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/bootmem.h> 34 #include <linux/memblock.h> 35 #include <linux/seq_file.h> 36 #include <linux/console.h> 37 #include <linux/root_dev.h> 38 #include <linux/highmem.h> 39 #include <linux/module.h> 40 #include <linux/efi.h> 41 #include <linux/init.h> 42 #include <linux/edd.h> 43 #include <linux/iscsi_ibft.h> 44 #include <linux/nodemask.h> 45 #include <linux/kexec.h> 46 #include <linux/dmi.h> 47 #include <linux/pfn.h> 48 #include <linux/pci.h> 49 #include <asm/pci-direct.h> 50 #include <linux/init_ohci1394_dma.h> 51 #include <linux/kvm_para.h> 52 #include <linux/dma-contiguous.h> 53 54 #include <linux/errno.h> 55 #include <linux/kernel.h> 56 #include <linux/stddef.h> 57 #include <linux/unistd.h> 58 #include <linux/ptrace.h> 59 #include <linux/user.h> 60 #include <linux/delay.h> 61 62 #include <linux/kallsyms.h> 63 #include <linux/cpufreq.h> 64 #include <linux/dma-mapping.h> 65 #include <linux/ctype.h> 66 #include <linux/uaccess.h> 67 68 #include <linux/percpu.h> 69 #include <linux/crash_dump.h> 70 #include <linux/tboot.h> 71 #include <linux/jiffies.h> 72 73 #include <video/edid.h> 74 75 #include <asm/mtrr.h> 76 #include <asm/apic.h> 77 #include <asm/realmode.h> 78 #include <asm/e820.h> 79 #include <asm/mpspec.h> 80 #include <asm/setup.h> 81 #include <asm/efi.h> 82 #include <asm/timer.h> 83 #include <asm/i8259.h> 84 #include <asm/sections.h> 85 #include <asm/dmi.h> 86 #include <asm/io_apic.h> 87 #include <asm/ist.h> 88 #include <asm/setup_arch.h> 89 #include <asm/bios_ebda.h> 90 #include <asm/cacheflush.h> 91 #include <asm/processor.h> 92 #include <asm/bugs.h> 93 94 #include <asm/vsyscall.h> 95 #include <asm/cpu.h> 96 #include <asm/desc.h> 97 #include <asm/dma.h> 98 #include <asm/iommu.h> 99 #include <asm/gart.h> 100 #include <asm/mmu_context.h> 101 #include <asm/proto.h> 102 103 #include <asm/paravirt.h> 104 #include <asm/hypervisor.h> 105 #include <asm/olpc_ofw.h> 106 107 #include <asm/percpu.h> 108 #include <asm/topology.h> 109 #include <asm/apicdef.h> 110 #include <asm/amd_nb.h> 111 #include <asm/mce.h> 112 #include <asm/alternative.h> 113 #include <asm/prom.h> 114 115 /* 116 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 117 * max_pfn_mapped: highest direct mapped pfn over 4GB 118 * 119 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are 120 * represented by pfn_mapped 121 */ 122 unsigned long max_low_pfn_mapped; 123 unsigned long max_pfn_mapped; 124 125 #ifdef CONFIG_DMI 126 RESERVE_BRK(dmi_alloc, 65536); 127 #endif 128 129 130 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 131 unsigned long _brk_end = (unsigned long)__brk_base; 132 133 #ifdef CONFIG_X86_64 134 int default_cpu_present_to_apicid(int mps_cpu) 135 { 136 return __default_cpu_present_to_apicid(mps_cpu); 137 } 138 139 int default_check_phys_apicid_present(int phys_apicid) 140 { 141 return __default_check_phys_apicid_present(phys_apicid); 142 } 143 #endif 144 145 struct boot_params boot_params; 146 147 /* 148 * Machine setup.. 149 */ 150 static struct resource data_resource = { 151 .name = "Kernel data", 152 .start = 0, 153 .end = 0, 154 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 155 }; 156 157 static struct resource code_resource = { 158 .name = "Kernel code", 159 .start = 0, 160 .end = 0, 161 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 162 }; 163 164 static struct resource bss_resource = { 165 .name = "Kernel bss", 166 .start = 0, 167 .end = 0, 168 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 169 }; 170 171 172 #ifdef CONFIG_X86_32 173 /* cpu data as detected by the assembly code in head.S */ 174 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 175 .wp_works_ok = -1, 176 .fdiv_bug = -1, 177 }; 178 /* common cpu data for all cpus */ 179 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 180 .wp_works_ok = -1, 181 .fdiv_bug = -1, 182 }; 183 EXPORT_SYMBOL(boot_cpu_data); 184 185 unsigned int def_to_bigsmp; 186 187 /* for MCA, but anyone else can use it if they want */ 188 unsigned int machine_id; 189 unsigned int machine_submodel_id; 190 unsigned int