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/efi.h> 35 #include <asm/numa.h> 36 #include <asm/bios_ebda.h> 37 #include <asm/bugs.h> 38 #include <asm/cacheinfo.h> 39 #include <asm/cpu.h> 40 #include <asm/efi.h> 41 #include <asm/gart.h> 42 #include <asm/hypervisor.h> 43 #include <asm/io_apic.h> 44 #include <asm/kasan.h> 45 #include <asm/kaslr.h> 46 #include <asm/mce.h> 47 #include <asm/memtype.h> 48 #include <asm/mtrr.h> 49 #include <asm/realmode.h> 50 #include <asm/olpc_ofw.h> 51 #include <asm/pci-direct.h> 52 #include <asm/prom.h> 53 #include <asm/proto.h> 54 #include <asm/thermal.h> 55 #include <asm/unwind.h> 56 #include <asm/vsyscall.h> 57 #include <linux/vmalloc.h> 58 59 /* 60 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 61 * max_pfn_mapped: highest directly mapped pfn > 4 GB 62 * 63 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 64 * represented by pfn_mapped[]. 65 */ 66 unsigned long max_low_pfn_mapped; 67 unsigned long max_pfn_mapped; 68 69 #ifdef CONFIG_DMI 70 RESERVE_BRK(dmi_alloc, 65536); 71 #endif 72 73 74 unsigned long _brk_start = (unsigned long)__brk_base; 75 unsigned long _brk_end = (unsigned long)__brk_base; 76 77 struct boot_params boot_params; 78 79 /* 80 * These are the four main kernel memory regions, we put them into 81 * the resource tree so that kdump tools and other debugging tools 82 * recover it: 83 */ 84 85 static struct resource rodata_resource = { 86 .name = "Kernel rodata", 87 .start = 0, 88 .end = 0, 89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 90 }; 91 92 static struct resource data_resource = { 93 .name = "Kernel data", 94 .start = 0, 95 .end = 0, 96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 97 }; 98 99 static struct resource code_resource = { 100 .name = "Kernel code", 101 .start = 0, 102 .end = 0, 103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 104 }; 105 106 static struct resource bss_resource = { 107 .name = "Kernel bss", 108 .start = 0, 109 .end = 0, 110 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 111 }; 112 113 114 #ifdef CONFIG_X86_32 115 /* CPU data as detected by the assembly code in head_32.S */ 116 struct cpuinfo_x86 new_cpu_data; 117 118 /* Common CPU data for all CPUs */ 119 struct cpuinfo_x86 boot_cpu_data __read_mostly; 120 EXPORT_SYMBOL(boot_cpu_data); 121 122 unsigned int def_to_bigsmp; 123 124 struct apm_info apm_info; 125 EXPORT_SYMBOL(apm_info); 126 127 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 128 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 129 struct ist_info ist_info; 130 EXPORT_SYMBOL(ist_info); 131 #else 132 struct ist_info ist_info; 133 #endif 134 135 #else 136 struct cpuinfo_x86 boot_cpu_data __read_mostly; 137 EXPORT_SYMBOL(boot_cpu_data); 138 #endif 139 140 141 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 142 __visible unsigned long mmu_cr4_features __ro_after_init; 143 #else 144 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 145 #endif 146 147 #ifdef CONFIG_IMA 148 static phys_addr_t ima_kexec_buffer_phys; 149 static size_t ima_kexec_buffer_size; 150 #endif 151 152 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 153 int bootloader_type, bootloader_version; 154 155 /* 156 * Setup options 157 */ 158 struct screen_info screen_info; 159 EXPORT_SYMBOL(screen_info); 160 struct edid_info edid_info; 161 EXPORT_SYMBOL_GPL(edid_info); 162 163 extern int root_mountflags; 164 165 unsigned long saved_video_mode; 166 167 #define RAMDISK_IMAGE_START_MASK 0x07FF 168 #define RAMDISK_PROMPT_FLAG 0x8000 169 #define RAMDISK_LOAD_FLAG 0x4000 170 171 static char __initdata command_line[COMMAND_LINE_SIZE]; 172 #ifdef CONFIG_CMDLINE_BOOL 173 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 174 #endif 175 176 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 177 struct edd edd; 178 #ifdef CONFIG_EDD_MODULE 179 EXPORT_SYMBOL(edd); 180 #endif 181 /** 182 * copy_edd() - Copy the BIOS EDD information 183 * from boot_params into a safe place. 184 * 185 */ 186 static inline void __init copy_edd(void) 187 { 188 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 189 sizeof(edd.mbr_signature)); 190 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 191 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 192 edd.edd_info_nr = boot_params.eddbuf_entries; 193 } 194 #else 195 static inline void __init copy_edd(void) 196 { 197 } 198 #endif 199 200 void * __init extend_brk(size_t size, size_t align) 201 { 202 size_t mask = align - 1; 203 void *ret; 204 205 BUG_ON(_brk_start == 0); 206 BUG_ON(align & mask); 207 208 _brk_end = (_brk_end + mask) & ~mask; 209 BUG_ON((char *)(_brk_end + size) > __brk_limit); 210 211 ret = (void *)_brk_end; 212 _brk_end += size; 213 214 memset(ret, 0, size); 215 216 return