1 /* 2 * Copyright (C) 1995 Linus Torvalds 3 * 4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 5 * 6 * Memory region support 7 * David Parsons <orc@pell.chi.il.us>, July-August 1999 8 * 9 * Added E820 sanitization routine (removes overlapping memory regions); 10 * Brian Moyle <bmoyle@mvista.com>, February 2001 11 * 12 * Moved CPU detection code to cpu/${cpu}.c 13 * Patrick Mochel <mochel@osdl.org>, March 2002 14 * 15 * Provisions for empty E820 memory regions (reported by certain BIOSes). 16 * Alex Achenbach <xela@slit.de>, December 2002. 17 * 18 */ 19 20 /* 21 * This file handles the architecture-dependent parts of initialization 22 */ 23 24 #include <linux/sched.h> 25 #include <linux/mm.h> 26 #include <linux/mmzone.h> 27 #include <linux/screen_info.h> 28 #include <linux/ioport.h> 29 #include <linux/acpi.h> 30 #include <linux/sfi.h> 31 #include <linux/apm_bios.h> 32 #include <linux/initrd.h> 33 #include <linux/memblock.h> 34 #include <linux/seq_file.h> 35 #include <linux/console.h> 36 #include <linux/root_dev.h> 37 #include <linux/highmem.h> 38 #include <linux/export.h> 39 #include <linux/efi.h> 40 #include <linux/init.h> 41 #include <linux/edd.h> 42 #include <linux/iscsi_ibft.h> 43 #include <linux/nodemask.h> 44 #include <linux/kexec.h> 45 #include <linux/dmi.h> 46 #include <linux/pfn.h> 47 #include <linux/pci.h> 48 #include <asm/pci-direct.h> 49 #include <linux/init_ohci1394_dma.h> 50 #include <linux/kvm_para.h> 51 #include <linux/dma-contiguous.h> 52 #include <xen/xen.h> 53 #include <uapi/linux/mount.h> 54 55 #include <linux/errno.h> 56 #include <linux/kernel.h> 57 #include <linux/stddef.h> 58 #include <linux/unistd.h> 59 #include <linux/ptrace.h> 60 #include <linux/user.h> 61 #include <linux/delay.h> 62 63 #include <linux/kallsyms.h> 64 #include <linux/cpufreq.h> 65 #include <linux/dma-mapping.h> 66 #include <linux/ctype.h> 67 #include <linux/uaccess.h> 68 69 #include <linux/percpu.h> 70 #include <linux/crash_dump.h> 71 #include <linux/tboot.h> 72 #include <linux/jiffies.h> 73 #include <linux/mem_encrypt.h> 74 #include <linux/sizes.h> 75 76 #include <linux/usb/xhci-dbgp.h> 77 #include <video/edid.h> 78 79 #include <asm/mtrr.h> 80 #include <asm/apic.h> 81 #include <asm/realmode.h> 82 #include <asm/e820/api.h> 83 #include <asm/mpspec.h> 84 #include <asm/setup.h> 85 #include <asm/efi.h> 86 #include <asm/timer.h> 87 #include <asm/i8259.h> 88 #include <asm/sections.h> 89 #include <asm/io_apic.h> 90 #include <asm/ist.h> 91 #include <asm/setup_arch.h> 92 #include <asm/bios_ebda.h> 93 #include <asm/cacheflush.h> 94 #include <asm/processor.h> 95 #include <asm/bugs.h> 96 #include <asm/kasan.h> 97 98 #include <asm/vsyscall.h> 99 #include <asm/cpu.h> 100 #include <asm/desc.h> 101 #include <asm/dma.h> 102 #include <asm/iommu.h> 103 #include <asm/gart.h> 104 #include <asm/mmu_context.h> 105 #include <asm/proto.h> 106 107 #include <asm/paravirt.h> 108 #include <asm/hypervisor.h> 109 #include <asm/olpc_ofw.h> 110 111 #include <asm/percpu.h> 112 #include <asm/topology.h> 113 #include <asm/apicdef.h> 114 #include <asm/amd_nb.h> 115 #include <asm/mce.h> 116 #include <asm/alternative.h> 117 #include <asm/prom.h> 118 #include <asm/microcode.h> 119 #include <asm/kaslr.h> 120 #include <asm/unwind.h> 121 122 /* 123 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 124 * max_pfn_mapped: highest direct mapped pfn over 4GB 125 * 126 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 127 * represented by pfn_mapped 128 */ 129 unsigned long max_low_pfn_mapped; 130 unsigned long max_pfn_mapped; 131 132 #ifdef CONFIG_DMI 133 RESERVE_BRK(dmi_alloc, 65536); 134 #endif 135 136 137 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 138 unsigned long _brk_end = (unsigned long)__brk_base; 139 140 struct boot_params boot_params; 141 142 /* 143 * Machine setup.. 144 */ 145 static struct resource data_resource = { 146 .name = "Kernel data", 147 .start = 0, 148 .end = 0, 149 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 150 }; 151 152 static struct resource code_resource = { 153 .name = "Kernel code", 154 .start = 0, 155 .end = 0, 156 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 157 }; 158 159 static struct resource bss_resource = { 160 .name = "Kernel bss", 161 .start = 0, 162 .end = 0, 163 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 164 }; 165 166 167 #ifdef CONFIG_X86_32 168 /* cpu data as detected by the assembly code in head_32.S */ 169 struct cpuinfo_x86 new_cpu_data; 170 171 /* common cpu data for all cpus */ 172 struct cpuinfo_x86 boot_cpu_data __read_mostly; 173 