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/pci.h> 18 #include <linux/root_dev.h> 19 #include <linux/hugetlb.h> 20 #include <linux/tboot.h> 21 #include <linux/usb/xhci-dbgp.h> 22 #include <linux/static_call.h> 23 #include <linux/swiotlb.h> 24 25 #include <uapi/linux/mount.h> 26 27 #include <xen/xen.h> 28 29 #include <asm/apic.h> 30 #include <asm/numa.h> 31 #include <asm/bios_ebda.h> 32 #include <asm/bugs.h> 33 #include <asm/cpu.h> 34 #include <asm/efi.h> 35 #include <asm/gart.h> 36 #include <asm/hypervisor.h> 37 #include <asm/io_apic.h> 38 #include <asm/kasan.h> 39 #include <asm/kaslr.h> 40 #include <asm/mce.h> 41 #include <asm/mtrr.h> 42 #include <asm/realmode.h> 43 #include <asm/olpc_ofw.h> 44 #include <asm/pci-direct.h> 45 #include <asm/prom.h> 46 #include <asm/proto.h> 47 #include <asm/thermal.h> 48 #include <asm/unwind.h> 49 #include <asm/vsyscall.h> 50 #include <linux/vmalloc.h> 51 52 /* 53 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 54 * max_pfn_mapped: highest directly mapped pfn > 4 GB 55 * 56 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 57 * represented by pfn_mapped[]. 58 */ 59 unsigned long max_low_pfn_mapped; 60 unsigned long max_pfn_mapped; 61 62 #ifdef CONFIG_DMI 63 RESERVE_BRK(dmi_alloc, 65536); 64 #endif 65 66 67 /* 68 * Range of the BSS area. The size of the BSS area is determined 69 * at link time, with RESERVE_BRK() facility reserving additional 70 * chunks. 71 */ 72 unsigned long _brk_start = (unsigned long)__brk_base; 73 unsigned long _brk_end = (unsigned long)__brk_base; 74 75 struct boot_params boot_params; 76 77 /* 78 * These are the four main kernel memory regions, we put them into 79 * the resource tree so that kdump tools and other debugging tools 80 * recover it: 81 */ 82 83 static struct resource rodata_resource = { 84 .name = "Kernel rodata", 85 .start = 0, 86 .end = 0, 87 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 88 }; 89 90 static struct resource data_resource = { 91 .name = "Kernel data", 92 .start = 0, 93 .end = 0, 94 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 95 }; 96 97 static struct resource code_resource = { 98 .name = "Kernel code", 99 .start = 0, 100 .end = 0, 101 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 102 }; 103 104 static struct resource bss_resource = { 105 .name = "Kernel bss", 106 .start = 0, 107 .end = 0, 108 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 109 }; 110 111 112 #ifdef CONFIG_X86_32 113 /* CPU data as detected by the assembly code in head_32.S */ 114 struct cpuinfo_x86 new_cpu_data; 115 116 /* Common CPU data for all CPUs */ 117 struct cpuinfo_x86 boot_cpu_data __read_mostly; 118 EXPORT_SYMBOL(boot_cpu_data); 119 120 unsigned int def_to_bigsmp; 121 122 struct apm_info apm_info; 123 EXPORT_SYMBOL(apm_info); 124 125 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 126 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 127 struct ist_info ist_info; 128 EXPORT_SYMBOL(ist_info); 129 #else 130 struct ist_info ist_info; 131 #endif 132 133 #else 134 struct cpuinfo_x86 boot_cpu_data __read_mostly; 135 EXPORT_SYMBOL(boot_cpu_data); 136 #endif 137 138 139 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 140 __visible unsigned long mmu_cr4_features __ro_after_init; 141 #else 142 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 143 #endif 144 145 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 146 int bootloader_type, bootloader_version; 147 148 /* 149 * Setup options 150 */ 151 struct screen_info screen_info; 152 EXPORT_SYMBOL(screen_info); 153 struct edid_info edid_info; 154 EXPORT_SYMBOL_GPL(edid_info); 155 156 extern int root_mountflags; 157 158 unsigned long saved_video_mode; 159 160 #define RAMDISK_IMAGE_START_MASK 0x07FF 161 #define RAMDISK_PROMPT_FLAG 0x8000 162 #define RAMDISK_LOAD_FLAG 0x4000 163 164 static char __initdata command_line[COMMAND_LINE_SIZE]; 165 #ifdef CONFIG_CMDLINE_BOOL 166 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 167 #endif 168 169 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 170 struct edd edd; 171 #ifdef CONFIG_EDD_MODULE 172 EXPORT_SYMBOL(edd); 173 #endif 174 /** 175 * copy_edd() - Copy the BIOS EDD information 176 * from boot_params into a safe place. 