include <asm/timer.h> 86 #include <asm/i8259.h> 87 #include <asm/sections.h> 88 #include <asm/io_apic.h> 89 #include <asm/ist.h> 90 #include <asm/setup_arch.h> 91 #include <asm/bios_ebda.h> 92 #include <asm/cacheflush.h> 93 #include <asm/processor.h> 94 #include <asm/bugs.h> 95 #include <asm/kasan.h> 96 97 #include <asm/vsyscall.h> 98 #include <asm/cpu.h> 99 #include <asm/desc.h> 100 #include <asm/dma.h> 101 #include <asm/iommu.h> 102 #include <asm/gart.h> 103 #include <asm/mmu_context.h> 104 #include <asm/proto.h> 105 106 #include <asm/paravirt.h> 107 #include <asm/hypervisor.h> 108 #include <asm/olpc_ofw.h> 109 110 #include <asm/percpu.h> 111 #include <asm/topology.h> 112 #include <asm/apicdef.h> 113 #include <asm/amd_nb.h> 114 #include <asm/mce.h> 115 #include <asm/alternative.h> 116 #include <asm/prom.h> 117 #include <asm/microcode.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 EXPORT_SYMBOL(boot_cpu_data); 195 #endif 196 197 198 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 199 __visible unsigned long mmu_cr4_features __ro_after_init; 200 #else 201 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 202 #endif 203 204 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 205 int bootloader_type, bootloader_version; 206 207 /* 208 * Setup options 209 */ 210 struct screen_info screen_info; 211 EXPORT_SYMBOL(screen_info); 212 struct edid_info edid_info; 213 EXPORT_SYMBOL_GPL(edid_info); 214 215 extern int root_mountflags; 216 217 unsigned long saved_video_mode; 218 219 #define RAMDISK_IMAGE_START_MASK 0x07FF 220 #define RAMDISK_PROMPT_FLAG 0x8000 221 #define RAMDISK_LOAD_FLAG 0x4000 222 223 static char __initdata command_line[COMMAND_LINE_SIZE]; 224 #ifdef CONFIG_CMDLINE_BOOL 225 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 226 #endif 227 228 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 229 struct edd edd; 230 #ifdef CONFIG_EDD_MODULE 231 EXPORT_SYMBOL(edd); 232 #endif 233 /** 234 * copy_edd() - Copy the BIOS EDD information 235 * from boot_params into a safe place. 236 * 237 */ 238 static inline void __init copy_edd(void) 239 { 240 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 241 sizeof(edd.mbr_signature)); 242 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 243 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 244 edd.edd_info_nr = boot_params.eddbuf_entries; 245 } 246 #else 247 static inline void __init copy_edd(void) 248 { 249 } 250 #endif 251 252 void * __init extend_brk(size_t size, size_t align) 253 { 254 size_t mask = align - 1; 255 void *ret; 256 257 BUG_ON(_brk_start == 0); 258 BUG_ON(align & mask); 259 260 _brk_end = (_brk_end + mask) & ~mask; 261 BUG_ON((char *)(_brk_end + size) > __brk_limit); 262 263 ret = (void *)_brk_end; 264 _brk_end += size; 265 266 memset(ret, 0, size); 267 268 return ret; 269 } 270 271 #ifdef CONFIG_X86_32 272 static void __init cleanup_highmap(void) 273 { 274 } 275 #endif 276 277 static void __init reserve_brk(void) 278 { 279 if (_brk_end > _brk_start) 280 memblock_reserve(__pa_symbol(_brk_start), 281 _brk_end - _brk_start); 282 283 /* Mark brk area as locked down and no longer taking any 284 new allocations */ 285 _brk_start = 0; 286 } 287 288 u64 relocated_ramdisk; 289 290 #ifdef CONFIG_BLK_DEV_INITRD 291 292 static u64 __init get_ramdisk_image(void) 293 { 294 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 295 296 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 297 298 return ramdisk_image; 299 } 300 static u64 __init get_ramdisk_size(void) 301 { 302 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 303 304 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 305 306 return ramdisk_size; 307 } 308 309 static void __init relocate_initrd(void) 310 { 311 /* Assume only end is not page aligned */ 312 u64 ramdisk_image = get_ramdisk_image(); 313 u64 ramdisk_size = get_ramdisk_size(); 314 u64 area_size = PAGE_ALIGN(ramdisk_size); 315 316 /* We need to move the initrd down into directly mapped mem */ 317 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 318 