| 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/apm_bios.h> |
| 31 | #include <linux/initrd.h> |
| 32 | #include <linux/bootmem.h> |
| 33 | #include <linux/seq_file.h> |
| 34 | #include <linux/console.h> |
| 35 | #include <linux/mca.h> |
| 36 | #include <linux/root_dev.h> |
| 37 | #include <linux/highmem.h> |
| 38 | #include <linux/module.h> |
| 39 | #include <linux/efi.h> |
| 40 | #include <linux/init.h> |
| 41 | #include <linux/edd.h> |
| 42 | #include <linux/nodemask.h> |
| 43 | #include <linux/kexec.h> |
| 44 | #include <linux/crash_dump.h> |
| 45 | #include <linux/dmi.h> |
| 46 | #include <linux/pfn.h> |
| 47 | #include <linux/pci.h> |
| 48 | #include <linux/init_ohci1394_dma.h> |
| 49 | |
| 50 | #include <video/edid.h> |
| 51 | |
| 52 | #include <asm/mtrr.h> |
| 53 | #include <asm/apic.h> |
| 54 | #include <asm/e820.h> |
| 55 | #include <asm/mpspec.h> |
| 56 | #include <asm/mmzone.h> |
| 57 | #include <asm/setup.h> |
| 58 | #include <asm/arch_hooks.h> |
| 59 | #include <asm/sections.h> |
| 60 | #include <asm/io_apic.h> |
| 61 | #include <asm/ist.h> |
| 62 | #include <asm/io.h> |
| 63 | #include <asm/vmi.h> |
| 64 | #include <setup_arch.h> |
| 65 | #include <asm/bios_ebda.h> |
| 66 | #include <asm/cacheflush.h> |
| 67 | #include <asm/processor.h> |
| 68 | |
| 69 | /* This value is set up by the early boot code to point to the value |
| 70 | immediately after the boot time page tables. It contains a *physical* |
| 71 | address, and must not be in the .bss segment! */ |
| 72 | unsigned long init_pg_tables_end __initdata = ~0UL; |
| 73 | |
| 74 | /* |
| 75 | * Machine setup.. |
| 76 | */ |
| 77 | static struct resource data_resource = { |
| 78 | .name = "Kernel data", |
| 79 | .start = 0, |
| 80 | .end = 0, |
| 81 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM |
| 82 | }; |
| 83 | |
| 84 | static struct resource code_resource = { |
| 85 | .name = "Kernel code", |
| 86 | .start = 0, |
| 87 | .end = 0, |
| 88 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM |
| 89 | }; |
| 90 | |
| 91 | static struct resource bss_resource = { |
| 92 | .name = "Kernel bss", |
| 93 | .start = 0, |
| 94 | .end = 0, |
| 95 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM |
| 96 | }; |
| 97 | |
| 98 | static struct resource video_ram_resource = { |
| 99 | .name = "Video RAM area", |
| 100 | .start = 0xa0000, |
| 101 | .end = 0xbffff, |
| 102 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM |
| 103 | }; |
| 104 | |
| 105 | static struct resource standard_io_resources[] = { { |
| 106 | .name = "dma1", |
| 107 | .start = 0x0000, |
| 108 | .end = 0x001f, |
| 109 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 110 | }, { |
| 111 | .name = "pic1", |
| 112 | .start = 0x0020, |
| 113 | .end = 0x0021, |
| 114 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 115 | }, { |
| 116 | .name = "timer0", |
| 117 | .start = 0x0040, |
| 118 | .end = 0x0043, |
| 119 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 120 | }, { |
| 121 | .name = "timer1", |
| 122 | .start = 0x0050, |
| 123 | .end = 0x0053, |
| 124 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 125 | }, { |
| 126 | .name = "keyboard", |
| 127 | .start = 0x0060, |
| 128 | .end = 0x006f, |
| 129 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 130 | }, { |
| 131 | .name = "dma page reg", |
| 132 | .start = 0x0080, |
| 133 | .end = 0x008f, |
| 134 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 135 | }, { |
| 136 | .name = "pic2", |
| 137 | .start = 0x00a0, |
| 138 | .end = 0x00a1, |
| 139 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 140 | }, { |
| 141 | .name = "dma2", |
| 142 | .start = 0x00c0, |
| 143 | .end = 0x00df, |
| 144 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 145 | }, { |
| 146 | .name = "fpu", |
| 147 | .start = 0x00f0, |
| 148 | .end = 0x00ff, |
| 149 | .flags = IORESOURCE_BUSY | IORESOURCE_IO |
| 150 | } }; |
| 151 | |
| 152 | /* cpu data as detected by the assembly code in head.S */ |
| 153 | struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 }; |
