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/memblock.h> 34 #include <linux/seq_file.h> 35 #include <linux/console.h> 36 #include <linux/root_dev.h> 37 #include <linux/highmem.h> 38 #include <linux/export.h> 39 #include <linux/efi.h> 40 #include <linux/init.h> 41 #include <linux/edd.h> 42 #include <linux/iscsi_ibft.h> 43 #include <linux/nodemask.h> 44 #include <linux/kexec.h> 45 #include <linux/dmi.h> 46 #include <linux/pfn.h> 47 #include <linux/pci.h> 48 #include <asm/pci-direct.h> 49 #include <linux/init_ohci1394_dma.h> 50 #include <linux/kvm_para.h> 51 #include <linux/dma-contiguous.h> 52 #include <xen/xen.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/kaslr.h> 118 #include <asm/unwind.h> 119 120 /* 121 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 122 * max_pfn_mapped: highest direct mapped pfn over 4GB 123 * 124 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 125 * represented by pfn_mapped 126 */ 127 unsigned long max_low_pfn_mapped; 128 unsigned long max_pfn_mapped; 129 130 #ifdef CONFIG_DMI 131 RESERVE_BRK(dmi_alloc, 65536); 132 #endif 133 134 135 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 136 unsigned long _brk_end = (unsigned long)__brk_base; 137 138 struct boot_params boot_params; 139 140 /* 141 * Machine setup.. 142 */ 143 static struct resource data_resource = { 144 .name = "Kernel data", 145 .start = 0, 146 .end = 0, 147 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 148 }; 149 150 static struct resource code_resource = { 151 .name = "Kernel code", 152 .start = 0, 153 .end = 0, 154 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 155 }; 156 157 static struct resource bss_resource = { 158 .name = "Kernel bss", 159 .start = 0, 160 .end = 0, 161 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 162 }; 163 164 165 #ifdef CONFIG_X86_32 166 /* cpu data as detected by the assembly code in head_32.S */ 167 struct cpuinfo_x86 new_cpu_data; 168 169 /* common cpu data for all cpus */ 170 struct cpuinfo_x86 boot_cpu_data __read_mostly; 171 EXPORT_SYMBOL(boot_cpu_data); 172 173 unsigned int def_to_bigsmp; 174 175 /* for MCA, but anyone else can use it if they want */ 176 unsigned int machine_id; 177 unsigned int machine_submodel_id; 178 unsigned int BIOS_revision; 179 180 struct apm_info apm_info; 181 EXPORT_SYMBOL(apm_info); 182 183 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 184 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 185 struct ist_info ist_info; 186 EXPORT_SYMBOL(ist_info); 187 #else 188 struct ist_info ist_info; 189 #endif 190 191 #else 192 struct cpuinfo_x86 boot_cpu_data __read_mostly; 193 EXPORT_SYMBOL(boot_cpu_data); 194 #endif 195 196 197 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 198 __visible unsigned long mmu_cr4_features __ro_after_init; 199 #else 200 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 201 #endif 202 203 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 204 int bootloader_type, bootloader_version; 205 206 /* 207 * Setup options 208 */ 209 struct screen_info screen_info; 210 EXPORT_SYMBOL(screen_info); 211 struct edid_info edid_info; 212 EXPORT_SYMBOL_GPL(edid_info); 213 214 extern int root_mountflags; 215 216 unsigned long saved_video_mode; 217 218 #define RAMDISK_IMAGE_START_MASK 0x07FF 219 #define RAMDISK_PROMPT_FLAG 0x8000 220 #define RAMDISK_LOAD_FLAG 0x4000 221 222 static char __initdata command_line[COMMAND_LINE_SIZE]; 223 #ifdef CONFIG_CMDLINE_BOOL 224 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 225 #endif 226 227 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 228 struct edd edd; 229 #ifdef CONFIG_EDD_MODULE 230 EXPORT_SYMBOL(edd); 231 #endif 232 /** 233 * copy_edd() - Copy the BIOS EDD information 234 * from boot_params into a safe place. 