177 * 178 */ 179 static inline void __init copy_edd(void) 180 { 181 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 182 sizeof(edd.mbr_signature)); 183 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 184 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 185 edd.edd_info_nr = boot_params.eddbuf_entries; 186 } 187 #else 188 static inline void __init copy_edd(void) 189 { 190 } 191 #endif 192 193 void * __init extend_brk(size_t size, size_t align) 194 { 195 size_t mask = align - 1; 196 void *ret; 197 198 BUG_ON(_brk_start == 0); 199 BUG_ON(align & mask); 200 201 _brk_end = (_brk_end + mask) & ~mask; 202 BUG_ON((char *)(_brk_end + size) > __brk_limit); 203 204 ret = (void *)_brk_end; 205 _brk_end += size; 206 207 memset(ret, 0, size); 208 209 return ret; 210 } 211 212 #ifdef CONFIG_X86_32 213 static void __init cleanup_highmap(void) 214 { 215 } 216 #endif 217 218 static void __init reserve_brk(void) 219 { 220 if (_brk_end > _brk_start) 221 memblock_reserve(__pa_symbol(_brk_start), 222 _brk_end - _brk_start); 223 224 /* Mark brk area as locked down and no longer taking any 225 new allocations */ 226 _brk_start = 0; 227 } 228 229 u64 relocated_ramdisk; 230 231 #ifdef CONFIG_BLK_DEV_INITRD 232 233 static u64 __init get_ramdisk_image(void) 234 { 235 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 236 237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 238 239 if (ramdisk_image == 0) 240 ramdisk_image = phys_initrd_start; 241 242 return ramdisk_image; 243 } 244 static u64 __init get_ramdisk_size(void) 245 { 246 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 247 248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 249 250 if (ramdisk_size == 0) 251 ramdisk_size = phys_initrd_size; 252 253 return ramdisk_size; 254 } 255 256 static void __init relocate_initrd(void) 257 { 258 /* Assume only end is not page aligned */ 259 u64 ramdisk_image = get_ramdisk_image(); 260 u64 ramdisk_size = get_ramdisk_size(); 261 u64 area_size = PAGE_ALIGN(ramdisk_size); 262 263 /* We need to move the initrd down into directly mapped mem */ 264 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 265 PFN_PHYS(max_pfn_mapped)); 266 if (!relocated_ramdisk) 267 panic("Cannot find place for new RAMDISK of size %lld\n", 268 ramdisk_size); 269 270 initrd_start = relocated_ramdisk + PAGE_OFFSET; 271 initrd_end = initrd_start + ramdisk_size; 272 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 273 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 274 275 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 276 277 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 278 " [mem %#010llx-%#010llx]\n", 279 ramdisk_image, ramdisk_image + ramdisk_size - 1, 280 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 281 } 282 283 static void __init early_reserve_initrd(void) 284 { 285 /* Assume only end is not page aligned */ 286 u64 ramdisk_image = get_ramdisk_image(); 287 u64 ramdisk_size = get_ramdisk_size(); 288 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 289 290 if (!boot_params.hdr.type_of_loader || 291 !ramdisk_image || !ramdisk_size) 292 return; /* No initrd provided by bootloader */ 293 294 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 295 } 296 297 static void __init reserve_initrd(void) 298 { 299 /* Assume only end is not page aligned */ 300 u64 ramdisk_image = get_ramdisk_image(); 301 u64 ramdisk_size = get_ramdisk_size(); 302 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 303 304 if (!boot_params.hdr.type_of_loader || 305 !ramdisk_image || !ramdisk_size) 306 return; /* No initrd provided by bootloader */ 307 308 initrd_start = 0; 309 310 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 311 ramdisk_end - 1); 312 313 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 314 PFN_DOWN(ramdisk_end))) { 315 /* All are mapped, easy case */ 316 initrd_start = ramdisk_image + PAGE_OFFSET; 317 initrd_end = initrd_start + ramdisk_size; 318 return; 319 } 320 321 relocate_initrd(); 322 323 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 324 } 325 326 #else 327 static void __init early_reserve_initrd(void) 328 { 329 } 330 static void __init reserve_initrd(void) 331 { 332 } 333 #endif /* CONFIG_BLK_DEV_INITRD */ 334 335 static void __init parse_setup_data(void) 336 { 337 struct setup_data *data; 338 u64 pa_data, pa_next; 339 340 pa_data = boot_params.hdr.setup_data; 341 while (pa_data) { 342 u32 data_len, data_type; 343 344 data = early_memremap(pa_data, sizeof(*data)); 345 data_len = data->len + sizeof(struct setup_data); 346 data_type = data->type; 347 pa_next = data->next; 348 early_memunmap(data, sizeof(*data)); 349 350 switch (data_type) { 351 case SETUP_E820_EXT: 352 e820__memory_setup_extended(pa_data, data_len); 353 break; 