1 /* 2 * Copyright (C) 1995 Linus Torvalds 3 * 4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 5 * 6 * Memory region support 7 * David Parsons <orc@pell.chi.il.us>, July-August 1999 8 * 9 * Added E820 sanitization routine (removes overlapping memory regions); 10 * Brian Moyle <bmoyle@mvista.com>, February 2001 11 * 12 * Moved CPU detection code to cpu/${cpu}.c 13 * Patrick Mochel <mochel@osdl.org>, March 2002 14 * 15 * Provisions for empty E820 memory regions (reported by certain BIOSes). 16 * Alex Achenbach <xela@slit.de>, December 2002. 17 * 18 */ 19 20 /* 21 * This file handles the architecture-dependent parts of initialization 22 */ 23 24 #include <linux/sched.h> 25 #include <linux/mm.h> 26 #include <linux/mmzone.h> 27 #include <linux/screen_info.h> 28 #include <linux/ioport.h> 29 #include <linux/acpi.h> 30 #include <linux/sfi.h> 31 #include <linux/apm_bios.h> 32 #include <linux/initrd.h> 33 #include <linux/bootmem.h> 34 #include <linux/memblock.h> 35 #include <linux/seq_file.h> 36 #include <linux/console.h> 37 #include <linux/root_dev.h> 38 #include <linux/highmem.h> 39 #include <linux/export.h> 40 #include <linux/efi.h> 41 #include <linux/init.h> 42 #include <linux/edd.h> 43 #include <linux/iscsi_ibft.h> 44 #include <linux/nodemask.h> 45 #include <linux/kexec.h> 46 #include <linux/dmi.h> 47 #include <linux/pfn.h> 48 #include <linux/pci.h> 49 #include <asm/pci-direct.h> 50 #include <linux/init_ohci1394_dma.h> 51 #include <linux/kvm_para.h> 52 #include <linux/dma-contiguous.h> 53 54 #include <linux/errno.h> 55 #include <linux/kernel.h> 56 #include <linux/stddef.h> 57 #include <linux/unistd.h> 58 #include <linux/ptrace.h> 59 #include <linux/user.h> 60 #include <linux/delay.h> 61 62 #include <linux/kallsyms.h> 63 #include <linux/cpufreq.h> 64 #include <linux/dma-mapping.h> 65 #include <linux/ctype.h> 66 #include <linux/uaccess.h> 67 68 #include <linux/percpu.h> 69 #include <linux/crash_dump.h> 70 #include <linux/tboot.h> 71 #include <linux/jiffies.h> 72 #include <linux/mem_encrypt.h> 73 74 #include <linux/usb/xhci-dbgp.h> 75 #include <video/edid.h> 76 77 #include <asm/mtrr.h> 78 #include <asm/apic.h> 79 #include <asm/realmode.h> 80 #include <asm/e820/api.h> 81 #include <asm/mpspec.h> 82 #include <asm/setup.h> 83 #include <asm/efi.h> 84 #include <asm/timer.h> 85 #include <asm/i8259.h> 86 #include <asm/sections.h> 87 #include <asm/io_apic.h> 88 #include <asm/ist.h> 89 #include <asm/setup_arch.h> 90 #include <asm/bios_ebda.h> 91 #include <asm/cacheflush.h> 92 #include <asm/processor.h> 93 #include <asm/bugs.h> 94 #include <asm/kasan.h> 95 96 #include <asm/vsyscall.h> 97 #include <asm/cpu.h> 98 #include <asm/desc.h> 99 #include <asm/dma.h> 100 #include <asm/iommu.h> 101 #include <asm/gart.h> 102 #include <asm/mmu_context.h> 103 #include <asm/proto.h> 104 105 #include <asm/paravirt.h> 106 #include <asm/hypervisor.h> 107 #include <asm/olpc_ofw.h> 108 109 #include <asm/percpu.h> 110 #include <asm/topology.h> 111 #include <asm/apicdef.h> 112 #include <asm/amd_nb.h> 113 #include <asm/mce.h> 114 #include <asm/alternative.h> 115 #include <asm/prom.h> 116 #include <asm/microcode.h> 117 #include <asm/mmu_context.h> 118 #include <asm/kaslr.h> 119 #include <asm/unwind.h> 120 121 /* 122 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 123 * max_pfn_mapped: highest direct mapped pfn over 4GB 124 * 125 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 126 * represented by pfn_mapped 127 */ 128 unsigned long max_low_pfn_mapped; 129 unsigned long max_pfn_mapped; 130 131 #ifdef CONFIG_DMI 132 RESERVE_BRK(dmi_alloc, 65536); 133 #endif 134 135 136 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 137 unsigned long _brk_end = (unsigned long)__brk_base; 138 139 struct boot_params boot_params; 140 141 /* 142 * Machine setup.. 143 */ 144 static struct resource data_resource = { 145 .name = "Kernel data", 146 .start = 0, 147 .end = 0, 148 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 149 }; 150 151 static struct resource code_resource = { 152 .name = "Kernel code", 153 .start = 0, 154 .end = 0, 155 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 156 }; 157 158 static struct resource bss_resource = { 159 .name = "Kernel bss", 160 .start = 0, 161 .end = 0, 162 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 163 }; 164 165 166 #ifdef CONFIG_X86_32 167 /* cpu data as detected by the assembly code in head_32.S */ 168 struct cpuinfo_x86 new_cpu_data; 169 170 /* common cpu data for all cpus */ 171 struct cpuinfo_x86 