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/console.h> 9 #include <linux/crash_dump.h> 10 #include <linux/dma-map-ops.h> 11 #include <linux/dmi.h> 12 #include <linux/efi.h> 13 #include <linux/init_ohci1394_dma.h> 14 #include <linux/initrd.h> 15 #include <linux/iscsi_ibft.h> 16 #include <linux/memblock.h> 17 #include <linux/panic_notifier.h> 18 #include <linux/pci.h> 19 #include <linux/root_dev.h> 20 #include <linux/hugetlb.h> 21 #include <linux/tboot.h> 22 #include <linux/usb/xhci-dbgp.h> 23 #include <linux/static_call.h> 24 #include <linux/swiotlb.h> 25 26 #include <uapi/linux/mount.h> 27 28 #include <xen/xen.h> 29 30 #include <asm/apic.h> 31 #include <asm/numa.h> 32 #include <asm/bios_ebda.h> 33 #include <asm/bugs.h> 34 #include <asm/cpu.h> 35 #include <asm/efi.h> 36 #include <asm/gart.h> 37 #include <asm/hypervisor.h> 38 #include <asm/io_apic.h> 39 #include <asm/kasan.h> 40 #include <asm/kaslr.h> 41 #include <asm/mce.h> 42 #include <asm/mtrr.h> 43 #include <asm/realmode.h> 44 #include <asm/olpc_ofw.h> 45 #include <asm/pci-direct.h> 46 #include <asm/prom.h> 47 #include <asm/proto.h> 48 #include <asm/thermal.h> 49 #include <asm/unwind.h> 50 #include <asm/vsyscall.h> 51 #include <linux/vmalloc.h> 52 53 /* 54 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 55 * max_pfn_mapped: highest directly mapped pfn > 4 GB 56 * 57 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 58 * represented by pfn_mapped[]. 59 */ 60 unsigned long max_low_pfn_mapped; 61 unsigned long max_pfn_mapped; 62 63 #ifdef CONFIG_DMI 64 RESERVE_BRK(dmi_alloc, 65536); 65 #endif 66 67 68 /* 69 * Range of the BSS area. The size of the BSS area is determined 70 * at link time, with RESERVE_BRK() facility reserving additional 71 * chunks. 72 */ 73 unsigned long _brk_start = (unsigned long)__brk_base; 74 unsigned long _brk_end = (unsigned long)__brk_base; 75 76 struct boot_params boot_params; 77 78 /* 79 * These are the four main kernel memory regions, we put them into 80 * the resource tree so that kdump tools and other debugging tools 81 * recover it: 82 */ 83 84 static struct resource rodata_resource = { 85 .name = "Kernel rodata", 86 .start = 0, 87 .end = 0, 88 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 89 }; 90 91 static struct resource data_resource = { 92 .name = "Kernel data", 93 .start = 0, 94 .end = 0, 95 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 96 }; 97 98 static struct resource code_resource = { 99 .name = "Kernel code", 100 .start = 0, 101 .end = 0, 102 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 103 }; 104 105 static struct resource bss_resource = { 106 .name = "Kernel bss", 107 .start = 0, 108 .end = 0, 109 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 110 }; 111 112 113 #ifdef CONFIG_X86_32 114 /* CPU data as detected by the assembly code in head_32.S */ 115 struct cpuinfo_x86 new_cpu_data; 116 117 /* Common CPU data for all CPUs */ 118 struct cpuinfo_x86 boot_cpu_data __read_mostly; 119 EXPORT_SYMBOL(boot_cpu_data); 120 121 unsigned int def_to_bigsmp; 122 123 struct apm_info apm_info; 124 EXPORT_SYMBOL(apm_info); 125 126 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 127 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 128 struct ist_info ist_info; 129 EXPORT_SYMBOL(ist_info); 130 #else 131 struct ist_info ist_info; 132 #endif 133 134 #else 135 struct cpuinfo_x86 boot_cpu_data __read_mostly; 136 EXPORT_SYMBOL(boot_cpu_data); 137 #endif 138 139 140 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 141 __visible unsigned long mmu_cr4_features __ro_after_init; 142 #else 143 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 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_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 parse_setup_data(void) 337 { 338 struct setup_data *data; 339 u64 pa_data, pa_next; 340 341 pa_data = boot_params.hdr.setup_data; 