1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1995 Linus Torvalds 4 * 5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 6 * 7 * Memory region support 8 * David Parsons <orc@pell.chi.il.us>, July-August 1999 9 * 10 * Added E820 sanitization routine (removes overlapping memory regions); 11 * Brian Moyle <bmoyle@mvista.com>, February 2001 12 * 13 * Moved CPU detection code to cpu/${cpu}.c 14 * Patrick Mochel <mochel@osdl.org>, March 2002 15 * 16 * Provisions for empty E820 memory regions (reported by certain BIOSes). 17 * Alex Achenbach <xela@slit.de>, December 2002. 18 * 19 */ 20 21 /* 22 * This file handles the architecture-dependent parts of initialization 23 */ 24 25 #include <linux/sched.h> 26 #include <linux/mm.h> 27 #include <linux/mmzone.h> 28 #include <linux/screen_info.h> 29 #include <linux/ioport.h> 30 #include <linux/acpi.h> 31 #include <linux/sfi.h> 32 #include <linux/apm_bios.h> 33 #include <linux/initrd.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/export.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 #include <xen/xen.h> 54 #include <uapi/linux/mount.h> 55 56 #include <linux/errno.h> 57 #include <linux/kernel.h> 58 #include <linux/stddef.h> 59 #include <linux/unistd.h> 60 #include <linux/ptrace.h> 61 #include <linux/user.h> 62 #include <linux/delay.h> 63 64 #include <linux/kallsyms.h> 65 #include <linux/cpufreq.h> 66 #include <linux/dma-mapping.h> 67 #include <linux/ctype.h> 68 #include <linux/uaccess.h> 69 70 #include <linux/percpu.h> 71 #include <linux/crash_dump.h> 72 #include <linux/tboot.h> 73 #include <linux/jiffies.h> 74 #include <linux/mem_encrypt.h> 75 #include <linux/sizes.h> 76 77 #include <linux/usb/xhci-dbgp.h> 78 #include <video/edid.h> 79 80 #include <asm/mtrr.h> 81 #include <asm/apic.h> 82 #include <asm/realmode.h> 83 #include <asm/e820/api.h> 84 #include <asm/mpspec.h> 85 #include <asm/setup.h> 86 #include <asm/efi.h> 87 #include <asm/timer.h> 88 #include <asm/i8259.h> 89 #include <asm/sections.h> 90 #include <asm/io_apic.h> 91 #include <asm/ist.h> 92 #include <asm/setup_arch.h> 93 #include <asm/bios_ebda.h> 94 #include <asm/cacheflush.h> 95 #include <asm/processor.h> 96 #include <asm/bugs.h> 97 #include <asm/kasan.h> 98 99 #include <asm/vsyscall.h> 100 #include <asm/cpu.h> 101 #include <asm/desc.h> 102 #include <asm/dma.h> 103 #include <asm/iommu.h> 104 #include <asm/gart.h> 105 #include <asm/mmu_context.h> 106 #include <asm/proto.h> 107 108 #include <asm/paravirt.h> 109 #include <asm/hypervisor.h> 110 #include <asm/olpc_ofw.h> 111 112 #include <asm/percpu.h> 113 #include <asm/topology.h> 114 #include <asm/apicdef.h> 115 #include <asm/amd_nb.h> 116 #include <asm/mce.h> 117 #include <asm/alternative.h> 118 #include <asm/prom.h> 119 #include <asm/microcode.h> 120 #include <asm/kaslr.h> 121 #include <asm/unwind.h> 122 123 /* 124 * max_low_pfn_mapped: highest direct mapped pfn under 4GB 125 * max_pfn_mapped: highest direct mapped pfn over 4GB 126 * 127 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 128 * represented by pfn_mapped 129 */ 130 unsigned long max_low_pfn_mapped; 131 unsigned long max_pfn_mapped; 132 133 #ifdef CONFIG_DMI 134 RESERVE_BRK(dmi_alloc, 65536); 135 #endif 136 137 138 static __initdata unsigned long _brk_start = (unsigned long)__brk_base; 139 unsigned long _brk_end = (unsigned long)__brk_base; 140 141 struct boot_params boot_params; 142 143 /* 144 * Machine setup.. 145 */ 146 static struct resource data_resource = { 147 .name = "Kernel data", 148 .start = 0, 149 .end = 0, 150 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 151 }; 152 153 static struct resource code_resource = { 154 .name = "Kernel code", 155 .start = 0, 156 .end = 0, 157 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 158 }; 159 160 static struct resource bss_resource = { 161 .name = "Kernel bss", 162 .start = 0, 163 .end = 0, 164 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 165 }; 166 167 168 #ifdef CONFIG_X86_32 169 /* cpu data as detected by the assembly code in head_32.S */ 170 struct cpuinfo_x86 new_cpu_data; 171 172 /* common cpu data for all cpus */ 173 struct cpuinfo_x86 boot_cpu_data __read_mostly; 174 EXPORT_SYMBOL(boot_cpu_data); 175 176 unsigned int def_to_bigsmp; 177 178 /* for MCA, but anyone else can use it if they want */ 179 unsigned int machine_id; 180 unsigned int machine_submodel_id; 181 unsigned int BIOS_revision; 182 183 struct apm_info apm_info; 184 EXPORT_SYMBOL(apm_info); 185 186 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 187 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 188 struct ist_info ist_info; 189 EXPORT_SYMBOL(ist_info); 