Loading...
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
72#include <video/edid.h>
73
74#include <asm/mtrr.h>
75#include <asm/apic.h>
76#include <asm/realmode.h>
77#include <asm/e820.h>
78#include <asm/mpspec.h>
79#include <asm/setup.h>
80#include <asm/efi.h>
81#include <asm/timer.h>
82#include <asm/i8259.h>
83#include <asm/sections.h>
84#include <asm/dmi.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#ifdef CONFIG_X86_64
111#include <asm/numa_64.h>
112#endif
113#include <asm/mce.h>
114#include <asm/alternative.h>
115#include <asm/prom.h>
116
117/*
118 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
119 * The direct mapping extends to max_pfn_mapped, so that we can directly access
120 * apertures, ACPI and other tables without having to play with fixmaps.
121 */
122unsigned long max_low_pfn_mapped;
123unsigned long max_pfn_mapped;
124
125#ifdef CONFIG_DMI
126RESERVE_BRK(dmi_alloc, 65536);
127#endif
128
129
130static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
131unsigned long _brk_end = (unsigned long)__brk_base;
132
133#ifdef CONFIG_X86_64
134int default_cpu_present_to_apicid(int mps_cpu)
135{
136 return __default_cpu_present_to_apicid(mps_cpu);
137}
138
139int default_check_phys_apicid_present(int phys_apicid)
140{
141 return __default_check_phys_apicid_present(phys_apicid);
142}
143#endif
144
145#ifndef CONFIG_DEBUG_BOOT_PARAMS
146struct boot_params __initdata boot_params;
147#else
148struct boot_params boot_params;
149#endif
150
151/*
152 * Machine setup..
153 */
154static struct resource data_resource = {
155 .name = "Kernel data",
156 .start = 0,
157 .end = 0,
158 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
159};
160
161static struct resource code_resource = {
162 .name = "Kernel code",
163 .start = 0,
164 .end = 0,
165 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
166};
167
168static struct resource bss_resource = {
169 .name = "Kernel bss",
170 .start = 0,
171 .end = 0,
172 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
173};
174
175
176#ifdef CONFIG_X86_32
177/* cpu data as detected by the assembly code in head.S */
178struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
179/* common cpu data for all cpus */
180struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
181EXPORT_SYMBOL(boot_cpu_data);
182
183unsigned int def_to_bigsmp;
184
185/* for MCA, but anyone else can use it if they want */
186unsigned int machine_id;
187unsigned int machine_submodel_id;
188unsigned int BIOS_revision;
189
190struct apm_info apm_info;
191EXPORT_SYMBOL(apm_info);
192
193#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
194 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
195struct ist_info ist_info;
196EXPORT_SYMBOL(ist_info);
197#else
198struct ist_info ist_info;
199#endif
200
201#else
202struct cpuinfo_x86 boot_cpu_data __read_mostly = {
203 .x86_phys_bits = MAX_PHYSMEM_BITS,
204};
205EXPORT_SYMBOL(boot_cpu_data);
206#endif
207
208
209#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
210unsigned long mmu_cr4_features;
211#else
212unsigned long mmu_cr4_features = X86_CR4_PAE;
213#endif
214
215/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
216int bootloader_type, bootloader_version;
217
218/*
219 * Setup options
220 */
221struct screen_info screen_info;
222EXPORT_SYMBOL(screen_info);
223struct edid_info edid_info;
224EXPORT_SYMBOL_GPL(edid_info);
225
226extern int root_mountflags;
227
228unsigned long saved_video_mode;
229
230#define RAMDISK_IMAGE_START_MASK 0x07FF
231#define RAMDISK_PROMPT_FLAG 0x8000
232#define RAMDISK_LOAD_FLAG 0x4000
233
234static char __initdata command_line[COMMAND_LINE_SIZE];
235#ifdef CONFIG_CMDLINE_BOOL
236static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
237#endif
238
239#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
240struct edd edd;
241#ifdef CONFIG_EDD_MODULE
242EXPORT_SYMBOL(edd);
243#endif
244/**
245 * copy_edd() - Copy the BIOS EDD information
246 * from boot_params into a safe place.
247 *
248 */
249static inline void __init copy_edd(void)
250{
251 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
252 sizeof(edd.mbr_signature));
253 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
254 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
255 edd.edd_info_nr = boot_params.eddbuf_entries;
256}
257#else
258static inline void __init copy_edd(void)
259{
260}
261#endif
262
263void * __init extend_brk(size_t size, size_t align)
264{
265 size_t mask = align - 1;
266 void *ret;
267
268 BUG_ON(_brk_start == 0);
269 BUG_ON(align & mask);
270
271 _brk_end = (_brk_end + mask) & ~mask;
272 BUG_ON((char *)(_brk_end + size) > __brk_limit);
273
274 ret = (void *)_brk_end;
275 _brk_end += size;
276
277 memset(ret, 0, size);
278
279 return ret;
280}
281
282#ifdef CONFIG_X86_64
283static void __init init_gbpages(void)
284{
285 if (direct_gbpages && cpu_has_gbpages)
286 printk(KERN_INFO "Using GB pages for direct mapping\n");
287 else
288 direct_gbpages = 0;
289}
290#else
291static inline void init_gbpages(void)
292{
293}
294static void __init cleanup_highmap(void)
295{
296}
297#endif
298
299static void __init reserve_brk(void)
300{
301 if (_brk_end > _brk_start)
302 memblock_reserve(__pa(_brk_start),
303 __pa(_brk_end) - __pa(_brk_start));
304
305 /* Mark brk area as locked down and no longer taking any
306 new allocations */
307 _brk_start = 0;
308}
309
310#ifdef CONFIG_BLK_DEV_INITRD
311
312#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
313static void __init relocate_initrd(void)
314{
315 /* Assume only end is not page aligned */
316 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
317 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
318 u64 area_size = PAGE_ALIGN(ramdisk_size);
319 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
320 u64 ramdisk_here;
321 unsigned long slop, clen, mapaddr;
322 char *p, *q;
323
324 /* We need to move the initrd down into lowmem */
325 ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size,
326 PAGE_SIZE);
327
328 if (!ramdisk_here)
329 panic("Cannot find place for new RAMDISK of size %lld\n",
330 ramdisk_size);
331
332 /* Note: this includes all the lowmem currently occupied by
333 the initrd, we rely on that fact to keep the data intact. */
334 memblock_reserve(ramdisk_here, area_size);
335 initrd_start = ramdisk_here + PAGE_OFFSET;
336 initrd_end = initrd_start + ramdisk_size;
337 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
338 ramdisk_here, ramdisk_here + ramdisk_size - 1);
339
340 q = (char *)initrd_start;
341
342 /* Copy any lowmem portion of the initrd */
343 if (ramdisk_image < end_of_lowmem) {
344 clen = end_of_lowmem - ramdisk_image;
345 p = (char *)__va(ramdisk_image);
346 memcpy(q, p, clen);
347 q += clen;
348 ramdisk_image += clen;
349 ramdisk_size -= clen;
350 }
351
352 /* Copy the highmem portion of the initrd */
353 while (ramdisk_size) {
354 slop = ramdisk_image & ~PAGE_MASK;
355 clen = ramdisk_size;
356 if (clen > MAX_MAP_CHUNK-slop)
357 clen = MAX_MAP_CHUNK-slop;
358 mapaddr = ramdisk_image & PAGE_MASK;
359 p = early_memremap(mapaddr, clen+slop);
360 memcpy(q, p+slop, clen);
361 early_iounmap(p, clen+slop);
362 q += clen;
363 ramdisk_image += clen;
364 ramdisk_size -= clen;
365 }
366 /* high pages is not converted by early_res_to_bootmem */
367 ramdisk_image = boot_params.hdr.ramdisk_image;
368 ramdisk_size = boot_params.hdr.ramdisk_size;
