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/crash_dump.h> 11 #include <linux/dma-map-ops.h> 12 #include <linux/dmi.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/cpu.h> 40 #include <asm/efi.h> 41 #include <asm/gart.h> 42 #include <asm/hypervisor.h> 43 #include <asm/io_apic.h> 44 #include <asm/kasan.h> 45 #include <asm/kaslr.h> 46 #include <asm/mce.h> 47 #include <asm/memtype.h> 48 #include <asm/mtrr.h> 49 #include <asm/realmode.h> 50 #include <asm/olpc_ofw.h> 51 #include <asm/pci-direct.h> 52 #include <asm/prom.h> 53 #include <asm/proto.h> 54 #include <asm/thermal.h> 55 #include <asm/unwind.h> 56 #include <asm/vsyscall.h> 57 #include <linux/vmalloc.h> 58 59 /* 60 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 61 * max_pfn_mapped: highest directly mapped pfn > 4 GB 62 * 63 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 64 * represented by pfn_mapped[]. 65 */ 66 unsigned long max_low_pfn_mapped; 67 unsigned long max_pfn_mapped; 68 69 #ifdef CONFIG_DMI 70 RESERVE_BRK(dmi_alloc, 65536); 71 #endif 72 73 74 unsigned long _brk_start = (unsigned long)__brk_base; 75 unsigned long _brk_end = (unsigned long)__brk_base; 76 77 struct boot_params boot_params; 78 79 /* 80 * These are the four main kernel memory regions, we put them into 81 * the resource tree so that kdump tools and other debugging tools 82 * recover it: 83 */ 84 85 static struct resource rodata_resource = { 86 .name = "Kernel rodata", 87 .start = 0, 88 .end = 0, 89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 90 }; 91 92 static struct resource data_resource = { 93 .name = "Kernel data", 94 .start = 0, 95 .end = 0, 96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 97 }; 98 99 static struct resource code_resource = { 100 .name = "Kernel code", 101 .start = 0, 102 .end = 0, 103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 104 }; 105 106 static struct resource bss_resource = { 107 .name = "Kernel bss", 108 .start = 0, 109 .end = 0, 110 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 111 }; 112 113 114 #ifdef CONFIG_X86_32 115 /* CPU data as detected by the assembly code in head_32.S */ 116 struct cpuinfo_x86 new_cpu_data; 117 unsigned int def_to_bigsmp; 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 u64 relocated_ramdisk; 231 232 #ifdef CONFIG_BLK_DEV_INITRD 233 234 static u64 __init get_ramdisk_image(void) 235 { 236 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 237 238 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 239 240 if (ramdisk_image == 0) 241 ramdisk_image = phys_initrd_start; 242 243 return ramdisk_image; 244 } 245 static u64 __init get_ramdisk_size(void) 246 { 247 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 248 249 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 250 251 if (ramdisk_size == 0) 252 ramdisk_size = phys_initrd_size; 253 254 return ramdisk_size; 255 } 256 257 static void __init relocate_initrd(void) 258 { 259 /* Assume only end is not page aligned */ 260 u64 ramdisk_image = get_ramdisk_image(); 261 u64 ramdisk_size = get_ramdisk_size(); 262 u64 area_size = PAGE_ALIGN(ramdisk_size); 263 264 /* We need to move the initrd down into directly mapped mem */ 265 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 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 277 278 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 279 " [mem %#010llx-%#010llx]\n", 280 ramdisk_image, ramdisk_image + ramdisk_size - 1, 281 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 282 } 283 284 static void __init early_reserve_initrd(void) 285 { 286 /* Assume only end is not page aligned */ 287 u64 ramdisk_image = get_ramdisk_image(); 288 u64 ramdisk_size = get_ramdisk_size(); 289 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 290 291 if (!boot_params.hdr.type_of_loader || 292 !ramdisk_image || !ramdisk_size) 293 return; /* No initrd provided by bootloader */ 294 295 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 296 } 297 298 static void __init reserve_initrd(void) 299 { 300 /* Assume only end is not page aligned */ 301 u64 ramdisk_image = get_ramdisk_image(); 302 u64 ramdisk_size = get_ramdisk_size(); 303 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 304 305 if (!boot_params.hdr.type_of_loader || 306 !ramdisk_image || !ramdisk_size) 307 return; /* No initrd provided by bootloader */ 308 309 