1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1995 Linus Torvalds 4 * 5 * This file contains the setup_arch() code, which handles the architecture-dependent 6 * parts of early kernel initialization. 7 */ 8 #include <linux/acpi.h> 9 #include <linux/console.h> 10 #include <linux/crash_dump.h> 11 #include <linux/dma-map-ops.h> 12 #include <linux/dmi.h> 13 #include <linux/efi.h> 14 #include <linux/ima.h> 15 #include <linux/init_ohci1394_dma.h> 16 #include <linux/initrd.h> 17 #include <linux/iscsi_ibft.h> 18 #include <linux/memblock.h> 19 #include <linux/panic_notifier.h> 20 #include <linux/pci.h> 21 #include <linux/root_dev.h> 22 #include <linux/hugetlb.h> 23 #include <linux/tboot.h> 24 #include <linux/usb/xhci-dbgp.h> 25 #include <linux/static_call.h> 26 #include <linux/swiotlb.h> 27 #include <linux/random.h> 28 29 #include <uapi/linux/mount.h> 30 31 #include <xen/xen.h> 32 33 #include <asm/apic.h> 34 #include <asm/numa.h> 35 #include <asm/bios_ebda.h> 36 #include <asm/bugs.h> 37 #include <asm/cpu.h> 38 #include <asm/efi.h> 39 #include <asm/gart.h> 40 #include <asm/hypervisor.h> 41 #include <asm/io_apic.h> 42 #include <asm/kasan.h> 43 #include <asm/kaslr.h> 44 #include <asm/mce.h> 45 #include <asm/memtype.h> 46 #include <asm/mtrr.h> 47 #include <asm/realmode.h> 48 #include <asm/olpc_ofw.h> 49 #include <asm/pci-direct.h> 50 #include <asm/prom.h> 51 #include <asm/proto.h> 52 #include <asm/thermal.h> 53 #include <asm/unwind.h> 54 #include <asm/vsyscall.h> 55 #include <linux/vmalloc.h> 56 57 /* 58 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 59 * max_pfn_mapped: highest directly mapped pfn > 4 GB 60 * 61 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 62 * represented by pfn_mapped[]. 63 */ 64 unsigned long max_low_pfn_mapped; 65 unsigned long max_pfn_mapped; 66 67 #ifdef CONFIG_DMI 68 RESERVE_BRK(dmi_alloc, 65536); 69 #endif 70 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 #ifdef CONFIG_IMA 146 static phys_addr_t ima_kexec_buffer_phys; 147 static size_t ima_kexec_buffer_size; 148 #endif 149 150 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 151 int bootloader_type, bootloader_version; 152 153 /* 154 * Setup options 155 */ 156 struct screen_info screen_info; 157 EXPORT_SYMBOL(screen_info); 158 struct edid_info edid_info; 159 EXPORT_SYMBOL_GPL(edid_info); 160 161 extern int root_mountflags; 162 163 unsigned long saved_video_mode; 164 165 #define RAMDISK_IMAGE_START_MASK 0x07FF 166 #define RAMDISK_PROMPT_FLAG 0x8000 167 #define RAMDISK_LOAD_FLAG 0x4000 168 169 static char __initdata command_line[COMMAND_LINE_SIZE]; 170 #ifdef CONFIG_CMDLINE_BOOL 171 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 172 #endif 173 174 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 175 struct edd edd; 176 #ifdef CONFIG_EDD_MODULE 177 EXPORT_SYMBOL(edd); 178 #endif 179 /** 180 * copy_edd() - Copy the BIOS EDD information 181 * from boot_params into a safe place. 182 * 183 */ 184 static inline void __init copy_edd(void) 185 { 186 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 187 sizeof(edd.mbr_signature)); 188 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 189 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 190 edd.edd_info_nr = boot_params.eddbuf_entries; 191 } 192 #else 193 static inline void __init copy_edd(void) 194 { 195 } 196 #endif 197 198 void * __init extend_brk(size_t size, size_t align) 199 { 200 size_t mask = align - 1; 201 void *ret; 202 203 BUG_ON(_brk_start == 0); 204 BUG_ON(align & mask); 205 206 _brk_end = (_brk_end + mask) & ~mask; 207 BUG_ON((char *)(_brk_end + size) > __brk_limit); 208 209 ret = (void *)_brk_end; 210 _brk_end += size; 211 212 memset(ret, 0, size); 213 214 return ret; 215 } 216 217 #ifdef CONFIG_X86_32 218 static void __init cleanup_highmap(void) 219 { 220 } 221 #endif 222 223 static void __init reserve_brk(void) 224 { 225 if (_brk_end > _brk_start) 226 memblock_reserve(__pa_symbol(_brk_start), 227 _brk_end - _brk_start); 228 229 /* Mark brk area as locked down and no longer taking any 230 new allocations */ 231 _brk_start = 0; 232 } 233 234 u64 relocated_ramdisk; 235 236 #ifdef CONFIG_BLK_DEV_INITRD 237 238 static u64 __init get_ramdisk_image(void) 239 { 240 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 241 242 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 243 244 if (ramdisk_image == 0) 245 ramdisk_image = phys_initrd_start; 246 247 return ramdisk_image; 248 } 249 static u64 __init get_ramdisk_size(void) 250 { 251 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 252 253 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 254 255 if (ramdisk_size == 0) 256 