235 * 236 */ 237 static inline void __init copy_edd(void) 238 { 239 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 240 sizeof(edd.mbr_signature)); 241 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 242 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 243 edd.edd_info_nr = boot_params.eddbuf_entries; 244 } 245 #else 246 static inline void __init copy_edd(void) 247 { 248 } 249 #endif 250 251 void * __init extend_brk(size_t size, size_t align) 252 { 253 size_t mask = align - 1; 254 void *ret; 255 256 BUG_ON(_brk_start == 0); 257 BUG_ON(align & mask); 258 259 _brk_end = (_brk_end + mask) & ~mask; 260 BUG_ON((char *)(_brk_end + size) > __brk_limit); 261 262 ret = (void *)_brk_end; 263 _brk_end += size; 264 265 memset(ret, 0, size); 266 267 return ret; 268 } 269 270 #ifdef CONFIG_X86_32 271 static void __init cleanup_highmap(void) 272 { 273 } 274 #endif 275 276 static void __init reserve_brk(void) 277 { 278 if (_brk_end > _brk_start) 279 memblock_reserve(__pa_symbol(_brk_start), 280 _brk_end - _brk_start); 281 282 /* Mark brk area as locked down and no longer taking any 283 new allocations */ 284 _brk_start = 0; 285 } 286 287 u64 relocated_ramdisk; 288 289 #ifdef CONFIG_BLK_DEV_INITRD 290 291 static u64 __init get_ramdisk_image(void) 292 { 293 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 294 295 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 296 297 return ramdisk_image; 298 } 299 static u64 __init get_ramdisk_size(void) 300 { 301 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 302 303 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 304 305 return ramdisk_size; 306 } 307 308 static void __init relocate_initrd(void) 309 { 310 /* Assume only end is not page aligned */ 311 u64 ramdisk_image = get_ramdisk_image(); 312 u64 ramdisk_size = get_ramdisk_size(); 313 u64 area_size = PAGE_ALIGN(ramdisk_size); 314 315 /* We need to move the initrd down into directly mapped mem */ 316 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 317 area_size, PAGE_SIZE); 318 319 if (!relocated_ramdisk) 320 panic("Cannot find place for new RAMDISK of size %lld\n", 321 ramdisk_size); 322 323 /* Note: this includes all the mem currently occupied by 324 the initrd, we rely on that fact to keep the data intact. */ 325 memblock_reserve(relocated_ramdisk, area_size); 326 initrd_start = relocated_ramdisk + PAGE_OFFSET; 327 initrd_end = initrd_start + ramdisk_size; 328 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 329 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 330 331 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 332 333 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 334 " [mem %#010llx-%#010llx]\n", 335 ramdisk_image, ramdisk_image + ramdisk_size - 1, 336 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 337 } 338 339 static void __init early_reserve_initrd(void) 340 { 341 /* Assume only end is not page aligned */ 342 u64 ramdisk_image = get_ramdisk_image(); 343 u64 ramdisk_size = get_ramdisk_size(); 344 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 345 346 if (!boot_params.hdr.type_of_loader || 347 !ramdisk_image || !ramdisk_size) 348 return; /* No initrd provided by bootloader */ 349 350 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 351 } 352 static void __init 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 u64 mapped_size; 359 360 if (!boot_params.hdr.type_of_loader || 361 !ramdisk_image || !ramdisk_size) 362 return; /* No initrd provided by bootloader */ 363 364 initrd_start = 0; 365 366 mapped_size = memblock_mem_size(max_pfn_mapped); 367 if (ramdisk_size >= (mapped_size>>1)) 368 panic("initrd too large to handle, " 369 "disabling initrd (%lld needed, %lld available)\n", 370 ramdisk_size, mapped_size>>1); 371 372 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 373 ramdisk_end - 1); 374 375 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 376 PFN_DOWN(ramdisk_end))) { 377 /* All are mapped, easy case */ 378 initrd_start = ramdisk_image + PAGE_OFFSET; 379 initrd_end = initrd_start + ramdisk_size; 380 return; 381 } 382 383 relocate_initrd(); 384 385 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 386 } 387 388 #else 389 static void __init early_reserve_initrd(void) 390 { 391 } 392 static void __init reserve_initrd(void) 393 { 394 } 395 #endif /* CONFIG_BLK_DEV_INITRD */ 396 397 static void __init parse_setup_data(void) 398 { 399 struct setup_data *data; 400 u64 pa_data, pa_next; 401 402 pa_data = boot_params.hdr.setup_data; 403 while (pa_data) { 404 u32 data_len, data_type; 405 406 data = early_memremap(pa_data, sizeof(*data)); 407 data_len = data->len + sizeof(struct setup_data); 408 data_type = data->type; 409 pa_next = data->next; 410 early_memunmap(data, sizeof(*data)); 411 412 switch (data_type) { 413 case SETUP_E820_EXT: 414 e820__memory_setup_extended(pa_data, data_len); 