EXPORT_SYMBOL(boot_cpu_data); 174 175 unsigned int def_to_bigsmp; 176 177 /* for MCA, but anyone else can use it if they want */ 178 unsigned int machine_id; 179 unsigned int machine_submodel_id; 180 unsigned int BIOS_revision; 181 182 struct apm_info apm_info; 183 EXPORT_SYMBOL(apm_info); 184 185 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 186 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 187 struct ist_info ist_info; 188 EXPORT_SYMBOL(ist_info); 189 #else 190 struct ist_info ist_info; 191 #endif 192 193 #else 194 struct cpuinfo_x86 boot_cpu_data __read_mostly; 195 EXPORT_SYMBOL(boot_cpu_data); 196 #endif 197 198 199 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 200 __visible unsigned long mmu_cr4_features __ro_after_init; 201 #else 202 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 203 #endif 204 205 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 206 int bootloader_type, bootloader_version; 207 208 /* 209 * Setup options 210 */ 211 struct screen_info screen_info; 212 EXPORT_SYMBOL(screen_info); 213 struct edid_info edid_info; 214 EXPORT_SYMBOL_GPL(edid_info); 215 216 extern int root_mountflags; 217 218 unsigned long saved_video_mode; 219 220 #define RAMDISK_IMAGE_START_MASK 0x07FF 221 #define RAMDISK_PROMPT_FLAG 0x8000 222 #define RAMDISK_LOAD_FLAG 0x4000 223 224 static char __initdata command_line[COMMAND_LINE_SIZE]; 225 #ifdef CONFIG_CMDLINE_BOOL 226 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 227 #endif 228 229 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 230 struct edd edd; 231 #ifdef CONFIG_EDD_MODULE 232 EXPORT_SYMBOL(edd); 233 #endif 234 /** 235 * copy_edd() - Copy the BIOS EDD information 236 * from boot_params into a safe place. 237 * 238 */ 239 static inline void __init copy_edd(void) 240 { 241 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 242 sizeof(edd.mbr_signature)); 243 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 244 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 245 edd.edd_info_nr = boot_params.eddbuf_entries; 246 } 247 #else 248 static inline void __init copy_edd(void) 249 { 250 } 251 #endif 252 253 void * __init extend_brk(size_t size, size_t align) 254 { 255 size_t mask = align - 1; 256 void *ret; 257 258 BUG_ON(_brk_start == 0); 259 BUG_ON(align & mask); 260 261 _brk_end = (_brk_end + mask) & ~mask; 262 BUG_ON((char *)(_brk_end + size) > __brk_limit); 263 264 ret = (void *)_brk_end; 265 _brk_end += size; 266 267 memset(ret, 0, size); 268 269 return ret; 270 } 271 272 #ifdef CONFIG_X86_32 273 static void __init cleanup_highmap(void) 274 { 275 } 276 #endif 277 278 static void __init reserve_brk(void) 279 { 280 if (_brk_end > _brk_start) 281 memblock_reserve(__pa_symbol(_brk_start), 282 _brk_end - _brk_start); 283 284 /* Mark brk area as locked down and no longer taking any 285 new allocations */ 286 _brk_start = 0; 287 } 288 289 u64 relocated_ramdisk; 290 291 #ifdef CONFIG_BLK_DEV_INITRD 292 293 static u64 __init get_ramdisk_image(void) 294 { 295 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 296 297 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 298 299 return ramdisk_image; 300 } 301 static u64 __init get_ramdisk_size(void) 302 { 303 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 304 305 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 306 307 return ramdisk_size; 308 } 309 310 static void __init relocate_initrd(void) 311 { 312 /* Assume only end is not page aligned */ 313 u64 ramdisk_image = get_ramdisk_image(); 314 u64 ramdisk_size = get_ramdisk_size(); 315 u64 area_size = PAGE_ALIGN(ramdisk_size); 316 317 /* We need to move the initrd down into directly mapped mem */ 318 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 319 area_size, PAGE_SIZE); 320 321 if (!relocated_ramdisk) 322 panic("Cannot find place for new RAMDISK of size %lld\n", 323 ramdisk_size); 324 325 /* Note: this includes all the mem currently occupied by 326 the initrd, we rely on that fact to keep the data intact. */ 327 memblock_reserve(relocated_ramdisk, area_size); 328 initrd_start = relocated_ramdisk + PAGE_OFFSET; 329 initrd_end = initrd_start + ramdisk_size; 330 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 331 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 332 333 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 334 335 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 336 " [mem %#010llx-%#010llx]\n", 337 ramdisk_image, ramdisk_image + ramdisk_size - 1, 338 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 339 } 340 341 static void __init early_reserve_initrd(void) 