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/dmi.h> 11 #include <linux/efi.h> 12 #include <linux/init_ohci1394_dma.h> 13 #include <linux/initrd.h> 14 #include <linux/iscsi_ibft.h> 15 #include <linux/memblock.h> 16 #include <linux/pci.h> 17 #include <linux/root_dev.h> 18 #include <linux/sfi.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 24 #include <uapi/linux/mount.h> 25 26 #include <xen/xen.h> 27 28 #include <asm/apic.h> 29 #include <asm/numa.h> 30 #include <asm/bios_ebda.h> 31 #include <asm/bugs.h> 32 #include <asm/cpu.h> 33 #include <asm/efi.h> 34 #include <asm/gart.h> 35 #include <asm/hypervisor.h> 36 #include <asm/io_apic.h> 37 #include <asm/kasan.h> 38 #include <asm/kaslr.h> 39 #include <asm/mce.h> 40 #include <asm/mtrr.h> 41 #include <asm/realmode.h> 42 #include <asm/olpc_ofw.h> 43 #include <asm/pci-direct.h> 44 #include <asm/prom.h> 45 #include <asm/proto.h> 46 #include <asm/unwind.h> 47 #include <asm/vsyscall.h> 48 #include <linux/vmalloc.h> 49 50 /* 51 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 52 * max_pfn_mapped: highest directly mapped pfn > 4 GB 53 * 54 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 55 * represented by pfn_mapped[]. 56 */ 57 unsigned long max_low_pfn_mapped; 58 unsigned long max_pfn_mapped; 59 60 #ifdef CONFIG_DMI 61 RESERVE_BRK(dmi_alloc, 65536); 62 #endif 63 64 65 /* 66 * Range of the BSS area. The size of the BSS area is determined 67 * at link time, with RESERVE_BRK*() facility reserving additional 68 * chunks. 69 */ 70 unsigned long _brk_start = (unsigned long)__brk_base; 71 unsigned long _brk_end = (unsigned long)__brk_base; 72 73 struct boot_params boot_params; 74 75 /* 76 * These are the four main kernel memory regions, we put them into 77 * the resource tree so that kdump tools and other debugging tools 78 * recover it: 79 */ 80 81 static struct resource rodata_resource = { 82 .name = "Kernel rodata", 83 .start = 0, 84 .end = 0, 85 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 86 }; 87 88 static struct resource data_resource = { 89 .name = "Kernel data", 90 .start = 0, 91 .end = 0, 92 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 93 }; 94 95 static struct resource code_resource = { 96 .name = "Kernel code", 97 .start = 0, 98 .end = 0, 99 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 100 }; 101 102 static struct resource bss_resource = { 103 .name = "Kernel bss", 104 .start = 0, 105 .end = 0, 106 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 107 }; 108 109 110 #ifdef CONFIG_X86_32 111 /* CPU data as detected by the assembly code in head_32.S */ 112 struct cpuinfo_x86 new_cpu_data; 113 114 /* Common CPU data for all CPUs */ 115 struct cpuinfo_x86 boot_cpu_data __read_mostly; 116 EXPORT_SYMBOL(boot_cpu_data); 117 118 unsigned int def_to_bigsmp; 119 120 /* For MCA, but anyone else can use it if they want */ 121 unsigned int machine_id; 122 unsigned int machine_submodel_id; 123 unsigned int BIOS_revision; 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_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_data *data; 373 u64 pa_data; 374 375 pa_data = boot_params.hdr.setup_data; 376 while (pa_data) { 377 data = early_memremap(pa_data, sizeof(*data)); 378 memblock_reserve(pa_data, sizeof(*data) + data->len); 379 380 if (data->type == SETUP_INDIRECT && 381 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) 382 memblock_reserve(((struct setup_indirect *)data->data)->addr, 383 ((struct setup_indirect *)data->data)->len); 384 385 pa_data = data->next; 386 early_memunmap(data, sizeof(*data)); 387 } 388 } 389 390 /* 391 * --------- Crashkernel reservation ------------------------------ 392 */ 393 394 #ifdef CONFIG_KEXEC_CORE 395 396 /* 16M alignment for crash kernel regions */ 397 #define CRASH_ALIGN SZ_16M 398 399 /* 400 * Keep the crash kernel below this limit. 401 * 402 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 403 * due to mapping restrictions. 