415 break; 416 case SETUP_DTB: 417 add_dtb(pa_data); 418 break; 419 case SETUP_EFI: 420 parse_efi_setup(pa_data, data_len); 421 break; 422 default: 423 break; 424 } 425 pa_data = pa_next; 426 } 427 } 428 429 static void __init memblock_x86_reserve_range_setup_data(void) 430 { 431 struct setup_data *data; 432 u64 pa_data; 433 434 pa_data = boot_params.hdr.setup_data; 435 while (pa_data) { 436 data = early_memremap(pa_data, sizeof(*data)); 437 memblock_reserve(pa_data, sizeof(*data) + data->len); 438 pa_data = data->next; 439 early_memunmap(data, sizeof(*data)); 440 } 441 } 442 443 /* 444 * --------- Crashkernel reservation ------------------------------ 445 */ 446 447 #ifdef CONFIG_KEXEC_CORE 448 449 /* 16M alignment for crash kernel regions */ 450 #define CRASH_ALIGN (16 << 20) 451 452 /* 453 * Keep the crash kernel below this limit. On 32 bits earlier kernels 454 * would limit the kernel to the low 512 MiB due to mapping restrictions. 455 * On 64bit, old kexec-tools need to under 896MiB. 456 */ 457 #ifdef CONFIG_X86_32 458 # define CRASH_ADDR_LOW_MAX (512 << 20) 459 # define CRASH_ADDR_HIGH_MAX (512 << 20) 460 #else 461 # define CRASH_ADDR_LOW_MAX (896UL << 20) 462 # define CRASH_ADDR_HIGH_MAX MAXMEM 463 #endif 464 465 static int __init reserve_crashkernel_low(void) 466 { 467 #ifdef CONFIG_X86_64 468 unsigned long long base, low_base = 0, low_size = 0; 469 unsigned long total_low_mem; 470 int ret; 471 472 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 473 474 /* crashkernel=Y,low */ 475 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 476 if (ret) { 477 /* 478 * two parts from lib/swiotlb.c: 479 * -swiotlb size: user-specified with swiotlb= or default. 480 * 481 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 482 * to 8M for other buffers that may need to stay low too. Also 483 * make sure we allocate enough extra low memory so that we 484 * don't run out of DMA buffers for 32-bit devices. 485 */ 486 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 487 } else { 488 /* passed with crashkernel=0,low ? */ 489 if (!low_size) 490 return 0; 491 } 492 493 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 494 if (!low_base) { 495 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 496 (unsigned long)(low_size >> 20)); 497 return -ENOMEM; 498 } 499 500 ret = memblock_reserve(low_base, low_size); 501 if (ret) { 502 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 503 return ret; 504 } 505 506 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 507 (unsigned long)(low_size >> 20), 508 (unsigned long)(low_base >> 20), 509 (unsigned long)(total_low_mem >> 20)); 510 511 crashk_low_res.start = low_base; 512 crashk_low_res.end = low_base + low_size - 1; 513 insert_resource(&iomem_resource, &crashk_low_res); 514 #endif 515 return 0; 516 } 517 518 static void __init reserve_crashkernel(void) 519 { 520 unsigned long long crash_size, crash_base, total_mem; 521 bool high = false; 522 int ret; 523 524 total_mem = memblock_phys_mem_size(); 525 526 /* crashkernel=XM */ 527 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 528 if (ret != 0 || crash_size <= 0) { 529 /* crashkernel=X,high */ 530 ret = parse_crashkernel_high(boot_command_line, total_mem, 531 &crash_size, &crash_base); 532 if (ret != 0 || crash_size <= 0) 533 return; 534 high = true; 535 } 536 537 if (xen_pv_domain()) { 538 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 539 return; 540 } 541 542 /* 0 means: find the address automatically */ 543 if (crash_base <= 0) { 544 /* 545 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 546 * as old kexec-tools loads bzImage below that, unless 547 * "crashkernel=size[KMG],high" is specified. 548 */ 549 crash_base = memblock_find_in_range(CRASH_ALIGN, 550 high ? CRASH_ADDR_HIGH_MAX 551 : CRASH_ADDR_LOW_MAX, 552 crash_size, CRASH_ALIGN); 553 if (!crash_base) { 554 pr_info("crashkernel reservation failed - No suitable area found.\n"); 555 return; 556 } 557 558 } else { 559 unsigned long long start; 560 561 start = memblock_find_in_range(crash_base, 562 crash_base + crash_size, 563 crash_size, 1 << 20); 564 if (start != crash_base) { 565 pr_info("crashkernel reservation failed - memory is in use.