1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1995 Linus Torvalds 4 * 5 * This file contains the setup_arch() code, which handles the architecture-dependent 6 * parts of early kernel initialization. 7 */ 8 #include <linux/console.h> 9 #include <linux/crash_dump.h> 10 #include <linux/dmi.h> 11 #include <linux/efi.h> 12 #include <linux/init_ohci1394_dma.h> 13 #include <linux/initrd.h> 14 #include <linux/iscsi_ibft.h> 15 #include <linux/memblock.h> 16 #include <linux/pci.h> 17 #include <linux/root_dev.h> 18 #include <linux/sfi.h> 19 #include <linux/tboot.h> 20 #include <linux/usb/xhci-dbgp.h> 21 22 #include <uapi/linux/mount.h> 23 24 #include <xen/xen.h> 25 26 #include <asm/apic.h> 27 #include <asm/bios_ebda.h> 28 #include <asm/bugs.h> 29 #include <asm/cpu.h> 30 #include <asm/efi.h> 31 #include <asm/gart.h> 32 #include <asm/hypervisor.h> 33 #include <asm/io_apic.h> 34 #include <asm/kasan.h> 35 #include <asm/kaslr.h> 36 #include <asm/mce.h> 37 #include <asm/mtrr.h> 38 #include <asm/realmode.h> 39 #include <asm/olpc_ofw.h> 40 #include <asm/pci-direct.h> 41 #include <asm/prom.h> 42 #include <asm/proto.h> 43 #include <asm/unwind.h> 44 #include <asm/vsyscall.h> 45 #include <linux/vmalloc.h> 46 47 /* 48 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 49 * max_pfn_mapped: highest directly mapped pfn > 4 GB 50 * 51 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 52 * represented by pfn_mapped[]. 53 */ 54 unsigned long max_low_pfn_mapped; 55 unsigned long max_pfn_mapped; 56 57 #ifdef CONFIG_DMI 58 RESERVE_BRK(dmi_alloc, 65536); 59 #endif 60 61 62 /* 63 * Range of the BSS area. The size of the BSS area is determined 64 * at link time, with RESERVE_BRK*() facility reserving additional 65 * chunks. 66 */ 67 unsigned long _brk_start = (unsigned long)__brk_base; 68 unsigned long _brk_end = (unsigned long)__brk_base; 69 70 struct boot_params boot_params; 71 72 /* 73 * These are the four main kernel memory regions, we put them into 74 * the resource tree so that kdump tools and other debugging tools 75 * recover it: 76 */ 77 78 static struct resource rodata_resource = { 79 .name = "Kernel rodata", 80 .start = 0, 81 .end = 0, 82 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 83 }; 84 85 static struct resource data_resource = { 86 .name = "Kernel data", 87 .start = 0, 88 .end = 0, 89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 90 }; 91 92 static struct resource code_resource = { 93 .name = "Kernel code", 94 .start = 0, 95 .end = 0, 96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 97 }; 98 99 static struct resource bss_resource = { 100 .name = "Kernel bss", 101 .start = 0, 102 .end = 0, 103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 104 }; 105 106 107 #ifdef CONFIG_X86_32 108 /* CPU data as detected by the assembly code in head_32.S */ 109 struct cpuinfo_x86 new_cpu_data; 110 111 /* Common CPU data for all CPUs */ 112 struct cpuinfo_x86 boot_cpu_data __read_mostly; 113 EXPORT_SYMBOL(boot_cpu_data); 114 115 unsigned int def_to_bigsmp; 116 117 /* For MCA, but anyone else can use it if they want */ 118 unsigned int machine_id; 119 unsigned int machine_submodel_id; 120 unsigned int BIOS_revision; 121 122 struct apm_info apm_info; 123 EXPORT_SYMBOL(apm_info); 124 125 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 126 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 127 struct ist_info ist_info; 128 EXPORT_SYMBOL(ist_info); 129 #else 130 struct ist_info ist_info; 131 #endif 132 133 #else 134 struct cpuinfo_x86 boot_cpu_data __read_mostly; 135 EXPORT_SYMBOL(boot_cpu_data); 136 #endif 137 138 139 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 140 __visible unsigned long mmu_cr4_features __ro_after_init; 141 #else 142 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 143 #endif 144 145 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 146 int bootloader_type, bootloader_version; 147 148 /* 149 * Setup options 150 */ 151 struct screen_info screen_info; 152 EXPORT_SYMBOL(screen_info); 153 struct edid_info edid_info; 154 EXPORT_SYMBOL_GPL(edid_info); 155 156 extern int root_mountflags; 157 158 unsigned long saved_video_mode; 159 160 #define RAMDISK_IMAGE_START_MASK 0x07FF 161 #define RAMDISK_PROMPT_FLAG 0x8000 162 #define RAMDISK_LOAD_FLAG 0x4000 163 164 static char __initdata command_line[COMMAND_LINE_SIZE]; 165 #ifdef CONFIG_CMDLINE_BOOL 166 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 167 #endif 168 169 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 170 struct edd edd; 171 #ifdef CONFIG_EDD_MODULE 172 EXPORT_SYMBOL(edd); 173 #endif 174 /** 175 * copy_edd() - Copy the BIOS EDD information 176 * from boot_params into a safe place. 