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/io_apic.h> 86 #include <asm/ist.h> 87 #include <asm/setup_arch.h> 88 #include <asm/bios_ebda.h> 89 #include <asm/cacheflush.h> 90 #include <asm/processor.h> 91 #include <asm/bugs.h> 92 93 #include <asm/vsyscall.h> 94 #include <asm/cpu.h> 95 #include <asm/desc.h> 96 #include <asm/dma.h> 97 #include <asm/iommu.h> 98 #include <asm/gart.h> 99 #include <asm/mmu_context.h> 100 #include <asm/proto.h> 101 102 #include <asm/paravirt.h> 103 #include <asm/hypervisor.h> 104 #include <asm/olpc_ofw.h> 105 106 #include <asm/percpu.h> 107 #include <asm/topology.h> 108 #include <asm/apicdef.h> 109 #include <asm/amd_nb.h> 110 #include <asm/mce.h> 111 #include <asm/alternative.h> 112 #include <asm/prom.h> 113 114 /* 115 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 116 * max_pfn_mapped: highest direct mapped pfn over 4GB 117 * 118 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are 119 * represented by pfn_mapped 120 */ 121 unsigned long max_low_pfn_mapped; 122 unsigned long max_pfn_mapped; 123 124 #ifdef CONFIG_DMI 125 RESERVE_BRK(dmi_alloc, 65536); 126 #endif 127 128 129 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 130 unsigned long _brk_end = (unsigned long)__brk_base; 131 132 #ifdef CONFIG_X86_64 133 int default_cpu_present_to_apicid(int mps_cpu) 134 { 135 return __default_cpu_present_to_apicid(mps_cpu); 136 } 137 138 int default_check_phys_apicid_present(int phys_apicid) 139 { 140 return __default_check_phys_apicid_present(phys_apicid); 141 } 142 #endif 143 144 struct boot_params boot_params; 145 146 /* 147 * Machine setup.. 148 */ 149 static struct resource data_resource = { 150 .name = "Kernel data", 151 .start = 0, 152 .end = 0, 153 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 154 }; 155 156 static struct resource code_resource = { 157 .name = "Kernel code", 158 .start = 0, 159 .end = 0, 160 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 161 }; 162 163 static struct resource bss_resource = { 164 .name = "Kernel bss", 165 .start = 0, 166 .end = 0, 167 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 168 }; 169 170 171 #ifdef CONFIG_X86_32 172 /* cpu data as detected by the assembly code in head.S */ 173 struct cpuinfo_x86 new_cpu_data = { 174 .wp_works_ok = -1, 175 }; 176 /* common cpu data for all cpus */ 177 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 178 .wp_works_ok = -1, 179 }; 180 EXPORT_SYMBOL(boot_cpu_data); 181 182 unsigned int def_to_bigsmp; 183 184 /* for MCA, but anyone else can use it if they want */ 185 unsigned int machine_id; 186 unsigned int machine_submodel_id; 187 unsigned int BIOS_revision; 188 189 struct apm_info apm_info; 190 EXPORT_SYMBOL(apm_info); 191 192 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 193 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 194 struct ist_info ist_info; 195 EXPORT_SYMBOL(ist_info); 196 #else 197 struct ist_info ist_info; 198 #endif 199 200 #else 201 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 202 .x86_phys_bits = MAX_PHYSMEM_BITS, 203 }; 204 EXPORT_SYMBOL(boot_cpu_data); 205 #endif 206 207 208 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 209 __visible unsigned long mmu_cr4_features; 210 #else 211 __visible unsigned long mmu_cr4_features = X86_CR4_PAE; 212 #endif 213 214 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 215 int bootloader_type, bootloader_version; 216 217 /* 218 * Setup options 219 */ 220 struct screen_info screen_info; 221 EXPORT_SYMBOL(screen_info); 222 struct edid_info edid_info; 223 EXPORT_SYMBOL_GPL(edid_info); 224 225 extern int root_mountflags; 226 227 unsigned long saved_video_mode; 228 229 #define RAMDISK_IMAGE_START_MASK 0x07FF 230 #define RAMDISK_PROMPT_FLAG 0x8000 231 #define RAMDISK_LOAD_FLAG 0x4000 232 233 static char __initdata command_line[COMMAND_LINE_SIZE]; 234 #ifdef CONFIG_CMDLINE_BOOL 235 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 236 #endif 237 238 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 239 struct edd edd; 240 #ifdef CONFIG_EDD_MODULE 241 EXPORT_SYMBOL(edd); 242 #endif 243 /** 244 * copy_edd() - Copy the BIOS EDD information 245 * from boot_params into a safe place. 