404 * 405 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 406 * the upper limit of system RAM in 4-level paging mode. Since the kdump 407 * jump could be from 5-level paging to 4-level paging, the jump will fail if 408 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 409 * no good way to detect the paging mode of the target kernel which will be 410 * loaded for dumping. 411 */ 412 #ifdef CONFIG_X86_32 413 # define CRASH_ADDR_LOW_MAX SZ_512M 414 # define CRASH_ADDR_HIGH_MAX SZ_512M 415 #else 416 # define CRASH_ADDR_LOW_MAX SZ_4G 417 # define CRASH_ADDR_HIGH_MAX SZ_64T 418 #endif 419 420 static int __init reserve_crashkernel_low(void) 421 { 422 #ifdef CONFIG_X86_64 423 unsigned long long base, low_base = 0, low_size = 0; 424 unsigned long low_mem_limit; 425 int ret; 426 427 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 428 429 /* crashkernel=Y,low */ 430 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 431 if (ret) { 432 /* 433 * two parts from kernel/dma/swiotlb.c: 434 * -swiotlb size: user-specified with swiotlb= or default. 435 * 436 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 437 * to 8M for other buffers that may need to stay low too. Also 438 * make sure we allocate enough extra low memory so that we 439 * don't run out of DMA buffers for 32-bit devices. 440 */ 441 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 442 } else { 443 /* passed with crashkernel=0,low ? */ 444 if (!low_size) 445 return 0; 446 } 447 448 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 449 if (!low_base) { 450 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 451 (unsigned long)(low_size >> 20)); 452 return -ENOMEM; 453 } 454 455 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 456 (unsigned long)(low_size >> 20), 457 (unsigned long)(low_base >> 20), 458 (unsigned long)(low_mem_limit >> 20)); 459 460 crashk_low_res.start = low_base; 461 crashk_low_res.end = low_base + low_size - 1; 462 insert_resource(&iomem_resource, &crashk_low_res); 463 #endif 464 return 0; 465 } 466 467 static void __init reserve_crashkernel(void) 468 { 469 unsigned long long crash_size, crash_base, total_mem; 470 bool high = false; 471 int ret; 472 473 total_mem = memblock_phys_mem_size(); 474 475 /* crashkernel=XM */ 476 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 477 if (ret != 0 || crash_size <= 0) { 478 /* crashkernel=X,high */ 479 ret = parse_crashkernel_high(boot_command_line, total_mem, 480 &crash_size, &crash_base); 481 if (ret != 0 || crash_size <= 0) 482 return; 483 high = true; 484 } 485 486 if (xen_pv_domain()) { 487 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 488 return; 489 } 490 491 /* 0 means: find the address automatically */ 492 if (!crash_base) { 493 /* 494 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 495 * crashkernel=x,high reserves memory over 4G, also allocates 496 * 256M extra low memory for DMA buffers and swiotlb. 497 * But the extra memory is not required for all machines. 498 * So try low memory first and fall back to high memory 499 * unless "crashkernel=size[KMG],high" is specified. 500 */ 501 if (!high) 502 crash_base = memblock_phys_alloc_range(crash_size, 503 CRASH_ALIGN, CRASH_ALIGN, 504 CRASH_ADDR_LOW_MAX); 505 if (!crash_base) 506 crash_base = memblock_phys_alloc_range(crash_size, 507 CRASH_ALIGN, CRASH_ALIGN, 508 CRASH_ADDR_HIGH_MAX); 509 if (!crash_base) { 510 pr_info("crashkernel reservation failed - No suitable area found.\n"); 511 return; 512 } 513 } else { 514 unsigned long long start; 515 516 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 517 crash_base + crash_size); 518 if (start != crash_base) { 519 pr_info("crashkernel reservation failed - memory is in use.