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#include <linux/jiffies.h>
72
73#include <video/edid.h>
74
75#include <asm/mtrr.h>
76#include <asm/apic.h>
77#include <asm/realmode.h>
78#include <asm/e820.h>
79#include <asm/mpspec.h>
80#include <asm/setup.h>
81#include <asm/efi.h>
82#include <asm/timer.h>
83#include <asm/i8259.h>
84#include <asm/sections.h>
85#include <asm/io_apic.h>
86#include <asm/ist.h>
87#include <asm/setup_arch.h>
88#include <asm/bios_ebda.h>
89#include <asm/cacheflush.h>
90#include <asm/processor.h>
91#include <asm/bugs.h>
92#include <asm/kasan.h>
93
94#include <asm/vsyscall.h>
95#include <asm/cpu.h>
96#include <asm/desc.h>
97#include <asm/dma.h>
98#include <asm/iommu.h>
99#include <asm/gart.h>
100#include <asm/mmu_context.h>
101#include <asm/proto.h>
102
103#include <asm/paravirt.h>
104#include <asm/hypervisor.h>
105#include <asm/olpc_ofw.h>
106
107#include <asm/percpu.h>
108#include <asm/topology.h>
109#include <asm/apicdef.h>
110#include <asm/amd_nb.h>
111#include <asm/mce.h>
112#include <asm/alternative.h>
113#include <asm/prom.h>
114#include <asm/microcode.h>
115#include <asm/mmu_context.h>
116
117/*
118 * max_low_pfn_mapped: highest direct mapped pfn under 4GB
119 * max_pfn_mapped: highest direct mapped pfn over 4GB
120 *
121 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
122 * represented by pfn_mapped
123 */
124unsigned long max_low_pfn_mapped;
125unsigned long max_pfn_mapped;
126
127#ifdef CONFIG_DMI
128RESERVE_BRK(dmi_alloc, 65536);
129#endif
130
131
132static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
133unsigned long _brk_end = (unsigned long)__brk_base;
134
135#ifdef CONFIG_X86_64
136int default_cpu_present_to_apicid(int mps_cpu)
137{
138 return __default_cpu_present_to_apicid(mps_cpu);
139}
140
141int default_check_phys_apicid_present(int phys_apicid)
142{
143 return __default_check_phys_apicid_present(phys_apicid);
144}
145#endif
146
147struct boot_params boot_params;
148
149/*
150 * Machine setup..
151 */
152static struct resource data_resource = {
153 .name = "Kernel data",
154 .start = 0,
155 .end = 0,
156 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
157};
158
159static struct resource code_resource = {
160 .name = "Kernel code",
161 .start = 0,
162 .end = 0,
163 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
164};
165
166static struct resource bss_resource = {
167 .name = "Kernel bss",
168 .start = 0,
169 .end = 0,
170 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
171};
172
173
174#ifdef CONFIG_X86_32
175/* cpu data as detected by the assembly code in head.S */
176struct cpuinfo_x86 new_cpu_data = {
177 .wp_works_ok = -1,
178};
179/* common cpu data for all cpus */
180struct cpuinfo_x86 boot_cpu_data __read_mostly = {
181 .wp_works_ok = -1,
182};
183EXPORT_SYMBOL(boot_cpu_data);
184
185unsigned int def_to_bigsmp;
186
187/* for MCA, but anyone else can use it if they want */
188unsigned int machine_id;
189unsigned int machine_submodel_id;
190unsigned int BIOS_revision;
191
192struct apm_info apm_info;
193EXPORT_SYMBOL(apm_info);
194
195#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
196 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
197struct ist_info ist_info;
198EXPORT_SYMBOL(ist_info);
199#else
200struct ist_info ist_info;
201#endif
202
203#else
204struct cpuinfo_x86 boot_cpu_data __read_mostly = {
205 .x86_phys_bits = MAX_PHYSMEM_BITS,
206};
207EXPORT_SYMBOL(boot_cpu_data);
208#endif
209
210
211#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
212__visible unsigned long mmu_cr4_features;
213#else
214__visible unsigned long mmu_cr4_features = X86_CR4_PAE;
215#endif
216
217/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
218int bootloader_type, bootloader_version;
219
220/*
221 * Setup options
222 */
223struct screen_info screen_info;
224EXPORT_SYMBOL(screen_info);
225struct edid_info edid_info;
226EXPORT_SYMBOL_GPL(edid_info);
227
228extern int root_mountflags;
229
230unsigned long saved_video_mode;
231
232#define RAMDISK_IMAGE_START_MASK 0x07FF
233#define RAMDISK_PROMPT_FLAG 0x8000
234#define RAMDISK_LOAD_FLAG 0x4000
235
236static char __initdata command_line[COMMAND_LINE_SIZE];
237#ifdef CONFIG_CMDLINE_BOOL
238static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
239#endif
240
241#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
242struct edd edd;
243#ifdef CONFIG_EDD_MODULE
244EXPORT_SYMBOL(edd);
245#endif
246/**
247 * copy_edd() - Copy the BIOS EDD information
248 * from boot_params into a safe place.