ret; 217 } 218 219 #ifdef CONFIG_X86_32 220 static void __init cleanup_highmap(void) 221 { 222 } 223 #endif 224 225 static void __init reserve_brk(void) 226 { 227 if (_brk_end > _brk_start) 228 memblock_reserve(__pa_symbol(_brk_start), 229 _brk_end - _brk_start); 230 231 /* Mark brk area as locked down and no longer taking any 232 new allocations */ 233 _brk_start = 0; 234 } 235 236 u64 relocated_ramdisk; 237 238 #ifdef CONFIG_BLK_DEV_INITRD 239 240 static u64 __init get_ramdisk_image(void) 241 { 242 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 243 244 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 245 246 if (ramdisk_image == 0) 247 ramdisk_image = phys_initrd_start; 248 249 return ramdisk_image; 250 } 251 static u64 __init get_ramdisk_size(void) 252 { 253 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 254 255 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 256 257 if (ramdisk_size == 0) 258 ramdisk_size = phys_initrd_size; 259 260 return ramdisk_size; 261 } 262 263 static void __init relocate_initrd(void) 264 { 265 /* Assume only end is not page aligned */ 266 u64 ramdisk_image = get_ramdisk_image(); 267 u64 ramdisk_size = get_ramdisk_size(); 268 u64 area_size = PAGE_ALIGN(ramdisk_size); 269 270 /* We need to move the initrd down into directly mapped mem */ 271 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 272 PFN_PHYS(max_pfn_mapped)); 273 if (!relocated_ramdisk) 274 panic("Cannot find place for new RAMDISK of size %lld\n", 275 ramdisk_size); 276 277 initrd_start = relocated_ramdisk + PAGE_OFFSET; 278 initrd_end = initrd_start + ramdisk_size; 279 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 280 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 281 282 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 283 284 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 285 " [mem %#010llx-%#010llx]\n", 286 ramdisk_image, ramdisk_image + ramdisk_size - 1, 287 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 288 } 289 290 static void __init early_reserve_initrd(void) 291 { 292 /* Assume only end is not page aligned */ 293 u64 ramdisk_image = get_ramdisk_image(); 294 u64 ramdisk_size = get_ramdisk_size(); 295 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 296 297 if (!boot_params.hdr.type_of_loader || 298 !ramdisk_image || !ramdisk_size) 299 return; /* No initrd provided by bootloader */ 300 301 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 302 } 303 304 static void __init reserve_initrd(void) 305 { 306 /* Assume only end is not page aligned */ 307 u64 ramdisk_image = get_ramdisk_image(); 308 u64 ramdisk_size = get_ramdisk_size(); 309 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 310 311 if (!boot_params.hdr.type_of_loader || 312 !ramdisk_image || !ramdisk_size) 313 return; /* No initrd provided by bootloader */ 314 315 initrd_start = 0; 316 317 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 318 ramdisk_end - 1); 319 320 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 321 PFN_DOWN(ramdisk_end))) { 322 /* All are mapped, easy case */ 323 initrd_start = ramdisk_image + PAGE_OFFSET; 324 initrd_end = initrd_start + ramdisk_size; 325 return; 326 } 327 328 relocate_initrd(); 329 330 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image); 331 } 332 333 #else 334 static void __init early_reserve_initrd(void) 335 { 336 } 337 static void __init reserve_initrd(void) 338 { 339 } 340 #endif /* CONFIG_BLK_DEV_INITRD */ 341 342 static void __init add_early_ima_buffer(u64 phys_addr) 343 { 344 #ifdef CONFIG_IMA 345 struct ima_setup_data *data; 346 347 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data)); 348 if (!data) { 349 pr_warn("setup: failed to memremap ima_setup_data entry\n"); 350 return; 351 } 352 353 if (data->size) { 354 memblock_reserve(data->addr, data->size); 355 ima_kexec_buffer_phys = data->addr; 356 ima_kexec_buffer_size = data->size; 357 } 358 359 early_memunmap(data, sizeof(*data)); 360 #else 361 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n"); 362 #endif 363 } 364 365 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE) 366 int __init ima_free_kexec_buffer(void) 367 { 368 int rc; 369 370 if (!ima_kexec_buffer_size) 371 return -ENOENT; 372 373 rc = memblock_phys_free(ima_kexec_buffer_phys, 374 ima_kexec_buffer_size); 375 if (rc) 376 return rc; 377 378 ima_kexec_buffer_phys = 0; 379 ima_kexec_buffer_size = 0; 380 381 return 0; 382 } 383 384 int __init ima_get_kexec_buffer(void **addr, size_t *size) 385 { 386 if (!ima_kexec_buffer_size) 387 return -ENOENT; 388 389 *addr = __va(ima_kexec_buffer_phys); 390 *size = ima_kexec_buffer_size; 391 392 return 0; 393 } 394 #endif 395 396 static void __init parse_setup_data(void) 397 { 398 struct