\n"); 520 return; 521 } 522 } 523 524 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 525 memblock_free(crash_base, crash_size); 526 return; 527 } 528 529 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 530 (unsigned long)(crash_size >> 20), 531 (unsigned long)(crash_base >> 20), 532 (unsigned long)(total_mem >> 20)); 533 534 crashk_res.start = crash_base; 535 crashk_res.end = crash_base + crash_size - 1; 536 insert_resource(&iomem_resource, &crashk_res); 537 } 538 #else 539 static void __init reserve_crashkernel(void) 540 { 541 } 542 #endif 543 544 static struct resource standard_io_resources[] = { 545 { .name = "dma1", .start = 0x00, .end = 0x1f, 546 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 547 { .name = "pic1", .start = 0x20, .end = 0x21, 548 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 549 { .name = "timer0", .start = 0x40, .end = 0x43, 550 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 551 { .name = "timer1", .start = 0x50, .end = 0x53, 552 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 553 { .name = "keyboard", .start = 0x60, .end = 0x60, 554 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 555 { .name = "keyboard", .start = 0x64, .end = 0x64, 556 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 557 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 558 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 559 { .name = "pic2", .start = 0xa0, .end = 0xa1, 560 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 561 { .name = "dma2", .start = 0xc0, .end = 0xdf, 562 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 563 { .name = "fpu", .start = 0xf0, .end = 0xff, 564 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 565 }; 566 567 void __init reserve_standard_io_resources(void) 568 { 569 int i; 570 571 /* request I/O space for devices used on all i[345]86 PCs */ 572 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 573 request_resource(&ioport_resource, &standard_io_resources[i]); 574 575 } 576 577 static __init void reserve_ibft_region(void) 578 { 579 unsigned long addr, size = 0; 580 581 addr = find_ibft_region(&size); 582 583 if (size) 584 memblock_reserve(addr, size); 585 } 586 587 static bool __init snb_gfx_workaround_needed(void) 588 { 589 #ifdef CONFIG_PCI 590 int i; 591 u16 vendor, devid; 592 static const __initconst u16 snb_ids[] = { 593 0x0102, 594 0x0112, 595 0x0122, 596 0x0106, 597 0x0116, 598 0x0126, 599 0x010a, 600 }; 601 602 /* Assume no if something weird is going on with PCI */ 603 if (!early_pci_allowed()) 604 return false; 605 606 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 607 if (vendor != 0x8086) 608 return false; 609 610 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 611 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 612 if (devid == snb_ids[i]) 613 return true; 614 #endif 615 616 return false; 617 } 618 619 /* 620 * Sandy Bridge graphics has trouble with certain ranges, exclude 621 * them from allocation. 622 */ 623 static void __init trim_snb_memory(void) 624 { 625 static const __initconst unsigned long bad_pages[] = { 626 0x20050000, 627 0x20110000, 628 0x20130000, 629 0x20138000, 630 0x40004000, 631 }; 632 int i; 633 634 if (!snb_gfx_workaround_needed()) 635 return; 636 637 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 638 639 /* 640 * Reserve all memory below the 1 MB mark that has not 641 * already been reserved. 642 */ 643 memblock_reserve(0, 1<<20); 644 645 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 646 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 647 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 648 bad_pages[i]); 649 } 650 } 651 652 /* 653 * Here we put platform-specific memory range workarounds, i.e. 654 * memory known to be corrupt or otherwise in need to be reserved on 655 * specific platforms. 656 * 657 * If this gets used more widely it could use a real dispatch mechanism. 658 */ 659 static void __init trim_platform_memory_ranges(void) 660 { 661 trim_snb_memory(); 662 } 663 664 static void __init trim_bios_range(void) 665 { 666 /* 667 * A special case is the first 4Kb of memory; 668 * This is a BIOS owned area, not kernel ram, but generally 669 * not listed as such in the E820 table. 