369 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
370 " [mem %#010llx-%#010llx]\n",
371 ramdisk_image, ramdisk_image + ramdisk_size - 1,
372 ramdisk_here, ramdisk_here + ramdisk_size - 1);
373}
374
375static void __init reserve_initrd(void)
376{
377 /* Assume only end is not page aligned */
378 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
379 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
380 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
381 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
382
383 if (!boot_params.hdr.type_of_loader ||
384 !ramdisk_image || !ramdisk_size)
385 return; /* No initrd provided by bootloader */
386
387 initrd_start = 0;
388
389 if (ramdisk_size >= (end_of_lowmem>>1)) {
390 panic("initrd too large to handle, "
391 "disabling initrd (%lld needed, %lld available)\n",
392 ramdisk_size, end_of_lowmem>>1);
393 }
394
395 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
396 ramdisk_end - 1);
397
398
399 if (ramdisk_end <= end_of_lowmem) {
400 /* All in lowmem, easy case */
401 /*
402 * don't need to reserve again, already reserved early
403 * in i386_start_kernel
404 */
405 initrd_start = ramdisk_image + PAGE_OFFSET;
406 initrd_end = initrd_start + ramdisk_size;
407 return;
408 }
409
410 relocate_initrd();
411
412 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
413}
414#else
415static void __init reserve_initrd(void)
416{
417}
418#endif /* CONFIG_BLK_DEV_INITRD */
419
420static void __init parse_setup_data(void)
421{
422 struct setup_data *data;
423 u64 pa_data;
424
425 if (boot_params.hdr.version < 0x0209)
426 return;
427 pa_data = boot_params.hdr.setup_data;
428 while (pa_data) {
429 u32 data_len, map_len;
430
431 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
432 (u64)sizeof(struct setup_data));
433 data = early_memremap(pa_data, map_len);
434 data_len = data->len + sizeof(struct setup_data);
435 if (data_len > map_len) {
436 early_iounmap(data, map_len);
437 data = early_memremap(pa_data, data_len);
438 map_len = data_len;
439 }
440
441 switch (data->type) {
442 case SETUP_E820_EXT:
443 parse_e820_ext(data);
444 break;
445 case SETUP_DTB:
446 add_dtb(pa_data);
447 break;
448 default:
449 break;
450 }
451 pa_data = data->next;
452 early_iounmap(data, map_len);
453 }
454}
455
456static void __init e820_reserve_setup_data(void)
457{
458 struct setup_data *data;
459 u64 pa_data;
460 int found = 0;
461
462 if (boot_params.hdr.version < 0x0209)
463 return;
464 pa_data = boot_params.hdr.setup_data;
465 while (pa_data) {
466 data = early_memremap(pa_data, sizeof(*data));
467 e820_update_range(pa_data, sizeof(*data)+data->len,
468 E820_RAM, E820_RESERVED_KERN);
469 found = 1;
470 pa_data = data->next;
471 early_iounmap(data, sizeof(*data));
472 }
473 if (!found)
474 return;
475
476 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
477 memcpy(&e820_saved, &e820, sizeof(struct e820map));
478 printk(KERN_INFO "extended physical RAM map:\n");
479 e820_print_map("reserve setup_data");
480}
481
482static void __init memblock_x86_reserve_range_setup_data(void)
483{
484 struct setup_data *data;
485 u64 pa_data;
486
487 if (boot_params.hdr.version < 0x0209)
488 return;
489 pa_data = boot_params.hdr.setup_data;
490 while (pa_data) {
491 data = early_memremap(pa_data, sizeof(*data));
492 memblock_reserve(pa_data, sizeof(*data) + data->len);
493 pa_data = data->next;
494 early_iounmap(data, sizeof(*data));
495 }
496}
497
498/*
499 * --------- Crashkernel reservation ------------------------------
500 */
501
502#ifdef CONFIG_KEXEC
503
504/*
505 * Keep the crash kernel below this limit. On 32 bits earlier kernels
506 * would limit the kernel to the low 512 MiB due to mapping restrictions.
507 * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this
508 * limit once kexec-tools are fixed.
509 */
510#ifdef CONFIG_X86_32
511# define CRASH_KERNEL_ADDR_MAX (512 << 20)
512#else
513# define CRASH_KERNEL_ADDR_MAX (896 << 20)
514#endif
515
516static void __init reserve_crashkernel(void)
517{
518 unsigned long long total_mem;
519 unsigned long long crash_size, crash_base;
520 int ret;
521
522 total_mem = memblock_phys_mem_size();
523
524 ret = parse_crashkernel(boot_command_line, total_mem,
525 &crash_size, &crash_base);
526 if (ret != 0 || crash_size <= 0)
527 return;
528
529 /* 0 means: find the address automatically */
530 if (crash_base <= 0) {
531 const unsigned long long alignment = 16<<20; /* 16M */
532
533 /*
534 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
535 */
536 crash_base = memblock_find_in_range(alignment,
537 CRASH_KERNEL_ADDR_MAX, crash_size, alignment);
538
539 if (!crash_base) {
540 pr_info("crashkernel reservation failed - No suitable area found.\n");
541 return;
542 }
543 } else {
544 unsigned long long start;
545
546 start = memblock_find_in_range(crash_base,
547 crash_base + crash_size, crash_size, 1<<20);
548 if (start != crash_base) {
549 pr_info("crashkernel reservation failed - memory is in use.\n");
550 return;
551 }
552 }
553 memblock_reserve(crash_base, crash_size);
554
555 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
556 "for crashkernel (System RAM: %ldMB)\n",
557 (unsigned long)(crash_size >> 20),
558 (unsigned long)(crash_base >> 20),
559 (unsigned long)(total_mem >> 20));
560
561 crashk_res.start = crash_base;
562 crashk_res.end = crash_base + crash_size - 1;
563 insert_resource(&iomem_resource, &crashk_res);
564}
565#else
566static void __init reserve_crashkernel(void)
567{
568}
569#endif
570
571static struct resource standard_io_resources[] = {
572 { .name = "dma1", .start = 0x00, .end = 0x1f,
573 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
574 { .name = "pic1", .start = 0x20, .end = 0x21,
575 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
576 { .name = "timer0", .start = 0x40, .end = 0x43,
577 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
578 { .name = "timer1", .start = 0x50, .end = 0x53,
579 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
580 { .name = "keyboard", .start = 0x60, .end = 0x60,
581 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
582 { .name = "keyboard", .start = 0x64, .end = 0x64,
583 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
584 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
585 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
586 { .name = "pic2", .start = 0xa0, .end = 0xa1,
587 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
588 { .name = "dma2", .start = 0xc0, .end = 0xdf,
589 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
590 { .name = "fpu", .start = 0xf0, .end = 0xff,
591 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
592};
593
594void __init reserve_standard_io_resources(void)
595{
596 int i;
597
598 /* request I/O space for devices used on all i[345]86 PCs */
599 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
600 request_resource(&ioport_resource, &standard_io_resources[i]);
601
602}
603
604static __init void reserve_ibft_region(void)
605{
606 unsigned long addr, size = 0;
607
608 addr = find_ibft_region(&size);
609
610 if (size)
611 memblock_reserve(addr, size);
612}
613
614static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
615
616static void __init trim_bios_range(void)
617{
618 /*
619 * A special case is the first 4Kb of memory;
620 * This is a BIOS owned area, not kernel ram, but generally
621 * not listed as such in the E820 table.