ramdisk_size = phys_initrd_size; 257 258 return ramdisk_size; 259 } 260 261 static void __init relocate_initrd(void) 262 { 263 /* Assume only end is not page aligned */ 264 u64 ramdisk_image = get_ramdisk_image(); 265 u64 ramdisk_size = get_ramdisk_size(); 266 u64 area_size = PAGE_ALIGN(ramdisk_size); 267 268 /* We need to move the initrd down into directly mapped mem */ 269 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 270 PFN_PHYS(max_pfn_mapped)); 271 if (!relocated_ramdisk) 272 panic("Cannot find place for new RAMDISK of size %lld\n", 273 ramdisk_size); 274 275 initrd_start = relocated_ramdisk + PAGE_OFFSET; 276 initrd_end = initrd_start + ramdisk_size; 277 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 278 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 279 280 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 281 282 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 283 " [mem %#010llx-%#010llx]\n", 284 ramdisk_image, ramdisk_image + ramdisk_size - 1, 285 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 286 } 287 288 static void __init early_reserve_initrd(void) 289 { 290 /* Assume only end is not page aligned */ 291 u64 ramdisk_image = get_ramdisk_image(); 292 u64 ramdisk_size = get_ramdisk_size(); 293 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 294 295 if (!boot_params.hdr.type_of_loader || 296 !ramdisk_image || !ramdisk_size) 297 return; /* No initrd provided by bootloader */ 298 299 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 300 } 301 302 static void __init reserve_initrd(void) 303 { 304 /* Assume only end is not page aligned */ 305 u64 ramdisk_image = get_ramdisk_image(); 306 u64 ramdisk_size = get_ramdisk_size(); 307 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 308 309 if (!boot_params.hdr.type_of_loader || 310 !ramdisk_image || !ramdisk_size) 311 return; /* No initrd provided by bootloader */ 312 313 initrd_start = 0; 314 315 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 316 ramdisk_end - 1); 317 318 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 319 PFN_DOWN(ramdisk_end))) { 320 /* All are mapped, easy case */ 321 initrd_start = ramdisk_image + PAGE_OFFSET; 322 initrd_end = initrd_start + ramdisk_size; 323 return; 324 } 325 326 relocate_initrd(); 327 328 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image); 329 } 330 331 #else 332 static void __init early_reserve_initrd(void) 333 { 334 } 335 static void __init reserve_initrd(void) 336 { 337 } 338 #endif /* CONFIG_BLK_DEV_INITRD */ 339 340 static void __init add_early_ima_buffer(u64 phys_addr) 341 { 342 #ifdef CONFIG_IMA 343 struct ima_setup_data *data; 344 345 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data)); 346 if (!data) { 347 pr_warn("setup: failed to memremap ima_setup_data entry\n"); 348 return; 349 } 350 351 if (data->size) { 352 memblock_reserve(data->addr, data->size); 353 ima_kexec_buffer_phys = data->addr; 354 ima_kexec_buffer_size = data->size; 355 } 356 357 early_memunmap(data, sizeof(*data)); 358 #else 359 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n"); 360 #endif 361 } 362 363 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE) 364 int __init ima_free_kexec_buffer(void) 365 { 366 int rc; 367 368 if (!ima_kexec_buffer_size) 369 return -ENOENT; 370 371 rc = memblock_phys_free(ima_kexec_buffer_phys, 372 ima_kexec_buffer_size); 373 if (rc) 374 return rc; 375 376 ima_kexec_buffer_phys = 0; 377 ima_kexec_buffer_size = 0; 378 379 return 0; 380 } 381 382 int __init ima_get_kexec_buffer(void **addr, size_t *size) 383 { 384 if (!ima_kexec_buffer_size) 385 return -ENOENT; 386 387 *addr = __va(ima_kexec_buffer_phys); 388 *size = ima_kexec_buffer_size; 389 390 return 0; 391 } 392 #endif 393 394 static void __init parse_setup_data(void) 395 { 396 struct setup_data *data; 397 u64 pa_data, pa_next; 398 399 pa_data = boot_params.hdr.setup_data; 400 while (pa_data) { 401 u32 data_len, data_type; 402 403 data = early_memremap(pa_data, sizeof(*data)); 404 data_len = data->len + sizeof(struct setup_data); 405 data_type = data->type; 406 pa_next = data->next; 407 early_memunmap(data, sizeof(*data)); 408 409 switch (data_type) { 410 case SETUP_E820_EXT: 411 e820__memory_setup_extended(pa_data, data_len); 412 break; 413 case SETUP_DTB: 414 add_dtb(pa_data); 415 break; 416 case SETUP_EFI: 417 parse_efi_setup(pa_data, data_len); 418 break; 419 case SETUP_IMA: 420 add_early_ima_buffer(pa_data); 421 break; 422 case SETUP_RNG_SEED: 423 data = early_memremap(pa_data, data_len); 424 add_bootloader_randomness(data->data, data->len); 425 /* Zero seed for forward secrecy. */ 426 memzero_explicit(data->data, data->len); 