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 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 168 bool builtin_cmdline_added __ro_after_init; 169 #endif 170 171 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 172 struct edd edd; 173 #ifdef CONFIG_EDD_MODULE 174 EXPORT_SYMBOL(edd); 175 #endif 176 /** 177 * copy_edd() - Copy the BIOS EDD information 178 * from boot_params into a safe place. 179 * 180 */ 181 static inline void __init copy_edd(void) 182 { 183 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 184 sizeof(edd.mbr_signature)); 185 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 186 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 187 edd.edd_info_nr = boot_params.eddbuf_entries; 188 } 189 #else 190 static inline void __init copy_edd(void) 191 { 192 } 193 #endif 194 195 void * __init extend_brk(size_t size, size_t align) 196 { 197 size_t mask = align - 1; 198 void *ret; 199 200 BUG_ON(_brk_start == 0); 201 BUG_ON(align & mask); 202 203 _brk_end = (_brk_end + mask) & ~mask; 204 BUG_ON((char *)(_brk_end + size) > __brk_limit); 205 206 ret = (void *)_brk_end; 207 _brk_end += size; 208 209 memset(ret, 0, size); 210 211 return ret; 212 } 213 214 #ifdef CONFIG_X86_32 215 static void __init cleanup_highmap(void) 216 { 217 } 218 #endif 219 220 static void __init reserve_brk(void) 221 { 222 if (_brk_end > _brk_start) 223 memblock_reserve(__pa_symbol(_brk_start), 224 _brk_end - _brk_start); 225 226 /* Mark brk area as locked down and no longer taking any 227 new allocations */ 228 _brk_start = 0; 229 } 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 if (ramdisk_image == 0) 240 ramdisk_image = phys_initrd_start; 241 242 return ramdisk_image; 243 } 244 static u64 __init get_ramdisk_size(void) 245 { 246 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 247 248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 249 250 if (ramdisk_size == 0) 251 ramdisk_size = phys_initrd_size; 252 253 return ramdisk_size; 254 } 255 256 static void __init relocate_initrd(void) 257 { 258 /* Assume only end is not page aligned */ 259 u64 ramdisk_image = get_ramdisk_image(); 260 u64 ramdisk_size = get_ramdisk_size(); 261 u64 area_size = PAGE_ALIGN(ramdisk_size); 262 int ret = 0; 263 264 /* We need to move the initrd down into directly mapped mem */ 265 u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 266 PFN_PHYS(max_pfn_mapped)); 267 if (!relocated_ramdisk) 268 panic("Cannot find place for new RAMDISK of size %lld\n", 269 ramdisk_size); 270 271 initrd_start = relocated_ramdisk + PAGE_OFFSET; 272 initrd_end = initrd_start + ramdisk_size; 273 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 274 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 275 276 ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 277 if (ret) 278 panic("Copy RAMDISK failed\n"); 279 280 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 281 " [mem %#010llx-%#010llx]\n", 282 ramdisk_image, ramdisk_image + ramdisk_size - 1, 283 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 284 } 285 286 static void __init early_reserve_initrd(void) 287 { 288 /* Assume only end is not page aligned */ 289 u64 ramdisk_image = get_ramdisk_image(); 290 u64 ramdisk_size = get_ramdisk_size(); 291 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 292 293 if (!boot_params.hdr.type_of_loader || 294 !ramdisk_image || !ramdisk_size) 295 return; /* No initrd provided by bootloader */ 296 297 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 298 } 299 300 static void __init reserve_initrd(void) 301 { 302 /* Assume only end is not page aligned */ 303 u64 ramdisk_image = get_ramdisk_image(); 304 u64 ramdisk_size = get_ramdisk_size(); 305 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 306 307 