1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1995 Linus Torvalds 4 * 5 * This file contains the setup_arch() code, which handles the architecture-dependent 6 * parts of early kernel initialization. 7 */ 8 #include <linux/console.h> 9 #include <linux/crash_dump.h> 10 #include <linux/dma-map-ops.h> 11 #include <linux/dmi.h> 12 #include <linux/efi.h> 13 #include <linux/init_ohci1394_dma.h> 14 #include <linux/initrd.h> 15 #include <linux/iscsi_ibft.h> 16 #include <linux/memblock.h> 17 #include <linux/pci.h> 18 #include <linux/root_dev.h> 19 #include <linux/sfi.h> 20 #include <linux/hugetlb.h> 21 #include <linux/tboot.h> 22 #include <linux/usb/xhci-dbgp.h> 23 #include <linux/static_call.h> 24 #include <linux/swiotlb.h> 25 26 #include <uapi/linux/mount.h> 27 28 #include <xen/xen.h> 29 30 #include <asm/apic.h> 31 #include <asm/numa.h> 32 #include <asm/bios_ebda.h> 33 #include <asm/bugs.h> 34 #include <asm/cpu.h> 35 #include <asm/efi.h> 36 #include <asm/gart.h> 37 #include <asm/hypervisor.h> 38 #include <asm/io_apic.h> 39 #include <asm/kasan.h> 40 #include <asm/kaslr.h> 41 #include <asm/mce.h> 42 #include <asm/mtrr.h> 43 #include <asm/realmode.h> 44 #include <asm/olpc_ofw.h> 45 #include <asm/pci-direct.h> 46 #include <asm/prom.h> 47 #include <asm/proto.h> 48 #include <asm/unwind.h> 49 #include <asm/vsyscall.h> 50 #include <linux/vmalloc.h> 51 52 /* 53 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 54 * max_pfn_mapped: highest directly mapped pfn > 4 GB 55 * 56 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 57 * represented by pfn_mapped[]. 58 */ 59 unsigned long max_low_pfn_mapped; 60 unsigned long max_pfn_mapped; 61 62 #ifdef CONFIG_DMI 63 RESERVE_BRK(dmi_alloc, 65536); 64 #endif 65 66 67 /* 68 * Range of the BSS area. The size of the BSS area is determined 69 * at link time, with RESERVE_BRK*() facility reserving additional 70 * chunks. 71 */ 72 unsigned long _brk_start = (unsigned long)__brk_base; 73 unsigned long _brk_end = (unsigned long)__brk_base; 74 75 struct boot_params boot_params; 76 77 /* 78 * These are the four main kernel memory regions, we put them into 79 * the resource tree so that kdump tools and other debugging tools 80 * recover it: 81 */ 82 83 static struct resource rodata_resource = { 84 .name = "Kernel rodata", 85 .start = 0, 86 .end = 0, 87 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 88 }; 89 90 static struct resource data_resource = { 91 .name = "Kernel data", 92 .start = 0, 93 .end = 0, 94 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 95 }; 96 97 static struct resource code_resource = { 98 .name = "Kernel code", 99 .start = 0, 100 .end = 0, 101 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 102 }; 103 104 static struct resource bss_resource = { 105 .name = "Kernel bss", 106 .start = 0, 107 .end = 0, 108 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 109 }; 110 111 112 #ifdef CONFIG_X86_32 113 /* CPU data as detected by the assembly code in head_32.S */ 114 struct cpuinfo_x86 new_cpu_data; 115 116 /* Common CPU data for all CPUs */ 117 struct cpuinfo_x86 boot_cpu_data __read_mostly; 118 EXPORT_SYMBOL(boot_cpu_data); 119 120 unsigned int def_to_bigsmp; 121 122 struct apm_info apm_info; 123 EXPORT_SYMBOL(apm_info); 124 125 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 126 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 127 struct ist_info ist_info; 128 EXPORT_SYMBOL(ist_info); 129 #else 130 struct ist_info ist_info; 131 #endif 132 133 #else 134 struct cpuinfo_x86 boot_cpu_data __read_mostly; 135 EXPORT_SYMBOL(boot_cpu_data); 136 #endif 137 138 139 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 140 __visible unsigned long mmu_cr4_features __ro_after_init; 141 #else 142 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 143 #endif 144 145 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 146 int bootloader_type, bootloader_version; 147 148 /* 149 * Setup options 150 */ 151 struct screen_info screen_info; 152 EXPORT_SYMBOL(screen_info); 153 struct edid_info edid_info; 154 EXPORT_SYMBOL_GPL(edid_info); 155 156 extern int root_mountflags; 157 158 unsigned long saved_video_mode; 159 160 #define RAMDISK_IMAGE_START_MASK 0x07FF 161 #define RAMDISK_PROMPT_FLAG 0x8000 162 #define RAMDISK_LOAD_FLAG 0x4000 163 164 static char __initdata command_line[COMMAND_LINE_SIZE]; 165 #ifdef CONFIG_CMDLINE_BOOL 166 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 167 #endif 168 169 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 170 struct edd edd; 171 #ifdef CONFIG_EDD_MODULE 172 EXPORT_SYMBOL(edd); 173 #endif 174 /** 175 * copy_edd() - Copy the BIOS EDD information 176 * from boot_params into a safe place. 