area_size, PAGE_SIZE); 319 320 if (!relocated_ramdisk) 321 panic("Cannot find place for new RAMDISK of size %lld\n", 322 ramdisk_size); 323 324 /* Note: this includes all the mem currently occupied by 325 the initrd, we rely on that fact to keep the data intact. */ 326 memblock_reserve(relocated_ramdisk, area_size); 327 initrd_start = relocated_ramdisk + PAGE_OFFSET; 328 initrd_end = initrd_start + ramdisk_size; 329 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 330 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 331 332 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 333 334 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 335 " [mem %#010llx-%#010llx]\n", 336 ramdisk_image, ramdisk_image + ramdisk_size - 1, 337 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 338 } 339 340 static void __init early_reserve_initrd(void) 341 { 342 /* Assume only end is not page aligned */ 343 u64 ramdisk_image = get_ramdisk_image(); 344 u64 ramdisk_size = get_ramdisk_size(); 345 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 346 347 if (!boot_params.hdr.type_of_loader || 348 !ramdisk_image || !ramdisk_size) 349 return; /* No initrd provided by bootloader */ 350 351 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 352 } 353 static void __init reserve_initrd(void) 354 { 355 /* Assume only end is not page aligned */ 356 u64 ramdisk_image = get_ramdisk_image(); 357 u64 ramdisk_size = get_ramdisk_size(); 358 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 359 u64 mapped_size; 360 361 if (!boot_params.hdr.type_of_loader || 362 !ramdisk_image || !ramdisk_size) 363 return; /* No initrd provided by bootloader */ 364 365 initrd_start = 0; 366 367 mapped_size = memblock_mem_size(max_pfn_mapped); 368 if (ramdisk_size >= (mapped_size>>1)) 369 panic("initrd too large to handle, " 370 "disabling initrd (%lld needed, %lld available)\n", 371 ramdisk_size, mapped_size>>1); 372 373 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 374 ramdisk_end - 1); 375 376 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 377 PFN_DOWN(ramdisk_end))) { 378 /* All are mapped, easy case */ 379 initrd_start = ramdisk_image + PAGE_OFFSET; 380 initrd_end = initrd_start + ramdisk_size; 381 return; 382 } 383 384 relocate_initrd(); 385 386 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 387 } 388 389 #else 390 static void __init early_reserve_initrd(void) 391 { 392 } 393 static void __init reserve_initrd(void) 394 { 395 } 396 #endif /* CONFIG_BLK_DEV_INITRD */ 397 398 static void __init parse_setup_data(void) 399 { 400 struct setup_data *data; 401 u64 pa_data, pa_next; 402 403 pa_data = boot_params.hdr.setup_data; 404 while (pa_data) { 405 u32 data_len, data_type; 406 407 data = early_memremap(pa_data, sizeof(*data)); 408 data_len = data->len + sizeof(struct setup_data); 409 data_type = data->type; 410 pa_next = data->next; 411 early_memunmap(data, sizeof(*data)); 412 413 switch (data_type) { 414 case SETUP_E820_EXT: 415 e820__memory_setup_extended(pa_data, data_len); 416 break; 417 case SETUP_DTB: 418 add_dtb(pa_data); 419 break; 420 case SETUP_EFI: 421 parse_efi_setup(pa_data, data_len); 422 break; 423 default: 424 break; 425 } 426 pa_data = pa_next; 427 } 428 } 429 430 static void __init memblock_x86_reserve_range_setup_data(void) 431 { 432 struct setup_data *data; 433 u64 pa_data; 434 435 pa_data = boot_params.hdr.setup_data; 436 while (pa_data) { 437 data = early_memremap(pa_data, sizeof(*data)); 438 memblock_reserve(pa_data, sizeof(*data) + data->len); 439 pa_data = data->next; 440 early_memunmap(data, sizeof(*data)); 441 } 442 } 443 444 /* 445 * --------- Crashkernel reservation ------------------------------ 446 */ 447 448 #ifdef CONFIG_KEXEC_CORE 449 450 /* 16M alignment for crash kernel regions */ 451 #define CRASH_ALIGN (16 << 20) 452 453 /* 454 * Keep the crash kernel below this limit. On 32 bits earlier kernels 455 * would limit the kernel to the low 512 MiB due to mapping restrictions. 