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/acpi.h> 9 #include <linux/console.h> 10 #include <linux/cpu.h> 11 #include <linux/crash_dump.h> 12 #include <linux/dma-map-ops.h> 13 #include <linux/efi.h> 14 #include <linux/ima.h> 15 #include <linux/init_ohci1394_dma.h> 16 #include <linux/initrd.h> 17 #include <linux/iscsi_ibft.h> 18 #include <linux/memblock.h> 19 #include <linux/panic_notifier.h> 20 #include <linux/pci.h> 21 #include <linux/root_dev.h> 22 #include <linux/hugetlb.h> 23 #include <linux/tboot.h> 24 #include <linux/usb/xhci-dbgp.h> 25 #include <linux/static_call.h> 26 #include <linux/swiotlb.h> 27 #include <linux/random.h> 28 29 #include <uapi/linux/mount.h> 30 31 #include <xen/xen.h> 32 33 #include <asm/apic.h> 34 #include <asm/efi.h> 35 #include <asm/numa.h> 36 #include <asm/bios_ebda.h> 37 #include <asm/bugs.h> 38 #include <asm/cacheinfo.h> 39 #include <asm/coco.h> 40 #include <asm/cpu.h> 41 #include <asm/efi.h> 42 #include <asm/gart.h> 43 #include <asm/hypervisor.h> 44 #include <asm/io_apic.h> 45 #include <asm/kasan.h> 46 #include <asm/kaslr.h> 47 #include <asm/mce.h> 48 #include <asm/memtype.h> 49 #include <asm/mtrr.h> 50 #include <asm/realmode.h> 51 #include <asm/olpc_ofw.h> 52 #include <asm/pci-direct.h> 53 #include <asm/prom.h> 54 #include <asm/proto.h> 55 #include <asm/thermal.h> 56 #include <asm/unwind.h> 57 #include <asm/vsyscall.h> 58 #include <linux/vmalloc.h> 59 60 /* 61 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 62 * max_pfn_mapped: highest directly mapped pfn > 4 GB 63 * 64 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 65 * represented by pfn_mapped[]. 66 */ 67 unsigned long max_low_pfn_mapped; 68 unsigned long max_pfn_mapped; 69 70 #ifdef CONFIG_DMI 71 RESERVE_BRK(dmi_alloc, 65536); 72 #endif 73 74 75 unsigned long _brk_start = (unsigned long)__brk_base; 76 unsigned long _brk_end = (unsigned long)__brk_base; 77 78 struct boot_params boot_params; 79 80 /* 81 * These are the four main kernel memory regions, we put them into 82 * the resource tree so that kdump tools and other debugging tools 83 * recover it: 84 */ 85 86 static struct resource rodata_resource = { 87 .name = "Kernel rodata", 88 .start = 0, 89 .end = 0, 90 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 91 }; 92 93 static struct resource data_resource = { 94 .name = "Kernel data", 95 .start = 0, 96 .end = 0, 97 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 98 }; 99 100 static struct resource code_resource = { 101 .name = "Kernel code", 102 .start = 0, 103 .end = 0, 104 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 105 }; 106 107 static struct resource bss_resource = { 108 .name = "Kernel bss", 109 .start = 0, 110 .end = 0, 111 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 112 }; 113 114 115 #ifdef CONFIG_X86_32 116 /* CPU data as detected by the assembly code in head_32.S */ 117 struct cpuinfo_x86 new_cpu_data; 118 119 struct apm_info apm_info; 120 EXPORT_SYMBOL(apm_info); 121 122 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 123 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 124 struct ist_info ist_info; 125 EXPORT_SYMBOL(ist_info); 126 #else 127 struct ist_info ist_info; 128 #endif 129 130 #endif 131 132 struct cpuinfo_x86 boot_cpu_data __read_mostly; 133 EXPORT_SYMBOL(boot_cpu_data); 134 135 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 136 __visible unsigned long mmu_cr4_features __ro_after_init; 137 #else 138 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 139 #endif 140 141 #ifdef CONFIG_IMA 142 static phys_addr_t ima_kexec_buffer_phys; 143 static size_t ima_kexec_buffer_size; 144 #endif 145 146 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 147 int bootloader_type, bootloader_version; 148 149 /* 150 * Setup options 151 */ 152 struct screen_info screen_info; 153 EXPORT_SYMBOL(screen_info); 154 struct edid_info edid_info; 155 EXPORT_SYMBOL_GPL(edid_info); 156 157 extern int root_mountflags; 158 159 unsigned long saved_video_mode; 160 161 #define RAMDISK_IMAGE_START_MASK 0x07FF 162 #define RAMDISK_PROMPT_FLAG 0x8000 163 #define RAMDISK_LOAD_FLAG 0x4000 164 165 static char __initdata command_line[COMMAND_LINE_SIZE]; 166 #ifdef CONFIG_CMDLINE_BOOL 167 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 168 #endif 169 170 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 171 struct edd edd; 172 #ifdef CONFIG_EDD_MODULE 173 EXPORT_SYMBOL(edd); 174 #endif 175 /** 176 * copy_edd() - Copy the BIOS EDD information 177 * from boot_params into a safe place. 