if (!boot_params.hdr.type_of_loader || 308 !ramdisk_image || !ramdisk_size) 309 return; /* No initrd provided by bootloader */ 310 311 initrd_start = 0; 312 313 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 314 ramdisk_end - 1); 315 316 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 317 PFN_DOWN(ramdisk_end))) { 318 /* All are mapped, easy case */ 319 initrd_start = ramdisk_image + PAGE_OFFSET; 320 initrd_end = initrd_start + ramdisk_size; 321 return; 322 } 323 324 relocate_initrd(); 325 326 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image); 327 } 328 329 #else 330 static void __init early_reserve_initrd(void) 331 { 332 } 333 static void __init reserve_initrd(void) 334 { 335 } 336 #endif /* CONFIG_BLK_DEV_INITRD */ 337 338 static void __init add_early_ima_buffer(u64 phys_addr) 339 { 340 #ifdef CONFIG_IMA 341 struct ima_setup_data *data; 342 343 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data)); 344 if (!data) { 345 pr_warn("setup: failed to memremap ima_setup_data entry\n"); 346 return; 347 } 348 349 if (data->size) { 350 memblock_reserve(data->addr, data->size); 351 ima_kexec_buffer_phys = data->addr; 352 ima_kexec_buffer_size = data->size; 353 } 354 355 early_memunmap(data, sizeof(*data)); 356 #else 357 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n"); 358 #endif 359 } 360 361 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE) 362 int __init ima_free_kexec_buffer(void) 363 { 364 if (!ima_kexec_buffer_size) 365 return -ENOENT; 366 367 memblock_free_late(ima_kexec_buffer_phys, 368 ima_kexec_buffer_size); 369 370 ima_kexec_buffer_phys = 0; 371 ima_kexec_buffer_size = 0; 372 373 return 0; 374 } 375 376 int __init ima_get_kexec_buffer(void **addr, size_t *size) 377 { 378 if (!ima_kexec_buffer_size) 379 return -ENOENT; 380 381 *addr = __va(ima_kexec_buffer_phys); 382 *size = ima_kexec_buffer_size; 383 384 return 0; 385 } 386 #endif 387 388 static void __init parse_setup_data(void) 389 { 390 struct setup_data *data; 391 u64 pa_data, pa_next; 392 393 pa_data = boot_params.hdr.setup_data; 394 while (pa_data) { 395 u32 data_len, data_type; 396 397 data = early_memremap(pa_data, sizeof(*data)); 398 data_len = data->len + sizeof(struct setup_data); 399 data_type = data->type; 400 pa_next = data->next; 401 early_memunmap(data, sizeof(*data)); 402 403 switch (data_type) { 404 case SETUP_E820_EXT: 405 e820__memory_setup_extended(pa_data, data_len); 406 break; 407 case SETUP_DTB: 408 add_dtb(pa_data); 409 break; 410 case SETUP_EFI: 411 parse_efi_setup(pa_data, data_len); 412 break; 413 case SETUP_IMA: 414 add_early_ima_buffer(pa_data); 415 break; 416 case SETUP_RNG_SEED: 417 data = early_memremap(pa_data, data_len); 418 add_bootloader_randomness(data->data, data->len); 419 /* Zero seed for forward secrecy. */ 420 memzero_explicit(data->data, data->len); 421 /* Zero length in case we find ourselves back here by accident. */ 422 memzero_explicit(&data->len, sizeof(data->len)); 423 early_memunmap(data, data_len); 424 break; 425 default: 426 break; 427 } 428 pa_data = pa_next; 429 } 430 } 431 432 /* 433 * Translate the fields of 'struct boot_param' into global variables 434 * representing these parameters. 435 */ 436 static void __init parse_boot_params(void) 437 { 438 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 439 screen_info = boot_params.screen_info; 440 edid_info = boot_params.edid_info; 441 #ifdef CONFIG_X86_32 442 apm_info.bios = boot_params.apm_bios_info; 443 ist_info = boot_params.ist_info; 444 #endif 445 saved_video_mode = boot_params.hdr.vid_mode; 446 bootloader_type = boot_params.hdr.type_of_loader; 447 if ((bootloader_type >> 4) == 0xe) { 448 bootloader_type &= 0xf; 449 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 450 } 451 