| 154 | /* common cpu data for all cpus */ |
| 155 | struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 }; |
| 156 | EXPORT_SYMBOL(boot_cpu_data); |
| 157 | |
| 158 | unsigned int def_to_bigsmp; |
| 159 | |
| 160 | #ifndef CONFIG_X86_PAE |
| 161 | unsigned long mmu_cr4_features; |
| 162 | #else |
| 163 | unsigned long mmu_cr4_features = X86_CR4_PAE; |
| 164 | #endif |
| 165 | |
| 166 | /* for MCA, but anyone else can use it if they want */ |
| 167 | unsigned int machine_id; |
| 168 | unsigned int machine_submodel_id; |
| 169 | unsigned int BIOS_revision; |
| 170 | |
| 171 | /* Boot loader ID as an integer, for the benefit of proc_dointvec */ |
| 172 | int bootloader_type; |
| 173 | |
| 174 | /* user-defined highmem size */ |
| 175 | static unsigned int highmem_pages = -1; |
| 176 | |
| 177 | /* |
| 178 | * Setup options |
| 179 | */ |
| 180 | struct screen_info screen_info; |
| 181 | EXPORT_SYMBOL(screen_info); |
| 182 | struct apm_info apm_info; |
| 183 | EXPORT_SYMBOL(apm_info); |
| 184 | struct edid_info edid_info; |
| 185 | EXPORT_SYMBOL_GPL(edid_info); |
| 186 | struct ist_info ist_info; |
| 187 | #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ |
| 188 | defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) |
| 189 | EXPORT_SYMBOL(ist_info); |
| 190 | #endif |
| 191 | |
| 192 | extern void early_cpu_init(void); |
| 193 | extern int root_mountflags; |
| 194 | |
| 195 | unsigned long saved_videomode; |
| 196 | |
| 197 | #define RAMDISK_IMAGE_START_MASK 0x07FF |
| 198 | #define RAMDISK_PROMPT_FLAG 0x8000 |
| 199 | #define RAMDISK_LOAD_FLAG 0x4000 |
| 200 | |
| 201 | static char __initdata command_line[COMMAND_LINE_SIZE]; |
| 202 | |
| 203 | #ifndef CONFIG_DEBUG_BOOT_PARAMS |
| 204 | struct boot_params __initdata boot_params; |
| 205 | #else |
| 206 | struct boot_params boot_params; |
| 207 | #endif |
| 208 | |
| 209 | #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) |
| 210 | struct edd edd; |
| 211 | #ifdef CONFIG_EDD_MODULE |
| 212 | EXPORT_SYMBOL(edd); |
| 213 | #endif |
| 214 | /** |
| 215 | * copy_edd() - Copy the BIOS EDD information |
| 216 | * from boot_params into a safe place. |
| 217 | * |
| 218 | */ |
| 219 | static inline void copy_edd(void) |
| 220 | { |
| 221 | memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, |
| 222 | sizeof(edd.mbr_signature)); |
| 223 | memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); |
| 224 | edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; |
| 225 | edd.edd_info_nr = boot_params.eddbuf_entries; |
| 226 | } |
| 227 | #else |
| 228 | static inline void copy_edd(void) |
| 229 | { |
| 230 | } |
| 231 | #endif |
| 232 | |
| 233 | int __initdata user_defined_memmap; |
| 234 | |
| 235 | /* |
| 236 | * "mem=nopentium" disables the 4MB page tables. |
| 237 | * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM |
| 238 | * to <mem>, overriding the bios size. |
| 239 | * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from |
| 240 | * <start> to <start>+<mem>, overriding the bios size. |
| 241 | * |
| 242 | * HPA tells me bootloaders need to parse mem=, so no new |
| 243 | * option should be mem= [also see Documentation/i386/boot.txt] |
| 244 | */ |
| 245 | static int __init parse_mem(char *arg) |
| 246 | { |
| 247 | if (!arg) |
| 248 | return -EINVAL; |
| 249 | |
| 250 | if (strcmp(arg, "nopentium") == 0) { |
| 251 | setup_clear_cpu_cap(X86_FEATURE_PSE); |
| 252 | } else { |
| 253 | /* If the user specifies memory size, we |
| 254 | * limit the BIOS-provided memory map to |
| 255 | * that size. exactmap can be used to specify |
| 256 | * the exact map. mem=number can be used to |
| 257 | * trim the existing memory map. |
| 258 | */ |
| 259 | unsigned long long mem_size; |
| 260 | |
| 261 | mem_size = memparse(arg, &arg); |
| 262 | limit_regions(mem_size); |
| 263 | user_defined_memmap = 1; |
| 264 | } |
| 265 | return 0; |
| 266 | } |
| 267 | early_param("mem", parse_mem); |
| 268 | |
| 269 | #ifdef CONFIG_PROC_VMCORE |
| 270 | /* elfcorehdr= specifies the location of elf core header |