177 * 178 */ 179 static inline void __init copy_edd(void) 180 { 181 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 182 sizeof(edd.mbr_signature)); 183 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 184 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 185 edd.edd_info_nr = boot_params.eddbuf_entries; 186 } 187 #else 188 static inline void __init copy_edd(void) 189 { 190 } 191 #endif 192 193 void * __init extend_brk(size_t size, size_t align) 194 { 195 size_t mask = align - 1; 196 void *ret; 197 198 BUG_ON(_brk_start == 0); 199 BUG_ON(align & mask); 200 201 _brk_end = (_brk_end + mask) & ~mask; 202 BUG_ON((char *)(_brk_end + size) > __brk_limit); 203 204 ret = (void *)_brk_end; 205 _brk_end += size; 206 207 memset(ret, 0, size); 208 209 return ret; 210 } 211 212 #ifdef CONFIG_X86_32 213 static void __init cleanup_highmap(void) 214 { 215 } 216 #endif 217 218 static void __init reserve_brk(void) 219 { 220 if (_brk_end > _brk_start) 221 memblock_reserve(__pa_symbol(_brk_start), 222 _brk_end - _brk_start); 223 224 /* Mark brk area as locked down and no longer taking any 225 new allocations */ 226 _brk_start = 0; 227 } 228 229 u64 relocated_ramdisk; 230 231 #ifdef CONFIG_BLK_DEV_INITRD 232 233 static u64 __init get_ramdisk_image(void) 234 { 235 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 236 237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 238 239 if (ramdisk_image == 0) 240 ramdisk_image = phys_initrd_start; 241 242 return ramdisk_image; 243 } 244 static u64 __init get_ramdisk_size(void) 245 { 246 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 247 248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 249 250 if (ramdisk_size == 0) 251 ramdisk_size = phys_initrd_size; 252 253 return ramdisk_size; 254 } 255 256 static void __init relocate_initrd(void) 257 { 258 /* Assume only end is not page aligned */ 259 u64 ramdisk_image = get_ramdisk_image(); 260 u64 ramdisk_size = get_ramdisk_size(); 261 u64 area_size = PAGE_ALIGN(ramdisk_size); 262 263 /* We need to move the initrd down into directly mapped mem */ 264 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 265 PFN_PHYS(max_pfn_mapped)); 266 if (!relocated_ramdisk) 267 panic("Cannot find place for new RAMDISK of size %lld\n", 268 ramdisk_size); 269 270 initrd_start = relocated_ramdisk + PAGE_OFFSET; 271 initrd_end = initrd_start + ramdisk_size; 272 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 273 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 274 275 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 276 277 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 278 " [mem %#010llx-%#010llx]\n", 279 ramdisk_image, ramdisk_image + ramdisk_size - 1, 280 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 281 } 282 283 static void __init early_reserve_initrd(void) 284 { 285 /* Assume only end is not page aligned */ 286 u64 ramdisk_image = get_ramdisk_image(); 287 u64 ramdisk_size = get_ramdisk_size(); 288 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 289 290 if (!boot_params.hdr.type_of_loader || 291 !ramdisk_image || !ramdisk_size) 292 return; /* No initrd provided by bootloader */ 293 294 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 295 } 296 297 static void __init reserve_initrd(void) 298 { 299 /* Assume only end is not page aligned */ 300 u64 ramdisk_image = get_ramdisk_image(); 301 u64 ramdisk_size = get_ramdisk_size(); 302 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 303 304 if (!boot_params.hdr.type_of_loader || 305 !ramdisk_image || !ramdisk_size) 306 return; /* No initrd provided by bootloader */ 307 308 initrd_start = 0; 309 310 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 311 ramdisk_end - 1); 312 313 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 314 PFN_DOWN(ramdisk_end))) { 315 /* All are mapped, easy case */ 316 initrd_start = ramdisk_image + PAGE_OFFSET; 317 initrd_end = initrd_start + ramdisk_size; 318 return; 319 } 320 321 relocate_initrd(); 322 323 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 324 } 325 326 #else 327 static void __init early_reserve_initrd(void) 328 { 329 } 330 static void __init reserve_initrd(void) 331 { 332 } 