246 * 247 */ 248 static inline void __init copy_edd(void) 249 { 250 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 251 sizeof(edd.mbr_signature)); 252 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 253 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 254 edd.edd_info_nr = boot_params.eddbuf_entries; 255 } 256 #else 257 static inline void __init copy_edd(void) 258 { 259 } 260 #endif 261 262 void * __init extend_brk(size_t size, size_t align) 263 { 264 size_t mask = align - 1; 265 void *ret; 266 267 BUG_ON(_brk_start == 0); 268 BUG_ON(align & mask); 269 270 _brk_end = (_brk_end + mask) & ~mask; 271 BUG_ON((char *)(_brk_end + size) > __brk_limit); 272 273 ret = (void *)_brk_end; 274 _brk_end += size; 275 276 memset(ret, 0, size); 277 278 return ret; 279 } 280 281 #ifdef CONFIG_X86_32 282 static void __init cleanup_highmap(void) 283 { 284 } 285 #endif 286 287 static void __init reserve_brk(void) 288 { 289 if (_brk_end > _brk_start) 290 memblock_reserve(__pa_symbol(_brk_start), 291 _brk_end - _brk_start); 292 293 /* Mark brk area as locked down and no longer taking any 294 new allocations */ 295 _brk_start = 0; 296 } 297 298 #ifdef CONFIG_BLK_DEV_INITRD 299 300 static u64 __init get_ramdisk_image(void) 301 { 302 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 303 304 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 305 306 return ramdisk_image; 307 } 308 static u64 __init get_ramdisk_size(void) 309 { 310 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 311 312 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 313 314 return ramdisk_size; 315 } 316 317 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT) 318 static void __init relocate_initrd(void) 319 { 320 /* Assume only end is not page aligned */ 321 u64 ramdisk_image = get_ramdisk_image(); 322 u64 ramdisk_size = get_ramdisk_size(); 323 u64 area_size = PAGE_ALIGN(ramdisk_size); 324 u64 ramdisk_here; 325 unsigned long slop, clen, mapaddr; 326 char *p, *q; 327 328 /* We need to move the initrd down into directly mapped mem */ 329 ramdisk_here = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 330 area_size, PAGE_SIZE); 331 332 if (!ramdisk_here) 333 panic("Cannot find place for new RAMDISK of size %lld\n", 334 ramdisk_size); 335 336 /* Note: this includes all the mem currently occupied by 337 the initrd, we rely on that fact to keep the data intact. */ 338 memblock_reserve(ramdisk_here, area_size); 339 initrd_start = ramdisk_here + PAGE_OFFSET; 340 initrd_end = initrd_start + ramdisk_size; 341 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 342 ramdisk_here, ramdisk_here + ramdisk_size - 1); 343 344 q = (char *)initrd_start; 345 346 /* Copy the initrd */ 347 while (ramdisk_size) { 348 slop = ramdisk_image & ~PAGE_MASK; 349 clen = ramdisk_size; 350 if (clen > MAX_MAP_CHUNK-slop) 351 clen = MAX_MAP_CHUNK-slop; 352 mapaddr = ramdisk_image & PAGE_MASK; 353 p = early_memremap(mapaddr, clen+slop); 354 memcpy(q, p+slop, clen); 355 early_iounmap(p, clen+slop); 356 q += clen; 357 ramdisk_image += clen; 358 ramdisk_size -= clen; 359 } 360 361 ramdisk_image = get_ramdisk_image(); 362 ramdisk_size = get_ramdisk_size(); 363 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 364 " [mem %#010llx-%#010llx]\n", 365 ramdisk_image, ramdisk_image + ramdisk_size - 1, 366 ramdisk_here, ramdisk_here + ramdisk_size - 1); 367 } 368 369 static void __init