setup_data *data; 399 u64 pa_data, pa_next; 400 401 pa_data = boot_params.hdr.setup_data; 402 while (pa_data) { 403 u32 data_len, data_type; 404 405 data = early_memremap(pa_data, sizeof(*data)); 406 data_len = data->len + sizeof(struct setup_data); 407 data_type = data->type; 408 pa_next = data->next; 409 early_memunmap(data, sizeof(*data)); 410 411 switch (data_type) { 412 case SETUP_E820_EXT: 413 e820__memory_setup_extended(pa_data, data_len); 414 break; 415 case SETUP_DTB: 416 add_dtb(pa_data); 417 break; 418 case SETUP_EFI: 419 parse_efi_setup(pa_data, data_len); 420 break; 421 case SETUP_IMA: 422 add_early_ima_buffer(pa_data); 423 break; 424 case SETUP_RNG_SEED: 425 data = early_memremap(pa_data, data_len); 426 add_bootloader_randomness(data->data, data->len); 427 /* Zero seed for forward secrecy. */ 428 memzero_explicit(data->data, data->len); 429 /* Zero length in case we find ourselves back here by accident. */ 430 memzero_explicit(&data->len, sizeof(data->len)); 431 early_memunmap(data, data_len); 432 break; 433 default: 434 break; 435 } 436 pa_data = pa_next; 437 } 438 } 439 440 static void __init memblock_x86_reserve_range_setup_data(void) 441 { 442 struct setup_indirect *indirect; 443 struct setup_data *data; 444 u64 pa_data, pa_next; 445 u32 len; 446 447 pa_data = boot_params.hdr.setup_data; 448 while (pa_data) { 449 data = early_memremap(pa_data, sizeof(*data)); 450 if (!data) { 451 pr_warn("setup: failed to memremap setup_data entry\n"); 452 return; 453 } 454 455 len = sizeof(*data); 456 pa_next = data->next; 457 458 memblock_reserve(pa_data, sizeof(*data) + data->len); 459 460 if (data->type == SETUP_INDIRECT) { 461 len += data->len; 462 early_memunmap(data, sizeof(*data)); 463 data = early_memremap(pa_data, len); 464 if (!data) { 465 pr_warn("setup: failed to memremap indirect setup_data\n"); 466 return; 467 } 468 469 indirect = (struct setup_indirect *)data->data; 470 471 if (indirect->type != SETUP_INDIRECT) 472 memblock_reserve(indirect->addr, indirect->len); 473 } 474 475 pa_data = pa_next; 476 early_memunmap(data, len); 477 } 478 } 479 480 /* 481 * --------- Crashkernel reservation ------------------------------ 482 */ 483 484 /* 16M alignment for crash kernel regions */ 485 #define CRASH_ALIGN SZ_16M 486 487 /* 488 * Keep the crash kernel below this limit. 489 * 490 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 491 * due to mapping restrictions. 492 * 493 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 494 * the upper limit of system RAM in 4-level paging mode. Since the kdump 495 * jump could be from 5-level paging to 4-level paging, the jump will fail if 496 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 497 * no good way to detect the paging mode of the target kernel which will be 498 * loaded for dumping. 499 */ 500 #ifdef CONFIG_X86_32 501 # define CRASH_ADDR_LOW_MAX SZ_512M 502 # define CRASH_ADDR_HIGH_MAX SZ_512M 503 #else 504 # define CRASH_ADDR_LOW_MAX SZ_4G 505 # define CRASH_ADDR_HIGH_MAX SZ_64T 506 #endif 507 508 static int __init reserve_crashkernel_low(void) 509 { 510 #ifdef CONFIG_X86_64 511 unsigned long long base, low_base = 0, low_size = 0; 512 unsigned long low_mem_limit; 513 int ret; 514 515 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 516 517 /* crashkernel=Y,low */ 518 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 519 if (ret) { 520 /* 521 * two parts from kernel/dma/swiotlb.c: 522 * -swiotlb size: user-specified with swiotlb= or default. 523 * 524 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 525 * to 8M for other buffers that may need to stay low too. Also 526 * make sure we allocate enough extra low memory so that we 527 * don't run out of DMA buffers for 32-bit devices. 528 */ 529 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 530 } else { 531 /* passed with crashkernel=0,low ? */ 532 if (!low_size) 533 return 0; 534 } 535 536 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 537 if (!low_base) { 538 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 539 (unsigned long)(low_size >> 20)); 540 return -ENOMEM; 541 } 542 543 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 544 (unsigned long)(low_size >> 20), 545 (unsigned long)(low_base >> 20), 546 (unsigned long)(low_mem_limit >> 20)); 547 548 crashk_low_res.start = low_base; 549 crashk_low_res.end = low_base + low_size - 1; 550 insert_resource(&iomem_resource, &crashk_low_res); 551 #endif 552 return 0; 553 } 554 555 static void __init reserve_crashkernel(void) 556 { 557 unsigned long long crash_size, crash_base, total_mem; 558 bool high = false; 559 int ret; 560 561 if (!IS_ENABLED(CONFIG_KEXEC_CORE)) 562 return; 563 564 total_mem = memblock_phys_mem_size(); 565 566 /* crashkernel=XM */ 567 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 568 if (ret != 0 || crash_size <= 0) { 569 /* crashkernel=X,high */ 570 ret = parse_crashkernel_high(boot_command_line, total_mem, 571 &crash_size, &crash_base); 572 if (ret != 0 || crash_size <= 0) 573 return; 574 high = true; 575 } 576 577 if (xen_pv_domain()) { 578 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 579 return; 580 } 581 582 /* 0 means: find the address automatically */ 583 if (!crash_base) { 584 /* 585 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 586 * crashkernel=x,high reserves memory over 4G, also allocates 587 * 256M extra low memory for DMA buffers and swiotlb. 