333 #endif /* CONFIG_BLK_DEV_INITRD */ 334 335 static void __init parse_setup_data(void) 336 { 337 struct setup_data *data; 338 u64 pa_data, pa_next; 339 340 pa_data = boot_params.hdr.setup_data; 341 while (pa_data) { 342 u32 data_len, data_type; 343 344 data = early_memremap(pa_data, sizeof(*data)); 345 data_len = data->len + sizeof(struct setup_data); 346 data_type = data->type; 347 pa_next = data->next; 348 early_memunmap(data, sizeof(*data)); 349 350 switch (data_type) { 351 case SETUP_E820_EXT: 352 e820__memory_setup_extended(pa_data, data_len); 353 break; 354 case SETUP_DTB: 355 add_dtb(pa_data); 356 break; 357 case SETUP_EFI: 358 parse_efi_setup(pa_data, data_len); 359 break; 360 default: 361 break; 362 } 363 pa_data = pa_next; 364 } 365 } 366 367 static void __init memblock_x86_reserve_range_setup_data(void) 368 { 369 struct setup_data *data; 370 u64 pa_data; 371 372 pa_data = boot_params.hdr.setup_data; 373 while (pa_data) { 374 data = early_memremap(pa_data, sizeof(*data)); 375 memblock_reserve(pa_data, sizeof(*data) + data->len); 376 377 if (data->type == SETUP_INDIRECT && 378 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT) 379 memblock_reserve(((struct setup_indirect *)data->data)->addr, 380 ((struct setup_indirect *)data->data)->len); 381 382 pa_data = data->next; 383 early_memunmap(data, sizeof(*data)); 384 } 385 } 386 387 /* 388 * --------- Crashkernel reservation ------------------------------ 389 */ 390 391 #ifdef CONFIG_KEXEC_CORE 392 393 /* 16M alignment for crash kernel regions */ 394 #define CRASH_ALIGN SZ_16M 395 396 /* 397 * Keep the crash kernel below this limit. 398 * 399 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range 400 * due to mapping restrictions. 401 * 402 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is 403 * the upper limit of system RAM in 4-level paging mode. Since the kdump 404 * jump could be from 5-level paging to 4-level paging, the jump will fail if 405 * the kernel is put above 64 TB, and during the 1st kernel bootup there's 406 * no good way to detect the paging mode of the target kernel which will be 407 * loaded for dumping. 408 */ 409 #ifdef CONFIG_X86_32 410 # define CRASH_ADDR_LOW_MAX SZ_512M 411 # define CRASH_ADDR_HIGH_MAX SZ_512M 412 #else 413 # define CRASH_ADDR_LOW_MAX SZ_4G 414 # define CRASH_ADDR_HIGH_MAX SZ_64T 415 #endif 416 417 static int __init reserve_crashkernel_low(void) 418 { 419 #ifdef CONFIG_X86_64 420 unsigned long long base, low_base = 0, low_size = 0; 421 unsigned long low_mem_limit; 422 int ret; 423 424 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX); 425 426 /* crashkernel=Y,low */ 427 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base); 428 if (ret) { 429 /* 430 * two parts from kernel/dma/swiotlb.c: 431 * -swiotlb size: user-specified with swiotlb= or default. 432 * 433 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 434 * to 8M for other buffers that may need to stay low too. Also 435 * make sure we allocate enough extra low memory so that we 436 * don't run out of DMA buffers for 32-bit devices. 437 */ 438 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 439 } else { 440 /* passed with crashkernel=0,low ? */ 441 if (!low_size) 442 return 0; 443 } 444 445 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 446 if (!low_base) { 447 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 448 (unsigned long)(low_size >> 20)); 449 return -ENOMEM; 450 } 451 452 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n", 453 (unsigned long)(low_size >> 20), 454 (unsigned long)(low_base >> 20), 455 (unsigned long)(low_mem_limit >> 20)); 456 457 crashk_low_res.start = low_base; 458 crashk_low_res.end = low_base + low_size - 1; 459 insert_resource(&iomem_resource, &crashk_low_res); 460 #endif 461 return 0; 462 } 463 464 static void __init reserve_crashkernel(void) 465 { 466 unsigned long long crash_size, crash_base, total_mem; 467 bool high = false; 468 int ret; 469 470 total_mem = memblock_phys_mem_size(); 471 472 /* crashkernel=XM */ 473 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 474 if (ret != 0 || crash_size <= 0) { 475 /* crashkernel=X,high */ 476 ret = parse_crashkernel_high(boot_command_line, total_mem, 477 &crash_size, &crash_base); 478 if (ret != 0 || crash_size <= 0) 479 return; 480 high = true; 481 } 482 483 if (xen_pv_domain()) { 484 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 485 return; 486 } 487 488 /* 0 means: find the address automatically */ 489 if (!crash_base) { 490 /* 491 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 492 * crashkernel=x,high reserves memory over 4G, also allocates 493 * 256M extra low memory for DMA buffers and swiotlb. 