early_reserve_initrd(void) 370 { 371 /* Assume only end is not page aligned */ 372 u64 ramdisk_image = get_ramdisk_image(); 373 u64 ramdisk_size = get_ramdisk_size(); 374 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 375 376 if (!boot_params.hdr.type_of_loader || 377 !ramdisk_image || !ramdisk_size) 378 return; /* No initrd provided by bootloader */ 379 380 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 381 } 382 static void __init reserve_initrd(void) 383 { 384 /* Assume only end is not page aligned */ 385 u64 ramdisk_image = get_ramdisk_image(); 386 u64 ramdisk_size = get_ramdisk_size(); 387 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 388 u64 mapped_size; 389 390 if (!boot_params.hdr.type_of_loader || 391 !ramdisk_image || !ramdisk_size) 392 return; /* No initrd provided by bootloader */ 393 394 initrd_start = 0; 395 396 mapped_size = memblock_mem_size(max_pfn_mapped); 397 if (ramdisk_size >= (mapped_size>>1)) 398 panic("initrd too large to handle, " 399 "disabling initrd (%lld needed, %lld available)\n", 400 ramdisk_size, mapped_size>>1); 401 402 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 403 ramdisk_end - 1); 404 405 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 406 PFN_DOWN(ramdisk_end))) { 407 /* All are mapped, easy case */ 408 initrd_start = ramdisk_image + PAGE_OFFSET; 409 initrd_end = initrd_start + ramdisk_size; 410 return; 411 } 412 413 relocate_initrd(); 414 415 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 416 } 417 #else 418 static void __init early_reserve_initrd(void) 419 { 420 } 421 static void __init reserve_initrd(void) 422 { 423 } 424 #endif /* CONFIG_BLK_DEV_INITRD */ 425 426 static void __init parse_setup_data(void) 427 { 428 struct setup_data *data; 429 u64 pa_data, pa_next; 430 431 pa_data = boot_params.hdr.setup_data; 432 while (pa_data) { 433 u32 data_len, map_len, data_type; 434 435 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK), 436 (u64)sizeof(struct setup_data)); 437 data = early_memremap(pa_data, map_len); 438 data_len = data->len + sizeof(struct setup_data); 439 data_type = data->type; 440 pa_next = data->next; 441 early_iounmap(data, map_len); 442 443 switch (data_type) { 444 case SETUP_E820_EXT: 445 parse_e820_ext(pa_data, data_len); 446 break; 447 case SETUP_DTB: 448 add_dtb(pa_data); 449 break; 450 default: 451 break; 452 } 453 pa_data = pa_next; 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 * On 64bit, old kexec-tools need to under 896MiB. 505 */ 506 #ifdef CONFIG_X86_32 507 # define CRASH_KERNEL_ADDR_LOW_MAX (512 << 20) 508 # define CRASH_KERNEL_ADDR_HIGH_MAX (512 << 20) 509 #else 510 # define CRASH_KERNEL_ADDR_LOW_MAX (896UL<<20) 511 # define CRASH_KERNEL_ADDR_HIGH_MAX MAXMEM 512 #endif 513 514 static void __init reserve_crashkernel_low(void) 515 { 516 #ifdef CONFIG_X86_64 517 const unsigned long long alignment = 16<<20; /* 16M */ 518 unsigned long long low_base = 0, low_size = 0; 519 unsigned long total_low_mem; 520 unsigned long long base; 521 bool auto_set = false; 522 int ret; 523 524 total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT)); 525 /* crashkernel=Y,low */ 526 ret = parse_crashkernel_low(boot_command_line, total_low_mem, 527 &low_size, &base); 528 if (ret != 0) { 529 /* 530 * two parts from lib/swiotlb.c: 531 * swiotlb size: user specified with swiotlb= or default. 532 * swiotlb overflow buffer: now is hardcoded to 32k. 533 * We round it to 8M for other buffers that 534 * may need to stay low too. 535 */ 536 low_size = swiotlb_size_or_default() + (8UL<<20); 537 auto_set = true; 538 } else { 539 /* passed with crashkernel=0,low ? */ 540 if (!low_size) 541 return; 542 } 543 544 low_base = memblock_find_in_range(low_size, (1ULL<<32), 545 low_size, alignment); 546 547 if (!low_base) { 548 if (!auto_set) 549 pr_info("crashkernel low reservation failed - No suitable area found.