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/init_ohci1394_dma.h> 15 #include <linux/initrd.h> 16 #include <linux/iscsi_ibft.h> 17 #include <linux/memblock.h> 18 #include <linux/panic_notifier.h> 19 #include <linux/pci.h> 20 #include <linux/root_dev.h> 21 #include <linux/hugetlb.h> 22 #include <linux/tboot.h> 23 #include <linux/usb/xhci-dbgp.h> 24 #include <linux/static_call.h> 25 #include <linux/swiotlb.h> 26 27 #include <uapi/linux/mount.h> 28 29 #include <xen/xen.h> 30 31 #include <asm/apic.h> 32 #include <asm/numa.h> 33 #include <asm/bios_ebda.h> 34 #include <asm/bugs.h> 35 #include <asm/cpu.h> 36 #include <asm/efi.h> 37 #include <asm/gart.h> 38 #include <asm/hypervisor.h> 39 #include <asm/io_apic.h> 40 #include <asm/kasan.h> 41 #include <asm/kaslr.h> 42 #include <asm/mce.h> 43 #include <asm/memtype.h> 44 #include <asm/mtrr.h> 45 #include <asm/realmode.h> 46 #include <asm/olpc_ofw.h> 47 #include <asm/pci-direct.h> 48 #include <asm/prom.h> 49 #include <asm/proto.h> 50 #include <asm/thermal.h> 51 #include <asm/unwind.h> 52 #include <asm/vsyscall.h> 53 #include <linux/vmalloc.h> 54 55 /* 56 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 57 * max_pfn_mapped: highest directly mapped pfn > 4 GB 58 * 59 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 60 * represented by pfn_mapped[]. 61 */ 62 unsigned long max_low_pfn_mapped; 63 unsigned long max_pfn_mapped; 64 65 #ifdef CONFIG_DMI 66 RESERVE_BRK(dmi_alloc, 65536); 67 #endif 68 69 70 /* 71 * Range of the BSS area. The size of the BSS area is determined 72 * at link time, with RESERVE_BRK() facility reserving additional 73 * chunks. 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 /* Common CPU data for all CPUs */ 120 struct cpuinfo_x86 boot_cpu_data __read_mostly; 121 EXPORT_SYMBOL(boot_cpu_data); 122 123 unsigned int def_to_bigsmp; 124 125 struct apm_info apm_info; 126 EXPORT_SYMBOL(apm_info); 127 128 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 129 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 130 struct ist_info ist_info; 131 EXPORT_SYMBOL(ist_info); 132 #else 133 struct ist_info ist_info; 134 #endif 135 136 #else 137 struct cpuinfo_x86 boot_cpu_data __read_mostly; 138 EXPORT_SYMBOL(boot_cpu_data); 139 #endif 140 141 142 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 143 __visible unsigned long mmu_cr4_features __ro_after_init; 144 #else 145 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 146 #endif 147 148 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 149 int bootloader_type, bootloader_version; 150 151 /* 152 * Setup options 153 */ 154 struct screen_info screen_info; 155 EXPORT_SYMBOL(screen_info); 156 struct edid_info edid_info; 157 EXPORT_SYMBOL_GPL(edid_info); 158 159 extern int root_mountflags; 160 161 unsigned long saved_video_mode; 162 163 #define RAMDISK_IMAGE_START_MASK 0x07FF 164 #define RAMDISK_PROMPT_FLAG 0x8000 165 #define RAMDISK_LOAD_FLAG 0x4000 166 167 static char __initdata command_line[COMMAND_LINE_SIZE]; 168 #ifdef CONFIG_CMDLINE_BOOL 169 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 170 #endif 171 172 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 173 struct edd edd; 174 #ifdef CONFIG_EDD_MODULE 175 EXPORT_SYMBOL(edd); 176 #endif 177 /** 178 * copy_edd() - Copy the BIOS EDD information 179 * from boot_params into a safe place. 180 * 181 */ 182 static inline void __init copy_edd(void) 183 { 184 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 185 sizeof(edd.mbr_signature)); 186 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 187 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 188 edd.edd_info_nr = boot_params.eddbuf_entries; 189 } 190 #else 191 static inline void __init copy_edd(void) 192 { 193 } 194 #endif 195 196 void * __init extend_brk(size_t size, size_t align) 197 { 198 size_t mask = align - 1; 199 void *ret; 200 201 BUG_ON(_brk_start == 0); 202 BUG_ON(align & mask); 203 204 _brk_end = (_brk_end + mask) & ~mask; 205 BUG_ON((char *)(_brk_end + size) > __brk_limit); 206 207 ret = (void *)_brk_end; 208 _brk_end += size; 209 210 memset(ret, 0, size); 211 212 return ret; 213 } 214 215 #ifdef CONFIG_X86_32 216 static void __init cleanup_highmap(void) 217 { 218 } 219 #endif 