initrd_start = 0; 310 311 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 312 ramdisk_end - 1); 313 314 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 315 PFN_DOWN(ramdisk_end))) { 316 /* All are mapped, easy case */ 317 initrd_start = ramdisk_image + PAGE_OFFSET; 318 initrd_end = initrd_start + ramdisk_size; 319 return; 320 } 321 322 relocate_initrd(); 323 324 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image); 325 } 326 327 #else 328 static void __init early_reserve_initrd(void) 329 { 330 } 331 static void __init reserve_initrd(void) 332 { 333 } 334 #endif /* CONFIG_BLK_DEV_INITRD */ 335 336 static void __init add_early_ima_buffer(u64 phys_addr) 337 { 338 #ifdef CONFIG_IMA 339 struct ima_setup_data *data; 340 341 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data)); 342 if (!data) { 343 pr_warn("setup: failed to memremap ima_setup_data entry\n"); 344 return; 345 } 346 347 if (data->size) { 348 memblock_reserve(data->addr, data->size); 349 ima_kexec_buffer_phys = data->addr; 350 ima_kexec_buffer_size = data->size; 351 } 352 353 early_memunmap(data, sizeof(*data)); 354 #else 355 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n"); 356 #endif 357 } 358 359 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE) 360 int __init ima_free_kexec_buffer(void) 361 { 362 int rc; 363 364 if (!ima_kexec_buffer_size) 365 return -ENOENT; 366 367 rc = memblock_phys_free(ima_kexec_buffer_phys, 368 ima_kexec_buffer_size); 369 if (rc) 370 return rc; 371 372 ima_kexec_buffer_phys = 0; 373 ima_kexec_buffer_size = 0; 374 375 return 0; 376 } 377 378 int __init ima_get_kexec_buffer(void **addr, size_t *size) 379 { 380 if (!ima_kexec_buffer_size) 381 return -ENOENT; 382 383 *addr = __va(ima_kexec_buffer_phys); 384 *size = ima_kexec_buffer_size; 385 386 return 0; 387 } 388 #endif 389 390 static void __init parse_setup_data(void) 391 { 392 struct setup_data *data; 393 u64 pa_data, pa_next; 394 395 pa_data = boot_params.hdr.setup_data; 396 while (pa_data) { 397 u32 data_len, data_type; 398 399 data = early_memremap(pa_data, sizeof(*data)); 400 data_len = data->len + sizeof(struct setup_data); 401 data_type = data->type; 402 pa_next = data->next; 403 early_memunmap(data, sizeof(*data)); 404 405 switch (data_type) { 406 case SETUP_E820_EXT: 407 e820__memory_setup_extended(pa_data, data_len); 408 break; 409 case SETUP_DTB: 410 add_dtb(pa_data); 411 break; 412 case SETUP_EFI: 413 parse_efi_setup(pa_data, data_len); 414 break; 415 case SETUP_IMA: 416 add_early_ima_buffer(pa_data); 417 break; 418 case SETUP_RNG_SEED: 419 data = early_memremap(pa_data, data_len); 420 add_bootloader_randomness(data->data, data->len); 421 /* Zero seed for forward secrecy. */ 422 memzero_explicit(data->data, data->len); 423 /* Zero length in case we find ourselves back here by accident. */ 424 memzero_explicit(&data->len, sizeof(data->len)); 425 early_memunmap(data, data_len); 426 break; 427 default: 428 break; 429 } 430 pa_data = pa_next; 431 } 432 } 433 434 static void __init memblock_x86_reserve_range_setup_data(void) 435 { 436 struct setup_indirect *indirect; 437 struct setup_data *data; 438 u64 pa_data, pa_next; 439 u32 len; 440 441 pa_data = boot_params.hdr.setup_data; 442 while (pa_data) { 443 data = early_memremap(pa_data, sizeof(*data)); 444 if (!data) { 445 pr_warn("setup: failed to memremap setup_data entry\n"); 446 return; 447 } 448 449 len = sizeof(*data); 450 pa_next = data->next; 451 452 memblock_reserve(pa_data, sizeof(*data) + data->len); 453 454 if (data->type == SETUP_INDIRECT) { 455 len += data->len; 456 early_memunmap(data, sizeof(*data)); 457 data = early_memremap(pa_data, len); 458 if (!data) { 459 pr_warn("setup: failed to memremap indirect setup_data\n"); 460 return; 461 } 462 463 indirect = (struct setup_indirect *)data->data; 464 465 if (indirect->type != SETUP_INDIRECT) 466 memblock_reserve(indirect->addr, indirect->len); 467 } 468 469 pa_data = pa_next; 470 early_memunmap(data, len); 471 } 472 } 473 474 /* 475 * --------- Crashkernel reservation ------------------------------ 476 */ 477 478 /* 16M alignment for crash kernel regions */ 479 #define CRASH_ALIGN SZ_16M 480 481 /* 482 * Keep the crash kernel below this limit. 483 * 484 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 485 * due to mapping restrictions. 