494 * But the extra memory is not required for all machines. 495 * So try low memory first and fall back to high memory 496 * unless "crashkernel=size[KMG],high" is specified. 497 */ 498 if (!high) 499 crash_base = memblock_phys_alloc_range(crash_size, 500 CRASH_ALIGN, CRASH_ALIGN, 501 CRASH_ADDR_LOW_MAX); 502 if (!crash_base) 503 crash_base = memblock_phys_alloc_range(crash_size, 504 CRASH_ALIGN, CRASH_ALIGN, 505 CRASH_ADDR_HIGH_MAX); 506 if (!crash_base) { 507 pr_info("crashkernel reservation failed - No suitable area found.\n"); 508 return; 509 } 510 } else { 511 unsigned long long start; 512 513 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base, 514 crash_base + crash_size); 515 if (start != crash_base) { 516 pr_info("crashkernel reservation failed - memory is in use.\n"); 517 return; 518 } 519 } 520 521 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 522 memblock_free(crash_base, crash_size); 523 return; 524 } 525 526 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 527 (unsigned long)(crash_size >> 20), 528 (unsigned long)(crash_base >> 20), 529 (unsigned long)(total_mem >> 20)); 530 531 crashk_res.start = crash_base; 532 crashk_res.end = crash_base + crash_size - 1; 533 insert_resource(&iomem_resource, &crashk_res); 534 } 535 #else 536 static void __init reserve_crashkernel(void) 537 { 538 } 539 #endif 540 541 static struct resource standard_io_resources[] = { 542 { .name = "dma1", .start = 0x00, .end = 0x1f, 543 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 544 { .name = "pic1", .start = 0x20, .end = 0x21, 545 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 546 { .name = "timer0", .start = 0x40, .end = 0x43, 547 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 548 { .name = "timer1", .start = 0x50, .end = 0x53, 549 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 550 { .name = "keyboard", .start = 0x60, .end = 0x60, 551 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 552 { .name = "keyboard", .start = 0x64, .end = 0x64, 553 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 554 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 555 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 556 { .name = "pic2", .start = 0xa0, .end = 0xa1, 557 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 558 { .name = "dma2", .start = 0xc0, .end = 0xdf, 559 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 560 { .name = "fpu", .start = 0xf0, .end = 0xff, 561 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 562 }; 563 564 void __init reserve_standard_io_resources(void) 565 { 566 int i; 567 568 /* request I/O space for devices used on all i[345]86 PCs */ 569 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 570 request_resource(&ioport_resource, &standard_io_resources[i]); 571 572 } 573 574 static __init void reserve_ibft_region(void) 575 { 576 unsigned long addr, size = 0; 577 578 addr = find_ibft_region(&size); 579 580 if (size) 581 memblock_reserve(addr, size); 582 } 583 584 static bool __init snb_gfx_workaround_needed(void) 585 { 586 #ifdef CONFIG_PCI 587 int i; 588 u16 vendor, devid; 589 static const __initconst u16 snb_ids[] = { 590 0x0102, 591 0x0112, 592 0x0122, 593 0x0106, 594 0x0116, 595 0x0126, 596 0x010a, 597 }; 598 599 /* Assume no if something weird is going on with PCI */ 600 if (!early_pci_allowed()) 601 return false; 602 603 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 604 if (vendor != 0x8086) 605 return false; 606 607 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 608 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 609 if (devid == snb_ids[i]) 610 return true; 611 #endif 612 613 return false; 614 } 615 616 /* 617 * Sandy Bridge graphics has trouble with certain ranges, exclude 618 * them from allocation. 619 */ 620 static void __init trim_snb_memory(void) 621 { 622 static const __initconst unsigned long bad_pages[] = { 623 0x20050000, 624 0x20110000, 625 0x20130000, 626 0x20138000, 627 0x40004000, 628 }; 629 int i; 630 631 if (!snb_gfx_workaround_needed()) 632 return; 633 634 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 635 636 /* 637 * Reserve all memory below the 1 MB mark that has not 638 * already been reserved. 