342 { 343 /* Assume only end is not page aligned */ 344 u64 ramdisk_image = get_ramdisk_image(); 345 u64 ramdisk_size = get_ramdisk_size(); 346 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 347 348 if (!boot_params.hdr.type_of_loader || 349 !ramdisk_image || !ramdisk_size) 350 return; /* No initrd provided by bootloader */ 351 352 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 353 } 354 static void __init reserve_initrd(void) 355 { 356 /* Assume only end is not page aligned */ 357 u64 ramdisk_image = get_ramdisk_image(); 358 u64 ramdisk_size = get_ramdisk_size(); 359 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 360 u64 mapped_size; 361 362 if (!boot_params.hdr.type_of_loader || 363 !ramdisk_image || !ramdisk_size) 364 return; /* No initrd provided by bootloader */ 365 366 initrd_start = 0; 367 368 mapped_size = memblock_mem_size(max_pfn_mapped); 369 if (ramdisk_size >= (mapped_size>>1)) 370 panic("initrd too large to handle, " 371 "disabling initrd (%lld needed, %lld available)\n", 372 ramdisk_size, mapped_size>>1); 373 374 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 375 ramdisk_end - 1); 376 377 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 378 PFN_DOWN(ramdisk_end))) { 379 /* All are mapped, easy case */ 380 initrd_start = ramdisk_image + PAGE_OFFSET; 381 initrd_end = initrd_start + ramdisk_size; 382 return; 383 } 384 385 relocate_initrd(); 386 387 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 388 } 389 390 #else 391 static void __init early_reserve_initrd(void) 392 { 393 } 394 static void __init reserve_initrd(void) 395 { 396 } 397 #endif /* CONFIG_BLK_DEV_INITRD */ 398 399 static void __init parse_setup_data(void) 400 { 401 struct setup_data *data; 402 u64 pa_data, pa_next; 403 404 pa_data = boot_params.hdr.setup_data; 405 while (pa_data) { 406 u32 data_len, data_type; 407 408 data = early_memremap(pa_data, sizeof(*data)); 409 data_len = data->len + sizeof(struct setup_data); 410 data_type = data->type; 411 pa_next = data->next; 412 early_memunmap(data, sizeof(*data)); 413 414 switch (data_type) { 415 case SETUP_E820_EXT: 416 e820__memory_setup_extended(pa_data, data_len); 417 break; 418 case SETUP_DTB: 419 add_dtb(pa_data); 420 break; 421 case SETUP_EFI: 422 parse_efi_setup(pa_data, data_len); 423 break; 424 default: 425 break; 426 } 427 pa_data = pa_next; 428 } 429 } 430 431 static void __init memblock_x86_reserve_range_setup_data(void) 432 { 433 struct setup_data *data; 434 u64 pa_data; 435 436 pa_data = boot_params.hdr.setup_data; 437 while (pa_data) { 438 data = early_memremap(pa_data, sizeof(*data)); 439 memblock_reserve(pa_data, sizeof(*data) + data->len); 440 pa_data = data->next; 441 early_memunmap(data, sizeof(*data)); 442 } 443 } 444 445 /* 446 * --------- Crashkernel reservation ------------------------------ 447 */ 448 449 #ifdef CONFIG_KEXEC_CORE 450 451 /* 16M alignment for crash kernel regions */ 452 #define CRASH_ALIGN SZ_16M 453 454 /* 455 * Keep the crash kernel below this limit. On 32 bits earlier kernels 456 * would limit the kernel to the low 512 MiB due to mapping restrictions. 457 */ 458 #ifdef CONFIG_X86_32 459 # define CRASH_ADDR_LOW_MAX SZ_512M 460 # define CRASH_ADDR_HIGH_MAX SZ_512M 461 #else 462 # define CRASH_ADDR_LOW_MAX SZ_4G 463 # define CRASH_ADDR_HIGH_MAX MAXMEM 464 #endif 465 466 static int __init reserve_crashkernel_low(void) 467 { 468 #ifdef CONFIG_X86_64 469 unsigned long long base, low_base = 0, low_size = 0; 470 unsigned long total_low_mem; 471 int ret; 472 473 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 474 475 /* crashkernel=Y,low */ 476 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 477 if (ret) { 478 /* 479 * two parts from lib/swiotlb.c: 480 * -swiotlb size: user-specified with swiotlb= or default. 481 * 482 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 483 * to 8M for other buffers that may need to stay low too. Also 484 * make sure we allocate enough extra low memory so that we 485 * don't run out of DMA buffers for 32-bit devices. 