249 *
250 */
251static inline void __init copy_edd(void)
252{
253 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
254 sizeof(edd.mbr_signature));
255 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
256 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
257 edd.edd_info_nr = boot_params.eddbuf_entries;
258}
259#else
260static inline void __init copy_edd(void)
261{
262}
263#endif
264
265void * __init extend_brk(size_t size, size_t align)
266{
267 size_t mask = align - 1;
268 void *ret;
269
270 BUG_ON(_brk_start == 0);
271 BUG_ON(align & mask);
272
273 _brk_end = (_brk_end + mask) & ~mask;
274 BUG_ON((char *)(_brk_end + size) > __brk_limit);
275
276 ret = (void *)_brk_end;
277 _brk_end += size;
278
279 memset(ret, 0, size);
280
281 return ret;
282}
283
284#ifdef CONFIG_X86_32
285static void __init cleanup_highmap(void)
286{
287}
288#endif
289
290static void __init reserve_brk(void)
291{
292 if (_brk_end > _brk_start)
293 memblock_reserve(__pa_symbol(_brk_start),
294 _brk_end - _brk_start);
295
296 /* Mark brk area as locked down and no longer taking any
297 new allocations */
298 _brk_start = 0;
299}
300
301u64 relocated_ramdisk;
302
303#ifdef CONFIG_BLK_DEV_INITRD
304
305static u64 __init get_ramdisk_image(void)
306{
307 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
308
309 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
310
311 return ramdisk_image;
312}
313static u64 __init get_ramdisk_size(void)
314{
315 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
316
317 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
318
319 return ramdisk_size;
320}
321
322static void __init relocate_initrd(void)
323{
324 /* Assume only end is not page aligned */
325 u64 ramdisk_image = get_ramdisk_image();
326 u64 ramdisk_size = get_ramdisk_size();
327 u64 area_size = PAGE_ALIGN(ramdisk_size);
328
329 /* We need to move the initrd down into directly mapped mem */
330 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
331 area_size, PAGE_SIZE);
332
333 if (!relocated_ramdisk)
334 panic("Cannot find place for new RAMDISK of size %lld\n",
335 ramdisk_size);
336
337 /* Note: this includes all the mem currently occupied by
338 the initrd, we rely on that fact to keep the data intact. */
339 memblock_reserve(relocated_ramdisk, area_size);
340 initrd_start = relocated_ramdisk + PAGE_OFFSET;
341 initrd_end = initrd_start + ramdisk_size;
342 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
343 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
344
345 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
346
347 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
348 " [mem %#010llx-%#010llx]\n",
349 ramdisk_image, ramdisk_image + ramdisk_size - 1,
350 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
351}
352
353static void __init early_reserve_initrd(void)
354{
355 /* Assume only end is not page aligned */
356 u64 ramdisk_image = get_ramdisk_image();
357 u64 ramdisk_size = get_ramdisk_size();
358 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
359
360 if (!boot_params.hdr.type_of_loader ||
361 !ramdisk_image || !ramdisk_size)
362 return; /* No initrd provided by bootloader */
363
364 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
365}
366static void __init reserve_initrd(void)
367{
368 /* Assume only end is not page aligned */
369 u64 ramdisk_image = get_ramdisk_image();
370 u64 ramdisk_size = get_ramdisk_size();
371 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
372 u64 mapped_size;
373
374 if (!boot_params.hdr.type_of_loader ||
375 !ramdisk_image || !ramdisk_size)
376 return; /* No initrd provided by bootloader */
377
378 initrd_start = 0;
379
380 mapped_size = memblock_mem_size(max_pfn_mapped);
381 if (ramdisk_size >= (mapped_size>>1))