427 /* Zero length in case we find ourselves back here by accident. */ 428 memzero_explicit(&data->len, sizeof(data->len)); 429 early_memunmap(data, data_len); 430 break; 431 default: 432 break; 433 } 434 pa_data = pa_next; 435 } 436 } 437 438 static void __init memblock_x86_reserve_range_setup_data(void) 439 { 440 struct setup_indirect *indirect; 441 struct setup_data *data; 442 u64 pa_data, pa_next; 443 u32 len; 444 445 pa_data = boot_params.hdr.setup_data; 446 while (pa_data) { 447 data = early_memremap(pa_data, sizeof(*data)); 448 if (!data) { 449 pr_warn("setup: failed to memremap setup_data entry\n"); 450 return; 451 } 452 453 len = sizeof(*data); 454 pa_next = data->next; 455 456 memblock_reserve(pa_data, sizeof(*data) + data->len); 457 458 if (data->type == SETUP_INDIRECT) { 459 len += data->len; 460 early_memunmap(data, sizeof(*data)); 461 data = early_memremap(pa_data, len); 462 if (!data) { 463 pr_warn("setup: failed to memremap indirect setup_data\n"); 464 return; 465 } 466 467 indirect = (struct setup_indirect *)data->data; 468 469 if (indirect->type != SETUP_INDIRECT) 470 memblock_reserve(indirect->addr, indirect->len); 471 } 472 473 pa_data = pa_next; 474 early_memunmap(data, len); 475 } 476 } 477 478 /* 479 * --------- Crashkernel reservation ------------------------------ 480 */ 481 482 /* 16M alignment for crash kernel regions */ 483 #define CRASH_ALIGN SZ_16M 484 485 /* 486 * Keep the crash kernel below this limit. 487 * 488 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 489 * due to mapping restrictions. 490 * 491 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 492 * the upper limit of system RAM in 4-level paging mode. Since the kdump 493 * jump could be from 5-level paging to 4-level paging, the jump will fail if 494 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 495 * no good way to detect the paging mode of the target kernel which will be 496 * loaded for dumping. 497 */ 498 #ifdef CONFIG_X86_32 499 # define CRASH_ADDR_LOW_MAX SZ_512M 500 # define CRASH_ADDR_HIGH_MAX SZ_512M 501 #else 502 # define CRASH_ADDR_LOW_MAX SZ_4G 503 # define CRASH_ADDR_HIGH_MAX SZ_64T 504 #endif 505 506 static int __init reserve_crashkernel_low(void) 507 { 508 #ifdef CONFIG_X86_64 509 unsigned long long base, low_base = 0, low_size = 0; 510 unsigned long low_mem_limit; 511 int ret; 512 513 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 514 515 /* crashkernel=Y,low */ 516 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 517 if (ret) { 518 /* 519 * two parts from kernel/dma/swiotlb.c: 520 * -swiotlb size: user-specified with swiotlb= or default. 521 * 522 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 523 * to 8M for other buffers that may need to stay low too. Also 524 * make sure we allocate enough extra low memory so that we 525 * don't run out of DMA buffers for 32-bit devices. 526 */ 527 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 528 } else { 529 /* passed with crashkernel=0,low ? */ 530 if (!low_size) 531 return 0; 532 } 533 534 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 535 if (!low_base) { 536 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 537 (unsigned long)(low_size >> 20)); 538 return -ENOMEM; 539 } 540 541 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 542 (unsigned long)(low_size >> 20), 543 (unsigned long)(low_base >> 20), 544 (unsigned long)(low_mem_limit >> 20)); 545 546 crashk_low_res.start = low_base; 547 crashk_low_res.end = low_base + low_size - 1; 548 insert_resource(&iomem_resource, &crashk_low_res); 549 #endif 550 return 0; 551 } 552 553 static void __init reserve_crashkernel(void) 554 { 555 unsigned long long crash_size, crash_base, total_mem; 556 bool high = false; 557 int ret; 558 559 if (!IS_ENABLED(CONFIG_KEXEC_CORE)) 560 return; 561 562 total_mem = memblock_phys_mem_size(); 563 564 /* crashkernel=XM */ 565 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 566 if (ret != 0 || crash_size <= 0) { 567 /* crashkernel=X,high */ 568 ret = parse_crashkernel_high(boot_command_line, total_mem, 569 &crash_size, &crash_base); 570 if (ret != 0 || crash_size <= 0) 571 return; 572 high = true; 573 } 574 575 if (xen_pv_domain()) { 576 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 577 return; 578 } 579 580 /* 0 means: find the address automatically */ 581 if (!crash_base) { 582 /* 583 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 584 * crashkernel=x,high reserves memory over 4G, also allocates 585 * 256M extra low memory for DMA buffers and swiotlb. 586 * But the extra memory is not required for all machines. 