220 221 static void __init reserve_brk(void) 222 { 223 if (_brk_end > _brk_start) 224 memblock_reserve(__pa_symbol(_brk_start), 225 _brk_end - _brk_start); 226 227 /* Mark brk area as locked down and no longer taking any 228 new allocations */ 229 _brk_start = 0; 230 } 231 232 u64 relocated_ramdisk; 233 234 #ifdef CONFIG_BLK_DEV_INITRD 235 236 static u64 __init get_ramdisk_image(void) 237 { 238 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 239 240 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 241 242 if (ramdisk_image == 0) 243 ramdisk_image = phys_initrd_start; 244 245 return ramdisk_image; 246 } 247 static u64 __init get_ramdisk_size(void) 248 { 249 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 250 251 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 252 253 if (ramdisk_size == 0) 254 ramdisk_size = phys_initrd_size; 255 256 return ramdisk_size; 257 } 258 259 static void __init relocate_initrd(void) 260 { 261 /* Assume only end is not page aligned */ 262 u64 ramdisk_image = get_ramdisk_image(); 263 u64 ramdisk_size = get_ramdisk_size(); 264 u64 area_size = PAGE_ALIGN(ramdisk_size); 265 266 /* We need to move the initrd down into directly mapped mem */ 267 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 268 PFN_PHYS(max_pfn_mapped)); 269 if (!relocated_ramdisk) 270 panic("Cannot find place for new RAMDISK of size %lld\n", 271 ramdisk_size); 272 273 initrd_start = relocated_ramdisk + PAGE_OFFSET; 274 initrd_end = initrd_start + ramdisk_size; 275 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 276 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 277 278 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 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 parse_setup_data(void) 339 { 340 struct setup_data *data; 341 u64 pa_data, pa_next; 342 343 pa_data = boot_params.hdr.setup_data; 344 while (pa_data) { 345 u32 data_len, data_type; 346 347 data = early_memremap(pa_data, sizeof(*data)); 348 data_len = data->len + sizeof(struct setup_data); 349 data_type = data->type; 350 pa_next = data->next; 351 early_memunmap(data, sizeof(*data)); 352 353 switch (data_type) { 354 case SETUP_E820_EXT: 355 e820__memory_setup_extended(pa_data, data_len); 356 break; 357 case SETUP_DTB: 358 add_dtb(pa_data); 359 break; 360 case SETUP_EFI: 361 parse_efi_setup(pa_data, data_len); 362 break; 363 default: 364 break; 365 } 366 pa_data = pa_next; 367 } 368 } 369 370 static void __init memblock_x86_reserve_range_setup_data(void) 371 { 372 struct setup_indirect *indirect; 373 struct setup_data *data; 374 u64 pa_data, pa_next; 375 u32 len; 376 377 pa_data = boot_params.hdr.setup_data; 378 while (pa_data) { 379 data = early_memremap(pa_data, sizeof(*data)); 380 if (!data) { 381 pr_warn("setup: failed to memremap setup_data entry\n"); 382 return; 383 } 384 385 len = sizeof(*data); 386 pa_next = data->next; 387 388 memblock_reserve(pa_data, sizeof(*data) + data->len); 389 390 if (data->type == SETUP_INDIRECT) { 391 len += data->len; 392 early_memunmap(data, sizeof(*data)); 393 data = early_memremap(pa_data, len); 394 if (!data) { 395 pr_warn("setup: failed to memremap indirect setup_data\n"); 396 return; 397 } 398 399 indirect = (struct setup_indirect *)data->data; 400 401 if (indirect->type != SETUP_INDIRECT) 402 memblock_reserve(indirect->addr, indirect->len); 403 } 404 405 pa_data = pa_next; 406 early_memunmap(data, len); 407 } 408 } 409 410 /* 411 * --------- Crashkernel reservation ------------------------------ 412 */ 413 414 #ifdef CONFIG_KEXEC_CORE 415 416 /* 16M alignment for crash kernel regions */ 417 #define CRASH_ALIGN SZ_16M 418 419 /* 420 * Keep the crash kernel below this limit. 421 * 422 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 423 * due to mapping restrictions. 424 * 425 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 426 * the upper limit of system RAM in 4-level paging mode. Since the kdump 427 * jump could be from 5-level paging to 4-level paging, the jump will fail if 428 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 429 * no good way to detect the paging mode of the target kernel which will be 430 * loaded for dumping. 