486 */ 487 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 488 } else { 489 /* passed with crashkernel=0,low ? */ 490 if (!low_size) 491 return 0; 492 } 493 494 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 495 if (!low_base) { 496 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 497 (unsigned long)(low_size >> 20)); 498 return -ENOMEM; 499 } 500 501 ret = memblock_reserve(low_base, low_size); 502 if (ret) { 503 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 504 return ret; 505 } 506 507 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 508 (unsigned long)(low_size >> 20), 509 (unsigned long)(low_base >> 20), 510 (unsigned long)(total_low_mem >> 20)); 511 512 crashk_low_res.start = low_base; 513 crashk_low_res.end = low_base + low_size - 1; 514 insert_resource(&iomem_resource, &crashk_low_res); 515 #endif 516 return 0; 517 } 518 519 static void __init reserve_crashkernel(void) 520 { 521 unsigned long long crash_size, crash_base, total_mem; 522 bool high = false; 523 int ret; 524 525 total_mem = memblock_phys_mem_size(); 526 527 /* crashkernel=XM */ 528 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 529 if (ret != 0 || crash_size <= 0) { 530 /* crashkernel=X,high */ 531 ret = parse_crashkernel_high(boot_command_line, total_mem, 532 &crash_size, &crash_base); 533 if (ret != 0 || crash_size <= 0) 534 return; 535 high = true; 536 } 537 538 if (xen_pv_domain()) { 539 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 540 return; 541 } 542 543 /* 0 means: find the address automatically */ 544 if (!crash_base) { 545 /* 546 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 547 * crashkernel=x,high reserves memory over 4G, also allocates 548 * 256M extra low memory for DMA buffers and swiotlb. 549 * But the extra memory is not required for all machines. 550 * So try low memory first and fall back to high memory 551 * unless "crashkernel=size[KMG],high" is specified. 552 */ 553 if (!high) 554 crash_base = memblock_find_in_range(CRASH_ALIGN, 555 CRASH_ADDR_LOW_MAX, 556 crash_size, CRASH_ALIGN); 557 if (!crash_base) 558 crash_base = memblock_find_in_range(CRASH_ALIGN, 559 CRASH_ADDR_HIGH_MAX, 560 crash_size, CRASH_ALIGN); 561 if (!crash_base) { 562 pr_info("crashkernel reservation failed - No suitable area found.\n"); 563 return; 564 } 565 } else { 566 unsigned long long start; 567 568 start = memblock_find_in_range(crash_base, 569 crash_base + crash_size, 570 crash_size, 1 << 20); 571 if (start != crash_base) { 572 pr_info("crashkernel reservation failed - memory is in use.\n"); 573 return; 574 } 575 } 576 ret = memblock_reserve(crash_base, crash_size); 577 if (ret) { 578 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 579 return; 580 } 581 582 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 583 memblock_free(crash_base, crash_size); 584 return; 585 } 586 587 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 588 (unsigned long)(crash_size >> 20), 589 (unsigned long)(crash_base >> 20), 590 (unsigned long)(total_mem >> 20)); 591 592 crashk_res.start = crash_base; 593 crashk_res.end = crash_base + crash_size - 1; 594 insert_resource(&iomem_resource, &crashk_res); 595 } 596 #else 597 static void __init reserve_crashkernel(void) 598 { 599 } 600 #endif 601 602 static struct resource standard_io_resources[] = { 603 { .name = "dma1", .start = 0x00, .end = 0x1f, 604 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 605 { .name = "pic1", .start = 0x20, .end = 0x21, 606 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 607 { .name = "timer0", .start = 0x40, .end = 0x43, 608 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 609 { .name = "timer1", .start = 0x50, .end = 0x53, 610 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 611 { .name = "keyboard", .start = 0x60, .end = 0x60, 612 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 613 { .name = "keyboard", .start = 0x64, .end = 0x64, 614 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 615 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 616 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 617 { .name = "pic2", .start = 0xa0, .end = 0xa1, 618 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 619 { .name = "dma2", .start = 0xc0, .end = 0xdf, 620 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 621 { .name = "fpu", .start = 0xf0, .end = 0xff, 622 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 623 }; 624 625 void __init reserve_standard_io_resources(void) 626 { 627 int i; 628 629 /* request I/O space for devices used on all i[345]86 PCs */ 630 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 631 request_resource(&ioport_resource, &standard_io_resources[i]); 632 633 } 634 635 static __init void reserve_ibft_region(void) 636 { 637 unsigned long addr, size = 0; 638 639 addr = find_ibft_region(&size); 640 641 if (size) 642 memblock_reserve(addr, size); 643 } 644 645 static bool __init snb_gfx_workaround_needed(void) 646 { 647 #ifdef CONFIG_PCI 648 int i; 649 u16 vendor, devid; 650 static const __initconst u16 snb_ids[] = { 651 0x0102, 652 0x0112, 653 0x0122, 654 0x0106, 655 0x0116, 656 0x0126, 657 0x010a, 658 }; 659 660 /* Assume no if something weird is going on with PCI */ 661 if (!early_pci_allowed()) 662 return false; 663 664 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 665 if (vendor != 0x8086) 666 return false; 667 668 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 669 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 670 if (devid == snb_ids[i]) 671 return true; 672 #endif 673 674 return false; 675 } 676 677 /* 678 * Sandy Bridge graphics has trouble with certain ranges, exclude 679 * them from allocation. 