456 * On 64bit, old kexec-tools need to under 896MiB. 457 */ 458 #ifdef CONFIG_X86_32 459 # define CRASH_ADDR_LOW_MAX (512 << 20) 460 # define CRASH_ADDR_HIGH_MAX (512 << 20) 461 #else 462 # define CRASH_ADDR_LOW_MAX (896UL << 20) 463 # define CRASH_ADDR_HIGH_MAX MAXMEM 464 #endif 465 466 static int __init reserve_crashkernel_low(void) 467 { 468 #ifdef CONFIG_X86_64 469 unsigned long long base, low_base = 0, low_size = 0; 470 unsigned long total_low_mem; 471 int ret; 472 473 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 474 475 /* crashkernel=Y,low */ 476 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 477 if (ret) { 478 /* 479 * two parts from lib/swiotlb.c: 480 * -swiotlb size: user-specified with swiotlb= or default. 481 * 482 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 483 * to 8M for other buffers that may need to stay low too. Also 484 * make sure we allocate enough extra low memory so that we 485 * don't run out of DMA buffers for 32-bit devices. 486 */ 487 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 488 } else { 489 /* passed with crashkernel=0,low ? */ 490 if (!low_size) 491 return 0; 492 } 493 494 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 495 if (!low_base) { 496 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 497 (unsigned long)(low_size >> 20)); 498 return -ENOMEM; 499 } 500 501 ret = memblock_reserve(low_base, low_size); 502 if (ret) { 503 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 504 return ret; 505 } 506 507 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 508 (unsigned long)(low_size >> 20), 509 (unsigned long)(low_base >> 20), 510 (unsigned long)(total_low_mem >> 20)); 511 512 crashk_low_res.start = low_base; 513 crashk_low_res.end = low_base + low_size - 1; 514 insert_resource(&iomem_resource, &crashk_low_res); 515 #endif 516 return 0; 517 } 518 519 static void __init reserve_crashkernel(void) 520 { 521 unsigned long long crash_size, crash_base, total_mem; 522 bool high = false; 523 int ret; 524 525 total_mem = memblock_phys_mem_size(); 526 527 /* crashkernel=XM */ 528 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 529 if (ret != 0 || crash_size <= 0) { 530 /* crashkernel=X,high */ 531 ret = parse_crashkernel_high(boot_command_line, total_mem, 532 &crash_size, &crash_base); 533 if (ret != 0 || crash_size <= 0) 534 return; 535 high = true; 536 } 537 538 if (xen_pv_domain()) { 539 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 540 return; 541 } 542 543 /* 0 means: find the address automatically */ 544 if (crash_base <= 0) { 545 /* 546 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 547 * as old kexec-tools loads bzImage below that, unless 548 * "crashkernel=size[KMG],high" is specified. 549 */ 550 crash_base = memblock_find_in_range(CRASH_ALIGN, 551 high ? CRASH_ADDR_HIGH_MAX 552 : CRASH_ADDR_LOW_MAX, 553 crash_size, CRASH_ALIGN); 554 if (!crash_base) { 555 pr_info("crashkernel reservation failed - No suitable area found.\n"); 556 return; 557 } 558 559 } else { 560 unsigned long long start; 561 562 start = memblock_find_in_range(crash_base, 563 crash_base + crash_size, 564 crash_size, 1 << 20); 565 if (start != crash_base) { 566 pr_info("crashkernel reservation failed - memory is in use.\n"); 567 return; 568 } 569 } 570 ret = memblock_reserve(crash_base, crash_size); 571 if (ret) { 572 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 573 return; 574 } 575 576 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 577 memblock_free(crash_base, crash_size); 578 return; 579 } 580 581 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 582 (unsigned long)(crash_size >> 20), 583 (unsigned long)(crash_base >> 20), 584 (unsigned long)(total_mem >> 20)); 585 586 crashk_res.start = crash_base; 587 crashk_res.end = crash_base + crash_size - 1; 588 insert_resource(&iomem_resource, &crashk_res); 589 } 590 #else 591 static void __init reserve_crashkernel(void) 592 { 593 } 594 #endif 595 596 static struct resource standard_io_resources[] = { 597 { .name = "dma1", .start = 0x00, .end = 0x1f, 598 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 599 { .name = "pic1", .start = 0x20, .end = 0x21, 600 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 601 { .name = "timer0", .start = 0x40, .end = 0x43, 602 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 603 { .name = "timer1", .start = 0x50, .end = 0x53, 604 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 605 { .name = "keyboard", .start = 0x60, .end = 0x60, 606 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 607 { .name = "keyboard", .start = 0x64, .end = 0x64, 608 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 609 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 610 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 611 { .name = "pic2", .start = 0xa0, .end = 0xa1, 612 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 613 { .name = "dma2", .start = 0xc0, .end = 0xdf, 614 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 615 { .name = "fpu", .start = 0xf0, .end = 0xff, 616 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 617 }; 618 619 void __init reserve_standard_io_resources(void) 620 { 621 int i; 622 623 /* request I/O space for devices used on all i[345]86 PCs */ 624 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 625 request_resource(&ioport_resource, &standard_io_resources[i]); 626 627 } 628 629 static __init void reserve_ibft_region(void) 630 { 631 unsigned long addr, size = 0; 632 633 addr = find_ibft_region(&size); 634 635 if (size) 636 memblock_reserve(addr, size); 637 } 638 639 static bool __init snb_gfx_workaround_needed(void) 640 { 641 #ifdef CONFIG_PCI 642 int i; 643 u16 vendor, devid; 644 static const __initconst u16 snb_ids[] = { 645 0x0102, 646 0x0112, 647 0x0122, 648 0x0106, 649 0x0116, 650 0x0126, 651 0x010a, 652 }; 653 654 /* Assume no if something weird is going on with PCI */ 655 if (!early_pci_allowed()) 656 return false; 657 658 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 659 if (vendor != 0x8086) 660 return false; 661 662 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 663 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 664 if (devid == snb_ids[i]) 665 return true; 666 #endif 667 668 return false; 669 } 670 671 /* 672 * Sandy Bridge graphics has trouble with certain ranges, exclude 673 * them from allocation. 674 */ 675 static void __init trim_snb_memory(void) 676 { 677 static const __initconst unsigned long bad_pages[] = { 678 0x20050000, 679 0x20110000, 680 0x20130000, 681 0x20138000, 682 0x40004000, 683 }; 684 int i; 685 686 if (!snb_gfx_workaround_needed()) 687 return; 688 689 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 690 691 /* 692 * Reserve all memory below the 1 MB mark that has not 693 * already been reserved. 694 */ 695 memblock_reserve(0, 1<<20); 696 697 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 698 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 699 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 700 bad_pages[i]); 701 } 702 } 703 704 /* 705 * Here we put platform-specific memory range workarounds, i.e. 706 * memory known to be corrupt or otherwise in need to be reserved on 707 * specific platforms. 708 * 709 * If this gets used more widely it could use a real dispatch mechanism. 710 */ 711 static void __init trim_platform_memory_ranges(void) 712 { 713 trim_snb_memory(); 714 } 715 716 static void __init trim_bios_range(void) 717 { 718 /* 719 * A special case is the first 4Kb of memory; 720 * This is a BIOS owned area, not kernel ram, but generally 721 * not listed as such in the E820 table. 722 * 723 * This typically reserves additional memory (64KiB by default) 724 * since some BIOSes are known to corrupt low memory. See the 725 * Kconfig help text for X86_RESERVE_LOW. 726 */ 727 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 728 729 /* 730 * special case: Some BIOSen report the PC BIOS 731 * area (640->1Mb) as ram even though it is not. 732 * take them out. 733 */ 734 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 735 736 e820__update_table(e820_table); 737 } 738 739 /* called before trim_bios_range() to spare extra sanitize */ 740 static void __init e820_add_kernel_range(void) 741 { 742 u64 start = __pa_symbol(_text); 743 u64 size = __pa_symbol(_end) - start; 744 745 /* 746 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 747 * attempt to fix it by adding the range. We may have a confused BIOS, 748 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 749 * exclude kernel range. If we really are running on top non-RAM, 750 * we will crash later anyways. 