178 * 179 */ 180 static inline void __init copy_edd(void) 181 { 182 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 183 sizeof(edd.mbr_signature)); 184 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 185 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 186 edd.edd_info_nr = boot_params.eddbuf_entries; 187 } 188 #else 189 static inline void __init copy_edd(void) 190 { 191 } 192 #endif 193 194 void * __init extend_brk(size_t size, size_t align) 195 { 196 size_t mask = align - 1; 197 void *ret; 198 199 BUG_ON(_brk_start == 0); 200 BUG_ON(align & mask); 201 202 _brk_end = (_brk_end + mask) & ~mask; 203 BUG_ON((char *)(_brk_end + size) > __brk_limit); 204 205 ret = (void *)_brk_end; 206 _brk_end += size; 207 208 memset(ret, 0, size); 209 210 return ret; 211 } 212 213 #ifdef CONFIG_X86_32 214 static void __init cleanup_highmap(void) 215 { 216 } 217 #endif 218 219 static void __init reserve_brk(void) 220 { 221 if (_brk_end > _brk_start) 222 memblock_reserve(__pa_symbol(_brk_start), 223 _brk_end - _brk_start); 224 225 /* Mark brk area as locked down and no longer taking any 226 new allocations */ 227 _brk_start = 0; 228 } 229 230 #ifdef CONFIG_BLK_DEV_INITRD 231 232 static u64 __init get_ramdisk_image(void) 233 { 234 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 235 236 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 237 238 if (ramdisk_image == 0) 239 ramdisk_image = phys_initrd_start; 240 241 return ramdisk_image; 242 } 243 static u64 __init get_ramdisk_size(void) 244 { 245 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 246 247 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 248 249 if (ramdisk_size == 0) 250 ramdisk_size = phys_initrd_size; 251 252 return ramdisk_size; 253 } 254 255 static void __init relocate_initrd(void) 256 { 257 /* Assume only end is not page aligned */ 258 u64 ramdisk_image = get_ramdisk_image(); 259 u64 ramdisk_size = get_ramdisk_size(); 260 u64 area_size = PAGE_ALIGN(ramdisk_size); 261 262 /* We need to move the initrd down into directly mapped mem */ 263 u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 264 PFN_PHYS(max_pfn_mapped)); 265 if (!relocated_ramdisk) 266 panic("Cannot find place for new RAMDISK of size %lld\n", 267 ramdisk_size); 268 269 initrd_start = relocated_ramdisk + PAGE_OFFSET; 270 initrd_end = initrd_start + ramdisk_size; 271 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 272 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 273 274 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 275 276 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 277 " [mem %#010llx-%#010llx]\n", 278 ramdisk_image, ramdisk_image + ramdisk_size - 1, 279 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 280 } 281 282 static void __init early_reserve_initrd(void) 283 { 284 /* Assume only end is not page aligned */ 285 u64 ramdisk_image = get_ramdisk_image(); 286 u64 ramdisk_size = get_ramdisk_size(); 287 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 288 289 if (!boot_params.hdr.type_of_loader || 290 !ramdisk_image || !ramdisk_size) 291 return; /* No initrd provided by bootloader */ 292 293 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 294 } 295 296 static void __init reserve_initrd(void) 297 { 298 /* Assume only end is not page aligned */ 299 u64 ramdisk_image = get_ramdisk_image(); 300 u64 ramdisk_size = get_ramdisk_size(); 301 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 302 303 if (!boot_params.hdr.type_of_loader || 304 !ramdisk_image || !ramdisk_size) 305 return; /* No initrd provided by bootloader */ 306 307 initrd_start = 0; 308 309 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 310 ramdisk_end - 1); 311 312 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 313 PFN_DOWN(ramdisk_end))) { 314 /* All are mapped, easy case */ 315 initrd_start = ramdisk_image + PAGE_OFFSET; 316 initrd_end = initrd_start + ramdisk_size; 317 return; 318 } 319 320 relocate_initrd(); 321 322 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image); 323 } 324 325 #else 326 static void __init early_reserve_initrd(void) 327 { 328 } 329 static void __init reserve_initrd(void) 330 { 331 } 332 #endif /* CONFIG_BLK_DEV_INITRD */ 333 334 static void __init add_early_ima_buffer(u64 phys_addr) 335 { 336 #ifdef CONFIG_IMA 337 struct ima_setup_data *data; 338 339 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data)); 340 if (!data) { 341 pr_warn("setup: failed to memremap ima_setup_data entry\n"); 342 return; 343 } 344 345 if (data->size) { 346 memblock_reserve(data->addr, data->size); 347 ima_kexec_buffer_phys = data->addr; 348 ima_kexec_buffer_size = data->size; 349 } 350 351 early_memunmap(data, sizeof(*data)); 352 #else 353 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n"); 354 #endif 355 } 356 357 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE) 358 int __init ima_free_kexec_buffer(void) 359 { 360 if (!ima_kexec_buffer_size) 361 return -ENOENT; 362 363 memblock_free_late(ima_kexec_buffer_phys, 364 ima_kexec_buffer_size); 365 366 ima_kexec_buffer_phys = 0; 367 ima_kexec_buffer_size = 0; 368 369 return 0; 370 } 371 372 int __init ima_get_kexec_buffer(void **addr, size_t *size) 373 { 374 if (!ima_kexec_buffer_size) 375 return -ENOENT; 376 377 *addr = __va(ima_kexec_buffer_phys); 378 *size = ima_kexec_buffer_size; 379 380 return 0; 381 } 382 #endif 383 384 static void __init parse_setup_data(void) 385 { 386 struct setup_data *data; 387 u64 pa_data, pa_next; 388 389 pa_data = boot_params.hdr.setup_data; 390 while (pa_data) { 391 u32 data_len, data_type; 392 393 data = early_memremap(pa_data, sizeof(*data)); 394 data_len = data->len + sizeof(struct setup_data); 395 data_type = data->type; 396 pa_next = data->next; 397 early_memunmap(data, sizeof(*data)); 398 399 switch (data_type) { 400 case SETUP_E820_EXT: 401 e820__memory_setup_extended(pa_data, data_len); 402 break; 403 case SETUP_DTB: 404 add_dtb(pa_data); 405 break; 406 case SETUP_EFI: 407 parse_efi_setup(pa_data, data_len); 408 break; 409 case SETUP_IMA: 410 add_early_ima_buffer(pa_data); 411 break; 412 case SETUP_RNG_SEED: 413 data = early_memremap(pa_data, data_len); 414 add_bootloader_randomness(data->data, data->len); 415 /* Zero seed for forward secrecy. */ 416 memzero_explicit(data->data, data->len); 417 /* Zero length in case we find ourselves back here by accident. */ 418 memzero_explicit(&data->len, sizeof(data->len)); 419 early_memunmap(data, data_len); 420 break; 421 default: 422 break; 423 } 424 pa_data = pa_next; 425 } 426 } 427 428 static void __init memblock_x86_reserve_range_setup_data(void) 429 { 430 struct setup_indirect *indirect; 431 struct setup_data *data; 432 u64 pa_data, pa_next; 433 u32 len; 434 435 pa_data = boot_params.hdr.setup_data; 436 while (pa_data) { 437 data = early_memremap(pa_data, sizeof(*data)); 438 if (!data) { 439 pr_warn("setup: failed to memremap setup_data entry\n"); 440 return; 441 } 442 443 len = sizeof(*data); 444 pa_next = data->next; 445 446 memblock_reserve(pa_data, sizeof(*data) + data->len); 447 448 if (data->type == SETUP_INDIRECT) { 449 len += data->len; 450 early_memunmap(data, sizeof(*data)); 451 data = early_memremap(pa_data, len); 452 if (!data) { 453 pr_warn("setup: failed to memremap indirect setup_data\n"); 454 return; 455 } 456 457 indirect = (struct setup_indirect *)data->data; 458 459 if (indirect->type != SETUP_INDIRECT) 460 memblock_reserve(indirect->addr, indirect->len); 461 } 462 463 pa_data = pa_next; 464 early_memunmap(data, len); 465 } 466 } 467 468 static void __init arch_reserve_crashkernel(void) 469 { 470 unsigned long long crash_base, crash_size, low_size = 0; 471 char *cmdline = boot_command_line; 472 bool high = false; 473 int ret; 474 475 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 476 return; 477 478 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 479 &crash_size, &crash_base, 480 &low_size, &high); 481 if (ret) 482 return; 483 484 if (xen_pv_domain()) { 485 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 486 return; 487 } 488 489 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 490 low_size, high); 491 } 492 493 static struct resource standard_io_resources[] = { 494 { .name = "dma1", .start = 0x00, .end = 0x1f, 495 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 496 { .name = "pic1", .start = 0x20, .end = 0x21, 497 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 498 { .name = "timer0", .start = 0x40, .end = 0x43, 499 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 