bootloader_version = bootloader_type & 0xf; 452 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 453 454 #ifdef CONFIG_BLK_DEV_RAM 455 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 456 #endif 457 #ifdef CONFIG_EFI 458 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 459 EFI32_LOADER_SIGNATURE, 4)) { 460 set_bit(EFI_BOOT, &efi.flags); 461 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 462 EFI64_LOADER_SIGNATURE, 4)) { 463 set_bit(EFI_BOOT, &efi.flags); 464 set_bit(EFI_64BIT, &efi.flags); 465 } 466 #endif 467 468 if (!boot_params.hdr.root_flags) 469 root_mountflags &= ~MS_RDONLY; 470 } 471 472 static void __init memblock_x86_reserve_range_setup_data(void) 473 { 474 struct setup_indirect *indirect; 475 struct setup_data *data; 476 u64 pa_data, pa_next; 477 u32 len; 478 479 pa_data = boot_params.hdr.setup_data; 480 while (pa_data) { 481 data = early_memremap(pa_data, sizeof(*data)); 482 if (!data) { 483 pr_warn("setup: failed to memremap setup_data entry\n"); 484 return; 485 } 486 487 len = sizeof(*data); 488 pa_next = data->next; 489 490 memblock_reserve(pa_data, sizeof(*data) + data->len); 491 492 if (data->type == SETUP_INDIRECT) { 493 len += data->len; 494 early_memunmap(data, sizeof(*data)); 495 data = early_memremap(pa_data, len); 496 if (!data) { 497 pr_warn("setup: failed to memremap indirect setup_data\n"); 498 return; 499 } 500 501 indirect = (struct setup_indirect *)data->data; 502 503 if (indirect->type != SETUP_INDIRECT) 504 memblock_reserve(indirect->addr, indirect->len); 505 } 506 507 pa_data = pa_next; 508 early_memunmap(data, len); 509 } 510 } 511 512 static void __init arch_reserve_crashkernel(void) 513 { 514 unsigned long long crash_base, crash_size, low_size = 0; 515 char *cmdline = boot_command_line; 516 bool high = false; 517 int ret; 518 519 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 520 return; 521 522 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 523 &crash_size, &crash_base, 524 &low_size, &high); 525 if (ret) 526 return; 527 528 if (xen_pv_domain()) { 529 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 530 return; 531 } 532 533 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 534 low_size, high); 535 } 536 537 static struct resource standard_io_resources[] = { 538 { .name = "dma1", .start = 0x00, .end = 0x1f, 539 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 540 { .name = "pic1", .start = 0x20, .end = 0x21, 541 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 542 { .name = "timer0", .start = 0x40, .end = 0x43, 543 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 544 { .name = "timer1", .start = 0x50, .end = 0x53, 545 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 546 { .name = "keyboard", .start = 0x60, .end = 0x60, 547 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 548 { .name = "keyboard", .start = 0x64, .end = 0x64, 549 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 550 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 551 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 552 { .name = "pic2", .start = 0xa0, .end = 0xa1, 553 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 554 { .name = "dma2", .start = 0xc0, .end = 0xdf, 555 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 556 { .name = "fpu", .start = 0xf0, .end = 0xff, 557 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 558 }; 559 560 void __init reserve_standard_io_resources(void) 561 { 562 int i; 563 564 /* request I/O space for devices used on all i[345]86 PCs */ 565 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 566 request_resource(&ioport_resource, &standard_io_resources[i]); 567 568 } 569 570 static void __init setup_kernel_resources(void) 571 { 572 code_resource.start = __pa_symbol(_text); 