431 */ 432 #ifdef CONFIG_X86_32 433 # define CRASH_ADDR_LOW_MAX SZ_512M 434 # define CRASH_ADDR_HIGH_MAX SZ_512M 435 #else 436 # define CRASH_ADDR_LOW_MAX SZ_4G 437 # define CRASH_ADDR_HIGH_MAX SZ_64T 438 #endif 439 440 static int __init reserve_crashkernel_low(void) 441 { 442 #ifdef CONFIG_X86_64 443 unsigned long long base, low_base = 0, low_size = 0; 444 unsigned long low_mem_limit; 445 int ret; 446 447 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 448 449 /* crashkernel=Y,low */ 450 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 451 if (ret) { 452 /* 453 * two parts from kernel/dma/swiotlb.c: 454 * -swiotlb size: user-specified with swiotlb= or default. 455 * 456 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 457 * to 8M for other buffers that may need to stay low too. Also 458 * make sure we allocate enough extra low memory so that we 459 * don't run out of DMA buffers for 32-bit devices. 460 */ 461 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 462 } else { 463 /* passed with crashkernel=0,low ? */ 464 if (!low_size) 465 return 0; 466 } 467 468 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 469 if (!low_base) { 470 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 471 (unsigned long)(low_size >> 20)); 472 return -ENOMEM; 473 } 474 475 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 476 (unsigned long)(low_size >> 20), 477 (unsigned long)(low_base >> 20), 478 (unsigned long)(low_mem_limit >> 20)); 479 480 crashk_low_res.start = low_base; 481 crashk_low_res.end = low_base + low_size - 1; 482 insert_resource(&iomem_resource, &crashk_low_res); 483 #endif 484 return 0; 485 } 486 487 static void __init reserve_crashkernel(void) 488 { 489 unsigned long long crash_size, crash_base, total_mem; 490 bool high = false; 491 int ret; 492 493 total_mem = memblock_phys_mem_size(); 494 495 /* crashkernel=XM */ 496 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 497 if (ret != 0 || crash_size <= 0) { 498 /* crashkernel=X,high */ 499 ret = parse_crashkernel_high(boot_command_line, total_mem, 500 &crash_size, &crash_base); 501 if (ret != 0 || crash_size <= 0) 502 return; 503 high = true; 504 } 505 506 if (xen_pv_domain()) { 507 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 508 return; 509 } 510 511 /* 0 means: find the address automatically */ 512 if (!crash_base) { 513 /* 514 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 515 * crashkernel=x,high reserves memory over 4G, also allocates 516 * 256M extra low memory for DMA buffers and swiotlb. 517 * But the extra memory is not required for all machines. 518 * So try low memory first and fall back to high memory 519 * unless "crashkernel=size[KMG],high" is specified. 520 */ 521 if (!high) 522 crash_base = memblock_phys_alloc_range(crash_size, 523 CRASH_ALIGN, CRASH_ALIGN, 524 CRASH_ADDR_LOW_MAX); 525 if (!crash_base) 526 crash_base = memblock_phys_alloc_range(crash_size, 527 CRASH_ALIGN, CRASH_ALIGN, 528 CRASH_ADDR_HIGH_MAX); 529 if (!crash_base) { 530 pr_info("crashkernel reservation failed - No suitable area found.\n"); 531 return; 532 } 533 } else { 534 unsigned long long start; 535 536 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 537 crash_base + crash_size); 538 if (start != crash_base) { 539 pr_info("crashkernel reservation failed - memory is in use.\n"); 540 return; 541 } 542 } 543 544 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 545 memblock_phys_free(crash_base, crash_size); 546 return; 547 } 548 549 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 550 (unsigned long)(crash_size >> 20), 551 (unsigned long)(crash_base >> 20), 552 (unsigned long)(total_mem >> 20)); 553 554 crashk_res.start = crash_base; 555 crashk_res.end = crash_base + crash_size - 1; 556 insert_resource(&iomem_resource, &crashk_res); 557 } 558 #else 559 static void __init reserve_crashkernel(void) 560 { 561 } 562 #endif 563 564 static struct resource standard_io_resources[] = { 565 { .name = "dma1", .start = 0x00, .end = 0x1f, 566 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 567 { .name = "pic1", .start = 0x20, .end = 0x21, 568 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 569 { .name = "timer0", .start = 0x40, .end = 0x43, 570 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 