500 { .name = "timer1", .start = 0x50, .end = 0x53, 501 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 502 { .name = "keyboard", .start = 0x60, .end = 0x60, 503 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 504 { .name = "keyboard", .start = 0x64, .end = 0x64, 505 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 506 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 507 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 508 { .name = "pic2", .start = 0xa0, .end = 0xa1, 509 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 510 { .name = "dma2", .start = 0xc0, .end = 0xdf, 511 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 512 { .name = "fpu", .start = 0xf0, .end = 0xff, 513 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 514 }; 515 516 void __init reserve_standard_io_resources(void) 517 { 518 int i; 519 520 /* request I/O space for devices used on all i[345]86 PCs */ 521 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 522 request_resource(&ioport_resource, &standard_io_resources[i]); 523 524 } 525 526 static bool __init snb_gfx_workaround_needed(void) 527 { 528 #ifdef CONFIG_PCI 529 int i; 530 u16 vendor, devid; 531 static const __initconst u16 snb_ids[] = { 532 0x0102, 533 0x0112, 534 0x0122, 535 0x0106, 536 0x0116, 537 0x0126, 538 0x010a, 539 }; 540 541 /* Assume no if something weird is going on with PCI */ 542 if (!early_pci_allowed()) 543 return false; 544 545 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 546 if (vendor != 0x8086) 547 return false; 548 549 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 550 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 551 if (devid == snb_ids[i]) 552 return true; 553 #endif 554 555 return false; 556 } 557 558 /* 559 * Sandy Bridge graphics has trouble with certain ranges, exclude 560 * them from allocation. 561 */ 562 static void __init trim_snb_memory(void) 563 { 564 static const __initconst unsigned long bad_pages[] = { 565 0x20050000, 566 0x20110000, 567 0x20130000, 568 0x20138000, 569 0x40004000, 570 }; 571 int i; 572 573 if (!snb_gfx_workaround_needed()) 574 return; 575 576 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 577 578 /* 579 * SandyBridge integrated graphics devices have a bug that prevents 580 * them from accessing certain memory ranges, namely anything below 581 * 1M and in the pages listed in bad_pages[] above. 582 * 583 * To avoid these pages being ever accessed by SNB gfx devices reserve 584 * bad_pages that have not already been reserved at boot time. 585 * All memory below the 1 MB mark is anyway reserved later during 586 * setup_arch(), so there is no need to reserve it here. 587 */ 588 589 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 590 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 591 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 592 bad_pages[i]); 593 } 594 } 595 596 static void __init trim_bios_range(void) 597 { 598 /* 599 * A special case is the first 4Kb of memory; 600 * This is a BIOS owned area, not kernel ram, but generally 601 * not listed as such in the E820 table. 602 * 603 * This typically reserves additional memory (64KiB by default) 604 * since some BIOSes are known to corrupt low memory. See the 605 * Kconfig help text for X86_RESERVE_LOW. 606 */ 607 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 608 609 /* 610 * special case: Some BIOSes report the PC BIOS 611 * area (640Kb -> 1Mb) as RAM even though it is not. 612 * take them out. 613 */ 614 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 615 616 e820__update_table(e820_table); 617 } 618 619 /* called before trim_bios_range() to spare extra sanitize */ 620 static void __init e820_add_kernel_range(void) 621 { 622 u64 start = __pa_symbol(_text); 623 u64 size = __pa_symbol(_end) - start; 624 625 /* 626 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 627 * attempt to fix it by adding the range. We may have a confused BIOS, 628 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 629 * exclude kernel range. If we really are running on top non-RAM, 630 * we will crash later anyways. 