\n"); 550 551 return; 552 } 553 554 memblock_reserve(low_base, low_size); 555 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 556 (unsigned long)(low_size >> 20), 557 (unsigned long)(low_base >> 20), 558 (unsigned long)(total_low_mem >> 20)); 559 crashk_low_res.start = low_base; 560 crashk_low_res.end = low_base + low_size - 1; 561 insert_resource(&iomem_resource, &crashk_low_res); 562 #endif 563 } 564 565 static void __init reserve_crashkernel(void) 566 { 567 const unsigned long long alignment = 16<<20; /* 16M */ 568 unsigned long long total_mem; 569 unsigned long long crash_size, crash_base; 570 bool high = false; 571 int ret; 572 573 total_mem = memblock_phys_mem_size(); 574 575 /* crashkernel=XM */ 576 ret = parse_crashkernel(boot_command_line, total_mem, 577 &crash_size, &crash_base); 578 if (ret != 0 || crash_size <= 0) { 579 /* crashkernel=X,high */ 580 ret = parse_crashkernel_high(boot_command_line, total_mem, 581 &crash_size, &crash_base); 582 if (ret != 0 || crash_size <= 0) 583 return; 584 high = true; 585 } 586 587 /* 0 means: find the address automatically */ 588 if (crash_base <= 0) { 589 /* 590 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX 591 */ 592 crash_base = memblock_find_in_range(alignment, 593 high ? CRASH_KERNEL_ADDR_HIGH_MAX : 594 CRASH_KERNEL_ADDR_LOW_MAX, 595 crash_size, alignment); 596 597 if (!crash_base) { 598 pr_info("crashkernel reservation failed - No suitable area found.\n"); 599 return; 600 } 601 602 } else { 603 unsigned long long start; 604 605 start = memblock_find_in_range(crash_base, 606 crash_base + crash_size, crash_size, 1<<20); 607 if (start != crash_base) { 608 pr_info("crashkernel reservation failed - memory is in use.\n"); 609 return; 610 } 611 } 612 memblock_reserve(crash_base, crash_size); 613 614 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " 615 "for crashkernel (System RAM: %ldMB)\n", 616 (unsigned long)(crash_size >> 20), 617 (unsigned long)(crash_base >> 20), 618 (unsigned long)(total_mem >> 20)); 619 620 crashk_res.start = crash_base; 621 crashk_res.end = crash_base + crash_size - 1; 622 insert_resource(&iomem_resource, &crashk_res); 623 624 if (crash_base >= (1ULL<<32)) 625 reserve_crashkernel_low(); 626 } 627 #else 628 static void __init reserve_crashkernel(void) 629 { 630 } 631 #endif 632 633 static struct resource standard_io_resources[] = { 634 { .name = "dma1", .start = 0x00, .end = 0x1f, 635 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 636 { .name = "pic1", .start = 0x20, .end = 0x21, 637 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 638 { .name = "timer0", .start = 0x40, .end = 0x43, 639 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 640 { .name = "timer1", .start = 0x50, .end = 0x53, 641 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 642 { .name = "keyboard", .start = 0x60, .end = 0x60, 643 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 644 { .name = "keyboard", .start = 0x64, .end = 0x64, 645 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 646 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 647 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 648 { .name = "pic2", .start = 0xa0, .end = 0xa1, 649 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 650 { .name = "dma2", .start = 0xc0, .end = 0xdf, 651 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 652 { .name = "fpu", .start = 0xf0, .end = 0xff, 653 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 654 }; 655 656 void __init reserve_standard_io_resources(void) 657 { 658 int i; 659 660 /* request I/O space for devices used on all i[345]86 PCs */ 661 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 662 