190 #else 191 struct ist_info ist_info; 192 #endif 193 194 #else 195 struct cpuinfo_x86 boot_cpu_data __read_mostly; 196 EXPORT_SYMBOL(boot_cpu_data); 197 #endif 198 199 200 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 201 __visible unsigned long mmu_cr4_features __ro_after_init; 202 #else 203 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 204 #endif 205 206 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 207 int bootloader_type, bootloader_version; 208 209 /* 210 * Setup options 211 */ 212 struct screen_info screen_info; 213 EXPORT_SYMBOL(screen_info); 214 struct edid_info edid_info; 215 EXPORT_SYMBOL_GPL(edid_info); 216 217 extern int root_mountflags; 218 219 unsigned long saved_video_mode; 220 221 #define RAMDISK_IMAGE_START_MASK 0x07FF 222 #define RAMDISK_PROMPT_FLAG 0x8000 223 #define RAMDISK_LOAD_FLAG 0x4000 224 225 static char __initdata command_line[COMMAND_LINE_SIZE]; 226 #ifdef CONFIG_CMDLINE_BOOL 227 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 228 #endif 229 230 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 231 struct edd edd; 232 #ifdef CONFIG_EDD_MODULE 233 EXPORT_SYMBOL(edd); 234 #endif 235 /** 236 * copy_edd() - Copy the BIOS EDD information 237 * from boot_params into a safe place. 238 * 239 */ 240 static inline void __init copy_edd(void) 241 { 242 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 243 sizeof(edd.mbr_signature)); 244 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 245 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 246 edd.edd_info_nr = boot_params.eddbuf_entries; 247 } 248 #else 249 static inline void __init copy_edd(void) 250 { 251 } 252 #endif 253 254 void * __init extend_brk(size_t size, size_t align) 255 { 256 size_t mask = align - 1; 257 void *ret; 258 259 BUG_ON(_brk_start == 0); 260 BUG_ON(align & mask); 261 262 _brk_end = (_brk_end + mask) & ~mask; 263 BUG_ON((char *)(_brk_end + size) > __brk_limit); 264 265 ret = (void *)_brk_end; 266 _brk_end += size; 267 268 memset(ret, 0, size); 269 270 return ret; 271 } 272 273 #ifdef CONFIG_X86_32 274 static void __init cleanup_highmap(void) 275 { 276 } 277 #endif 278 279 static void __init reserve_brk(void) 280 { 281 if (_brk_end > _brk_start) 282 memblock_reserve(__pa_symbol(_brk_start), 283 _brk_end - _brk_start); 284 285 /* Mark brk area as locked down and no longer taking any 286 new allocations */ 287 _brk_start = 0; 288 } 289 290 u64 relocated_ramdisk; 291 292 #ifdef CONFIG_BLK_DEV_INITRD 293 294 static u64 __init get_ramdisk_image(void) 295 { 296 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 297 298 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 299 300 return ramdisk_image; 301 } 302 static u64 __init get_ramdisk_size(void) 303 { 304 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 305 306 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 307 308 return ramdisk_size; 309 } 310 311 static void __init relocate_initrd(void) 312 { 313 /* Assume only end is not page aligned */ 314 u64 ramdisk_image = get_ramdisk_image(); 315 u64 ramdisk_size = get_ramdisk_size(); 316 u64 area_size = PAGE_ALIGN(ramdisk_size); 317 318 /* We need to move the initrd down into directly mapped mem */ 319 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped), 320 area_size, PAGE_SIZE); 321 322 if (!relocated_ramdisk) 323 panic("Cannot find place for new RAMDISK of size %lld\n", 324 ramdisk_size); 325 326 /* Note: this includes all the mem currently occupied by 327 the initrd, we rely on that fact to keep the data intact. */ 328 memblock_reserve(relocated_ramdisk, area_size); 329 initrd_start = relocated_ramdisk + PAGE_OFFSET; 330 initrd_end = initrd_start + ramdisk_size; 331 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 332 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 333 334 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 335 336 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 337 " [mem %#010llx-%#010llx]\n", 338 ramdisk_image, ramdisk_image + ramdisk_size - 1, 339 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 340 } 341 342 static void __init early_reserve_initrd(void) 343 { 344 /* Assume only end is not page aligned */ 345 u64 ramdisk_image = get_ramdisk_image(); 346 u64 ramdisk_size = get_ramdisk_size(); 347 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 348 349 if (!boot_params.hdr.type_of_loader || 350 !ramdisk_image || !ramdisk_size) 351 return; /* No initrd provided by bootloader */ 352 353 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 354 } 355 static void __init reserve_initrd(void) 356 { 357 /* Assume only end is not page aligned */ 358 u64 ramdisk_image = get_ramdisk_image(); 359 u64 ramdisk_size = get_ramdisk_size(); 360 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 361 u64 mapped_size; 362 363 if (!boot_params.hdr.type_of_loader || 364 !ramdisk_image || !ramdisk_size) 365 return; /* No initrd provided by bootloader */ 366 367 initrd_start = 0; 368 369 mapped_size = memblock_mem_size(max_pfn_mapped); 370 if (ramdisk_size >= (mapped_size>>1)) 371 panic("initrd too large to handle, " 372 "disabling initrd (%lld needed, %lld available)\n", 373 ramdisk_size, mapped_size>>1); 374 375 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 376 ramdisk_end - 1); 377 378 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 379 PFN_DOWN(ramdisk_end))) { 380 /* All are mapped, easy case */ 381 initrd_start = ramdisk_image + PAGE_OFFSET; 382 initrd_end = initrd_start + ramdisk_size; 383 return; 384 } 385 386 relocate_initrd(); 387 388 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image); 389 } 390 391 #else 392 static void __init early_reserve_initrd(void) 393 { 394 } 395 static void __init reserve_initrd(void) 396 { 397 } 398 #endif /* CONFIG_BLK_DEV_INITRD */ 399 400 static void __init parse_setup_data(void) 401 { 402 struct setup_data *data; 403 u64 pa_data, pa_next; 404 405 pa_data = boot_params.hdr.setup_data; 406 while (pa_data) { 407 u32 data_len, data_type; 408 409 data = early_memremap(pa_data, sizeof(*data)); 410 data_len = data->len + sizeof(struct setup_data); 411 data_type = data->type; 412 pa_next = data->next; 413 early_memunmap(data, sizeof(*data)); 414 415 switch (data_type) { 416 case SETUP_E820_EXT: 417 e820__memory_setup_extended(pa_data, data_len); 418 break; 419 case SETUP_DTB: 420 add_dtb(pa_data); 421 break; 422 case SETUP_EFI: 423 parse_efi_setup(pa_data, data_len); 424 break; 425 default: 426 break; 427 } 428 pa_data = pa_next; 429 } 430 } 431 432 static void __init memblock_x86_reserve_range_setup_data(void) 433 { 434 struct setup_data *data; 435 u64 pa_data; 436 437 pa_data = boot_params.hdr.setup_data; 438 while (pa_data) { 439 data = early_memremap(pa_data, sizeof(*data)); 440 memblock_reserve(pa_data, sizeof(*data) + data->len); 441 pa_data = data->next; 442 early_memunmap(data, sizeof(*data)); 443 } 444 } 445 446 /* 447 * --------- Crashkernel reservation ------------------------------ 448 */ 449 450 #ifdef CONFIG_KEXEC_CORE 451 452 /* 16M alignment for crash kernel regions */ 453 #define CRASH_ALIGN SZ_16M 454 455 /* 456 * Keep the crash kernel below this limit. 457 * 458 * On 32 bits earlier kernels would limit the kernel to the low 512 MiB 459 * due to mapping restrictions. 460 * 461 * On 64bit, kdump kernel need be restricted to be under 64TB, which is 462 * the upper limit of system RAM in 4-level paing mode. Since the kdump 463 * jumping could be from 5-level to 4-level, the jumping will fail if 464 * kernel is put above 64TB, and there's no way to detect the paging mode 465 * of the kernel which will be loaded for dumping during the 1st kernel 466 * bootup. 467 */ 468 #ifdef CONFIG_X86_32 469 # define CRASH_ADDR_LOW_MAX SZ_512M 470 # define CRASH_ADDR_HIGH_MAX SZ_512M 471 #else 472 # define CRASH_ADDR_LOW_MAX SZ_4G 473 # define CRASH_ADDR_HIGH_MAX SZ_64T 474 #endif 475 476 static int __init reserve_crashkernel_low(void) 477 { 478 #ifdef CONFIG_X86_64 479 unsigned long long base, low_base = 0, low_size = 0; 480 unsigned long total_low_mem; 481 int ret; 482 483 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT)); 484 485 /* crashkernel=Y,low */ 486 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base); 487 if (ret) { 488 /* 489 * two parts from lib/swiotlb.c: 490 * -swiotlb size: user-specified with swiotlb= or default. 491 * 492 * -swiotlb overflow buffer: now hardcoded to 32k. We round it 493 * to 8M for other buffers that may need to stay low too. Also 494 * make sure we allocate enough extra low memory so that we 495 * don't run out of DMA buffers for 32-bit devices. 