631 */ 632 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 633 return; 634 635 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 636 e820__range_remove(start, size, E820_TYPE_RAM, 0); 637 e820__range_add(start, size, E820_TYPE_RAM); 638 } 639 640 static void __init early_reserve_memory(void) 641 { 642 /* 643 * Reserve the memory occupied by the kernel between _text and 644 * __end_of_kernel_reserve symbols. Any kernel sections after the 645 * __end_of_kernel_reserve symbol must be explicitly reserved with a 646 * separate memblock_reserve() or they will be discarded. 647 */ 648 memblock_reserve(__pa_symbol(_text), 649 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 650 651 /* 652 * The first 4Kb of memory is a BIOS owned area, but generally it is 653 * not listed as such in the E820 table. 654 * 655 * Reserve the first 64K of memory since some BIOSes are known to 656 * corrupt low memory. After the real mode trampoline is allocated the 657 * rest of the memory below 640k is reserved. 658 * 659 * In addition, make sure page 0 is always reserved because on 660 * systems with L1TF its contents can be leaked to user processes. 661 */ 662 memblock_reserve(0, SZ_64K); 663 664 early_reserve_initrd(); 665 666 memblock_x86_reserve_range_setup_data(); 667 668 reserve_bios_regions(); 669 trim_snb_memory(); 670 } 671 672 /* 673 * Dump out kernel offset information on panic. 674 */ 675 static int 676 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 677 { 678 if (kaslr_enabled()) { 679 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 680 kaslr_offset(), 681 __START_KERNEL, 682 __START_KERNEL_map, 683 MODULES_VADDR-1); 684 } else { 685 pr_emerg("Kernel Offset: disabled\n"); 686 } 687 688 return 0; 689 } 690 691 void x86_configure_nx(void) 692 { 693 if (boot_cpu_has(X86_FEATURE_NX)) 694 __supported_pte_mask |= _PAGE_NX; 695 else 696 __supported_pte_mask &= ~_PAGE_NX; 697 } 698 699 static void __init x86_report_nx(void) 700 { 701 if (!boot_cpu_has(X86_FEATURE_NX)) { 702 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 703 "missing in CPU!\n"); 704 } else { 705 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 706 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 707 #else 708 /* 32bit non-PAE kernel, NX cannot be used */ 709 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 710 "cannot be enabled: non-PAE kernel!\n"); 711 #endif 712 } 713 } 714 715 /* 716 * Determine if we were loaded by an EFI loader. If so, then we have also been 717 * passed the efi memmap, systab, etc., so we should use these data structures 718 * for initialization. Note, the efi init code path is determined by the 719 * global efi_enabled. This allows the same kernel image to be used on existing 720 * systems (with a traditional BIOS) as well as on EFI systems. 721 */ 722 /* 723 * setup_arch - architecture-specific boot-time initializations 724 * 725 * Note: On x86_64, fixmaps are ready for use even before this is called. 726 */ 727 728 void __init setup_arch(char **cmdline_p) 729 { 730 #ifdef CONFIG_X86_32 731 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 732 733 /* 734 * copy kernel address range established so far and switch 735 * to the proper swapper page table 736 */ 737 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 738 initial_page_table + KERNEL_PGD_BOUNDARY, 739 KERNEL_PGD_PTRS); 740 741 load_cr3(swapper_pg_dir); 742 /* 743 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 744 * a cr3 based tlb flush, so the following __flush_tlb_all() 745 * will not flush anything because the CPU quirk which clears 746 * X86_FEATURE_PGE has not been invoked yet. Though due to the 747 * load_cr3() above the TLB has been flushed already. The 748 * quirk is invoked before subsequent calls to __flush_tlb_all() 749 * so proper operation is guaranteed. 750 */ 751 __flush_tlb_all(); 752 #else 753 printk(KERN_INFO "Command line: %s\n", boot_command_line); 754 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 755 #endif 756 757 #ifdef CONFIG_CMDLINE_BOOL 758 #ifdef CONFIG_CMDLINE_OVERRIDE 759 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 760 #else 761 if (builtin_cmdline[0]) { 762 /* append boot loader cmdline to builtin */ 763 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 764 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 765 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 766 } 767 #endif 768 #endif 769 770 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 771 *cmdline_p = command_line; 772 773 /* 774 * If we have OLPC OFW, we might end up relocating the