\n"); 566 return; 567 } 568 } 569 ret = memblock_reserve(crash_base, crash_size); 570 if (ret) { 571 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 572 return; 573 } 574 575 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 576 memblock_free(crash_base, crash_size); 577 return; 578 } 579 580 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 581 (unsigned long)(crash_size >> 20), 582 (unsigned long)(crash_base >> 20), 583 (unsigned long)(total_mem >> 20)); 584 585 crashk_res.start = crash_base; 586 crashk_res.end = crash_base + crash_size - 1; 587 insert_resource(&iomem_resource, &crashk_res); 588 } 589 #else 590 static void __init reserve_crashkernel(void) 591 { 592 } 593 #endif 594 595 static struct resource standard_io_resources[] = { 596 { .name = "dma1", .start = 0x00, .end = 0x1f, 597 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 598 { .name = "pic1", .start = 0x20, .end = 0x21, 599 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 600 { .name = "timer0", .start = 0x40, .end = 0x43, 601 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 602 { .name = "timer1", .start = 0x50, .end = 0x53, 603 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 604 { .name = "keyboard", .start = 0x60, .end = 0x60, 605 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 606 { .name = "keyboard", .start = 0x64, .end = 0x64, 607 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 608 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 609 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 610 { .name = "pic2", .start = 0xa0, .end = 0xa1, 611 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 612 { .name = "dma2", .start = 0xc0, .end = 0xdf, 613 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 614 { .name = "fpu", .start = 0xf0, .end = 0xff, 615 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 616 }; 617 618 void __init reserve_standard_io_resources(void) 619 { 620 int i; 621 622 /* request I/O space for devices used on all i[345]86 PCs */ 623 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 624 request_resource(&ioport_resource, &standard_io_resources[i]); 625 626 } 627 628 static __init void reserve_ibft_region(void) 629 { 630 unsigned long addr, size = 0; 631 632 addr = find_ibft_region(&size); 633 634 if (size) 635 memblock_reserve(addr, size); 636 } 637 638 static bool __init snb_gfx_workaround_needed(void) 639 { 640 #ifdef CONFIG_PCI 641 int i; 642 u16 vendor, devid; 643 static const __initconst u16 snb_ids[] = { 644 0x0102, 645 0x0112, 646 0x0122, 647 0x0106, 648 0x0116, 649 0x0126, 650 0x010a, 651 }; 652 653 /* Assume no if something weird is going on with PCI */ 654 if (!early_pci_allowed()) 655 return false; 656 657 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 658 if (vendor != 0x8086) 659 return false; 660 661 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 662 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 663 if (devid == snb_ids[i]) 664 return true; 665 #endif 666 667 return false; 668 } 669 670 /* 671 * Sandy Bridge graphics has trouble with certain ranges, exclude 672 * them from allocation. 673 */ 674 static void __init trim_snb_memory(void) 675 { 676 static const __initconst unsigned long bad_pages[] = { 677 0x20050000, 678 0x20110000, 679 0x20130000, 680 0x20138000, 681 0x40004000, 682 }; 683 int i; 684 685 if (!snb_gfx_workaround_needed()) 686 return; 687 688 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 689 690 /* 691 * Reserve all memory below the 1 MB mark that has not 692 * already been reserved. 693 */ 694 memblock_reserve(0, 1<<20); 695 696 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 697 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 698 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 699 bad_pages[i]); 700 } 701 } 702 703 /* 704 * Here we put platform-specific memory range workarounds, i.e. 705 * memory known to be corrupt or otherwise in need to be reserved on 706 * specific platforms. 707 * 708 * If this gets used more widely it could use a real dispatch mechanism. 709 */ 710 static void __init trim_platform_memory_ranges(void) 711 { 712 trim_snb_memory(); 713 } 714 715 static void __init trim_bios_range(void) 716 { 717 /* 718 * A special case is the first 4Kb of memory; 719 * This is a BIOS owned area, not kernel ram, but generally 720 * not listed as such in the E820 table. 