573 code_resource.end = __pa_symbol(_etext)-1; 574 rodata_resource.start = __pa_symbol(__start_rodata); 575 rodata_resource.end = __pa_symbol(__end_rodata)-1; 576 data_resource.start = __pa_symbol(_sdata); 577 data_resource.end = __pa_symbol(_edata)-1; 578 bss_resource.start = __pa_symbol(__bss_start); 579 bss_resource.end = __pa_symbol(__bss_stop)-1; 580 581 insert_resource(&iomem_resource, &code_resource); 582 insert_resource(&iomem_resource, &rodata_resource); 583 insert_resource(&iomem_resource, &data_resource); 584 insert_resource(&iomem_resource, &bss_resource); 585 } 586 587 static bool __init snb_gfx_workaround_needed(void) 588 { 589 #ifdef CONFIG_PCI 590 int i; 591 u16 vendor, devid; 592 static const __initconst u16 snb_ids[] = { 593 0x0102, 594 0x0112, 595 0x0122, 596 0x0106, 597 0x0116, 598 0x0126, 599 0x010a, 600 }; 601 602 /* Assume no if something weird is going on with PCI */ 603 if (!early_pci_allowed()) 604 return false; 605 606 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 607 if (vendor != 0x8086) 608 return false; 609 610 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 611 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 612 if (devid == snb_ids[i]) 613 return true; 614 #endif 615 616 return false; 617 } 618 619 /* 620 * Sandy Bridge graphics has trouble with certain ranges, exclude 621 * them from allocation. 622 */ 623 static void __init trim_snb_memory(void) 624 { 625 static const __initconst unsigned long bad_pages[] = { 626 0x20050000, 627 0x20110000, 628 0x20130000, 629 0x20138000, 630 0x40004000, 631 }; 632 int i; 633 634 if (!snb_gfx_workaround_needed()) 635 return; 636 637 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 638 639 /* 640 * SandyBridge integrated graphics devices have a bug that prevents 641 * them from accessing certain memory ranges, namely anything below 642 * 1M and in the pages listed in bad_pages[] above. 643 * 644 * To avoid these pages being ever accessed by SNB gfx devices reserve 645 * bad_pages that have not already been reserved at boot time. 646 * All memory below the 1 MB mark is anyway reserved later during 647 * setup_arch(), so there is no need to reserve it here. 648 */ 649 650 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 651 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 652 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 653 bad_pages[i]); 654 } 655 } 656 657 static void __init trim_bios_range(void) 658 { 659 /* 660 * A special case is the first 4Kb of memory; 661 * This is a BIOS owned area, not kernel ram, but generally 662 * not listed as such in the E820 table. 663 * 664 * This typically reserves additional memory (64KiB by default) 665 * since some BIOSes are known to corrupt low memory. See the 666 * Kconfig help text for X86_RESERVE_LOW. 667 */ 668 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 669 670 /* 671 * special case: Some BIOSes report the PC BIOS 672 * area (640Kb -> 1Mb) as RAM even though it is not. 673 * take them out. 674 */ 675 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 676 677 e820__update_table(e820_table); 678 } 679 680 /* called before trim_bios_range() to spare extra sanitize */ 681 static void __init e820_add_kernel_range(void) 682 { 683 u64 start = __pa_symbol(_text); 684 u64 size = __pa_symbol(_end) - start; 685 686 /* 687 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 688 * attempt to fix it by adding the range. We may have a confused BIOS, 689 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 690 * exclude kernel range. If we really are running on top non-RAM, 691 * we will crash later anyways. 