354 case SETUP_DTB: 355 add_dtb(pa_data); 356 break; 357 case SETUP_EFI: 358 parse_efi_setup(pa_data, data_len); 359 break; 360 default: 361 break; 362 } 363 pa_data = pa_next; 364 } 365 } 366 367 static void __init memblock_x86_reserve_range_setup_data(void) 368 { 369 struct setup_data *data; 370 u64 pa_data; 371 372 pa_data = boot_params.hdr.setup_data; 373 while (pa_data) { 374 data = early_memremap(pa_data, sizeof(*data)); 375 memblock_reserve(pa_data, sizeof(*data) + data->len); 376 377 if (data->type == SETUP_INDIRECT && 378 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) 379 memblock_reserve(((struct setup_indirect *)data->data)->addr, 380 ((struct setup_indirect *)data->data)->len); 381 382 pa_data = data->next; 383 early_memunmap(data, sizeof(*data)); 384 } 385 } 386 387 /* 388 * --------- Crashkernel reservation ------------------------------ 389 */ 390 391 #ifdef CONFIG_KEXEC_CORE 392 393 /* 16M alignment for crash kernel regions */ 394 #define CRASH_ALIGN SZ_16M 395 396 /* 397 * Keep the crash kernel below this limit. 398 * 399 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 400 * due to mapping restrictions. 401 * 402 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 403 * the upper limit of system RAM in 4-level paging mode. Since the kdump 404 * jump could be from 5-level paging to 4-level paging, the jump will fail if 405 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 406 * no good way to detect the paging mode of the target kernel which will be 407 * loaded for dumping. 408 */ 409 #ifdef CONFIG_X86_32 410 # define CRASH_ADDR_LOW_MAX SZ_512M 411 # define CRASH_ADDR_HIGH_MAX SZ_512M 412 #else 413 # define CRASH_ADDR_LOW_MAX SZ_4G 414 # define CRASH_ADDR_HIGH_MAX SZ_64T 415 #endif 416 417 static int __init reserve_crashkernel_low(void) 418 { 419 #ifdef CONFIG_X86_64 420 unsigned long long base, low_base = 0, low_size = 0; 421 unsigned long low_mem_limit; 422 int ret; 423 424 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 425 426 /* crashkernel=Y,low */ 427 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 428 if (ret) { 429 /* 430 * two parts from kernel/dma/swiotlb.c: 431 * -swiotlb size: user-specified with swiotlb= or default. 432 * 433 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 434 * to 8M for other buffers that may need to stay low too. Also 435 * make sure we allocate enough extra low memory so that we 436 * don't run out of DMA buffers for 32-bit devices. 437 */ 438 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 439 } else { 440 /* passed with crashkernel=0,low ? */ 441 if (!low_size) 442 return 0; 443 } 444 445 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 446 if (!low_base) { 447 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 448 (unsigned long)(low_size >> 20)); 449 return -ENOMEM; 450 } 451 452 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 453 (unsigned long)(low_size >> 20), 454 (unsigned long)(low_base >> 20), 455 (unsigned long)(low_mem_limit >> 20)); 456 457 crashk_low_res.start = low_base; 458 crashk_low_res.end = low_base + low_size - 1; 459 insert_resource(&iomem_resource, &crashk_low_res); 460 #endif 461 return 0; 462 } 463 464 static void __init reserve_crashkernel(void) 465 { 466 unsigned long long crash_size, crash_base, total_mem; 467 bool high = false; 468 int ret; 469 470 total_mem = memblock_phys_mem_size(); 471 472 /* crashkernel=XM */ 473 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 474 if (ret != 0 || crash_size <= 0) { 475 /* crashkernel=X,high */ 476 ret = parse_crashkernel_high(boot_command_line, total_mem, 477 &crash_size, &crash_base); 478 if (ret != 0 || crash_size <= 0) 479 return; 480 high = true; 481 } 482 483 if (xen_pv_domain()) { 484 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 485 return; 486 } 487 488 /* 0 means: find the address automatically */ 489 if (!crash_base) { 490 /* 491 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 492 * crashkernel=x,high reserves memory over 4G, also allocates 493 * 256M extra low memory for DMA buffers and swiotlb. 494 * But the extra memory is not required for all machines. 495 * So try low memory first and fall back to high memory 496 * unless "crashkernel=size[KMG],high" is specified. 497 */ 498 if (!high) 499 crash_base = memblock_phys_alloc_range(crash_size, 500 CRASH_ALIGN, CRASH_ALIGN, 501 CRASH_ADDR_LOW_MAX); 502 if (!crash_base) 503 crash_base = memblock_phys_alloc_range(crash_size, 504 CRASH_ALIGN, CRASH_ALIGN, 505 CRASH_ADDR_HIGH_MAX); 506 if (!crash_base) { 507 pr_info("crashkernel reservation failed - No suitable area found.