670 * 671 * This typically reserves additional memory (64KiB by default) 672 * since some BIOSes are known to corrupt low memory. See the 673 * Kconfig help text for X86_RESERVE_LOW. 674 */ 675 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 676 677 /* 678 * special case: Some BIOSes report the PC BIOS 679 * area (640Kb -> 1Mb) as RAM even though it is not. 680 * take them out. 681 */ 682 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 683 684 e820__update_table(e820_table); 685 } 686 687 /* called before trim_bios_range() to spare extra sanitize */ 688 static void __init e820_add_kernel_range(void) 689 { 690 u64 start = __pa_symbol(_text); 691 u64 size = __pa_symbol(_end) - start; 692 693 /* 694 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 695 * attempt to fix it by adding the range. We may have a confused BIOS, 696 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 697 * exclude kernel range. If we really are running on top non-RAM, 698 * we will crash later anyways. 699 */ 700 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 701 return; 702 703 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 704 e820__range_remove(start, size, E820_TYPE_RAM, 0); 705 e820__range_add(start, size, E820_TYPE_RAM); 706 } 707 708 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 709 710 static int __init parse_reservelow(char *p) 711 { 712 unsigned long long size; 713 714 if (!p) 715 return -EINVAL; 716 717 size = memparse(p, &p); 718 719 if (size < 4096) 720 size = 4096; 721 722 if (size > 640*1024) 723 size = 640*1024; 724 725 reserve_low = size; 726 727 return 0; 728 } 729 730 early_param("reservelow", parse_reservelow); 731 732 static void __init trim_low_memory_range(void) 733 { 734 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 735 } 736 737 /* 738 * Dump out kernel offset information on panic. 739 */ 740 static int 741 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 742 { 743 if (kaslr_enabled()) { 744 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 745 kaslr_offset(), 746 __START_KERNEL, 747 __START_KERNEL_map, 748 MODULES_VADDR-1); 749 } else { 750 pr_emerg("Kernel Offset: disabled\n"); 751 } 752 753 return 0; 754 } 755 756 /* 757 * Determine if we were loaded by an EFI loader. If so, then we have also been 758 * passed the efi memmap, systab, etc., so we should use these data structures 759 * for initialization. Note, the efi init code path is determined by the 760 * global efi_enabled. This allows the same kernel image to be used on existing 761 * systems (with a traditional BIOS) as well as on EFI systems. 762 */ 763 /* 764 * setup_arch - architecture-specific boot-time initializations 765 * 766 * Note: On x86_64, fixmaps are ready for use even before this is called. 767 */ 768 769 void __init setup_arch(char **cmdline_p) 770 { 771 /* 772 * Reserve the memory occupied by the kernel between _text and 773 * __end_of_kernel_reserve symbols. Any kernel sections after the 774 * __end_of_kernel_reserve symbol must be explicitly reserved with a 775 * separate memblock_reserve() or they will be discarded. 776 */ 777 memblock_reserve(__pa_symbol(_text), 778 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 779 780 /* 781 * Make sure page 0 is always reserved because on systems with 782 * L1TF its contents can be leaked to user processes. 783 */ 784 memblock_reserve(0, PAGE_SIZE); 785 786 early_reserve_initrd(); 787 788 /* 789 * At this point everything still needed from the boot loader 790 * or BIOS or kernel text should be early reserved or marked not 791 * RAM in e820. All other memory is free game. 792 */ 793 794 #ifdef CONFIG_X86_32 795 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 796 797 /* 798 * copy kernel address range established so far and switch 799 * to the proper swapper page table 800 */ 801 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 802 initial_page_table + KERNEL_PGD_BOUNDARY, 803 KERNEL_PGD_PTRS); 804 805 load_cr3(swapper_pg_dir); 806 /* 807 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 808 * a cr3 based tlb flush, so the following __flush_tlb_all() 809 * will not flush anything because the CPU quirk which clears 810 * X86_FEATURE_PGE has not been invoked yet. Though due to the 811 * load_cr3() above the TLB has been flushed already. The 812 * quirk is invoked before subsequent calls to __flush_tlb_all() 813 * so proper operation is guaranteed. 