692 */ 693 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 694 return; 695 696 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 697 e820__range_remove(start, size, E820_TYPE_RAM, 0); 698 e820__range_add(start, size, E820_TYPE_RAM); 699 } 700 701 static void __init early_reserve_memory(void) 702 { 703 /* 704 * Reserve the memory occupied by the kernel between _text and 705 * __end_of_kernel_reserve symbols. Any kernel sections after the 706 * __end_of_kernel_reserve symbol must be explicitly reserved with a 707 * separate memblock_reserve() or they will be discarded. 708 */ 709 memblock_reserve(__pa_symbol(_text), 710 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 711 712 /* 713 * The first 4Kb of memory is a BIOS owned area, but generally it is 714 * not listed as such in the E820 table. 715 * 716 * Reserve the first 64K of memory since some BIOSes are known to 717 * corrupt low memory. After the real mode trampoline is allocated the 718 * rest of the memory below 640k is reserved. 719 * 720 * In addition, make sure page 0 is always reserved because on 721 * systems with L1TF its contents can be leaked to user processes. 722 */ 723 memblock_reserve(0, SZ_64K); 724 725 early_reserve_initrd(); 726 727 memblock_x86_reserve_range_setup_data(); 728 729 reserve_bios_regions(); 730 trim_snb_memory(); 731 } 732 733 /* 734 * Dump out kernel offset information on panic. 735 */ 736 static int 737 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 738 { 739 if (kaslr_enabled()) { 740 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 741 kaslr_offset(), 742 __START_KERNEL, 743 __START_KERNEL_map, 744 MODULES_VADDR-1); 745 } else { 746 pr_emerg("Kernel Offset: disabled\n"); 747 } 748 749 return 0; 750 } 751 752 void x86_configure_nx(void) 753 { 754 if (boot_cpu_has(X86_FEATURE_NX)) 755 __supported_pte_mask |= _PAGE_NX; 756 else 757 __supported_pte_mask &= ~_PAGE_NX; 758 } 759 760 static void __init x86_report_nx(void) 761 { 762 if (!boot_cpu_has(X86_FEATURE_NX)) { 763 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 764 "missing in CPU!\n"); 765 } else { 766 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 767 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 768 #else 769 /* 32bit non-PAE kernel, NX cannot be used */ 770 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 771 "cannot be enabled: non-PAE kernel!\n"); 772 #endif 773 } 774 } 775 776 /* 777 * Determine if we were loaded by an EFI loader. If so, then we have also been 778 * passed the efi memmap, systab, etc., so we should use these data structures 779 * for initialization. Note, the efi init code path is determined by the 780 * global efi_enabled. This allows the same kernel image to be used on existing 781 * systems (with a traditional BIOS) as well as on EFI systems. 782 */ 783 /* 784 * setup_arch - architecture-specific boot-time initializations 785 * 786 * Note: On x86_64, fixmaps are ready for use even before this is called. 787 */ 788 789 void __init setup_arch(char **cmdline_p) 790 { 791 #ifdef CONFIG_X86_32 792 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 793 794 /* 795 * copy kernel address range established so far and switch 796 * to the proper swapper page table 797 */ 798 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 799 initial_page_table + KERNEL_PGD_BOUNDARY, 800 KERNEL_PGD_PTRS); 801 802 load_cr3(swapper_pg_dir); 803 /* 804 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 805 * a cr3 based tlb flush, so the following __flush_tlb_all() 806 * will not flush anything because the CPU quirk which clears 807 * X86_FEATURE_PGE has not been invoked yet. Though due to the 808 * load_cr3() above the TLB has been flushed already. The 809 * quirk is invoked before subsequent calls to __flush_tlb_all() 810 * so proper operation is guaranteed. 