boot_cpu_data __read_mostly; 172 EXPORT_SYMBOL(boot_cpu_data); 173 174 unsigned int def_to_bigsmp; 175 176 /* for MCA, but anyone else can use it if they want */ 177 unsigned int machine_id; 178 unsigned int machine_submodel_id; 179 unsigned int BIOS_revision; 180 181 struct apm_info apm_info; 182 EXPORT_SYMBOL(apm_info); 183 184 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 185 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 186 struct ist_info ist_info; 187 EXPORT_SYMBOL(ist_info); 188 #else 189 struct ist_info ist_info; 190 #endif 191 192 #else 193 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 194 .x86_phys_bits = MAX_PHYSMEM_BITS, 195 }; 196 EXPORT_SYMBOL(boot_cpu_data); 197 #endif 198 199 200 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 201 __visible unsigned long mmu_cr4_features __ro_after_init; 202 #else 203 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 204 #endif 205 206 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 207 int bootloader_type, bootloader_version; 208 209 /* 210 * Setup options 211 */ 212 struct screen_info screen_info; 213 EXPORT_SYMBOL(screen_info); 214 struct edid_info edid_info; 215 EXPORT_SYMBOL_GPL(edid_info); 216 217 extern int root_mountflags; 218 219 unsigned long saved_video_mode; 220 221 #define RAMDISK_IMAGE_START_MASK 0x07FF 222 #define RAMDISK_PROMPT_FLAG 0x8000 223 #define RAMDISK_LOAD_FLAG 0x4000 224 225 static char __initdata command_line[COMMAND_LINE_SIZE]; 226 #ifdef CONFIG_CMDLINE_BOOL 227 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 228 #endif 229 230 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 231 struct edd edd; 232 #ifdef CONFIG_EDD_MODULE 233 EXPORT_SYMBOL(edd); 234 #endif 235 /** 236 * copy_edd() - Copy the BIOS EDD information 237 * from boot_params into a safe place. 238 * 239 */ 240 static inline void __init copy_edd(void) 241 { 242 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 243 sizeof(edd.mbr_signature)); 244 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 245 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 246 edd.edd_info_nr = boot_params.eddbuf_entries; 247 } 248 #else 249 static inline void __init copy_edd(void) 250 { 251 } 252 #endif 253 254 void * __init extend_brk(size_t size, size_t align) 255 { 256 size_t mask = align - 1; 257 void *ret; 258 259 BUG_ON(_brk_start == 0); 260 BUG_ON(align & mask); 261 262 _brk_end = (_brk_end + mask) & ~mask; 263 BUG_ON((char *)(_brk_end + size) > __brk_limit); 264 265 ret = (void *)_brk_end; 266 _brk_end += size; 267 268 memset(ret, 0, size); 269 270 return ret; 271 } 272 273 #ifdef CONFIG_X86_32 274 static void __init cleanup_highmap(void) 275 { 276 } 277 #endif 278 279 static void __init reserve_brk(void) 280 { 281 if (_brk_end > _brk_start) 282 memblock_reserve(__pa_symbol(_brk_start), 283 _brk_end - _brk_start); 284 285 /* Mark brk area as locked down and no longer taking any 286 new allocations */ 287 _brk_start = 0; 288 } 289 290 u64 relocated_ramdisk; 291 292 #ifdef CONFIG_BLK_DEV_INITRD 293 294 static u64 __init get_ramdisk_image(void) 295 { 296 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 297 298 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 299 300 return ramdisk_image; 301 } 302 static u64 __init get_ramdisk_size(void) 303 { 304 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 305 306 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 307 308 return ramdisk_size; 309 } 310 311 static void __init relocate_initrd(void) 312 { 313 /* Assume only end is not page aligned */ 314 u64 ramdisk_image = get_ramdisk_image(); 315 u64 ramdisk_size = get_ramdisk_size(); 316 u64 area_size = PAGE_ALIGN(ramdisk_size); 317 318 /* We need to move the initrd down into directly mapped mem */ 319 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 320 area_size, PAGE_SIZE); 321 322 if (!relocated_ramdisk) 323 panic("Cannot find place for new RAMDISK of size %lld\n", 324 ramdisk_size); 325 326 /* Note: this includes all the mem currently occupied by 327 the initrd, we rely on that fact to keep the data intact. */ 328 memblock_reserve(relocated_ramdisk, area_size); 329 initrd_start = relocated_ramdisk + PAGE_OFFSET; 330 initrd_end = initrd_start + ramdisk_size; 331 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 332 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 333 334 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 335 336 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 337 " [mem %#010llx-%#010llx]\n", 338 