622 *
623 * This typically reserves additional memory (64KiB by default)
624 * since some BIOSes are known to corrupt low memory. See the
625 * Kconfig help text for X86_RESERVE_LOW.
626 */
627 e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE),
628 E820_RAM, E820_RESERVED);
629
630 /*
631 * special case: Some BIOSen report the PC BIOS
632 * area (640->1Mb) as ram even though it is not.
633 * take them out.
634 */
635 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
636 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
637}
638
639static int __init parse_reservelow(char *p)
640{
641 unsigned long long size;
642
643 if (!p)
644 return -EINVAL;
645
646 size = memparse(p, &p);
647
648 if (size < 4096)
649 size = 4096;
650
651 if (size > 640*1024)
652 size = 640*1024;
653
654 reserve_low = size;
655
656 return 0;
657}
658
659early_param("reservelow", parse_reservelow);
660
661/*
662 * Determine if we were loaded by an EFI loader. If so, then we have also been
663 * passed the efi memmap, systab, etc., so we should use these data structures
664 * for initialization. Note, the efi init code path is determined by the
665 * global efi_enabled. This allows the same kernel image to be used on existing
666 * systems (with a traditional BIOS) as well as on EFI systems.
667 */
668/*
669 * setup_arch - architecture-specific boot-time initializations
670 *
671 * Note: On x86_64, fixmaps are ready for use even before this is called.
672 */
673
674void __init setup_arch(char **cmdline_p)
675{
676#ifdef CONFIG_X86_32
677 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
678 visws_early_detect();
679
680 /*
681 * copy kernel address range established so far and switch
682 * to the proper swapper page table
683 */
684 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
685 initial_page_table + KERNEL_PGD_BOUNDARY,
686 KERNEL_PGD_PTRS);
687
688 load_cr3(swapper_pg_dir);
689 __flush_tlb_all();
690#else
691 printk(KERN_INFO "Command line: %s\n", boot_command_line);
692#endif
693
694 /*
695 * If we have OLPC OFW, we might end up relocating the fixmap due to
696 * reserve_top(), so do this before touching the ioremap area.
697 */
698 olpc_ofw_detect();
699
700 early_trap_init();
701 early_cpu_init();
702 early_ioremap_init();
703
704 setup_olpc_ofw_pgd();
705
706 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
707 screen_info = boot_params.screen_info;
708 edid_info = boot_params.edid_info;
709#ifdef CONFIG_X86_32
710 apm_info.bios = boot_params.apm_bios_info;
711 ist_info = boot_params.ist_info;
712 if (boot_params.sys_desc_table.length != 0) {
713 machine_id = boot_params.sys_desc_table.table[0];
714 machine_submodel_id = boot_params.sys_desc_table.table[1];
715 BIOS_revision = boot_params.sys_desc_table.table[2];
716 }
717#endif
718 saved_video_mode = boot_params.hdr.vid_mode;
719 bootloader_type = boot_params.hdr.type_of_loader;
720 if ((bootloader_type >> 4) == 0xe) {
721 bootloader_type &= 0xf;
722 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
723 }
724 bootloader_version = bootloader_type & 0xf;
725 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
726
727#ifdef CONFIG_BLK_DEV_RAM
728 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
729 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
730 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
731#endif
732#ifdef CONFIG_EFI
733 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
734 "EL32", 4)) {
735 efi_enabled = 1;
736 efi_64bit = false;
737 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
738 "EL64", 4)) {
739 efi_enabled = 1;
740 efi_64bit = true;
741 }
742 if (efi_enabled && efi_memblock_x86_reserve_range())
743 efi_enabled = 0;
744#endif
745
746 x86_init.oem.arch_setup();
747
748 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
749 setup_memory_map();
750 parse_setup_data();
751 /* update the e820_saved too */
752 e820_reserve_setup_data();
753
754 copy_edd();
755
756 if (!boot_params.hdr.root_flags)
757 root_mountflags &= ~MS_RDONLY;
758 init_mm.start_code = (unsigned long) _text;
759 init_mm.end_code = (unsigned long) _etext;
760 init_mm.end_data = (unsigned long) _edata;
761 init_mm.brk = _brk_end;
762
763 code_resource.start = virt_to_phys(_text);
764 code_resource.end = virt_to_phys(_etext)-1;
765 data_resource.start = virt_to_phys(_etext);
766 data_resource.end = virt_to_phys(_edata)-1;
767 bss_resource.start = virt_to_phys(&__bss_start);
768 bss_resource.end = virt_to_phys(&__bss_stop)-1;
769
770#ifdef CONFIG_CMDLINE_BOOL
771#ifdef CONFIG_CMDLINE_OVERRIDE
772 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
773#else
774 if (builtin_cmdline[0]) {
775 /* append boot loader cmdline to builtin */
776 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
777 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
778 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
779 }
780#endif
781#endif
782
783 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
784 *cmdline_p = command_line;
785
786 /*
787 * x86_configure_nx() is called before parse_early_param() to detect
788 * whether hardware doesn't support NX (so that the early EHCI debug
789 * console setup can safely call set_fixmap()). It may then be called
790 * again from within noexec_setup() during parsing early parameters
791 * to honor the respective command line option.
792 */
793 x86_configure_nx();
794
795 parse_early_param();
796
797 x86_report_nx();
798
799 /* after early param, so could get panic from serial */
800 memblock_x86_reserve_range_setup_data();
801
802 if (acpi_mps_check()) {
803#ifdef CONFIG_X86_LOCAL_APIC
804 disable_apic = 1;
805#endif
806 setup_clear_cpu_cap(X86_FEATURE_APIC);
807 }
808
809#ifdef CONFIG_PCI
810 if (pci_early_dump_regs)
811 early_dump_pci_devices();
812#endif
813
814 finish_e820_parsing();
815
816 if (efi_enabled)
817 efi_init();
818
819 dmi_scan_machine();
820
821 /*
822 * VMware detection requires dmi to be available, so this
823 * needs to be done after dmi_scan_machine, for the BP.