| 271 | * stored by the crashed kernel. |
| 272 | */ |
| 273 | static int __init parse_elfcorehdr(char *arg) |
| 274 | { |
| 275 | if (!arg) |
| 276 | return -EINVAL; |
| 277 | |
| 278 | elfcorehdr_addr = memparse(arg, &arg); |
| 279 | return 0; |
| 280 | } |
| 281 | early_param("elfcorehdr", parse_elfcorehdr); |
| 282 | #endif /* CONFIG_PROC_VMCORE */ |
| 283 | |
| 284 | /* |
| 285 | * highmem=size forces highmem to be exactly 'size' bytes. |
| 286 | * This works even on boxes that have no highmem otherwise. |
| 287 | * This also works to reduce highmem size on bigger boxes. |
| 288 | */ |
| 289 | static int __init parse_highmem(char *arg) |
| 290 | { |
| 291 | if (!arg) |
| 292 | return -EINVAL; |
| 293 | |
| 294 | highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; |
| 295 | return 0; |
| 296 | } |
| 297 | early_param("highmem", parse_highmem); |
| 298 | |
| 299 | /* |
| 300 | * vmalloc=size forces the vmalloc area to be exactly 'size' |
| 301 | * bytes. This can be used to increase (or decrease) the |
| 302 | * vmalloc area - the default is 128m. |
| 303 | */ |
| 304 | static int __init parse_vmalloc(char *arg) |
| 305 | { |
| 306 | if (!arg) |
| 307 | return -EINVAL; |
| 308 | |
| 309 | __VMALLOC_RESERVE = memparse(arg, &arg); |
| 310 | return 0; |
| 311 | } |
| 312 | early_param("vmalloc", parse_vmalloc); |
| 313 | |
| 314 | /* |
| 315 | * reservetop=size reserves a hole at the top of the kernel address space which |
| 316 | * a hypervisor can load into later. Needed for dynamically loaded hypervisors, |
| 317 | * so relocating the fixmap can be done before paging initialization. |
| 318 | */ |
| 319 | static int __init parse_reservetop(char *arg) |
| 320 | { |
| 321 | unsigned long address; |
| 322 | |
| 323 | if (!arg) |
| 324 | return -EINVAL; |
| 325 | |
| 326 | address = memparse(arg, &arg); |
| 327 | reserve_top_address(address); |
| 328 | return 0; |
| 329 | } |
| 330 | early_param("reservetop", parse_reservetop); |
| 331 | |
| 332 | /* |
| 333 | * Determine low and high memory ranges: |
| 334 | */ |
| 335 | unsigned long __init find_max_low_pfn(void) |
| 336 | { |
| 337 | unsigned long max_low_pfn; |
| 338 | |
| 339 | max_low_pfn = max_pfn; |
| 340 | if (max_low_pfn > MAXMEM_PFN) { |
| 341 | if (highmem_pages == -1) |
| 342 | highmem_pages = max_pfn - MAXMEM_PFN; |
| 343 | if (highmem_pages + MAXMEM_PFN < max_pfn) |
| 344 | max_pfn = MAXMEM_PFN + highmem_pages; |
| 345 | if (highmem_pages + MAXMEM_PFN > max_pfn) { |
| 346 | printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages)); |
| 347 | highmem_pages = 0; |
| 348 | } |
| 349 | max_low_pfn = MAXMEM_PFN; |
| 350 | #ifndef CONFIG_HIGHMEM |
| 351 | /* Maximum memory usable is what is directly addressable */ |
| 352 | printk(KERN_WARNING "Warning only %ldMB will be used.\n", |
| 353 | MAXMEM>>20); |
| 354 | if (max_pfn > MAX_NONPAE_PFN) |
| 355 | printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n"); |
| 356 | else |
| 357 | printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); |
| 358 | max_pfn = MAXMEM_PFN; |
| 359 | #else /* !CONFIG_HIGHMEM */ |
| 360 | #ifndef CONFIG_HIGHMEM64G |
| 361 | if (max_pfn > MAX_NONPAE_PFN) { |
| 362 | max_pfn = MAX_NONPAE_PFN; |
| 363 | printk(KERN_WARNING "Warning only 4GB will be used.\n"); |
| 364 | printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n"); |
| 365 | } |
| 366 | #endif /* !CONFIG_HIGHMEM64G */ |
| 367 | #endif /* !CONFIG_HIGHMEM */ |
| 368 | } else { |
| 369 | if (highmem_pages == -1) |
| 370 | highmem_pages = 0; |
| 371 | #ifdef CONFIG_HIGHMEM |
| 372 | if (highmem_pages >= max_pfn) { |
| 373 | printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); |
| 374 | highmem_pages = 0; |
| 375 | } |
| 376 | if (highmem_pages) { |
| 377 | if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){ |
| 378 | printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages)); |
| 379 | highmem_pages = 0; |
| 380 | } |
| 381 | max_low_pfn -= highmem_pages; |
| 382 | } |
| 383 | #else |
| 384 | if (highmem_pages) |
| 385 | printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); |