811 */ 812 __flush_tlb_all(); 813 #else 814 printk(KERN_INFO "Command line: %s\n", boot_command_line); 815 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 816 #endif 817 818 #ifdef CONFIG_CMDLINE_BOOL 819 #ifdef CONFIG_CMDLINE_OVERRIDE 820 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 821 #else 822 if (builtin_cmdline[0]) { 823 /* append boot loader cmdline to builtin */ 824 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 825 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 826 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 827 } 828 #endif 829 builtin_cmdline_added = true; 830 #endif 831 832 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 833 *cmdline_p = command_line; 834 835 /* 836 * If we have OLPC OFW, we might end up relocating the fixmap due to 837 * reserve_top(), so do this before touching the ioremap area. 838 */ 839 olpc_ofw_detect(); 840 841 idt_setup_early_traps(); 842 early_cpu_init(); 843 jump_label_init(); 844 static_call_init(); 845 early_ioremap_init(); 846 847 setup_olpc_ofw_pgd(); 848 849 parse_boot_params(); 850 851 x86_init.oem.arch_setup(); 852 853 /* 854 * Do some memory reservations *before* memory is added to memblock, so 855 * memblock allocations won't overwrite it. 856 * 857 * After this point, everything still needed from the boot loader or 858 * firmware or kernel text should be early reserved or marked not RAM in 859 * e820. All other memory is free game. 860 * 861 * This call needs to happen before e820__memory_setup() which calls the 862 * xen_memory_setup() on Xen dom0 which relies on the fact that those 863 * early reservations have happened already. 864 */ 865 early_reserve_memory(); 866 867 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 868 e820__memory_setup(); 869 parse_setup_data(); 870 871 copy_edd(); 872 873 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 874 875 /* 876 * x86_configure_nx() is called before parse_early_param() to detect 877 * whether hardware doesn't support NX (so that the early EHCI debug 878 * console setup can safely call set_fixmap()). 879 */ 880 x86_configure_nx(); 881 882 parse_early_param(); 883 884 if (efi_enabled(EFI_BOOT)) 885 efi_memblock_x86_reserve_range(); 886 887 x86_report_nx(); 888 889 apic_setup_apic_calls(); 890 891 if (acpi_mps_check()) { 892 #ifdef CONFIG_X86_LOCAL_APIC 893 apic_is_disabled = true; 894 #endif 895 setup_clear_cpu_cap(X86_FEATURE_APIC); 896 } 897 898 e820__finish_early_params(); 899 900 if (efi_enabled(EFI_BOOT)) 901 efi_init(); 902 903 reserve_ibft_region(); 904 x86_init.resources.dmi_setup(); 905 906 /* 907 * VMware detection requires dmi to be available, so this 908 * needs to be done after dmi_setup(), for the boot CPU. 909 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be 910 * called before cache_bp_init() for setting up MTRR state. 911 */ 912 init_hypervisor_platform(); 913 914 tsc_early_init(); 915 x86_init.resources.probe_roms(); 916 917 /* 918 * Add resources for kernel text and data to the iomem_resource. 919 * Do it after parse_early_param, so it can be debugged. 920 */ 921 setup_kernel_resources(); 922 923 e820_add_kernel_range(); 924 trim_bios_range(); 925 #ifdef CONFIG_X86_32 926 if (ppro_with_ram_bug()) { 927 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 928 E820_TYPE_RESERVED); 929 e820__update_table(e820_table); 930 printk(KERN_INFO "fixed physical RAM map:\n"); 931 e820__print_table("bad_ppro"); 932 } 933 #else 934 early_gart_iommu_check(); 935 #endif 936 937 /* 938 * partially used pages are not usable - thus 939 * we are rounding upwards: 940 */ 941 max_pfn = e820__end_of_ram_pfn(); 942 943 /* update e820 for memory not covered by WB MTRRs */ 944 cache_bp_init(); 945 if (mtrr_trim_uncached_memory(max_pfn)) 946 max_pfn = e820__end_of_ram_pfn(); 947 948 max_possible_pfn = max_pfn; 949 950 /* 951 * Define random base addresses for memory sections after max_pfn is 952 * defined and before each memory section base is used. 953 */ 954 kernel_randomize_memory(); 955 956 #ifdef CONFIG_X86_32 957 /* max_low_pfn get updated here */ 958 find_low_pfn_range(); 959 #else 960 check_x2apic(); 961 962 /* How many end-of-memory variables you have, grandma! */ 963 /* need this before calling reserve_initrd */ 964 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 965 max_low_pfn = e820__end_of_low_ram_pfn(); 966 else 967 max_low_pfn = max_pfn; 968 969 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 970 #endif 971 972 /* Find and reserve MPTABLE area */ 973 x86_init.mpparse.find_mptable(); 974 975 early_alloc_pgt_buf(); 976 977 /* 978 * Need to conclude brk, before e820__memblock_setup() 979 * it could use memblock_find_in_range, could overlap with 980 * brk area. 981 */ 982 reserve_brk(); 983 984 cleanup_highmap(); 985 986 e820__memblock_setup(); 987 988 /* 989 * Needs to run after memblock setup because it needs the physical 990 * memory size. 991 */ 992 mem_encrypt_setup_arch(); 993 cc_random_init(); 994 995 efi_find_mirror(); 996 efi_esrt_init(); 997 efi_mokvar_table_init(); 998 999 /* 1000 * The EFI specification says that boot service code won't be 1001 * called after ExitBootServices(). This is, in fact, a lie. 1002 */ 1003 efi_reserve_boot_services(); 1004 1005 /* preallocate 4k for mptable mpc */ 1006 e820__memblock_alloc_reserved_mpc_new(); 1007 1008 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1009 setup_bios_corruption_check(); 1010 #endif 1011 1012 #ifdef CONFIG_X86_32 1013 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1014 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1015 #endif 1016 1017 /* 1018 * Find free memory for the real mode trampoline and place it there. If 1019 * there is not enough free memory under 1M, on EFI-enabled systems 1020 * there will be additional attempt to reclaim the memory for the real 1021 * mode trampoline at efi_free_boot_services(). 1022 * 1023 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1024 * are known to corrupt low memory and several hundred kilobytes are not 1025 * worth complex detection what memory gets clobbered. Windows does the 1026 * same thing for very similar reasons. 1027 * 1028 * Moreover, on machines with SandyBridge graphics or in setups that use 1029 * crashkernel the entire 1M is reserved anyway. 1030 * 1031 * Note the host kernel TDX also requires the first 1MB being reserved. 1032 */ 1033 x86_platform.realmode_reserve(); 1034 1035 init_mem_mapping(); 1036 1037 /* 1038 * init_mem_mapping() relies on the early IDT page fault handling. 1039 * Now either enable FRED or install the real page fault handler 1040 * for 64-bit in the IDT. 1041 */ 1042 cpu_init_replace_early_idt(); 1043 1044 /* 1045 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1046 * with the current CR4 value. This may not be necessary, but 1047 * auditing all the early-boot CR4 manipulation would be needed to 1048 * rule it out. 1049 * 1050 * Mask off features that don't work outside long mode (just 1051 * PCIDE for now). 1052 */ 1053 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1054 1055 memblock_set_current_limit(get_max_mapped()); 1056 1057 /* 1058 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1059 */ 1060 1061 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1062 if (init_ohci1394_dma_early) 1063 init_ohci1394_dma_on_all_controllers(); 1064 #endif 1065 /* Allocate bigger log buffer */ 1066 setup_log_buf(1); 1067 1068 if (efi_enabled(EFI_BOOT)) { 1069 switch (boot_params.secure_boot) { 1070 case efi_secureboot_mode_disabled: 1071 pr_info("Secure boot disabled\n"); 1072 break; 1073 case