BIOS_revision; 191 192 struct apm_info apm_info; 193 EXPORT_SYMBOL(apm_info); 194 195 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 196 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 197 struct ist_info ist_info; 198 EXPORT_SYMBOL(ist_info); 199 #else 200 struct ist_info ist_info; 201 #endif 202 203 #else 204 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 205 .x86_phys_bits = MAX_PHYSMEM_BITS, 206 }; 207 EXPORT_SYMBOL(boot_cpu_data); 208 #endif 209 210 211 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 212 unsigned long mmu_cr4_features; 213 #else 214 unsigned long mmu_cr4_features = X86_CR4_PAE; 215 #endif 216 217 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 218 int bootloader_type, bootloader_version; 219 220 /* 221 * Setup options 222 */ 223 struct screen_info screen_info; 224 EXPORT_SYMBOL(screen_info); 225 struct edid_info edid_info; 226 EXPORT_SYMBOL_GPL(edid_info); 227 228 extern int root_mountflags; 229 230 unsigned long saved_video_mode; 231 232 #define RAMDISK_IMAGE_START_MASK 0x07FF 233 #define RAMDISK_PROMPT_FLAG 0x8000 234 #define RAMDISK_LOAD_FLAG 0x4000 235 236 static char __initdata command_line[COMMAND_LINE_SIZE]; 237 #ifdef CONFIG_CMDLINE_BOOL 238 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 239 #endif 240 241 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 242 struct edd edd; 243 #ifdef CONFIG_EDD_MODULE 244 EXPORT_SYMBOL(edd); 245 #endif 246 /** 247 * copy_edd() - Copy the BIOS EDD information 248 * from boot_params into a safe place. 249 * 250 */ 251 static inline void __init copy_edd(void) 252 { 253 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 254 sizeof(edd.mbr_signature)); 255 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 256 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 257 edd.edd_info_nr = boot_params.eddbuf_entries; 258 } 259 #else 260 static inline void __init copy_edd(void) 261 { 262 } 263 #endif 264 265 void * __init extend_brk(size_t size, size_t align) 266 { 267 size_t mask = align - 1; 268 void *ret; 269 270 BUG_ON(_brk_start == 0); 271 BUG_ON(align & mask); 272 273 _brk_end = (_brk_end + mask) & ~mask; 274 BUG_ON((char *)(_brk_end + size) > __brk_limit); 275 276 ret = (void *)_brk_end; 277 _brk_end += size; 278 279 memset(ret, 0, size); 280 281 return ret; 282 } 283 284 #ifdef CONFIG_X86_32 285 static void __init cleanup_highmap(void) 286 { 287 } 288 #endif 289 290 static void __init reserve_brk(void) 291 { 292 if (_brk_end > _brk_start) 293 memblock_reserve(__pa_symbol(_brk_start), 294 _brk_end - _brk_start); 295 296 /* Mark brk area as locked down and no longer taking any 297 new allocations */ 298 _brk_start = 0; 299 } 300 301 #ifdef CONFIG_BLK_DEV_INITRD 302 303 static u64 __init get_ramdisk_image(void) 304 { 305 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 306 307 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 308 309 return ramdisk_image; 310 } 311 static u64 __init get_ramdisk_size(void) 312 { 313 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 314 315 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 316 317 return ramdisk_size; 318 } 319 320 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT) 321 static void __init relocate_initrd(void) 322 { 323 /* Assume only end is not page aligned */ 324 u64 ramdisk_image = get_ramdisk_image(); 325 u64 ramdisk_size = get_ramdisk_size(); 326 u64 area_size = PAGE_ALIGN(ramdisk_size); 327 u64 ramdisk_here; 328 unsigned long slop, clen, mapaddr; 329 char *p, *q; 330 331 /* We need to move the initrd down into directly mapped mem */ 332 ramdisk_here = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 333 area_size, PAGE_SIZE); 334 335 if (!ramdisk_here) 336 panic("Cannot find place for new RAMDISK of size %lld\n", 337 ramdisk_size); 338 339 /* Note: this includes all the mem currently occupied by 340 the initrd, we rely on that fact to keep the data intact. */ 341 memblock_reserve(ramdisk_here, area_size); 342 initrd_start = ramdisk_here + PAGE_OFFSET; 343 initrd_end = initrd_start + ramdisk_size; 344 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 