| 386 | #endif |
| 387 | } |
| 388 | return max_low_pfn; |
| 389 | } |
| 390 | |
| 391 | #define BIOS_EBDA_SEGMENT 0x40E |
| 392 | #define BIOS_LOWMEM_KILOBYTES 0x413 |
| 393 | |
| 394 | /* |
| 395 | * The BIOS places the EBDA/XBDA at the top of conventional |
| 396 | * memory, and usually decreases the reported amount of |
| 397 | * conventional memory (int 0x12) too. This also contains a |
| 398 | * workaround for Dell systems that neglect to reserve EBDA. |
| 399 | * The same workaround also avoids a problem with the AMD768MPX |
| 400 | * chipset: reserve a page before VGA to prevent PCI prefetch |
| 401 | * into it (errata #56). Usually the page is reserved anyways, |
| 402 | * unless you have no PS/2 mouse plugged in. |
| 403 | */ |
| 404 | static void __init reserve_ebda_region(void) |
| 405 | { |
| 406 | unsigned int lowmem, ebda_addr; |
| 407 | |
| 408 | /* To determine the position of the EBDA and the */ |
| 409 | /* end of conventional memory, we need to look at */ |
| 410 | /* the BIOS data area. In a paravirtual environment */ |
| 411 | /* that area is absent. We'll just have to assume */ |
| 412 | /* that the paravirt case can handle memory setup */ |
| 413 | /* correctly, without our help. */ |
| 414 | if (paravirt_enabled()) |
| 415 | return; |
| 416 | |
| 417 | /* end of low (conventional) memory */ |
| 418 | lowmem = *(unsigned short *)__va(BIOS_LOWMEM_KILOBYTES); |
| 419 | lowmem <<= 10; |
| 420 | |
| 421 | /* start of EBDA area */ |
| 422 | ebda_addr = *(unsigned short *)__va(BIOS_EBDA_SEGMENT); |
| 423 | ebda_addr <<= 4; |
| 424 | |
| 425 | /* Fixup: bios puts an EBDA in the top 64K segment */ |
| 426 | /* of conventional memory, but does not adjust lowmem. */ |
| 427 | if ((lowmem - ebda_addr) <= 0x10000) |
| 428 | lowmem = ebda_addr; |
| 429 | |
| 430 | /* Fixup: bios does not report an EBDA at all. */ |
| 431 | /* Some old Dells seem to need 4k anyhow (bugzilla 2990) */ |
| 432 | if ((ebda_addr == 0) && (lowmem >= 0x9f000)) |
| 433 | lowmem = 0x9f000; |
| 434 | |
| 435 | /* Paranoia: should never happen, but... */ |
| 436 | if ((lowmem == 0) || (lowmem >= 0x100000)) |
| 437 | lowmem = 0x9f000; |
| 438 | |
| 439 | /* reserve all memory between lowmem and the 1MB mark */ |
| 440 | reserve_bootmem(lowmem, 0x100000 - lowmem, BOOTMEM_DEFAULT); |
| 441 | } |
| 442 | |
| 443 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
| 444 | void __init setup_bootmem_allocator(void); |
| 445 | static unsigned long __init setup_memory(void) |
| 446 | { |
| 447 | /* |
| 448 | * partially used pages are not usable - thus |
| 449 | * we are rounding upwards: |
| 450 | */ |
| 451 | min_low_pfn = PFN_UP(init_pg_tables_end); |
| 452 | |
| 453 | max_low_pfn = find_max_low_pfn(); |
| 454 | |
| 455 | #ifdef CONFIG_HIGHMEM |
| 456 | highstart_pfn = highend_pfn = max_pfn; |
| 457 | if (max_pfn > max_low_pfn) { |
| 458 | highstart_pfn = max_low_pfn; |
| 459 | } |
| 460 | printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", |
| 461 | pages_to_mb(highend_pfn - highstart_pfn)); |
| 462 | num_physpages = highend_pfn; |
| 463 | high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; |
| 464 | #else |
| 465 | num_physpages = max_low_pfn; |
| 466 | high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; |
| 467 | #endif |
| 468 | #ifdef CONFIG_FLATMEM |
| 469 | max_mapnr = num_physpages; |
| 470 | #endif |
| 471 | printk(KERN_NOTICE "%ldMB LOWMEM available.\n", |
| 472 | pages_to_mb(max_low_pfn)); |
| 473 | |
| 474 | setup_bootmem_allocator(); |
| 475 | |
| 476 | return max_low_pfn; |
| 477 | } |
| 478 | |
| 479 | void __init zone_sizes_init(void) |
| 480 | { |
| 481 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
| 482 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
| 483 | max_zone_pfns[ZONE_DMA] = |
| 484 | virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; |
| 485 | max_zone_pfns[ZONE_NORMAL] = max_low_pfn; |
| 486 | #ifdef CONFIG_HIGHMEM |
| 487 | max_zone_pfns[ZONE_HIGHMEM] = highend_pfn; |
| 488 | add_active_range(0, 0, highend_pfn); |