587 * So try low memory first and fall back to high memory 588 * unless "crashkernel=size[KMG],high" is specified. 589 */ 590 if (!high) 591 crash_base = memblock_phys_alloc_range(crash_size, 592 CRASH_ALIGN, CRASH_ALIGN, 593 CRASH_ADDR_LOW_MAX); 594 if (!crash_base) 595 crash_base = memblock_phys_alloc_range(crash_size, 596 CRASH_ALIGN, CRASH_ALIGN, 597 CRASH_ADDR_HIGH_MAX); 598 if (!crash_base) { 599 pr_info("crashkernel reservation failed - No suitable area found.\n"); 600 return; 601 } 602 } else { 603 unsigned long long start; 604 605 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 606 crash_base + crash_size); 607 if (start != crash_base) { 608 pr_info("crashkernel reservation failed - memory is in use.\n"); 609 return; 610 } 611 } 612 613 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 614 memblock_phys_free(crash_base, crash_size); 615 return; 616 } 617 618 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 619 (unsigned long)(crash_size >> 20), 620 (unsigned long)(crash_base >> 20), 621 (unsigned long)(total_mem >> 20)); 622 623 crashk_res.start = crash_base; 624 crashk_res.end = crash_base + crash_size - 1; 625 insert_resource(&iomem_resource, &crashk_res); 626 } 627 628 static struct resource standard_io_resources[] = { 629 { .name = "dma1", .start = 0x00, .end = 0x1f, 630 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 631 { .name = "pic1", .start = 0x20, .end = 0x21, 632 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 633 { .name = "timer0", .start = 0x40, .end = 0x43, 634 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 635 { .name = "timer1", .start = 0x50, .end = 0x53, 636 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 637 { .name = "keyboard", .start = 0x60, .end = 0x60, 638 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 639 { .name = "keyboard", .start = 0x64, .end = 0x64, 640 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 641 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 642 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 643 { .name = "pic2", .start = 0xa0, .end = 0xa1, 644 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 645 { .name = "dma2", .start = 0xc0, .end = 0xdf, 646 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 647 { .name = "fpu", .start = 0xf0, .end = 0xff, 648 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 649 }; 650 651 void __init reserve_standard_io_resources(void) 652 { 653 int i; 654 655 /* request I/O space for devices used on all i[345]86 PCs */ 656 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 657 request_resource(&ioport_resource, &standard_io_resources[i]); 658 659 } 660 661 static bool __init snb_gfx_workaround_needed(void) 662 { 663 #ifdef CONFIG_PCI 664 int i; 665 u16 vendor, devid; 666 static const __initconst u16 snb_ids[] = { 667 0x0102, 668 0x0112, 669 0x0122, 670 0x0106, 671 0x0116, 672 0x0126, 673 0x010a, 674 }; 675 676 /* Assume no if something weird is going on with PCI */ 677 if (!early_pci_allowed()) 678 return false; 679 680 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 681 if (vendor != 0x8086) 682 return false; 683 684 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 685 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 686 if (devid == snb_ids[i]) 687 return true; 688 #endif 689 690 return false; 691 } 692 693 /* 694 * Sandy Bridge graphics has trouble with certain ranges, exclude 695 * them from allocation. 696 */ 697 static void __init trim_snb_memory(void) 698 { 699 static const __initconst unsigned long bad_pages[] = { 700 0x20050000, 701 0x20110000, 702 0x20130000, 703 0x20138000, 704 0x40004000, 705 }; 706 int i; 707 708 if (!snb_gfx_workaround_needed()) 709 return; 710 711 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 712 713 /* 714 * SandyBridge integrated graphics devices have a bug that prevents 715 * them from accessing certain memory ranges, namely anything below 716 * 1M and in the pages listed in bad_pages[] above. 717 * 718 * To avoid these pages being ever accessed by SNB gfx devices reserve 719 * bad_pages that have not already been reserved at boot time. 720 * All memory below the 1 MB mark is anyway reserved later during 721 * setup_arch(), so there is no need to reserve it here. 722 */ 723 724 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 725 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 726 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 727 bad_pages[i]); 728 } 729 } 730 731 static void __init trim_bios_range(void) 732 { 733 /* 734 * A special case is the first 4Kb of memory; 735 * This is a BIOS owned area, not kernel ram, but generally 736 * not listed as such in the E820 table. 