request_resource(&ioport_resource, &standard_io_resources[i]); 663 664 } 665 666 static __init void reserve_ibft_region(void) 667 { 668 unsigned long addr, size = 0; 669 670 addr = find_ibft_region(&size); 671 672 if (size) 673 memblock_reserve(addr, size); 674 } 675 676 static bool __init snb_gfx_workaround_needed(void) 677 { 678 #ifdef CONFIG_PCI 679 int i; 680 u16 vendor, devid; 681 static const __initconst u16 snb_ids[] = { 682 0x0102, 683 0x0112, 684 0x0122, 685 0x0106, 686 0x0116, 687 0x0126, 688 0x010a, 689 }; 690 691 /* Assume no if something weird is going on with PCI */ 692 if (!early_pci_allowed()) 693 return false; 694 695 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 696 if (vendor != 0x8086) 697 return false; 698 699 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 700 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 701 if (devid == snb_ids[i]) 702 return true; 703 #endif 704 705 return false; 706 } 707 708 /* 709 * Sandy Bridge graphics has trouble with certain ranges, exclude 710 * them from allocation. 711 */ 712 static void __init trim_snb_memory(void) 713 { 714 static const __initconst unsigned long bad_pages[] = { 715 0x20050000, 716 0x20110000, 717 0x20130000, 718 0x20138000, 719 0x40004000, 720 }; 721 int i; 722 723 if (!snb_gfx_workaround_needed()) 724 return; 725 726 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 727 728 /* 729 * Reserve all memory below the 1 MB mark that has not 730 * already been reserved. 731 */ 732 memblock_reserve(0, 1<<20); 733 734 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 735 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 736 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 737 bad_pages[i]); 738 } 739 } 740 741 /* 742 * Here we put platform-specific memory range workarounds, i.e. 743 * memory known to be corrupt or otherwise in need to be reserved on 744 * specific platforms. 745 * 746 * If this gets used more widely it could use a real dispatch mechanism. 747 */ 748 static void __init trim_platform_memory_ranges(void) 749 { 750 trim_snb_memory(); 751 } 752 753 static void __init trim_bios_range(void) 754 { 755 /* 756 * A special case is the first 4Kb of memory; 757 * This is a BIOS owned area, not kernel ram, but generally 758 * not listed as such in the E820 table. 759 * 760 * This typically reserves additional memory (64KiB by default) 761 * since some BIOSes are known to corrupt low memory. See the 762 * Kconfig help text for X86_RESERVE_LOW. 763 */ 764 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED); 765 766 /* 767 * special case: Some BIOSen report the PC BIOS 768 * area (640->1Mb) as ram even though it is not. 769 * take them out. 770 */ 771 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1); 772 773 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 774 } 775 776 /* called before trim_bios_range() to spare extra sanitize */ 777 static void __init e820_add_kernel_range(void) 778 { 779 u64 start = __pa_symbol(_text); 780 u64 size = __pa_symbol(_end) - start; 781 782 /* 783 * Complain if .text .data and .bss are not marked as E820_RAM and 784 * attempt to fix it by adding the range. We may have a confused BIOS, 785 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 786 * exclude kernel range. If we really are running on top non-RAM, 787 * we will crash later anyways. 788 */ 789 if (e820_all_mapped(start, start + size, E820_RAM)) 790 return; 791 792 pr_warn(".text .data .bss are not marked as E820_RAM!