486 * 487 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 488 * the upper limit of system RAM in 4-level paging mode. Since the kdump 489 * jump could be from 5-level paging to 4-level paging, the jump will fail if 490 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 491 * no good way to detect the paging mode of the target kernel which will be 492 * loaded for dumping. 493 */ 494 #ifdef CONFIG_X86_32 495 # define CRASH_ADDR_LOW_MAX SZ_512M 496 # define CRASH_ADDR_HIGH_MAX SZ_512M 497 #else 498 # define CRASH_ADDR_LOW_MAX SZ_4G 499 # define CRASH_ADDR_HIGH_MAX SZ_64T 500 #endif 501 502 static int __init reserve_crashkernel_low(void) 503 { 504 #ifdef CONFIG_X86_64 505 unsigned long long base, low_base = 0, low_size = 0; 506 unsigned long low_mem_limit; 507 int ret; 508 509 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 510 511 /* crashkernel=Y,low */ 512 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 513 if (ret) { 514 /* 515 * two parts from kernel/dma/swiotlb.c: 516 * -swiotlb size: user-specified with swiotlb= or default. 517 * 518 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 519 * to 8M for other buffers that may need to stay low too. Also 520 * make sure we allocate enough extra low memory so that we 521 * don't run out of DMA buffers for 32-bit devices. 522 */ 523 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 524 } else { 525 /* passed with crashkernel=0,low ? */ 526 if (!low_size) 527 return 0; 528 } 529 530 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 531 if (!low_base) { 532 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 533 (unsigned long)(low_size >> 20)); 534 return -ENOMEM; 535 } 536 537 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 538 (unsigned long)(low_size >> 20), 539 (unsigned long)(low_base >> 20), 540 (unsigned long)(low_mem_limit >> 20)); 541 542 crashk_low_res.start = low_base; 543 crashk_low_res.end = low_base + low_size - 1; 544 insert_resource(&iomem_resource, &crashk_low_res); 545 #endif 546 return 0; 547 } 548 549 static void __init reserve_crashkernel(void) 550 { 551 unsigned long long crash_size, crash_base, total_mem; 552 bool high = false; 553 int ret; 554 555 if (!IS_ENABLED(CONFIG_KEXEC_CORE)) 556 return; 557 558 total_mem = memblock_phys_mem_size(); 559 560 /* crashkernel=XM */ 561 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 562 if (ret != 0 || crash_size <= 0) { 563 /* crashkernel=X,high */ 564 ret = parse_crashkernel_high(boot_command_line, total_mem, 565 &crash_size, &crash_base); 566 if (ret != 0 || crash_size <= 0) 567 return; 568 high = true; 569 } 570 571 if (xen_pv_domain()) { 572 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 573 return; 574 } 575 576 /* 0 means: find the address automatically */ 577 if (!crash_base) { 578 /* 579 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 580 * crashkernel=x,high reserves memory over 4G, also allocates 581 * 256M extra low memory for DMA buffers and swiotlb. 582 * But the extra memory is not required for all machines. 583 * So try low memory first and fall back to high memory 584 * unless "crashkernel=size[KMG],high" is specified. 585 */ 586 if (!high) 587 crash_base = memblock_phys_alloc_range(crash_size, 588 CRASH_ALIGN, CRASH_ALIGN, 589 CRASH_ADDR_LOW_MAX); 590 if (!crash_base) 591 crash_base = memblock_phys_alloc_range(crash_size, 592 CRASH_ALIGN, CRASH_ALIGN, 593 CRASH_ADDR_HIGH_MAX); 594 if (!crash_base) { 595 pr_info("crashkernel reservation failed - No suitable area found.\n"); 596 return; 597 } 598 } else { 599 unsigned long long start; 600 601 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 602 crash_base + crash_size); 603 if (start != crash_base) { 604 pr_info("crashkernel reservation failed - memory is in use.