571 { .name = "timer1", .start = 0x50, .end = 0x53, 572 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 573 { .name = "keyboard", .start = 0x60, .end = 0x60, 574 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 575 { .name = "keyboard", .start = 0x64, .end = 0x64, 576 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 577 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 578 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 579 { .name = "pic2", .start = 0xa0, .end = 0xa1, 580 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 581 { .name = "dma2", .start = 0xc0, .end = 0xdf, 582 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 583 { .name = "fpu", .start = 0xf0, .end = 0xff, 584 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 585 }; 586 587 void __init reserve_standard_io_resources(void) 588 { 589 int i; 590 591 /* request I/O space for devices used on all i[345]86 PCs */ 592 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 593 request_resource(&ioport_resource, &standard_io_resources[i]); 594 595 } 596 597 static bool __init snb_gfx_workaround_needed(void) 598 { 599 #ifdef CONFIG_PCI 600 int i; 601 u16 vendor, devid; 602 static const __initconst u16 snb_ids[] = { 603 0x0102, 604 0x0112, 605 0x0122, 606 0x0106, 607 0x0116, 608 0x0126, 609 0x010a, 610 }; 611 612 /* Assume no if something weird is going on with PCI */ 613 if (!early_pci_allowed()) 614 return false; 615 616 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 617 if (vendor != 0x8086) 618 return false; 619 620 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 621 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 622 if (devid == snb_ids[i]) 623 return true; 624 #endif 625 626 return false; 627 } 628 629 /* 630 * Sandy Bridge graphics has trouble with certain ranges, exclude 631 * them from allocation. 632 */ 633 static void __init trim_snb_memory(void) 634 { 635 static const __initconst unsigned long bad_pages[] = { 636 0x20050000, 637 0x20110000, 638 0x20130000, 639 0x20138000, 640 0x40004000, 641 }; 642 int i; 643 644 if (!snb_gfx_workaround_needed()) 645 return; 646 647 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 648 649 /* 650 * SandyBridge integrated graphics devices have a bug that prevents 651 * them from accessing certain memory ranges, namely anything below 652 * 1M and in the pages listed in bad_pages[] above. 653 * 654 * To avoid these pages being ever accessed by SNB gfx devices reserve 655 * bad_pages that have not already been reserved at boot time. 656 * All memory below the 1 MB mark is anyway reserved later during 657 * setup_arch(), so there is no need to reserve it here. 658 */ 659 660 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 661 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 662 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 663 bad_pages[i]); 664 } 665 } 666 667 static void __init trim_bios_range(void) 668 { 669 /* 670 * A special case is the first 4Kb of memory; 671 * This is a BIOS owned area, not kernel ram, but generally 672 * not listed as such in the E820 table. 673 * 674 * This typically reserves additional memory (64KiB by default) 675 * since some BIOSes are known to corrupt low memory. See the 676 * Kconfig help text for X86_RESERVE_LOW. 677 */ 678 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 679 680 /* 681 * special case: Some BIOSes report the PC BIOS 682 * area (640Kb -> 1Mb) as RAM even though it is not. 683 * take them out. 684 */ 685 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 686 687 e820__update_table(e820_table); 688 } 689 690 /* called before trim_bios_range() to spare extra sanitize */ 691 static void __init e820_add_kernel_range(void) 692 { 693 u64 start = __pa_symbol(_text); 694 u64 size = __pa_symbol(_end) - start; 695 696 /* 697 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 698 * attempt to fix it by adding the range. We may have a confused BIOS, 699 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 700 * exclude kernel range. If we really are running on top non-RAM, 701 * we will crash later anyways. 