824 */
825 init_hypervisor_platform();
826
827 x86_init.resources.probe_roms();
828
829 /* after parse_early_param, so could debug it */
830 insert_resource(&iomem_resource, &code_resource);
831 insert_resource(&iomem_resource, &data_resource);
832 insert_resource(&iomem_resource, &bss_resource);
833
834 trim_bios_range();
835#ifdef CONFIG_X86_32
836 if (ppro_with_ram_bug()) {
837 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
838 E820_RESERVED);
839 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
840 printk(KERN_INFO "fixed physical RAM map:\n");
841 e820_print_map("bad_ppro");
842 }
843#else
844 early_gart_iommu_check();
845#endif
846
847 /*
848 * partially used pages are not usable - thus
849 * we are rounding upwards:
850 */
851 max_pfn = e820_end_of_ram_pfn();
852
853 /* update e820 for memory not covered by WB MTRRs */
854 mtrr_bp_init();
855 if (mtrr_trim_uncached_memory(max_pfn))
856 max_pfn = e820_end_of_ram_pfn();
857
858#ifdef CONFIG_X86_32
859 /* max_low_pfn get updated here */
860 find_low_pfn_range();
861#else
862 num_physpages = max_pfn;
863
864 check_x2apic();
865
866 /* How many end-of-memory variables you have, grandma! */
867 /* need this before calling reserve_initrd */
868 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
869 max_low_pfn = e820_end_of_low_ram_pfn();
870 else
871 max_low_pfn = max_pfn;
872
873 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
874#endif
875
876 /*
877 * Find and reserve possible boot-time SMP configuration:
878 */
879 find_smp_config();
880
881 reserve_ibft_region();
882
883 /*
884 * Need to conclude brk, before memblock_x86_fill()
885 * it could use memblock_find_in_range, could overlap with
886 * brk area.
887 */
888 reserve_brk();
889
890 cleanup_highmap();
891
892 memblock.current_limit = get_max_mapped();
893 memblock_x86_fill();
894
895 /*
896 * The EFI specification says that boot service code won't be called
897 * after ExitBootServices(). This is, in fact, a lie.
898 */
899 if (efi_enabled)
900 efi_reserve_boot_services();
901
902 /* preallocate 4k for mptable mpc */
903 early_reserve_e820_mpc_new();
904
905#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
906 setup_bios_corruption_check();
907#endif
908
909 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
910 (max_pfn_mapped<<PAGE_SHIFT) - 1);
911
912 setup_real_mode();
913
914 init_gbpages();
915
916 /* max_pfn_mapped is updated here */
917 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
918 max_pfn_mapped = max_low_pfn_mapped;
919
920#ifdef CONFIG_X86_64
921 if (max_pfn > max_low_pfn) {
922 max_pfn_mapped = init_memory_mapping(1UL<<32,
923 max_pfn<<PAGE_SHIFT);
924 /* can we preseve max_low_pfn ?*/
925 max_low_pfn = max_pfn;
926 }
927#endif
928 memblock.current_limit = get_max_mapped();
929 dma_contiguous_reserve(0);
930
931 /*
932 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
933 */
934
935#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
936 if (init_ohci1394_dma_early)
937 init_ohci1394_dma_on_all_controllers();
938#endif
939 /* Allocate bigger log buffer */
940 setup_log_buf(1);
941
942 reserve_initrd();
943
944 reserve_crashkernel();
945
946 vsmp_init();
947
948 io_delay_init();
949
950 /*
951 * Parse the ACPI tables for possible boot-time SMP configuration.
952 */
953 acpi_boot_table_init();
954
955 early_acpi_boot_init();
956
957 initmem_init();
958 memblock_find_dma_reserve();
959
960#ifdef CONFIG_KVM_CLOCK
961 kvmclock_init();
962#endif
963
964 x86_init.paging.pagetable_setup_start(swapper_pg_dir);
965 paging_init();
966 x86_init.paging.pagetable_setup_done(swapper_pg_dir);
967
968 if (boot_cpu_data.cpuid_level >= 0) {
969 /* A CPU has %cr4 if and only if it has CPUID */
970 mmu_cr4_features = read_cr4();
971 if (trampoline_cr4_features)
972 *trampoline_cr4_features = mmu_cr4_features;
973 }
974
975#ifdef CONFIG_X86_32
976 /* sync back kernel address range */
977 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
978 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
979 KERNEL_PGD_PTRS);
980#endif
981
982 tboot_probe();
983
984#ifdef CONFIG_X86_64
985 map_vsyscall();
986#endif
987
988 generic_apic_probe();
989
990 early_quirks();
991
992 /*
993 * Read APIC and some other early information from ACPI tables.
994 */
995 acpi_boot_init();
996 sfi_init();
997 x86_dtb_init();
998
999 /*
1000 * get boot-time SMP configuration:
1001 */
1002 if (smp_found_config)
1003 get_smp_config();
1004
1005 prefill_possible_map();
1006
1007 init_cpu_to_node();
1008
1009 init_apic_mappings();
1010 if (x86_io_apic_ops.init)
1011 x86_io_apic_ops.init();
1012
1013 kvm_guest_init();
1014
1015 e820_reserve_resources();
1016 e820_mark_nosave_regions(max_low_pfn);
1017
1018 x86_init.resources.reserve_resources();
1019
1020 e820_setup_gap();
1021
1022#ifdef CONFIG_VT
1023#if defined(CONFIG_VGA_CONSOLE)
1024 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1025 conswitchp = &vga_con;
1026#elif defined(CONFIG_DUMMY_CONSOLE)
1027 conswitchp = &dummy_con;
1028#endif
1029#endif
1030 x86_init.oem.banner();
1031
1032 x86_init.timers.wallclock_init();
1033
1034 x86_platform.wallclock_init();
1035
1036 mcheck_init();
1037
1038 arch_init_ideal_nops();
1039}
1040
1041#ifdef CONFIG_X86_32
1042
1043static struct resource video_ram_resource = {
1044 .name = "Video RAM area",
1045 .start = 0xa0000,
1046 .end = 0xbffff,
1047 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1048};
1049
1050void __init i386_reserve_resources(void)
1051{
1052 request_resource(&iomem_resource, &video_ram_resource);
1053 reserve_standard_io_resources();
1054}
1055
1056#endif /* CONFIG_X86_32 */
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/mca.h>
38#include <linux/root_dev.h>
39#include <linux/highmem.h>
40#include <linux/module.h>
41#include <linux/efi.h>
42#include <linux/init.h>
43#include <linux/edd.h>
44#include <linux/iscsi_ibft.h>
45#include <linux/nodemask.h>
46#include <linux/kexec.h>
47#include <linux/dmi.h>
48#include <linux/pfn.h>
49#include <linux/pci.h>
50#include <asm/pci-direct.h>
51#include <linux/init_ohci1394_dma.h>
52#include <linux/kvm_para.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
72#include <video/edid.h>
73
74#include <asm/mtrr.h>
75#include <asm/apic.h>
76#include <asm/trampoline.h>
77#include <asm/e820.h>
78#include <asm/mpspec.h>
79#include <asm/setup.h>
80#include <asm/efi.h>
81#include <asm/timer.h>
82#include <asm/i8259.h>
83#include <asm/sections.h>
84#include <asm/dmi.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/system.h>
94#include <asm/vsyscall.h>
95#include <asm/cpu.h>
96#include <asm/desc.h>
97#include <asm/dma.h>
98#include <asm/iommu.h>
99#include <asm/gart.h>
100#include <asm/mmu_context.h>
101#include <asm/proto.h>
102
103#include <asm/paravirt.h>
104#include <asm/hypervisor.h>
105#include <asm/olpc_ofw.h>
106
107#include <asm/percpu.h>
108#include <asm/topology.h>
109#include <asm/apicdef.h>
110#include <asm/amd_nb.h>
111#ifdef CONFIG_X86_64
112#include <asm/numa_64.h>
113#endif
114#include <asm/mce.h>
115#include <asm/alternative.h>
116#include <asm/prom.h>
117
118/*
119 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
120 * The direct mapping extends to max_pfn_mapped, so that we can directly access
121 * apertures, ACPI and other tables without having to play with fixmaps.