588 * But the extra memory is not required for all machines. 589 * So try low memory first and fall back to high memory 590 * unless "crashkernel=size[KMG],high" is specified. 591 */ 592 if (!high) 593 crash_base = memblock_phys_alloc_range(crash_size, 594 CRASH_ALIGN, CRASH_ALIGN, 595 CRASH_ADDR_LOW_MAX); 596 if (!crash_base) 597 crash_base = memblock_phys_alloc_range(crash_size, 598 CRASH_ALIGN, CRASH_ALIGN, 599 CRASH_ADDR_HIGH_MAX); 600 if (!crash_base) { 601 pr_info("crashkernel reservation failed - No suitable area found.\n"); 602 return; 603 } 604 } else { 605 unsigned long long start; 606 607 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 608 crash_base + crash_size); 609 if (start != crash_base) { 610 pr_info("crashkernel reservation failed - memory is in use.\n"); 611 return; 612 } 613 } 614 615 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 616 memblock_phys_free(crash_base, crash_size); 617 return; 618 } 619 620 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 621 (unsigned long)(crash_size >> 20), 622 (unsigned long)(crash_base >> 20), 623 (unsigned long)(total_mem >> 20)); 624 625 crashk_res.start = crash_base; 626 crashk_res.end = crash_base + crash_size - 1; 627 insert_resource(&iomem_resource, &crashk_res); 628 } 629 630 static struct resource standard_io_resources[] = { 631 { .name = "dma1", .start = 0x00, .end = 0x1f, 632 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 633 { .name = "pic1", .start = 0x20, .end = 0x21, 634 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 635 { .name = "timer0", .start = 0x40, .end = 0x43, 636 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 637 { .name = "timer1", .start = 0x50, .end = 0x53, 638 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 639 { .name = "keyboard", .start = 0x60, .end = 0x60, 640 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 641 { .name = "keyboard", .start = 0x64, .end = 0x64, 642 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 643 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 644 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 645 { .name = "pic2", .start = 0xa0, .end = 0xa1, 646 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 647 { .name = "dma2", .start = 0xc0, .end = 0xdf, 648 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 649 { .name = "fpu", .start = 0xf0, .end = 0xff, 650 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 651 }; 652 653 void __init reserve_standard_io_resources(void) 654 { 655 int i; 656 657 /* request I/O space for devices used on all i[345]86 PCs */ 658 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 659 request_resource(&ioport_resource, &standard_io_resources[i]); 660 661 } 662 663 static bool __init snb_gfx_workaround_needed(void) 664 { 665 #ifdef CONFIG_PCI 666 int i; 667 u16 vendor, devid; 668 static const __initconst u16 snb_ids[] = { 669 0x0102, 670 0x0112, 671 0x0122, 672 0x0106, 673 0x0116, 674 0x0126, 675 0x010a, 676 }; 677 678 /* Assume no if something weird is going on with PCI */ 679 if (!early_pci_allowed()) 680 return false; 681 682 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 683 if (vendor != 0x8086) 684 return false; 685 686 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 687 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 688 if (devid == snb_ids[i]) 689 return true; 690 #endif 691 692 return false; 693 } 694 695 /* 696 * Sandy Bridge graphics has trouble with certain ranges, exclude 697 * them from allocation. 698 */ 699 static void __init trim_snb_memory(void) 700 { 701 static const __initconst unsigned long bad_pages[] = { 702 0x20050000, 703 0x20110000, 704 0x20130000, 705 0x20138000, 706 0x40004000, 707 }; 708 int i; 709 710 if (!snb_gfx_workaround_needed()) 711 return; 712 713 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 714 715 /* 716 * SandyBridge integrated graphics devices have a bug that prevents 717 * them from accessing certain memory ranges, namely anything below 718 * 1M and in the pages listed in bad_pages[] above. 