639 */ 640 memblock_reserve(0, 1<<20); 641 642 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 643 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 644 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 645 bad_pages[i]); 646 } 647 } 648 649 /* 650 * Here we put platform-specific memory range workarounds, i.e. 651 * memory known to be corrupt or otherwise in need to be reserved on 652 * specific platforms. 653 * 654 * If this gets used more widely it could use a real dispatch mechanism. 655 */ 656 static void __init trim_platform_memory_ranges(void) 657 { 658 trim_snb_memory(); 659 } 660 661 static void __init trim_bios_range(void) 662 { 663 /* 664 * special case: Some BIOSes report the PC BIOS 665 * area (640Kb -> 1Mb) as RAM even though it is not. 666 * take them out. 667 */ 668 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 669 670 e820__update_table(e820_table); 671 } 672 673 /* called before trim_bios_range() to spare extra sanitize */ 674 static void __init e820_add_kernel_range(void) 675 { 676 u64 start = __pa_symbol(_text); 677 u64 size = __pa_symbol(_end) - start; 678 679 /* 680 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 681 * attempt to fix it by adding the range. We may have a confused BIOS, 682 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 683 * exclude kernel range. If we really are running on top non-RAM, 684 * we will crash later anyways. 685 */ 686 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 687 return; 688 689 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 690 e820__range_remove(start, size, E820_TYPE_RAM, 0); 691 e820__range_add(start, size, E820_TYPE_RAM); 692 } 693 694 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 695 696 static int __init parse_reservelow(char *p) 697 { 698 unsigned long long size; 699 700 if (!p) 701 return -EINVAL; 702 703 size = memparse(p, &p); 704 705 if (size < 4096) 706 size = 4096; 707 708 if (size > 640*1024) 709 size = 640*1024; 710 711 reserve_low = size; 712 713 return 0; 714 } 715 716 early_param("reservelow", parse_reservelow); 717 718 static void __init trim_low_memory_range(void) 719 { 720 /* 721 * A special case is the first 4Kb of memory; 722 * This is a BIOS owned area, not kernel ram, but generally 723 * not listed as such in the E820 table. 724 * 725 * This typically reserves additional memory (64KiB by default) 726 * since some BIOSes are known to corrupt low memory. See the 727 * Kconfig help text for X86_RESERVE_LOW. 728 */ 729 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 730 } 731 732 /* 733 * Dump out kernel offset information on panic. 734 */ 735 static int 736 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 737 { 738 if (kaslr_enabled()) { 739 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 740 kaslr_offset(), 741 __START_KERNEL, 742 __START_KERNEL_map, 743 MODULES_VADDR-1); 744 } else { 745 pr_emerg("Kernel Offset: disabled\n"); 746 } 747 748 return 0; 749 } 750 751 /* 752 * Determine if we were loaded by an EFI loader. If so, then we have also been 753 * passed the efi memmap, systab, etc., so we should use these data structures 754 * for initialization. Note, the efi init code path is determined by the 755 * global efi_enabled. This allows the same kernel image to be used on existing 756 * systems (with a traditional BIOS) as well as on EFI systems. 757 */ 758 /* 759 * setup_arch - architecture-specific boot-time initializations 760 * 761 * Note: On x86_64, fixmaps are ready for use even before this is called. 762 */ 763 764 void __init setup_arch(char **cmdline_p) 765 { 766 /* 767 * Reserve the memory occupied by the kernel between _text and 768 * __end_of_kernel_reserve symbols. Any kernel sections after the 769 * __end_of_kernel_reserve symbol must be explicitly reserved with a 770 * separate memblock_reserve() or they will be discarded. 771 */ 772 memblock_reserve(__pa_symbol(_text), 773 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 774 775 /* 776 * Make sure page 0 is always reserved because on systems with 777 * L1TF its contents can be leaked to user processes. 