496 */ 497 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20); 498 } else { 499 /* passed with crashkernel=0,low ? */ 500 if (!low_size) 501 return 0; 502 } 503 504 low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN); 505 if (!low_base) { 506 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n", 507 (unsigned long)(low_size >> 20)); 508 return -ENOMEM; 509 } 510 511 ret = memblock_reserve(low_base, low_size); 512 if (ret) { 513 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__); 514 return ret; 515 } 516 517 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n", 518 (unsigned long)(low_size >> 20), 519 (unsigned long)(low_base >> 20), 520 (unsigned long)(total_low_mem >> 20)); 521 522 crashk_low_res.start = low_base; 523 crashk_low_res.end = low_base + low_size - 1; 524 insert_resource(&iomem_resource, &crashk_low_res); 525 #endif 526 return 0; 527 } 528 529 static void __init reserve_crashkernel(void) 530 { 531 unsigned long long crash_size, crash_base, total_mem; 532 bool high = false; 533 int ret; 534 535 total_mem = memblock_phys_mem_size(); 536 537 /* crashkernel=XM */ 538 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base); 539 if (ret != 0 || crash_size <= 0) { 540 /* crashkernel=X,high */ 541 ret = parse_crashkernel_high(boot_command_line, total_mem, 542 &crash_size, &crash_base); 543 if (ret != 0 || crash_size <= 0) 544 return; 545 high = true; 546 } 547 548 if (xen_pv_domain()) { 549 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 550 return; 551 } 552 553 /* 0 means: find the address automatically */ 554 if (!crash_base) { 555 /* 556 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory, 557 * crashkernel=x,high reserves memory over 4G, also allocates 558 * 256M extra low memory for DMA buffers and swiotlb. 559 * But the extra memory is not required for all machines. 560 * So try low memory first and fall back to high memory 561 * unless "crashkernel=size[KMG],high" is specified. 562 */ 563 if (!high) 564 crash_base = memblock_find_in_range(CRASH_ALIGN, 565 CRASH_ADDR_LOW_MAX, 566 crash_size, CRASH_ALIGN); 567 if (!crash_base) 568 crash_base = memblock_find_in_range(CRASH_ALIGN, 569 CRASH_ADDR_HIGH_MAX, 570 crash_size, CRASH_ALIGN); 571 if (!crash_base) { 572 pr_info("crashkernel reservation failed - No suitable area found.\n"); 573 return; 574 } 575 } else { 576 unsigned long long start; 577 578 start = memblock_find_in_range(crash_base, 579 crash_base + crash_size, 580 crash_size, 1 << 20); 581 if (start != crash_base) { 582 pr_info("crashkernel reservation failed - memory is in use.\n"); 583 return; 584 } 585 } 586 ret = memblock_reserve(crash_base, crash_size); 587 if (ret) { 588 pr_err("%s: Error reserving crashkernel memblock.\n", __func__); 589 return; 590 } 591 592 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) { 593 memblock_free(crash_base, crash_size); 594 return; 595 } 596 597 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n", 598 (unsigned long)(crash_size >> 20), 599 (unsigned long)(crash_base >> 20), 600 (unsigned long)(total_mem >> 20)); 601 602 crashk_res.start = crash_base; 603 crashk_res.end = crash_base + crash_size - 1; 604 insert_resource(&iomem_resource, &crashk_res); 605 } 606 #else 607 static void __init reserve_crashkernel(void) 608 { 609 } 610 #endif 611 612 static struct resource standard_io_resources[] = { 613 { .name = "dma1", .start = 0x00, .end = 0x1f, 614 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 615 { .name = "pic1", .start = 0x20, .end = 0x21, 616 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 617 { .name = "timer0", .start = 0x40, .end = 0x43, 618 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 619 { .name = "timer1", .start = 0x50, .end = 0x53, 620 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 621 { .name = "keyboard", .start = 0x60, .end = 0x60, 622 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 623 { .name = "keyboard", .start = 0x64, .end = 0x64, 624 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 625 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 626 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 627 { .name = "pic2", .start = 0xa0, .end = 0xa1, 628 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 629 { .name = "dma2", .start = 0xc0, .end = 0xdf, 630 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 631 { .name = "fpu", .start = 0xf0, .end = 0xff, 632 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 633 }; 634 635 void __init reserve_standard_io_resources(void) 636 { 637 int i; 638 639 /* request I/O space for devices used on all i[345]86 PCs */ 640 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 641 request_resource(&ioport_resource, &standard_io_resources[i]); 642 643 } 644 645 static __init void reserve_ibft_region(void) 646 { 647 unsigned long addr, size = 0; 648 649 addr = find_ibft_region(&size); 650 651 if (size) 652 memblock_reserve(addr, size); 653 } 654 655 static bool __init snb_gfx_workaround_needed(void) 656 { 657 #ifdef CONFIG_PCI 658 int i; 659 u16 vendor, devid; 660 static const __initconst u16 snb_ids[] = { 661 0x0102, 662 0x0112, 663 0x0122, 664 0x0106, 665 0x0116, 666 0x0126, 667 0x010a, 668 }; 669 670 /* Assume no if something weird is going on with PCI */ 671 if (!early_pci_allowed()) 672 return false; 673 674 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 675 if (vendor != 0x8086) 676 return false; 677 678 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 679 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 680 if (devid == snb_ids[i]) 681 return true; 682 #endif 683 684 return false; 685 } 686 687 /* 688 * Sandy Bridge graphics has trouble with certain ranges, exclude 689 * them from allocation. 