\n"); 508 return; 509 } 510 } else { 511 unsigned long long start; 512 513 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 514 crash_base + crash_size); 515 if (start != crash_base) { 516 pr_info("crashkernel reservation failed - memory is in use.\n"); 517 return; 518 } 519 } 520 521 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 522 memblock_free(crash_base, crash_size); 523 return; 524 } 525 526 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 527 (unsigned long)(crash_size >> 20), 528 (unsigned long)(crash_base >> 20), 529 (unsigned long)(total_mem >> 20)); 530 531 crashk_res.start = crash_base; 532 crashk_res.end = crash_base + crash_size - 1; 533 insert_resource(&iomem_resource, &crashk_res); 534 } 535 #else 536 static void __init reserve_crashkernel(void) 537 { 538 } 539 #endif 540 541 static struct resource standard_io_resources[] = { 542 { .name = "dma1", .start = 0x00, .end = 0x1f, 543 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 544 { .name = "pic1", .start = 0x20, .end = 0x21, 545 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 546 { .name = "timer0", .start = 0x40, .end = 0x43, 547 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 548 { .name = "timer1", .start = 0x50, .end = 0x53, 549 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 550 { .name = "keyboard", .start = 0x60, .end = 0x60, 551 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 552 { .name = "keyboard", .start = 0x64, .end = 0x64, 553 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 554 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 555 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 556 { .name = "pic2", .start = 0xa0, .end = 0xa1, 557 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 558 { .name = "dma2", .start = 0xc0, .end = 0xdf, 559 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 560 { .name = "fpu", .start = 0xf0, .end = 0xff, 561 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 562 }; 563 564 void __init reserve_standard_io_resources(void) 565 { 566 int i; 567 568 /* request I/O space for devices used on all i[345]86 PCs */ 569 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 570 request_resource(&ioport_resource, &standard_io_resources[i]); 571 572 } 573 574 static __init void reserve_ibft_region(void) 575 { 576 unsigned long addr, size = 0; 577 578 addr = find_ibft_region(&size); 579 580 if (size) 581 memblock_reserve(addr, size); 582 } 583 584 static bool __init snb_gfx_workaround_needed(void) 585 { 586 #ifdef CONFIG_PCI 587 int i; 588 u16 vendor, devid; 589 static const __initconst u16 snb_ids[] = { 590 0x0102, 591 0x0112, 592 0x0122, 593 0x0106, 594 0x0116, 595 0x0126, 596 0x010a, 597 }; 598 599 /* Assume no if something weird is going on with PCI */ 600 if (!early_pci_allowed()) 601 return false; 602 603 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 604 if (vendor != 0x8086) 605 return false; 606 607 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 608 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 609 if (devid == snb_ids[i]) 610 return true; 611 #endif 612 613 return false; 614 } 615 616 /* 617 * Sandy Bridge graphics has trouble with certain ranges, exclude 618 * them from allocation. 619 */ 620 static void __init trim_snb_memory(void) 621 { 622 static const __initconst unsigned long bad_pages[] = { 623 0x20050000, 624 0x20110000, 625 0x20130000, 626 0x20138000, 627 0x40004000, 628 }; 629 int i; 630 631 if (!snb_gfx_workaround_needed()) 632 return; 633 634 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 635 636 /* 637 * SandyBridge integrated graphics devices have a bug that prevents 638 * them from accessing certain memory ranges, namely anything below 639 * 1M and in the pages listed in bad_pages[] above. 640 * 641 * To avoid these pages being ever accessed by SNB gfx devices reserve 642 * bad_pages that have not already been reserved at boot time. 643 * All memory below the 1 MB mark is anyway reserved later during 644 * setup_arch(), so there is no need to reserve it here. 645 */ 646 647 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 648 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 649 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 650 bad_pages[i]); 651 } 652 } 653 654 static void __init trim_bios_range(void) 655 { 656 /* 657 * A special case is the first 4Kb of memory; 658 * This is a BIOS owned area, not kernel ram, but generally 659 * not listed as such in the E820 table. 