680 */ 681 static void __init trim_snb_memory(void) 682 { 683 static const __initconst unsigned long bad_pages[] = { 684 0x20050000, 685 0x20110000, 686 0x20130000, 687 0x20138000, 688 0x40004000, 689 }; 690 int i; 691 692 if (!snb_gfx_workaround_needed()) 693 return; 694 695 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 696 697 /* 698 * Reserve all memory below the 1 MB mark that has not 699 * already been reserved. 700 */ 701 memblock_reserve(0, 1<<20); 702 703 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 704 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 705 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 706 bad_pages[i]); 707 } 708 } 709 710 /* 711 * Here we put platform-specific memory range workarounds, i.e. 712 * memory known to be corrupt or otherwise in need to be reserved on 713 * specific platforms. 714 * 715 * If this gets used more widely it could use a real dispatch mechanism. 716 */ 717 static void __init trim_platform_memory_ranges(void) 718 { 719 trim_snb_memory(); 720 } 721 722 static void __init trim_bios_range(void) 723 { 724 /* 725 * A special case is the first 4Kb of memory; 726 * This is a BIOS owned area, not kernel ram, but generally 727 * not listed as such in the E820 table. 728 * 729 * This typically reserves additional memory (64KiB by default) 730 * since some BIOSes are known to corrupt low memory. See the 731 * Kconfig help text for X86_RESERVE_LOW. 732 */ 733 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 734 735 /* 736 * special case: Some BIOSen report the PC BIOS 737 * area (640->1Mb) as ram even though it is not. 738 * take them out. 739 */ 740 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 741 742 e820__update_table(e820_table); 743 } 744 745 /* called before trim_bios_range() to spare extra sanitize */ 746 static void __init e820_add_kernel_range(void) 747 { 748 u64 start = __pa_symbol(_text); 749 u64 size = __pa_symbol(_end) - start; 750 751 /* 752 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 753 * attempt to fix it by adding the range. We may have a confused BIOS, 754 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 755 * exclude kernel range. If we really are running on top non-RAM, 756 * we will crash later anyways. 757 */ 758 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 759 return; 760 761 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 762 e820__range_remove(start, size, E820_TYPE_RAM, 0); 763 e820__range_add(start, size, E820_TYPE_RAM); 764 } 765 766 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 767 768 static int __init parse_reservelow(char *p) 769 { 770 unsigned long long size; 771 772 if (!p) 773 return -EINVAL; 774 775 size = memparse(p, &p); 776 777 if (size < 4096) 778 size = 4096; 779 780 if (size > 640*1024) 781 size = 640*1024; 782 783 reserve_low = size; 784 785 return 0; 786 } 787 788 early_param("reservelow", parse_reservelow); 789 790 static void __init trim_low_memory_range(void) 791 { 792 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 793 } 794 795 /* 796 * Dump out kernel offset information on panic. 797 */ 798 static int 799 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 800 { 801 if (kaslr_enabled()) { 802 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 803 kaslr_offset(), 804 __START_KERNEL, 805 __START_KERNEL_map, 806 MODULES_VADDR-1); 807 } else { 808 pr_emerg("Kernel Offset: disabled\n"); 809 } 810 811 return 0; 812 } 813 814 /* 815 * Determine if we were loaded by an EFI loader. If so, then we have also been 816 * passed the efi memmap, systab, etc., so we should use these data structures 817 * for initialization. Note, the efi init code path is determined by the 818 * global efi_enabled. This allows the same kernel image to be used on existing 819 * systems (with a traditional BIOS) as well as on EFI systems. 820 */ 821 /* 822 * setup_arch - architecture-specific boot-time initializations 823 * 824 * Note: On x86_64, fixmaps are ready for use even before this is called. 