382 panic("initrd too large to handle, "
383 "disabling initrd (%lld needed, %lld available)\n",
384 ramdisk_size, mapped_size>>1);
385
386 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
387 ramdisk_end - 1);
388
389 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
390 PFN_DOWN(ramdisk_end))) {
391 /* All are mapped, easy case */
392 initrd_start = ramdisk_image + PAGE_OFFSET;
393 initrd_end = initrd_start + ramdisk_size;
394 return;
395 }
396
397 relocate_initrd();
398
399 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
400}
401#else
402static void __init early_reserve_initrd(void)
403{
404}
405static void __init reserve_initrd(void)
406{
407}
408#endif /* CONFIG_BLK_DEV_INITRD */
409
410static void __init parse_setup_data(void)
411{
412 struct setup_data *data;
413 u64 pa_data, pa_next;
414
415 pa_data = boot_params.hdr.setup_data;
416 while (pa_data) {
417 u32 data_len, data_type;
418
419 data = early_memremap(pa_data, sizeof(*data));
420 data_len = data->len + sizeof(struct setup_data);
421 data_type = data->type;
422 pa_next = data->next;
423 early_memunmap(data, sizeof(*data));
424
425 switch (data_type) {
426 case SETUP_E820_EXT:
427 parse_e820_ext(pa_data, data_len);
428 break;
429 case SETUP_DTB:
430 add_dtb(pa_data);
431 break;
432 case SETUP_EFI:
433 parse_efi_setup(pa_data, data_len);
434 break;
435 default:
436 break;
437 }
438 pa_data = pa_next;
439 }
440}
441
442static void __init e820_reserve_setup_data(void)
443{
444 struct setup_data *data;
445 u64 pa_data;
446
447 pa_data = boot_params.hdr.setup_data;
448 if (!pa_data)
449 return;
450
451 while (pa_data) {
452 data = early_memremap(pa_data, sizeof(*data));
453 e820_update_range(pa_data, sizeof(*data)+data->len,
454 E820_RAM, E820_RESERVED_KERN);
455 pa_data = data->next;
456 early_memunmap(data, sizeof(*data));
457 }
458
459 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
460 memcpy(&e820_saved, &e820, sizeof(struct e820map));
461 printk(KERN_INFO "extended physical RAM map:\n");
462 e820_print_map("reserve setup_data");
463}
464
465static void __init memblock_x86_reserve_range_setup_data(void)
466{
467 struct setup_data *data;
468 u64 pa_data;
469
470 pa_data = boot_params.hdr.setup_data;
471 while (pa_data) {
472 data = early_memremap(pa_data, sizeof(*data));
473 memblock_reserve(pa_data, sizeof(*data) + data->len);
474 pa_data = data->next;
475 early_memunmap(data, sizeof(*data));
476 }
477}
478
479/*
480 * --------- Crashkernel reservation ------------------------------
481 */
482
483#ifdef CONFIG_KEXEC_CORE
484
485/* 16M alignment for crash kernel regions */
486#define CRASH_ALIGN (16 << 20)
487
488/*
489 * Keep the crash kernel below this limit. On 32 bits earlier kernels
490 * would limit the kernel to the low 512 MiB due to mapping restrictions.
491 * On 64bit, old kexec-tools need to under 896MiB.
492 */
493#ifdef CONFIG_X86_32
494# define CRASH_ADDR_LOW_MAX (512 << 20)
495# define CRASH_ADDR_HIGH_MAX (512 << 20)
496#else
497# define CRASH_ADDR_LOW_MAX (896UL << 20)
498# define CRASH_ADDR_HIGH_MAX MAXMEM
499#endif
500
501static int __init reserve_crashkernel_low(void)
502{
503#ifdef CONFIG_X86_64
504 unsigned long long base, low_base = 0, low_size = 0;
505 unsigned long total_low_mem;
506 int ret;
507
508 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
509
510 /* crashkernel=Y,low */
511 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
512 if (ret) {
513 /*
514 * two parts from lib/swiotlb.c:
515 * -swiotlb size: user-specified with swiotlb= or default.
516 *
517 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
518 * to 8M for other buffers that may need to stay low too. Also
519 * make sure we allocate enough extra low memory so that we
520 * don't run out of DMA buffers for 32-bit devices.