721 * 722 * This typically reserves additional memory (64KiB by default) 723 * since some BIOSes are known to corrupt low memory. See the 724 * Kconfig help text for X86_RESERVE_LOW. 725 */ 726 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 727 728 /* 729 * special case: Some BIOSen report the PC BIOS 730 * area (640->1Mb) as ram even though it is not. 731 * take them out. 732 */ 733 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 734 735 e820__update_table(e820_table); 736 } 737 738 /* called before trim_bios_range() to spare extra sanitize */ 739 static void __init e820_add_kernel_range(void) 740 { 741 u64 start = __pa_symbol(_text); 742 u64 size = __pa_symbol(_end) - start; 743 744 /* 745 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 746 * attempt to fix it by adding the range. We may have a confused BIOS, 747 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 748 * exclude kernel range. If we really are running on top non-RAM, 749 * we will crash later anyways. 750 */ 751 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 752 return; 753 754 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 755 e820__range_remove(start, size, E820_TYPE_RAM, 0); 756 e820__range_add(start, size, E820_TYPE_RAM); 757 } 758 759 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 760 761 static int __init parse_reservelow(char *p) 762 { 763 unsigned long long size; 764 765 if (!p) 766 return -EINVAL; 767 768 size = memparse(p, &p); 769 770 if (size < 4096) 771 size = 4096; 772 773 if (size > 640*1024) 774 size = 640*1024; 775 776 reserve_low = size; 777 778 return 0; 779 } 780 781 early_param("reservelow", parse_reservelow); 782 783 static void __init trim_low_memory_range(void) 784 { 785 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 786 } 787 788 /* 789 * Dump out kernel offset information on panic. 790 */ 791 static int 792 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 793 { 794 if (kaslr_enabled()) { 795 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 796 kaslr_offset(), 797 __START_KERNEL, 798 __START_KERNEL_map, 799 MODULES_VADDR-1); 800 } else { 801 pr_emerg("Kernel Offset: disabled\n"); 802 } 803 804 return 0; 805 } 806 807 /* 808 * Determine if we were loaded by an EFI loader. If so, then we have also been 809 * passed the efi memmap, systab, etc., so we should use these data structures 810 * for initialization. Note, the efi init code path is determined by the 811 * global efi_enabled. This allows the same kernel image to be used on existing 812 * systems (with a traditional BIOS) as well as on EFI systems. 813 */ 814 /* 815 * setup_arch - architecture-specific boot-time initializations 816 * 817 * Note: On x86_64, fixmaps are ready for use even before this is called. 818 */ 819 820 void __init setup_arch(char **cmdline_p) 821 { 822 memblock_reserve(__pa_symbol(_text), 823 (unsigned long)__bss_stop - (unsigned long)_text); 824 825 /* 826 * Make sure page 0 is always reserved because on systems with 827 * L1TF its contents can be leaked to user processes. 828 */ 829 memblock_reserve(0, PAGE_SIZE); 830 831 early_reserve_initrd(); 832 833 /* 834 * At this point everything still needed from the boot loader 835 * or BIOS or kernel text should be early reserved or marked not 836 * RAM in e820. All other memory is free game. 837 */ 838 839 #ifdef CONFIG_X86_32 840 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 841 842 /* 843 * copy kernel address range established so far and switch 844 * to the proper swapper page table 845 */ 846 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 847 initial_page_table + KERNEL_PGD_BOUNDARY, 848 KERNEL_PGD_PTRS); 849 850 load_cr3(swapper_pg_dir); 851 /* 852 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 853 * a cr3 based tlb flush, so the following __flush_tlb_all() 854 * will not flush anything because the cpu quirk which clears 855 * X86_FEATURE_PGE has not been invoked yet. Though due to the 856 * load_cr3() above the TLB has been flushed already. The 857 * quirk is invoked before subsequent calls to __flush_tlb_all() 858 * so proper operation is guaranteed. 