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