342 while (pa_data) { 343 u32 data_len, data_type; 344 345 data = early_memremap(pa_data, sizeof(*data)); 346 data_len = data->len + sizeof(struct setup_data); 347 data_type = data->type; 348 pa_next = data->next; 349 early_memunmap(data, sizeof(*data)); 350 351 switch (data_type) { 352 case SETUP_E820_EXT: 353 e820__memory_setup_extended(pa_data, data_len); 354 break; 355 case SETUP_DTB: 356 add_dtb(pa_data); 357 break; 358 case SETUP_EFI: 359 parse_efi_setup(pa_data, data_len); 360 break; 361 default: 362 break; 363 } 364 pa_data = pa_next; 365 } 366 } 367 368 static void __init memblock_x86_reserve_range_setup_data(void) 369 { 370 struct setup_data *data; 371 u64 pa_data; 372 373 pa_data = boot_params.hdr.setup_data; 374 while (pa_data) { 375 data = early_memremap(pa_data, sizeof(*data)); 376 memblock_reserve(pa_data, sizeof(*data) + data->len); 377 378 if (data->type == SETUP_INDIRECT && 379 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) 380 memblock_reserve(((struct setup_indirect *)data->data)->addr, 381 ((struct setup_indirect *)data->data)->len); 382 383 pa_data = data->next; 384 early_memunmap(data, sizeof(*data)); 385 } 386 } 387 388 /* 389 * --------- Crashkernel reservation ------------------------------ 390 */ 391 392 #ifdef CONFIG_KEXEC_CORE 393 394 /* 16M alignment for crash kernel regions */ 395 #define CRASH_ALIGN SZ_16M 396 397 /* 398 * Keep the crash kernel below this limit. 399 * 400 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 401 * due to mapping restrictions. 402 * 403 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 404 * the upper limit of system RAM in 4-level paging mode. Since the kdump 405 * jump could be from 5-level paging to 4-level paging, the jump will fail if 406 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 407 * no good way to detect the paging mode of the target kernel which will be 408 * loaded for dumping. 409 */ 410 #ifdef CONFIG_X86_32 411 # define CRASH_ADDR_LOW_MAX SZ_512M 412 # define CRASH_ADDR_HIGH_MAX SZ_512M 413 #else 414 # define CRASH_ADDR_LOW_MAX SZ_4G 415 # define CRASH_ADDR_HIGH_MAX SZ_64T 416 #endif 417 418 static int __init reserve_crashkernel_low(void) 419 { 420 #ifdef CONFIG_X86_64 421 unsigned long long base, low_base = 0, low_size = 0; 422 unsigned long low_mem_limit; 423 int ret; 424 425 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 426 427 /* crashkernel=Y,low */ 428 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 429 if (ret) { 430 /* 431 * two parts from kernel/dma/swiotlb.c: 432 * -swiotlb size: user-specified with swiotlb= or default. 433 * 434 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 435 * to 8M for other buffers that may need to stay low too. Also 436 * make sure we allocate enough extra low memory so that we 437 * don't run out of DMA buffers for 32-bit devices. 438 */ 439 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 440 } else { 441 /* passed with crashkernel=0,low ? */ 442 if (!low_size) 443 return 0; 444 } 445 446 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 447 if (!low_base) { 448 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 449 (unsigned long)(low_size >> 20)); 450 return -ENOMEM; 451 } 452 453 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 454 (unsigned long)(low_size >> 20), 455 (unsigned long)(low_base >> 20), 456 (unsigned long)(low_mem_limit >> 20)); 457 458 crashk_low_res.start = low_base; 459 crashk_low_res.end = low_base + low_size - 1; 460 insert_resource(&iomem_resource, &crashk_low_res); 461 #endif 462 return 0; 463 } 464 465 static void __init reserve_crashkernel(void) 466 { 467 unsigned long long crash_size, crash_base, total_mem; 468 bool high = false; 469 int ret; 470 471 total_mem = memblock_phys_mem_size(); 472 473 /* crashkernel=XM */ 474 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 475 if (ret != 0 || crash_size <= 0) { 476 /* crashkernel=X,high */ 477 ret = parse_crashkernel_high(boot_command_line, total_mem, 478 &crash_size, &crash_base); 479 if (ret != 0 || crash_size <= 0) 480 return; 481 high = true; 482 } 483 484 if (xen_pv_domain()) { 485 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 486 return; 487 } 488 489 /* 0 means: find the address automatically */ 490 if (!crash_base) { 491 /* 492 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 493 * crashkernel=x,high reserves memory over 4G, also allocates 494 * 256M extra low memory for DMA buffers and swiotlb. 