690 */ 691 static void __init trim_snb_memory(void) 692 { 693 static const __initconst unsigned long bad_pages[] = { 694 0x20050000, 695 0x20110000, 696 0x20130000, 697 0x20138000, 698 0x40004000, 699 }; 700 int i; 701 702 if (!snb_gfx_workaround_needed()) 703 return; 704 705 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 706 707 /* 708 * Reserve all memory below the 1 MB mark that has not 709 * already been reserved. 710 */ 711 memblock_reserve(0, 1<<20); 712 713 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 714 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 715 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 716 bad_pages[i]); 717 } 718 } 719 720 /* 721 * Here we put platform-specific memory range workarounds, i.e. 722 * memory known to be corrupt or otherwise in need to be reserved on 723 * specific platforms. 724 * 725 * If this gets used more widely it could use a real dispatch mechanism. 726 */ 727 static void __init trim_platform_memory_ranges(void) 728 { 729 trim_snb_memory(); 730 } 731 732 static void __init trim_bios_range(void) 733 { 734 /* 735 * A special case is the first 4Kb of memory; 736 * This is a BIOS owned area, not kernel ram, but generally 737 * not listed as such in the E820 table. 738 * 739 * This typically reserves additional memory (64KiB by default) 740 * since some BIOSes are known to corrupt low memory. See the 741 * Kconfig help text for X86_RESERVE_LOW. 742 */ 743 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 744 745 /* 746 * special case: Some BIOSen report the PC BIOS 747 * area (640->1Mb) as ram even though it is not. 748 * take them out. 749 */ 750 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 751 752 e820__update_table(e820_table); 753 } 754 755 /* called before trim_bios_range() to spare extra sanitize */ 756 static void __init e820_add_kernel_range(void) 757 { 758 u64 start = __pa_symbol(_text); 759 u64 size = __pa_symbol(_end) - start; 760 761 /* 762 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 763 * attempt to fix it by adding the range. We may have a confused BIOS, 764 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 765 * exclude kernel range. If we really are running on top non-RAM, 766 * we will crash later anyways. 767 */ 768 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 769 return; 770 771 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 772 e820__range_remove(start, size, E820_TYPE_RAM, 0); 773 e820__range_add(start, size, E820_TYPE_RAM); 774 } 775 776 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10; 777 778 static int __init parse_reservelow(char *p) 779 { 780 unsigned long long size; 781 782 if (!p) 783 return -EINVAL; 784 785 size = memparse(p, &p); 786 787 if (size < 4096) 788 size = 4096; 789 790 if (size > 640*1024) 791 size = 640*1024; 792 793 reserve_low = size; 794 795 return 0; 796 } 797 798 early_param("reservelow", parse_reservelow); 799 800 static void __init trim_low_memory_range(void) 801 { 802 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE)); 803 } 804 805 /* 806 * Dump out kernel offset information on panic. 807 */ 808 static int 809 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 810 { 811 if (kaslr_enabled()) { 812 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 813 kaslr_offset(), 814 __START_KERNEL, 815 __START_KERNEL_map, 816 MODULES_VADDR-1); 817 } else { 818 pr_emerg("Kernel Offset: disabled\n"); 819 } 820 821 return 0; 822 } 823 824 /* 825 * Determine if we were loaded by an EFI loader. If so, then we have also been 826 * passed the efi memmap, systab, etc., so we should use these data structures 827 * for initialization. Note, the efi init code path is determined by the 828 * global efi_enabled. This allows the same kernel image to be used on existing 829 * systems (with a traditional BIOS) as well as on EFI systems. 830 */ 831 /* 832 * setup_arch - architecture-specific boot-time initializations 833 * 834 * Note: On x86_64, fixmaps are ready for use even before this is called. 