| 489 | #else |
| 490 | add_active_range(0, 0, max_low_pfn); |
| 491 | #endif |
| 492 | |
| 493 | free_area_init_nodes(max_zone_pfns); |
| 494 | } |
| 495 | #else |
| 496 | extern unsigned long __init setup_memory(void); |
| 497 | extern void zone_sizes_init(void); |
| 498 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
| 499 | |
| 500 | static inline unsigned long long get_total_mem(void) |
| 501 | { |
| 502 | unsigned long long total; |
| 503 | |
| 504 | total = max_low_pfn - min_low_pfn; |
| 505 | #ifdef CONFIG_HIGHMEM |
| 506 | total += highend_pfn - highstart_pfn; |
| 507 | #endif |
| 508 | |
| 509 | return total << PAGE_SHIFT; |
| 510 | } |
| 511 | |
| 512 | #ifdef CONFIG_KEXEC |
| 513 | static void __init reserve_crashkernel(void) |
| 514 | { |
| 515 | unsigned long long total_mem; |
| 516 | unsigned long long crash_size, crash_base; |
| 517 | int ret; |
| 518 | |
| 519 | total_mem = get_total_mem(); |
| 520 | |
| 521 | ret = parse_crashkernel(boot_command_line, total_mem, |
| 522 | &crash_size, &crash_base); |
| 523 | if (ret == 0 && crash_size > 0) { |
| 524 | if (crash_base > 0) { |
| 525 | printk(KERN_INFO "Reserving %ldMB of memory at %ldMB " |
| 526 | "for crashkernel (System RAM: %ldMB)\n", |
| 527 | (unsigned long)(crash_size >> 20), |
| 528 | (unsigned long)(crash_base >> 20), |
| 529 | (unsigned long)(total_mem >> 20)); |
| 530 | crashk_res.start = crash_base; |
| 531 | crashk_res.end = crash_base + crash_size - 1; |
| 532 | reserve_bootmem(crash_base, crash_size, |
| 533 | BOOTMEM_DEFAULT); |
| 534 | } else |
| 535 | printk(KERN_INFO "crashkernel reservation failed - " |
| 536 | "you have to specify a base address\n"); |
| 537 | } |
| 538 | } |
| 539 | #else |
| 540 | static inline void __init reserve_crashkernel(void) |
| 541 | {} |
| 542 | #endif |
| 543 | |
| 544 | #ifdef CONFIG_BLK_DEV_INITRD |
| 545 | |
| 546 | static bool do_relocate_initrd = false; |
| 547 | |
| 548 | static void __init reserve_initrd(void) |
| 549 | { |
| 550 | unsigned long ramdisk_image = boot_params.hdr.ramdisk_image; |
| 551 | unsigned long ramdisk_size = boot_params.hdr.ramdisk_size; |
| 552 | unsigned long ramdisk_end = ramdisk_image + ramdisk_size; |
| 553 | unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT; |
| 554 | unsigned long ramdisk_here; |
| 555 | |
| 556 | initrd_start = 0; |
| 557 | |
| 558 | if (!boot_params.hdr.type_of_loader || |
| 559 | !ramdisk_image || !ramdisk_size) |
| 560 | return; /* No initrd provided by bootloader */ |
| 561 | |
| 562 | if (ramdisk_end < ramdisk_image) { |
| 563 | printk(KERN_ERR "initrd wraps around end of memory, " |
| 564 | "disabling initrd\n"); |
| 565 | return; |
| 566 | } |
| 567 | if (ramdisk_size >= end_of_lowmem/2) { |
| 568 | printk(KERN_ERR "initrd too large to handle, " |
| 569 | "disabling initrd\n"); |
| 570 | return; |
| 571 | } |
| 572 | if (ramdisk_end <= end_of_lowmem) { |
| 573 | /* All in lowmem, easy case */ |
| 574 | reserve_bootmem(ramdisk_image, ramdisk_size, BOOTMEM_DEFAULT); |
| 575 | initrd_start = ramdisk_image + PAGE_OFFSET; |
| 576 | initrd_end = initrd_start+ramdisk_size; |
| 577 | return; |
| 578 | } |
| 579 | |
| 580 | /* We need to move the initrd down into lowmem */ |
| 581 | ramdisk_here = (end_of_lowmem - ramdisk_size) & PAGE_MASK; |
| 582 | |
| 583 | /* Note: this includes all the lowmem currently occupied by |
| 584 | the initrd, we rely on that fact to keep the data intact. */ |
| 585 | reserve_bootmem(ramdisk_here, ramdisk_size, BOOTMEM_DEFAULT); |
| 586 | initrd_start = ramdisk_here + PAGE_OFFSET; |
| 587 | initrd_end = initrd_start + ramdisk_size; |
| 588 | |
| 589 | do_relocate_initrd = true; |
| 590 | } |
| 591 | |
| 592 | #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT) |
| 593 | |
| 594 | static void __init relocate_initrd(void) |
| 595 | { |
| 596 | unsigned long ramdisk_image = boot_params.hdr.ramdisk_image; |
| 597 | unsigned long ramdisk_size = boot_params.hdr.ramdisk_size; |
| 598 | unsigned long end_of_lowmem = max_low_pfn << PAGE_SHIFT; |
| 599 | unsigned long ramdisk_here; |