814 */ 815 __flush_tlb_all(); 816 #else 817 printk(KERN_INFO "Command line: %s\n", boot_command_line); 818 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 819 #endif 820 821 /* 822 * If we have OLPC OFW, we might end up relocating the fixmap due to 823 * reserve_top(), so do this before touching the ioremap area. 824 */ 825 olpc_ofw_detect(); 826 827 idt_setup_early_traps(); 828 early_cpu_init(); 829 arch_init_ideal_nops(); 830 jump_label_init(); 831 static_call_init(); 832 early_ioremap_init(); 833 834 setup_olpc_ofw_pgd(); 835 836 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 837 screen_info = boot_params.screen_info; 838 edid_info = boot_params.edid_info; 839 #ifdef CONFIG_X86_32 840 apm_info.bios = boot_params.apm_bios_info; 841 ist_info = boot_params.ist_info; 842 #endif 843 saved_video_mode = boot_params.hdr.vid_mode; 844 bootloader_type = boot_params.hdr.type_of_loader; 845 if ((bootloader_type >> 4) == 0xe) { 846 bootloader_type &= 0xf; 847 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 848 } 849 bootloader_version = bootloader_type & 0xf; 850 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 851 852 #ifdef CONFIG_BLK_DEV_RAM 853 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 854 #endif 855 #ifdef CONFIG_EFI 856 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 857 EFI32_LOADER_SIGNATURE, 4)) { 858 set_bit(EFI_BOOT, &efi.flags); 859 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 860 EFI64_LOADER_SIGNATURE, 4)) { 861 set_bit(EFI_BOOT, &efi.flags); 862 set_bit(EFI_64BIT, &efi.flags); 863 } 864 #endif 865 866 x86_init.oem.arch_setup(); 867 868 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 869 e820__memory_setup(); 870 parse_setup_data(); 871 872 copy_edd(); 873 874 if (!boot_params.hdr.root_flags) 875 root_mountflags &= ~MS_RDONLY; 876 init_mm.start_code = (unsigned long) _text; 877 init_mm.end_code = (unsigned long) _etext; 878 init_mm.end_data = (unsigned long) _edata; 879 init_mm.brk = _brk_end; 880 881 code_resource.start = __pa_symbol(_text); 882 code_resource.end = __pa_symbol(_etext)-1; 883 rodata_resource.start = __pa_symbol(__start_rodata); 884 rodata_resource.end = __pa_symbol(__end_rodata)-1; 885 data_resource.start = __pa_symbol(_sdata); 886 data_resource.end = __pa_symbol(_edata)-1; 887 bss_resource.start = __pa_symbol(__bss_start); 888 bss_resource.end = __pa_symbol(__bss_stop)-1; 889 890 #ifdef CONFIG_CMDLINE_BOOL 891 #ifdef CONFIG_CMDLINE_OVERRIDE 892 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 893 #else 894 if (builtin_cmdline[0]) { 895 /* append boot loader cmdline to builtin */ 896 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 897 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 898 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 899 } 900 #endif 901 #endif 902 903 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 904 *cmdline_p = command_line; 905 906 /* 907 * x86_configure_nx() is called before parse_early_param() to detect 908 * whether hardware doesn't support NX (so that the early EHCI debug 909 * console setup can safely call set_fixmap()). It may then be called 910 * again from within noexec_setup() during parsing early parameters 911 * to honor the respective command line option. 912 */ 913 x86_configure_nx(); 914 915 parse_early_param(); 916 917 if (efi_enabled(EFI_BOOT)) 918 efi_memblock_x86_reserve_range(); 919 #ifdef CONFIG_MEMORY_HOTPLUG 920 /* 921 * Memory used by the kernel cannot be hot-removed because Linux 922 * cannot migrate the kernel pages. When memory hotplug is 923 * enabled, we should prevent memblock from allocating memory 924 * for the kernel. 925 * 926 * ACPI SRAT records all hotpluggable memory ranges. But before 927 * SRAT is parsed, we don't know about it. 928 * 929 * The kernel image is loaded into memory at very early time. We 930 * cannot prevent this anyway. So on NUMA system, we set any 931 * node the kernel resides in as un-hotpluggable. 932 * 933 * Since on modern servers, one node could have double-digit 934 * gigabytes memory, we can assume the memory around the kernel 935 * image is also un-hotpluggable. So before SRAT is parsed, just 936 * allocate memory near the kernel image to try the best to keep 937 * the kernel away from hotpluggable memory. 