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