859 */ 860 __flush_tlb_all(); 861 #else 862 printk(KERN_INFO "Command line: %s\n", boot_command_line); 863 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 864 #endif 865 866 /* 867 * If we have OLPC OFW, we might end up relocating the fixmap due to 868 * reserve_top(), so do this before touching the ioremap area. 869 */ 870 olpc_ofw_detect(); 871 872 idt_setup_early_traps(); 873 early_cpu_init(); 874 arch_init_ideal_nops(); 875 jump_label_init(); 876 early_ioremap_init(); 877 878 setup_olpc_ofw_pgd(); 879 880 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 881 screen_info = boot_params.screen_info; 882 edid_info = boot_params.edid_info; 883 #ifdef CONFIG_X86_32 884 apm_info.bios = boot_params.apm_bios_info; 885 ist_info = boot_params.ist_info; 886 #endif 887 saved_video_mode = boot_params.hdr.vid_mode; 888 bootloader_type = boot_params.hdr.type_of_loader; 889 if ((bootloader_type >> 4) == 0xe) { 890 bootloader_type &= 0xf; 891 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 892 } 893 bootloader_version = bootloader_type & 0xf; 894 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 895 896 #ifdef CONFIG_BLK_DEV_RAM 897 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 898 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 899 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 900 #endif 901 #ifdef CONFIG_EFI 902 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 903 EFI32_LOADER_SIGNATURE, 4)) { 904 set_bit(EFI_BOOT, &efi.flags); 905 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 906 EFI64_LOADER_SIGNATURE, 4)) { 907 set_bit(EFI_BOOT, &efi.flags); 908 set_bit(EFI_64BIT, &efi.flags); 909 } 910 #endif 911 912 x86_init.oem.arch_setup(); 913 914 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 915 e820__memory_setup(); 916 parse_setup_data(); 917 918 copy_edd(); 919 920 if (!boot_params.hdr.root_flags) 921 root_mountflags &= ~MS_RDONLY; 922 init_mm.start_code = (unsigned long) _text; 923 init_mm.end_code = (unsigned long) _etext; 924 init_mm.end_data = (unsigned long) _edata; 925 init_mm.brk = _brk_end; 926 927 mpx_mm_init(&init_mm); 928 929 code_resource.start = __pa_symbol(_text); 930 code_resource.end = __pa_symbol(_etext)-1; 931 data_resource.start = __pa_symbol(_etext); 932 data_resource.end = __pa_symbol(_edata)-1; 933 bss_resource.start = __pa_symbol(__bss_start); 934 bss_resource.end = __pa_symbol(__bss_stop)-1; 935 936 #ifdef CONFIG_CMDLINE_BOOL 937 #ifdef CONFIG_CMDLINE_OVERRIDE 938 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 939 #else 940 if (builtin_cmdline[0]) { 941 /* append boot loader cmdline to builtin */ 942 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 943 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 944 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 945 } 946 #endif 947 #endif 948 949 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 950 *cmdline_p = command_line; 951 952 /* 953 * x86_configure_nx() is called before parse_early_param() to detect 954 * whether hardware doesn't support NX (so that the early EHCI debug 955 * console setup can safely call set_fixmap()). It may then be called 956 * again from within noexec_setup() during parsing early parameters 957 * to honor the respective command line option. 958 */ 959 x86_configure_nx(); 960 961 parse_early_param(); 962 963 if (efi_enabled(EFI_BOOT)) 964 efi_memblock_x86_reserve_range(); 965 #ifdef CONFIG_MEMORY_HOTPLUG 966 /* 967 * Memory used by the kernel cannot be hot-removed because Linux 968 * cannot migrate the kernel pages. When memory hotplug is 969 * enabled, we should prevent memblock from allocating memory 970 * for the kernel. 971 * 972 * ACPI SRAT records all hotpluggable memory ranges. But before 973 * SRAT is parsed, we don't know about it. 974 * 975 * The kernel image is loaded into memory at very early time. We 976 * cannot prevent this anyway. So on NUMA system, we set any 977 * node the kernel resides in as un-hotpluggable. 978 * 979 * Since on modern servers, one node could have double-digit 980 * gigabytes memory, we can assume the memory around the kernel 981 * image is also un-hotpluggable. So before SRAT is parsed, just 982 * allocate memory near the kernel image to try the best to keep 983 * the kernel away from hotpluggable memory. 