825 */ 826 827 void __init setup_arch(char **cmdline_p) 828 { 829 memblock_reserve(__pa_symbol(_text), 830 (unsigned long)__bss_stop - (unsigned long)_text); 831 832 /* 833 * Make sure page 0 is always reserved because on systems with 834 * L1TF its contents can be leaked to user processes. 835 */ 836 memblock_reserve(0, PAGE_SIZE); 837 838 early_reserve_initrd(); 839 840 /* 841 * At this point everything still needed from the boot loader 842 * or BIOS or kernel text should be early reserved or marked not 843 * RAM in e820. All other memory is free game. 844 */ 845 846 #ifdef CONFIG_X86_32 847 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 848 849 /* 850 * copy kernel address range established so far and switch 851 * to the proper swapper page table 852 */ 853 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 854 initial_page_table + KERNEL_PGD_BOUNDARY, 855 KERNEL_PGD_PTRS); 856 857 load_cr3(swapper_pg_dir); 858 /* 859 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 860 * a cr3 based tlb flush, so the following __flush_tlb_all() 861 * will not flush anything because the cpu quirk which clears 862 * X86_FEATURE_PGE has not been invoked yet. Though due to the 863 * load_cr3() above the TLB has been flushed already. The 864 * quirk is invoked before subsequent calls to __flush_tlb_all() 865 * so proper operation is guaranteed. 866 */ 867 __flush_tlb_all(); 868 #else 869 printk(KERN_INFO "Command line: %s\n", boot_command_line); 870 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 871 #endif 872 873 /* 874 * If we have OLPC OFW, we might end up relocating the fixmap due to 875 * reserve_top(), so do this before touching the ioremap area. 876 */ 877 olpc_ofw_detect(); 878 879 idt_setup_early_traps(); 880 early_cpu_init(); 881 arch_init_ideal_nops(); 882 jump_label_init(); 883 early_ioremap_init(); 884 885 setup_olpc_ofw_pgd(); 886 887 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 888 screen_info = boot_params.screen_info; 889 edid_info = boot_params.edid_info; 890 #ifdef CONFIG_X86_32 891 apm_info.bios = boot_params.apm_bios_info; 892 ist_info = boot_params.ist_info; 893 #endif 894 saved_video_mode = boot_params.hdr.vid_mode; 895 bootloader_type = boot_params.hdr.type_of_loader; 896 if ((bootloader_type >> 4) == 0xe) { 897 bootloader_type &= 0xf; 898 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 899 } 900 bootloader_version = bootloader_type & 0xf; 901 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 902 903 #ifdef CONFIG_BLK_DEV_RAM 904 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 905 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 906 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 907 #endif 908 #ifdef CONFIG_EFI 909 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 910 EFI32_LOADER_SIGNATURE, 4)) { 911 set_bit(EFI_BOOT, &efi.flags); 912 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 913 EFI64_LOADER_SIGNATURE, 4)) { 914 set_bit(EFI_BOOT, &efi.flags); 915 set_bit(EFI_64BIT, &efi.flags); 916 } 917 #endif 918 919 x86_init.oem.arch_setup(); 920 921 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 922 e820__memory_setup(); 923 parse_setup_data(); 924 925 copy_edd(); 926 927 if (!boot_params.hdr.root_flags) 928 root_mountflags &= ~MS_RDONLY; 929 init_mm.start_code = (unsigned long) _text; 930 init_mm.end_code = (unsigned long) _etext; 931 init_mm.end_data = (unsigned long) _edata; 932 init_mm.brk = _brk_end; 933 934 mpx_mm_init(&init_mm); 935 936 code_resource.start = __pa_symbol(_text); 937 code_resource.end = __pa_symbol(_etext)-1; 938 data_resource.start = __pa_symbol(_etext); 939 data_resource.end = __pa_symbol(_edata)-1; 940 bss_resource.start = __pa_symbol(__bss_start); 941 bss_resource.end = __pa_symbol(__bss_stop)-1; 942 943 #ifdef CONFIG_CMDLINE_BOOL 944 #ifdef CONFIG_CMDLINE_OVERRIDE 945 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 946 #else 947 if (builtin_cmdline[0]) { 948 /* append boot loader cmdline to builtin */ 949 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 950 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 951 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 952 } 953 #endif 954 #endif 955 956 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 957 *cmdline_p = command_line; 958 959 /* 960 * x86_configure_nx() is called before parse_early_param() to detect 961 * whether hardware doesn't support NX (so that the early EHCI debug 962 * console setup can safely call set_fixmap()). It may then be called 963 * again from within noexec_setup() during parsing early parameters 964 * to honor the respective command line option. 