\n"); 605 return; 606 } 607 } 608 609 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 610 memblock_phys_free(crash_base, crash_size); 611 return; 612 } 613 614 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 615 (unsigned long)(crash_size >> 20), 616 (unsigned long)(crash_base >> 20), 617 (unsigned long)(total_mem >> 20)); 618 619 crashk_res.start = crash_base; 620 crashk_res.end = crash_base + crash_size - 1; 621 insert_resource(&iomem_resource, &crashk_res); 622 } 623 624 static struct resource standard_io_resources[] = { 625 { .name = "dma1", .start = 0x00, .end = 0x1f, 626 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 627 { .name = "pic1", .start = 0x20, .end = 0x21, 628 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 629 { .name = "timer0", .start = 0x40, .end = 0x43, 630 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 631 { .name = "timer1", .start = 0x50, .end = 0x53, 632 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 633 { .name = "keyboard", .start = 0x60, .end = 0x60, 634 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 635 { .name = "keyboard", .start = 0x64, .end = 0x64, 636 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 637 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 638 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 639 { .name = "pic2", .start = 0xa0, .end = 0xa1, 640 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 641 { .name = "dma2", .start = 0xc0, .end = 0xdf, 642 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 643 { .name = "fpu", .start = 0xf0, .end = 0xff, 644 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 645 }; 646 647 void __init reserve_standard_io_resources(void) 648 { 649 int i; 650 651 /* request I/O space for devices used on all i[345]86 PCs */ 652 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 653 request_resource(&ioport_resource, &standard_io_resources[i]); 654 655 } 656 657 static bool __init snb_gfx_workaround_needed(void) 658 { 659 #ifdef CONFIG_PCI 660 int i; 661 u16 vendor, devid; 662 static const __initconst u16 snb_ids[] = { 663 0x0102, 664 0x0112, 665 0x0122, 666 0x0106, 667 0x0116, 668 0x0126, 669 0x010a, 670 }; 671 672 /* Assume no if something weird is going on with PCI */ 673 if (!early_pci_allowed()) 674 return false; 675 676 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 677 if (vendor != 0x8086) 678 return false; 679 680 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 681 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 682 if (devid == snb_ids[i]) 683 return true; 684 #endif 685 686 return false; 687 } 688 689 /* 690 * Sandy Bridge graphics has trouble with certain ranges, exclude 691 * them from allocation. 692 */ 693 static void __init trim_snb_memory(void) 694 { 695 static const __initconst unsigned long bad_pages[] = { 696 0x20050000, 697 0x20110000, 698 0x20130000, 699 0x20138000, 700 0x40004000, 701 }; 702 int i; 703 704 if (!snb_gfx_workaround_needed()) 705 return; 706 707 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 708 709 /* 710 * SandyBridge integrated graphics devices have a bug that prevents 711 * them from accessing certain memory ranges, namely anything below 712 * 1M and in the pages listed in bad_pages[] above. 713 * 714 * To avoid these pages being ever accessed by SNB gfx devices reserve 715 * bad_pages that have not already been reserved at boot time. 716 * All memory below the 1 MB mark is anyway reserved later during 717 * setup_arch(), so there is no need to reserve it here. 718 */ 719 720 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 721 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 722 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 723 bad_pages[i]); 724 } 725 } 726 727 static void __init trim_bios_range(void) 728 { 729 /* 730 * A special case is the first 4Kb of memory; 731 * This is a BIOS owned area, not kernel ram, but generally 732 * not listed as such in the E820 table. 733 * 734 * This typically reserves additional memory (64KiB by default) 735 * since some BIOSes are known to corrupt low memory. See the 736 * Kconfig help text for X86_RESERVE_LOW. 737 */ 738 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 739 740 /* 741 * special case: Some BIOSes report the PC BIOS 742 * area (640Kb -> 1Mb) as RAM even though it is not. 743 * take them out. 744 */ 745 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 746 747 e820__update_table(e820_table); 748 } 749 750 /* called before trim_bios_range() to spare extra sanitize */ 751 static void __init e820_add_kernel_range(void) 752 { 753 u64 start = __pa_symbol(_text); 754 u64 size = __pa_symbol(_end) - start; 755 756 /* 757 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 758 * attempt to fix it by adding the range. We may have a confused BIOS, 759 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 760 * exclude kernel range. If we really are running on top non-RAM, 761 * we will crash later anyways. 