984 */ 985 if (movable_node_is_enabled()) 986 memblock_set_bottom_up(true); 987 #endif 988 989 x86_report_nx(); 990 991 /* after early param, so could get panic from serial */ 992 memblock_x86_reserve_range_setup_data(); 993 994 if (acpi_mps_check()) { 995 #ifdef CONFIG_X86_LOCAL_APIC 996 disable_apic = 1; 997 #endif 998 setup_clear_cpu_cap(X86_FEATURE_APIC); 999 } 1000 1001 e820__reserve_setup_data(); 1002 e820__finish_early_params(); 1003 1004 if (efi_enabled(EFI_BOOT)) 1005 efi_init(); 1006 1007 dmi_scan_machine(); 1008 dmi_memdev_walk(); 1009 dmi_set_dump_stack_arch_desc(); 1010 1011 /* 1012 * VMware detection requires dmi to be available, so this 1013 * needs to be done after dmi_scan_machine(), for the boot CPU. 1014 */ 1015 init_hypervisor_platform(); 1016 1017 tsc_early_init(); 1018 x86_init.resources.probe_roms(); 1019 1020 /* after parse_early_param, so could debug it */ 1021 insert_resource(&iomem_resource, &code_resource); 1022 insert_resource(&iomem_resource, &data_resource); 1023 insert_resource(&iomem_resource, &bss_resource); 1024 1025 e820_add_kernel_range(); 1026 trim_bios_range(); 1027 #ifdef CONFIG_X86_32 1028 if (ppro_with_ram_bug()) { 1029 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1030 E820_TYPE_RESERVED); 1031 e820__update_table(e820_table); 1032 printk(KERN_INFO "fixed physical RAM map:\n"); 1033 e820__print_table("bad_ppro"); 1034 } 1035 #else 1036 early_gart_iommu_check(); 1037 #endif 1038 1039 /* 1040 * partially used pages are not usable - thus 1041 * we are rounding upwards: 1042 */ 1043 max_pfn = e820__end_of_ram_pfn(); 1044 1045 /* update e820 for memory not covered by WB MTRRs */ 1046 mtrr_bp_init(); 1047 if (mtrr_trim_uncached_memory(max_pfn)) 1048 max_pfn = e820__end_of_ram_pfn(); 1049 1050 max_possible_pfn = max_pfn; 1051 1052 /* 1053 * This call is required when the CPU does not support PAT. If 1054 * mtrr_bp_init() invoked it already via pat_init() the call has no 1055 * effect. 1056 */ 1057 init_cache_modes(); 1058 1059 /* 1060 * Define random base addresses for memory sections after max_pfn is 1061 * defined and before each memory section base is used. 1062 */ 1063 kernel_randomize_memory(); 1064 1065 #ifdef CONFIG_X86_32 1066 /* max_low_pfn get updated here */ 1067 find_low_pfn_range(); 1068 #else 1069 check_x2apic(); 1070 1071 /* How many end-of-memory variables you have, grandma! */ 1072 /* need this before calling reserve_initrd */ 1073 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1074 max_low_pfn = e820__end_of_low_ram_pfn(); 1075 else 1076 max_low_pfn = max_pfn; 1077 1078 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1079 #endif 1080 1081 /* 1082 * Find and reserve possible boot-time SMP configuration: 1083 */ 1084 find_smp_config(); 1085 1086 reserve_ibft_region(); 1087 1088 early_alloc_pgt_buf(); 1089 1090 /* 1091 * Need to conclude brk, before e820__memblock_setup() 1092 * it could use memblock_find_in_range, could overlap with 1093 * brk area. 1094 */ 1095 reserve_brk(); 1096 1097 cleanup_highmap(); 1098 1099 memblock_set_current_limit(ISA_END_ADDRESS); 1100 e820__memblock_setup(); 1101 1102 reserve_bios_regions(); 1103 1104 if (efi_enabled(EFI_MEMMAP)) { 1105 efi_fake_memmap(); 1106 efi_find_mirror(); 1107 efi_esrt_init(); 1108 1109 /* 1110 * The EFI specification says that boot service code won't be 1111 * called after ExitBootServices(). This is, in fact, a lie. 1112 */ 1113 efi_reserve_boot_services(); 1114 } 1115 1116 /* preallocate 4k for mptable mpc */ 1117 e820__memblock_alloc_reserved_mpc_new(); 1118 1119 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1120 setup_bios_corruption_check(); 1121 #endif 1122 1123 #ifdef CONFIG_X86_32 1124 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1125 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1126 #endif 1127 1128 reserve_real_mode(); 1129 1130 trim_platform_memory_ranges(); 1131 trim_low_memory_range(); 1132 1133 init_mem_mapping(); 1134 1135 idt_setup_early_pf(); 1136 1137 /* 1138 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1139 * with the current CR4 value. This may not be necessary, but 1140 * auditing all the early-boot CR4 manipulation would be needed to 1141 * rule it out. 1142 * 1143 * Mask off features that don't work outside long mode (just 1144 * PCIDE for now). 