fixmap due to 775 * reserve_top(), so do this before touching the ioremap area. 776 */ 777 olpc_ofw_detect(); 778 779 idt_setup_early_traps(); 780 early_cpu_init(); 781 jump_label_init(); 782 static_call_init(); 783 early_ioremap_init(); 784 785 setup_olpc_ofw_pgd(); 786 787 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 788 screen_info = boot_params.screen_info; 789 edid_info = boot_params.edid_info; 790 #ifdef CONFIG_X86_32 791 apm_info.bios = boot_params.apm_bios_info; 792 ist_info = boot_params.ist_info; 793 #endif 794 saved_video_mode = boot_params.hdr.vid_mode; 795 bootloader_type = boot_params.hdr.type_of_loader; 796 if ((bootloader_type >> 4) == 0xe) { 797 bootloader_type &= 0xf; 798 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 799 } 800 bootloader_version = bootloader_type & 0xf; 801 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 802 803 #ifdef CONFIG_BLK_DEV_RAM 804 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 805 #endif 806 #ifdef CONFIG_EFI 807 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 808 EFI32_LOADER_SIGNATURE, 4)) { 809 set_bit(EFI_BOOT, &efi.flags); 810 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 811 EFI64_LOADER_SIGNATURE, 4)) { 812 set_bit(EFI_BOOT, &efi.flags); 813 set_bit(EFI_64BIT, &efi.flags); 814 } 815 #endif 816 817 x86_init.oem.arch_setup(); 818 819 /* 820 * Do some memory reservations *before* memory is added to memblock, so 821 * memblock allocations won't overwrite it. 822 * 823 * After this point, everything still needed from the boot loader or 824 * firmware or kernel text should be early reserved or marked not RAM in 825 * e820. All other memory is free game. 826 * 827 * This call needs to happen before e820__memory_setup() which calls the 828 * xen_memory_setup() on Xen dom0 which relies on the fact that those 829 * early reservations have happened already. 830 */ 831 early_reserve_memory(); 832 833 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 834 e820__memory_setup(); 835 parse_setup_data(); 836 837 copy_edd(); 838 839 if (!boot_params.hdr.root_flags) 840 root_mountflags &= ~MS_RDONLY; 841 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 842 843 code_resource.start = __pa_symbol(_text); 844 code_resource.end = __pa_symbol(_etext)-1; 845 rodata_resource.start = __pa_symbol(__start_rodata); 846 rodata_resource.end = __pa_symbol(__end_rodata)-1; 847 data_resource.start = __pa_symbol(_sdata); 848 data_resource.end = __pa_symbol(_edata)-1; 849 bss_resource.start = __pa_symbol(__bss_start); 850 bss_resource.end = __pa_symbol(__bss_stop)-1; 851 852 /* 853 * x86_configure_nx() is called before parse_early_param() to detect 854 * whether hardware doesn't support NX (so that the early EHCI debug 855 * console setup can safely call set_fixmap()). 856 */ 857 x86_configure_nx(); 858 859 parse_early_param(); 860 861 if (efi_enabled(EFI_BOOT)) 862 efi_memblock_x86_reserve_range(); 863 864 #ifdef CONFIG_MEMORY_HOTPLUG 865 /* 866 * Memory used by the kernel cannot be hot-removed because Linux 867 * cannot migrate the kernel pages. When memory hotplug is 868 * enabled, we should prevent memblock from allocating memory 869 * for the kernel. 870 * 871 * ACPI SRAT records all hotpluggable memory ranges. But before 872 * SRAT is parsed, we don't know about it. 873 * 874 * The kernel image is loaded into memory at very early time. We 875 * cannot prevent this anyway. So on NUMA system, we set any 876 * node the kernel resides in as un-hotpluggable. 877 * 878 * Since on modern servers, one node could have double-digit 879 * gigabytes memory, we can assume the memory around the kernel 880 * image is also un-hotpluggable. So before SRAT is parsed, just 881 * allocate memory near the kernel image to try the best to keep 882 * the kernel away from hotpluggable memory. 883 */ 884 if (movable_node_is_enabled()) 885 memblock_set_bottom_up(true); 886 #endif 887 888 x86_report_nx(); 889 890 apic_setup_apic_calls(); 891 892 if (acpi_mps_check()) { 893 #ifdef CONFIG_X86_LOCAL_APIC 894 apic_is_disabled = true; 895 #endif 896 setup_clear_cpu_cap(X86_FEATURE_APIC); 897 } 898 899 e820__reserve_setup_data(); 900 e820__finish_early_params(); 901 902 if (efi_enabled(EFI_BOOT)) 903 efi_init(); 904 905 reserve_ibft_region(); 906 x86_init.resources.dmi_setup(); 907 908 /* 909 * VMware detection requires dmi to be available, so this 910 * needs to be done after dmi_setup(), for the boot CPU. 911 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be 912 * called before cache_bp_init() for setting up MTRR state. 