177 * 178 */ 179 static inline void __init copy_edd(void) 180 { 181 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 182 sizeof(edd.mbr_signature)); 183 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 184 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 185 edd.edd_info_nr = boot_params.eddbuf_entries; 186 } 187 #else 188 static inline void __init copy_edd(void) 189 { 190 } 191 #endif 192 193 void * __init extend_brk(size_t size, size_t align) 194 { 195 size_t mask = align - 1; 196 void *ret; 197 198 BUG_ON(_brk_start == 0); 199 BUG_ON(align & mask); 200 201 _brk_end = (_brk_end + mask) & ~mask; 202 BUG_ON((char *)(_brk_end + size) > __brk_limit); 203 204 ret = (void *)_brk_end; 205 _brk_end += size; 206 207 memset(ret, 0, size); 208 209 return ret; 210 } 211 212 #ifdef CONFIG_X86_32 213 static void __init cleanup_highmap(void) 214 { 215 } 216 #endif 217 218 static void __init reserve_brk(void) 219 { 220 if (_brk_end > _brk_start) 221 memblock_reserve(__pa_symbol(_brk_start), 222 _brk_end - _brk_start); 223 224 /* Mark brk area as locked down and no longer taking any 225 new allocations */ 226 _brk_start = 0; 227 } 228 229 u64 relocated_ramdisk; 230 231 #ifdef CONFIG_BLK_DEV_INITRD 232 233 static u64 __init get_ramdisk_image(void) 234 { 235 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 236 237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 238 239 return ramdisk_image; 240 } 241 static u64 __init get_ramdisk_size(void) 242 { 243 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 244 245 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 246 247 return ramdisk_size; 248 } 249 250 static void __init relocate_initrd(void) 251 { 252 /* Assume only end is not page aligned */ 253 u64 ramdisk_image = get_ramdisk_image(); 254 u64 ramdisk_size = get_ramdisk_size(); 255 u64 area_size = PAGE_ALIGN(ramdisk_size); 256 257 /* We need to move the initrd down into directly mapped mem */ 258 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 259 area_size, PAGE_SIZE); 260 261 if (!relocated_ramdisk) 262 panic("Cannot find place for new RAMDISK of size %lld\n", 263 ramdisk_size); 264 265 /* Note: this includes all the mem currently occupied by 266 the initrd, we rely on that fact to keep the data intact. */ 267 memblock_reserve(relocated_ramdisk, area_size); 268 initrd_start = relocated_ramdisk + PAGE_OFFSET; 269 initrd_end = initrd_start + ramdisk_size; 270 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 271 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 272 273 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 274 275 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 276 " [mem %#010llx-%#010llx]\n", 277 ramdisk_image, ramdisk_image + ramdisk_size - 1, 278 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 279 } 280 281 static void __init early_reserve_initrd(void) 282 { 283 /* Assume only end is not page aligned */ 284 u64 ramdisk_image = get_ramdisk_image(); 285 u64 ramdisk_size = get_ramdisk_size(); 286 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 287 288 if (!boot_params.hdr.type_of_loader || 289 !ramdisk_image || !ramdisk_size) 290 return; /* No initrd provided by bootloader */ 291 292 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 293 } 294 static void __init reserve_initrd(void) 295 { 296 /* Assume only end is not page aligned */ 297 u64 ramdisk_image = get_ramdisk_image(); 298 u64 ramdisk_size = get_ramdisk_size(); 299 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 300 u64 mapped_size; 301 302 if (!boot_params.hdr.type_of_loader || 303 !ramdisk_image || !ramdisk_size) 304 return; /* No initrd provided by bootloader */ 305 306 initrd_start = 0; 307 308 mapped_size = memblock_mem_size(max_pfn_mapped); 309 if (ramdisk_size >= (mapped_size>>1)) 310 panic("initrd too large to handle, " 311 "disabling initrd (%lld needed, %lld available)\n", 312 ramdisk_size, mapped_size>>1); 313 314 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 315 ramdisk_end - 1); 316 317 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 318 PFN_DOWN(ramdisk_end))) { 319 /* All are mapped, easy case */ 320 initrd_start = ramdisk_image + PAGE_OFFSET; 321 initrd_end = initrd_start + ramdisk_size; 322 return; 323 } 324 325 relocate_initrd(); 326 327 