ramdisk_image, ramdisk_image + ramdisk_size - 1, 339 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 340 } 341 342 static void __init early_reserve_initrd(void) 343 { 344 /* Assume only end is not page aligned */ 345 u64 ramdisk_image = get_ramdisk_image(); 346 u64 ramdisk_size = get_ramdisk_size(); 347 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 348 349 if (!boot_params.hdr.type_of_loader || 350 !ramdisk_image || !ramdisk_size) 351 return; /* No initrd provided by bootloader */ 352 353 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 354 } 355 static void __init reserve_initrd(void) 356 { 357 /* Assume only end is not page aligned */ 358 u64 ramdisk_image = get_ramdisk_image(); 359 u64 ramdisk_size = get_ramdisk_size(); 360 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 361 u64 mapped_size; 362 363 if (!boot_params.hdr.type_of_loader || 364 !ramdisk_image || !ramdisk_size) 365 return; /* No initrd provided by bootloader */ 366 367 /* 368 * If SME is active, this memory will be marked encrypted by the 369 * kernel when it is accessed (including relocation). However, the 370 * ramdisk image was loaded decrypted by the bootloader, so make 371 * sure that it is encrypted before accessing it. For SEV the 372 * ramdisk will already be encrypted, so only do this for SME. 373 */ 374 if (sme_active()) 375 sme_early_encrypt(ramdisk_image, ramdisk_end - ramdisk_image); 376 377 initrd_start = 0; 378 379 mapped_size = memblock_mem_size(max_pfn_mapped); 380 if (ramdisk_size >= (mapped_size>>1)) 381 panic("initrd too large to handle, " 382 "disabling initrd (%lld needed, %lld available)\n", 383 ramdisk_size, mapped_size>>1); 384 385 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 386 ramdisk_end - 1); 387 388 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 389 PFN_DOWN(ramdisk_end))) { 390 /* All are mapped, easy case */ 391 initrd_start = ramdisk_image + PAGE_OFFSET; 392 initrd_end = initrd_start + ramdisk_size; 393 return; 394 } 395 396 relocate_initrd(); 397 398 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 399 } 400 401 #else 402 static void __init early_reserve_initrd(void) 403 { 404 } 405 static void __init reserve_initrd(void) 406 { 407 } 408 #endif /* CONFIG_BLK_DEV_INITRD */ 409 410 static void __init parse_setup_data(void) 411 { 412 struct setup_data *data; 413 u64 pa_data, pa_next; 414 415 pa_data = boot_params.hdr.setup_data; 416 while (pa_data) { 417 u32 data_len, data_type; 418 419 data = early_memremap(pa_data, sizeof(*data)); 420 data_len = data->len + sizeof(struct setup_data); 421 data_type = data->type; 422 pa_next = data->next; 423 early_memunmap(data, sizeof(*data)); 424 425 switch (data_type) { 426 case SETUP_E820_EXT: 427 e820__memory_setup_extended(pa_data, data_len); 428 break; 429 case SETUP_DTB: 430 add_dtb(pa_data); 431 break; 432 case SETUP_EFI: 433 parse_efi_setup(pa_data, data_len); 434 break; 435 default: 436 break; 437 } 438 pa_data = pa_next; 439 } 440 } 441 442 static void __init memblock_x86_reserve_range_setup_data(void) 443 { 444 struct setup_data *data; 445 u64 pa_data; 446 447 pa_data = boot_params.hdr.setup_data; 448 while (pa_data) { 449 data = early_memremap(pa_data, sizeof(*data)); 450 memblock_reserve(pa_data, sizeof(*data) + data->len); 451 pa_data = data->next; 452 early_memunmap(data, sizeof(*data)); 453 } 454 } 455 456 /* 457 * --------- Crashkernel reservation ------------------------------ 458 */ 459 460 #ifdef CONFIG_KEXEC_CORE 461 462 /* 16M alignment for crash kernel regions */ 463 #define CRASH_ALIGN (16 << 20) 464 465 /* 466 * Keep the crash kernel below this limit. On 32 bits earlier kernels 467 * would limit the kernel to the low 512 MiB due to mapping restrictions. 468 * On 64bit, old kexec-tools need to under 896MiB. 469 */ 470 #ifdef CONFIG_X86_32 471 # define CRASH_ADDR_LOW_MAX (512 << 20) 472 # define CRASH_ADDR_HIGH_MAX (512 << 20) 473 #else 474 # define CRASH_ADDR_LOW_MAX (896UL << 20) 475 # define CRASH_ADDR_HIGH_MAX MAXMEM 476 #endif 477 478 static int __init reserve_crashkernel_low(void) 479 { 480 #ifdef CONFIG_X86_64 481 unsigned long long base, low_base = 0, low_size = 0; 482 unsigned long total_low_mem; 483 int ret; 484 485 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 486 487 /* crashkernel=Y,low */ 488 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 489 if (ret) { 490 /* 491 * two parts from lib/swiotlb.c: 492 * -swiotlb size: user-specified with swiotlb= or default. 