495 * But the extra memory is not required for all machines. 496 * So try low memory first and fall back to high memory 497 * unless "crashkernel=size[KMG],high" is specified. 498 */ 499 if (!high) 500 crash_base = memblock_phys_alloc_range(crash_size, 501 CRASH_ALIGN, CRASH_ALIGN, 502 CRASH_ADDR_LOW_MAX); 503 if (!crash_base) 504 crash_base = memblock_phys_alloc_range(crash_size, 505 CRASH_ALIGN, CRASH_ALIGN, 506 CRASH_ADDR_HIGH_MAX); 507 if (!crash_base) { 508 pr_info("crashkernel reservation failed - No suitable area found.\n"); 509 return; 510 } 511 } else { 512 unsigned long long start; 513 514 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 515 crash_base + crash_size); 516 if (start != crash_base) { 517 pr_info("crashkernel reservation failed - memory is in use.\n"); 518 return; 519 } 520 } 521 522 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 523 memblock_free(crash_base, crash_size); 524 return; 525 } 526 527 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 528 (unsigned long)(crash_size >> 20), 529 (unsigned long)(crash_base >> 20), 530 (unsigned long)(total_mem >> 20)); 531 532 crashk_res.start = crash_base; 533 crashk_res.end = crash_base + crash_size - 1; 534 insert_resource(&iomem_resource, &crashk_res); 535 } 536 #else 537 static void __init reserve_crashkernel(void) 538 { 539 } 540 #endif 541 542 static struct resource standard_io_resources[] = { 543 { .name = "dma1", .start = 0x00, .end = 0x1f, 544 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 545 { .name = "pic1", .start = 0x20, .end = 0x21, 546 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 547 { .name = "timer0", .start = 0x40, .end = 0x43, 548 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 549 { .name = "timer1", .start = 0x50, .end = 0x53, 550 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 551 { .name = "keyboard", .start = 0x60, .end = 0x60, 552 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 553 { .name = "keyboard", .start = 0x64, .end = 0x64, 554 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 555 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 556 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 557 { .name = "pic2", .start = 0xa0, .end = 0xa1, 558 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 559 { .name = "dma2", .start = 0xc0, .end = 0xdf, 560 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 561 { .name = "fpu", .start = 0xf0, .end = 0xff, 562 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 563 }; 564 565 void __init reserve_standard_io_resources(void) 566 { 567 int i; 568 569 /* request I/O space for devices used on all i[345]86 PCs */ 570 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 571 request_resource(&ioport_resource, &standard_io_resources[i]); 572 573 } 574 575 static __init void reserve_ibft_region(void) 576 { 577 unsigned long addr, size = 0; 578 579 addr = find_ibft_region(&size); 580 581 if (size) 582 memblock_reserve(addr, size); 583 } 584 585 static bool __init snb_gfx_workaround_needed(void) 586 { 587 #ifdef CONFIG_PCI 588 int i; 589 u16 vendor, devid; 590 static const __initconst u16 snb_ids[] = { 591 0x0102, 592 0x0112, 593 0x0122, 594 0x0106, 595 0x0116, 596 0x0126, 597 0x010a, 598 }; 599 600 /* Assume no if something weird is going on with PCI */ 601 if (!early_pci_allowed()) 602 return false; 603 604 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 605 if (vendor != 0x8086) 606 return false; 607 608 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 609 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 610 if (devid == snb_ids[i]) 611 return true; 612 #endif 613 614 return false; 615 } 616 617 /* 618 * Sandy Bridge graphics has trouble with certain ranges, exclude 619 * them from allocation. 