719 * 720 * To avoid these pages being ever accessed by SNB gfx devices reserve 721 * bad_pages that have not already been reserved at boot time. 722 * All memory below the 1 MB mark is anyway reserved later during 723 * setup_arch(), so there is no need to reserve it here. 724 */ 725 726 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 727 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 728 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 729 bad_pages[i]); 730 } 731 } 732 733 static void __init trim_bios_range(void) 734 { 735 /* 736 * A special case is the first 4Kb of memory; 737 * This is a BIOS owned area, not kernel ram, but generally 738 * not listed as such in the E820 table. 739 * 740 * This typically reserves additional memory (64KiB by default) 741 * since some BIOSes are known to corrupt low memory. See the 742 * Kconfig help text for X86_RESERVE_LOW. 743 */ 744 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 745 746 /* 747 * special case: Some BIOSes report the PC BIOS 748 * area (640Kb -> 1Mb) as RAM even though it is not. 749 * take them out. 750 */ 751 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 752 753 e820__update_table(e820_table); 754 } 755 756 /* called before trim_bios_range() to spare extra sanitize */ 757 static void __init e820_add_kernel_range(void) 758 { 759 u64 start = __pa_symbol(_text); 760 u64 size = __pa_symbol(_end) - start; 761 762 /* 763 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 764 * attempt to fix it by adding the range. We may have a confused BIOS, 765 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 766 * exclude kernel range. If we really are running on top non-RAM, 767 * we will crash later anyways. 768 */ 769 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 770 return; 771 772 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 773 e820__range_remove(start, size, E820_TYPE_RAM, 0); 774 e820__range_add(start, size, E820_TYPE_RAM); 775 } 776 777 static void __init early_reserve_memory(void) 778 { 779 /* 780 * Reserve the memory occupied by the kernel between _text and 781 * __end_of_kernel_reserve symbols. Any kernel sections after the 782 * __end_of_kernel_reserve symbol must be explicitly reserved with a 783 * separate memblock_reserve() or they will be discarded. 784 */ 785 memblock_reserve(__pa_symbol(_text), 786 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 787 788 /* 789 * The first 4Kb of memory is a BIOS owned area, but generally it is 790 * not listed as such in the E820 table. 791 * 792 * Reserve the first 64K of memory since some BIOSes are known to 793 * corrupt low memory. After the real mode trampoline is allocated the 794 * rest of the memory below 640k is reserved. 795 * 796 * In addition, make sure page 0 is always reserved because on 797 * systems with L1TF its contents can be leaked to user processes. 798 */ 799 memblock_reserve(0, SZ_64K); 800 801 early_reserve_initrd(); 802 803 memblock_x86_reserve_range_setup_data(); 804 805 reserve_ibft_region(); 806 reserve_bios_regions(); 807 trim_snb_memory(); 808 } 809 810 /* 811 * Dump out kernel offset information on panic. 812 */ 813 static int 814 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 815 { 816 if (kaslr_enabled()) { 817 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 818 kaslr_offset(), 819 __START_KERNEL, 820 __START_KERNEL_map, 821 MODULES_VADDR-1); 822 } else { 823 pr_emerg("Kernel Offset: disabled\n"); 824 } 825 826 return 0; 827 } 828 829 void x86_configure_nx(void) 830 { 831 if (boot_cpu_has(X86_FEATURE_NX)) 832 __supported_pte_mask |= _PAGE_NX; 833 else 834 __supported_pte_mask &= ~_PAGE_NX; 835 } 836 837 static void __init x86_report_nx(void) 838 { 839 if (!boot_cpu_has(X86_FEATURE_NX)) { 840 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 841 "missing in CPU!\n"); 842 } else { 843 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 844 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 845 #else 846 /* 32bit non-PAE kernel, NX cannot be used */ 847 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 848 "cannot be enabled: non-PAE kernel!\n"); 849 #endif 850 } 851 } 852 853 /* 854 * Determine if we were loaded by an EFI loader. If so, then we have also been 855 * passed the efi memmap, systab, etc., so we should use these data structures 856 * for initialization. Note, the efi init code path is determined by the 857 * global efi_enabled. This allows the same kernel image to be used on existing 858 * systems (with a traditional BIOS) as well as on EFI systems. 859 */ 860 /* 861 * setup_arch - architecture-specific boot-time initializations 862 * 863 * Note: On x86_64, fixmaps are ready for use even before this is called. 