122 */
123unsigned long max_low_pfn_mapped;
124unsigned long max_pfn_mapped;
125
126#ifdef CONFIG_DMI
127RESERVE_BRK(dmi_alloc, 65536);
128#endif
129
130
131static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
132unsigned long _brk_end = (unsigned long)__brk_base;
133
134#ifdef CONFIG_X86_64
135int default_cpu_present_to_apicid(int mps_cpu)
136{
137 return __default_cpu_present_to_apicid(mps_cpu);
138}
139
140int default_check_phys_apicid_present(int phys_apicid)
141{
142 return __default_check_phys_apicid_present(phys_apicid);
143}
144#endif
145
146#ifndef CONFIG_DEBUG_BOOT_PARAMS
147struct boot_params __initdata boot_params;
148#else
149struct boot_params boot_params;
150#endif
151
152/*
153 * Machine setup..
154 */
155static struct resource data_resource = {
156 .name = "Kernel data",
157 .start = 0,
158 .end = 0,
159 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
160};
161
162static struct resource code_resource = {
163 .name = "Kernel code",
164 .start = 0,
165 .end = 0,
166 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
167};
168
169static struct resource bss_resource = {
170 .name = "Kernel bss",
171 .start = 0,
172 .end = 0,
173 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
174};
175
176
177#ifdef CONFIG_X86_32
178/* cpu data as detected by the assembly code in head.S */
179struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
180/* common cpu data for all cpus */
181struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
182EXPORT_SYMBOL(boot_cpu_data);
183static void set_mca_bus(int x)
184{
185#ifdef CONFIG_MCA
186 MCA_bus = x;
187#endif
188}
189
190unsigned int def_to_bigsmp;
191
192/* for MCA, but anyone else can use it if they want */
193unsigned int machine_id;
194unsigned int machine_submodel_id;
195unsigned int BIOS_revision;
196
197struct apm_info apm_info;
198EXPORT_SYMBOL(apm_info);
199
200#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
201 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
202struct ist_info ist_info;
203EXPORT_SYMBOL(ist_info);
204#else
205struct ist_info ist_info;
206#endif
207
208#else
209struct cpuinfo_x86 boot_cpu_data __read_mostly = {
210 .x86_phys_bits = MAX_PHYSMEM_BITS,
211};
212EXPORT_SYMBOL(boot_cpu_data);
213#endif
214
215
216#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
217unsigned long mmu_cr4_features;
218#else
219unsigned long mmu_cr4_features = X86_CR4_PAE;
220#endif
221
222/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
223int bootloader_type, bootloader_version;
224
225/*
226 * Setup options
227 */
228struct screen_info screen_info;
229EXPORT_SYMBOL(screen_info);
230struct edid_info edid_info;
231EXPORT_SYMBOL_GPL(edid_info);
232
233extern int root_mountflags;
234
235unsigned long saved_video_mode;
236
237#define RAMDISK_IMAGE_START_MASK 0x07FF
238#define RAMDISK_PROMPT_FLAG 0x8000
239#define RAMDISK_LOAD_FLAG 0x4000
240
241static char __initdata command_line[COMMAND_LINE_SIZE];
242#ifdef CONFIG_CMDLINE_BOOL
243static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
244#endif
245
246#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
247struct edd edd;
248#ifdef CONFIG_EDD_MODULE
249EXPORT_SYMBOL(edd);
250#endif
251/**
252 * copy_edd() - Copy the BIOS EDD information
253 * from boot_params into a safe place.
254 *
255 */
256static inline void __init copy_edd(void)
257{
258 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
259 sizeof(edd.mbr_signature));
260 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
261 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
262 edd.edd_info_nr = boot_params.eddbuf_entries;
263}
264#else
265static inline void __init copy_edd(void)
266{
267}
268#endif
269
270void * __init extend_brk(size_t size, size_t align)
271{
272 size_t mask = align - 1;
273 void *ret;
274
275 BUG_ON(_brk_start == 0);
276 BUG_ON(align & mask);
277
278 _brk_end = (_brk_end + mask) & ~mask;
279 BUG_ON((char *)(_brk_end + size) > __brk_limit);
280
281 ret = (void *)_brk_end;
282 _brk_end += size;
283
284 memset(ret, 0, size);
285
286 return ret;
287}
288
289#ifdef CONFIG_X86_64
290static void __init init_gbpages(void)
291{
292 if (direct_gbpages && cpu_has_gbpages)
293 printk(KERN_INFO "Using GB pages for direct mapping\n");
294 else
295 direct_gbpages = 0;
296}
297#else
298static inline void init_gbpages(void)
299{
300}
301static void __init cleanup_highmap(void)
302{
303}
304#endif
305
306static void __init reserve_brk(void)
307{
308 if (_brk_end > _brk_start)
309 memblock_x86_reserve_range(__pa(_brk_start), __pa(_brk_end), "BRK");
310
311 /* Mark brk area as locked down and no longer taking any
312 new allocations */
313 _brk_start = 0;
314}
315
316#ifdef CONFIG_BLK_DEV_INITRD
317
318#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
319static void __init relocate_initrd(void)
320{
321 /* Assume only end is not page aligned */
322 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
323 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
324 u64 area_size = PAGE_ALIGN(ramdisk_size);
325 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
326 u64 ramdisk_here;
327 unsigned long slop, clen, mapaddr;
328 char *p, *q;
329
330 /* We need to move the initrd down into lowmem */
331 ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size,
332 PAGE_SIZE);
333
334 if (ramdisk_here == MEMBLOCK_ERROR)
335 panic("Cannot find place for new RAMDISK of size %lld\n",