620 */ 621 static void __init trim_snb_memory(void) 622 { 623 static const __initconst unsigned long bad_pages[] = { 624 0x20050000, 625 0x20110000, 626 0x20130000, 627 0x20138000, 628 0x40004000, 629 }; 630 int i; 631 632 if (!snb_gfx_workaround_needed()) 633 return; 634 635 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 636 637 /* 638 * SandyBridge integrated graphics devices have a bug that prevents 639 * them from accessing certain memory ranges, namely anything below 640 * 1M and in the pages listed in bad_pages[] above. 641 * 642 * To avoid these pages being ever accessed by SNB gfx devices reserve 643 * bad_pages that have not already been reserved at boot time. 644 * All memory below the 1 MB mark is anyway reserved later during 645 * setup_arch(), so there is no need to reserve it here. 646 */ 647 648 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 649 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 650 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 651 bad_pages[i]); 652 } 653 } 654 655 static void __init trim_bios_range(void) 656 { 657 /* 658 * A special case is the first 4Kb of memory; 659 * This is a BIOS owned area, not kernel ram, but generally 660 * not listed as such in the E820 table. 661 * 662 * This typically reserves additional memory (64KiB by default) 663 * since some BIOSes are known to corrupt low memory. See the 664 * Kconfig help text for X86_RESERVE_LOW. 665 */ 666 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 667 668 /* 669 * special case: Some BIOSes report the PC BIOS 670 * area (640Kb -> 1Mb) as RAM even though it is not. 671 * take them out. 672 */ 673 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 674 675 e820__update_table(e820_table); 676 } 677 678 /* called before trim_bios_range() to spare extra sanitize */ 679 static void __init e820_add_kernel_range(void) 680 { 681 u64 start = __pa_symbol(_text); 682 u64 size = __pa_symbol(_end) - start; 683 684 /* 685 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 686 * attempt to fix it by adding the range. We may have a confused BIOS, 687 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 688 * exclude kernel range. If we really are running on top non-RAM, 689 * we will crash later anyways. 690 */ 691 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 692 return; 693 694 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 695 e820__range_remove(start, size, E820_TYPE_RAM, 0); 696 e820__range_add(start, size, E820_TYPE_RAM); 697 } 698 699 static void __init early_reserve_memory(void) 700 { 701 /* 702 * Reserve the memory occupied by the kernel between _text and 703 * __end_of_kernel_reserve symbols. Any kernel sections after the 704 * __end_of_kernel_reserve symbol must be explicitly reserved with a 705 * separate memblock_reserve() or they will be discarded. 706 */ 707 memblock_reserve(__pa_symbol(_text), 708 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 709 710 /* 711 * The first 4Kb of memory is a BIOS owned area, but generally it is 712 * not listed as such in the E820 table. 713 * 714 * Reserve the first 64K of memory since some BIOSes are known to 715 * corrupt low memory. After the real mode trampoline is allocated the 716 * rest of the memory below 640k is reserved. 717 * 718 * In addition, make sure page 0 is always reserved because on 719 * systems with L1TF its contents can be leaked to user processes. 