864 */ 865 866 void __init setup_arch(char **cmdline_p) 867 { 868 #ifdef CONFIG_X86_32 869 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 870 871 /* 872 * copy kernel address range established so far and switch 873 * to the proper swapper page table 874 */ 875 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 876 initial_page_table + KERNEL_PGD_BOUNDARY, 877 KERNEL_PGD_PTRS); 878 879 load_cr3(swapper_pg_dir); 880 /* 881 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 882 * a cr3 based tlb flush, so the following __flush_tlb_all() 883 * will not flush anything because the CPU quirk which clears 884 * X86_FEATURE_PGE has not been invoked yet. Though due to the 885 * load_cr3() above the TLB has been flushed already. The 886 * quirk is invoked before subsequent calls to __flush_tlb_all() 887 * so proper operation is guaranteed. 888 */ 889 __flush_tlb_all(); 890 #else 891 printk(KERN_INFO "Command line: %s\n", boot_command_line); 892 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 893 #endif 894 895 /* 896 * If we have OLPC OFW, we might end up relocating the fixmap due to 897 * reserve_top(), so do this before touching the ioremap area. 898 */ 899 olpc_ofw_detect(); 900 901 idt_setup_early_traps(); 902 early_cpu_init(); 903 jump_label_init(); 904 static_call_init(); 905 early_ioremap_init(); 906 907 setup_olpc_ofw_pgd(); 908 909 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 910 screen_info = boot_params.screen_info; 911 edid_info = boot_params.edid_info; 912 #ifdef CONFIG_X86_32 913 apm_info.bios = boot_params.apm_bios_info; 914 ist_info = boot_params.ist_info; 915 #endif 916 saved_video_mode = boot_params.hdr.vid_mode; 917 bootloader_type = boot_params.hdr.type_of_loader; 918 if ((bootloader_type >> 4) == 0xe) { 919 bootloader_type &= 0xf; 920 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 921 } 922 bootloader_version = bootloader_type & 0xf; 923 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 924 925 #ifdef CONFIG_BLK_DEV_RAM 926 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 927 #endif 928 #ifdef CONFIG_EFI 929 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 930 EFI32_LOADER_SIGNATURE, 4)) { 931 set_bit(EFI_BOOT, &efi.flags); 932 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 933 EFI64_LOADER_SIGNATURE, 4)) { 934 set_bit(EFI_BOOT, &efi.flags); 935 set_bit(EFI_64BIT, &efi.flags); 936 } 937 #endif 938 939 x86_init.oem.arch_setup(); 940 941 /* 942 * Do some memory reservations *before* memory is added to memblock, so 943 * memblock allocations won't overwrite it. 944 * 945 * After this point, everything still needed from the boot loader or 946 * firmware or kernel text should be early reserved or marked not RAM in 947 * e820. All other memory is free game. 948 * 949 * This call needs to happen before e820__memory_setup() which calls the 950 * xen_memory_setup() on Xen dom0 which relies on the fact that those 951 * early reservations have happened already. 952 */ 953 early_reserve_memory(); 954 955 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 956 e820__memory_setup(); 957 parse_setup_data(); 958 959 copy_edd(); 960 961 if (!boot_params.hdr.root_flags) 962 root_mountflags &= ~MS_RDONLY; 963 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 964 965 code_resource.start = __pa_symbol(_text); 966 code_resource.end = __pa_symbol(_etext)-1; 967 rodata_resource.start = __pa_symbol(__start_rodata); 968 rodata_resource.end = __pa_symbol(__end_rodata)-1; 969 data_resource.start = __pa_symbol(_sdata); 970 data_resource.end = __pa_symbol(_edata)-1; 971 bss_resource.start = __pa_symbol(__bss_start); 972 bss_resource.end = __pa_symbol(__bss_stop)-1; 973 974 #ifdef CONFIG_CMDLINE_BOOL 975 #ifdef CONFIG_CMDLINE_OVERRIDE 976 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 977 #else 978 if (builtin_cmdline[0]) { 979 /* append boot loader cmdline to builtin */ 980 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 981 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 982 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 983 } 984 #endif 985 #endif 986 987 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 988 *cmdline_p = command_line; 989 990 /* 991 * x86_configure_nx() is called before parse_early_param() to detect 992 * whether hardware doesn't support NX (so that the early EHCI debug 993 * console setup can safely call set_fixmap()). 994 */ 995 x86_configure_nx(); 996 997 parse_early_param(); 998 999 if (efi_enabled(EFI_BOOT)) 1000 efi_memblock_x86_reserve_range(); 1001 1002 #ifdef CONFIG_MEMORY_HOTPLUG 1003 /* 1004 * Memory used by the kernel cannot be hot-removed because Linux 1005 * cannot migrate the kernel pages. When memory hotplug is 1006 * enabled, we should prevent memblock from allocating memory 1007 * for the kernel. 