660 * 661 * This typically reserves additional memory (64KiB by default) 662 * since some BIOSes are known to corrupt low memory. See the 663 * Kconfig help text for X86_RESERVE_LOW. 664 */ 665 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 666 667 /* 668 * special case: Some BIOSes report the PC BIOS 669 * area (640Kb -> 1Mb) as RAM even though it is not. 670 * take them out. 671 */ 672 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 673 674 e820__update_table(e820_table); 675 } 676 677 /* called before trim_bios_range() to spare extra sanitize */ 678 static void __init e820_add_kernel_range(void) 679 { 680 u64 start = __pa_symbol(_text); 681 u64 size = __pa_symbol(_end) - start; 682 683 /* 684 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 685 * attempt to fix it by adding the range. We may have a confused BIOS, 686 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 687 * exclude kernel range. If we really are running on top non-RAM, 688 * we will crash later anyways. 689 */ 690 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 691 return; 692 693 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 694 e820__range_remove(start, size, E820_TYPE_RAM, 0); 695 e820__range_add(start, size, E820_TYPE_RAM); 696 } 697 698 static void __init early_reserve_memory(void) 699 { 700 /* 701 * Reserve the memory occupied by the kernel between _text and 702 * __end_of_kernel_reserve symbols. Any kernel sections after the 703 * __end_of_kernel_reserve symbol must be explicitly reserved with a 704 * separate memblock_reserve() or they will be discarded. 705 */ 706 memblock_reserve(__pa_symbol(_text), 707 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 708 709 /* 710 * The first 4Kb of memory is a BIOS owned area, but generally it is 711 * not listed as such in the E820 table. 712 * 713 * Reserve the first 64K of memory since some BIOSes are known to 714 * corrupt low memory. After the real mode trampoline is allocated the 715 * rest of the memory below 640k is reserved. 716 * 717 * In addition, make sure page 0 is always reserved because on 718 * systems with L1TF its contents can be leaked to user processes. 719 */ 720 memblock_reserve(0, SZ_64K); 721 722 early_reserve_initrd(); 723 724 if (efi_enabled(EFI_BOOT)) 725 efi_memblock_x86_reserve_range(); 726 727 memblock_x86_reserve_range_setup_data(); 728 729 reserve_ibft_region(); 730 reserve_bios_regions(); 731 trim_snb_memory(); 732 } 733 734 /* 735 * Dump out kernel offset information on panic. 736 */ 737 static int 738 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 739 { 740 if (kaslr_enabled()) { 741 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 742 kaslr_offset(), 743 __START_KERNEL, 744 __START_KERNEL_map, 745 MODULES_VADDR-1); 746 } else { 747 pr_emerg("Kernel Offset: disabled\n"); 748 } 749 750 return 0; 751 } 752 753 /* 754 * Determine if we were loaded by an EFI loader. If so, then we have also been 755 * passed the efi memmap, systab, etc., so we should use these data structures 756 * for initialization. Note, the efi init code path is determined by the 757 * global efi_enabled. This allows the same kernel image to be used on existing 758 * systems (with a traditional BIOS) as well as on EFI systems. 759 */ 760 /* 761 * setup_arch - architecture-specific boot-time initializations 762 * 763 * Note: On x86_64, fixmaps are ready for use even before this is called. 764 */ 765 766 void __init setup_arch(char **cmdline_p) 767 { 768 #ifdef CONFIG_X86_32 769 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 770 771 /* 772 * copy kernel address range established so far and switch 773 * to the proper swapper page table 774 */ 775 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 776 initial_page_table + KERNEL_PGD_BOUNDARY, 777 KERNEL_PGD_PTRS); 778 779 load_cr3(swapper_pg_dir); 780 /* 781 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 782 * a cr3 based tlb flush, so the following __flush_tlb_all() 783 * will not flush anything because the CPU quirk which clears 784 * X86_FEATURE_PGE has not been invoked yet. Though due to the 785 * load_cr3() above the TLB has been flushed already. The 786 * quirk is invoked before subsequent calls to __flush_tlb_all() 787 * so proper operation is guaranteed. 