1145 */ 1146 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1147 1148 memblock_set_current_limit(get_max_mapped()); 1149 1150 /* 1151 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1152 */ 1153 1154 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1155 if (init_ohci1394_dma_early) 1156 init_ohci1394_dma_on_all_controllers(); 1157 #endif 1158 /* Allocate bigger log buffer */ 1159 setup_log_buf(1); 1160 1161 if (efi_enabled(EFI_BOOT)) { 1162 switch (boot_params.secure_boot) { 1163 case efi_secureboot_mode_disabled: 1164 pr_info("Secure boot disabled\n"); 1165 break; 1166 case efi_secureboot_mode_enabled: 1167 pr_info("Secure boot enabled\n"); 1168 break; 1169 default: 1170 pr_info("Secure boot could not be determined\n"); 1171 break; 1172 } 1173 } 1174 1175 reserve_initrd(); 1176 1177 acpi_table_upgrade(); 1178 1179 vsmp_init(); 1180 1181 io_delay_init(); 1182 1183 early_platform_quirks(); 1184 1185 /* 1186 * Parse the ACPI tables for possible boot-time SMP configuration. 1187 */ 1188 acpi_boot_table_init(); 1189 1190 early_acpi_boot_init(); 1191 1192 initmem_init(); 1193 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1194 1195 /* 1196 * Reserve memory for crash kernel after SRAT is parsed so that it 1197 * won't consume hotpluggable memory. 1198 */ 1199 reserve_crashkernel(); 1200 1201 memblock_find_dma_reserve(); 1202 1203 if (!early_xdbc_setup_hardware()) 1204 early_xdbc_register_console(); 1205 1206 x86_init.paging.pagetable_init(); 1207 1208 kasan_init(); 1209 1210 /* 1211 * Sync back kernel address range. 1212 * 1213 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1214 * this call? 1215 */ 1216 sync_initial_page_table(); 1217 1218 tboot_probe(); 1219 1220 map_vsyscall(); 1221 1222 generic_apic_probe(); 1223 1224 early_quirks(); 1225 1226 /* 1227 * Read APIC and some other early information from ACPI tables. 1228 */ 1229 acpi_boot_init(); 1230 sfi_init(); 1231 x86_dtb_init(); 1232 1233 /* 1234 * get boot-time SMP configuration: 1235 */ 1236 get_smp_config(); 1237 1238 /* 1239 * Systems w/o ACPI and mptables might not have it mapped the local 1240 * APIC yet, but prefill_possible_map() might need to access it. 1241 */ 1242 init_apic_mappings(); 1243 1244 prefill_possible_map(); 1245 1246 init_cpu_to_node(); 1247 1248 io_apic_init_mappings(); 1249 1250 x86_init.hyper.guest_late_init(); 1251 1252 e820__reserve_resources(); 1253 e820__register_nosave_regions(max_pfn); 1254 1255 x86_init.resources.reserve_resources(); 1256 1257 e820__setup_pci_gap(); 1258 1259 #ifdef CONFIG_VT 1260 #if defined(CONFIG_VGA_CONSOLE) 1261 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1262 conswitchp = &vga_con; 1263 #elif defined(CONFIG_DUMMY_CONSOLE) 1264 conswitchp = &dummy_con; 1265 #endif 1266 #endif 1267 x86_init.oem.banner(); 1268 1269 x86_init.timers.wallclock_init(); 1270 1271 mcheck_init(); 1272 1273 register_refined_jiffies(CLOCK_TICK_RATE); 1274 1275 #ifdef CONFIG_EFI 1276 if (efi_enabled(EFI_BOOT)) 1277 efi_apply_memmap_quirks(); 1278 #endif 1279 1280 unwind_init(); 1281 } 1282 1283 /* 1284 * From boot protocol 2.14 onwards we expect the bootloader to set the 1285 * version to "0x8000 | <used version>". In case we find a version >= 2.14 1286 * without the 0x8000 we assume the boot loader supports 2.13 only and 1287 * reset the version accordingly. The 0x8000 flag is removed in any case. 1288 */ 1289 void __init x86_verify_bootdata_version(void) 1290 { 1291 if (boot_params.hdr.version & VERSION_WRITTEN) 1292 boot_params.hdr.version &= ~VERSION_WRITTEN; 1293 else if (boot_params.hdr.version >= 0x020e) 1294 boot_params.hdr.version = 0x020d; 1295 1296 if (boot_params.hdr.version < 0x020e) 1297 boot_params.hdr.acpi_rsdp_addr = 0; 1298 } 1299 1300 #ifdef CONFIG_X86_32 1301 1302 static struct resource video_ram_resource = { 1303 .name = "Video RAM area", 1304 .start = 0xa0000, 1305 .end = 0xbffff, 1306 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1307 }; 1308 1309 void __init i386_reserve_resources(void) 1310 { 1311 request_resource(&iomem_resource, &video_ram_resource); 1312 reserve_standard_io_resources(); 1313 } 1314 1315 #endif /* CONFIG_X86_32 */ 1316 1317 static struct notifier_block kernel_offset_notifier = { 1318 .notifier_call = dump_kernel_offset 1319 }; 1320 1321 static int __init register_kernel_offset_dumper(void) 1322 { 1323 atomic_notifier_chain_register(&panic_notifier_list, 1324 &kernel_offset_notifier); 1325 return 0; 1326 } 1327 __initcall(register_kernel_offset_dumper); 1328