336 ramdisk_size);
337
338 /* Note: this includes all the lowmem currently occupied by
339 the initrd, we rely on that fact to keep the data intact. */
340 memblock_x86_reserve_range(ramdisk_here, ramdisk_here + area_size, "NEW RAMDISK");
341 initrd_start = ramdisk_here + PAGE_OFFSET;
342 initrd_end = initrd_start + ramdisk_size;
343 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
344 ramdisk_here, ramdisk_here + ramdisk_size);
345
346 q = (char *)initrd_start;
347
348 /* Copy any lowmem portion of the initrd */
349 if (ramdisk_image < end_of_lowmem) {
350 clen = end_of_lowmem - ramdisk_image;
351 p = (char *)__va(ramdisk_image);
352 memcpy(q, p, clen);
353 q += clen;
354 ramdisk_image += clen;
355 ramdisk_size -= clen;
356 }
357
358 /* Copy the highmem portion of the initrd */
359 while (ramdisk_size) {
360 slop = ramdisk_image & ~PAGE_MASK;
361 clen = ramdisk_size;
362 if (clen > MAX_MAP_CHUNK-slop)
363 clen = MAX_MAP_CHUNK-slop;
364 mapaddr = ramdisk_image & PAGE_MASK;
365 p = early_memremap(mapaddr, clen+slop);
366 memcpy(q, p+slop, clen);
367 early_iounmap(p, clen+slop);
368 q += clen;
369 ramdisk_image += clen;
370 ramdisk_size -= clen;
371 }
372 /* high pages is not converted by early_res_to_bootmem */
373 ramdisk_image = boot_params.hdr.ramdisk_image;
374 ramdisk_size = boot_params.hdr.ramdisk_size;
375 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
376 " %08llx - %08llx\n",
377 ramdisk_image, ramdisk_image + ramdisk_size - 1,
378 ramdisk_here, ramdisk_here + ramdisk_size - 1);
379}
380
381static void __init reserve_initrd(void)
382{
383 /* Assume only end is not page aligned */
384 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
385 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
386 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
387 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
388
389 if (!boot_params.hdr.type_of_loader ||
390 !ramdisk_image || !ramdisk_size)
391 return; /* No initrd provided by bootloader */
392
393 initrd_start = 0;
394
395 if (ramdisk_size >= (end_of_lowmem>>1)) {
396 memblock_x86_free_range(ramdisk_image, ramdisk_end);
397 printk(KERN_ERR "initrd too large to handle, "
398 "disabling initrd\n");
399 return;
400 }
401
402 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
403 ramdisk_end);
404
405
406 if (ramdisk_end <= end_of_lowmem) {
407 /* All in lowmem, easy case */
408 /*
409 * don't need to reserve again, already reserved early
410 * in i386_start_kernel
411 */
412 initrd_start = ramdisk_image + PAGE_OFFSET;
413 initrd_end = initrd_start + ramdisk_size;
414 return;
415 }
416
417 relocate_initrd();
418
419 memblock_x86_free_range(ramdisk_image, ramdisk_end);
420}
421#else
422static void __init reserve_initrd(void)
423{
424}
425#endif /* CONFIG_BLK_DEV_INITRD */
426
427static void __init parse_setup_data(void)
428{
429 struct setup_data *data;
430 u64 pa_data;
431
432 if (boot_params.hdr.version < 0x0209)
433 return;
434 pa_data = boot_params.hdr.setup_data;
435 while (pa_data) {
436 u32 data_len, map_len;
437
438 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
439 (u64)sizeof(struct setup_data));
440 data = early_memremap(pa_data, map_len);
441 data_len = data->len + sizeof(struct setup_data);
442 if (data_len > map_len) {
443 early_iounmap(data, map_len);
444 data = early_memremap(pa_data, data_len);
445 map_len = data_len;
446 }
447
448 switch (data->type) {
449 case SETUP_E820_EXT:
450 parse_e820_ext(data);
451 break;
452 case SETUP_DTB:
453 add_dtb(pa_data);
454 break;
455 default:
456 break;
457 }
458 pa_data = data->next;
459 early_iounmap(data, map_len);
460 }
461}
462
463static void __init e820_reserve_setup_data(void)
464{
465 struct setup_data *data;
466 u64 pa_data;
467 int found = 0;
468
469 if (boot_params.hdr.version < 0x0209)
470 return;
471 pa_data = boot_params.hdr.setup_data;
472 while (pa_data) {
473 data = early_memremap(pa_data, sizeof(*data));
474 e820_update_range(pa_data, sizeof(*data)+data->len,
475 E820_RAM, E820_RESERVED_KERN);
476 found = 1;
477 pa_data = data->next;
478 early_iounmap(data, sizeof(*data));
479 }
480 if (!found)
481 return;
482
483 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
484 memcpy(&e820_saved, &e820, sizeof(struct e820map));
485 printk(KERN_INFO "extended physical RAM map:\n");
486 e820_print_map("reserve setup_data");
487}
488
489static void __init memblock_x86_reserve_range_setup_data(void)
490{
491 struct setup_data *data;
492 u64 pa_data;
493 char buf[32];
494
495 if (boot_params.hdr.version < 0x0209)
496 return;
497 pa_data = boot_params.hdr.setup_data;
498 while (pa_data) {
499 data = early_memremap(pa_data, sizeof(*data));
500 sprintf(buf, "setup data %x", data->type);
501 memblock_x86_reserve_range(pa_data, pa_data+sizeof(*data)+data->len, buf);
502 pa_data = data->next;
503 early_iounmap(data, sizeof(*data));
504 }
505}
506
507/*
508 * --------- Crashkernel reservation ------------------------------
509 */
510
511#ifdef CONFIG_KEXEC
512
513static inline unsigned long long get_total_mem(void)
514{
515 unsigned long long total;
516
517 total = max_pfn - min_low_pfn;
518
519 return total << PAGE_SHIFT;
520}
521
522/*
523 * Keep the crash kernel below this limit. On 32 bits earlier kernels
524 * would limit the kernel to the low 512 MiB due to mapping restrictions.
525 * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this
526 * limit once kexec-tools are fixed.