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 328 } 329 330 #else 331 static void __init early_reserve_initrd(void) 332 { 333 } 334 static void __init reserve_initrd(void) 335 { 336 } 337 #endif /* CONFIG_BLK_DEV_INITRD */ 338 339 static void __init parse_setup_data(void) 340 { 341 struct setup_data *data; 342 u64 pa_data, pa_next; 343 344 pa_data = boot_params.hdr.setup_data; 345 while (pa_data) { 346 u32 data_len, data_type; 347 348 data = early_memremap(pa_data, sizeof(*data)); 349 data_len = data->len + sizeof(struct setup_data); 350 data_type = data->type; 351 pa_next = data->next; 352 early_memunmap(data, sizeof(*data)); 353 354 switch (data_type) { 355 case SETUP_E820_EXT: 356 e820__memory_setup_extended(pa_data, data_len); 357 break; 358 case SETUP_DTB: 359 add_dtb(pa_data); 360 break; 361 case SETUP_EFI: 362 parse_efi_setup(pa_data, data_len); 363 break; 364 default: 365 break; 366 } 367 pa_data = pa_next; 368 } 369 } 370 371 static void __init memblock_x86_reserve_range_setup_data(void) 372 { 373 struct setup_data *data; 374 u64 pa_data; 375 376 pa_data = boot_params.hdr.setup_data; 377 while (pa_data) { 378 data = early_memremap(pa_data, sizeof(*data)); 379 memblock_reserve(pa_data, sizeof(*data) + data->len); 380 381 if (data->type == SETUP_INDIRECT && 382 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) 383 memblock_reserve(((struct setup_indirect *)data->data)->addr, 384 ((struct setup_indirect *)data->data)->len); 385 386 pa_data = data->next; 387 early_memunmap(data, sizeof(*data)); 388 } 389 } 390 391 /* 392 * --------- Crashkernel reservation ------------------------------ 393 */ 394 395 #ifdef CONFIG_KEXEC_CORE 396 397 /* 16M alignment for crash kernel regions */ 398 #define CRASH_ALIGN SZ_16M 399 400 /* 401 * Keep the crash kernel below this limit. 402 * 403 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 404 * due to mapping restrictions. 405 * 406 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 407 * the upper limit of system RAM in 4-level paging mode. Since the kdump 408 * jump could be from 5-level paging to 4-level paging, the jump will fail if 409 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 410 * no good way to detect the paging mode of the target kernel which will be 411 * loaded for dumping. 412 */ 413 #ifdef CONFIG_X86_32 414 # define CRASH_ADDR_LOW_MAX SZ_512M 415 # define CRASH_ADDR_HIGH_MAX SZ_512M 416 #else 417 # define CRASH_ADDR_LOW_MAX SZ_4G 418 # define CRASH_ADDR_HIGH_MAX SZ_64T 419 #endif 420 421 static int __init reserve_crashkernel_low(void) 422 { 423 #ifdef CONFIG_X86_64 424 unsigned long long base, low_base = 0, low_size = 0; 425 unsigned long total_low_mem; 426 int ret; 427 428 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 429 430 /* crashkernel=Y,low */ 431 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 432 if (ret) { 433 /* 434 * two parts from kernel/dma/swiotlb.c: 435 * -swiotlb size: user-specified with swiotlb= or default. 436 * 437 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 438 * to 8M for other buffers that may need to stay low too. Also 439 * make sure we allocate enough extra low memory so that we 440 * don't run out of DMA buffers for 32-bit devices. 441 */ 442 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 443 } else { 444 /* passed with crashkernel=0,low ? */ 445 if (!low_size) 446 return 0; 447 } 448 449 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 450 if (!low_base) { 451 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 452 (unsigned long)(low_size >> 20)); 453 return -ENOMEM; 454 } 455 456 ret = memblock_reserve(low_base, low_size); 457 if (ret) { 458 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 459 return ret; 460 } 461 462 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 463 (unsigned long)(low_size >> 20), 464 (unsigned long)(low_base >> 20), 465 (unsigned long)(total_low_mem >> 20)); 466 467 crashk_low_res.start = low_base; 468 crashk_low_res.end = low_base + low_size - 1; 469 insert_resource(&iomem_resource, &crashk_low_res); 470 #endif 471 return 0; 472 } 473 474 static void __init reserve_crashkernel(void) 475 { 476 unsigned long long crash_size, crash_base, total_mem; 477 bool high = false; 478 int ret; 479 480 total_mem = memblock_phys_mem_size(); 481 482 /* crashkernel=XM */ 483 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 484 if (ret != 0 || crash_size <= 0) { 485 /* crashkernel=X,high */ 486 ret = parse_crashkernel_high(boot_command_line, total_mem, 487 &crash_size, &crash_base); 488 if (ret != 0 || crash_size <= 0) 489 return; 490 high = true; 491 } 492 493 if (xen_pv_domain()) { 494 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 495 return; 496 } 497 498 /* 0 means: find the address automatically */ 499 if (!crash_base) { 500 /* 501 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 502 * crashkernel=x,high reserves memory over 4G, also allocates 503 * 256M extra low memory for DMA buffers and swiotlb. 