\n"); 793 e820_remove_range(start, size, E820_RAM, 0); 794 e820_add_region(start, size, E820_RAM); 795 } 796 797 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 798 799 static int __init parse_reservelow(char *p) 800 { 801 unsigned long long size; 802 803 if (!p) 804 return -EINVAL; 805 806 size = memparse(p, &p); 807 808 if (size < 4096) 809 size = 4096; 810 811 if (size > 640*1024) 812 size = 640*1024; 813 814 reserve_low = size; 815 816 return 0; 817 } 818 819 early_param("reservelow", parse_reservelow); 820 821 static void __init trim_low_memory_range(void) 822 { 823 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 824 } 825 826 /* 827 * Determine if we were loaded by an EFI loader. If so, then we have also been 828 * passed the efi memmap, systab, etc., so we should use these data structures 829 * for initialization. Note, the efi init code path is determined by the 830 * global efi_enabled. This allows the same kernel image to be used on existing 831 * systems (with a traditional BIOS) as well as on EFI systems. 832 */ 833 /* 834 * setup_arch - architecture-specific boot-time initializations 835 * 836 * Note: On x86_64, fixmaps are ready for use even before this is called. 837 */ 838 839 void __init setup_arch(char **cmdline_p) 840 { 841 memblock_reserve(__pa_symbol(_text), 842 (unsigned long)__bss_stop - (unsigned long)_text); 843 844 early_reserve_initrd(); 845 846 /* 847 * At this point everything still needed from the boot loader 848 * or BIOS or kernel text should be early reserved or marked not 849 * RAM in e820. All other memory is free game. 850 */ 851 852 #ifdef CONFIG_X86_32 853 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 854 visws_early_detect(); 855 856 /* 857 * copy kernel address range established so far and switch 858 * to the proper swapper page table 859 */ 860 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 861 initial_page_table + KERNEL_PGD_BOUNDARY, 862 KERNEL_PGD_PTRS); 863 864 load_cr3(swapper_pg_dir); 865 __flush_tlb_all(); 866 #else 867 printk(KERN_INFO "Command line: %s\n", boot_command_line); 868 #endif 869 870 /* 871 * If we have OLPC OFW, we might end up relocating the fixmap due to 872 * reserve_top(), so do this before touching the ioremap area. 873 */ 874 olpc_ofw_detect(); 875 876 early_trap_init(); 877 early_cpu_init(); 878 early_ioremap_init(); 879 880 setup_olpc_ofw_pgd(); 881 882 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 883 screen_info = boot_params.screen_info; 884 edid_info = boot_params.edid_info; 885 #ifdef CONFIG_X86_32 886 apm_info.bios = boot_params.apm_bios_info; 887 ist_info = boot_params.ist_info; 888 if (boot_params.sys_desc_table.length != 0) { 889 machine_id = boot_params.sys_desc_table.table[0]; 890 machine_submodel_id = boot_params.sys_desc_table.table[1]; 891 BIOS_revision = boot_params.sys_desc_table.table[2]; 892 } 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 "EL32", 4)) { 911 set_bit(EFI_BOOT, &x86_efi_facility); 912 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 913 "EL64", 4)) { 914 set_bit(EFI_BOOT, &x86_efi_facility); 915 set_bit(EFI_64BIT, &x86_efi_facility); 916 } 917 918 if (efi_enabled(EFI_BOOT)) 919 efi_memblock_x86_reserve_range(); 920 #endif 921 922 x86_init.oem.arch_setup(); 923 924 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 925 setup_memory_map(); 926 parse_setup_data(); 927 /* update the e820_saved too */ 928 e820_reserve_setup_data(); 929 930 copy_edd(); 931 932 if (!boot_params.hdr.root_flags) 933 root_mountflags &= ~MS_RDONLY; 934 init_mm.start_code = (unsigned long) _text; 935 init_mm.end_code = (unsigned long) _etext; 936 init_mm.end_data = (unsigned long) _edata; 937 init_mm.brk = _brk_end; 938 939 code_resource.start = __pa_symbol(_text); 940 code_resource.end = __pa_symbol(_etext)-1; 941 data_resource.start = __pa_symbol(_etext); 942 data_resource.end = __pa_symbol(_edata)-1; 943 bss_resource.start = __pa_symbol(__bss_start); 944 bss_resource.end = __pa_symbol(__bss_stop)-1; 945 946 #ifdef CONFIG_CMDLINE_BOOL 947 #ifdef CONFIG_CMDLINE_OVERRIDE 948 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 