702 */ 703 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 704 return; 705 706 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 707 e820__range_remove(start, size, E820_TYPE_RAM, 0); 708 e820__range_add(start, size, E820_TYPE_RAM); 709 } 710 711 static void __init early_reserve_memory(void) 712 { 713 /* 714 * Reserve the memory occupied by the kernel between _text and 715 * __end_of_kernel_reserve symbols. Any kernel sections after the 716 * __end_of_kernel_reserve symbol must be explicitly reserved with a 717 * separate memblock_reserve() or they will be discarded. 718 */ 719 memblock_reserve(__pa_symbol(_text), 720 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 721 722 /* 723 * The first 4Kb of memory is a BIOS owned area, but generally it is 724 * not listed as such in the E820 table. 725 * 726 * Reserve the first 64K of memory since some BIOSes are known to 727 * corrupt low memory. After the real mode trampoline is allocated the 728 * rest of the memory below 640k is reserved. 729 * 730 * In addition, make sure page 0 is always reserved because on 731 * systems with L1TF its contents can be leaked to user processes. 732 */ 733 memblock_reserve(0, SZ_64K); 734 735 early_reserve_initrd(); 736 737 memblock_x86_reserve_range_setup_data(); 738 739 reserve_ibft_region(); 740 reserve_bios_regions(); 741 trim_snb_memory(); 742 } 743 744 /* 745 * Dump out kernel offset information on panic. 746 */ 747 static int 748 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 749 { 750 if (kaslr_enabled()) { 751 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 752 kaslr_offset(), 753 __START_KERNEL, 754 __START_KERNEL_map, 755 MODULES_VADDR-1); 756 } else { 757 pr_emerg("Kernel Offset: disabled\n"); 758 } 759 760 return 0; 761 } 762 763 /* 764 * Determine if we were loaded by an EFI loader. If so, then we have also been 765 * passed the efi memmap, systab, etc., so we should use these data structures 766 * for initialization. Note, the efi init code path is determined by the 767 * global efi_enabled. This allows the same kernel image to be used on existing 768 * systems (with a traditional BIOS) as well as on EFI systems. 769 */ 770 /* 771 * setup_arch - architecture-specific boot-time initializations 772 * 773 * Note: On x86_64, fixmaps are ready for use even before this is called. 774 */ 775 776 void __init setup_arch(char **cmdline_p) 777 { 778 #ifdef CONFIG_X86_32 779 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 780 781 /* 782 * copy kernel address range established so far and switch 783 * to the proper swapper page table 784 */ 785 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 786 initial_page_table + KERNEL_PGD_BOUNDARY, 787 KERNEL_PGD_PTRS); 788 789 load_cr3(swapper_pg_dir); 790 /* 791 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 792 * a cr3 based tlb flush, so the following __flush_tlb_all() 793 * will not flush anything because the CPU quirk which clears 794 * X86_FEATURE_PGE has not been invoked yet. Though due to the 795 * load_cr3() above the TLB has been flushed already. The 796 * quirk is invoked before subsequent calls to __flush_tlb_all() 797 * so proper operation is guaranteed. 798 */ 799 __flush_tlb_all(); 800 #else 801 printk(KERN_INFO "Command line: %s\n", boot_command_line); 802 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 803 #endif 804 805 /* 806 * If we have OLPC OFW, we might end up relocating the fixmap due to 807 * reserve_top(), so do this before touching the ioremap area. 808 */ 809 olpc_ofw_detect(); 810 811 idt_setup_early_traps(); 812 early_cpu_init(); 813 jump_label_init(); 814 static_call_init(); 815 early_ioremap_init(); 816 817 setup_olpc_ofw_pgd(); 818 819 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 820 screen_info = boot_params.screen_info; 821 edid_info = boot_params.edid_info; 822 #ifdef CONFIG_X86_32 823 apm_info.bios = boot_params.apm_bios_info; 824 ist_info = boot_params.ist_info; 825 #endif 826 saved_video_mode = boot_params.hdr.vid_mode; 827 bootloader_type = boot_params.hdr.type_of_loader; 828 if ((bootloader_type >> 4) == 0xe) { 829 bootloader_type &= 0xf; 830 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 831 } 832 bootloader_version = bootloader_type & 0xf; 833 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 834 835 #ifdef