504 * But the extra memory is not required for all machines. 505 * So try low memory first and fall back to high memory 506 * unless "crashkernel=size[KMG],high" is specified. 507 */ 508 if (!high) 509 crash_base = memblock_find_in_range(CRASH_ALIGN, 510 CRASH_ADDR_LOW_MAX, 511 crash_size, CRASH_ALIGN); 512 if (!crash_base) 513 crash_base = memblock_find_in_range(CRASH_ALIGN, 514 CRASH_ADDR_HIGH_MAX, 515 crash_size, CRASH_ALIGN); 516 if (!crash_base) { 517 pr_info("crashkernel reservation failed - No suitable area found.\n"); 518 return; 519 } 520 } else { 521 unsigned long long start; 522 523 start = memblock_find_in_range(crash_base, 524 crash_base + crash_size, 525 crash_size, 1 << 20); 526 if (start != crash_base) { 527 pr_info("crashkernel reservation failed - memory is in use.\n"); 528 return; 529 } 530 } 531 ret = memblock_reserve(crash_base, crash_size); 532 if (ret) { 533 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 534 return; 535 } 536 537 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 538 memblock_free(crash_base, crash_size); 539 return; 540 } 541 542 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 543 (unsigned long)(crash_size >> 20), 544 (unsigned long)(crash_base >> 20), 545 (unsigned long)(total_mem >> 20)); 546 547 crashk_res.start = crash_base; 548 crashk_res.end = crash_base + crash_size - 1; 549 insert_resource(&iomem_resource, &crashk_res); 550 } 551 #else 552 static void __init reserve_crashkernel(void) 553 { 554 } 555 #endif 556 557 static struct resource standard_io_resources[] = { 558 { .name = "dma1", .start = 0x00, .end = 0x1f, 559 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 560 { .name = "pic1", .start = 0x20, .end = 0x21, 561 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 562 { .name = "timer0", .start = 0x40, .end = 0x43, 563 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 564 { .name = "timer1", .start = 0x50, .end = 0x53, 565 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 566 { .name = "keyboard", .start = 0x60, .end = 0x60, 567 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 568 { .name = "keyboard", .start = 0x64, .end = 0x64, 569 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 570 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 571 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 572 { .name = "pic2", .start = 0xa0, .end = 0xa1, 573 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 574 { .name = "dma2", .start = 0xc0, .end = 0xdf, 575 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 576 { .name = "fpu", .start = 0xf0, .end = 0xff, 577 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 578 }; 579 580 void __init reserve_standard_io_resources(void) 581 { 582 int i; 583 584 /* request I/O space for devices used on all i[345]86 PCs */ 585 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 586 request_resource(&ioport_resource, &standard_io_resources[i]); 587 588 } 589 590 static __init void reserve_ibft_region(void) 591 { 592 unsigned long addr, size = 0; 593 594 addr = find_ibft_region(&size); 595 596 if (size) 597 memblock_reserve(addr, size); 598 } 599 600 static bool __init snb_gfx_workaround_needed(void) 601 { 602 #ifdef CONFIG_PCI 603 int i; 604 u16 vendor, devid; 605 static const __initconst u16 snb_ids[] = { 606 0x0102, 607 0x0112, 608 0x0122, 609 0x0106, 610 0x0116, 611 0x0126, 612 0x010a, 613 }; 614 615 /* Assume no if something weird is going on with PCI */ 616 if (!early_pci_allowed()) 617 return false; 618 619 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 620 if (vendor != 0x8086) 621 return false; 622 623 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 624 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 625 if (devid == snb_ids[i]) 626 return true; 627 #endif 628 629 return false; 630 } 631 632 /* 633 * Sandy Bridge graphics has trouble with certain ranges, exclude 634 * them from allocation. 