949 #else 950 if (builtin_cmdline[0]) { 951 /* append boot loader cmdline to builtin */ 952 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 953 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 954 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 955 } 956 #endif 957 #endif 958 959 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 960 *cmdline_p = command_line; 961 962 /* 963 * x86_configure_nx() is called before parse_early_param() to detect 964 * whether hardware doesn't support NX (so that the early EHCI debug 965 * console setup can safely call set_fixmap()). It may then be called 966 * again from within noexec_setup() during parsing early parameters 967 * to honor the respective command line option. 968 */ 969 x86_configure_nx(); 970 971 parse_early_param(); 972 973 x86_report_nx(); 974 975 /* after early param, so could get panic from serial */ 976 memblock_x86_reserve_range_setup_data(); 977 978 if (acpi_mps_check()) { 979 #ifdef CONFIG_X86_LOCAL_APIC 980 disable_apic = 1; 981 #endif 982 setup_clear_cpu_cap(X86_FEATURE_APIC); 983 } 984 985 #ifdef CONFIG_PCI 986 if (pci_early_dump_regs) 987 early_dump_pci_devices(); 988 #endif 989 990 finish_e820_parsing(); 991 992 if (efi_enabled(EFI_BOOT)) 993 efi_init(); 994 995 dmi_scan_machine(); 996 dmi_set_dump_stack_arch_desc(); 997 998 /* 999 * VMware detection requires dmi to be available, so this 1000 * needs to be done after dmi_scan_machine, for the BP. 1001 */ 1002 init_hypervisor_platform(); 1003 1004 x86_init.resources.probe_roms(); 1005 1006 /* after parse_early_param, so could debug it */ 1007 insert_resource(&iomem_resource, &code_resource); 1008 insert_resource(&iomem_resource, &data_resource); 1009 insert_resource(&iomem_resource, &bss_resource); 1010 1011 e820_add_kernel_range(); 1012 trim_bios_range(); 1013 #ifdef CONFIG_X86_32 1014 if (ppro_with_ram_bug()) { 1015 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM, 1016 E820_RESERVED); 1017 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); 1018 printk(KERN_INFO "fixed physical RAM map:\n"); 1019 e820_print_map("bad_ppro"); 1020 } 1021 #else 1022 early_gart_iommu_check(); 1023 #endif 1024 1025 /* 1026 * partially used pages are not usable - thus 1027 * we are rounding upwards: 1028 */ 1029 max_pfn = e820_end_of_ram_pfn(); 1030 1031 /* update e820 for memory not covered by WB MTRRs */ 1032 mtrr_bp_init(); 1033 if (mtrr_trim_uncached_memory(max_pfn)) 1034 max_pfn = e820_end_of_ram_pfn(); 1035 1036 #ifdef CONFIG_X86_32 1037 /* max_low_pfn get updated here */ 1038 find_low_pfn_range(); 1039 #else 1040 check_x2apic(); 1041 1042 /* How many end-of-memory variables you have, grandma! */ 1043 /* need this before calling reserve_initrd */ 1044 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1045 max_low_pfn = e820_end_of_low_ram_pfn(); 1046 else 1047 max_low_pfn = max_pfn; 1048 1049 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1050 #endif 1051 1052 /* 1053 * Find and reserve possible boot-time SMP configuration: 1054 */ 1055 find_smp_config(); 1056 1057 reserve_ibft_region(); 1058 1059 early_alloc_pgt_buf(); 1060 1061 /* 1062 * Need to conclude brk, before memblock_x86_fill() 1063 * it could use memblock_find_in_range, could overlap with 1064 * brk area. 1065 */ 1066 reserve_brk(); 1067 1068 cleanup_highmap(); 1069 1070 memblock_set_current_limit(ISA_END_ADDRESS); 1071 memblock_x86_fill(); 1072 1073 /* 1074 * The EFI specification says that boot service code won't be called 1075 * after ExitBootServices(). This is, in fact, a lie. 1076 */ 1077 if (efi_enabled(EFI_MEMMAP)) 1078 efi_reserve_boot_services(); 1079 1080 /* preallocate 4k for mptable mpc */ 1081 early_reserve_e820_mpc_new(); 1082 1083 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1084 