345 ramdisk_here, ramdisk_here + ramdisk_size - 1); 346 347 q = (char *)initrd_start; 348 349 /* Copy the initrd */ 350 while (ramdisk_size) { 351 slop = ramdisk_image & ~PAGE_MASK; 352 clen = ramdisk_size; 353 if (clen > MAX_MAP_CHUNK-slop) 354 clen = MAX_MAP_CHUNK-slop; 355 mapaddr = ramdisk_image & PAGE_MASK; 356 p = early_memremap(mapaddr, clen+slop); 357 memcpy(q, p+slop, clen); 358 early_iounmap(p, clen+slop); 359 q += clen; 360 ramdisk_image += clen; 361 ramdisk_size -= clen; 362 } 363 364 ramdisk_image = get_ramdisk_image(); 365 ramdisk_size = get_ramdisk_size(); 366 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 367 " [mem %#010llx-%#010llx]\n", 368 ramdisk_image, ramdisk_image + ramdisk_size - 1, 369 ramdisk_here, ramdisk_here + ramdisk_size - 1); 370 } 371 372 static void __init early_reserve_initrd(void) 373 { 374 /* Assume only end is not page aligned */ 375 u64 ramdisk_image = get_ramdisk_image(); 376 u64 ramdisk_size = get_ramdisk_size(); 377 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 378 379 if (!boot_params.hdr.type_of_loader || 380 !ramdisk_image || !ramdisk_size) 381 return; /* No initrd provided by bootloader */ 382 383 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 384 } 385 static void __init reserve_initrd(void) 386 { 387 /* Assume only end is not page aligned */ 388 u64 ramdisk_image = get_ramdisk_image(); 389 u64 ramdisk_size = get_ramdisk_size(); 390 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 391 u64 mapped_size; 392 393 if (!boot_params.hdr.type_of_loader || 394 !ramdisk_image || !ramdisk_size) 395 return; /* No initrd provided by bootloader */ 396 397 initrd_start = 0; 398 399 mapped_size = memblock_mem_size(max_pfn_mapped); 400 if (ramdisk_size >= (mapped_size>>1)) 401 panic("initrd too large to handle, " 402 "disabling initrd (%lld needed, %lld available)\n", 403 ramdisk_size, mapped_size>>1); 404 405 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 406 ramdisk_end - 1); 407 408 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 409 PFN_DOWN(ramdisk_end))) { 410 /* All are mapped, easy case */ 411 initrd_start = ramdisk_image + PAGE_OFFSET; 412 initrd_end = initrd_start + ramdisk_size; 413 return; 414 } 415 416 relocate_initrd(); 417 418 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 419 } 420 #else 421 static void __init early_reserve_initrd(void) 422 { 423 } 424 static void __init reserve_initrd(void) 425 { 426 } 427 #endif /* CONFIG_BLK_DEV_INITRD */ 428 429 static void __init parse_setup_data(void) 430 { 431 struct setup_data *data; 432 u64 pa_data; 433 434 pa_data = boot_params.hdr.setup_data; 435 while (pa_data) { 436 u32 data_len, map_len; 437 438 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK), 439 (u64)sizeof(struct setup_data)); 440 data = early_memremap(pa_data, map_len); 441 data_len = data->len + sizeof(struct setup_data); 442 if (data_len > map_len) { 443 early_iounmap(data, map_len); 444 data = early_memremap(pa_data, data_len); 445 map_len = data_len; 446 } 447 448 switch (data->type) { 449 case SETUP_E820_EXT: 450 parse_e820_ext(data); 451 break; 452 case SETUP_DTB: 453 add_dtb(pa_data); 454 break; 455 default: 456 break; 457 } 458 pa_data = data->next; 459 early_iounmap(data, map_len); 460 } 461 } 462 463 static void __init e820_reserve_setup_data(void) 464 { 465 struct setup_data *data; 466 u64 pa_data; 467 int found = 0; 468 469 pa_data = boot_params.hdr.setup_data; 470 while (pa_data) { 471 data = early_memremap(pa_data, sizeof(*data)); 472 e820_update_range(pa_data, sizeof(*data)+data->len, 473 E820_RAM, E820_RESERVED_KERN); 474 found = 1; 475 pa_data = data->next; 476 early_iounmap(data, sizeof(*data)); 477 } 478 if (!found) 479 return; 480 481 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 482 memcpy(&e820_saved, &e820, sizeof(struct e820map)); 483 printk(KERN_INFO "extended physical RAM map:\n"); 484 e820_print_map("reserve setup_data"); 485 } 486 487 static void __init memblock_x86_reserve_range_setup_data(void) 488 { 489 struct setup_data *data; 490 u64 pa_data; 491 492 pa_data = boot_params.hdr.setup_data; 493 while (pa_data) { 494 data = early_memremap(pa_data, sizeof(*data)); 495 memblock_reserve(pa_data, sizeof(*data) + data->len); 496 pa_data = data->next; 497 early_iounmap(data, sizeof(*data)); 498 } 499 } 500 501 /* 502 * --------- Crashkernel reservation ------------------------------ 503 */ 504 505 #ifdef CONFIG_KEXEC 506 507 /* 508 * Keep the crash kernel below this limit. On 32 bits earlier kernels 509 * would limit the kernel to the low 512 MiB due to mapping restrictions. 