521 */
522 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
523 } else {
524 /* passed with crashkernel=0,low ? */
525 if (!low_size)
526 return 0;
527 }
528
529 low_base = memblock_find_in_range(low_size, 1ULL << 32, low_size, CRASH_ALIGN);
530 if (!low_base) {
531 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
532 (unsigned long)(low_size >> 20));
533 return -ENOMEM;
534 }
535
536 ret = memblock_reserve(low_base, low_size);
537 if (ret) {
538 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
539 return ret;
540 }
541
542 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
543 (unsigned long)(low_size >> 20),
544 (unsigned long)(low_base >> 20),
545 (unsigned long)(total_low_mem >> 20));
546
547 crashk_low_res.start = low_base;
548 crashk_low_res.end = low_base + low_size - 1;
549 insert_resource(&iomem_resource, &crashk_low_res);
550#endif
551 return 0;
552}
553
554static void __init reserve_crashkernel(void)
555{
556 unsigned long long crash_size, crash_base, total_mem;
557 bool high = false;
558 int ret;
559
560 total_mem = memblock_phys_mem_size();
561
562 /* crashkernel=XM */
563 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
564 if (ret != 0 || crash_size <= 0) {
565 /* crashkernel=X,high */
566 ret = parse_crashkernel_high(boot_command_line, total_mem,
567 &crash_size, &crash_base);
568 if (ret != 0 || crash_size <= 0)
569 return;
570 high = true;
571 }
572
573 /* 0 means: find the address automatically */
574 if (crash_base <= 0) {
575 /*
576 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
577 */
578 crash_base = memblock_find_in_range(CRASH_ALIGN,
579 high ? CRASH_ADDR_HIGH_MAX
580 : CRASH_ADDR_LOW_MAX,
581 crash_size, CRASH_ALIGN);
582 if (!crash_base) {
583 pr_info("crashkernel reservation failed - No suitable area found.\n");
584 return;
585 }
586
587 } else {
588 unsigned long long start;
589
590 start = memblock_find_in_range(crash_base,
591 crash_base + crash_size,
592 crash_size, 1 << 20);
593 if (start != crash_base) {
594 pr_info("crashkernel reservation failed - memory is in use.\n");
595 return;
596 }
597 }
598 ret = memblock_reserve(crash_base, crash_size);
599 if (ret) {
600 pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
601 return;
602 }
603
604 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
605 memblock_free(crash_base, crash_size);
606 return;
607 }
608
609 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
610 (unsigned long)(crash_size >> 20),
611 (unsigned long)(crash_base >> 20),
612 (unsigned long)(total_mem >> 20));
613
614 crashk_res.start = crash_base;
615 crashk_res.end = crash_base + crash_size - 1;
616 insert_resource(&iomem_resource, &crashk_res);
617}
618#else
619static void __init reserve_crashkernel(void)
620{
621}
622#endif
623
624static struct resource standard_io_resources[] = {
625 { .name = "dma1", .start = 0x00, .end = 0x1f,
626 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
627 { .name = "pic1", .start = 0x20, .end = 0x21,
628 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
629 { .name = "timer0", .start = 0x40, .end = 0x43,
630 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
631 { .name = "timer1", .start = 0x50, .end = 0x53,
632 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
633 { .name = "keyboard", .start = 0x60, .end = 0x60,
634 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
635 { .name = "keyboard", .start = 0x64, .end = 0x64,
636 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
637 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
638 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
639 { .name = "pic2", .start = 0xa0, .end = 0xa1,
640 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
641 { .name = "dma2", .start = 0xc0, .end = 0xdf,
642 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
643 { .name = "fpu", .start = 0xf0, .end = 0xff,
644 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
645};
646
647void __init reserve_standard_io_resources(void)
648{
649 int i;
650
651 /* request I/O space for devices used on all i[345]86 PCs */
652 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
653 request_resource(&ioport_resource, &standard_io_resources[i]);
654
655}
656
657static __init void reserve_ibft_region(void)
658{
659 unsigned long addr, size = 0;
660
661 addr = find_ibft_region(&size);
662
663 if (size)
664 memblock_reserve(addr, size);
665}
666
667static bool __init snb_gfx_workaround_needed(void)
668{
669#ifdef CONFIG_PCI
670 int i;
671 u16 vendor, devid;
672 static const __initconst u16 snb_ids[] = {
673 0x0102,
674 0x0112,
675 0x0122,
676 0x0106,
677 0x0116,
678 0x0126,
679 0x010a,
680 };
681
682 /* Assume no if something weird is going on with PCI */
683 if (!early_pci_allowed())
684 return false;
685
686 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
687 if (vendor != 0x8086)
688 return false;
689
690 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
691 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
692 if (devid == snb_ids[i])
693 return true;
694#endif
695
696 return false;
697}
698
699/*
700 * Sandy Bridge graphics has trouble with certain ranges, exclude
701 * them from allocation.
702 */
703static void __init trim_snb_memory(void)
704{
705 static const __initconst unsigned long bad_pages[] = {
706 0x20050000,
707 0x20110000,
708 0x20130000,
709 0x20138000,
710 0x40004000,
711 };
712 int i;
713
714 if (!snb_gfx_workaround_needed())
715 return;
716
717 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
718
719 /*
720 * Reserve all memory below the 1 MB mark that has not
721 * already been reserved.
722 */
723 memblock_reserve(0, 1<<20);
724
725 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
726 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
727 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
728 bad_pages[i]);
729 }
730}
731
732/*
733 * Here we put platform-specific memory range workarounds, i.e.
734 * memory known to be corrupt or otherwise in need to be reserved on
735 * specific platforms.
736 *
737 * If this gets used more widely it could use a real dispatch mechanism.
738 */
739static void __init trim_platform_memory_ranges(void)
740{
741 trim_snb_memory();
742}
743
744static void __init trim_bios_range(void)
745{
746 /*
747 * A special case is the first 4Kb of memory;
748 * This is a BIOS owned area, not kernel ram, but generally
749 * not listed as such in the E820 table.
750 *
751 * This typically reserves additional memory (64KiB by default)
752 * since some BIOSes are known to corrupt low memory. See the
753 * Kconfig help text for X86_RESERVE_LOW.