778 */ 779 memblock_reserve(0, PAGE_SIZE); 780 781 early_reserve_initrd(); 782 783 /* 784 * At this point everything still needed from the boot loader 785 * or BIOS or kernel text should be early reserved or marked not 786 * RAM in e820. All other memory is free game. 787 */ 788 789 #ifdef CONFIG_X86_32 790 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 791 792 /* 793 * copy kernel address range established so far and switch 794 * to the proper swapper page table 795 */ 796 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 797 initial_page_table + KERNEL_PGD_BOUNDARY, 798 KERNEL_PGD_PTRS); 799 800 load_cr3(swapper_pg_dir); 801 /* 802 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 803 * a cr3 based tlb flush, so the following __flush_tlb_all() 804 * will not flush anything because the CPU quirk which clears 805 * X86_FEATURE_PGE has not been invoked yet. Though due to the 806 * load_cr3() above the TLB has been flushed already. The 807 * quirk is invoked before subsequent calls to __flush_tlb_all() 808 * so proper operation is guaranteed. 809 */ 810 __flush_tlb_all(); 811 #else 812 printk(KERN_INFO "Command line: %s\n", boot_command_line); 813 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 814 #endif 815 816 /* 817 * If we have OLPC OFW, we might end up relocating the fixmap due to 818 * reserve_top(), so do this before touching the ioremap area. 819 */ 820 olpc_ofw_detect(); 821 822 idt_setup_early_traps(); 823 early_cpu_init(); 824 arch_init_ideal_nops(); 825 jump_label_init(); 826 static_call_init(); 827 early_ioremap_init(); 828 829 setup_olpc_ofw_pgd(); 830 831 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 832 screen_info = boot_params.screen_info; 833 edid_info = boot_params.edid_info; 834 #ifdef CONFIG_X86_32 835 apm_info.bios = boot_params.apm_bios_info; 836 ist_info = boot_params.ist_info; 837 #endif 838 saved_video_mode = boot_params.hdr.vid_mode; 839 bootloader_type = boot_params.hdr.type_of_loader; 840 if ((bootloader_type >> 4) == 0xe) { 841 bootloader_type &= 0xf; 842 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 843 } 844 bootloader_version = bootloader_type & 0xf; 845 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 846 847 #ifdef CONFIG_BLK_DEV_RAM 848 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 849 #endif 850 #ifdef CONFIG_EFI 851 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 852 EFI32_LOADER_SIGNATURE, 4)) { 853 set_bit(EFI_BOOT, &efi.flags); 854 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 855 EFI64_LOADER_SIGNATURE, 4)) { 856 set_bit(EFI_BOOT, &efi.flags); 857 set_bit(EFI_64BIT, &efi.flags); 858 } 859 #endif 860 861 x86_init.oem.arch_setup(); 862 863 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 864 e820__memory_setup(); 865 parse_setup_data(); 866 867 copy_edd(); 868 869 if (!boot_params.hdr.root_flags) 870 root_mountflags &= ~MS_RDONLY; 871 init_mm.start_code = (unsigned long) _text; 872 init_mm.end_code = (unsigned long) _etext; 873 init_mm.end_data = (unsigned long) _edata; 874 init_mm.brk = _brk_end; 875 876 code_resource.start = __pa_symbol(_text); 877 code_resource.end = __pa_symbol(_etext)-1; 878 rodata_resource.start = __pa_symbol(__start_rodata); 879 rodata_resource.end = __pa_symbol(__end_rodata)-1; 880 data_resource.start = __pa_symbol(_sdata); 881 data_resource.end = __pa_symbol(_edata)-1; 882 bss_resource.start = __pa_symbol(__bss_start); 883 bss_resource.end = __pa_symbol(__bss_stop)-1; 884 885 #ifdef CONFIG_CMDLINE_BOOL 886 #ifdef CONFIG_CMDLINE_OVERRIDE 887 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 888 #else 889 if (builtin_cmdline[0]) { 890 /* append boot loader cmdline to builtin */ 891 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 892 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 893 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 894 } 895 #endif 896 #endif 897 898 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 899 *cmdline_p = command_line; 900 901 /* 902 * x86_configure_nx() is called before parse_early_param() to detect 903 * whether hardware doesn't support NX (so that the early EHCI debug 904 * console setup can safely call set_fixmap()). It may then be called 905 * again from within noexec_setup() during parsing early parameters 906 * to honor the respective command line option. 907 */ 908 x86_configure_nx(); 909 910 parse_early_param(); 911 912 if (efi_enabled(EFI_BOOT)) 913 efi_memblock_x86_reserve_range(); 914 #ifdef CONFIG_MEMORY_HOTPLUG 915 /* 916 * Memory used by the kernel cannot be hot-removed because Linux 917 * cannot migrate the kernel pages. When memory hotplug is 918 * enabled, we should prevent memblock from allocating memory 919 * for the kernel. 