788 */ 789 __flush_tlb_all(); 790 #else 791 printk(KERN_INFO "Command line: %s\n", boot_command_line); 792 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 793 #endif 794 795 /* 796 * If we have OLPC OFW, we might end up relocating the fixmap due to 797 * reserve_top(), so do this before touching the ioremap area. 798 */ 799 olpc_ofw_detect(); 800 801 idt_setup_early_traps(); 802 early_cpu_init(); 803 jump_label_init(); 804 static_call_init(); 805 early_ioremap_init(); 806 807 setup_olpc_ofw_pgd(); 808 809 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 810 screen_info = boot_params.screen_info; 811 edid_info = boot_params.edid_info; 812 #ifdef CONFIG_X86_32 813 apm_info.bios = boot_params.apm_bios_info; 814 ist_info = boot_params.ist_info; 815 #endif 816 saved_video_mode = boot_params.hdr.vid_mode; 817 bootloader_type = boot_params.hdr.type_of_loader; 818 if ((bootloader_type >> 4) == 0xe) { 819 bootloader_type &= 0xf; 820 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 821 } 822 bootloader_version = bootloader_type & 0xf; 823 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 824 825 #ifdef CONFIG_BLK_DEV_RAM 826 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 827 #endif 828 #ifdef CONFIG_EFI 829 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 830 EFI32_LOADER_SIGNATURE, 4)) { 831 set_bit(EFI_BOOT, &efi.flags); 832 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 833 EFI64_LOADER_SIGNATURE, 4)) { 834 set_bit(EFI_BOOT, &efi.flags); 835 set_bit(EFI_64BIT, &efi.flags); 836 } 837 #endif 838 839 x86_init.oem.arch_setup(); 840 841 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 842 e820__memory_setup(); 843 parse_setup_data(); 844 845 copy_edd(); 846 847 if (!boot_params.hdr.root_flags) 848 root_mountflags &= ~MS_RDONLY; 849 init_mm.start_code = (unsigned long) _text; 850 init_mm.end_code = (unsigned long) _etext; 851 init_mm.end_data = (unsigned long) _edata; 852 init_mm.brk = _brk_end; 853 854 code_resource.start = __pa_symbol(_text); 855 code_resource.end = __pa_symbol(_etext)-1; 856 rodata_resource.start = __pa_symbol(__start_rodata); 857 rodata_resource.end = __pa_symbol(__end_rodata)-1; 858 data_resource.start = __pa_symbol(_sdata); 859 data_resource.end = __pa_symbol(_edata)-1; 860 bss_resource.start = __pa_symbol(__bss_start); 861 bss_resource.end = __pa_symbol(__bss_stop)-1; 862 863 #ifdef CONFIG_CMDLINE_BOOL 864 #ifdef CONFIG_CMDLINE_OVERRIDE 865 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 866 #else 867 if (builtin_cmdline[0]) { 868 /* append boot loader cmdline to builtin */ 869 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 870 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 871 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 872 } 873 #endif 874 #endif 875 876 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 877 *cmdline_p = command_line; 878 879 /* 880 * x86_configure_nx() is called before parse_early_param() to detect 881 * whether hardware doesn't support NX (so that the early EHCI debug 882 * console setup can safely call set_fixmap()). It may then be called 883 * again from within noexec_setup() during parsing early parameters 884 * to honor the respective command line option. 885 */ 886 x86_configure_nx(); 887 888 parse_early_param(); 889 890 /* 891 * Do some memory reservations *before* memory is added to 892 * memblock, so memblock allocations won't overwrite it. 893 * Do it after early param, so we could get (unlikely) panic from 894 * serial. 895 * 896 * After this point everything still needed from the boot loader or 897 * firmware or kernel text should be early reserved or marked not 898 * RAM in e820. All other memory is free game. 899 */ 900 early_reserve_memory(); 901 902 #ifdef CONFIG_MEMORY_HOTPLUG 903 /* 904 * Memory used by the kernel cannot be hot-removed because Linux 905 * cannot migrate the kernel pages. When memory hotplug is 906 * enabled, we should prevent memblock from allocating memory 907 * for the kernel. 908 * 909 * ACPI SRAT records all hotpluggable memory ranges. But before 910 * SRAT is parsed, we don't know about it. 911 * 912 * The kernel image is loaded into memory at very early time. We 913 * cannot prevent this anyway. So on NUMA system, we set any 914 * node the kernel resides in as un-hotpluggable. 915 * 916 * Since on modern servers, one node could have double-digit 917 * gigabytes memory, we can assume the memory around the kernel 918 * image is also un-hotpluggable. So before SRAT is parsed, just 919 * allocate memory near the kernel image to try the best to keep 920 * the kernel away from hotpluggable memory. 