setup_bios_corruption_check(); 1085 #endif 1086 1087 #ifdef CONFIG_X86_32 1088 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1089 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1090 #endif 1091 1092 reserve_real_mode(); 1093 1094 trim_platform_memory_ranges(); 1095 trim_low_memory_range(); 1096 1097 init_mem_mapping(); 1098 1099 early_trap_pf_init(); 1100 1101 setup_real_mode(); 1102 1103 memblock_set_current_limit(get_max_mapped()); 1104 dma_contiguous_reserve(0); 1105 1106 /* 1107 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1108 */ 1109 1110 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1111 if (init_ohci1394_dma_early) 1112 init_ohci1394_dma_on_all_controllers(); 1113 #endif 1114 /* Allocate bigger log buffer */ 1115 setup_log_buf(1); 1116 1117 reserve_initrd(); 1118 1119 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD) 1120 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start); 1121 #endif 1122 1123 reserve_crashkernel(); 1124 1125 vsmp_init(); 1126 1127 io_delay_init(); 1128 1129 /* 1130 * Parse the ACPI tables for possible boot-time SMP configuration. 1131 */ 1132 acpi_boot_table_init(); 1133 1134 early_acpi_boot_init(); 1135 1136 initmem_init(); 1137 memblock_find_dma_reserve(); 1138 1139 #ifdef CONFIG_KVM_GUEST 1140 kvmclock_init(); 1141 #endif 1142 1143 x86_init.paging.pagetable_init(); 1144 1145 if (boot_cpu_data.cpuid_level >= 0) { 1146 /* A CPU has %cr4 if and only if it has CPUID */ 1147 mmu_cr4_features = read_cr4(); 1148 if (trampoline_cr4_features) 1149 *trampoline_cr4_features = mmu_cr4_features; 1150 } 1151 1152 #ifdef CONFIG_X86_32 1153 /* sync back kernel address range */ 1154 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 1155 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 1156 KERNEL_PGD_PTRS); 1157 #endif 1158 1159 tboot_probe(); 1160 1161 #ifdef CONFIG_X86_64 1162 map_vsyscall(); 1163 #endif 1164 1165 generic_apic_probe(); 1166 1167 early_quirks(); 1168 1169 /* 1170 * Read APIC and some other early information from ACPI tables. 1171 */ 1172 acpi_boot_init(); 1173 sfi_init(); 1174 x86_dtb_init(); 1175 1176 /* 1177 * get boot-time SMP configuration: 1178 */ 1179 if (smp_found_config) 1180 get_smp_config(); 1181 1182 prefill_possible_map(); 1183 1184 init_cpu_to_node(); 1185 1186 init_apic_mappings(); 1187 if (x86_io_apic_ops.init) 1188 x86_io_apic_ops.init(); 1189 1190 kvm_guest_init(); 1191 1192 e820_reserve_resources(); 1193 e820_mark_nosave_regions(max_low_pfn); 1194 1195 x86_init.resources.reserve_resources(); 1196 1197 e820_setup_gap(); 1198 1199 #ifdef CONFIG_VT 1200 #if defined(CONFIG_VGA_CONSOLE) 1201 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1202 conswitchp = &vga_con; 1203 #elif defined(CONFIG_DUMMY_CONSOLE) 1204 conswitchp = &dummy_con; 1205 #endif 1206 #endif 1207 x86_init.oem.banner(); 1208 1209 x86_init.timers.wallclock_init(); 1210 1211 mcheck_init(); 1212 1213 arch_init_ideal_nops(); 1214 1215 register_refined_jiffies(CLOCK_TICK_RATE); 1216 1217 #ifdef CONFIG_EFI 1218 /* Once setup is done above, unmap the EFI memory map on 1219 * mismatched firmware/kernel archtectures since there is no 1220 * support for runtime services. 1221 */ 1222 if (efi_enabled(EFI_BOOT) && !efi_is_native()) { 1223 pr_info("efi: Setup done, disabling due to 32/64-bit mismatch\n"); 1224 efi_unmap_memmap(); 1225 } 1226 #endif 1227 } 1228 1229 #ifdef CONFIG_X86_32 1230 1231 static struct resource video_ram_resource = { 1232 .name = "Video RAM area", 1233 .start = 0xa0000, 1234 .end = 0xbffff, 1235 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1236 }; 1237 1238 void __init i386_reserve_resources(void) 1239 { 1240 request_resource(&iomem_resource, &video_ram_resource); 1241 reserve_standard_io_resources(); 1242 } 1243 1244 #endif /* CONFIG_X86_32 */ 1245