754 */
755 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
756
757 /*
758 * special case: Some BIOSen report the PC BIOS
759 * area (640->1Mb) as ram even though it is not.
760 * take them out.
761 */
762 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
763
764 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
765}
766
767/* called before trim_bios_range() to spare extra sanitize */
768static void __init e820_add_kernel_range(void)
769{
770 u64 start = __pa_symbol(_text);
771 u64 size = __pa_symbol(_end) - start;
772
773 /*
774 * Complain if .text .data and .bss are not marked as E820_RAM and
775 * attempt to fix it by adding the range. We may have a confused BIOS,
776 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
777 * exclude kernel range. If we really are running on top non-RAM,
778 * we will crash later anyways.
779 */
780 if (e820_all_mapped(start, start + size, E820_RAM))
781 return;
782
783 pr_warn(".text .data .bss are not marked as E820_RAM!\n");
784 e820_remove_range(start, size, E820_RAM, 0);
785 e820_add_region(start, size, E820_RAM);
786}
787
788static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
789
790static int __init parse_reservelow(char *p)
791{
792 unsigned long long size;
793
794 if (!p)
795 return -EINVAL;
796
797 size = memparse(p, &p);
798
799 if (size < 4096)
800 size = 4096;
801
802 if (size > 640*1024)
803 size = 640*1024;
804
805 reserve_low = size;
806
807 return 0;
808}
809
810early_param("reservelow", parse_reservelow);
811
812static void __init trim_low_memory_range(void)
813{
814 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
815}
816
817/*
818 * Dump out kernel offset information on panic.
819 */
820static int
821dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
822{
823 if (kaslr_enabled()) {
824 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
825 kaslr_offset(),
826 __START_KERNEL,
827 __START_KERNEL_map,
828 MODULES_VADDR-1);
829 } else {
830 pr_emerg("Kernel Offset: disabled\n");
831 }
832
833 return 0;
834}
835
836/*
837 * Determine if we were loaded by an EFI loader. If so, then we have also been
838 * passed the efi memmap, systab, etc., so we should use these data structures
839 * for initialization. Note, the efi init code path is determined by the
840 * global efi_enabled. This allows the same kernel image to be used on existing
841 * systems (with a traditional BIOS) as well as on EFI systems.
842 */
843/*
844 * setup_arch - architecture-specific boot-time initializations
845 *
846 * Note: On x86_64, fixmaps are ready for use even before this is called.
847 */
848
849void __init setup_arch(char **cmdline_p)
850{
851 memblock_reserve(__pa_symbol(_text),
852 (unsigned long)__bss_stop - (unsigned long)_text);
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#endif
887
888 /*
889 * If we have OLPC OFW, we might end up relocating the fixmap due to
890 * reserve_top(), so do this before touching the ioremap area.
891 */
892 olpc_ofw_detect();
893
894 early_trap_init();
895 early_cpu_init();
896 early_ioremap_init();
897
898 setup_olpc_ofw_pgd();
899
900 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
901 screen_info = boot_params.screen_info;
902 edid_info = boot_params.edid_info;
903#ifdef CONFIG_X86_32
904 apm_info.bios = boot_params.apm_bios_info;
905 ist_info = boot_params.ist_info;
906#endif
907 saved_video_mode = boot_params.hdr.vid_mode;
908 bootloader_type = boot_params.hdr.type_of_loader;
909 if ((bootloader_type >> 4) == 0xe) {
910 bootloader_type &= 0xf;
911 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
912 }
913 bootloader_version = bootloader_type & 0xf;
914 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
915
916#ifdef CONFIG_BLK_DEV_RAM
917 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
918 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
919 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
920#endif
921#ifdef CONFIG_EFI
922 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
923 EFI32_LOADER_SIGNATURE, 4)) {
924 set_bit(EFI_BOOT, &efi.flags);
925 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
926 EFI64_LOADER_SIGNATURE, 4)) {
927 set_bit(EFI_BOOT, &efi.flags);
928 set_bit(EFI_64BIT, &efi.flags);
929 }
930
931 if (efi_enabled(EFI_BOOT))
932 efi_memblock_x86_reserve_range();
933#endif
934
935 x86_init.oem.arch_setup();
936
937 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
938 setup_memory_map();
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 x86_report_nx();
987
988 /* after early param, so could get panic from serial */
989 memblock_x86_reserve_range_setup_data();
990
991 if (acpi_mps_check()) {
992#ifdef CONFIG_X86_LOCAL_APIC
993 disable_apic = 1;
994#endif
995 setup_clear_cpu_cap(X86_FEATURE_APIC);
996 }
997
998#ifdef CONFIG_PCI
999 if (pci_early_dump_regs)
1000 early_dump_pci_devices();
1001#endif
1002
1003 /* update the e820_saved too */
1004 e820_reserve_setup_data();
1005 finish_e820_parsing();
1006
1007 if (efi_enabled(EFI_BOOT))
1008 efi_init();
1009
1010 dmi_scan_machine();
1011 dmi_memdev_walk();
1012 dmi_set_dump_stack_arch_desc();
1013
1014 /*
1015 * VMware detection requires dmi to be available, so this
1016 * needs to be done after dmi_scan_machine, for the BP.