762 */ 763 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 764 return; 765 766 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 767 e820__range_remove(start, size, E820_TYPE_RAM, 0); 768 e820__range_add(start, size, E820_TYPE_RAM); 769 } 770 771 static void __init early_reserve_memory(void) 772 { 773 /* 774 * Reserve the memory occupied by the kernel between _text and 775 * __end_of_kernel_reserve symbols. Any kernel sections after the 776 * __end_of_kernel_reserve symbol must be explicitly reserved with a 777 * separate memblock_reserve() or they will be discarded. 778 */ 779 memblock_reserve(__pa_symbol(_text), 780 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 781 782 /* 783 * The first 4Kb of memory is a BIOS owned area, but generally it is 784 * not listed as such in the E820 table. 785 * 786 * Reserve the first 64K of memory since some BIOSes are known to 787 * corrupt low memory. After the real mode trampoline is allocated the 788 * rest of the memory below 640k is reserved. 789 * 790 * In addition, make sure page 0 is always reserved because on 791 * systems with L1TF its contents can be leaked to user processes. 792 */ 793 memblock_reserve(0, SZ_64K); 794 795 early_reserve_initrd(); 796 797 memblock_x86_reserve_range_setup_data(); 798 799 reserve_bios_regions(); 800 trim_snb_memory(); 801 } 802 803 /* 804 * Dump out kernel offset information on panic. 805 */ 806 static int 807 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 808 { 809 if (kaslr_enabled()) { 810 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 811 kaslr_offset(), 812 __START_KERNEL, 813 __START_KERNEL_map, 814 MODULES_VADDR-1); 815 } else { 816 pr_emerg("Kernel Offset: disabled\n"); 817 } 818 819 return 0; 820 } 821 822 void x86_configure_nx(void) 823 { 824 if (boot_cpu_has(X86_FEATURE_NX)) 825 __supported_pte_mask |= _PAGE_NX; 826 else 827 __supported_pte_mask &= ~_PAGE_NX; 828 } 829 830 static void __init x86_report_nx(void) 831 { 832 if (!boot_cpu_has(X86_FEATURE_NX)) { 833 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 834 "missing in CPU!\n"); 835 } else { 836 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 837 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 838 #else 839 /* 32bit non-PAE kernel, NX cannot be used */ 840 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 841 "cannot be enabled: non-PAE kernel!\n"); 842 #endif 843 } 844 } 845 846 /* 847 * Determine if we were loaded by an EFI loader. If so, then we have also been 848 * passed the efi memmap, systab, etc., so we should use these data structures 849 * for initialization. Note, the efi init code path is determined by the 850 * global efi_enabled. This allows the same kernel image to be used on existing 851 * systems (with a traditional BIOS) as well as on EFI systems. 852 */ 853 /* 854 * setup_arch - architecture-specific boot-time initializations 855 * 856 * Note: On x86_64, fixmaps are ready for use even before this is called. 857 */ 858 859 void __init setup_arch(char **cmdline_p) 860 { 861 #ifdef CONFIG_X86_32 862 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 863 864 /* 865 * copy kernel address range established so far and switch 866 * to the proper swapper page table 867 */ 868 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 869 initial_page_table + KERNEL_PGD_BOUNDARY, 870 KERNEL_PGD_PTRS); 871 872 load_cr3(swapper_pg_dir); 873 /* 874 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 875 * a cr3 based tlb flush, so the following __flush_tlb_all() 876 * will not flush anything because the CPU quirk which clears 877 * X86_FEATURE_PGE has not been invoked yet. Though due to the 878 * load_cr3() above the TLB has been flushed already. The 879 * quirk is invoked before subsequent calls to __flush_tlb_all() 880 * so proper operation is guaranteed. 