835 */ 836 837 void __init setup_arch(char **cmdline_p) 838 { 839 /* 840 * Reserve the memory occupied by the kernel between _text and 841 * __end_of_kernel_reserve symbols. Any kernel sections after the 842 * __end_of_kernel_reserve symbol must be explicitly reserved with a 843 * separate memblock_reserve() or they will be discarded. 844 */ 845 memblock_reserve(__pa_symbol(_text), 846 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 847 848 /* 849 * Make sure page 0 is always reserved because on systems with 850 * L1TF its contents can be leaked to user processes. 851 */ 852 memblock_reserve(0, PAGE_SIZE); 853 854 early_reserve_initrd(); 855 856 /* 857 * At this point everything still needed from the boot loader 858 * or BIOS or kernel text should be early reserved or marked not 859 * RAM in e820. All other memory is free game. 860 */ 861 862 #ifdef CONFIG_X86_32 863 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 864 865 /* 866 * copy kernel address range established so far and switch 867 * to the proper swapper page table 868 */ 869 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 870 initial_page_table + KERNEL_PGD_BOUNDARY, 871 KERNEL_PGD_PTRS); 872 873 load_cr3(swapper_pg_dir); 874 /* 875 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 876 * a cr3 based tlb flush, so the following __flush_tlb_all() 877 * will not flush anything because the cpu quirk which clears 878 * X86_FEATURE_PGE has not been invoked yet. Though due to the 879 * load_cr3() above the TLB has been flushed already. The 880 * quirk is invoked before subsequent calls to __flush_tlb_all() 881 * so proper operation is guaranteed. 882 */ 883 __flush_tlb_all(); 884 #else 885 printk(KERN_INFO "Command line: %s\n", boot_command_line); 886 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 887 #endif 888 889 /* 890 * If we have OLPC OFW, we might end up relocating the fixmap due to 891 * reserve_top(), so do this before touching the ioremap area. 892 */ 893 olpc_ofw_detect(); 894 895 idt_setup_early_traps(); 896 early_cpu_init(); 897 arch_init_ideal_nops(); 898 jump_label_init(); 899 early_ioremap_init(); 900 901 setup_olpc_ofw_pgd(); 902 903 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 904 screen_info = boot_params.screen_info; 905 edid_info = boot_params.edid_info; 906 #ifdef CONFIG_X86_32 907 apm_info.bios = boot_params.apm_bios_info; 908 ist_info = boot_params.ist_info; 909 #endif 910 saved_video_mode = boot_params.hdr.vid_mode; 911 bootloader_type = boot_params.hdr.type_of_loader; 912 if ((bootloader_type >> 4) == 0xe) { 913 bootloader_type &= 0xf; 914 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 915 } 916 bootloader_version = bootloader_type & 0xf; 917 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 918 919 #ifdef CONFIG_BLK_DEV_RAM 920 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 921 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); 922 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); 923 #endif 924 #ifdef CONFIG_EFI 925 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 926 EFI32_LOADER_SIGNATURE, 4)) { 927 set_bit(EFI_BOOT, &efi.flags); 928 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 929 EFI64_LOADER_SIGNATURE, 4)) { 930 set_bit(EFI_BOOT, &efi.flags); 931 set_bit(EFI_64BIT, &efi.flags); 932 } 933 #endif 934 935 x86_init.oem.arch_setup(); 936 937 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 938 e820__memory_setup(); 939 parse_setup_data(); 940 941 copy_edd(); 942 943 if (!boot_params.hdr.root_flags) 944 root_mountflags &= ~MS_RDONLY; 945 init_mm.start_code = (unsigned long) _text; 946 init_mm.end_code = (unsigned long) _etext; 947 init_mm.end_data = (unsigned long) _edata; 948 init_mm.brk = _brk_end; 949 950 mpx_mm_init(&init_mm); 951 952 code_resource.start = __pa_symbol(_text); 953 code_resource.end = __pa_symbol(_etext)-1; 954 data_resource.start = __pa_symbol(_etext); 955 data_resource.end = __pa_symbol(_edata)-1; 956 bss_resource.start = __pa_symbol(__bss_start); 957 bss_resource.end = __pa_symbol(__bss_stop)-1; 958 959 #ifdef CONFIG_CMDLINE_BOOL 960 #ifdef CONFIG_CMDLINE_OVERRIDE 961 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 962 #else 963 if (builtin_cmdline[0]) { 964 /* append boot loader cmdline to builtin */ 965 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 966 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 967 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 968 } 969 #endif 970 #endif 971 972 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 973 *cmdline_p = command_line; 974 975 /* 976 * x86_configure_nx() is called before parse_early_param() to detect 977 * whether hardware doesn't support NX (so that the early EHCI debug 978 * console setup can safely call set_fixmap()). It may then be called 979 * again from within noexec_setup() during parsing early parameters 980 * to honor the respective command line option. 