921 */ 922 if (movable_node_is_enabled()) 923 memblock_set_bottom_up(true); 924 #endif 925 926 x86_report_nx(); 927 928 if (acpi_mps_check()) { 929 #ifdef CONFIG_X86_LOCAL_APIC 930 disable_apic = 1; 931 #endif 932 setup_clear_cpu_cap(X86_FEATURE_APIC); 933 } 934 935 e820__reserve_setup_data(); 936 e820__finish_early_params(); 937 938 if (efi_enabled(EFI_BOOT)) 939 efi_init(); 940 941 dmi_setup(); 942 943 /* 944 * VMware detection requires dmi to be available, so this 945 * needs to be done after dmi_setup(), for the boot CPU. 946 */ 947 init_hypervisor_platform(); 948 949 tsc_early_init(); 950 x86_init.resources.probe_roms(); 951 952 /* after parse_early_param, so could debug it */ 953 insert_resource(&iomem_resource, &code_resource); 954 insert_resource(&iomem_resource, &rodata_resource); 955 insert_resource(&iomem_resource, &data_resource); 956 insert_resource(&iomem_resource, &bss_resource); 957 958 e820_add_kernel_range(); 959 trim_bios_range(); 960 #ifdef CONFIG_X86_32 961 if (ppro_with_ram_bug()) { 962 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 963 E820_TYPE_RESERVED); 964 e820__update_table(e820_table); 965 printk(KERN_INFO "fixed physical RAM map:\n"); 966 e820__print_table("bad_ppro"); 967 } 968 #else 969 early_gart_iommu_check(); 970 #endif 971 972 /* 973 * partially used pages are not usable - thus 974 * we are rounding upwards: 975 */ 976 max_pfn = e820__end_of_ram_pfn(); 977 978 /* update e820 for memory not covered by WB MTRRs */ 979 mtrr_bp_init(); 980 if (mtrr_trim_uncached_memory(max_pfn)) 981 max_pfn = e820__end_of_ram_pfn(); 982 983 max_possible_pfn = max_pfn; 984 985 /* 986 * This call is required when the CPU does not support PAT. If 987 * mtrr_bp_init() invoked it already via pat_init() the call has no 988 * effect. 989 */ 990 init_cache_modes(); 991 992 /* 993 * Define random base addresses for memory sections after max_pfn is 994 * defined and before each memory section base is used. 995 */ 996 kernel_randomize_memory(); 997 998 #ifdef CONFIG_X86_32 999 /* max_low_pfn get updated here */ 1000 find_low_pfn_range(); 1001 #else 1002 check_x2apic(); 1003 1004 /* How many end-of-memory variables you have, grandma! */ 1005 /* need this before calling reserve_initrd */ 1006 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1007 max_low_pfn = e820__end_of_low_ram_pfn(); 1008 else 1009 max_low_pfn = max_pfn; 1010 1011 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1012 #endif 1013 1014 /* 1015 * Find and reserve possible boot-time SMP configuration: 1016 */ 1017 find_smp_config(); 1018 1019 early_alloc_pgt_buf(); 1020 1021 /* 1022 * Need to conclude brk, before e820__memblock_setup() 1023 * it could use memblock_find_in_range, could overlap with 1024 * brk area. 1025 */ 1026 reserve_brk(); 1027 1028 cleanup_highmap(); 1029 1030 memblock_set_current_limit(ISA_END_ADDRESS); 1031 e820__memblock_setup(); 1032 1033 /* 1034 * Needs to run after memblock setup because it needs the physical 1035 * memory size. 1036 */ 1037 sev_setup_arch(); 1038 1039 efi_fake_memmap(); 1040 efi_find_mirror(); 1041 efi_esrt_init(); 1042 efi_mokvar_table_init(); 1043 1044 /* 1045 * The EFI specification says that boot service code won't be 1046 * called after ExitBootServices(). This is, in fact, a lie. 1047 */ 1048 efi_reserve_boot_services(); 1049 1050 /* preallocate 4k for mptable mpc */ 1051 e820__memblock_alloc_reserved_mpc_new(); 1052 1053 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1054 setup_bios_corruption_check(); 1055 #endif 1056 1057 #ifdef CONFIG_X86_32 1058 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1059 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1060 #endif 1061 1062 /* 1063 * Find free memory for the real mode trampoline and place it there. If 1064 * there is not enough free memory under 1M, on EFI-enabled systems 1065 * there will be additional attempt to reclaim the memory for the real 1066 * mode trampoline at efi_free_boot_services(). 1067 * 1068 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1069 * are known to corrupt low memory and several hundred kilobytes are not 1070 * worth complex detection what memory gets clobbered. Windows does the 1071 * same thing for very similar reasons. 1072 * 1073 * Moreover, on machines with SandyBridge graphics or in setups that use 1074 * crashkernel the entire 1M is reserved anyway. 