737 * 738 * This typically reserves additional memory (64KiB by default) 739 * since some BIOSes are known to corrupt low memory. See the 740 * Kconfig help text for X86_RESERVE_LOW. 741 */ 742 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 743 744 /* 745 * special case: Some BIOSes report the PC BIOS 746 * area (640Kb -> 1Mb) as RAM even though it is not. 747 * take them out. 748 */ 749 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 750 751 e820__update_table(e820_table); 752 } 753 754 /* called before trim_bios_range() to spare extra sanitize */ 755 static void __init e820_add_kernel_range(void) 756 { 757 u64 start = __pa_symbol(_text); 758 u64 size = __pa_symbol(_end) - start; 759 760 /* 761 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 762 * attempt to fix it by adding the range. We may have a confused BIOS, 763 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 764 * exclude kernel range. If we really are running on top non-RAM, 765 * we will crash later anyways. 766 */ 767 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 768 return; 769 770 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 771 e820__range_remove(start, size, E820_TYPE_RAM, 0); 772 e820__range_add(start, size, E820_TYPE_RAM); 773 } 774 775 static void __init early_reserve_memory(void) 776 { 777 /* 778 * Reserve the memory occupied by the kernel between _text and 779 * __end_of_kernel_reserve symbols. Any kernel sections after the 780 * __end_of_kernel_reserve symbol must be explicitly reserved with a 781 * separate memblock_reserve() or they will be discarded. 782 */ 783 memblock_reserve(__pa_symbol(_text), 784 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 785 786 /* 787 * The first 4Kb of memory is a BIOS owned area, but generally it is 788 * not listed as such in the E820 table. 789 * 790 * Reserve the first 64K of memory since some BIOSes are known to 791 * corrupt low memory. After the real mode trampoline is allocated the 792 * rest of the memory below 640k is reserved. 793 * 794 * In addition, make sure page 0 is always reserved because on 795 * systems with L1TF its contents can be leaked to user processes. 796 */ 797 memblock_reserve(0, SZ_64K); 798 799 early_reserve_initrd(); 800 801 memblock_x86_reserve_range_setup_data(); 802 803 reserve_ibft_region(); 804 reserve_bios_regions(); 805 trim_snb_memory(); 806 } 807 808 /* 809 * Dump out kernel offset information on panic. 810 */ 811 static int 812 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 813 { 814 if (kaslr_enabled()) { 815 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 816 kaslr_offset(), 817 __START_KERNEL, 818 __START_KERNEL_map, 819 MODULES_VADDR-1); 820 } else { 821 pr_emerg("Kernel Offset: disabled\n"); 822 } 823 824 return 0; 825 } 826 827 void x86_configure_nx(void) 828 { 829 if (boot_cpu_has(X86_FEATURE_NX)) 830 __supported_pte_mask |= _PAGE_NX; 831 else 832 __supported_pte_mask &= ~_PAGE_NX; 833 } 834 835 static void __init x86_report_nx(void) 836 { 837 if (!boot_cpu_has(X86_FEATURE_NX)) { 838 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 839 "missing in CPU!\n"); 840 } else { 841 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 842 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 843 #else 844 /* 32bit non-PAE kernel, NX cannot be used */ 845 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 846 "cannot be enabled: non-PAE kernel!\n"); 847 #endif 848 } 849 } 850 851 /* 852 * Determine if we were loaded by an EFI loader. If so, then we have also been 853 * passed the efi memmap, systab, etc., so we should use these data structures 854 * for initialization. Note, the efi init code path is determined by the 855 * global efi_enabled. This allows the same kernel image to be used on existing 856 * systems (with a traditional BIOS) as well as on EFI systems. 857 */ 858 /* 859 * setup_arch - architecture-specific boot-time initializations 860 * 861 * Note: On x86_64, fixmaps are ready for use even before this is called. 862 */ 863 864 void __init setup_arch(char **cmdline_p) 865 { 866 #ifdef CONFIG_X86_32 867 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 868 869 /* 870 * copy kernel address range established so far and switch 871 * to the proper swapper page table 872 */ 873 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 874 initial_page_table + KERNEL_PGD_BOUNDARY, 875 KERNEL_PGD_PTRS); 876 877 load_cr3(swapper_pg_dir); 878 /* 879 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 880 * a cr3 based tlb flush, so the following __flush_tlb_all() 881 * will not flush anything because the CPU quirk which clears 882 * X86_FEATURE_PGE has not been invoked yet. Though due to the 883 * load_cr3() above the TLB has been flushed already. The 884 * quirk is invoked before subsequent calls to __flush_tlb_all() 885 * so proper operation is guaranteed. 