| 600 | unsigned long slop, clen, mapaddr; |
| 601 | char *p, *q; |
| 602 | |
| 603 | if (!do_relocate_initrd) |
| 604 | return; |
| 605 | |
| 606 | ramdisk_here = initrd_start - PAGE_OFFSET; |
| 607 | |
| 608 | q = (char *)initrd_start; |
| 609 | |
| 610 | /* Copy any lowmem portion of the initrd */ |
| 611 | if (ramdisk_image < end_of_lowmem) { |
| 612 | clen = end_of_lowmem - ramdisk_image; |
| 613 | p = (char *)__va(ramdisk_image); |
| 614 | memcpy(q, p, clen); |
| 615 | q += clen; |
| 616 | ramdisk_image += clen; |
| 617 | ramdisk_size -= clen; |
| 618 | } |
| 619 | |
| 620 | /* Copy the highmem portion of the initrd */ |
| 621 | while (ramdisk_size) { |
| 622 | slop = ramdisk_image & ~PAGE_MASK; |
| 623 | clen = ramdisk_size; |
| 624 | if (clen > MAX_MAP_CHUNK-slop) |
| 625 | clen = MAX_MAP_CHUNK-slop; |
| 626 | mapaddr = ramdisk_image & PAGE_MASK; |
| 627 | p = early_ioremap(mapaddr, clen+slop); |
| 628 | memcpy(q, p+slop, clen); |
| 629 | early_iounmap(p, clen+slop); |
| 630 | q += clen; |
| 631 | ramdisk_image += clen; |
| 632 | ramdisk_size -= clen; |
| 633 | } |
| 634 | } |
| 635 | |
| 636 | #endif /* CONFIG_BLK_DEV_INITRD */ |
| 637 | |
| 638 | void __init setup_bootmem_allocator(void) |
| 639 | { |
| 640 | unsigned long bootmap_size; |
| 641 | /* |
| 642 | * Initialize the boot-time allocator (with low memory only): |
| 643 | */ |
| 644 | bootmap_size = init_bootmem(min_low_pfn, max_low_pfn); |
| 645 | |
| 646 | register_bootmem_low_pages(max_low_pfn); |
| 647 | |
| 648 | /* |
| 649 | * Reserve the bootmem bitmap itself as well. We do this in two |
| 650 | * steps (first step was init_bootmem()) because this catches |
| 651 | * the (very unlikely) case of us accidentally initializing the |
| 652 | * bootmem allocator with an invalid RAM area. |
| 653 | */ |
| 654 | reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) + |
| 655 | bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text), |
| 656 | BOOTMEM_DEFAULT); |
| 657 | |
| 658 | /* |
| 659 | * reserve physical page 0 - it's a special BIOS page on many boxes, |
| 660 | * enabling clean reboots, SMP operation, laptop functions. |
| 661 | */ |
| 662 | reserve_bootmem(0, PAGE_SIZE, BOOTMEM_DEFAULT); |
| 663 | |
| 664 | /* reserve EBDA region */ |
| 665 | reserve_ebda_region(); |
| 666 | |
| 667 | #ifdef CONFIG_SMP |
| 668 | /* |
| 669 | * But first pinch a few for the stack/trampoline stuff |
| 670 | * FIXME: Don't need the extra page at 4K, but need to fix |
| 671 | * trampoline before removing it. (see the GDT stuff) |
| 672 | */ |
| 673 | reserve_bootmem(PAGE_SIZE, PAGE_SIZE, BOOTMEM_DEFAULT); |
| 674 | #endif |
| 675 | #ifdef CONFIG_ACPI_SLEEP |
| 676 | /* |
| 677 | * Reserve low memory region for sleep support. |
| 678 | */ |
| 679 | acpi_reserve_bootmem(); |
| 680 | #endif |
| 681 | #ifdef CONFIG_X86_FIND_SMP_CONFIG |
| 682 | /* |
| 683 | * Find and reserve possible boot-time SMP configuration: |
| 684 | */ |
| 685 | find_smp_config(); |
| 686 | #endif |
| 687 | #ifdef CONFIG_BLK_DEV_INITRD |
| 688 | reserve_initrd(); |
| 689 | #endif |
| 690 | numa_kva_reserve(); |
| 691 | reserve_crashkernel(); |
| 692 | } |
| 693 | |
| 694 | /* |
| 695 | * The node 0 pgdat is initialized before all of these because |
| 696 | * it's needed for bootmem. node>0 pgdats have their virtual |
| 697 | * space allocated before the pagetables are in place to access |
| 698 | * them, so they can't be cleared then. |
| 699 | * |
| 700 | * This should all compile down to nothing when NUMA is off. |
| 701 | */ |
| 702 | static void __init remapped_pgdat_init(void) |
| 703 | { |
| 704 | int nid; |
| 705 | |
| 706 | for_each_online_node(nid) { |
| 707 | if (nid != 0) |
| 708 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); |
| 709 | } |
| 710 | } |
| 711 | |
| 712 | #ifdef CONFIG_MCA |
| 713 | static void set_mca_bus(int x) |
| 714 | { |
| 715 | MCA_bus = x; |
| 716 | } |
| 717 | #else |
| 718 | static void set_mca_bus(int x) { } |
| 719 | #endif |
| 720 | |
| 721 | /* Overridden in paravirt.c if CONFIG_PARAVIRT */ |