981 */ 982 x86_configure_nx(); 983 984 parse_early_param(); 985 986 if (efi_enabled(EFI_BOOT)) 987 efi_memblock_x86_reserve_range(); 988 #ifdef CONFIG_MEMORY_HOTPLUG 989 /* 990 * Memory used by the kernel cannot be hot-removed because Linux 991 * cannot migrate the kernel pages. When memory hotplug is 992 * enabled, we should prevent memblock from allocating memory 993 * for the kernel. 994 * 995 * ACPI SRAT records all hotpluggable memory ranges. But before 996 * SRAT is parsed, we don't know about it. 997 * 998 * The kernel image is loaded into memory at very early time. We 999 * cannot prevent this anyway. So on NUMA system, we set any 1000 * node the kernel resides in as un-hotpluggable. 1001 * 1002 * Since on modern servers, one node could have double-digit 1003 * gigabytes memory, we can assume the memory around the kernel 1004 * image is also un-hotpluggable. So before SRAT is parsed, just 1005 * allocate memory near the kernel image to try the best to keep 1006 * the kernel away from hotpluggable memory. 1007 */ 1008 if (movable_node_is_enabled()) 1009 memblock_set_bottom_up(true); 1010 #endif 1011 1012 x86_report_nx(); 1013 1014 /* after early param, so could get panic from serial */ 1015 memblock_x86_reserve_range_setup_data(); 1016 1017 if (acpi_mps_check()) { 1018 #ifdef CONFIG_X86_LOCAL_APIC 1019 disable_apic = 1; 1020 #endif 1021 setup_clear_cpu_cap(X86_FEATURE_APIC); 1022 } 1023 1024 e820__reserve_setup_data(); 1025 e820__finish_early_params(); 1026 1027 if (efi_enabled(EFI_BOOT)) 1028 efi_init(); 1029 1030 dmi_setup(); 1031 1032 /* 1033 * VMware detection requires dmi to be available, so this 1034 * needs to be done after dmi_setup(), for the boot CPU. 1035 */ 1036 init_hypervisor_platform(); 1037 1038 tsc_early_init(); 1039 x86_init.resources.probe_roms(); 1040 1041 /* after parse_early_param, so could debug it */ 1042 insert_resource(&iomem_resource, &code_resource); 1043 insert_resource(&iomem_resource, &data_resource); 1044 insert_resource(&iomem_resource, &bss_resource); 1045 1046 e820_add_kernel_range(); 1047 trim_bios_range(); 1048 #ifdef CONFIG_X86_32 1049 if (ppro_with_ram_bug()) { 1050 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1051 E820_TYPE_RESERVED); 1052 e820__update_table(e820_table); 1053 printk(KERN_INFO "fixed physical RAM map:\n"); 1054 e820__print_table("bad_ppro"); 1055 } 1056 #else 1057 early_gart_iommu_check(); 1058 #endif 1059 1060 /* 1061 * partially used pages are not usable - thus 1062 * we are rounding upwards: 1063 */ 1064 max_pfn = e820__end_of_ram_pfn(); 1065 1066 /* update e820 for memory not covered by WB MTRRs */ 1067 mtrr_bp_init(); 1068 if (mtrr_trim_uncached_memory(max_pfn)) 1069 max_pfn = e820__end_of_ram_pfn(); 1070 1071 max_possible_pfn = max_pfn; 1072 1073 /* 1074 * This call is required when the CPU does not support PAT. If 1075 * mtrr_bp_init() invoked it already via pat_init() the call has no 1076 * effect. 1077 */ 1078 init_cache_modes(); 1079 1080 /* 1081 * Define random base addresses for memory sections after max_pfn is 1082 * defined and before each memory section base is used. 1083 */ 1084 kernel_randomize_memory(); 1085 1086 #ifdef CONFIG_X86_32 1087 /* max_low_pfn get updated here */ 1088 find_low_pfn_range(); 1089 #else 1090 check_x2apic(); 1091 1092 /* How many end-of-memory variables you have, grandma! */ 1093 /* need this before calling reserve_initrd */ 1094 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1095 max_low_pfn = e820__end_of_low_ram_pfn(); 1096 else 1097 max_low_pfn = max_pfn; 1098 1099 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1100 #endif 1101 1102 /* 1103 * Find and reserve possible boot-time SMP configuration: 1104 */ 1105 find_smp_config(); 1106 1107 reserve_ibft_region(); 1108 1109 early_alloc_pgt_buf(); 1110 1111 /* 1112 * Need to conclude brk, before e820__memblock_setup() 1113 * it could use memblock_find_in_range, could overlap with 1114 * brk area. 1115 */ 1116 reserve_brk(); 1117 1118 cleanup_highmap(); 1119 1120 memblock_set_current_limit(ISA_END_ADDRESS); 1121 e820__memblock_setup(); 1122 1123 reserve_bios_regions(); 1124 1125 if (efi_enabled(EFI_MEMMAP)) { 1126 efi_fake_memmap(); 1127 efi_find_mirror(); 1128 efi_esrt_init(); 1129 1130 /* 1131 * The EFI specification says that boot service code won't be 1132 * called after ExitBootServices(). This is, in fact, a lie. 1133 */ 1134 efi_reserve_boot_services(); 1135 } 1136 1137 /* preallocate 4k for mptable mpc */ 1138 