965 */ 966 x86_configure_nx(); 967 968 parse_early_param(); 969 970 if (efi_enabled(EFI_BOOT)) 971 efi_memblock_x86_reserve_range(); 972 #ifdef CONFIG_MEMORY_HOTPLUG 973 /* 974 * Memory used by the kernel cannot be hot-removed because Linux 975 * cannot migrate the kernel pages. When memory hotplug is 976 * enabled, we should prevent memblock from allocating memory 977 * for the kernel. 978 * 979 * ACPI SRAT records all hotpluggable memory ranges. But before 980 * SRAT is parsed, we don't know about it. 981 * 982 * The kernel image is loaded into memory at very early time. We 983 * cannot prevent this anyway. So on NUMA system, we set any 984 * node the kernel resides in as un-hotpluggable. 985 * 986 * Since on modern servers, one node could have double-digit 987 * gigabytes memory, we can assume the memory around the kernel 988 * image is also un-hotpluggable. So before SRAT is parsed, just 989 * allocate memory near the kernel image to try the best to keep 990 * the kernel away from hotpluggable memory. 991 */ 992 if (movable_node_is_enabled()) 993 memblock_set_bottom_up(true); 994 #endif 995 996 x86_report_nx(); 997 998 /* after early param, so could get panic from serial */ 999 memblock_x86_reserve_range_setup_data(); 1000 1001 if (acpi_mps_check()) { 1002 #ifdef CONFIG_X86_LOCAL_APIC 1003 disable_apic = 1; 1004 #endif 1005 setup_clear_cpu_cap(X86_FEATURE_APIC); 1006 } 1007 1008 e820__reserve_setup_data(); 1009 e820__finish_early_params(); 1010 1011 if (efi_enabled(EFI_BOOT)) 1012 efi_init(); 1013 1014 dmi_setup(); 1015 1016 /* 1017 * VMware detection requires dmi to be available, so this 1018 * needs to be done after dmi_setup(), for the boot CPU. 1019 */ 1020 init_hypervisor_platform(); 1021 1022 tsc_early_init(); 1023 x86_init.resources.probe_roms(); 1024 1025 /* after parse_early_param, so could debug it */ 1026 insert_resource(&iomem_resource, &code_resource); 1027 insert_resource(&iomem_resource, &data_resource); 1028 insert_resource(&iomem_resource, &bss_resource); 1029 1030 e820_add_kernel_range(); 1031 trim_bios_range(); 1032 #ifdef CONFIG_X86_32 1033 if (ppro_with_ram_bug()) { 1034 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1035 E820_TYPE_RESERVED); 1036 e820__update_table(e820_table); 1037 printk(KERN_INFO "fixed physical RAM map:\n"); 1038 e820__print_table("bad_ppro"); 1039 } 1040 #else 1041 early_gart_iommu_check(); 1042 #endif 1043 1044 /* 1045 * partially used pages are not usable - thus 1046 * we are rounding upwards: 1047 */ 1048 max_pfn = e820__end_of_ram_pfn(); 1049 1050 /* update e820 for memory not covered by WB MTRRs */ 1051 mtrr_bp_init(); 1052 if (mtrr_trim_uncached_memory(max_pfn)) 1053 max_pfn = e820__end_of_ram_pfn(); 1054 1055 max_possible_pfn = max_pfn; 1056 1057 /* 1058 * This call is required when the CPU does not support PAT. If 1059 * mtrr_bp_init() invoked it already via pat_init() the call has no 1060 * effect. 1061 */ 1062 init_cache_modes(); 1063 1064 /* 1065 * Define random base addresses for memory sections after max_pfn is 1066 * defined and before each memory section base is used. 1067 */ 1068 kernel_randomize_memory(); 1069 1070 #ifdef CONFIG_X86_32 1071 /* max_low_pfn get updated here */ 1072 find_low_pfn_range(); 1073 #else 1074 check_x2apic(); 1075 1076 /* How many end-of-memory variables you have, grandma! */ 1077 /* need this before calling reserve_initrd */ 1078 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1079 max_low_pfn = e820__end_of_low_ram_pfn(); 1080 else 1081 max_low_pfn = max_pfn; 1082 1083 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1084 #endif 1085 1086 /* 1087 * Find and reserve possible boot-time SMP configuration: 1088 */ 1089 find_smp_config(); 1090 1091 reserve_ibft_region(); 1092 1093 early_alloc_pgt_buf(); 1094 1095 /* 1096 * Need to conclude brk, before e820__memblock_setup() 1097 * it could use memblock_find_in_range, could overlap with 1098 * brk area. 1099 */ 1100 reserve_brk(); 1101 1102 cleanup_highmap(); 1103 1104 memblock_set_current_limit(ISA_END_ADDRESS); 1105 e820__memblock_setup(); 1106 1107 reserve_bios_regions(); 1108 1109 if (efi_enabled(EFI_MEMMAP)) { 1110 efi_fake_memmap(); 1111 efi_find_mirror(); 1112 efi_esrt_init(); 1113 1114 /* 1115 * The EFI specification says that boot service code won't be 1116 * called after ExitBootServices(). This is, in fact, a lie. 1117 */ 1118 efi_reserve_boot_services(); 1119 } 1120 1121 /* preallocate 4k for mptable mpc */ 1122 e820__memblock_alloc_reserved_mpc_new(); 1123 1124 