| 722 | char * __init __attribute__((weak)) memory_setup(void) |
| 723 | { |
| 724 | return machine_specific_memory_setup(); |
| 725 | } |
| 726 | |
| 727 | #ifdef CONFIG_NUMA |
| 728 | /* |
| 729 | * In the golden day, when everything among i386 and x86_64 will be |
| 730 | * integrated, this will not live here |
| 731 | */ |
| 732 | void *x86_cpu_to_node_map_early_ptr; |
| 733 | int x86_cpu_to_node_map_init[NR_CPUS] = { |
| 734 | [0 ... NR_CPUS-1] = NUMA_NO_NODE |
| 735 | }; |
| 736 | DEFINE_PER_CPU(int, x86_cpu_to_node_map) = NUMA_NO_NODE; |
| 737 | #endif |
| 738 | |
| 739 | /* |
| 740 | * Determine if we were loaded by an EFI loader. If so, then we have also been |
| 741 | * passed the efi memmap, systab, etc., so we should use these data structures |
| 742 | * for initialization. Note, the efi init code path is determined by the |
| 743 | * global efi_enabled. This allows the same kernel image to be used on existing |
| 744 | * systems (with a traditional BIOS) as well as on EFI systems. |
| 745 | */ |
| 746 | void __init setup_arch(char **cmdline_p) |
| 747 | { |
| 748 | unsigned long max_low_pfn; |
| 749 | |
| 750 | memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); |
| 751 | pre_setup_arch_hook(); |
| 752 | early_cpu_init(); |
| 753 | early_ioremap_init(); |
| 754 | |
| 755 | #ifdef CONFIG_EFI |
| 756 | if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, |
| 757 | "EL32", 4)) |
| 758 | efi_enabled = 1; |
| 759 | #endif |
| 760 | |
| 761 | ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); |
| 762 | screen_info = boot_params.screen_info; |
| 763 | edid_info = boot_params.edid_info; |
| 764 | apm_info.bios = boot_params.apm_bios_info; |
| 765 | ist_info = boot_params.ist_info; |
| 766 | saved_videomode = boot_params.hdr.vid_mode; |
| 767 | if( boot_params.sys_desc_table.length != 0 ) { |
| 768 | set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2); |
| 769 | machine_id = boot_params.sys_desc_table.table[0]; |
| 770 | machine_submodel_id = boot_params.sys_desc_table.table[1]; |
| 771 | BIOS_revision = boot_params.sys_desc_table.table[2]; |
| 772 | } |
| 773 | bootloader_type = boot_params.hdr.type_of_loader; |
| 774 | |
| 775 | #ifdef CONFIG_BLK_DEV_RAM |
| 776 | rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; |
| 777 | rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0); |
| 778 | rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0); |
| 779 | #endif |
| 780 | ARCH_SETUP |
| 781 | |
| 782 | printk(KERN_INFO "BIOS-provided physical RAM map:\n"); |
| 783 | print_memory_map(memory_setup()); |
| 784 | |
| 785 | copy_edd(); |
| 786 | |
| 787 | if (!boot_params.hdr.root_flags) |
| 788 | root_mountflags &= ~MS_RDONLY; |
| 789 | init_mm.start_code = (unsigned long) _text; |
| 790 | init_mm.end_code = (unsigned long) _etext; |
| 791 | init_mm.end_data = (unsigned long) _edata; |
| 792 | init_mm.brk = init_pg_tables_end + PAGE_OFFSET; |
| 793 | |
| 794 | code_resource.start = virt_to_phys(_text); |
| 795 | code_resource.end = virt_to_phys(_etext)-1; |
| 796 | data_resource.start = virt_to_phys(_etext); |
| 797 | data_resource.end = virt_to_phys(_edata)-1; |
| 798 | bss_resource.start = virt_to_phys(&__bss_start); |
| 799 | bss_resource.end = virt_to_phys(&__bss_stop)-1; |
| 800 | |
| 801 | parse_early_param(); |
| 802 | |
| 803 | if (user_defined_memmap) { |
| 804 | printk(KERN_INFO "user-defined physical RAM map:\n"); |
| 805 | print_memory_map("user"); |
| 806 | } |
| 807 | |
| 808 | strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); |
| 809 | *cmdline_p = command_line; |
| 810 | |
| 811 | if (efi_enabled) |
| 812 | efi_init(); |
| 813 | |
| 814 | /* update e820 for memory not covered by WB MTRRs */ |
| 815 | find_max_pfn(); |
| 816 | mtrr_bp_init(); |
| 817 | if (mtrr_trim_uncached_memory(max_pfn)) |
| 818 | find_max_pfn(); |
| 819 | |
| 820 | max_low_pfn = setup_memory(); |
| 821 | |
| 822 | #ifdef CONFIG_VMI |
| 823 | /* |
| 824 | * Must be after max_low_pfn is determined, and before kernel |
| 825 | * pagetables are setup. |
| 826 | */ |