527 */
528#ifdef CONFIG_X86_32
529# define CRASH_KERNEL_ADDR_MAX (512 << 20)
530#else
531# define CRASH_KERNEL_ADDR_MAX (896 << 20)
532#endif
533
534static void __init reserve_crashkernel(void)
535{
536 unsigned long long total_mem;
537 unsigned long long crash_size, crash_base;
538 int ret;
539
540 total_mem = get_total_mem();
541
542 ret = parse_crashkernel(boot_command_line, total_mem,
543 &crash_size, &crash_base);
544 if (ret != 0 || crash_size <= 0)
545 return;
546
547 /* 0 means: find the address automatically */
548 if (crash_base <= 0) {
549 const unsigned long long alignment = 16<<20; /* 16M */
550
551 /*
552 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
553 */
554 crash_base = memblock_find_in_range(alignment,
555 CRASH_KERNEL_ADDR_MAX, crash_size, alignment);
556
557 if (crash_base == MEMBLOCK_ERROR) {
558 pr_info("crashkernel reservation failed - No suitable area found.\n");
559 return;
560 }
561 } else {
562 unsigned long long start;
563
564 start = memblock_find_in_range(crash_base,
565 crash_base + crash_size, crash_size, 1<<20);
566 if (start != crash_base) {
567 pr_info("crashkernel reservation failed - memory is in use.\n");
568 return;
569 }
570 }
571 memblock_x86_reserve_range(crash_base, crash_base + crash_size, "CRASH KERNEL");
572
573 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
574 "for crashkernel (System RAM: %ldMB)\n",
575 (unsigned long)(crash_size >> 20),
576 (unsigned long)(crash_base >> 20),
577 (unsigned long)(total_mem >> 20));
578
579 crashk_res.start = crash_base;
580 crashk_res.end = crash_base + crash_size - 1;
581 insert_resource(&iomem_resource, &crashk_res);
582}
583#else
584static void __init reserve_crashkernel(void)
585{
586}
587#endif
588
589static struct resource standard_io_resources[] = {
590 { .name = "dma1", .start = 0x00, .end = 0x1f,
591 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
592 { .name = "pic1", .start = 0x20, .end = 0x21,
593 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
594 { .name = "timer0", .start = 0x40, .end = 0x43,
595 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
596 { .name = "timer1", .start = 0x50, .end = 0x53,
597 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
598 { .name = "keyboard", .start = 0x60, .end = 0x60,
599 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
600 { .name = "keyboard", .start = 0x64, .end = 0x64,
601 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
602 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
603 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
604 { .name = "pic2", .start = 0xa0, .end = 0xa1,
605 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
606 { .name = "dma2", .start = 0xc0, .end = 0xdf,
607 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
608 { .name = "fpu", .start = 0xf0, .end = 0xff,
609 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
610};
611
612void __init reserve_standard_io_resources(void)
613{
614 int i;
615
616 /* request I/O space for devices used on all i[345]86 PCs */
617 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
618 request_resource(&ioport_resource, &standard_io_resources[i]);
619
620}
621
622static __init void reserve_ibft_region(void)
623{
624 unsigned long addr, size = 0;
625
626 addr = find_ibft_region(&size);
627
628 if (size)
629 memblock_x86_reserve_range(addr, addr + size, "* ibft");
630}
631
632static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
633
634static void __init trim_bios_range(void)
635{
636 /*
637 * A special case is the first 4Kb of memory;
638 * This is a BIOS owned area, not kernel ram, but generally
639 * not listed as such in the E820 table.
640 *
641 * This typically reserves additional memory (64KiB by default)
642 * since some BIOSes are known to corrupt low memory. See the
643 * Kconfig help text for X86_RESERVE_LOW.
644 */
645 e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE),
646 E820_RAM, E820_RESERVED);
647
648 /*
649 * special case: Some BIOSen report the PC BIOS
650 * area (640->1Mb) as ram even though it is not.
651 * take them out.
652 */
653 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
654 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
655}
656
657static int __init parse_reservelow(char *p)
658{
659 unsigned long long size;
660
661 if (!p)
662 return -EINVAL;
663
664 size = memparse(p, &p);
665
666 if (size < 4096)
667 size = 4096;
668
669 if (size > 640*1024)
670 size = 640*1024;
671
672 reserve_low = size;
673
674 return 0;
675}
676
677early_param("reservelow", parse_reservelow);
678
679/*
680 * Determine if we were loaded by an EFI loader. If so, then we have also been
681 * passed the efi memmap, systab, etc., so we should use these data structures
682 * for initialization. Note, the efi init code path is determined by the
683 * global efi_enabled. This allows the same kernel image to be used on existing
684 * systems (with a traditional BIOS) as well as on EFI systems.
685 */
686/*
687 * setup_arch - architecture-specific boot-time initializations
688 *
689 * Note: On x86_64, fixmaps are ready for use even before this is called.
690 */
691
692void __init setup_arch(char **cmdline_p)
693{
694#ifdef CONFIG_X86_32
695 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
696 visws_early_detect();
697
698 /*
699 * copy kernel address range established so far and switch
700 * to the proper swapper page table
701 */
702 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
703 initial_page_table + KERNEL_PGD_BOUNDARY,
704 KERNEL_PGD_PTRS);
705
706 load_cr3(swapper_pg_dir);
707 __flush_tlb_all();
708#else
709 printk(KERN_INFO "Command line: %s\n", boot_command_line);
710#endif
711
712 /*
713 * If we have OLPC OFW, we might end up relocating the fixmap due to
714 * reserve_top(), so do this before touching the ioremap area.
715 */
716 olpc_ofw_detect();
717
718 early_trap_init();
719 early_cpu_init();
720 early_ioremap_init();
721
722 setup_olpc_ofw_pgd();
723
724 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
725 screen_info = boot_params.screen_info;
726 edid_info = boot_params.edid_info;
727#ifdef CONFIG_X86_32
728 apm_info.bios = boot_params.apm_bios_info;
729 ist_info = boot_params.ist_info;
730 if (boot_params.sys_desc_table.length != 0) {
731 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
732 machine_id = boot_params.sys_desc_table.table[0];
733 machine_submodel_id = boot_params.sys_desc_table.table[1];
734 BIOS_revision = boot_params.sys_desc_table.table[2];
735 }
736#endif
737 saved_video_mode = boot_params.hdr.vid_mode;
738 bootloader_type = boot_params.hdr.type_of_loader;
739 if ((bootloader_type >> 4) == 0xe) {
740 bootloader_type &= 0xf;
741 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
742 }
743 bootloader_version = bootloader_type & 0xf;
744 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
745
746#ifdef CONFIG_BLK_DEV_RAM
747 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
748 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
749 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
750#endif
751#ifdef CONFIG_EFI
752 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
753#ifdef CONFIG_X86_32
754 "EL32",
755#else
756 "EL64",
757#endif
758 4)) {
759 efi_enabled = 1;
760 efi_memblock_x86_reserve_range();
761 }
762#endif
763
764 x86_init.oem.arch_setup();
765
766 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
767 setup_memory_map();
768 parse_setup_data();
769 /* update the e820_saved too */
770 e820_reserve_setup_data();
771
772 copy_edd();
773
774 if (!boot_params.hdr.root_flags)
775 root_mountflags &= ~MS_RDONLY;
776 init_mm.start_code = (unsigned long) _text;
777 init_mm.end_code = (unsigned long) _etext;
778 init_mm.end_data = (unsigned long) _edata;
779 init_mm.brk = _brk_end;
780
781 code_resource.start = virt_to_phys(_text);
782 code_resource.end = virt_to_phys(_etext)-1;
783 data_resource.start = virt_to_phys(_etext);
784 data_resource.end = virt_to_phys(_edata)-1;
785 bss_resource.start = virt_to_phys(&__bss_start);
786 bss_resource.end = virt_to_phys(&__bss_stop)-1;
787
788#ifdef CONFIG_CMDLINE_BOOL
789#ifdef CONFIG_CMDLINE_OVERRIDE
790 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
791#else
792 if (builtin_cmdline[0]) {
793 /* append boot loader cmdline to builtin */
794 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
795 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
796 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
797 }
798#endif
799#endif
800
801 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
802 *cmdline_p = command_line;
803
804 /*
805 * x86_configure_nx() is called before parse_early_param() to detect
806 * whether hardware doesn't support NX (so that the early EHCI debug
807 * console setup can safely call set_fixmap()). It may then be called
808 * again from within noexec_setup() during parsing early parameters
809 * to honor the respective command line option.