635 */ 636 static void __init trim_snb_memory(void) 637 { 638 static const __initconst unsigned long bad_pages[] = { 639 0x20050000, 640 0x20110000, 641 0x20130000, 642 0x20138000, 643 0x40004000, 644 }; 645 int i; 646 647 if (!snb_gfx_workaround_needed()) 648 return; 649 650 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 651 652 /* 653 * Reserve all memory below the 1 MB mark that has not 654 * already been reserved. 655 */ 656 memblock_reserve(0, 1<<20); 657 658 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 659 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 660 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 661 bad_pages[i]); 662 } 663 } 664 665 /* 666 * Here we put platform-specific memory range workarounds, i.e. 667 * memory known to be corrupt or otherwise in need to be reserved on 668 * specific platforms. 669 * 670 * If this gets used more widely it could use a real dispatch mechanism. 671 */ 672 static void __init trim_platform_memory_ranges(void) 673 { 674 trim_snb_memory(); 675 } 676 677 static void __init trim_bios_range(void) 678 { 679 /* 680 * A special case is the first 4Kb of memory; 681 * This is a BIOS owned area, not kernel ram, but generally 682 * not listed as such in the E820 table. 683 * 684 * This typically reserves additional memory (64KiB by default) 685 * since some BIOSes are known to corrupt low memory. See the 686 * Kconfig help text for X86_RESERVE_LOW. 687 */ 688 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 689 690 /* 691 * special case: Some BIOSes report the PC BIOS 692 * area (640Kb -> 1Mb) as RAM even though it is not. 693 * take them out. 694 */ 695 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 696 697 e820__update_table(e820_table); 698 } 699 700 /* called before trim_bios_range() to spare extra sanitize */ 701 static void __init e820_add_kernel_range(void) 702 { 703 u64 start = __pa_symbol(_text); 704 u64 size = __pa_symbol(_end) - start; 705 706 /* 707 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 708 * attempt to fix it by adding the range. We may have a confused BIOS, 709 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 710 * exclude kernel range. If we really are running on top non-RAM, 711 * we will crash later anyways. 712 */ 713 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 714 return; 715 716 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 717 e820__range_remove(start, size, E820_TYPE_RAM, 0); 718 e820__range_add(start, size, E820_TYPE_RAM); 719 } 720 721 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 722 723 static int __init parse_reservelow(char *p) 724 { 725 unsigned long long size; 726 727 if (!p) 728 return -EINVAL; 729 730 size = memparse(p, &p); 731 732 if (size < 4096) 733 size = 4096; 734 735 if (size > 640*1024) 736 size = 640*1024; 737 738 reserve_low = size; 739 740 return 0; 741 } 742 743 early_param("reservelow", parse_reservelow); 744 745 static void __init trim_low_memory_range(void) 746 { 747 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 748 } 749 750 /* 751 * Dump out kernel offset information on panic. 752 */ 753 static int 754 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 755 { 756 if (kaslr_enabled()) { 757 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 758 kaslr_offset(), 759 __START_KERNEL, 760 __START_KERNEL_map, 761 MODULES_VADDR-1); 762 } else { 763 pr_emerg("Kernel Offset: disabled\n"); 764 } 765 766 return 0; 767 } 768 769 /* 770 * Determine if we were loaded by an EFI loader. If so, then we have also been 771 * passed the efi memmap, systab, etc., so we should use these data structures 772 * for initialization. Note, the efi init code path is determined by the 773 * global efi_enabled. This allows the same kernel image to be used on existing 774 * systems (with a traditional BIOS) as well as on EFI systems. 775 */ 776 /* 777 * setup_arch - architecture-specific boot-time initializations 778 * 779 * Note: On x86_64, fixmaps are ready for use even before this is called. 780 */ 781 782 void __init setup_arch(char **cmdline_p) 783 { 784 /* 785 * Reserve the memory occupied by the kernel between _text and 786 * __end_of_kernel_reserve symbols. Any kernel sections after the 787 * __end_of_kernel_reserve symbol must be explicitly reserved with a 788 * separate memblock_reserve() or they will be discarded. 789 */ 790 memblock_reserve(__pa_symbol(_text), 791 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 792 793 /* 794 * Make sure page 0 is always reserved because on systems with 795 * L1TF its contents can be leaked to user processes. 