938 */ 939 if (movable_node_is_enabled()) 940 memblock_set_bottom_up(true); 941 #endif 942 943 x86_report_nx(); 944 945 /* after early param, so could get panic from serial */ 946 memblock_x86_reserve_range_setup_data(); 947 948 if (acpi_mps_check()) { 949 #ifdef CONFIG_X86_LOCAL_APIC 950 disable_apic = 1; 951 #endif 952 setup_clear_cpu_cap(X86_FEATURE_APIC); 953 } 954 955 e820__reserve_setup_data(); 956 e820__finish_early_params(); 957 958 if (efi_enabled(EFI_BOOT)) 959 efi_init(); 960 961 dmi_setup(); 962 963 /* 964 * VMware detection requires dmi to be available, so this 965 * needs to be done after dmi_setup(), for the boot CPU. 966 */ 967 init_hypervisor_platform(); 968 969 tsc_early_init(); 970 x86_init.resources.probe_roms(); 971 972 /* after parse_early_param, so could debug it */ 973 insert_resource(&iomem_resource, &code_resource); 974 insert_resource(&iomem_resource, &rodata_resource); 975 insert_resource(&iomem_resource, &data_resource); 976 insert_resource(&iomem_resource, &bss_resource); 977 978 e820_add_kernel_range(); 979 trim_bios_range(); 980 #ifdef CONFIG_X86_32 981 if (ppro_with_ram_bug()) { 982 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 983 E820_TYPE_RESERVED); 984 e820__update_table(e820_table); 985 printk(KERN_INFO "fixed physical RAM map:\n"); 986 e820__print_table("bad_ppro"); 987 } 988 #else 989 early_gart_iommu_check(); 990 #endif 991 992 /* 993 * partially used pages are not usable - thus 994 * we are rounding upwards: 995 */ 996 max_pfn = e820__end_of_ram_pfn(); 997 998 /* update e820 for memory not covered by WB MTRRs */ 999 mtrr_bp_init(); 1000 if (mtrr_trim_uncached_memory(max_pfn)) 1001 max_pfn = e820__end_of_ram_pfn(); 1002 1003 max_possible_pfn = max_pfn; 1004 1005 /* 1006 * This call is required when the CPU does not support PAT. If 1007 * mtrr_bp_init() invoked it already via pat_init() the call has no 1008 * effect. 1009 */ 1010 init_cache_modes(); 1011 1012 /* 1013 * Define random base addresses for memory sections after max_pfn is 1014 * defined and before each memory section base is used. 1015 */ 1016 kernel_randomize_memory(); 1017 1018 #ifdef CONFIG_X86_32 1019 /* max_low_pfn get updated here */ 1020 find_low_pfn_range(); 1021 #else 1022 check_x2apic(); 1023 1024 /* How many end-of-memory variables you have, grandma! */ 1025 /* need this before calling reserve_initrd */ 1026 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1027 max_low_pfn = e820__end_of_low_ram_pfn(); 1028 else 1029 max_low_pfn = max_pfn; 1030 1031 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1032 #endif 1033 1034 /* 1035 * Find and reserve possible boot-time SMP configuration: 1036 */ 1037 find_smp_config(); 1038 1039 reserve_ibft_region(); 1040 1041 early_alloc_pgt_buf(); 1042 1043 /* 1044 * Need to conclude brk, before e820__memblock_setup() 1045 * it could use memblock_find_in_range, could overlap with 1046 * brk area. 1047 */ 1048 reserve_brk(); 1049 1050 cleanup_highmap(); 1051 1052 memblock_set_current_limit(ISA_END_ADDRESS); 1053 e820__memblock_setup(); 1054 1055 reserve_bios_regions(); 1056 1057 efi_fake_memmap(); 1058 efi_find_mirror(); 1059 efi_esrt_init(); 1060 efi_mokvar_table_init(); 1061 1062 /* 1063 * The EFI specification says that boot service code won't be 1064 * called after ExitBootServices(). This is, in fact, a lie. 1065 */ 1066 efi_reserve_boot_services(); 1067 1068 /* preallocate 4k for mptable mpc */ 1069 e820__memblock_alloc_reserved_mpc_new(); 1070 1071 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1072 setup_bios_corruption_check(); 1073 #endif 1074 1075 #ifdef CONFIG_X86_32 1076 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1077 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1078 #endif 1079 1080 reserve_real_mode(); 1081 1082 trim_platform_memory_ranges(); 1083 trim_low_memory_range(); 1084 1085 init_mem_mapping(); 1086 1087 idt_setup_early_pf(); 1088 1089 /* 1090 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1091 * with the current CR4 value. This may not be necessary, but 1092 * auditing all the early-boot CR4 manipulation would be needed to 1093 * rule it out. 1094 * 1095 * Mask off features that don't work outside long mode (just 1096 * PCIDE for now). 