1017 */
1018 init_hypervisor_platform();
1019
1020 x86_init.resources.probe_roms();
1021
1022 /* after parse_early_param, so could debug it */
1023 insert_resource(&iomem_resource, &code_resource);
1024 insert_resource(&iomem_resource, &data_resource);
1025 insert_resource(&iomem_resource, &bss_resource);
1026
1027 e820_add_kernel_range();
1028 trim_bios_range();
1029#ifdef CONFIG_X86_32
1030 if (ppro_with_ram_bug()) {
1031 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1032 E820_RESERVED);
1033 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1034 printk(KERN_INFO "fixed physical RAM map:\n");
1035 e820_print_map("bad_ppro");
1036 }
1037#else
1038 early_gart_iommu_check();
1039#endif
1040
1041 /*
1042 * partially used pages are not usable - thus
1043 * we are rounding upwards:
1044 */
1045 max_pfn = e820_end_of_ram_pfn();
1046
1047 /* update e820 for memory not covered by WB MTRRs */
1048 mtrr_bp_init();
1049 if (mtrr_trim_uncached_memory(max_pfn))
1050 max_pfn = e820_end_of_ram_pfn();
1051
1052 max_possible_pfn = max_pfn;
1053
1054#ifdef CONFIG_X86_32
1055 /* max_low_pfn get updated here */
1056 find_low_pfn_range();
1057#else
1058 check_x2apic();
1059
1060 /* How many end-of-memory variables you have, grandma! */
1061 /* need this before calling reserve_initrd */
1062 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1063 max_low_pfn = e820_end_of_low_ram_pfn();
1064 else
1065 max_low_pfn = max_pfn;
1066
1067 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1068#endif
1069
1070 /*
1071 * Find and reserve possible boot-time SMP configuration:
1072 */
1073 find_smp_config();
1074
1075 reserve_ibft_region();
1076
1077 early_alloc_pgt_buf();
1078
1079 /*
1080 * Need to conclude brk, before memblock_x86_fill()
1081 * it could use memblock_find_in_range, could overlap with
1082 * brk area.
1083 */
1084 reserve_brk();
1085
1086 cleanup_highmap();
1087
1088 memblock_set_current_limit(ISA_END_ADDRESS);
1089 memblock_x86_fill();
1090
1091 if (efi_enabled(EFI_BOOT)) {
1092 efi_fake_memmap();
1093 efi_find_mirror();
1094 }
1095
1096 /*
1097 * The EFI specification says that boot service code won't be called
1098 * after ExitBootServices(). This is, in fact, a lie.
1099 */
1100 if (efi_enabled(EFI_MEMMAP))
1101 efi_reserve_boot_services();
1102
1103 /* preallocate 4k for mptable mpc */
1104 early_reserve_e820_mpc_new();
1105
1106#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1107 setup_bios_corruption_check();
1108#endif
1109
1110#ifdef CONFIG_X86_32
1111 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1112 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1113#endif
1114
1115 reserve_real_mode();
1116
1117 trim_platform_memory_ranges();
1118 trim_low_memory_range();
1119
1120 init_mem_mapping();
1121
1122 early_trap_pf_init();
1123
1124 setup_real_mode();
1125
1126 memblock_set_current_limit(get_max_mapped());
1127
1128 /*
1129 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1130 */
1131
1132#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1133 if (init_ohci1394_dma_early)
1134 init_ohci1394_dma_on_all_controllers();
1135#endif
1136 /* Allocate bigger log buffer */
1137 setup_log_buf(1);
1138
1139 reserve_initrd();
1140
1141#if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1142 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1143#endif
1144
1145 vsmp_init();
1146
1147 io_delay_init();
1148
1149 /*
1150 * Parse the ACPI tables for possible boot-time SMP configuration.
1151 */
1152 acpi_boot_table_init();
1153
1154 early_acpi_boot_init();
1155
1156 initmem_init();
1157 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1158
1159 /*
1160 * Reserve memory for crash kernel after SRAT is parsed so that it
1161 * won't consume hotpluggable memory.