796 */ 797 memblock_reserve(0, PAGE_SIZE); 798 799 early_reserve_initrd(); 800 801 /* 802 * At this point everything still needed from the boot loader 803 * or BIOS or kernel text should be early reserved or marked not 804 * RAM in e820. All other memory is free game. 805 */ 806 807 #ifdef CONFIG_X86_32 808 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 809 810 /* 811 * copy kernel address range established so far and switch 812 * to the proper swapper page table 813 */ 814 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 815 initial_page_table + KERNEL_PGD_BOUNDARY, 816 KERNEL_PGD_PTRS); 817 818 load_cr3(swapper_pg_dir); 819 /* 820 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 821 * a cr3 based tlb flush, so the following __flush_tlb_all() 822 * will not flush anything because the CPU quirk which clears 823 * X86_FEATURE_PGE has not been invoked yet. Though due to the 824 * load_cr3() above the TLB has been flushed already. The 825 * quirk is invoked before subsequent calls to __flush_tlb_all() 826 * so proper operation is guaranteed. 827 */ 828 __flush_tlb_all(); 829 #else 830 printk(KERN_INFO "Command line: %s\n", boot_command_line); 831 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 832 #endif 833 834 /* 835 * If we have OLPC OFW, we might end up relocating the fixmap due to 836 * reserve_top(), so do this before touching the ioremap area. 837 */ 838 olpc_ofw_detect(); 839 840 idt_setup_early_traps(); 841 early_cpu_init(); 842 arch_init_ideal_nops(); 843 jump_label_init(); 844 early_ioremap_init(); 845 846 setup_olpc_ofw_pgd(); 847 848 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 849 screen_info = boot_params.screen_info; 850 edid_info = boot_params.edid_info; 851 #ifdef CONFIG_X86_32 852 apm_info.bios = boot_params.apm_bios_info; 853 ist_info = boot_params.ist_info; 854 #endif 855 saved_video_mode = boot_params.hdr.vid_mode; 856 bootloader_type = boot_params.hdr.type_of_loader; 857 if ((bootloader_type >> 4) == 0xe) { 858 bootloader_type &= 0xf; 859 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 860 } 861 bootloader_version = bootloader_type & 0xf; 862 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 863 864 #ifdef CONFIG_BLK_DEV_RAM 865 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 866 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 867 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 868 #endif 869 #ifdef CONFIG_EFI 870 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 871 EFI32_LOADER_SIGNATURE, 4)) { 872 set_bit(EFI_BOOT, &efi.flags); 873 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 874 EFI64_LOADER_SIGNATURE, 4)) { 875 set_bit(EFI_BOOT, &efi.flags); 876 set_bit(EFI_64BIT, &efi.flags); 877 } 878 #endif 879 880 x86_init.oem.arch_setup(); 881 882 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 883 e820__memory_setup(); 884 parse_setup_data(); 885 886 copy_edd(); 887 888 if (!boot_params.hdr.root_flags) 889 root_mountflags &= ~MS_RDONLY; 890 init_mm.start_code = (unsigned long) _text; 891 init_mm.end_code = (unsigned long) _etext; 892 init_mm.end_data = (unsigned long) _edata; 893 init_mm.brk = _brk_end; 894 895 code_resource.start = __pa_symbol(_text); 896 code_resource.end = __pa_symbol(_etext)-1; 897 rodata_resource.start = __pa_symbol(__start_rodata); 898 rodata_resource.end = __pa_symbol(__end_rodata)-1; 899 data_resource.start = __pa_symbol(_sdata); 900 data_resource.end = __pa_symbol(_edata)-1; 901 bss_resource.start = __pa_symbol(__bss_start); 902 bss_resource.end = __pa_symbol(__bss_stop)-1; 903 904 #ifdef CONFIG_CMDLINE_BOOL 905 #ifdef CONFIG_CMDLINE_OVERRIDE 906 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 907 #else 908 if (builtin_cmdline[0]) { 909 /* append boot loader cmdline to builtin */ 910 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 911 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 912 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 913 } 914 #endif 915 #endif 916 917 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 918 *cmdline_p = command_line; 919 920 /* 921 * x86_configure_nx() is called before parse_early_param() to detect 922 * whether hardware doesn't support NX (so that the early EHCI debug 923 * console setup can safely call set_fixmap()). It may then be called 924 * again from within noexec_setup() during parsing early parameters 925 * to honor the respective command line option. 