efi_secureboot_mode_enabled: 1074 pr_info("Secure boot enabled\n"); 1075 break; 1076 default: 1077 pr_info("Secure boot could not be determined\n"); 1078 break; 1079 } 1080 } 1081 1082 reserve_initrd(); 1083 1084 acpi_table_upgrade(); 1085 /* Look for ACPI tables and reserve memory occupied by them. */ 1086 acpi_boot_table_init(); 1087 1088 vsmp_init(); 1089 1090 io_delay_init(); 1091 1092 early_platform_quirks(); 1093 1094 /* Some platforms need the APIC registered for NUMA configuration */ 1095 early_acpi_boot_init(); 1096 x86_init.mpparse.early_parse_smp_cfg(); 1097 1098 x86_flattree_get_config(); 1099 1100 initmem_init(); 1101 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1102 1103 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1104 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1105 1106 /* 1107 * Reserve memory for crash kernel after SRAT is parsed so that it 1108 * won't consume hotpluggable memory. 1109 */ 1110 arch_reserve_crashkernel(); 1111 1112 if (!early_xdbc_setup_hardware()) 1113 early_xdbc_register_console(); 1114 1115 x86_init.paging.pagetable_init(); 1116 1117 kasan_init(); 1118 1119 /* 1120 * Sync back kernel address range. 1121 * 1122 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1123 * this call? 1124 */ 1125 sync_initial_page_table(); 1126 1127 tboot_probe(); 1128 1129 map_vsyscall(); 1130 1131 x86_32_probe_apic(); 1132 1133 early_quirks(); 1134 1135 topology_apply_cmdline_limits_early(); 1136 1137 /* 1138 * Parse SMP configuration. Try ACPI first and then the platform 1139 * specific parser. 1140 */ 1141 acpi_boot_init(); 1142 x86_init.mpparse.parse_smp_cfg(); 1143 1144 /* Last opportunity to detect and map the local APIC */ 1145 init_apic_mappings(); 1146 1147 topology_init_possible_cpus(); 1148 1149 init_cpu_to_node(); 1150 init_gi_nodes(); 1151 1152 io_apic_init_mappings(); 1153 1154 x86_init.hyper.guest_late_init(); 1155 1156 e820__reserve_resources(); 1157 e820__register_nosave_regions(max_pfn); 1158 1159 x86_init.resources.reserve_resources(); 1160 1161 e820__setup_pci_gap(); 1162 1163 #ifdef CONFIG_VT 1164 #if defined(CONFIG_VGA_CONSOLE) 1165 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1166 vgacon_register_screen(&screen_info); 1167 #endif 1168 #endif 1169 x86_init.oem.banner(); 1170 1171 x86_init.timers.wallclock_init(); 1172 1173 /* 1174 * This needs to run before setup_local_APIC() which soft-disables the 1175 * local APIC temporarily and that masks the thermal LVT interrupt, 1176 * leading to softlockups on machines which have configured SMI 1177 * interrupt delivery. 1178 */ 1179 therm_lvt_init(); 1180 1181 mcheck_init(); 1182 1183 register_refined_jiffies(CLOCK_TICK_RATE); 1184 1185 #ifdef CONFIG_EFI 1186 if (efi_enabled(EFI_BOOT)) 1187 efi_apply_memmap_quirks(); 1188 #endif 1189 1190 unwind_init(); 1191 } 1192 1193 #ifdef CONFIG_X86_32 1194 1195 static struct resource video_ram_resource = { 1196 .name = "Video RAM area", 1197 .start = 0xa0000, 1198 .end = 0xbffff, 1199 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1200 }; 1201 1202 void __init i386_reserve_resources(void) 1203 { 1204 request_resource(&iomem_resource, &video_ram_resource); 1205 reserve_standard_io_resources(); 1206 } 1207 1208 #endif /* CONFIG_X86_32 */ 1209 1210 static struct notifier_block kernel_offset_notifier = { 1211 .notifier_call = dump_kernel_offset 1212 }; 1213 1214 static int __init register_kernel_offset_dumper(void) 1215 { 1216 atomic_notifier_chain_register(&panic_notifier_list, 1217 &kernel_offset_notifier); 1218 return 0; 1219 } 1220 __initcall(register_kernel_offset_dumper); 1221 1222 #ifdef CONFIG_HOTPLUG_CPU 1223 bool arch_cpu_is_hotpluggable(int cpu) 1224 { 1225 return cpu > 0; 1226 } 1227 #endif /* CONFIG_HOTPLUG_CPU */ 1228