751 */ 752 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 753 return; 754 755 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 756 e820__range_remove(start, size, E820_TYPE_RAM, 0); 757 e820__range_add(start, size, E820_TYPE_RAM); 758 } 759 760 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 761 762 static int __init parse_reservelow(char *p) 763 { 764 unsigned long long size; 765 766 if (!p) 767 return -EINVAL; 768 769 size = memparse(p, &p); 770 771 if (size < 4096) 772 size = 4096; 773 774 if (size > 640*1024) 775 size = 640*1024; 776 777 reserve_low = size; 778 779 return 0; 780 } 781 782 early_param("reservelow", parse_reservelow); 783 784 static void __init trim_low_memory_range(void) 785 { 786 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 787 } 788 789 /* 790 * Dump out kernel offset information on panic. 791 */ 792 static int 793 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 794 { 795 if (kaslr_enabled()) { 796 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 797 kaslr_offset(), 798 __START_KERNEL, 799 __START_KERNEL_map, 800 MODULES_VADDR-1); 801 } else { 802 pr_emerg("Kernel Offset: disabled\n"); 803 } 804 805 return 0; 806 } 807 808 /* 809 * Determine if we were loaded by an EFI loader. If so, then we have also been 810 * passed the efi memmap, systab, etc., so we should use these data structures 811 * for initialization. Note, the efi init code path is determined by the 812 * global efi_enabled. This allows the same kernel image to be used on existing 813 * systems (with a traditional BIOS) as well as on EFI systems. 814 */ 815 /* 816 * setup_arch - architecture-specific boot-time initializations 817 * 818 * Note: On x86_64, fixmaps are ready for use even before this is called. 819 */ 820 821 void __init setup_arch(char **cmdline_p) 822 { 823 memblock_reserve(__pa_symbol(_text), 824 (unsigned long)__bss_stop - (unsigned long)_text); 825 826 early_reserve_initrd(); 827 828 /* 829 * At this point everything still needed from the boot loader 830 * or BIOS or kernel text should be early reserved or marked not 831 * RAM in e820. All other memory is free game. 832 */ 833 834 #ifdef CONFIG_X86_32 835 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 836 837 /* 838 * copy kernel address range established so far and switch 839 * to the proper swapper page table 840 */ 841 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 842 initial_page_table + KERNEL_PGD_BOUNDARY, 843 KERNEL_PGD_PTRS); 844 845 load_cr3(swapper_pg_dir); 846 /* 847 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 848 * a cr3 based tlb flush, so the following __flush_tlb_all() 849 * will not flush anything because the cpu quirk which clears 850 * X86_FEATURE_PGE has not been invoked yet. Though due to the 851 * load_cr3() above the TLB has been flushed already. The 852 * quirk is invoked before subsequent calls to __flush_tlb_all() 853 * so proper operation is guaranteed. 854 */ 855 __flush_tlb_all(); 856 #else 857 printk(KERN_INFO "Command line: %s\n", boot_command_line); 858 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 859 #endif 860 861 /* 862 * If we have OLPC OFW, we might end up relocating the fixmap due to 863 * reserve_top(), so do this before touching the ioremap area. 864 */ 865 olpc_ofw_detect(); 866 867 idt_setup_early_traps(); 868 early_cpu_init(); 869 early_ioremap_init(); 870 871 setup_olpc_ofw_pgd(); 872 873 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 874 screen_info = boot_params.screen_info; 875 edid_info = boot_params.edid_info; 876 #ifdef CONFIG_X86_32 877 apm_info.bios = boot_params.apm_bios_info; 878 ist_info = boot_params.ist_info; 879 #endif 880 saved_video_mode = boot_params.hdr.vid_mode; 881 bootloader_type = boot_params.hdr.type_of_loader; 882 if ((bootloader_type >> 4) == 0xe) { 883 bootloader_type &= 0xf; 884 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 885 } 886 bootloader_version = bootloader_type & 0xf; 887 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 888 889 #ifdef CONFIG_BLK_DEV_RAM 890 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 891 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 892 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 893 #endif 894 #ifdef CONFIG_EFI 895 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 896 EFI32_LOADER_SIGNATURE, 4)) { 897 set_bit(EFI_BOOT, &efi.flags); 898 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 899 EFI64_LOADER_SIGNATURE, 4)) { 900 set_bit(EFI_BOOT, &efi.flags); 901 set_bit(EFI_64BIT, &efi.flags); 902 } 903 #endif 904 905 x86_init.oem.arch_setup(); 906 907 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 908 e820__memory_setup(); 909 parse_setup_data(); 910 911 copy_edd(); 912 913 if (!boot_params.hdr.root_flags) 914 root_mountflags &= ~MS_RDONLY; 915 init_mm.start_code = (unsigned long) _text; 916 init_mm.end_code = (unsigned long) _etext; 917 init_mm.end_data = (unsigned long) _edata; 918 init_mm.brk = _brk_end; 919 920 mpx_mm_init(&init_mm); 921 922 code_resource.start = __pa_symbol(_text); 923 code_resource.end = __pa_symbol(_etext)-1; 924 data_resource.start = __pa_symbol(_etext); 925 data_resource.end = __pa_symbol(_edata)-1; 926 bss_resource.start = __pa_symbol(__bss_start); 927 bss_resource.end = __pa_symbol(__bss_stop)-1; 928 929 #ifdef CONFIG_CMDLINE_BOOL 930 #ifdef CONFIG_CMDLINE_OVERRIDE 931 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 932 #else 933 if (builtin_cmdline[0]) { 934 /* append boot loader cmdline to builtin */ 935 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 936 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 937 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 938 } 939 #endif 940 #endif 941 942 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 943 *cmdline_p = command_line; 944 945 /* 946 * x86_configure_nx() is called before parse_early_param() to detect 947 * whether hardware doesn't support NX (so that the early EHCI debug 948 * console setup can safely call set_fixmap()). It may then be called 949 * again from within noexec_setup() during parsing early parameters 950 * to honor the respective command line option. 951 */ 952 x86_configure_nx(); 953 954 parse_early_param(); 955 956 if (efi_enabled(EFI_BOOT)) 957 efi_memblock_x86_reserve_range(); 958 #ifdef CONFIG_MEMORY_HOTPLUG 959 /* 960 * Memory used by the kernel cannot be hot-removed because Linux 961 * cannot migrate the kernel pages. When memory hotplug is 962 * enabled, we should prevent memblock from allocating memory 963 * for the kernel. 964 * 965 * ACPI SRAT records all hotpluggable memory ranges. But before 966 * SRAT is parsed, we don't know about it. 967 * 968 * The kernel image is loaded into memory at very early time. We 969 * cannot prevent this anyway. So on NUMA system, we set any 970 * node the kernel resides in as un-hotpluggable. 971 * 972 * Since on modern servers, one node could have double-digit 973 * gigabytes memory, we can assume the memory around the kernel 974 * image is also un-hotpluggable. So before SRAT is parsed, just 975 * allocate memory near the kernel image to try the best to keep 976 * the kernel away from hotpluggable memory. 