510 */ 511 #ifdef CONFIG_X86_32 512 # define CRASH_KERNEL_ADDR_MAX (512 << 20) 513 #else 514 # define CRASH_KERNEL_ADDR_MAX MAXMEM 515 #endif 516 517 static void __init reserve_crashkernel_low(void) 518 { 519 #ifdef CONFIG_X86_64 520 const unsigned long long alignment = 16<<20; /* 16M */ 521 unsigned long long low_base = 0, low_size = 0; 522 unsigned long total_low_mem; 523 unsigned long long base; 524 int ret; 525 526 total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT)); 527 ret = parse_crashkernel_low(boot_command_line, total_low_mem, 528 &low_size, &base); 529 if (ret != 0 || low_size <= 0) 530 return; 531 532 low_base = memblock_find_in_range(low_size, (1ULL<<32), 533 low_size, alignment); 534 535 if (!low_base) { 536 pr_info("crashkernel low reservation failed - No suitable area found.\n"); 537 538 return; 539 } 540 541 memblock_reserve(low_base, low_size); 542 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 543 (unsigned long)(low_size >> 20), 544 (unsigned long)(low_base >> 20), 545 (unsigned long)(total_low_mem >> 20)); 546 crashk_low_res.start = low_base; 547 crashk_low_res.end = low_base + low_size - 1; 548 insert_resource(&iomem_resource, &crashk_low_res); 549 #endif 550 } 551 552 static void __init reserve_crashkernel(void) 553 { 554 const unsigned long long alignment = 16<<20; /* 16M */ 555 unsigned long long total_mem; 556 unsigned long long crash_size, crash_base; 557 int ret; 558 559 total_mem = memblock_phys_mem_size(); 560 561 ret = parse_crashkernel(boot_command_line, total_mem, 562 &crash_size, &crash_base); 563 if (ret != 0 || crash_size <= 0) 564 return; 565 566 /* 0 means: find the address automatically */ 567 if (crash_base <= 0) { 568 /* 569 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 570 */ 571 crash_base = memblock_find_in_range(alignment, 572 CRASH_KERNEL_ADDR_MAX, crash_size, alignment); 573 574 if (!crash_base) { 575 pr_info("crashkernel reservation failed - No suitable area found.\n"); 576 return; 577 } 578 579 } else { 580 unsigned long long start; 581 582 start = memblock_find_in_range(crash_base, 583 crash_base + crash_size, crash_size, 1<<20); 584 if (start != crash_base) { 585 pr_info("crashkernel reservation failed - memory is in use.\n"); 586 return; 587 } 588 } 589 memblock_reserve(crash_base, crash_size); 590 591 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " 592 "for crashkernel (System RAM: %ldMB)\n", 593 (unsigned long)(crash_size >> 20), 594 (unsigned long)(crash_base >> 20), 595 (unsigned long)(total_mem >> 20)); 596 597 crashk_res.start = crash_base; 598 crashk_res.end = crash_base + crash_size - 1; 599 insert_resource(&iomem_resource, &crashk_res); 600 601 if (crash_base >= (1ULL<<32)) 602 reserve_crashkernel_low(); 603 } 604 #else 605 static void __init reserve_crashkernel(void) 606 { 607 } 608 #endif 609 610 static struct resource standard_io_resources[] = { 611 { .name = "dma1", .start = 0x00, .end = 0x1f, 612 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 613 { .name = "pic1", .start = 0x20, .end = 0x21, 614 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 615 { .name = "timer0", .start = 0x40, .end = 0x43, 616 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 617 { .name = "timer1", .start = 0x50, .end = 0x53, 618 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 619 { .name = "keyboard", .start = 0x60, .end = 0x60, 620 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 621 { .name = "keyboard", .start = 0x64, .end = 0x64, 622 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 623 