720 */ 721 memblock_reserve(0, SZ_64K); 722 723 early_reserve_initrd(); 724 725 if (efi_enabled(EFI_BOOT)) 726 efi_memblock_x86_reserve_range(); 727 728 memblock_x86_reserve_range_setup_data(); 729 730 reserve_ibft_region(); 731 reserve_bios_regions(); 732 trim_snb_memory(); 733 } 734 735 /* 736 * Dump out kernel offset information on panic. 737 */ 738 static int 739 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 740 { 741 if (kaslr_enabled()) { 742 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 743 kaslr_offset(), 744 __START_KERNEL, 745 __START_KERNEL_map, 746 MODULES_VADDR-1); 747 } else { 748 pr_emerg("Kernel Offset: disabled\n"); 749 } 750 751 return 0; 752 } 753 754 /* 755 * Determine if we were loaded by an EFI loader. If so, then we have also been 756 * passed the efi memmap, systab, etc., so we should use these data structures 757 * for initialization. Note, the efi init code path is determined by the 758 * global efi_enabled. This allows the same kernel image to be used on existing 759 * systems (with a traditional BIOS) as well as on EFI systems. 760 */ 761 /* 762 * setup_arch - architecture-specific boot-time initializations 763 * 764 * Note: On x86_64, fixmaps are ready for use even before this is called. 765 */ 766 767 void __init setup_arch(char **cmdline_p) 768 { 769 #ifdef CONFIG_X86_32 770 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 771 772 /* 773 * copy kernel address range established so far and switch 774 * to the proper swapper page table 775 */ 776 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 777 initial_page_table + KERNEL_PGD_BOUNDARY, 778 KERNEL_PGD_PTRS); 779 780 load_cr3(swapper_pg_dir); 781 /* 782 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 783 * a cr3 based tlb flush, so the following __flush_tlb_all() 784 * will not flush anything because the CPU quirk which clears 785 * X86_FEATURE_PGE has not been invoked yet. Though due to the 786 * load_cr3() above the TLB has been flushed already. The 787 * quirk is invoked before subsequent calls to __flush_tlb_all() 788 * so proper operation is guaranteed. 789 */ 790 __flush_tlb_all(); 791 #else 792 printk(KERN_INFO "Command line: %s\n", boot_command_line); 793 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 794 #endif 795 796 /* 797 * If we have OLPC OFW, we might end up relocating the fixmap due to 798 * reserve_top(), so do this before touching the ioremap area. 799 */ 800 olpc_ofw_detect(); 801 802 idt_setup_early_traps(); 803 early_cpu_init(); 804 jump_label_init(); 805 static_call_init(); 806 early_ioremap_init(); 807 808 setup_olpc_ofw_pgd(); 809 810 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 811 screen_info = boot_params.screen_info; 812 edid_info = boot_params.edid_info; 813 #ifdef CONFIG_X86_32 814 apm_info.bios = boot_params.apm_bios_info; 815 ist_info = boot_params.ist_info; 816 #endif 817 saved_video_mode = boot_params.hdr.vid_mode; 818 bootloader_type = boot_params.hdr.type_of_loader; 819 if ((bootloader_type >> 4) == 0xe) { 820 bootloader_type &= 0xf; 821 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 822 } 823 bootloader_version = bootloader_type & 0xf; 824 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 825 826 #ifdef CONFIG_BLK_DEV_RAM 827 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 828 #endif 829 #ifdef CONFIG_EFI 830 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 831 EFI32_LOADER_SIGNATURE, 4)) { 832 set_bit(EFI_BOOT, &efi.flags); 833 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 834 EFI64_LOADER_SIGNATURE, 4)) { 835 set_bit(EFI_BOOT, &efi.flags); 836 set_bit(EFI_64BIT, &efi.flags); 837 } 838 #endif 839 840 x86_init.oem.arch_setup(); 841 842 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 843 e820__memory_setup(); 844 parse_setup_data(); 845 846 copy_edd(); 847 848 if (!boot_params.hdr.root_flags) 849 root_mountflags &= ~MS_RDONLY; 850 