1008 * 1009 * ACPI SRAT records all hotpluggable memory ranges. But before 1010 * SRAT is parsed, we don't know about it. 1011 * 1012 * The kernel image is loaded into memory at very early time. We 1013 * cannot prevent this anyway. So on NUMA system, we set any 1014 * node the kernel resides in as un-hotpluggable. 1015 * 1016 * Since on modern servers, one node could have double-digit 1017 * gigabytes memory, we can assume the memory around the kernel 1018 * image is also un-hotpluggable. So before SRAT is parsed, just 1019 * allocate memory near the kernel image to try the best to keep 1020 * the kernel away from hotpluggable memory. 1021 */ 1022 if (movable_node_is_enabled()) 1023 memblock_set_bottom_up(true); 1024 #endif 1025 1026 x86_report_nx(); 1027 1028 if (acpi_mps_check()) { 1029 #ifdef CONFIG_X86_LOCAL_APIC 1030 disable_apic = 1; 1031 #endif 1032 setup_clear_cpu_cap(X86_FEATURE_APIC); 1033 } 1034 1035 e820__reserve_setup_data(); 1036 e820__finish_early_params(); 1037 1038 if (efi_enabled(EFI_BOOT)) 1039 efi_init(); 1040 1041 dmi_setup(); 1042 1043 /* 1044 * VMware detection requires dmi to be available, so this 1045 * needs to be done after dmi_setup(), for the boot CPU. 1046 */ 1047 init_hypervisor_platform(); 1048 1049 tsc_early_init(); 1050 x86_init.resources.probe_roms(); 1051 1052 /* after parse_early_param, so could debug it */ 1053 insert_resource(&iomem_resource, &code_resource); 1054 insert_resource(&iomem_resource, &rodata_resource); 1055 insert_resource(&iomem_resource, &data_resource); 1056 insert_resource(&iomem_resource, &bss_resource); 1057 1058 e820_add_kernel_range(); 1059 trim_bios_range(); 1060 #ifdef CONFIG_X86_32 1061 if (ppro_with_ram_bug()) { 1062 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1063 E820_TYPE_RESERVED); 1064 e820__update_table(e820_table); 1065 printk(KERN_INFO "fixed physical RAM map:\n"); 1066 e820__print_table("bad_ppro"); 1067 } 1068 #else 1069 early_gart_iommu_check(); 1070 #endif 1071 1072 /* 1073 * partially used pages are not usable - thus 1074 * we are rounding upwards: 1075 */ 1076 max_pfn = e820__end_of_ram_pfn(); 1077 1078 /* update e820 for memory not covered by WB MTRRs */ 1079 cache_bp_init(); 1080 if (mtrr_trim_uncached_memory(max_pfn)) 1081 max_pfn = e820__end_of_ram_pfn(); 1082 1083 max_possible_pfn = max_pfn; 1084 1085 /* 1086 * Define random base addresses for memory sections after max_pfn is 1087 * defined and before each memory section base is used. 1088 */ 1089 kernel_randomize_memory(); 1090 1091 #ifdef CONFIG_X86_32 1092 /* max_low_pfn get updated here */ 1093 find_low_pfn_range(); 1094 #else 1095 check_x2apic(); 1096 1097 /* How many end-of-memory variables you have, grandma! */ 1098 /* need this before calling reserve_initrd */ 1099 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1100 max_low_pfn = e820__end_of_low_ram_pfn(); 1101 else 1102 max_low_pfn = max_pfn; 1103 1104 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1105 #endif 1106 1107 /* 1108 * Find and reserve possible boot-time SMP configuration: 1109 */ 1110 find_smp_config(); 1111 1112 early_alloc_pgt_buf(); 1113 1114 /* 1115 * Need to conclude brk, before e820__memblock_setup() 1116 * it could use memblock_find_in_range, could overlap with 1117 * brk area. 1118 */ 1119 reserve_brk(); 1120 1121 cleanup_highmap(); 1122 1123 memblock_set_current_limit(ISA_END_ADDRESS); 1124 e820__memblock_setup(); 1125 1126 /* 1127 * Needs to run after memblock setup because it needs the physical 1128 * memory size. 1129 */ 1130 sev_setup_arch(); 1131 1132 efi_fake_memmap(); 1133 efi_find_mirror(); 1134 efi_esrt_init(); 1135 efi_mokvar_table_init(); 1136 1137 /* 1138 * The EFI specification says that boot service code won't be 1139 * called after ExitBootServices(). This is, in fact, a lie. 1140 */ 1141 efi_reserve_boot_services(); 1142 1143 /* preallocate 4k for mptable mpc */ 1144 e820__memblock_alloc_reserved_mpc_new(); 1145 1146 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1147 setup_bios_corruption_check(); 1148 #endif 1149 1150 #ifdef CONFIG_X86_32 1151 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1152 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1153 #endif 1154 1155 /* 1156 * Find free memory for the real mode trampoline and place it there. If 1157 * there is not enough free memory under 1M, on EFI-enabled systems 1158 * there will be additional attempt to reclaim the memory for the real 1159 * mode trampoline at efi_free_boot_services(). 1160 * 1161 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1162 * are known to corrupt low memory and several hundred kilobytes are not 1163 * worth complex detection what memory gets clobbered. Windows does the 1164 * same thing for very similar reasons. 