493 * 494 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 495 * to 8M for other buffers that may need to stay low too. Also 496 * make sure we allocate enough extra low memory so that we 497 * don't run out of DMA buffers for 32-bit devices. 498 */ 499 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 500 } else { 501 /* passed with crashkernel=0,low ? */ 502 if (!low_size) 503 return 0; 504 } 505 506 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 507 if (!low_base) { 508 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 509 (unsigned long)(low_size >> 20)); 510 return -ENOMEM; 511 } 512 513 ret = memblock_reserve(low_base, low_size); 514 if (ret) { 515 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 516 return ret; 517 } 518 519 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 520 (unsigned long)(low_size >> 20), 521 (unsigned long)(low_base >> 20), 522 (unsigned long)(total_low_mem >> 20)); 523 524 crashk_low_res.start = low_base; 525 crashk_low_res.end = low_base + low_size - 1; 526 insert_resource(&iomem_resource, &crashk_low_res); 527 #endif 528 return 0; 529 } 530 531 static void __init reserve_crashkernel(void) 532 { 533 unsigned long long crash_size, crash_base, total_mem; 534 bool high = false; 535 int ret; 536 537 total_mem = memblock_phys_mem_size(); 538 539 /* crashkernel=XM */ 540 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 541 if (ret != 0 || crash_size <= 0) { 542 /* crashkernel=X,high */ 543 ret = parse_crashkernel_high(boot_command_line, total_mem, 544 &crash_size, &crash_base); 545 if (ret != 0 || crash_size <= 0) 546 return; 547 high = true; 548 } 549 550 /* 0 means: find the address automatically */ 551 if (crash_base <= 0) { 552 /* 553 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 554 * as old kexec-tools loads bzImage below that, unless 555 * "crashkernel=size[KMG],high" is specified. 556 */ 557 crash_base = memblock_find_in_range(CRASH_ALIGN, 558 high ? CRASH_ADDR_HIGH_MAX 559 : CRASH_ADDR_LOW_MAX, 560 crash_size, CRASH_ALIGN); 561 if (!crash_base) { 562 pr_info("crashkernel reservation failed - No suitable area found.\n"); 563 return; 564 } 565 566 } else { 567 unsigned long long start; 568 569 start = memblock_find_in_range(crash_base, 570 crash_base + crash_size, 571 crash_size, 1 << 20); 572 if (start != crash_base) { 573 pr_info("crashkernel reservation failed - memory is in use.\n"); 574 return; 575 } 576 } 577 ret = memblock_reserve(crash_base, crash_size); 578 if (ret) { 579 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 580 return; 581 } 582 583 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 584 memblock_free(crash_base, crash_size); 585 return; 586 } 587 588 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 589 (unsigned long)(crash_size >> 20), 590 (unsigned long)(crash_base >> 20), 591 (unsigned long)(total_mem >> 20)); 592 593 crashk_res.start = crash_base; 594 crashk_res.end = crash_base + crash_size - 1; 595 insert_resource(&iomem_resource, &crashk_res); 596 } 597 #else 598 static void __init reserve_crashkernel(void) 599 { 600 } 601 #endif 602 603 static struct resource standard_io_resources[] = { 604 { .name = "dma1", .start = 0x00, .end = 0x1f, 605 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 606 { .name = "pic1", .start = 0x20, .end = 0x21, 607 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 608 { .name = "timer0", .start = 0x40, .end = 0x43, 609 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 610 { .name = "timer1", .start = 0x50, .end = 0x53, 611 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 612 { .name = "keyboard", .start = 0x60, .end = 0x60, 613 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 614 { .name = "keyboard", .start = 0x64, .end = 0x64, 615 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 616 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 617 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 618 { .name = "pic2", .start = 0xa0, .end = 0xa1, 619 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 620 { .name = "dma2", .start = 0xc0, .end = 0xdf, 621 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 622 { .name = "fpu", .start = 0xf0, .end = 0xff, 623 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 624 }; 625 626 void __init reserve_standard_io_resources(void) 627 { 628 int i; 629 630 /* request I/O space for devices used on all i[345]86 PCs */ 631 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 632 request_resource(&ioport_resource, &standard_io_resources[i]); 633 634 } 635 636 