CONFIG_BLK_DEV_RAM 836 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 837 #endif 838 #ifdef CONFIG_EFI 839 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 840 EFI32_LOADER_SIGNATURE, 4)) { 841 set_bit(EFI_BOOT, &efi.flags); 842 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 843 EFI64_LOADER_SIGNATURE, 4)) { 844 set_bit(EFI_BOOT, &efi.flags); 845 set_bit(EFI_64BIT, &efi.flags); 846 } 847 #endif 848 849 x86_init.oem.arch_setup(); 850 851 /* 852 * Do some memory reservations *before* memory is added to memblock, so 853 * memblock allocations won't overwrite it. 854 * 855 * After this point, everything still needed from the boot loader or 856 * firmware or kernel text should be early reserved or marked not RAM in 857 * e820. All other memory is free game. 858 * 859 * This call needs to happen before e820__memory_setup() which calls the 860 * xen_memory_setup() on Xen dom0 which relies on the fact that those 861 * early reservations have happened already. 862 */ 863 early_reserve_memory(); 864 865 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 866 e820__memory_setup(); 867 parse_setup_data(); 868 869 copy_edd(); 870 871 if (!boot_params.hdr.root_flags) 872 root_mountflags &= ~MS_RDONLY; 873 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 874 875 code_resource.start = __pa_symbol(_text); 876 code_resource.end = __pa_symbol(_etext)-1; 877 rodata_resource.start = __pa_symbol(__start_rodata); 878 rodata_resource.end = __pa_symbol(__end_rodata)-1; 879 data_resource.start = __pa_symbol(_sdata); 880 data_resource.end = __pa_symbol(_edata)-1; 881 bss_resource.start = __pa_symbol(__bss_start); 882 bss_resource.end = __pa_symbol(__bss_stop)-1; 883 884 #ifdef CONFIG_CMDLINE_BOOL 885 #ifdef CONFIG_CMDLINE_OVERRIDE 886 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 887 #else 888 if (builtin_cmdline[0]) { 889 /* append boot loader cmdline to builtin */ 890 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 891 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 892 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 893 } 894 #endif 895 #endif 896 897 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 898 *cmdline_p = command_line; 899 900 /* 901 * x86_configure_nx() is called before parse_early_param() to detect 902 * whether hardware doesn't support NX (so that the early EHCI debug 903 * console setup can safely call set_fixmap()). It may then be called 904 * again from within noexec_setup() during parsing early parameters 905 * to honor the respective command line option. 906 */ 907 x86_configure_nx(); 908 909 parse_early_param(); 910 911 if (efi_enabled(EFI_BOOT)) 912 efi_memblock_x86_reserve_range(); 913 914 #ifdef CONFIG_MEMORY_HOTPLUG 915 /* 916 * Memory used by the kernel cannot be hot-removed because Linux 917 * cannot migrate the kernel pages. When memory hotplug is 918 * enabled, we should prevent memblock from allocating memory 919 * for the kernel. 920 * 921 * ACPI SRAT records all hotpluggable memory ranges. But before 922 * SRAT is parsed, we don't know about it. 923 * 924 * The kernel image is loaded into memory at very early time. We 925 * cannot prevent this anyway. So on NUMA system, we set any 926 * node the kernel resides in as un-hotpluggable. 927 * 928 * Since on modern servers, one node could have double-digit 929 * gigabytes memory, we can assume the memory around the kernel 930 * image is also un-hotpluggable. So before SRAT is parsed, just 931 * allocate memory near the kernel image to try the best to keep 932 * the kernel away from hotpluggable memory. 933 */ 934 if (movable_node_is_enabled()) 935 memblock_set_bottom_up(true); 936 #endif 937 938 x86_report_nx(); 939 940 if (acpi_mps_check()) { 941 #ifdef CONFIG_X86_LOCAL_APIC 942 disable_apic = 1; 943 #endif 944 setup_clear_cpu_cap(X86_FEATURE_APIC); 945 } 946 947 e820__reserve_setup_data(); 948 e820__finish_early_params(); 949 950 if (efi_enabled(EFI_BOOT)) 951 efi_init(); 952 953 dmi_setup(); 954 955 /* 956 * VMware detection requires dmi to be available, so this 957 * needs to be done after dmi_setup(), for the boot CPU. 958 */ 959 init_hypervisor_platform(); 960 961 tsc_early_init(); 962 x86_init.resources.probe_roms(); 963 964 /* after parse_early_param, so could debug it */ 965 insert_resource(&iomem_resource, &code_resource); 966 insert_resource(&iomem_resource, &rodata_resource); 967 insert_resource(&iomem_resource, &data_resource); 968 insert_resource(&iomem_resource, &bss_resource); 969 970 e820_add_kernel_range(); 971 trim_bios_range(); 972 #ifdef CONFIG_X86_32 973 if (ppro_with_ram_bug()) { 974 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 975 E820_TYPE_RESERVED); 976 e820__update_table(e820_table); 977 printk(KERN_INFO "fixed physical RAM map:\n"); 978 e820__print_table("bad_ppro"); 979 } 980 #else 981 early_gart_iommu_check(); 982 #endif 983 984 /* 985 * partially used pages are not usable - thus 986 * we are rounding upwards: 987 */ 988 max_pfn = e820__end_of_ram_pfn(); 989 990 /* update e820 for memory not covered by WB MTRRs */ 991 if (IS_ENABLED(CONFIG_MTRR)) 992 mtrr_bp_init(); 993 else 994 pat_disable("PAT support disabled because CONFIG_MTRR is disabled in the kernel."); 995 996 if (mtrr_trim_uncached_memory(max_pfn)) 997 max_pfn = e820__end_of_ram_pfn(); 998 999 max_possible_pfn = max_pfn; 1000 1001 /* 1002 * This call is required when the CPU does not support PAT. If 1003 * mtrr_bp_init() invoked it already via pat_init() the call has no 1004 * effect. 1005 */ 1006 init_cache_modes(); 1007 1008 /* 1009 * Define random base addresses for memory sections after max_pfn is 1010 * defined and before each memory section base is used. 1011 */ 1012 kernel_randomize_memory(); 1013 1014 #ifdef CONFIG_X86_32 1015 /* max_low_pfn get updated here */ 1016 find_low_pfn_range(); 1017 #else 1018 check_x2apic(); 1019 1020 /* How many end-of-memory variables you have, grandma! */ 1021 /* need this before calling reserve_initrd */ 1022 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1023 max_low_pfn = e820__end_of_low_ram_pfn(); 1024 else 1025 max_low_pfn = max_pfn; 1026 1027 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1028 #endif 1029 1030 /* 1031 * Find and reserve possible boot-time SMP configuration: 1032 */ 1033 find_smp_config(); 1034 1035 early_alloc_pgt_buf(); 1036 1037 /* 1038 * Need to conclude brk, before e820__memblock_setup() 1039 * it could use memblock_find_in_range, could overlap with 1040 * brk area. 1041 */ 1042 reserve_brk(); 1043 1044 cleanup_highmap(); 1045 1046 memblock_set_current_limit(ISA_END_ADDRESS); 1047 e820__memblock_setup(); 1048 1049 /* 1050 * Needs to run after memblock setup because it needs the physical 1051 * memory size. 1052 */ 1053 sev_setup_arch(); 1054 1055 efi_fake_memmap(); 1056 efi_find_mirror(); 1057 efi_esrt_init(); 1058 efi_mokvar_table_init(); 1059 1060 /* 1061 * The EFI specification says that boot service code won't be 1062 * called after ExitBootServices(). This is, in fact, a lie. 1063 */ 1064 efi_reserve_boot_services(); 1065 1066 /* preallocate 4k for mptable mpc */ 1067 e820__memblock_alloc_reserved_mpc_new(); 1068 1069 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1070 setup_bios_corruption_check(); 1071 #endif 1072 1073 #ifdef CONFIG_X86_32 1074 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1075 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1076 #endif 1077 1078 /* 1079 * Find free memory for the real mode trampoline and place it there. If 1080 * there is not enough free memory under 1M, on EFI-enabled systems 1081 * there will be additional attempt to reclaim the memory for the real 1082 * mode trampoline at efi_free_boot_services(). 1083 * 1084 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1085 * are known to corrupt low memory and several hundred kilobytes are not 1086 * worth complex detection what memory gets clobbered. Windows does the 1087 * same thing for very similar reasons. 1088 * 1089 * Moreover, on machines with SandyBridge graphics or in setups that use 1090 * crashkernel the entire 1M is reserved anyway. 1091 */ 1092 reserve_real_mode(); 1093 1094 init_mem_mapping(); 1095 1096 idt_setup_early_pf(); 1097 1098 /* 1099 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1100 * with the current CR4 value. This may not be necessary, but 1101 * auditing all the early-boot CR4 manipulation would be needed to 1102 * rule it out. 1103 * 1104 * Mask off features that don't work outside long mode (just 1105 * PCIDE for now). 