913 */ 914 init_hypervisor_platform(); 915 916 tsc_early_init(); 917 x86_init.resources.probe_roms(); 918 919 /* after parse_early_param, so could debug it */ 920 insert_resource(&iomem_resource, &code_resource); 921 insert_resource(&iomem_resource, &rodata_resource); 922 insert_resource(&iomem_resource, &data_resource); 923 insert_resource(&iomem_resource, &bss_resource); 924 925 e820_add_kernel_range(); 926 trim_bios_range(); 927 #ifdef CONFIG_X86_32 928 if (ppro_with_ram_bug()) { 929 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 930 E820_TYPE_RESERVED); 931 e820__update_table(e820_table); 932 printk(KERN_INFO "fixed physical RAM map:\n"); 933 e820__print_table("bad_ppro"); 934 } 935 #else 936 early_gart_iommu_check(); 937 #endif 938 939 /* 940 * partially used pages are not usable - thus 941 * we are rounding upwards: 942 */ 943 max_pfn = e820__end_of_ram_pfn(); 944 945 /* update e820 for memory not covered by WB MTRRs */ 946 cache_bp_init(); 947 if (mtrr_trim_uncached_memory(max_pfn)) 948 max_pfn = e820__end_of_ram_pfn(); 949 950 max_possible_pfn = max_pfn; 951 952 /* 953 * Define random base addresses for memory sections after max_pfn is 954 * defined and before each memory section base is used. 955 */ 956 kernel_randomize_memory(); 957 958 #ifdef CONFIG_X86_32 959 /* max_low_pfn get updated here */ 960 find_low_pfn_range(); 961 #else 962 check_x2apic(); 963 964 /* How many end-of-memory variables you have, grandma! */ 965 /* need this before calling reserve_initrd */ 966 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 967 max_low_pfn = e820__end_of_low_ram_pfn(); 968 else 969 max_low_pfn = max_pfn; 970 971 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 972 #endif 973 974 /* Find and reserve MPTABLE area */ 975 x86_init.mpparse.find_mptable(); 976 977 early_alloc_pgt_buf(); 978 979 /* 980 * Need to conclude brk, before e820__memblock_setup() 981 * it could use memblock_find_in_range, could overlap with 982 * brk area. 983 */ 984 reserve_brk(); 985 986 cleanup_highmap(); 987 988 memblock_set_current_limit(ISA_END_ADDRESS); 989 e820__memblock_setup(); 990 991 /* 992 * Needs to run after memblock setup because it needs the physical 993 * memory size. 994 */ 995 mem_encrypt_setup_arch(); 996 cc_random_init(); 997 998 efi_fake_memmap(); 999 efi_find_mirror(); 1000 efi_esrt_init(); 1001 efi_mokvar_table_init(); 1002 1003 /* 1004 * The EFI specification says that boot service code won't be 1005 * called after ExitBootServices(). This is, in fact, a lie. 1006 */ 1007 efi_reserve_boot_services(); 1008 1009 /* preallocate 4k for mptable mpc */ 1010 e820__memblock_alloc_reserved_mpc_new(); 1011 1012 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1013 setup_bios_corruption_check(); 1014 #endif 1015 1016 #ifdef CONFIG_X86_32 1017 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1018 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1019 #endif 1020 1021 /* 1022 * Find free memory for the real mode trampoline and place it there. If 1023 * there is not enough free memory under 1M, on EFI-enabled systems 1024 * there will be additional attempt to reclaim the memory for the real 1025 * mode trampoline at efi_free_boot_services(). 1026 * 1027 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1028 * are known to corrupt low memory and several hundred kilobytes are not 1029 * worth complex detection what memory gets clobbered. Windows does the 1030 * same thing for very similar reasons. 1031 * 1032 * Moreover, on machines with SandyBridge graphics or in setups that use 1033 * crashkernel the entire 1M is reserved anyway. 1034 * 1035 * Note the host kernel TDX also requires the first 1MB being reserved. 1036 */ 1037 x86_platform.realmode_reserve(); 1038 1039 init_mem_mapping(); 1040 1041 idt_setup_early_pf(); 1042 1043 /* 1044 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1045 * with the current CR4 value. This may not be necessary, but 1046 * auditing all the early-boot CR4 manipulation would be needed to 1047 * rule it out. 1048 * 1049 * Mask off features that don't work outside long mode (just 1050 * PCIDE for now). 