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 851 852 code_resource.start = __pa_symbol(_text); 853 code_resource.end = __pa_symbol(_etext)-1; 854 rodata_resource.start = __pa_symbol(__start_rodata); 855 rodata_resource.end = __pa_symbol(__end_rodata)-1; 856 data_resource.start = __pa_symbol(_sdata); 857 data_resource.end = __pa_symbol(_edata)-1; 858 bss_resource.start = __pa_symbol(__bss_start); 859 bss_resource.end = __pa_symbol(__bss_stop)-1; 860 861 #ifdef CONFIG_CMDLINE_BOOL 862 #ifdef CONFIG_CMDLINE_OVERRIDE 863 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 864 #else 865 if (builtin_cmdline[0]) { 866 /* append boot loader cmdline to builtin */ 867 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 868 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 869 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 870 } 871 #endif 872 #endif 873 874 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 875 *cmdline_p = command_line; 876 877 /* 878 * x86_configure_nx() is called before parse_early_param() to detect 879 * whether hardware doesn't support NX (so that the early EHCI debug 880 * console setup can safely call set_fixmap()). It may then be called 881 * again from within noexec_setup() during parsing early parameters 882 * to honor the respective command line option. 883 */ 884 x86_configure_nx(); 885 886 parse_early_param(); 887 888 /* 889 * Do some memory reservations *before* memory is added to 890 * memblock, so memblock allocations won't overwrite it. 891 * Do it after early param, so we could get (unlikely) panic from 892 * serial. 893 * 894 * After this point everything still needed from the boot loader or 895 * firmware or kernel text should be early reserved or marked not 896 * RAM in e820. All other memory is free game. 897 */ 898 early_reserve_memory(); 899 900 #ifdef CONFIG_MEMORY_HOTPLUG 901 /* 902 * Memory used by the kernel cannot be hot-removed because Linux 903 * cannot migrate the kernel pages. When memory hotplug is 904 * enabled, we should prevent memblock from allocating memory 905 * for the kernel. 906 * 907 * ACPI SRAT records all hotpluggable memory ranges. But before 908 * SRAT is parsed, we don't know about it. 909 * 910 * The kernel image is loaded into memory at very early time. We 911 * cannot prevent this anyway. So on NUMA system, we set any 912 * node the kernel resides in as un-hotpluggable. 913 * 914 * Since on modern servers, one node could have double-digit 915 * gigabytes memory, we can assume the memory around the kernel 916 * image is also un-hotpluggable. So before SRAT is parsed, just 917 * allocate memory near the kernel image to try the best to keep 918 * the kernel away from hotpluggable memory. 919 */ 920 if (movable_node_is_enabled()) 921 memblock_set_bottom_up(true); 922 #endif 923 924 x86_report_nx(); 925 926 if (acpi_mps_check()) { 927 #ifdef CONFIG_X86_LOCAL_APIC 928 disable_apic = 1; 929 #endif 930 setup_clear_cpu_cap(X86_FEATURE_APIC); 931 } 932 933 e820__reserve_setup_data(); 934 e820__finish_early_params(); 935 936 if (efi_enabled(EFI_BOOT)) 937 efi_init(); 938 939 dmi_setup(); 940 941 /* 942 * VMware detection requires dmi to be available, so this 943 * needs to be done after dmi_setup(), for the boot CPU. 944 */ 945 init_hypervisor_platform(); 946 947 tsc_early_init(); 948 x86_init.resources.probe_roms(); 949 950 /* after parse_early_param, so could debug it */ 951 insert_resource(&iomem_resource, &code_resource); 952 insert_resource(&iomem_resource, &rodata_resource); 953 insert_resource(&iomem_resource, &data_resource); 954 insert_resource(&iomem_resource, &bss_resource); 955 956 e820_add_kernel_range(); 957 trim_bios_range(); 958 #ifdef CONFIG_X86_32 959 if (ppro_with_ram_bug()) { 960 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 961 E820_TYPE_RESERVED); 962 e820__update_table(e820_table); 963 printk(KERN_INFO "fixed physical RAM map:\n"); 964 e820__print_table("bad_ppro"); 965 } 966 #else 967 early_gart_iommu_check(); 968 #endif 969 970 /* 971 * partially used pages are not usable - thus 972 * we are rounding upwards: 973 */ 974 max_pfn = e820__end_of_ram_pfn(); 975 976 /* update e820 for memory not covered by WB MTRRs */ 977 mtrr_bp_init(); 978 if (mtrr_trim_uncached_memory(max_pfn)) 979 max_pfn = e820__end_of_ram_pfn(); 980 981 max_possible_pfn = max_pfn; 982 983 /* 984 * This call is required when the CPU does not support PAT. If 985 * mtrr_bp_init() invoked it already via pat_init() the call has no 986 * effect. 987 */ 988 init_cache_modes(); 989 990 /* 991 * Define random base addresses for memory sections after max_pfn is 992 * defined and before each memory section base is used. 993 */ 994 kernel_randomize_memory(); 995 996 #ifdef CONFIG_X86_32 997 /* max_low_pfn get updated here */ 998 find_low_pfn_range(); 999 #else 1000 check_x2apic(); 1001 1002 /* How many end-of-memory variables you have, grandma! */ 1003 /* need this before calling reserve_initrd */ 1004 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1005 max_low_pfn = e820__end_of_low_ram_pfn(); 1006 else 1007 max_low_pfn = max_pfn; 1008 1009 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1010 #endif 1011 1012 /* 1013 * Find and reserve possible boot-time SMP configuration: 1014 */ 1015 find_smp_config(); 1016 1017 early_alloc_pgt_buf(); 1018 1019 /* 1020 * Need to conclude brk, before e820__memblock_setup() 1021 * it could use memblock_find_in_range, could overlap with 1022 * brk area. 1023 */ 1024 reserve_brk(); 1025 1026 cleanup_highmap(); 1027 1028 memblock_set_current_limit(ISA_END_ADDRESS); 1029 e820__memblock_setup(); 1030 1031 /* 1032 * Needs to run after memblock setup because it needs the physical 1033 * memory size. 1034 */ 1035 sev_setup_arch(); 1036 1037 efi_fake_memmap(); 1038 efi_find_mirror(); 1039 efi_esrt_init(); 1040 efi_mokvar_table_init(); 1041 1042 /* 1043 * The EFI specification says that boot service code won't be 1044 * called after ExitBootServices(). This is, in fact, a lie. 1045 */ 1046 efi_reserve_boot_services(); 1047 1048 /* preallocate 4k for mptable mpc */ 1049 e820__memblock_alloc_reserved_mpc_new(); 1050 1051 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1052 setup_bios_corruption_check(); 1053 #endif 1054 1055 #ifdef CONFIG_X86_32 1056 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1057 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1058 #endif 1059 1060 /* 1061 * Find free memory for the real mode trampoline and place it there. If 1062 * there is not enough free memory under 1M, on EFI-enabled systems 1063 * there will be additional attempt to reclaim the memory for the real 1064 * mode trampoline at efi_free_boot_services(). 1065 * 1066 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1067 * are known to corrupt low memory and several hundred kilobytes are not 1068 * worth complex detection what memory gets clobbered. Windows does the 1069 * same thing for very similar reasons. 1070 * 1071 * Moreover, on machines with SandyBridge graphics or in setups that use 1072 * crashkernel the entire 1M is reserved anyway. 1073 */ 1074 reserve_real_mode(); 1075 1076 init_mem_mapping(); 1077 1078 idt_setup_early_pf(); 1079 1080 /* 1081 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1082 * with the current CR4 value. This may not be necessary, but 1083 * auditing all the early-boot CR4 manipulation would be needed to 1084 * rule it out. 1085 * 1086 * Mask off features that don't work outside long mode (just 1087 * PCIDE for now). 