| 827 | vmi_init(); |
| 828 | #endif |
| 829 | |
| 830 | /* |
| 831 | * NOTE: before this point _nobody_ is allowed to allocate |
| 832 | * any memory using the bootmem allocator. Although the |
| 833 | * allocator is now initialised only the first 8Mb of the kernel |
| 834 | * virtual address space has been mapped. All allocations before |
| 835 | * paging_init() has completed must use the alloc_bootmem_low_pages() |
| 836 | * variant (which allocates DMA'able memory) and care must be taken |
| 837 | * not to exceed the 8Mb limit. |
| 838 | */ |
| 839 | |
| 840 | #ifdef CONFIG_SMP |
| 841 | smp_alloc_memory(); /* AP processor realmode stacks in low memory*/ |
| 842 | #endif |
| 843 | paging_init(); |
| 844 | |
| 845 | /* |
| 846 | * NOTE: On x86-32, only from this point on, fixmaps are ready for use. |
| 847 | */ |
| 848 | |
| 849 | #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT |
| 850 | if (init_ohci1394_dma_early) |
| 851 | init_ohci1394_dma_on_all_controllers(); |
| 852 | #endif |
| 853 | |
| 854 | remapped_pgdat_init(); |
| 855 | sparse_init(); |
| 856 | zone_sizes_init(); |
| 857 | |
| 858 | /* |
| 859 | * NOTE: at this point the bootmem allocator is fully available. |
| 860 | */ |
| 861 | |
| 862 | #ifdef CONFIG_BLK_DEV_INITRD |
| 863 | relocate_initrd(); |
| 864 | #endif |
| 865 | |
| 866 | paravirt_post_allocator_init(); |
| 867 | |
| 868 | dmi_scan_machine(); |
| 869 | |
| 870 | io_delay_init(); |
| 871 | |
| 872 | #ifdef CONFIG_X86_SMP |
| 873 | /* |
| 874 | * setup to use the early static init tables during kernel startup |
| 875 | * X86_SMP will exclude sub-arches that don't deal well with it. |
| 876 | */ |
| 877 | x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init; |
| 878 | x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init; |
| 879 | #ifdef CONFIG_NUMA |
| 880 | x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init; |
| 881 | #endif |
| 882 | #endif |
| 883 | |
| 884 | #ifdef CONFIG_X86_GENERICARCH |
| 885 | generic_apic_probe(); |
| 886 | #endif |
| 887 | |
| 888 | #ifdef CONFIG_ACPI |
| 889 | /* |
| 890 | * Parse the ACPI tables for possible boot-time SMP configuration. |
| 891 | */ |
| 892 | acpi_boot_table_init(); |
| 893 | #endif |
| 894 | |
| 895 | early_quirks(); |
| 896 | |
| 897 | #ifdef CONFIG_ACPI |
| 898 | acpi_boot_init(); |
| 899 | |
| 900 | #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC) |
| 901 | if (def_to_bigsmp) |
| 902 | printk(KERN_WARNING "More than 8 CPUs detected and " |
| 903 | "CONFIG_X86_PC cannot handle it.\nUse " |
| 904 | "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n"); |
| 905 | #endif |
| 906 | #endif |
| 907 | #ifdef CONFIG_X86_LOCAL_APIC |
| 908 | if (smp_found_config) |
| 909 | get_smp_config(); |
| 910 | #endif |
| 911 | |
| 912 | e820_register_memory(); |
| 913 | e820_mark_nosave_regions(); |
| 914 | |
| 915 | #ifdef CONFIG_VT |
| 916 | #if defined(CONFIG_VGA_CONSOLE) |
| 917 | if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) |
| 918 | conswitchp = &vga_con; |
| 919 | #elif defined(CONFIG_DUMMY_CONSOLE) |
| 920 | conswitchp = &dummy_con; |
| 921 | #endif |
| 922 | #endif |
| 923 | } |
| 924 | |
| 925 | /* |
| 926 | * Request address space for all standard resources |
| 927 | * |
| 928 | * This is called just before pcibios_init(), which is also a |
| 929 | * subsys_initcall, but is linked in later (in arch/i386/pci/common.c). |
| 930 | */ |
| 931 | static int __init request_standard_resources(void) |
| 932 | { |
| 933 | int i; |
| 934 | |
| 935 | printk(KERN_INFO "Setting up standard PCI resources\n"); |
| 936 | init_iomem_resources(&code_resource, &data_resource, &bss_resource); |
| 937 | |
| 938 | request_resource(&iomem_resource, &video_ram_resource); |
| 939 | |
| 940 | /* request I/O space for devices used on all i[345]86 PCs */ |
| 941 | for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) |
| 942 | request_resource(&ioport_resource, &standard_io_resources[i]); |
| 943 | return 0; |
| 944 | } |
| 945 | |
| 946 | subsys_initcall(request_standard_resources); |