1075 */ 1076 reserve_real_mode(); 1077 1078 init_mem_mapping(); 1079 1080 idt_setup_early_pf(); 1081 1082 /* 1083 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1084 * with the current CR4 value. This may not be necessary, but 1085 * auditing all the early-boot CR4 manipulation would be needed to 1086 * rule it out. 1087 * 1088 * Mask off features that don't work outside long mode (just 1089 * PCIDE for now). 1090 */ 1091 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1092 1093 memblock_set_current_limit(get_max_mapped()); 1094 1095 /* 1096 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1097 */ 1098 1099 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1100 if (init_ohci1394_dma_early) 1101 init_ohci1394_dma_on_all_controllers(); 1102 #endif 1103 /* Allocate bigger log buffer */ 1104 setup_log_buf(1); 1105 1106 if (efi_enabled(EFI_BOOT)) { 1107 switch (boot_params.secure_boot) { 1108 case efi_secureboot_mode_disabled: 1109 pr_info("Secure boot disabled\n"); 1110 break; 1111 case efi_secureboot_mode_enabled: 1112 pr_info("Secure boot enabled\n"); 1113 break; 1114 default: 1115 pr_info("Secure boot could not be determined\n"); 1116 break; 1117 } 1118 } 1119 1120 reserve_initrd(); 1121 1122 acpi_table_upgrade(); 1123 /* Look for ACPI tables and reserve memory occupied by them. */ 1124 acpi_boot_table_init(); 1125 1126 vsmp_init(); 1127 1128 io_delay_init(); 1129 1130 early_platform_quirks(); 1131 1132 early_acpi_boot_init(); 1133 1134 initmem_init(); 1135 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1136 1137 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1138 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1139 1140 /* 1141 * Reserve memory for crash kernel after SRAT is parsed so that it 1142 * won't consume hotpluggable memory. 1143 */ 1144 reserve_crashkernel(); 1145 1146 memblock_find_dma_reserve(); 1147 1148 if (!early_xdbc_setup_hardware()) 1149 early_xdbc_register_console(); 1150 1151 x86_init.paging.pagetable_init(); 1152 1153 kasan_init(); 1154 1155 /* 1156 * Sync back kernel address range. 1157 * 1158 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1159 * this call? 1160 */ 1161 sync_initial_page_table(); 1162 1163 tboot_probe(); 1164 1165 map_vsyscall(); 1166 1167 generic_apic_probe(); 1168 1169 early_quirks(); 1170 1171 /* 1172 * Read APIC and some other early information from ACPI tables. 1173 */ 1174 acpi_boot_init(); 1175 x86_dtb_init(); 1176 1177 /* 1178 * get boot-time SMP configuration: 1179 */ 1180 get_smp_config(); 1181 1182 /* 1183 * Systems w/o ACPI and mptables might not have it mapped the local 1184 * APIC yet, but prefill_possible_map() might need to access it. 1185 */ 1186 init_apic_mappings(); 1187 1188 prefill_possible_map(); 1189 1190 init_cpu_to_node(); 1191 init_gi_nodes(); 1192 1193 io_apic_init_mappings(); 1194 1195 x86_init.hyper.guest_late_init(); 1196 1197 e820__reserve_resources(); 1198 e820__register_nosave_regions(max_pfn); 1199 1200 x86_init.resources.reserve_resources(); 1201 1202 e820__setup_pci_gap(); 1203 1204 #ifdef CONFIG_VT 1205 #if defined(CONFIG_VGA_CONSOLE) 1206 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1207 conswitchp = &vga_con; 1208 #endif 1209 #endif 1210 x86_init.oem.banner(); 1211 1212 x86_init.timers.wallclock_init(); 1213 1214 /* 1215 * This needs to run before setup_local_APIC() which soft-disables the 1216 * local APIC temporarily and that masks the thermal LVT interrupt, 1217 * leading to softlockups on machines which have configured SMI 1218 * interrupt delivery. 1219 */ 1220 therm_lvt_init(); 1221 1222 mcheck_init(); 1223 1224 register_refined_jiffies(CLOCK_TICK_RATE); 1225 1226 #ifdef CONFIG_EFI 1227 if (efi_enabled(EFI_BOOT)) 1228 efi_apply_memmap_quirks(); 1229 #endif 1230 1231 unwind_init(); 1232 } 1233 1234 #ifdef CONFIG_X86_32 1235 1236 static struct resource video_ram_resource = { 1237 .name = "Video RAM area", 1238 .start = 0xa0000, 1239 .end = 0xbffff, 1240 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1241 }; 1242 1243 void __init i386_reserve_resources(void) 1244 { 1245 request_resource(&iomem_resource, &video_ram_resource); 1246 reserve_standard_io_resources(); 1247 } 1248 1249 #endif /* CONFIG_X86_32 */ 1250 1251 static struct notifier_block kernel_offset_notifier = { 1252 .notifier_call = dump_kernel_offset 1253 }; 1254 1255 static int __init register_kernel_offset_dumper(void) 1256 { 1257 atomic_notifier_chain_register(&panic_notifier_list, 1258 &kernel_offset_notifier); 1259 return 0; 1260 } 1261 __initcall(register_kernel_offset_dumper); 1262