static __init void reserve_ibft_region(void) 637 { 638 unsigned long addr, size = 0; 639 640 addr = find_ibft_region(&size); 641 642 if (size) 643 memblock_reserve(addr, size); 644 } 645 646 static bool __init snb_gfx_workaround_needed(void) 647 { 648 #ifdef CONFIG_PCI 649 int i; 650 u16 vendor, devid; 651 static const __initconst u16 snb_ids[] = { 652 0x0102, 653 0x0112, 654 0x0122, 655 0x0106, 656 0x0116, 657 0x0126, 658 0x010a, 659 }; 660 661 /* Assume no if something weird is going on with PCI */ 662 if (!early_pci_allowed()) 663 return false; 664 665 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 666 if (vendor != 0x8086) 667 return false; 668 669 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 670 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 671 if (devid == snb_ids[i]) 672 return true; 673 #endif 674 675 return false; 676 } 677 678 /* 679 * Sandy Bridge graphics has trouble with certain ranges, exclude 680 * them from allocation. 681 */ 682 static void __init trim_snb_memory(void) 683 { 684 static const __initconst unsigned long bad_pages[] = { 685 0x20050000, 686 0x20110000, 687 0x20130000, 688 0x20138000, 689 0x40004000, 690 }; 691 int i; 692 693 if (!snb_gfx_workaround_needed()) 694 return; 695 696 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 697 698 /* 699 * Reserve all memory below the 1 MB mark that has not 700 * already been reserved. 701 */ 702 memblock_reserve(0, 1<<20); 703 704 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 705 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 706 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 707 bad_pages[i]); 708 } 709 } 710 711 /* 712 * Here we put platform-specific memory range workarounds, i.e. 713 * memory known to be corrupt or otherwise in need to be reserved on 714 * specific platforms. 715 * 716 * If this gets used more widely it could use a real dispatch mechanism. 717 */ 718 static void __init trim_platform_memory_ranges(void) 719 { 720 trim_snb_memory(); 721 } 722 723 static void __init trim_bios_range(void) 724 { 725 /* 726 * A special case is the first 4Kb of memory; 727 * This is a BIOS owned area, not kernel ram, but generally 728 * not listed as such in the E820 table. 729 * 730 * This typically reserves additional memory (64KiB by default) 731 * since some BIOSes are known to corrupt low memory. See the 732 * Kconfig help text for X86_RESERVE_LOW. 733 */ 734 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 735 736 /* 737 * special case: Some BIOSen report the PC BIOS 738 * area (640->1Mb) as ram even though it is not. 739 * take them out. 740 */ 741 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 742 743 e820__update_table(e820_table); 744 } 745 746 /* called before trim_bios_range() to spare extra sanitize */ 747 static void __init e820_add_kernel_range(void) 748 { 749 u64 start = __pa_symbol(_text); 750 u64 size = __pa_symbol(_end) - start; 751 752 /* 753 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 754 * attempt to fix it by adding the range. We may have a confused BIOS, 755 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 756 * exclude kernel range. If we really are running on top non-RAM, 757 * we will crash later anyways. 758 */ 759 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 760 return; 761 762 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 763 e820__range_remove(start, size, E820_TYPE_RAM, 0); 764 e820__range_add(start, size, E820_TYPE_RAM); 765 } 766 767 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 768 769 static int __init parse_reservelow(char *p) 770 { 771 unsigned long long size; 772 773 if (!p) 774 return -EINVAL; 775 776 size = memparse(p, &p); 777 778 if (size < 4096) 779 size = 4096; 780 781 if (size > 640*1024) 782 size = 640*1024; 783 784 reserve_low = size; 785 786 return 0; 787 } 788 789 early_param("reservelow", parse_reservelow); 790 791 static void __init trim_low_memory_range(void) 792 { 793 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 794 } 795 796 /* 797 * Dump out kernel offset information on panic. 798 */ 799 static int 800 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 801 { 802 if (kaslr_enabled()) { 803 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 804 kaslr_offset(), 805 __START_KERNEL, 806 __START_KERNEL_map, 807 MODULES_VADDR-1); 808 } else { 809 pr_emerg("Kernel Offset: disabled\n"); 810 } 811 812 return 0; 813 } 814 815 /* 816 * Determine if we were loaded by an EFI loader. If so, then we have also been 817 * passed the efi memmap, systab, etc., so we should use these data structures 818 * for initialization. Note, the efi init code path is determined by the 819 * global efi_enabled. This allows the same kernel image to be used on existing 820 * systems (with a traditional BIOS) as well as on EFI systems. 