881 */ 882 __flush_tlb_all(); 883 #else 884 printk(KERN_INFO "Command line: %s\n", boot_command_line); 885 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 886 #endif 887 888 /* 889 * If we have OLPC OFW, we might end up relocating the fixmap due to 890 * reserve_top(), so do this before touching the ioremap area. 891 */ 892 olpc_ofw_detect(); 893 894 idt_setup_early_traps(); 895 early_cpu_init(); 896 jump_label_init(); 897 static_call_init(); 898 early_ioremap_init(); 899 900 setup_olpc_ofw_pgd(); 901 902 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 903 screen_info = boot_params.screen_info; 904 edid_info = boot_params.edid_info; 905 #ifdef CONFIG_X86_32 906 apm_info.bios = boot_params.apm_bios_info; 907 ist_info = boot_params.ist_info; 908 #endif 909 saved_video_mode = boot_params.hdr.vid_mode; 910 bootloader_type = boot_params.hdr.type_of_loader; 911 if ((bootloader_type >> 4) == 0xe) { 912 bootloader_type &= 0xf; 913 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 914 } 915 bootloader_version = bootloader_type & 0xf; 916 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 917 918 #ifdef CONFIG_BLK_DEV_RAM 919 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 920 #endif 921 #ifdef CONFIG_EFI 922 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 923 EFI32_LOADER_SIGNATURE, 4)) { 924 set_bit(EFI_BOOT, &efi.flags); 925 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 926 EFI64_LOADER_SIGNATURE, 4)) { 927 set_bit(EFI_BOOT, &efi.flags); 928 set_bit(EFI_64BIT, &efi.flags); 929 } 930 #endif 931 932 x86_init.oem.arch_setup(); 933 934 /* 935 * Do some memory reservations *before* memory is added to memblock, so 936 * memblock allocations won't overwrite it. 937 * 938 * After this point, everything still needed from the boot loader or 939 * firmware or kernel text should be early reserved or marked not RAM in 940 * e820. All other memory is free game. 941 * 942 * This call needs to happen before e820__memory_setup() which calls the 943 * xen_memory_setup() on Xen dom0 which relies on the fact that those 944 * early reservations have happened already. 945 */ 946 early_reserve_memory(); 947 948 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 949 e820__memory_setup(); 950 parse_setup_data(); 951 952 copy_edd(); 953 954 if (!boot_params.hdr.root_flags) 955 root_mountflags &= ~MS_RDONLY; 956 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 957 958 code_resource.start = __pa_symbol(_text); 959 code_resource.end = __pa_symbol(_etext)-1; 960 rodata_resource.start = __pa_symbol(__start_rodata); 961 rodata_resource.end = __pa_symbol(__end_rodata)-1; 962 data_resource.start = __pa_symbol(_sdata); 963 data_resource.end = __pa_symbol(_edata)-1; 964 bss_resource.start = __pa_symbol(__bss_start); 965 bss_resource.end = __pa_symbol(__bss_stop)-1; 966 967 #ifdef CONFIG_CMDLINE_BOOL 968 #ifdef CONFIG_CMDLINE_OVERRIDE 969 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 970 #else 971 if (builtin_cmdline[0]) { 972 /* append boot loader cmdline to builtin */ 973 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 974 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 975 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 976 } 977 #endif 978 #endif 979 980 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 981 *cmdline_p = command_line; 982 983 /* 984 * x86_configure_nx() is called before parse_early_param() to detect 985 * whether hardware doesn't support NX (so that the early EHCI debug 986 * console setup can safely call set_fixmap()). 987 */ 988 x86_configure_nx(); 989 990 parse_early_param(); 991 992 if (efi_enabled(EFI_BOOT)) 993 efi_memblock_x86_reserve_range(); 994 995 #ifdef CONFIG_MEMORY_HOTPLUG 996 /* 997 * Memory used by the kernel cannot be hot-removed because Linux 998 * cannot migrate the kernel pages. When memory hotplug is 999 * enabled, we should prevent memblock from allocating memory 1000 * for the kernel. 1001 * 1002 * ACPI SRAT records all hotpluggable memory ranges. But before 1003 * SRAT is parsed, we don't know about it. 1004 * 1005 * The kernel image is loaded into memory at very early time. We 1006 * cannot prevent this anyway. So on NUMA system, we set any 1007 * node the kernel resides in as un-hotpluggable. 1008 * 1009 * Since on modern servers, one node could have double-digit 1010 * gigabytes memory, we can assume the memory around the kernel 1011 * image is also un-hotpluggable. So before SRAT is parsed, just 1012 * allocate memory near the kernel image to try the best to keep 1013 * the kernel away from hotpluggable memory. 