977 */ 978 if (movable_node_is_enabled()) 979 memblock_set_bottom_up(true); 980 #endif 981 982 x86_report_nx(); 983 984 /* after early param, so could get panic from serial */ 985 memblock_x86_reserve_range_setup_data(); 986 987 if (acpi_mps_check()) { 988 #ifdef CONFIG_X86_LOCAL_APIC 989 disable_apic = 1; 990 #endif 991 setup_clear_cpu_cap(X86_FEATURE_APIC); 992 } 993 994 #ifdef CONFIG_PCI 995 if (pci_early_dump_regs) 996 early_dump_pci_devices(); 997 #endif 998 999 e820__reserve_setup_data(); 1000 e820__finish_early_params(); 1001 1002 if (efi_enabled(EFI_BOOT)) 1003 efi_init(); 1004 1005 dmi_scan_machine(); 1006 dmi_memdev_walk(); 1007 dmi_set_dump_stack_arch_desc(); 1008 1009 /* 1010 * VMware detection requires dmi to be available, so this 1011 * needs to be done after dmi_scan_machine(), for the boot CPU. 1012 */ 1013 init_hypervisor_platform(); 1014 1015 x86_init.resources.probe_roms(); 1016 1017 /* after parse_early_param, so could debug it */ 1018 insert_resource(&iomem_resource, &code_resource); 1019 insert_resource(&iomem_resource, &data_resource); 1020 insert_resource(&iomem_resource, &bss_resource); 1021 1022 e820_add_kernel_range(); 1023 trim_bios_range(); 1024 #ifdef CONFIG_X86_32 1025 if (ppro_with_ram_bug()) { 1026 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1027 E820_TYPE_RESERVED); 1028 e820__update_table(e820_table); 1029 printk(KERN_INFO "fixed physical RAM map:\n"); 1030 e820__print_table("bad_ppro"); 1031 } 1032 #else 1033 early_gart_iommu_check(); 1034 #endif 1035 1036 /* 1037 * partially used pages are not usable - thus 1038 * we are rounding upwards: 1039 */ 1040 max_pfn = e820__end_of_ram_pfn(); 1041 1042 /* update e820 for memory not covered by WB MTRRs */ 1043 mtrr_bp_init(); 1044 if (mtrr_trim_uncached_memory(max_pfn)) 1045 max_pfn = e820__end_of_ram_pfn(); 1046 1047 max_possible_pfn = max_pfn; 1048 1049 /* 1050 * This call is required when the CPU does not support PAT. If 1051 * mtrr_bp_init() invoked it already via pat_init() the call has no 1052 * effect. 1053 */ 1054 init_cache_modes(); 1055 1056 /* 1057 * Define random base addresses for memory sections after max_pfn is 1058 * defined and before each memory section base is used. 1059 */ 1060 kernel_randomize_memory(); 1061 1062 #ifdef CONFIG_X86_32 1063 /* max_low_pfn get updated here */ 1064 find_low_pfn_range(); 1065 #else 1066 check_x2apic(); 1067 1068 /* How many end-of-memory variables you have, grandma! */ 1069 /* need this before calling reserve_initrd */ 1070 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1071 max_low_pfn = e820__end_of_low_ram_pfn(); 1072 else 1073 max_low_pfn = max_pfn; 1074 1075 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1076 #endif 1077 1078 /* 1079 * Find and reserve possible boot-time SMP configuration: 1080 */ 1081 find_smp_config(); 1082 1083 reserve_ibft_region(); 1084 1085 early_alloc_pgt_buf(); 1086 1087 /* 1088 * Need to conclude brk, before e820__memblock_setup() 1089 * it could use memblock_find_in_range, could overlap with 1090 * brk area. 1091 */ 1092 reserve_brk(); 1093 1094 cleanup_highmap(); 1095 1096 memblock_set_current_limit(ISA_END_ADDRESS); 1097 e820__memblock_setup(); 1098 1099 reserve_bios_regions(); 1100 1101 if (efi_enabled(EFI_MEMMAP)) { 1102 efi_fake_memmap(); 1103 efi_find_mirror(); 1104 efi_esrt_init(); 1105 1106 /* 1107 * The EFI specification says that boot service code won't be 1108 * called after ExitBootServices(). This is, in fact, a lie. 1109 */ 1110 efi_reserve_boot_services(); 1111 } 1112 1113 /* preallocate 4k for mptable mpc */ 1114 e820__memblock_alloc_reserved_mpc_new(); 1115 1116 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1117 setup_bios_corruption_check(); 1118 #endif 1119 1120 #ifdef CONFIG_X86_32 1121 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1122 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1123 #endif 1124 1125 reserve_real_mode(); 1126 1127 trim_platform_memory_ranges(); 1128 trim_low_memory_range(); 1129 1130 init_mem_mapping(); 1131 1132 idt_setup_early_pf(); 1133 1134 /* 1135 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1136 * with the current CR4 value. This may not be necessary, but 1137 * auditing all the early-boot CR4 manipulation would be needed to 1138 * rule it out. 1139 * 1140 * Mask off features that don't work outside long mode (just 1141 * PCIDE for now). 