821 */ 822 /* 823 * setup_arch - architecture-specific boot-time initializations 824 * 825 * Note: On x86_64, fixmaps are ready for use even before this is called. 826 */ 827 828 void __init setup_arch(char **cmdline_p) 829 { 830 memblock_reserve(__pa_symbol(_text), 831 (unsigned long)__bss_stop - (unsigned long)_text); 832 833 early_reserve_initrd(); 834 835 /* 836 * At this point everything still needed from the boot loader 837 * or BIOS or kernel text should be early reserved or marked not 838 * RAM in e820. All other memory is free game. 839 */ 840 841 #ifdef CONFIG_X86_32 842 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 843 844 /* 845 * copy kernel address range established so far and switch 846 * to the proper swapper page table 847 */ 848 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 849 initial_page_table + KERNEL_PGD_BOUNDARY, 850 KERNEL_PGD_PTRS); 851 852 load_cr3(swapper_pg_dir); 853 /* 854 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 855 * a cr3 based tlb flush, so the following __flush_tlb_all() 856 * will not flush anything because the cpu quirk which clears 857 * X86_FEATURE_PGE has not been invoked yet. Though due to the 858 * load_cr3() above the TLB has been flushed already. The 859 * quirk is invoked before subsequent calls to __flush_tlb_all() 860 * so proper operation is guaranteed. 861 */ 862 __flush_tlb_all(); 863 #else 864 printk(KERN_INFO "Command line: %s\n", boot_command_line); 865 #endif 866 867 /* 868 * If we have OLPC OFW, we might end up relocating the fixmap due to 869 * reserve_top(), so do this before touching the ioremap area. 870 */ 871 olpc_ofw_detect(); 872 873 idt_setup_early_traps(); 874 early_cpu_init(); 875 early_ioremap_init(); 876 877 setup_olpc_ofw_pgd(); 878 879 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 880 screen_info = boot_params.screen_info; 881 edid_info = boot_params.edid_info; 882 #ifdef CONFIG_X86_32 883 apm_info.bios = boot_params.apm_bios_info; 884 ist_info = boot_params.ist_info; 885 #endif 886 saved_video_mode = boot_params.hdr.vid_mode; 887 bootloader_type = boot_params.hdr.type_of_loader; 888 if ((bootloader_type >> 4) == 0xe) { 889 bootloader_type &= 0xf; 890 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 891 } 892 bootloader_version = bootloader_type & 0xf; 893 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 894 895 #ifdef CONFIG_BLK_DEV_RAM 896 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 897 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 898 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 899 #endif 900 #ifdef CONFIG_EFI 901 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 902 EFI32_LOADER_SIGNATURE, 4)) { 903 set_bit(EFI_BOOT, &efi.flags); 904 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 905 EFI64_LOADER_SIGNATURE, 4)) { 906 set_bit(EFI_BOOT, &efi.flags); 907 set_bit(EFI_64BIT, &efi.flags); 908 } 909 910 if (efi_enabled(EFI_BOOT)) 911 efi_memblock_x86_reserve_range(); 912 #endif 913 914 x86_init.oem.arch_setup(); 915 916 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 917 e820__memory_setup(); 918 parse_setup_data(); 919 920 copy_edd(); 921 922 if (!boot_params.hdr.root_flags) 923 root_mountflags &= ~MS_RDONLY; 924 init_mm.start_code = (unsigned long) _text; 925 init_mm.end_code = (unsigned long) _etext; 926 init_mm.end_data = (unsigned long) _edata; 927 init_mm.brk = _brk_end; 928 929 mpx_mm_init(&init_mm); 930 931 code_resource.start = __pa_symbol(_text); 932 code_resource.end = __pa_symbol(_etext)-1; 933 data_resource.start = __pa_symbol(_etext); 934 data_resource.end = __pa_symbol(_edata)-1; 935 bss_resource.start = __pa_symbol(__bss_start); 936 bss_resource.end = __pa_symbol(__bss_stop)-1; 937 938 #ifdef CONFIG_CMDLINE_BOOL 939 #ifdef CONFIG_CMDLINE_OVERRIDE 940 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 941 #else 942 if (builtin_cmdline[0]) { 943 /* append boot loader cmdline to builtin */ 944 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 945 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 946 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 947 } 948 #endif 949 #endif 950 951 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 952 *cmdline_p = command_line; 953 954 /* 955 * x86_configure_nx() is called before parse_early_param() to detect 956 * whether hardware doesn't support NX (so that the early EHCI debug 957 * console setup can safely call set_fixmap()). It may then be called 958 * again from within noexec_setup() during parsing early parameters 959 * to honor the respective command line option. 