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