810 */
811 x86_configure_nx();
812
813 parse_early_param();
814
815 x86_report_nx();
816
817 /* after early param, so could get panic from serial */
818 memblock_x86_reserve_range_setup_data();
819
820 if (acpi_mps_check()) {
821#ifdef CONFIG_X86_LOCAL_APIC
822 disable_apic = 1;
823#endif
824 setup_clear_cpu_cap(X86_FEATURE_APIC);
825 }
826
827#ifdef CONFIG_PCI
828 if (pci_early_dump_regs)
829 early_dump_pci_devices();
830#endif
831
832 finish_e820_parsing();
833
834 if (efi_enabled)
835 efi_init();
836
837 dmi_scan_machine();
838
839 /*
840 * VMware detection requires dmi to be available, so this
841 * needs to be done after dmi_scan_machine, for the BP.
842 */
843 init_hypervisor_platform();
844
845 x86_init.resources.probe_roms();
846
847 /* after parse_early_param, so could debug it */
848 insert_resource(&iomem_resource, &code_resource);
849 insert_resource(&iomem_resource, &data_resource);
850 insert_resource(&iomem_resource, &bss_resource);
851
852 trim_bios_range();
853#ifdef CONFIG_X86_32
854 if (ppro_with_ram_bug()) {
855 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
856 E820_RESERVED);
857 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
858 printk(KERN_INFO "fixed physical RAM map:\n");
859 e820_print_map("bad_ppro");
860 }
861#else
862 early_gart_iommu_check();
863#endif
864
865 /*
866 * partially used pages are not usable - thus
867 * we are rounding upwards:
868 */
869 max_pfn = e820_end_of_ram_pfn();
870
871 /* update e820 for memory not covered by WB MTRRs */
872 mtrr_bp_init();
873 if (mtrr_trim_uncached_memory(max_pfn))
874 max_pfn = e820_end_of_ram_pfn();
875
876#ifdef CONFIG_X86_32
877 /* max_low_pfn get updated here */
878 find_low_pfn_range();
879#else
880 num_physpages = max_pfn;
881
882 check_x2apic();
883
884 /* How many end-of-memory variables you have, grandma! */
885 /* need this before calling reserve_initrd */
886 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
887 max_low_pfn = e820_end_of_low_ram_pfn();
888 else
889 max_low_pfn = max_pfn;
890
891 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
892#endif
893
894 /*
895 * Find and reserve possible boot-time SMP configuration:
896 */
897 find_smp_config();
898
899 reserve_ibft_region();
900
901 /*
902 * Need to conclude brk, before memblock_x86_fill()
903 * it could use memblock_find_in_range, could overlap with
904 * brk area.
905 */
906 reserve_brk();
907
908 cleanup_highmap();
909
910 memblock.current_limit = get_max_mapped();
911 memblock_x86_fill();
912
913 /*
914 * The EFI specification says that boot service code won't be called
915 * after ExitBootServices(). This is, in fact, a lie.
916 */
917 if (efi_enabled)
918 efi_reserve_boot_services();
919
920 /* preallocate 4k for mptable mpc */
921 early_reserve_e820_mpc_new();
922
923#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
924 setup_bios_corruption_check();
925#endif
926
927 printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
928 max_pfn_mapped<<PAGE_SHIFT);
929
930 setup_trampolines();
931
932 init_gbpages();
933
934 /* max_pfn_mapped is updated here */
935 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
936 max_pfn_mapped = max_low_pfn_mapped;
937
938#ifdef CONFIG_X86_64
939 if (max_pfn > max_low_pfn) {
940 max_pfn_mapped = init_memory_mapping(1UL<<32,
941 max_pfn<<PAGE_SHIFT);
942 /* can we preseve max_low_pfn ?*/
943 max_low_pfn = max_pfn;
944 }
945#endif
946 memblock.current_limit = get_max_mapped();
947
948 /*
949 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
950 */
951
952#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
953 if (init_ohci1394_dma_early)
954 init_ohci1394_dma_on_all_controllers();
955#endif
956 /* Allocate bigger log buffer */
957 setup_log_buf(1);
958
959 reserve_initrd();
960
961 reserve_crashkernel();
962
963 vsmp_init();
964
965 io_delay_init();
966
967 /*
968 * Parse the ACPI tables for possible boot-time SMP configuration.
969 */
970 acpi_boot_table_init();
971
972 early_acpi_boot_init();
973
974 initmem_init();
975 memblock_find_dma_reserve();
976
977#ifdef CONFIG_KVM_CLOCK
978 kvmclock_init();
979#endif
980
981 x86_init.paging.pagetable_setup_start(swapper_pg_dir);
982 paging_init();
983 x86_init.paging.pagetable_setup_done(swapper_pg_dir);
984
985 if (boot_cpu_data.cpuid_level >= 0) {
986 /* A CPU has %cr4 if and only if it has CPUID */
987 mmu_cr4_features = read_cr4();
988 }
989
990#ifdef CONFIG_X86_32
991 /* sync back kernel address range */
992 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
993 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
994 KERNEL_PGD_PTRS);
995#endif
996
997 tboot_probe();
998
999#ifdef CONFIG_X86_64
1000 map_vsyscall();
1001#endif
1002
1003 generic_apic_probe();
1004
1005 early_quirks();
1006
1007 /*
1008 * Read APIC and some other early information from ACPI tables.
1009 */
1010 acpi_boot_init();
1011 sfi_init();
1012 x86_dtb_init();
1013
1014 /*
1015 * get boot-time SMP configuration:
1016 */
1017 if (smp_found_config)
1018 get_smp_config();
1019
1020 prefill_possible_map();
1021
1022 init_cpu_to_node();
1023
1024 init_apic_mappings();
1025 ioapic_and_gsi_init();
1026
1027 kvm_guest_init();
1028
1029 e820_reserve_resources();
1030 e820_mark_nosave_regions(max_low_pfn);
1031
1032 x86_init.resources.reserve_resources();
1033
1034 e820_setup_gap();
1035
1036#ifdef CONFIG_VT
1037#if defined(CONFIG_VGA_CONSOLE)
1038 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1039 conswitchp = &vga_con;
1040#elif defined(CONFIG_DUMMY_CONSOLE)
1041 conswitchp = &dummy_con;
1042#endif
1043#endif
1044 x86_init.oem.banner();
1045
1046 x86_init.timers.wallclock_init();
1047
1048 mcheck_init();
1049
1050 arch_init_ideal_nops();
1051}
1052
1053#ifdef CONFIG_X86_32
1054
1055static struct resource video_ram_resource = {
1056 .name = "Video RAM area",
1057 .start = 0xa0000,
1058 .end = 0xbffff,
1059 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1060};
1061
1062void __init i386_reserve_resources(void)
1063{
1064 request_resource(&iomem_resource, &video_ram_resource);
1065 reserve_standard_io_resources();
1066}
1067
1068#endif /* CONFIG_X86_32 */