960 */ 961 x86_configure_nx(); 962 963 parse_early_param(); 964 965 #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 #ifdef CONFIG_PCI 1002 if (pci_early_dump_regs) 1003 early_dump_pci_devices(); 1004 #endif 1005 1006 e820__reserve_setup_data(); 1007 e820__finish_early_params(); 1008 1009 if (efi_enabled(EFI_BOOT)) 1010 efi_init(); 1011 1012 dmi_scan_machine(); 1013 dmi_memdev_walk(); 1014 dmi_set_dump_stack_arch_desc(); 1015 1016 /* 1017 * VMware detection requires dmi to be available, so this 1018 * needs to be done after dmi_scan_machine(), for the boot CPU. 1019 */ 1020 init_hypervisor_platform(); 1021 1022 x86_init.resources.probe_roms(); 1023 1024 /* after parse_early_param, so could debug it */ 1025 insert_resource(&iomem_resource, &code_resource); 1026 insert_resource(&iomem_resource, &data_resource); 1027 insert_resource(&iomem_resource, &bss_resource); 1028 1029 e820_add_kernel_range(); 1030 trim_bios_range(); 1031 #ifdef CONFIG_X86_32 1032 if (ppro_with_ram_bug()) { 1033 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1034 E820_TYPE_RESERVED); 1035 e820__update_table(e820_table); 1036 printk(KERN_INFO "fixed physical RAM map:\n"); 1037 e820__print_table("bad_ppro"); 1038 } 1039 #else 1040 early_gart_iommu_check(); 1041 #endif 1042 1043 /* 1044 * partially used pages are not usable - thus 1045 * we are rounding upwards: 1046 */ 1047 max_pfn = e820__end_of_ram_pfn(); 1048 1049 /* update e820 for memory not covered by WB MTRRs */ 1050 mtrr_bp_init(); 1051 if (mtrr_trim_uncached_memory(max_pfn)) 1052 max_pfn = e820__end_of_ram_pfn(); 1053 1054 max_possible_pfn = max_pfn; 1055 1056 /* 1057 * This call is required when the CPU does not support PAT. If 1058 * mtrr_bp_init() invoked it already via pat_init() the call has no 1059 * effect. 1060 */ 1061 init_cache_modes(); 1062 1063 /* 1064 * Define random base addresses for memory sections after max_pfn is 1065 * defined and before each memory section base is used. 1066 */ 1067 kernel_randomize_memory(); 1068 1069 #ifdef CONFIG_X86_32 1070 /* max_low_pfn get updated here */ 1071 find_low_pfn_range(); 1072 #else 1073 check_x2apic(); 1074 1075 /* How many end-of-memory variables you have, grandma! */ 1076 /* need this before calling reserve_initrd */ 1077 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1078 max_low_pfn = e820__end_of_low_ram_pfn(); 1079 else 1080 max_low_pfn = max_pfn; 1081 1082 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1083 #endif 1084 1085 /* 1086 * Find and reserve possible boot-time SMP configuration: 1087 */ 1088 find_smp_config(); 1089 1090 reserve_ibft_region(); 1091 1092 early_alloc_pgt_buf(); 1093 1094 /* 1095 * Need to conclude brk, before e820__memblock_setup() 1096 * it could use memblock_find_in_range, could overlap with 1097 * brk area. 1098 */ 1099 reserve_brk(); 1100 1101 cleanup_highmap(); 1102 1103 memblock_set_current_limit(ISA_END_ADDRESS); 1104 e820__memblock_setup(); 1105 1106 reserve_bios_regions(); 1107 1108 if (efi_enabled(EFI_MEMMAP)) { 1109 efi_fake_memmap(); 1110 efi_find_mirror(); 1111 efi_esrt_init(); 1112 1113 /* 1114 * The EFI specification says that boot service code won't be 1115 * called after ExitBootServices(). This is, in fact, a lie. 1116 */ 1117 efi_reserve_boot_services(); 1118 } 1119 1120 /* preallocate 4k for mptable mpc */ 1121 e820__memblock_alloc_reserved_mpc_new(); 1122 1123 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1124 setup_bios_corruption_check(); 1125 #endif 1126 1127 #ifdef CONFIG_X86_32 1128 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1129 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1130 #endif 1131 1132 reserve_real_mode(); 1133 1134 trim_platform_memory_ranges(); 1135 trim_low_memory_range(); 1136 1137 init_mem_mapping(); 1138 1139 idt_setup_early_pf(); 1140 1141 /* 1142 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1143 * with the current CR4 value. This may not be necessary, but 1144 * auditing all the early-boot CR4 manipulation would be needed to 1145 * rule it out. 1146 * 1147 * Mask off features that don't work outside long mode (just 1148 * PCIDE for now). 