920 * 921 * ACPI SRAT records all hotpluggable memory ranges. But before 922 * SRAT is parsed, we don't know about it. 923 * 924 * The kernel image is loaded into memory at very early time. We 925 * cannot prevent this anyway. So on NUMA system, we set any 926 * node the kernel resides in as un-hotpluggable. 927 * 928 * Since on modern servers, one node could have double-digit 929 * gigabytes memory, we can assume the memory around the kernel 930 * image is also un-hotpluggable. So before SRAT is parsed, just 931 * allocate memory near the kernel image to try the best to keep 932 * the kernel away from hotpluggable memory. 933 */ 934 if (movable_node_is_enabled()) 935 memblock_set_bottom_up(true); 936 #endif 937 938 x86_report_nx(); 939 940 /* after early param, so could get panic from serial */ 941 memblock_x86_reserve_range_setup_data(); 942 943 if (acpi_mps_check()) { 944 #ifdef CONFIG_X86_LOCAL_APIC 945 disable_apic = 1; 946 #endif 947 setup_clear_cpu_cap(X86_FEATURE_APIC); 948 } 949 950 e820__reserve_setup_data(); 951 e820__finish_early_params(); 952 953 if (efi_enabled(EFI_BOOT)) 954 efi_init(); 955 956 dmi_setup(); 957 958 /* 959 * VMware detection requires dmi to be available, so this 960 * needs to be done after dmi_setup(), for the boot CPU. 961 */ 962 init_hypervisor_platform(); 963 964 tsc_early_init(); 965 x86_init.resources.probe_roms(); 966 967 /* after parse_early_param, so could debug it */ 968 insert_resource(&iomem_resource, &code_resource); 969 insert_resource(&iomem_resource, &rodata_resource); 970 insert_resource(&iomem_resource, &data_resource); 971 insert_resource(&iomem_resource, &bss_resource); 972 973 e820_add_kernel_range(); 974 trim_bios_range(); 975 #ifdef CONFIG_X86_32 976 if (ppro_with_ram_bug()) { 977 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 978 E820_TYPE_RESERVED); 979 e820__update_table(e820_table); 980 printk(KERN_INFO "fixed physical RAM map:\n"); 981 e820__print_table("bad_ppro"); 982 } 983 #else 984 early_gart_iommu_check(); 985 #endif 986 987 /* 988 * partially used pages are not usable - thus 989 * we are rounding upwards: 990 */ 991 max_pfn = e820__end_of_ram_pfn(); 992 993 /* update e820 for memory not covered by WB MTRRs */ 994 mtrr_bp_init(); 995 if (mtrr_trim_uncached_memory(max_pfn)) 996 max_pfn = e820__end_of_ram_pfn(); 997 998 max_possible_pfn = max_pfn; 999 1000 /* 1001 * This call is required when the CPU does not support PAT. If 1002 * mtrr_bp_init() invoked it already via pat_init() the call has no 1003 * effect. 1004 */ 1005 init_cache_modes(); 1006 1007 /* 1008 * Define random base addresses for memory sections after max_pfn is 1009 * defined and before each memory section base is used. 1010 */ 1011 kernel_randomize_memory(); 1012 1013 #ifdef CONFIG_X86_32 1014 /* max_low_pfn get updated here */ 1015 find_low_pfn_range(); 1016 #else 1017 check_x2apic(); 1018 1019 /* How many end-of-memory variables you have, grandma! */ 1020 /* need this before calling reserve_initrd */ 1021 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1022 max_low_pfn = e820__end_of_low_ram_pfn(); 1023 else 1024 max_low_pfn = max_pfn; 1025 1026 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1027 #endif 1028 1029 /* 1030 * Find and reserve possible boot-time SMP configuration: 1031 */ 1032 find_smp_config(); 1033 1034 reserve_ibft_region(); 1035 1036 early_alloc_pgt_buf(); 1037 1038 /* 1039 * Need to conclude brk, before e820__memblock_setup() 1040 * it could use memblock_find_in_range, could overlap with 1041 * brk area. 1042 */ 1043 reserve_brk(); 1044 1045 cleanup_highmap(); 1046 1047 memblock_set_current_limit(ISA_END_ADDRESS); 1048 e820__memblock_setup(); 1049 1050 /* 1051 * Needs to run after memblock setup because it needs the physical 1052 * memory size. 1053 */ 1054 sev_setup_arch(); 1055 1056 reserve_bios_regions(); 1057 1058 efi_fake_memmap(); 1059 efi_find_mirror(); 1060 efi_esrt_init(); 1061 efi_mokvar_table_init(); 1062 1063 /* 1064 * The EFI specification says that boot service code won't be 1065 * called after ExitBootServices(). This is, in fact, a lie. 1066 */ 1067 efi_reserve_boot_services(); 1068 1069 /* preallocate 4k for mptable mpc */ 1070 e820__memblock_alloc_reserved_mpc_new(); 1071 1072 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1073 setup_bios_corruption_check(); 1074 #endif 1075 1076 #ifdef CONFIG_X86_32 1077 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1078 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1079 #endif 1080 1081 reserve_real_mode(); 1082 1083 trim_platform_memory_ranges(); 1084 trim_low_memory_range(); 1085 1086 init_mem_mapping(); 1087 1088 idt_setup_early_pf(); 1089 1090 /* 1091 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1092 * with the current CR4 value. This may not be necessary, but 1093 * auditing all the early-boot CR4 manipulation would be needed to 1094 * rule it out. 1095 * 1096 * Mask off features that don't work outside long mode (just 1097 * PCIDE for now). 