1106 */ 1107 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1108 1109 memblock_set_current_limit(get_max_mapped()); 1110 1111 /* 1112 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1113 */ 1114 1115 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1116 if (init_ohci1394_dma_early) 1117 init_ohci1394_dma_on_all_controllers(); 1118 #endif 1119 /* Allocate bigger log buffer */ 1120 setup_log_buf(1); 1121 1122 if (efi_enabled(EFI_BOOT)) { 1123 switch (boot_params.secure_boot) { 1124 case efi_secureboot_mode_disabled: 1125 pr_info("Secure boot disabled\n"); 1126 break; 1127 case efi_secureboot_mode_enabled: 1128 pr_info("Secure boot enabled\n"); 1129 break; 1130 default: 1131 pr_info("Secure boot could not be determined\n"); 1132 break; 1133 } 1134 } 1135 1136 reserve_initrd(); 1137 1138 acpi_table_upgrade(); 1139 /* Look for ACPI tables and reserve memory occupied by them. */ 1140 acpi_boot_table_init(); 1141 1142 vsmp_init(); 1143 1144 io_delay_init(); 1145 1146 early_platform_quirks(); 1147 1148 early_acpi_boot_init(); 1149 1150 initmem_init(); 1151 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1152 1153 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1154 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1155 1156 /* 1157 * Reserve memory for crash kernel after SRAT is parsed so that it 1158 * won't consume hotpluggable memory. 1159 */ 1160 reserve_crashkernel(); 1161 1162 memblock_find_dma_reserve(); 1163 1164 if (!early_xdbc_setup_hardware()) 1165 early_xdbc_register_console(); 1166 1167 x86_init.paging.pagetable_init(); 1168 1169 kasan_init(); 1170 1171 /* 1172 * Sync back kernel address range. 1173 * 1174 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1175 * this call? 1176 */ 1177 sync_initial_page_table(); 1178 1179 tboot_probe(); 1180 1181 map_vsyscall(); 1182 1183 generic_apic_probe(); 1184 1185 early_quirks(); 1186 1187 /* 1188 * Read APIC and some other early information from ACPI tables. 1189 */ 1190 acpi_boot_init(); 1191 x86_dtb_init(); 1192 1193 /* 1194 * get boot-time SMP configuration: 1195 */ 1196 get_smp_config(); 1197 1198 /* 1199 * Systems w/o ACPI and mptables might not have it mapped the local 1200 * APIC yet, but prefill_possible_map() might need to access it. 1201 */ 1202 init_apic_mappings(); 1203 1204 prefill_possible_map(); 1205 1206 init_cpu_to_node(); 1207 init_gi_nodes(); 1208 1209 io_apic_init_mappings(); 1210 1211 x86_init.hyper.guest_late_init(); 1212 1213 e820__reserve_resources(); 1214 e820__register_nosave_regions(max_pfn); 1215 1216 x86_init.resources.reserve_resources(); 1217 1218 e820__setup_pci_gap(); 1219 1220 #ifdef CONFIG_VT 1221 #if defined(CONFIG_VGA_CONSOLE) 1222 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1223 conswitchp = &vga_con; 1224 #endif 1225 #endif 1226 x86_init.oem.banner(); 1227 1228 x86_init.timers.wallclock_init(); 1229 1230 /* 1231 * This needs to run before setup_local_APIC() which soft-disables the 1232 * local APIC temporarily and that masks the thermal LVT interrupt, 1233 * leading to softlockups on machines which have configured SMI 1234 * interrupt delivery. 1235 */ 1236 therm_lvt_init(); 1237 1238 mcheck_init(); 1239 1240 register_refined_jiffies(CLOCK_TICK_RATE); 1241 1242 #ifdef CONFIG_EFI 1243 if (efi_enabled(EFI_BOOT)) 1244 efi_apply_memmap_quirks(); 1245 #endif 1246 1247 unwind_init(); 1248 } 1249 1250 #ifdef CONFIG_X86_32 1251 1252 static struct resource video_ram_resource = { 1253 .name = "Video RAM area", 1254 .start = 0xa0000, 1255 .end = 0xbffff, 1256 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1257 }; 1258 1259 void __init i386_reserve_resources(void) 1260 { 1261 request_resource(&iomem_resource, &video_ram_resource); 1262 reserve_standard_io_resources(); 1263 } 1264 1265 #endif /* CONFIG_X86_32 */ 1266 1267 static struct notifier_block kernel_offset_notifier = { 1268 .notifier_call = dump_kernel_offset 1269 }; 1270 1271 static int __init register_kernel_offset_dumper(void) 1272 { 1273 atomic_notifier_chain_register(&panic_notifier_list, 1274 &kernel_offset_notifier); 1275 return 0; 1276 } 1277 __initcall(register_kernel_offset_dumper); 1278