926 */ 927 x86_configure_nx(); 928 929 parse_early_param(); 930 931 if (efi_enabled(EFI_BOOT)) 932 efi_memblock_x86_reserve_range(); 933 #ifdef CONFIG_MEMORY_HOTPLUG 934 /* 935 * Memory used by the kernel cannot be hot-removed because Linux 936 * cannot migrate the kernel pages. When memory hotplug is 937 * enabled, we should prevent memblock from allocating memory 938 * for the kernel. 939 * 940 * ACPI SRAT records all hotpluggable memory ranges. But before 941 * SRAT is parsed, we don't know about it. 942 * 943 * The kernel image is loaded into memory at very early time. We 944 * cannot prevent this anyway. So on NUMA system, we set any 945 * node the kernel resides in as un-hotpluggable. 946 * 947 * Since on modern servers, one node could have double-digit 948 * gigabytes memory, we can assume the memory around the kernel 949 * image is also un-hotpluggable. So before SRAT is parsed, just 950 * allocate memory near the kernel image to try the best to keep 951 * the kernel away from hotpluggable memory. 952 */ 953 if (movable_node_is_enabled()) 954 memblock_set_bottom_up(true); 955 #endif 956 957 x86_report_nx(); 958 959 /* after early param, so could get panic from serial */ 960 memblock_x86_reserve_range_setup_data(); 961 962 if (acpi_mps_check()) { 963 #ifdef CONFIG_X86_LOCAL_APIC 964 disable_apic = 1; 965 #endif 966 setup_clear_cpu_cap(X86_FEATURE_APIC); 967 } 968 969 e820__reserve_setup_data(); 970 e820__finish_early_params(); 971 972 if (efi_enabled(EFI_BOOT)) 973 efi_init(); 974 975 dmi_setup(); 976 977 /* 978 * VMware detection requires dmi to be available, so this 979 * needs to be done after dmi_setup(), for the boot CPU. 980 */ 981 init_hypervisor_platform(); 982 983 tsc_early_init(); 984 x86_init.resources.probe_roms(); 985 986 /* after parse_early_param, so could debug it */ 987 insert_resource(&iomem_resource, &code_resource); 988 insert_resource(&iomem_resource, &rodata_resource); 989 insert_resource(&iomem_resource, &data_resource); 990 insert_resource(&iomem_resource, &bss_resource); 991 992 e820_add_kernel_range(); 993 trim_bios_range(); 994 #ifdef CONFIG_X86_32 995 if (ppro_with_ram_bug()) { 996 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 997 E820_TYPE_RESERVED); 998 e820__update_table(e820_table); 999 printk(KERN_INFO "fixed physical RAM map:\n"); 1000 e820__print_table("bad_ppro"); 1001 } 1002 #else 1003 early_gart_iommu_check(); 1004 #endif 1005 1006 /* 1007 * partially used pages are not usable - thus 1008 * we are rounding upwards: 1009 */ 1010 max_pfn = e820__end_of_ram_pfn(); 1011 1012 /* update e820 for memory not covered by WB MTRRs */ 1013 mtrr_bp_init(); 1014 if (mtrr_trim_uncached_memory(max_pfn)) 1015 max_pfn = e820__end_of_ram_pfn(); 1016 1017 max_possible_pfn = max_pfn; 1018 1019 /* 1020 * This call is required when the CPU does not support PAT. If 1021 * mtrr_bp_init() invoked it already via pat_init() the call has no 1022 * effect. 1023 */ 1024 init_cache_modes(); 1025 1026 /* 1027 * Define random base addresses for memory sections after max_pfn is 1028 * defined and before each memory section base is used. 1029 */ 1030 kernel_randomize_memory(); 1031 1032 #ifdef CONFIG_X86_32 1033 /* max_low_pfn get updated here */ 1034 find_low_pfn_range(); 1035 #else 1036 check_x2apic(); 1037 1038 /* How many end-of-memory variables you have, grandma! */ 1039 /* need this before calling reserve_initrd */ 1040 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1041 max_low_pfn = e820__end_of_low_ram_pfn(); 1042 else 1043 max_low_pfn = max_pfn; 1044 1045 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1046 #endif 1047 1048 /* 1049 * Find and reserve possible boot-time SMP configuration: 1050 */ 1051 find_smp_config(); 1052 1053 reserve_ibft_region(); 1054 1055 early_alloc_pgt_buf(); 1056 1057 /* 1058 * Need to conclude brk, before e820__memblock_setup() 1059 * it could use memblock_find_in_range, could overlap with 1060 * brk area. 1061 */ 1062 reserve_brk(); 1063 1064 cleanup_highmap(); 1065 1066 memblock_set_current_limit(ISA_END_ADDRESS); 1067 e820__memblock_setup(); 1068 1069 reserve_bios_regions(); 1070 1071 efi_fake_memmap(); 1072 efi_find_mirror(); 1073 efi_esrt_init(); 1074 1075 /* 1076 * The EFI specification says that boot service code won't be 1077 * called after ExitBootServices(). This is, in fact, a lie. 1078 */ 1079 efi_reserve_boot_services(); 1080 1081 /* preallocate 4k for mptable mpc */ 1082 e820__memblock_alloc_reserved_mpc_new(); 1083 1084 