1088 */ 1089 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1090 1091 memblock_set_current_limit(get_max_mapped()); 1092 1093 /* 1094 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1095 */ 1096 1097 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1098 if (init_ohci1394_dma_early) 1099 init_ohci1394_dma_on_all_controllers(); 1100 #endif 1101 /* Allocate bigger log buffer */ 1102 setup_log_buf(1); 1103 1104 if (efi_enabled(EFI_BOOT)) { 1105 switch (boot_params.secure_boot) { 1106 case efi_secureboot_mode_disabled: 1107 pr_info("Secure boot disabled\n"); 1108 break; 1109 case efi_secureboot_mode_enabled: 1110 pr_info("Secure boot enabled\n"); 1111 break; 1112 default: 1113 pr_info("Secure boot could not be determined\n"); 1114 break; 1115 } 1116 } 1117 1118 reserve_initrd(); 1119 1120 acpi_table_upgrade(); 1121 /* Look for ACPI tables and reserve memory occupied by them. */ 1122 acpi_boot_table_init(); 1123 1124 vsmp_init(); 1125 1126 io_delay_init(); 1127 1128 early_platform_quirks(); 1129 1130 early_acpi_boot_init(); 1131 1132 initmem_init(); 1133 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1134 1135 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1136 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1137 1138 /* 1139 * Reserve memory for crash kernel after SRAT is parsed so that it 1140 * won't consume hotpluggable memory. 1141 */ 1142 reserve_crashkernel(); 1143 1144 memblock_find_dma_reserve(); 1145 1146 if (!early_xdbc_setup_hardware()) 1147 early_xdbc_register_console(); 1148 1149 x86_init.paging.pagetable_init(); 1150 1151 kasan_init(); 1152 1153 /* 1154 * Sync back kernel address range. 1155 * 1156 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1157 * this call? 1158 */ 1159 sync_initial_page_table(); 1160 1161 tboot_probe(); 1162 1163 map_vsyscall(); 1164 1165 generic_apic_probe(); 1166 1167 early_quirks(); 1168 1169 /* 1170 * Read APIC and some other early information from ACPI tables. 1171 */ 1172 acpi_boot_init(); 1173 x86_dtb_init(); 1174 1175 /* 1176 * get boot-time SMP configuration: 1177 */ 1178 get_smp_config(); 1179 1180 /* 1181 * Systems w/o ACPI and mptables might not have it mapped the local 1182 * APIC yet, but prefill_possible_map() might need to access it. 1183 */ 1184 init_apic_mappings(); 1185 1186 prefill_possible_map(); 1187 1188 init_cpu_to_node(); 1189 init_gi_nodes(); 1190 1191 io_apic_init_mappings(); 1192 1193 x86_init.hyper.guest_late_init(); 1194 1195 e820__reserve_resources(); 1196 e820__register_nosave_regions(max_pfn); 1197 1198 x86_init.resources.reserve_resources(); 1199 1200 e820__setup_pci_gap(); 1201 1202 #ifdef CONFIG_VT 1203 #if defined(CONFIG_VGA_CONSOLE) 1204 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1205 conswitchp = &vga_con; 1206 #endif 1207 #endif 1208 x86_init.oem.banner(); 1209 1210 x86_init.timers.wallclock_init(); 1211 1212 /* 1213 * This needs to run before setup_local_APIC() which soft-disables the 1214 * local APIC temporarily and that masks the thermal LVT interrupt, 1215 * leading to softlockups on machines which have configured SMI 1216 * interrupt delivery. 1217 */ 1218 therm_lvt_init(); 1219 1220 mcheck_init(); 1221 1222 register_refined_jiffies(CLOCK_TICK_RATE); 1223 1224 #ifdef CONFIG_EFI 1225 if (efi_enabled(EFI_BOOT)) 1226 efi_apply_memmap_quirks(); 1227 #endif 1228 1229 unwind_init(); 1230 } 1231 1232 #ifdef CONFIG_X86_32 1233 1234 static struct resource video_ram_resource = { 1235 .name = "Video RAM area", 1236 .start = 0xa0000, 1237 .end = 0xbffff, 1238 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1239 }; 1240 1241 void __init i386_reserve_resources(void) 1242 { 1243 request_resource(&iomem_resource, &video_ram_resource); 1244 reserve_standard_io_resources(); 1245 } 1246 1247 #endif /* CONFIG_X86_32 */ 1248 1249 static struct notifier_block kernel_offset_notifier = { 1250 .notifier_call = dump_kernel_offset 1251 }; 1252 1253 static int __init register_kernel_offset_dumper(void) 1254 { 1255 atomic_notifier_chain_register(&panic_notifier_list, 1256 &kernel_offset_notifier); 1257 return 0; 1258 } 1259 __initcall(register_kernel_offset_dumper); 1260