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 624 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 625 { .name = "pic2", .start = 0xa0, .end = 0xa1, 626 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 627 { .name = "dma2", .start = 0xc0, .end = 0xdf, 628 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 629 { .name = "fpu", .start = 0xf0, .end = 0xff, 630 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 631 }; 632 633 void __init reserve_standard_io_resources(void) 634 { 635 int i; 636 637 /* request I/O space for devices used on all i[345]86 PCs */ 638 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 639 request_resource(&ioport_resource, &standard_io_resources[i]); 640 641 } 642 643 static __init void reserve_ibft_region(void) 644 { 645 unsigned long addr, size = 0; 646 647 addr = find_ibft_region(&size); 648 649 if (size) 650 memblock_reserve(addr, size); 651 } 652 653 static bool __init snb_gfx_workaround_needed(void) 654 { 655 #ifdef CONFIG_PCI 656 int i; 657 u16 vendor, devid; 658 static const __initconst u16 snb_ids[] = { 659 0x0102, 660 0x0112, 661 0x0122, 662 0x0106, 663 0x0116, 664 0x0126, 665 0x010a, 666 }; 667 668 /* Assume no if something weird is going on with PCI */ 669 if (!early_pci_allowed()) 670 return false; 671 672 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 673 if (vendor != 0x8086) 674 return false; 675 676 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 677 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 678 if (devid == snb_ids[i]) 679 return true; 680 #endif 681 682 return false; 683 } 684 685 /* 686 * Sandy Bridge graphics has trouble with certain ranges, exclude 687 * them from allocation. 688 */ 689 static void __init trim_snb_memory(void) 690 { 691 static const __initconst unsigned long bad_pages[] = { 692 0x20050000, 693 0x20110000, 694 0x20130000, 695 0x20138000, 696 0x40004000, 697 }; 698 int i; 699 700 if (!snb_gfx_workaround_needed()) 701 return; 702 703 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 704 705 /* 706 * Reserve all memory below the 1 MB mark that has not 707 * already been reserved. 708 */ 709 memblock_reserve(0, 1<<20); 710 711 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 712 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 713 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 714 bad_pages[i]); 715 } 716 } 717 718 /* 719 * Here we put platform-specific memory range workarounds, i.e. 720 * memory known to be corrupt or otherwise in need to be reserved on 721 * specific platforms. 722 * 723 * If this gets used more widely it could use a real dispatch mechanism. 724 */ 725 static void __init trim_platform_memory_ranges(void) 726 { 727 trim_snb_memory(); 728 } 729 730 static void __init trim_bios_range(void) 731 { 732 /* 733 * A special case is the first 4Kb of memory; 734 * This is a BIOS owned area, not kernel ram, but generally 735 * not listed as such in the E820 table. 736 * 737 * This typically reserves additional memory (64KiB by default) 738 * since some BIOSes are known to corrupt low memory. See the 739 * Kconfig help text for X86_RESERVE_LOW. 740 */ 741 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 742 743 /* 744 * special case: Some BIOSen report the PC BIOS 745 * area (640->1Mb) as ram even though it is not. 746 * take them out. 747 */ 748 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 749 750 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 751 } 752 753 /* called before trim_bios_range() to spare extra sanitize */ 754 static void __init e820_add_kernel_range(void) 755 { 756 u64 start = __pa_symbol(_text); 757 u64 size = __pa_symbol(_end) - start; 758 759 /* 760 * Complain if .text .data and .bss are not marked as E820_RAM and 761 * attempt to fix it by adding the range. We may have a confused BIOS, 762 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 763 * exclude kernel range. If we really are running on top non-RAM, 764 * we will crash later anyways. 765 */ 766 if (e820_all_mapped(start, start + size, E820_RAM)) 767 return; 768 769 pr_warn(".text .data .bss are not marked as E820_RAM!