1142 */ 1143 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1144 1145 memblock_set_current_limit(get_max_mapped()); 1146 1147 /* 1148 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1149 */ 1150 1151 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1152 if (init_ohci1394_dma_early) 1153 init_ohci1394_dma_on_all_controllers(); 1154 #endif 1155 /* Allocate bigger log buffer */ 1156 setup_log_buf(1); 1157 1158 if (efi_enabled(EFI_BOOT)) { 1159 switch (boot_params.secure_boot) { 1160 case efi_secureboot_mode_disabled: 1161 pr_info("Secure boot disabled\n"); 1162 break; 1163 case efi_secureboot_mode_enabled: 1164 pr_info("Secure boot enabled\n"); 1165 break; 1166 default: 1167 pr_info("Secure boot could not be determined\n"); 1168 break; 1169 } 1170 } 1171 1172 reserve_initrd(); 1173 1174 acpi_table_upgrade(); 1175 1176 vsmp_init(); 1177 1178 io_delay_init(); 1179 1180 early_platform_quirks(); 1181 1182 /* 1183 * Parse the ACPI tables for possible boot-time SMP configuration. 1184 */ 1185 acpi_boot_table_init(); 1186 1187 early_acpi_boot_init(); 1188 1189 initmem_init(); 1190 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1191 1192 /* 1193 * Reserve memory for crash kernel after SRAT is parsed so that it 1194 * won't consume hotpluggable memory. 1195 */ 1196 reserve_crashkernel(); 1197 1198 memblock_find_dma_reserve(); 1199 1200 #ifdef CONFIG_KVM_GUEST 1201 kvmclock_init(); 1202 #endif 1203 1204 tsc_early_delay_calibrate(); 1205 if (!early_xdbc_setup_hardware()) 1206 early_xdbc_register_console(); 1207 1208 x86_init.paging.pagetable_init(); 1209 1210 kasan_init(); 1211 1212 /* 1213 * Sync back kernel address range. 1214 * 1215 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1216 * this call? 1217 */ 1218 sync_initial_page_table(); 1219 1220 tboot_probe(); 1221 1222 map_vsyscall(); 1223 1224 generic_apic_probe(); 1225 1226 early_quirks(); 1227 1228 /* 1229 * Read APIC and some other early information from ACPI tables. 1230 */ 1231 acpi_boot_init(); 1232 sfi_init(); 1233 x86_dtb_init(); 1234 1235 /* 1236 * get boot-time SMP configuration: 1237 */ 1238 get_smp_config(); 1239 1240 /* 1241 * Systems w/o ACPI and mptables might not have it mapped the local 1242 * APIC yet, but prefill_possible_map() might need to access it. 1243 */ 1244 init_apic_mappings(); 1245 1246 prefill_possible_map(); 1247 1248 init_cpu_to_node(); 1249 1250 io_apic_init_mappings(); 1251 1252 x86_init.hyper.guest_late_init(); 1253 1254 e820__reserve_resources(); 1255 e820__register_nosave_regions(max_low_pfn); 1256 1257 x86_init.resources.reserve_resources(); 1258 1259 e820__setup_pci_gap(); 1260 1261 #ifdef CONFIG_VT 1262 #if defined(CONFIG_VGA_CONSOLE) 1263 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1264 conswitchp = &vga_con; 1265 #elif defined(CONFIG_DUMMY_CONSOLE) 1266 conswitchp = &dummy_con; 1267 #endif 1268 #endif 1269 x86_init.oem.banner(); 1270 1271 x86_init.timers.wallclock_init(); 1272 1273 mcheck_init(); 1274 1275 arch_init_ideal_nops(); 1276 1277 register_refined_jiffies(CLOCK_TICK_RATE); 1278 1279 #ifdef CONFIG_EFI 1280 if (efi_enabled(EFI_BOOT)) 1281 efi_apply_memmap_quirks(); 1282 #endif 1283 1284 unwind_init(); 1285 } 1286 1287 #ifdef CONFIG_X86_32 1288 1289 static struct resource video_ram_resource = { 1290 .name = "Video RAM area", 1291 .start = 0xa0000, 1292 .end = 0xbffff, 1293 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1294 }; 1295 1296 void __init i386_reserve_resources(void) 1297 { 1298 request_resource(&iomem_resource, &video_ram_resource); 1299 reserve_standard_io_resources(); 1300 } 1301 1302 #endif /* CONFIG_X86_32 */ 1303 1304 static struct notifier_block kernel_offset_notifier = { 1305 .notifier_call = dump_kernel_offset 1306 }; 1307 1308 static int __init register_kernel_offset_dumper(void) 1309 { 1310 atomic_notifier_chain_register(&panic_notifier_list, 1311 &kernel_offset_notifier); 1312 return 0; 1313 } 1314 __initcall(register_kernel_offset_dumper); 1315 1316 void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 1317 { 1318 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 1319 return; 1320 1321 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1322 } 1323