1162 */
1163 reserve_crashkernel();
1164
1165 memblock_find_dma_reserve();
1166
1167#ifdef CONFIG_KVM_GUEST
1168 kvmclock_init();
1169#endif
1170
1171 x86_init.paging.pagetable_init();
1172
1173 kasan_init();
1174
1175 if (boot_cpu_data.cpuid_level >= 0) {
1176 /* A CPU has %cr4 if and only if it has CPUID */
1177 mmu_cr4_features = __read_cr4();
1178 if (trampoline_cr4_features)
1179 *trampoline_cr4_features = mmu_cr4_features;
1180 }
1181
1182#ifdef CONFIG_X86_32
1183 /* sync back kernel address range */
1184 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1185 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1186 KERNEL_PGD_PTRS);
1187
1188 /*
1189 * sync back low identity map too. It is used for example
1190 * in the 32-bit EFI stub.
1191 */
1192 clone_pgd_range(initial_page_table,
1193 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1194 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1195#endif
1196
1197 tboot_probe();
1198
1199 map_vsyscall();
1200
1201 generic_apic_probe();
1202
1203 early_quirks();
1204
1205 /*
1206 * Read APIC and some other early information from ACPI tables.
1207 */
1208 acpi_boot_init();
1209 sfi_init();
1210 x86_dtb_init();
1211
1212 /*
1213 * get boot-time SMP configuration:
1214 */
1215 if (smp_found_config)
1216 get_smp_config();
1217
1218 prefill_possible_map();
1219
1220 init_cpu_to_node();
1221
1222 init_apic_mappings();
1223 io_apic_init_mappings();
1224
1225 kvm_guest_init();
1226
1227 e820_reserve_resources();
1228 e820_mark_nosave_regions(max_low_pfn);
1229
1230 x86_init.resources.reserve_resources();
1231
1232 e820_setup_gap();
1233
1234#ifdef CONFIG_VT
1235#if defined(CONFIG_VGA_CONSOLE)
1236 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1237 conswitchp = &vga_con;
1238#elif defined(CONFIG_DUMMY_CONSOLE)
1239 conswitchp = &dummy_con;
1240#endif
1241#endif
1242 x86_init.oem.banner();
1243
1244 x86_init.timers.wallclock_init();
1245
1246 mcheck_init();
1247
1248 arch_init_ideal_nops();
1249
1250 register_refined_jiffies(CLOCK_TICK_RATE);
1251
1252#ifdef CONFIG_EFI
1253 if (efi_enabled(EFI_BOOT))
1254 efi_apply_memmap_quirks();
1255#endif
1256}
1257
1258#ifdef CONFIG_X86_32
1259
1260static struct resource video_ram_resource = {
1261 .name = "Video RAM area",
1262 .start = 0xa0000,
1263 .end = 0xbffff,
1264 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1265};
1266
1267void __init i386_reserve_resources(void)
1268{
1269 request_resource(&iomem_resource, &video_ram_resource);
1270 reserve_standard_io_resources();
1271}
1272
1273#endif /* CONFIG_X86_32 */
1274
1275static struct notifier_block kernel_offset_notifier = {
1276 .notifier_call = dump_kernel_offset
1277};
1278
1279static int __init register_kernel_offset_dumper(void)
1280{
1281 atomic_notifier_chain_register(&panic_notifier_list,
1282 &kernel_offset_notifier);
1283 return 0;
1284}
1285__initcall(register_kernel_offset_dumper);
1286
1287void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
1288{
1289 if (!boot_cpu_has(X86_FEATURE_OSPKE))
1290 return;
1291
1292 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
1293}
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 *
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
8 *
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
11 *
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
14 *
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
17 *
18 */
19
20/*
21 * This file handles the architecture-dependent parts of initialization
22 */
23
24#include <linux/sched.h>
25#include <linux/mm.h>
26#include <linux/mmzone.h>
27#include <linux/screen_info.h>
28#include <linux/ioport.h>
29#include <linux/acpi.h>
30#include <linux/sfi.h>
31#include <linux/apm_bios.h>
32#include <linux/initrd.h>
33#include <linux/bootmem.h>
34#include <linux/memblock.h>
35#include <linux/seq_file.h>
36#include <linux/console.h>
37#include <linux/root_dev.h>
38#include <linux/highmem.h>
39#include <linux/module.h>
40#include <linux/efi.h>
41#include <linux/init.h>
42#include <linux/edd.h>
43#include <linux/iscsi_ibft.h>
44#include <linux/nodemask.h>
45#include <linux/kexec.h>
46#include <linux/dmi.h>
47#include <linux/pfn.h>
48#include <linux/pci.h>
49#include <asm/pci-direct.h>
50#include <linux/init_ohci1394_dma.h>
51#include <linux/kvm_para.h>
52#include <linux/dma-contiguous.h>
53
54#include <linux/errno.h>