\n"); 770 e820_remove_range(start, size, E820_RAM, 0); 771 e820_add_region(start, size, E820_RAM); 772 } 773 774 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 775 776 static int __init parse_reservelow(char *p) 777 { 778 unsigned long long size; 779 780 if (!p) 781 return -EINVAL; 782 783 size = memparse(p, &p); 784 785 if (size < 4096) 786 size = 4096; 787 788 if (size > 640*1024) 789 size = 640*1024; 790 791 reserve_low = size; 792 793 return 0; 794 } 795 796 early_param("reservelow", parse_reservelow); 797 798 static void __init trim_low_memory_range(void) 799 { 800 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 801 } 802 803 /* 804 * Determine if we were loaded by an EFI loader. If so, then we have also been 805 * passed the efi memmap, systab, etc., so we should use these data structures 806 * for initialization. Note, the efi init code path is determined by the 807 * global efi_enabled. This allows the same kernel image to be used on existing 808 * systems (with a traditional BIOS) as well as on EFI systems. 809 */ 810 /* 811 * setup_arch - architecture-specific boot-time initializations 812 * 813 * Note: On x86_64, fixmaps are ready for use even before this is called. 814 */ 815 816 void __init setup_arch(char **cmdline_p) 817 { 818 memblock_reserve(__pa_symbol(_text), 819 (unsigned long)__bss_stop - (unsigned long)_text); 820 821 early_reserve_initrd(); 822 823 /* 824 * At this point everything still needed from the boot loader 825 * or BIOS or kernel text should be early reserved or marked not 826 * RAM in e820. All other memory is free game. 827 */ 828 829 #ifdef CONFIG_X86_32 830 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 831 visws_early_detect(); 832 833 /* 834 * copy kernel address range established so far and switch 835 * to the proper swapper page table 836 */ 837 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 838 initial_page_table + KERNEL_PGD_BOUNDARY, 839 KERNEL_PGD_PTRS); 840 841 load_cr3(swapper_pg_dir); 842 __flush_tlb_all(); 843 #else 844 printk(KERN_INFO "Command line: %s\n", boot_command_line); 845 #endif 846 847 /* 848 * If we have OLPC OFW, we might end up relocating the fixmap due to 849 * reserve_top(), so do this before touching the ioremap area. 850 */ 851 olpc_ofw_detect(); 852 853 early_trap_init(); 854 early_cpu_init(); 855 early_ioremap_init(); 856 857 setup_olpc_ofw_pgd(); 858 859 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 860 screen_info = boot_params.screen_info; 861 edid_info = boot_params.edid_info; 862 #ifdef CONFIG_X86_32 863 apm_info.bios = boot_params.apm_bios_info; 864 ist_info = boot_params.ist_info; 865 if (boot_params.sys_desc_table.length != 0) { 866 machine_id = boot_params.sys_desc_table.table[0]; 867 machine_submodel_id = boot_params.sys_desc_table.table[1]; 868 BIOS_revision = boot_params.sys_desc_table.table[2]; 869 } 870 #endif 871 saved_video_mode = boot_params.hdr.vid_mode; 872 bootloader_type = boot_params.hdr.type_of_loader; 873 if ((bootloader_type >> 4) == 0xe) { 874 bootloader_type &= 0xf; 875 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 876 } 877 bootloader_version = bootloader_type & 0xf; 878 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 879 880 #ifdef CONFIG_BLK_DEV_RAM 881 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 882 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 883 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 884 #endif 885 #ifdef CONFIG_EFI 886 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 887 "EL32", 4)) { 888 set_bit(EFI_BOOT, &x86_efi_facility); 889 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 890 "EL64", 4)) { 891 set_bit(EFI_BOOT, &x86_efi_facility); 892 set_bit(EFI_64BIT, &x86_efi_facility); 893 } 894 895 if (efi_enabled(EFI_BOOT)) 896 efi_memblock_x86_reserve_range(); 897 #endif 898 899 x86_init.oem.arch_setup(); 900 901 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 902 setup_memory_map(); 903 parse_setup_data(); 904 /* update the e820_saved too */ 905 e820_reserve_setup_data(); 906 907 copy_edd(); 908 909 if (!boot_params.hdr.root_flags) 910 root_mountflags &= ~MS_RDONLY; 911 init_mm.start_code = (unsigned long) _text; 912 init_mm.end_code = (unsigned long) _etext; 913 init_mm.end_data = (unsigned long) _edata; 914 init_mm.brk = _brk_end; 915 916 code_resource.start = __pa_symbol(_text); 917 code_resource.end = __pa_symbol(_etext)-1; 918 data_resource.start = __pa_symbol(_etext); 919 data_resource.end = __pa_symbol(_edata)-1; 920 bss_resource.start = __pa_symbol(__bss_start); 921 bss_resource.end = __pa_symbol(__bss_stop)-1; 922 923 #ifdef CONFIG_CMDLINE_BOOL 924 #ifdef CONFIG_CMDLINE_OVERRIDE 925 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 926 #else 927 if (builtin_cmdline[0]) { 928 /* append boot loader cmdline to builtin */ 929 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 930 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 931 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 932 } 933 #endif 934 #endif 935 936 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 937 *cmdline_p = command_line; 938 939 /* 940 * x86_configure_nx() is called before parse_early_param() to detect 941 * whether hardware doesn't support NX (so that the early EHCI debug 942 * console setup can safely call set_fixmap()). It may then be called 943 * again from within noexec_setup() during parsing early parameters 944 * to honor the respective command line option. 945 */ 946 x86_configure_nx(); 947 948 parse_early_param(); 949 950 x86_report_nx(); 951 952 /* after early param, so could get panic from serial */ 953 memblock_x86_reserve_range_setup_data(); 954 955 if (acpi_mps_check()) { 956 #ifdef CONFIG_X86_LOCAL_APIC 957 disable_apic = 1; 958 #endif 959 setup_clear_cpu_cap(X86_FEATURE_APIC); 960 } 961 962 #ifdef CONFIG_PCI 963 if (pci_early_dump_regs) 964 early_dump_pci_devices(); 965 #endif 966 967 finish_e820_parsing(); 968 969 if (efi_enabled(EFI_BOOT)) 970 efi_init(); 971 972 dmi_scan_machine(); 973 974 /* 975 * VMware detection requires dmi to be available, so this 976 * needs to be done after dmi_scan_machine, for the BP. 977 */ 978 init_hypervisor_platform(); 979 980 x86_init.resources.probe_roms(); 981 982 /* after parse_early_param, so could debug it */ 983 insert_resource(&iomem_resource, &code_resource); 984 insert_resource(&iomem_resource, &data_resource); 985 insert_resource(&iomem_resource, &bss_resource); 986 987 e820_add_kernel_range(); 988 trim_bios_range(); 989 #ifdef CONFIG_X86_32 990 if (ppro_with_ram_bug()) { 991 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 992 E820_RESERVED); 993 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 994 printk(KERN_INFO "fixed physical RAM map:\n"); 995 e820_print_map("bad_ppro"); 996 } 997 #else 998 early_gart_iommu_check(); 999 #endif 1000 1001 /* 1002 * partially used pages are not usable - thus 1003 * we are rounding upwards: 1004 */ 1005 max_pfn = e820_end_of_ram_pfn(); 1006 1007 /* update e820 for memory not covered by WB MTRRs */ 1008 mtrr_bp_init(); 1009 if (mtrr_trim_uncached_memory(max_pfn)) 1010 max_pfn = e820_end_of_ram_pfn(); 1011 1012 #ifdef CONFIG_X86_32 1013 /* max_low_pfn get updated here */ 1014 find_low_pfn_range(); 1015 #else 1016 num_physpages = max_pfn; 1017 1018 check_x2apic(); 1019 1020 /* How many end-of-memory variables you have, grandma! */ 1021 /* need this before calling reserve_initrd */ 1022 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1023 max_low_pfn = e820_end_of_low_ram_pfn(); 1024 else 1025 max_low_pfn = max_pfn; 1026 1027 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1028 #endif 1029 1030 /* 1031 * Find and reserve possible boot-time SMP configuration: 1032 */ 1033 find_smp_config(); 1034 1035 reserve_ibft_region(); 1036 1037 early_alloc_pgt_buf(); 1038 1039 /* 1040 * Need to conclude brk, before memblock_x86_fill() 1041 * it could use memblock_find_in_range, could overlap with 1042 * brk area. 1043 */ 1044 reserve_brk(); 1045 1046 cleanup_highmap(); 1047 1048 memblock.current_limit = ISA_END_ADDRESS; 1049 memblock_x86_fill(); 1050 1051 /* 1052 * The EFI specification says that boot service code won't