1051 */ 1052 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1053 1054 memblock_set_current_limit(get_max_mapped()); 1055 1056 /* 1057 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1058 */ 1059 1060 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1061 if (init_ohci1394_dma_early) 1062 init_ohci1394_dma_on_all_controllers(); 1063 #endif 1064 /* Allocate bigger log buffer */ 1065 setup_log_buf(1); 1066 1067 if (efi_enabled(EFI_BOOT)) { 1068 switch (boot_params.secure_boot) { 1069 case efi_secureboot_mode_disabled: 1070 pr_info("Secure boot disabled\n"); 1071 break; 1072 case efi_secureboot_mode_enabled: 1073 pr_info("Secure boot enabled\n"); 1074 break; 1075 default: 1076 pr_info("Secure boot could not be determined\n"); 1077 break; 1078 } 1079 } 1080 1081 reserve_initrd(); 1082 1083 acpi_table_upgrade(); 1084 /* Look for ACPI tables and reserve memory occupied by them. */ 1085 acpi_boot_table_init(); 1086 1087 vsmp_init(); 1088 1089 io_delay_init(); 1090 1091 early_platform_quirks(); 1092 1093 /* Some platforms need the APIC registered for NUMA configuration */ 1094 early_acpi_boot_init(); 1095 x86_init.mpparse.early_parse_smp_cfg(); 1096 1097 x86_flattree_get_config(); 1098 1099 initmem_init(); 1100 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1101 1102 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1103 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1104 1105 /* 1106 * Reserve memory for crash kernel after SRAT is parsed so that it 1107 * won't consume hotpluggable memory. 1108 */ 1109 arch_reserve_crashkernel(); 1110 1111 if (!early_xdbc_setup_hardware()) 1112 early_xdbc_register_console(); 1113 1114 x86_init.paging.pagetable_init(); 1115 1116 kasan_init(); 1117 1118 /* 1119 * Sync back kernel address range. 1120 * 1121 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1122 * this call? 1123 */ 1124 sync_initial_page_table(); 1125 1126 tboot_probe(); 1127 1128 map_vsyscall(); 1129 1130 x86_32_probe_apic(); 1131 1132 early_quirks(); 1133 1134 topology_apply_cmdline_limits_early(); 1135 1136 /* 1137 * Parse SMP configuration. Try ACPI first and then the platform 1138 * specific parser. 1139 */ 1140 acpi_boot_init(); 1141 x86_init.mpparse.parse_smp_cfg(); 1142 1143 /* Last opportunity to detect and map the local APIC */ 1144 init_apic_mappings(); 1145 1146 topology_init_possible_cpus(); 1147 1148 init_cpu_to_node(); 1149 init_gi_nodes(); 1150 1151 io_apic_init_mappings(); 1152 1153 x86_init.hyper.guest_late_init(); 1154 1155 e820__reserve_resources(); 1156 e820__register_nosave_regions(max_pfn); 1157 1158 x86_init.resources.reserve_resources(); 1159 1160 e820__setup_pci_gap(); 1161 1162 #ifdef CONFIG_VT 1163 #if defined(CONFIG_VGA_CONSOLE) 1164 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1165 vgacon_register_screen(&screen_info); 1166 #endif 1167 #endif 1168 x86_init.oem.banner(); 1169 1170 x86_init.timers.wallclock_init(); 1171 1172 /* 1173 * This needs to run before setup_local_APIC() which soft-disables the 1174 * local APIC temporarily and that masks the thermal LVT interrupt, 1175 * leading to softlockups on machines which have configured SMI 1176 * interrupt delivery. 1177 */ 1178 therm_lvt_init(); 1179 1180 mcheck_init(); 1181 1182 register_refined_jiffies(CLOCK_TICK_RATE); 1183 1184 #ifdef CONFIG_EFI 1185 if (efi_enabled(EFI_BOOT)) 1186 efi_apply_memmap_quirks(); 1187 #endif 1188 1189 unwind_init(); 1190 } 1191 1192 #ifdef CONFIG_X86_32 1193 1194 static struct resource video_ram_resource = { 1195 .name = "Video RAM area", 1196 .start = 0xa0000, 1197 .end = 0xbffff, 1198 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1199 }; 1200 1201 void __init i386_reserve_resources(void) 1202 { 1203 request_resource(&iomem_resource, &video_ram_resource); 1204 reserve_standard_io_resources(); 1205 } 1206 1207 #endif /* CONFIG_X86_32 */ 1208 1209 static struct notifier_block kernel_offset_notifier = { 1210 .notifier_call = dump_kernel_offset 1211 }; 1212 1213 static int __init register_kernel_offset_dumper(void) 1214 { 1215 atomic_notifier_chain_register(&panic_notifier_list, 1216 &kernel_offset_notifier); 1217 return 0; 1218 } 1219 __initcall(register_kernel_offset_dumper); 1220 1221 #ifdef CONFIG_HOTPLUG_CPU 1222 bool arch_cpu_is_hotpluggable(int cpu) 1223 { 1224 return cpu > 0; 1225 } 1226 #endif /* CONFIG_HOTPLUG_CPU */ 1227