1165 * 1166 * Moreover, on machines with SandyBridge graphics or in setups that use 1167 * crashkernel the entire 1M is reserved anyway. 1168 */ 1169 x86_platform.realmode_reserve(); 1170 1171 init_mem_mapping(); 1172 1173 idt_setup_early_pf(); 1174 1175 /* 1176 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1177 * with the current CR4 value. This may not be necessary, but 1178 * auditing all the early-boot CR4 manipulation would be needed to 1179 * rule it out. 1180 * 1181 * Mask off features that don't work outside long mode (just 1182 * PCIDE for now). 1183 */ 1184 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1185 1186 memblock_set_current_limit(get_max_mapped()); 1187 1188 /* 1189 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1190 */ 1191 1192 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1193 if (init_ohci1394_dma_early) 1194 init_ohci1394_dma_on_all_controllers(); 1195 #endif 1196 /* Allocate bigger log buffer */ 1197 setup_log_buf(1); 1198 1199 if (efi_enabled(EFI_BOOT)) { 1200 switch (boot_params.secure_boot) { 1201 case efi_secureboot_mode_disabled: 1202 pr_info("Secure boot disabled\n"); 1203 break; 1204 case efi_secureboot_mode_enabled: 1205 pr_info("Secure boot enabled\n"); 1206 break; 1207 default: 1208 pr_info("Secure boot could not be determined\n"); 1209 break; 1210 } 1211 } 1212 1213 reserve_initrd(); 1214 1215 acpi_table_upgrade(); 1216 /* Look for ACPI tables and reserve memory occupied by them. */ 1217 acpi_boot_table_init(); 1218 1219 vsmp_init(); 1220 1221 io_delay_init(); 1222 1223 early_platform_quirks(); 1224 1225 early_acpi_boot_init(); 1226 1227 initmem_init(); 1228 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1229 1230 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1231 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1232 1233 /* 1234 * Reserve memory for crash kernel after SRAT is parsed so that it 1235 * won't consume hotpluggable memory. 1236 */ 1237 reserve_crashkernel(); 1238 1239 memblock_find_dma_reserve(); 1240 1241 if (!early_xdbc_setup_hardware()) 1242 early_xdbc_register_console(); 1243 1244 x86_init.paging.pagetable_init(); 1245 1246 kasan_init(); 1247 1248 /* 1249 * Sync back kernel address range. 1250 * 1251 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1252 * this call? 1253 */ 1254 sync_initial_page_table(); 1255 1256 tboot_probe(); 1257 1258 map_vsyscall(); 1259 1260 generic_apic_probe(); 1261 1262 early_quirks(); 1263 1264 /* 1265 * Read APIC and some other early information from ACPI tables. 1266 */ 1267 acpi_boot_init(); 1268 x86_dtb_init(); 1269 1270 /* 1271 * get boot-time SMP configuration: 1272 */ 1273 get_smp_config(); 1274 1275 /* 1276 * Systems w/o ACPI and mptables might not have it mapped the local 1277 * APIC yet, but prefill_possible_map() might need to access it. 1278 */ 1279 init_apic_mappings(); 1280 1281 prefill_possible_map(); 1282 1283 init_cpu_to_node(); 1284 init_gi_nodes(); 1285 1286 io_apic_init_mappings(); 1287 1288 x86_init.hyper.guest_late_init(); 1289 1290 e820__reserve_resources(); 1291 e820__register_nosave_regions(max_pfn); 1292 1293 x86_init.resources.reserve_resources(); 1294 1295 e820__setup_pci_gap(); 1296 1297 #ifdef CONFIG_VT 1298 #if defined(CONFIG_VGA_CONSOLE) 1299 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1300 conswitchp = &vga_con; 1301 #endif 1302 #endif 1303 x86_init.oem.banner(); 1304 1305 x86_init.timers.wallclock_init(); 1306 1307 /* 1308 * This needs to run before setup_local_APIC() which soft-disables the 1309 * local APIC temporarily and that masks the thermal LVT interrupt, 1310 * leading to softlockups on machines which have configured SMI 1311 * interrupt delivery. 1312 */ 1313 therm_lvt_init(); 1314 1315 mcheck_init(); 1316 1317 register_refined_jiffies(CLOCK_TICK_RATE); 1318 1319 #ifdef CONFIG_EFI 1320 if (efi_enabled(EFI_BOOT)) 1321 efi_apply_memmap_quirks(); 1322 #endif 1323 1324 unwind_init(); 1325 } 1326 1327 #ifdef CONFIG_X86_32 1328 1329 static struct resource video_ram_resource = { 1330 .name = "Video RAM area", 1331 .start = 0xa0000, 1332 .end = 0xbffff, 1333 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1334 }; 1335 1336 void __init i386_reserve_resources(void) 1337 { 1338 request_resource(&iomem_resource, &video_ram_resource); 1339 reserve_standard_io_resources(); 1340 } 1341 1342 #endif /* CONFIG_X86_32 */ 1343 1344 static struct notifier_block kernel_offset_notifier = { 1345 .notifier_call = dump_kernel_offset 1346 }; 1347 1348 static int __init register_kernel_offset_dumper(void) 1349 { 1350 atomic_notifier_chain_register(&panic_notifier_list, 1351 &kernel_offset_notifier); 1352 return 0; 1353 } 1354 __initcall(register_kernel_offset_dumper); 1355