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1085 setup_bios_corruption_check(); 1086 #endif 1087 1088 #ifdef CONFIG_X86_32 1089 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1090 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1091 #endif 1092 1093 reserve_real_mode(); 1094 1095 trim_platform_memory_ranges(); 1096 trim_low_memory_range(); 1097 1098 init_mem_mapping(); 1099 1100 idt_setup_early_pf(); 1101 1102 /* 1103 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1104 * with the current CR4 value. This may not be necessary, but 1105 * auditing all the early-boot CR4 manipulation would be needed to 1106 * rule it out. 1107 * 1108 * Mask off features that don't work outside long mode (just 1109 * PCIDE for now). 1110 */ 1111 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1112 1113 memblock_set_current_limit(get_max_mapped()); 1114 1115 /* 1116 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1117 */ 1118 1119 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1120 if (init_ohci1394_dma_early) 1121 init_ohci1394_dma_on_all_controllers(); 1122 #endif 1123 /* Allocate bigger log buffer */ 1124 setup_log_buf(1); 1125 1126 if (efi_enabled(EFI_BOOT)) { 1127 switch (boot_params.secure_boot) { 1128 case efi_secureboot_mode_disabled: 1129 pr_info("Secure boot disabled\n"); 1130 break; 1131 case efi_secureboot_mode_enabled: 1132 pr_info("Secure boot enabled\n"); 1133 break; 1134 default: 1135 pr_info("Secure boot could not be determined\n"); 1136 break; 1137 } 1138 } 1139 1140 reserve_initrd(); 1141 1142 acpi_table_upgrade(); 1143 1144 vsmp_init(); 1145 1146 io_delay_init(); 1147 1148 early_platform_quirks(); 1149 1150 /* 1151 * Parse the ACPI tables for possible boot-time SMP configuration. 1152 */ 1153 acpi_boot_table_init(); 1154 1155 early_acpi_boot_init(); 1156 1157 initmem_init(); 1158 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1159 1160 /* 1161 * Reserve memory for crash kernel after SRAT is parsed so that it 1162 * won't consume hotpluggable memory. 1163 */ 1164 reserve_crashkernel(); 1165 1166 memblock_find_dma_reserve(); 1167 1168 if (!early_xdbc_setup_hardware()) 1169 early_xdbc_register_console(); 1170 1171 x86_init.paging.pagetable_init(); 1172 1173 kasan_init(); 1174 1175 /* 1176 * Sync back kernel address range. 1177 * 1178 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1179 * this call? 1180 */ 1181 sync_initial_page_table(); 1182 1183 tboot_probe(); 1184 1185 map_vsyscall(); 1186 1187 generic_apic_probe(); 1188 1189 early_quirks(); 1190 1191 /* 1192 * Read APIC and some other early information from ACPI tables. 1193 */ 1194 acpi_boot_init(); 1195 sfi_init(); 1196 x86_dtb_init(); 1197 1198 /* 1199 * get boot-time SMP configuration: 1200 */ 1201 get_smp_config(); 1202 1203 /* 1204 * Systems w/o ACPI and mptables might not have it mapped the local 1205 * APIC yet, but prefill_possible_map() might need to access it. 1206 */ 1207 init_apic_mappings(); 1208 1209 prefill_possible_map(); 1210 1211 init_cpu_to_node(); 1212 1213 io_apic_init_mappings(); 1214 1215 x86_init.hyper.guest_late_init(); 1216 1217 e820__reserve_resources(); 1218 e820__register_nosave_regions(max_pfn); 1219 1220 x86_init.resources.reserve_resources(); 1221 1222 e820__setup_pci_gap(); 1223 1224 #ifdef CONFIG_VT 1225 #if defined(CONFIG_VGA_CONSOLE) 1226 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1227 conswitchp = &vga_con; 1228 #endif 1229 #endif 1230 x86_init.oem.banner(); 1231 1232 x86_init.timers.wallclock_init(); 1233 1234 mcheck_init(); 1235 1236 register_refined_jiffies(CLOCK_TICK_RATE); 1237 1238 #ifdef CONFIG_EFI 1239 if (efi_enabled(EFI_BOOT)) 1240 efi_apply_memmap_quirks(); 1241 #endif 1242 1243 unwind_init(); 1244 } 1245 1246 #ifdef CONFIG_X86_32 1247 1248 static struct resource video_ram_resource = { 1249 .name = "Video RAM area", 1250 .start = 0xa0000, 1251 .end = 0xbffff, 1252 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1253 }; 1254 1255 void __init i386_reserve_resources(void) 1256 { 1257 request_resource(&iomem_resource, &video_ram_resource); 1258 reserve_standard_io_resources(); 1259 } 1260 1261 #endif /* CONFIG_X86_32 */ 1262 1263 static struct notifier_block kernel_offset_notifier = { 1264 .notifier_call = dump_kernel_offset 1265 }; 1266 1267 static int __init register_kernel_offset_dumper(void) 1268 { 1269 atomic_notifier_chain_register(&panic_notifier_list, 1270 &kernel_offset_notifier); 1271 return 0; 1272 } 1273 __initcall(register_kernel_offset_dumper); 1274