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1125 setup_bios_corruption_check(); 1126 #endif 1127 1128 #ifdef CONFIG_X86_32 1129 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1130 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1131 #endif 1132 1133 reserve_real_mode(); 1134 1135 trim_platform_memory_ranges(); 1136 trim_low_memory_range(); 1137 1138 init_mem_mapping(); 1139 1140 idt_setup_early_pf(); 1141 1142 /* 1143 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1144 * with the current CR4 value. This may not be necessary, but 1145 * auditing all the early-boot CR4 manipulation would be needed to 1146 * rule it out. 1147 * 1148 * Mask off features that don't work outside long mode (just 1149 * PCIDE for now). 1150 */ 1151 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1152 1153 memblock_set_current_limit(get_max_mapped()); 1154 1155 /* 1156 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1157 */ 1158 1159 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1160 if (init_ohci1394_dma_early) 1161 init_ohci1394_dma_on_all_controllers(); 1162 #endif 1163 /* Allocate bigger log buffer */ 1164 setup_log_buf(1); 1165 1166 if (efi_enabled(EFI_BOOT)) { 1167 switch (boot_params.secure_boot) { 1168 case efi_secureboot_mode_disabled: 1169 pr_info("Secure boot disabled\n"); 1170 break; 1171 case efi_secureboot_mode_enabled: 1172 pr_info("Secure boot enabled\n"); 1173 break; 1174 default: 1175 pr_info("Secure boot could not be determined\n"); 1176 break; 1177 } 1178 } 1179 1180 reserve_initrd(); 1181 1182 acpi_table_upgrade(); 1183 1184 vsmp_init(); 1185 1186 io_delay_init(); 1187 1188 early_platform_quirks(); 1189 1190 /* 1191 * Parse the ACPI tables for possible boot-time SMP configuration. 1192 */ 1193 acpi_boot_table_init(); 1194 1195 early_acpi_boot_init(); 1196 1197 initmem_init(); 1198 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1199 1200 /* 1201 * Reserve memory for crash kernel after SRAT is parsed so that it 1202 * won't consume hotpluggable memory. 1203 */ 1204 reserve_crashkernel(); 1205 1206 memblock_find_dma_reserve(); 1207 1208 if (!early_xdbc_setup_hardware()) 1209 early_xdbc_register_console(); 1210 1211 x86_init.paging.pagetable_init(); 1212 1213 kasan_init(); 1214 1215 /* 1216 * Sync back kernel address range. 1217 * 1218 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1219 * this call? 1220 */ 1221 sync_initial_page_table(); 1222 1223 tboot_probe(); 1224 1225 map_vsyscall(); 1226 1227 generic_apic_probe(); 1228 1229 early_quirks(); 1230 1231 /* 1232 * Read APIC and some other early information from ACPI tables. 1233 */ 1234 acpi_boot_init(); 1235 sfi_init(); 1236 x86_dtb_init(); 1237 1238 /* 1239 * get boot-time SMP configuration: 1240 */ 1241 get_smp_config(); 1242 1243 /* 1244 * Systems w/o ACPI and mptables might not have it mapped the local 1245 * APIC yet, but prefill_possible_map() might need to access it. 1246 */ 1247 init_apic_mappings(); 1248 1249 prefill_possible_map(); 1250 1251 init_cpu_to_node(); 1252 1253 io_apic_init_mappings(); 1254 1255 x86_init.hyper.guest_late_init(); 1256 1257 e820__reserve_resources(); 1258 e820__register_nosave_regions(max_pfn); 1259 1260 x86_init.resources.reserve_resources(); 1261 1262 e820__setup_pci_gap(); 1263 1264 #ifdef CONFIG_VT 1265 #if defined(CONFIG_VGA_CONSOLE) 1266 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1267 conswitchp = &vga_con; 1268 #elif defined(CONFIG_DUMMY_CONSOLE) 1269 conswitchp = &dummy_con; 1270 #endif 1271 #endif 1272 x86_init.oem.banner(); 1273 1274 x86_init.timers.wallclock_init(); 1275 1276 mcheck_init(); 1277 1278 register_refined_jiffies(CLOCK_TICK_RATE); 1279 1280 #ifdef CONFIG_EFI 1281 if (efi_enabled(EFI_BOOT)) 1282 efi_apply_memmap_quirks(); 1283 #endif 1284 1285 unwind_init(); 1286 } 1287 1288 #ifdef CONFIG_X86_32 1289 1290 static struct resource video_ram_resource = { 1291 .name = "Video RAM area", 1292 .start = 0xa0000, 1293 .end = 0xbffff, 1294 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1295 }; 1296 1297 void __init i386_reserve_resources(void) 1298 { 1299 request_resource(&iomem_resource, &video_ram_resource); 1300 reserve_standard_io_resources(); 1301 } 1302 1303 #endif /* CONFIG_X86_32 */ 1304 1305 static struct notifier_block kernel_offset_notifier = { 1306 .notifier_call = dump_kernel_offset 1307 }; 1308 1309 static int __init register_kernel_offset_dumper(void) 1310 { 1311 atomic_notifier_chain_register(&panic_notifier_list, 1312 &kernel_offset_notifier); 1313 return 0; 1314 } 1315 __initcall(register_kernel_offset_dumper); 1316