886 */ 887 __flush_tlb_all(); 888 #else 889 printk(KERN_INFO "Command line: %s\n", boot_command_line); 890 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 891 #endif 892 893 /* 894 * If we have OLPC OFW, we might end up relocating the fixmap due to 895 * reserve_top(), so do this before touching the ioremap area. 896 */ 897 olpc_ofw_detect(); 898 899 idt_setup_early_traps(); 900 early_cpu_init(); 901 jump_label_init(); 902 static_call_init(); 903 early_ioremap_init(); 904 905 setup_olpc_ofw_pgd(); 906 907 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 908 screen_info = boot_params.screen_info; 909 edid_info = boot_params.edid_info; 910 #ifdef CONFIG_X86_32 911 apm_info.bios = boot_params.apm_bios_info; 912 ist_info = boot_params.ist_info; 913 #endif 914 saved_video_mode = boot_params.hdr.vid_mode; 915 bootloader_type = boot_params.hdr.type_of_loader; 916 if ((bootloader_type >> 4) == 0xe) { 917 bootloader_type &= 0xf; 918 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 919 } 920 bootloader_version = bootloader_type & 0xf; 921 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 922 923 #ifdef CONFIG_BLK_DEV_RAM 924 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 925 #endif 926 #ifdef CONFIG_EFI 927 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 928 EFI32_LOADER_SIGNATURE, 4)) { 929 set_bit(EFI_BOOT, &efi.flags); 930 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 931 EFI64_LOADER_SIGNATURE, 4)) { 932 set_bit(EFI_BOOT, &efi.flags); 933 set_bit(EFI_64BIT, &efi.flags); 934 } 935 #endif 936 937 x86_init.oem.arch_setup(); 938 939 /* 940 * Do some memory reservations *before* memory is added to memblock, so 941 * memblock allocations won't overwrite it. 942 * 943 * After this point, everything still needed from the boot loader or 944 * firmware or kernel text should be early reserved or marked not RAM in 945 * e820. All other memory is free game. 946 * 947 * This call needs to happen before e820__memory_setup() which calls the 948 * xen_memory_setup() on Xen dom0 which relies on the fact that those 949 * early reservations have happened already. 950 */ 951 early_reserve_memory(); 952 953 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 954 e820__memory_setup(); 955 parse_setup_data(); 956 957 copy_edd(); 958 959 if (!boot_params.hdr.root_flags) 960 root_mountflags &= ~MS_RDONLY; 961 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 962 963 code_resource.start = __pa_symbol(_text); 964 code_resource.end = __pa_symbol(_etext)-1; 965 rodata_resource.start = __pa_symbol(__start_rodata); 966 rodata_resource.end = __pa_symbol(__end_rodata)-1; 967 data_resource.start = __pa_symbol(_sdata); 968 data_resource.end = __pa_symbol(_edata)-1; 969 bss_resource.start = __pa_symbol(__bss_start); 970 bss_resource.end = __pa_symbol(__bss_stop)-1; 971 972 #ifdef CONFIG_CMDLINE_BOOL 973 #ifdef CONFIG_CMDLINE_OVERRIDE 974 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 975 #else 976 if (builtin_cmdline[0]) { 977 /* append boot loader cmdline to builtin */ 978 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 979 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 980 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 981 } 982 #endif 983 #endif 984 985 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 986 *cmdline_p = command_line; 987 988 /* 989 * x86_configure_nx() is called before parse_early_param() to detect 990 * whether hardware doesn't support NX (so that the early EHCI debug 991 * console setup can safely call set_fixmap()). 992 */ 993 x86_configure_nx(); 994 995 parse_early_param(); 996 997 if (efi_enabled(EFI_BOOT)) 998 efi_memblock_x86_reserve_range(); 999 1000 #ifdef CONFIG_MEMORY_HOTPLUG 1001 /* 1002 * Memory used by the kernel cannot be hot-removed because Linux 1003 * cannot migrate the kernel pages. When memory hotplug is 1004 * enabled, we should prevent memblock from allocating memory 1005 * for the kernel. 1006 * 1007 * ACPI SRAT records all hotpluggable memory ranges. But before 1008 * SRAT is parsed, we don't know about it. 1009 * 1010 * The kernel image is loaded into memory at very early time. We 1011 * cannot prevent this anyway. So on NUMA system, we set any 1012 * node the kernel resides in as un-hotpluggable. 1013 * 1014 * Since on modern servers, one node could have double-digit 1015 * gigabytes memory, we can assume the memory around the kernel 1016 * image is also un-hotpluggable. So before SRAT is parsed, just 1017 * allocate memory near the kernel image to try the best to keep 1018 * the kernel away from hotpluggable memory. 