1098 */ 1099 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1100 1101 memblock_set_current_limit(get_max_mapped()); 1102 1103 /* 1104 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1105 */ 1106 1107 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1108 if (init_ohci1394_dma_early) 1109 init_ohci1394_dma_on_all_controllers(); 1110 #endif 1111 /* Allocate bigger log buffer */ 1112 setup_log_buf(1); 1113 1114 if (efi_enabled(EFI_BOOT)) { 1115 switch (boot_params.secure_boot) { 1116 case efi_secureboot_mode_disabled: 1117 pr_info("Secure boot disabled\n"); 1118 break; 1119 case efi_secureboot_mode_enabled: 1120 pr_info("Secure boot enabled\n"); 1121 break; 1122 default: 1123 pr_info("Secure boot could not be determined\n"); 1124 break; 1125 } 1126 } 1127 1128 reserve_initrd(); 1129 1130 acpi_table_upgrade(); 1131 1132 vsmp_init(); 1133 1134 io_delay_init(); 1135 1136 early_platform_quirks(); 1137 1138 /* 1139 * Parse the ACPI tables for possible boot-time SMP configuration. 1140 */ 1141 acpi_boot_table_init(); 1142 1143 early_acpi_boot_init(); 1144 1145 initmem_init(); 1146 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1147 1148 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1149 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1150 1151 /* 1152 * Reserve memory for crash kernel after SRAT is parsed so that it 1153 * won't consume hotpluggable memory. 1154 */ 1155 reserve_crashkernel(); 1156 1157 memblock_find_dma_reserve(); 1158 1159 if (!early_xdbc_setup_hardware()) 1160 early_xdbc_register_console(); 1161 1162 x86_init.paging.pagetable_init(); 1163 1164 kasan_init(); 1165 1166 /* 1167 * Sync back kernel address range. 1168 * 1169 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1170 * this call? 1171 */ 1172 sync_initial_page_table(); 1173 1174 tboot_probe(); 1175 1176 map_vsyscall(); 1177 1178 generic_apic_probe(); 1179 1180 early_quirks(); 1181 1182 /* 1183 * Read APIC and some other early information from ACPI tables. 1184 */ 1185 acpi_boot_init(); 1186 sfi_init(); 1187 x86_dtb_init(); 1188 1189 /* 1190 * get boot-time SMP configuration: 1191 */ 1192 get_smp_config(); 1193 1194 /* 1195 * Systems w/o ACPI and mptables might not have it mapped the local 1196 * APIC yet, but prefill_possible_map() might need to access it. 1197 */ 1198 init_apic_mappings(); 1199 1200 prefill_possible_map(); 1201 1202 init_cpu_to_node(); 1203 init_gi_nodes(); 1204 1205 io_apic_init_mappings(); 1206 1207 x86_init.hyper.guest_late_init(); 1208 1209 e820__reserve_resources(); 1210 e820__register_nosave_regions(max_pfn); 1211 1212 x86_init.resources.reserve_resources(); 1213 1214 e820__setup_pci_gap(); 1215 1216 #ifdef CONFIG_VT 1217 #if defined(CONFIG_VGA_CONSOLE) 1218 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1219 conswitchp = &vga_con; 1220 #endif 1221 #endif 1222 x86_init.oem.banner(); 1223 1224 x86_init.timers.wallclock_init(); 1225 1226 mcheck_init(); 1227 1228 register_refined_jiffies(CLOCK_TICK_RATE); 1229 1230 #ifdef CONFIG_EFI 1231 if (efi_enabled(EFI_BOOT)) 1232 efi_apply_memmap_quirks(); 1233 #endif 1234 1235 unwind_init(); 1236 } 1237 1238 #ifdef CONFIG_X86_32 1239 1240 static struct resource video_ram_resource = { 1241 .name = "Video RAM area", 1242 .start = 0xa0000, 1243 .end = 0xbffff, 1244 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1245 }; 1246 1247 void __init i386_reserve_resources(void) 1248 { 1249 request_resource(&iomem_resource, &video_ram_resource); 1250 reserve_standard_io_resources(); 1251 } 1252 1253 #endif /* CONFIG_X86_32 */ 1254 1255 static struct notifier_block kernel_offset_notifier = { 1256 .notifier_call = dump_kernel_offset 1257 }; 1258 1259 static int __init register_kernel_offset_dumper(void) 1260 { 1261 atomic_notifier_chain_register(&panic_notifier_list, 1262 &kernel_offset_notifier); 1263 return 0; 1264 } 1265 __initcall(register_kernel_offset_dumper); 1266