1149 */ 1150 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1151 1152 memblock_set_current_limit(get_max_mapped()); 1153 1154 /* 1155 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1156 */ 1157 1158 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1159 if (init_ohci1394_dma_early) 1160 init_ohci1394_dma_on_all_controllers(); 1161 #endif 1162 /* Allocate bigger log buffer */ 1163 setup_log_buf(1); 1164 1165 if (efi_enabled(EFI_BOOT)) { 1166 switch (boot_params.secure_boot) { 1167 case efi_secureboot_mode_disabled: 1168 pr_info("Secure boot disabled\n"); 1169 break; 1170 case efi_secureboot_mode_enabled: 1171 pr_info("Secure boot enabled\n"); 1172 break; 1173 default: 1174 pr_info("Secure boot could not be determined\n"); 1175 break; 1176 } 1177 } 1178 1179 reserve_initrd(); 1180 1181 acpi_table_upgrade(); 1182 1183 vsmp_init(); 1184 1185 io_delay_init(); 1186 1187 early_platform_quirks(); 1188 1189 /* 1190 * Parse the ACPI tables for possible boot-time SMP configuration. 1191 */ 1192 acpi_boot_table_init(); 1193 1194 early_acpi_boot_init(); 1195 1196 initmem_init(); 1197 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1198 1199 /* 1200 * Reserve memory for crash kernel after SRAT is parsed so that it 1201 * won't consume hotpluggable memory. 1202 */ 1203 reserve_crashkernel(); 1204 1205 memblock_find_dma_reserve(); 1206 1207 #ifdef CONFIG_KVM_GUEST 1208 kvmclock_init(); 1209 #endif 1210 1211 tsc_early_delay_calibrate(); 1212 if (!early_xdbc_setup_hardware()) 1213 early_xdbc_register_console(); 1214 1215 x86_init.paging.pagetable_init(); 1216 1217 kasan_init(); 1218 1219 #ifdef CONFIG_X86_32 1220 /* sync back kernel address range */ 1221 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1222 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1223 KERNEL_PGD_PTRS); 1224 1225 /* 1226 * sync back low identity map too. It is used for example 1227 * in the 32-bit EFI stub. 1228 */ 1229 clone_pgd_range(initial_page_table, 1230 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1231 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 1232 #endif 1233 1234 tboot_probe(); 1235 1236 map_vsyscall(); 1237 1238 generic_apic_probe(); 1239 1240 early_quirks(); 1241 1242 /* 1243 * Read APIC and some other early information from ACPI tables. 1244 */ 1245 acpi_boot_init(); 1246 sfi_init(); 1247 x86_dtb_init(); 1248 1249 /* 1250 * get boot-time SMP configuration: 1251 */ 1252 get_smp_config(); 1253 1254 /* 1255 * Systems w/o ACPI and mptables might not have it mapped the local 1256 * APIC yet, but prefill_possible_map() might need to access it. 1257 */ 1258 init_apic_mappings(); 1259 1260 prefill_possible_map(); 1261 1262 init_cpu_to_node(); 1263 1264 io_apic_init_mappings(); 1265 1266 x86_init.hyper.guest_late_init(); 1267 1268 e820__reserve_resources(); 1269 e820__register_nosave_regions(max_low_pfn); 1270 1271 x86_init.resources.reserve_resources(); 1272 1273 e820__setup_pci_gap(); 1274 1275 #ifdef CONFIG_VT 1276 #if defined(CONFIG_VGA_CONSOLE) 1277 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1278 conswitchp = &vga_con; 1279 #elif defined(CONFIG_DUMMY_CONSOLE) 1280 conswitchp = &dummy_con; 1281 #endif 1282 #endif 1283 x86_init.oem.banner(); 1284 1285 x86_init.timers.wallclock_init(); 1286 1287 mcheck_init(); 1288 1289 arch_init_ideal_nops(); 1290 1291 register_refined_jiffies(CLOCK_TICK_RATE); 1292 1293 #ifdef CONFIG_EFI 1294 if (efi_enabled(EFI_BOOT)) 1295 efi_apply_memmap_quirks(); 1296 #endif 1297 1298 unwind_init(); 1299 } 1300 1301 #ifdef CONFIG_X86_32 1302 1303 static struct resource video_ram_resource = { 1304 .name = "Video RAM area", 1305 .start = 0xa0000, 1306 .end = 0xbffff, 1307 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1308 }; 1309 1310 void __init i386_reserve_resources(void) 1311 { 1312 request_resource(&iomem_resource, &video_ram_resource); 1313 reserve_standard_io_resources(); 1314 } 1315 1316 #endif /* CONFIG_X86_32 */ 1317 1318 static struct notifier_block kernel_offset_notifier = { 1319 .notifier_call = dump_kernel_offset 1320 }; 1321 1322 static int __init register_kernel_offset_dumper(void) 1323 { 1324 atomic_notifier_chain_register(&panic_notifier_list, 1325 &kernel_offset_notifier); 1326 return 0; 1327 } 1328 __initcall(register_kernel_offset_dumper); 1329 1330 void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) 1331 { 1332 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 1333 return; 1334 1335 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma)); 1336 } 1337