1019 */ 1020 if (movable_node_is_enabled()) 1021 memblock_set_bottom_up(true); 1022 #endif 1023 1024 x86_report_nx(); 1025 1026 if (acpi_mps_check()) { 1027 #ifdef CONFIG_X86_LOCAL_APIC 1028 disable_apic = 1; 1029 #endif 1030 setup_clear_cpu_cap(X86_FEATURE_APIC); 1031 } 1032 1033 e820__reserve_setup_data(); 1034 e820__finish_early_params(); 1035 1036 if (efi_enabled(EFI_BOOT)) 1037 efi_init(); 1038 1039 dmi_setup(); 1040 1041 /* 1042 * VMware detection requires dmi to be available, so this 1043 * needs to be done after dmi_setup(), for the boot CPU. 1044 */ 1045 init_hypervisor_platform(); 1046 1047 tsc_early_init(); 1048 x86_init.resources.probe_roms(); 1049 1050 /* after parse_early_param, so could debug it */ 1051 insert_resource(&iomem_resource, &code_resource); 1052 insert_resource(&iomem_resource, &rodata_resource); 1053 insert_resource(&iomem_resource, &data_resource); 1054 insert_resource(&iomem_resource, &bss_resource); 1055 1056 e820_add_kernel_range(); 1057 trim_bios_range(); 1058 #ifdef CONFIG_X86_32 1059 if (ppro_with_ram_bug()) { 1060 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1061 E820_TYPE_RESERVED); 1062 e820__update_table(e820_table); 1063 printk(KERN_INFO "fixed physical RAM map:\n"); 1064 e820__print_table("bad_ppro"); 1065 } 1066 #else 1067 early_gart_iommu_check(); 1068 #endif 1069 1070 /* 1071 * partially used pages are not usable - thus 1072 * we are rounding upwards: 1073 */ 1074 max_pfn = e820__end_of_ram_pfn(); 1075 1076 /* update e820 for memory not covered by WB MTRRs */ 1077 if (IS_ENABLED(CONFIG_MTRR)) 1078 mtrr_bp_init(); 1079 else 1080 pat_disable("PAT support disabled because CONFIG_MTRR is disabled in the kernel."); 1081 1082 if (mtrr_trim_uncached_memory(max_pfn)) 1083 max_pfn = e820__end_of_ram_pfn(); 1084 1085 max_possible_pfn = max_pfn; 1086 1087 /* 1088 * This call is required when the CPU does not support PAT. If 1089 * mtrr_bp_init() invoked it already via pat_init() the call has no 1090 * effect. 1091 */ 1092 init_cache_modes(); 1093 1094 /* 1095 * Define random base addresses for memory sections after max_pfn is 1096 * defined and before each memory section base is used. 1097 */ 1098 kernel_randomize_memory(); 1099 1100 #ifdef CONFIG_X86_32 1101 /* max_low_pfn get updated here */ 1102 find_low_pfn_range(); 1103 #else 1104 check_x2apic(); 1105 1106 /* How many end-of-memory variables you have, grandma! */ 1107 /* need this before calling reserve_initrd */ 1108 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1109 max_low_pfn = e820__end_of_low_ram_pfn(); 1110 else 1111 max_low_pfn = max_pfn; 1112 1113 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1114 #endif 1115 1116 /* 1117 * Find and reserve possible boot-time SMP configuration: 1118 */ 1119 find_smp_config(); 1120 1121 early_alloc_pgt_buf(); 1122 1123 /* 1124 * Need to conclude brk, before e820__memblock_setup() 1125 * it could use memblock_find_in_range, could overlap with 1126 * brk area. 1127 */ 1128 reserve_brk(); 1129 1130 cleanup_highmap(); 1131 1132 memblock_set_current_limit(ISA_END_ADDRESS); 1133 e820__memblock_setup(); 1134 1135 /* 1136 * Needs to run after memblock setup because it needs the physical 1137 * memory size. 1138 */ 1139 sev_setup_arch(); 1140 1141 efi_fake_memmap(); 1142 efi_find_mirror(); 1143 efi_esrt_init(); 1144 efi_mokvar_table_init(); 1145 1146 /* 1147 * The EFI specification says that boot service code won't be 1148 * called after ExitBootServices(). This is, in fact, a lie. 1149 */ 1150 efi_reserve_boot_services(); 1151 1152 /* preallocate 4k for mptable mpc */ 1153 e820__memblock_alloc_reserved_mpc_new(); 1154 1155 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1156 setup_bios_corruption_check(); 1157 #endif 1158 1159 #ifdef CONFIG_X86_32 1160 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1161 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1162 #endif 1163 1164 /* 1165 * Find free memory for the real mode trampoline and place it there. If 1166 * there is not enough free memory under 1M, on EFI-enabled systems 1167 * there will be additional attempt to reclaim the memory for the real 1168 * mode trampoline at efi_free_boot_services(). 1169 * 1170 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1171 * are known to corrupt low memory and several hundred kilobytes are not 1172 * worth complex detection what memory gets clobbered. Windows does the 1173 * same thing for very similar reasons. 