be called 1053 * after ExitBootServices(). This is, in fact, a lie. 1054 */ 1055 if (efi_enabled(EFI_MEMMAP)) 1056 efi_reserve_boot_services(); 1057 1058 /* preallocate 4k for mptable mpc */ 1059 early_reserve_e820_mpc_new(); 1060 1061 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1062 setup_bios_corruption_check(); 1063 #endif 1064 1065 #ifdef CONFIG_X86_32 1066 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1067 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1068 #endif 1069 1070 reserve_real_mode(); 1071 1072 trim_platform_memory_ranges(); 1073 trim_low_memory_range(); 1074 1075 init_mem_mapping(); 1076 1077 early_trap_pf_init(); 1078 1079 setup_real_mode(); 1080 1081 memblock.current_limit = get_max_mapped(); 1082 dma_contiguous_reserve(0); 1083 1084 /* 1085 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1086 */ 1087 1088 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1089 if (init_ohci1394_dma_early) 1090 init_ohci1394_dma_on_all_controllers(); 1091 #endif 1092 /* Allocate bigger log buffer */ 1093 setup_log_buf(1); 1094 1095 reserve_initrd(); 1096 1097 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD) 1098 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start); 1099 #endif 1100 1101 reserve_crashkernel(); 1102 1103 vsmp_init(); 1104 1105 io_delay_init(); 1106 1107 /* 1108 * Parse the ACPI tables for possible boot-time SMP configuration. 1109 */ 1110 acpi_boot_table_init(); 1111 1112 early_acpi_boot_init(); 1113 1114 initmem_init(); 1115 memblock_find_dma_reserve(); 1116 1117 #ifdef CONFIG_KVM_GUEST 1118 kvmclock_init(); 1119 #endif 1120 1121 x86_init.paging.pagetable_init(); 1122 1123 if (boot_cpu_data.cpuid_level >= 0) { 1124 /* A CPU has %cr4 if and only if it has CPUID */ 1125 mmu_cr4_features = read_cr4(); 1126 if (trampoline_cr4_features) 1127 *trampoline_cr4_features = mmu_cr4_features; 1128 } 1129 1130 #ifdef CONFIG_X86_32 1131 /* sync back kernel address range */ 1132 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1133 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1134 KERNEL_PGD_PTRS); 1135 #endif 1136 1137 tboot_probe(); 1138 1139 #ifdef CONFIG_X86_64 1140 map_vsyscall(); 1141 #endif 1142 1143 generic_apic_probe(); 1144 1145 early_quirks(); 1146 1147 /* 1148 * Read APIC and some other early information from ACPI tables. 1149 */ 1150 acpi_boot_init(); 1151 sfi_init(); 1152 x86_dtb_init(); 1153 1154 /* 1155 * get boot-time SMP configuration: 1156 */ 1157 if (smp_found_config) 1158 get_smp_config(); 1159 1160 prefill_possible_map(); 1161 1162 init_cpu_to_node(); 1163 1164 init_apic_mappings(); 1165 if (x86_io_apic_ops.init) 1166 x86_io_apic_ops.init(); 1167 1168 kvm_guest_init(); 1169 1170 e820_reserve_resources(); 1171 e820_mark_nosave_regions(max_low_pfn); 1172 1173 x86_init.resources.reserve_resources(); 1174 1175 e820_setup_gap(); 1176 1177 #ifdef CONFIG_VT 1178 #if defined(CONFIG_VGA_CONSOLE) 1179 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1180 conswitchp = &vga_con; 1181 #elif defined(CONFIG_DUMMY_CONSOLE) 1182 conswitchp = &dummy_con; 1183 #endif 1184 #endif 1185 x86_init.oem.banner(); 1186 1187 x86_init.timers.wallclock_init(); 1188 1189 mcheck_init(); 1190 1191 arch_init_ideal_nops(); 1192 1193 register_refined_jiffies(CLOCK_TICK_RATE); 1194 1195 #ifdef CONFIG_EFI 1196 /* Once setup is done above, unmap the EFI memory map on 1197 * mismatched firmware/kernel archtectures since there is no 1198 * support for runtime services. 1199 */ 1200 if (efi_enabled(EFI_BOOT) && !efi_is_native()) { 1201 pr_info("efi: Setup done, disabling due to 32/64-bit mismatch\n"); 1202 efi_unmap_memmap(); 1203 } 1204 #endif 1205 } 1206 1207 #ifdef CONFIG_X86_32 1208 1209 static struct resource video_ram_resource = { 1210 .name = "Video RAM area", 1211 .start = 0xa0000, 1212 .end = 0xbffff, 1213 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1214 }; 1215 1216 void __init i386_reserve_resources(void) 1217 { 1218 request_resource(&iomem_resource, &video_ram_resource); 1219 reserve_standard_io_resources(); 1220 } 1221 1222 #endif /* CONFIG_X86_32 */ 1223