1097 */ 1098 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1099 1100 memblock_set_current_limit(get_max_mapped()); 1101 1102 /* 1103 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1104 */ 1105 1106 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1107 if (init_ohci1394_dma_early) 1108 init_ohci1394_dma_on_all_controllers(); 1109 #endif 1110 /* Allocate bigger log buffer */ 1111 setup_log_buf(1); 1112 1113 if (efi_enabled(EFI_BOOT)) { 1114 switch (boot_params.secure_boot) { 1115 case efi_secureboot_mode_disabled: 1116 pr_info("Secure boot disabled\n"); 1117 break; 1118 case efi_secureboot_mode_enabled: 1119 pr_info("Secure boot enabled\n"); 1120 break; 1121 default: 1122 pr_info("Secure boot could not be determined\n"); 1123 break; 1124 } 1125 } 1126 1127 reserve_initrd(); 1128 1129 acpi_table_upgrade(); 1130 1131 vsmp_init(); 1132 1133 io_delay_init(); 1134 1135 early_platform_quirks(); 1136 1137 /* 1138 * Parse the ACPI tables for possible boot-time SMP configuration. 1139 */ 1140 acpi_boot_table_init(); 1141 1142 early_acpi_boot_init(); 1143 1144 initmem_init(); 1145 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1146 1147 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1148 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1149 1150 /* 1151 * Reserve memory for crash kernel after SRAT is parsed so that it 1152 * won't consume hotpluggable memory. 1153 */ 1154 reserve_crashkernel(); 1155 1156 memblock_find_dma_reserve(); 1157 1158 if (!early_xdbc_setup_hardware()) 1159 early_xdbc_register_console(); 1160 1161 x86_init.paging.pagetable_init(); 1162 1163 kasan_init(); 1164 1165 /* 1166 * Sync back kernel address range. 1167 * 1168 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1169 * this call? 1170 */ 1171 sync_initial_page_table(); 1172 1173 tboot_probe(); 1174 1175 map_vsyscall(); 1176 1177 generic_apic_probe(); 1178 1179 early_quirks(); 1180 1181 /* 1182 * Read APIC and some other early information from ACPI tables. 1183 */ 1184 acpi_boot_init(); 1185 sfi_init(); 1186 x86_dtb_init(); 1187 1188 /* 1189 * get boot-time SMP configuration: 1190 */ 1191 get_smp_config(); 1192 1193 /* 1194 * Systems w/o ACPI and mptables might not have it mapped the local 1195 * APIC yet, but prefill_possible_map() might need to access it. 1196 */ 1197 init_apic_mappings(); 1198 1199 prefill_possible_map(); 1200 1201 init_cpu_to_node(); 1202 init_gi_nodes(); 1203 1204 io_apic_init_mappings(); 1205 1206 x86_init.hyper.guest_late_init(); 1207 1208 e820__reserve_resources(); 1209 e820__register_nosave_regions(max_pfn); 1210 1211 x86_init.resources.reserve_resources(); 1212 1213 e820__setup_pci_gap(); 1214 1215 #ifdef CONFIG_VT 1216 #if defined(CONFIG_VGA_CONSOLE) 1217 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1218 conswitchp = &vga_con; 1219 #endif 1220 #endif 1221 x86_init.oem.banner(); 1222 1223 x86_init.timers.wallclock_init(); 1224 1225 mcheck_init(); 1226 1227 register_refined_jiffies(CLOCK_TICK_RATE); 1228 1229 #ifdef CONFIG_EFI 1230 if (efi_enabled(EFI_BOOT)) 1231 efi_apply_memmap_quirks(); 1232 #endif 1233 1234 unwind_init(); 1235 } 1236 1237 #ifdef CONFIG_X86_32 1238 1239 static struct resource video_ram_resource = { 1240 .name = "Video RAM area", 1241 .start = 0xa0000, 1242 .end = 0xbffff, 1243 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1244 }; 1245 1246 void __init i386_reserve_resources(void) 1247 { 1248 request_resource(&iomem_resource, &video_ram_resource); 1249 reserve_standard_io_resources(); 1250 } 1251 1252 #endif /* CONFIG_X86_32 */ 1253 1254 static struct notifier_block kernel_offset_notifier = { 1255 .notifier_call = dump_kernel_offset 1256 }; 1257 1258 static int __init register_kernel_offset_dumper(void) 1259 { 1260 atomic_notifier_chain_register(&panic_notifier_list, 1261 &kernel_offset_notifier); 1262 return 0; 1263 } 1264 __initcall(register_kernel_offset_dumper); 1265