1014 */ 1015 if (movable_node_is_enabled()) 1016 memblock_set_bottom_up(true); 1017 #endif 1018 1019 x86_report_nx(); 1020 1021 if (acpi_mps_check()) { 1022 #ifdef CONFIG_X86_LOCAL_APIC 1023 disable_apic = 1; 1024 #endif 1025 setup_clear_cpu_cap(X86_FEATURE_APIC); 1026 } 1027 1028 e820__reserve_setup_data(); 1029 e820__finish_early_params(); 1030 1031 if (efi_enabled(EFI_BOOT)) 1032 efi_init(); 1033 1034 reserve_ibft_region(); 1035 dmi_setup(); 1036 1037 /* 1038 * VMware detection requires dmi to be available, so this 1039 * needs to be done after dmi_setup(), for the boot CPU. 1040 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be 1041 * called before cache_bp_init() for setting up MTRR state. 1042 */ 1043 init_hypervisor_platform(); 1044 1045 tsc_early_init(); 1046 x86_init.resources.probe_roms(); 1047 1048 /* after parse_early_param, so could debug it */ 1049 insert_resource(&iomem_resource, &code_resource); 1050 insert_resource(&iomem_resource, &rodata_resource); 1051 insert_resource(&iomem_resource, &data_resource); 1052 insert_resource(&iomem_resource, &bss_resource); 1053 1054 e820_add_kernel_range(); 1055 trim_bios_range(); 1056 #ifdef CONFIG_X86_32 1057 if (ppro_with_ram_bug()) { 1058 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1059 E820_TYPE_RESERVED); 1060 e820__update_table(e820_table); 1061 printk(KERN_INFO "fixed physical RAM map:\n"); 1062 e820__print_table("bad_ppro"); 1063 } 1064 #else 1065 early_gart_iommu_check(); 1066 #endif 1067 1068 /* 1069 * partially used pages are not usable - thus 1070 * we are rounding upwards: 1071 */ 1072 max_pfn = e820__end_of_ram_pfn(); 1073 1074 /* update e820 for memory not covered by WB MTRRs */ 1075 cache_bp_init(); 1076 if (mtrr_trim_uncached_memory(max_pfn)) 1077 max_pfn = e820__end_of_ram_pfn(); 1078 1079 max_possible_pfn = max_pfn; 1080 1081 /* 1082 * Define random base addresses for memory sections after max_pfn is 1083 * defined and before each memory section base is used. 1084 */ 1085 kernel_randomize_memory(); 1086 1087 #ifdef CONFIG_X86_32 1088 /* max_low_pfn get updated here */ 1089 find_low_pfn_range(); 1090 #else 1091 check_x2apic(); 1092 1093 /* How many end-of-memory variables you have, grandma! */ 1094 /* need this before calling reserve_initrd */ 1095 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1096 max_low_pfn = e820__end_of_low_ram_pfn(); 1097 else 1098 max_low_pfn = max_pfn; 1099 1100 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1101 #endif 1102 1103 /* 1104 * Find and reserve possible boot-time SMP configuration: 1105 */ 1106 find_smp_config(); 1107 1108 early_alloc_pgt_buf(); 1109 1110 /* 1111 * Need to conclude brk, before e820__memblock_setup() 1112 * it could use memblock_find_in_range, could overlap with 1113 * brk area. 1114 */ 1115 reserve_brk(); 1116 1117 cleanup_highmap(); 1118 1119 memblock_set_current_limit(ISA_END_ADDRESS); 1120 e820__memblock_setup(); 1121 1122 /* 1123 * Needs to run after memblock setup because it needs the physical 1124 * memory size. 1125 */ 1126 sev_setup_arch(); 1127 1128 efi_fake_memmap(); 1129 efi_find_mirror(); 1130 efi_esrt_init(); 1131 efi_mokvar_table_init(); 1132 1133 /* 1134 * The EFI specification says that boot service code won't be 1135 * called after ExitBootServices(). This is, in fact, a lie. 1136 */ 1137 efi_reserve_boot_services(); 1138 1139 /* preallocate 4k for mptable mpc */ 1140 e820__memblock_alloc_reserved_mpc_new(); 1141 1142 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1143 setup_bios_corruption_check(); 1144 #endif 1145 1146 #ifdef CONFIG_X86_32 1147 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1148 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1149 #endif 1150 1151 /* 1152 * Find free memory for the real mode trampoline and place it there. If 1153 * there is not enough free memory under 1M, on EFI-enabled systems 1154 * there will be additional attempt to reclaim the memory for the real 1155 * mode trampoline at efi_free_boot_services(). 1156 * 1157 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1158 * are known to corrupt low memory and several hundred kilobytes are not 1159 * worth complex detection what memory gets clobbered. Windows does the 1160 * same thing for very similar reasons. 1161 * 1162 * Moreover, on machines with SandyBridge graphics or in setups that use 1163 * crashkernel the entire 1M is reserved anyway. 1164 */ 1165 x86_platform.realmode_reserve(); 1166 1167 init_mem_mapping(); 1168 1169 idt_setup_early_pf(); 1170 1171 /* 1172 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1173 * with the current CR4 value. This may not be necessary, but 1174 * auditing all the early-boot CR4 manipulation would be needed to 1175 * rule it out. 1176 * 1177 * Mask off features that don't work outside long mode (just 1178 * PCIDE for now). 