1174 * 1175 * Moreover, on machines with SandyBridge graphics or in setups that use 1176 * crashkernel the entire 1M is reserved anyway. 1177 */ 1178 reserve_real_mode(); 1179 1180 init_mem_mapping(); 1181 1182 idt_setup_early_pf(); 1183 1184 /* 1185 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1186 * with the current CR4 value. This may not be necessary, but 1187 * auditing all the early-boot CR4 manipulation would be needed to 1188 * rule it out. 1189 * 1190 * Mask off features that don't work outside long mode (just 1191 * PCIDE for now). 1192 */ 1193 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1194 1195 memblock_set_current_limit(get_max_mapped()); 1196 1197 /* 1198 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1199 */ 1200 1201 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1202 if (init_ohci1394_dma_early) 1203 init_ohci1394_dma_on_all_controllers(); 1204 #endif 1205 /* Allocate bigger log buffer */ 1206 setup_log_buf(1); 1207 1208 if (efi_enabled(EFI_BOOT)) { 1209 switch (boot_params.secure_boot) { 1210 case efi_secureboot_mode_disabled: 1211 pr_info("Secure boot disabled\n"); 1212 break; 1213 case efi_secureboot_mode_enabled: 1214 pr_info("Secure boot enabled\n"); 1215 break; 1216 default: 1217 pr_info("Secure boot could not be determined\n"); 1218 break; 1219 } 1220 } 1221 1222 reserve_initrd(); 1223 1224 acpi_table_upgrade(); 1225 /* Look for ACPI tables and reserve memory occupied by them. */ 1226 acpi_boot_table_init(); 1227 1228 vsmp_init(); 1229 1230 io_delay_init(); 1231 1232 early_platform_quirks(); 1233 1234 early_acpi_boot_init(); 1235 1236 initmem_init(); 1237 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1238 1239 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1240 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1241 1242 /* 1243 * Reserve memory for crash kernel after SRAT is parsed so that it 1244 * won't consume hotpluggable memory. 1245 */ 1246 reserve_crashkernel(); 1247 1248 memblock_find_dma_reserve(); 1249 1250 if (!early_xdbc_setup_hardware()) 1251 early_xdbc_register_console(); 1252 1253 x86_init.paging.pagetable_init(); 1254 1255 kasan_init(); 1256 1257 /* 1258 * Sync back kernel address range. 1259 * 1260 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1261 * this call? 1262 */ 1263 sync_initial_page_table(); 1264 1265 tboot_probe(); 1266 1267 map_vsyscall(); 1268 1269 generic_apic_probe(); 1270 1271 early_quirks(); 1272 1273 /* 1274 * Read APIC and some other early information from ACPI tables. 1275 */ 1276 acpi_boot_init(); 1277 x86_dtb_init(); 1278 1279 /* 1280 * get boot-time SMP configuration: 1281 */ 1282 get_smp_config(); 1283 1284 /* 1285 * Systems w/o ACPI and mptables might not have it mapped the local 1286 * APIC yet, but prefill_possible_map() might need to access it. 1287 */ 1288 init_apic_mappings(); 1289 1290 prefill_possible_map(); 1291 1292 init_cpu_to_node(); 1293 init_gi_nodes(); 1294 1295 io_apic_init_mappings(); 1296 1297 x86_init.hyper.guest_late_init(); 1298 1299 e820__reserve_resources(); 1300 e820__register_nosave_regions(max_pfn); 1301 1302 x86_init.resources.reserve_resources(); 1303 1304 e820__setup_pci_gap(); 1305 1306 #ifdef CONFIG_VT 1307 #if defined(CONFIG_VGA_CONSOLE) 1308 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1309 conswitchp = &vga_con; 1310 #endif 1311 #endif 1312 x86_init.oem.banner(); 1313 1314 x86_init.timers.wallclock_init(); 1315 1316 /* 1317 * This needs to run before setup_local_APIC() which soft-disables the 1318 * local APIC temporarily and that masks the thermal LVT interrupt, 1319 * leading to softlockups on machines which have configured SMI 1320 * interrupt delivery. 1321 */ 1322 therm_lvt_init(); 1323 1324 mcheck_init(); 1325 1326 register_refined_jiffies(CLOCK_TICK_RATE); 1327 1328 #ifdef CONFIG_EFI 1329 if (efi_enabled(EFI_BOOT)) 1330 efi_apply_memmap_quirks(); 1331 #endif 1332 1333 unwind_init(); 1334 } 1335 1336 #ifdef CONFIG_X86_32 1337 1338 static struct resource video_ram_resource = { 1339 .name = "Video RAM area", 1340 .start = 0xa0000, 1341 .end = 0xbffff, 1342 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1343 }; 1344 1345 void __init i386_reserve_resources(void) 1346 { 1347 request_resource(&iomem_resource, &video_ram_resource); 1348 reserve_standard_io_resources(); 1349 } 1350 1351 #endif /* CONFIG_X86_32 */ 1352 1353 static struct notifier_block kernel_offset_notifier = { 1354 .notifier_call = dump_kernel_offset 1355 }; 1356 1357 static int __init register_kernel_offset_dumper(void) 1358 { 1359 atomic_notifier_chain_register(&panic_notifier_list, 1360 &kernel_offset_notifier); 1361 return 0; 1362 } 1363 __initcall(register_kernel_offset_dumper); 1364