e820__memblock_alloc_reserved_mpc_new(); 1139 1140 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1141 setup_bios_corruption_check(); 1142 #endif 1143 1144 #ifdef CONFIG_X86_32 1145 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1146 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1147 #endif 1148 1149 reserve_real_mode(); 1150 1151 trim_platform_memory_ranges(); 1152 trim_low_memory_range(); 1153 1154 init_mem_mapping(); 1155 1156 idt_setup_early_pf(); 1157 1158 /* 1159 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1160 * with the current CR4 value. This may not be necessary, but 1161 * auditing all the early-boot CR4 manipulation would be needed to 1162 * rule it out. 1163 * 1164 * Mask off features that don't work outside long mode (just 1165 * PCIDE for now). 1166 */ 1167 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1168 1169 memblock_set_current_limit(get_max_mapped()); 1170 1171 /* 1172 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1173 */ 1174 1175 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1176 if (init_ohci1394_dma_early) 1177 init_ohci1394_dma_on_all_controllers(); 1178 #endif 1179 /* Allocate bigger log buffer */ 1180 setup_log_buf(1); 1181 1182 if (efi_enabled(EFI_BOOT)) { 1183 switch (boot_params.secure_boot) { 1184 case efi_secureboot_mode_disabled: 1185 pr_info("Secure boot disabled\n"); 1186 break; 1187 case efi_secureboot_mode_enabled: 1188 pr_info("Secure boot enabled\n"); 1189 break; 1190 default: 1191 pr_info("Secure boot could not be determined\n"); 1192 break; 1193 } 1194 } 1195 1196 reserve_initrd(); 1197 1198 acpi_table_upgrade(); 1199 1200 vsmp_init(); 1201 1202 io_delay_init(); 1203 1204 early_platform_quirks(); 1205 1206 /* 1207 * Parse the ACPI tables for possible boot-time SMP configuration. 1208 */ 1209 acpi_boot_table_init(); 1210 1211 early_acpi_boot_init(); 1212 1213 initmem_init(); 1214 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1215 1216 /* 1217 * Reserve memory for crash kernel after SRAT is parsed so that it 1218 * won't consume hotpluggable memory. 1219 */ 1220 reserve_crashkernel(); 1221 1222 memblock_find_dma_reserve(); 1223 1224 if (!early_xdbc_setup_hardware()) 1225 early_xdbc_register_console(); 1226 1227 x86_init.paging.pagetable_init(); 1228 1229 kasan_init(); 1230 1231 /* 1232 * Sync back kernel address range. 1233 * 1234 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1235 * this call? 1236 */ 1237 sync_initial_page_table(); 1238 1239 tboot_probe(); 1240 1241 map_vsyscall(); 1242 1243 generic_apic_probe(); 1244 1245 early_quirks(); 1246 1247 /* 1248 * Read APIC and some other early information from ACPI tables. 1249 */ 1250 acpi_boot_init(); 1251 sfi_init(); 1252 x86_dtb_init(); 1253 1254 /* 1255 * get boot-time SMP configuration: 1256 */ 1257 get_smp_config(); 1258 1259 /* 1260 * Systems w/o ACPI and mptables might not have it mapped the local 1261 * APIC yet, but prefill_possible_map() might need to access it. 1262 */ 1263 init_apic_mappings(); 1264 1265 prefill_possible_map(); 1266 1267 init_cpu_to_node(); 1268 1269 io_apic_init_mappings(); 1270 1271 x86_init.hyper.guest_late_init(); 1272 1273 e820__reserve_resources(); 1274 e820__register_nosave_regions(max_pfn); 1275 1276 x86_init.resources.reserve_resources(); 1277 1278 e820__setup_pci_gap(); 1279 1280 #ifdef CONFIG_VT 1281 #if defined(CONFIG_VGA_CONSOLE) 1282 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1283 conswitchp = &vga_con; 1284 #elif defined(CONFIG_DUMMY_CONSOLE) 1285 conswitchp = &dummy_con; 1286 #endif 1287 #endif 1288 x86_init.oem.banner(); 1289 1290 x86_init.timers.wallclock_init(); 1291 1292 mcheck_init(); 1293 1294 register_refined_jiffies(CLOCK_TICK_RATE); 1295 1296 #ifdef CONFIG_EFI 1297 if (efi_enabled(EFI_BOOT)) 1298 efi_apply_memmap_quirks(); 1299 #endif 1300 1301 unwind_init(); 1302 } 1303 1304 #ifdef CONFIG_X86_32 1305 1306 static struct resource video_ram_resource = { 1307 .name = "Video RAM area", 1308 .start = 0xa0000, 1309 .end = 0xbffff, 1310 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1311 }; 1312 1313 void __init i386_reserve_resources(void) 1314 { 1315 request_resource(&iomem_resource, &video_ram_resource); 1316 reserve_standard_io_resources(); 1317 } 1318 1319 #endif /* CONFIG_X86_32 */ 1320 1321 static struct notifier_block kernel_offset_notifier = { 1322 .notifier_call = dump_kernel_offset 1323 }; 1324 1325 static int __init register_kernel_offset_dumper(void) 1326 { 1327 atomic_notifier_chain_register(&panic_notifier_list, 1328 &kernel_offset_notifier); 1329 return 0; 1330 } 1331 __initcall(register_kernel_offset_dumper); 1332