1179 */ 1180 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1181 1182 memblock_set_current_limit(get_max_mapped()); 1183 1184 /* 1185 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1186 */ 1187 1188 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1189 if (init_ohci1394_dma_early) 1190 init_ohci1394_dma_on_all_controllers(); 1191 #endif 1192 /* Allocate bigger log buffer */ 1193 setup_log_buf(1); 1194 1195 if (efi_enabled(EFI_BOOT)) { 1196 switch (boot_params.secure_boot) { 1197 case efi_secureboot_mode_disabled: 1198 pr_info("Secure boot disabled\n"); 1199 break; 1200 case efi_secureboot_mode_enabled: 1201 pr_info("Secure boot enabled\n"); 1202 break; 1203 default: 1204 pr_info("Secure boot could not be determined\n"); 1205 break; 1206 } 1207 } 1208 1209 reserve_initrd(); 1210 1211 acpi_table_upgrade(); 1212 /* Look for ACPI tables and reserve memory occupied by them. */ 1213 acpi_boot_table_init(); 1214 1215 vsmp_init(); 1216 1217 io_delay_init(); 1218 1219 early_platform_quirks(); 1220 1221 early_acpi_boot_init(); 1222 1223 initmem_init(); 1224 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1225 1226 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1227 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1228 1229 /* 1230 * Reserve memory for crash kernel after SRAT is parsed so that it 1231 * won't consume hotpluggable memory. 1232 */ 1233 reserve_crashkernel(); 1234 1235 memblock_find_dma_reserve(); 1236 1237 if (!early_xdbc_setup_hardware()) 1238 early_xdbc_register_console(); 1239 1240 x86_init.paging.pagetable_init(); 1241 1242 kasan_init(); 1243 1244 /* 1245 * Sync back kernel address range. 1246 * 1247 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1248 * this call? 1249 */ 1250 sync_initial_page_table(); 1251 1252 tboot_probe(); 1253 1254 map_vsyscall(); 1255 1256 generic_apic_probe(); 1257 1258 early_quirks(); 1259 1260 /* 1261 * Read APIC and some other early information from ACPI tables. 1262 */ 1263 acpi_boot_init(); 1264 x86_dtb_init(); 1265 1266 /* 1267 * get boot-time SMP configuration: 1268 */ 1269 get_smp_config(); 1270 1271 /* 1272 * Systems w/o ACPI and mptables might not have it mapped the local 1273 * APIC yet, but prefill_possible_map() might need to access it. 1274 */ 1275 init_apic_mappings(); 1276 1277 prefill_possible_map(); 1278 1279 init_cpu_to_node(); 1280 init_gi_nodes(); 1281 1282 io_apic_init_mappings(); 1283 1284 x86_init.hyper.guest_late_init(); 1285 1286 e820__reserve_resources(); 1287 e820__register_nosave_regions(max_pfn); 1288 1289 x86_init.resources.reserve_resources(); 1290 1291 e820__setup_pci_gap(); 1292 1293 #ifdef CONFIG_VT 1294 #if defined(CONFIG_VGA_CONSOLE) 1295 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1296 conswitchp = &vga_con; 1297 #endif 1298 #endif 1299 x86_init.oem.banner(); 1300 1301 x86_init.timers.wallclock_init(); 1302 1303 /* 1304 * This needs to run before setup_local_APIC() which soft-disables the 1305 * local APIC temporarily and that masks the thermal LVT interrupt, 1306 * leading to softlockups on machines which have configured SMI 1307 * interrupt delivery. 1308 */ 1309 therm_lvt_init(); 1310 1311 mcheck_init(); 1312 1313 register_refined_jiffies(CLOCK_TICK_RATE); 1314 1315 #ifdef CONFIG_EFI 1316 if (efi_enabled(EFI_BOOT)) 1317 efi_apply_memmap_quirks(); 1318 #endif 1319 1320 unwind_init(); 1321 } 1322 1323 #ifdef CONFIG_X86_32 1324 1325 static struct resource video_ram_resource = { 1326 .name = "Video RAM area", 1327 .start = 0xa0000, 1328 .end = 0xbffff, 1329 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1330 }; 1331 1332 void __init i386_reserve_resources(void) 1333 { 1334 request_resource(&iomem_resource, &video_ram_resource); 1335 reserve_standard_io_resources(); 1336 } 1337 1338 #endif /* CONFIG_X86_32 */ 1339 1340 static struct notifier_block kernel_offset_notifier = { 1341 .notifier_call = dump_kernel_offset 1342 }; 1343 1344 static int __init register_kernel_offset_dumper(void) 1345 { 1346 atomic_notifier_chain_register(&panic_notifier_list, 1347 &kernel_offset_notifier); 1348 return 0; 1349 } 1350 __initcall(register_kernel_offset_dumper); 1351