55#include <linux/kernel.h>
56#include <linux/stddef.h>
57#include <linux/unistd.h>
58#include <linux/ptrace.h>
59#include <linux/user.h>
60#include <linux/delay.h>
61
62#include <linux/kallsyms.h>
63#include <linux/cpufreq.h>
64#include <linux/dma-mapping.h>
65#include <linux/ctype.h>
66#include <linux/uaccess.h>
67
68#include <linux/percpu.h>
69#include <linux/crash_dump.h>
70#include <linux/tboot.h>
71#include <linux/jiffies.h>
72
73#include <video/edid.h>
74
75#include <asm/mtrr.h>
76#include <asm/apic.h>
77#include <asm/realmode.h>
78#include <asm/e820.h>
79#include <asm/mpspec.h>
80#include <asm/setup.h>
81#include <asm/efi.h>
82#include <asm/timer.h>
83#include <asm/i8259.h>
84#include <asm/sections.h>
85#include <asm/io_apic.h>
86#include <asm/ist.h>
87#include <asm/setup_arch.h>
88#include <asm/bios_ebda.h>
89#include <asm/cacheflush.h>
90#include <asm/processor.h>
91#include <asm/bugs.h>
92
93#include <asm/vsyscall.h>
94#include <asm/cpu.h>
95#include <asm/desc.h>
96#include <asm/dma.h>
97#include <asm/iommu.h>
98#include <asm/gart.h>
99#include <asm/mmu_context.h>
100#include <asm/proto.h>
101
102#include <asm/paravirt.h>
103#include <asm/hypervisor.h>
104#include <asm/olpc_ofw.h>
105
106#include <asm/percpu.h>
107#include <asm/topology.h>
108#include <asm/apicdef.h>
109#include <asm/amd_nb.h>
110#include <asm/mce.h>
111#include <asm/alternative.h>
112#include <asm/prom.h>
113
114/*
115 * max_low_pfn_mapped: highest direct mapped pfn under 4GB
116 * max_pfn_mapped: highest direct mapped pfn over 4GB
117 *
118 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
119 * represented by pfn_mapped
120 */
121unsigned long max_low_pfn_mapped;
122unsigned long max_pfn_mapped;
123
124#ifdef CONFIG_DMI
125RESERVE_BRK(dmi_alloc, 65536);
126#endif
127
128
129static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
130unsigned long _brk_end = (unsigned long)__brk_base;
131
132#ifdef CONFIG_X86_64
133int default_cpu_present_to_apicid(int mps_cpu)
134{
135 return __default_cpu_present_to_apicid(mps_cpu);
136}
137
138int default_check_phys_apicid_present(int phys_apicid)
139{
140 return __default_check_phys_apicid_present(phys_apicid);
141}
142#endif
143
144struct boot_params boot_params;
145
146/*
147 * Machine setup..
148 */
149static struct resource data_resource = {
150 .name = "Kernel data",
151 .start = 0,
152 .end = 0,
153 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
154};
155
156static struct resource code_resource = {
157 .name = "Kernel code",
158 .start = 0,
159 .end = 0,
160 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
161};
162
163static struct resource bss_resource = {
164 .name = "Kernel bss",
165 .start = 0,
166 .end = 0,
167 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
168};
169
170
171#ifdef CONFIG_X86_32
172/* cpu data as detected by the assembly code in head.S */
173struct cpuinfo_x86 new_cpu_data = {
174 .wp_works_ok = -1,
175};
176/* common cpu data for all cpus */
177struct cpuinfo_x86 boot_cpu_data __read_mostly = {
178 .wp_works_ok = -1,
179};
180EXPORT_SYMBOL(boot_cpu_data);
181
182unsigned int def_to_bigsmp;
183
184/* for MCA, but anyone else can use it if they want */
185unsigned int machine_id;
186unsigned int machine_submodel_id;
187unsigned int BIOS_revision;
188
189struct apm_info apm_info;
190EXPORT_SYMBOL(apm_info);
191
192#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
193 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
194struct ist_info ist_info;
195EXPORT_SYMBOL(ist_info);
196#else
197struct ist_info ist_info;
198#endif
199
200#else
201struct cpuinfo_x86 boot_cpu_data __read_mostly = {
202 .x86_phys_bits = MAX_PHYSMEM_BITS,
203};
204EXPORT_SYMBOL(boot_cpu_data);
205#endif
206
207
208#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
209__visible unsigned long mmu_cr4_features;
210#else
211__visible unsigned long mmu_cr4_features = X86_CR4_PAE;
212#endif
213
214/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
215int bootloader_type, bootloader_version;
216
217/*
218 * Setup options
219 */
220struct screen_info screen_info;
221EXPORT_SYMBOL(screen_info);
222struct edid_info edid_info;
223EXPORT_SYMBOL_GPL(edid_info);
224
225extern int root_mountflags;
226
227unsigned long saved_video_mode;
228
229#define RAMDISK_IMAGE_START_MASK 0x07FF
230#define RAMDISK_PROMPT_FLAG 0x8000
231#define RAMDISK_LOAD_FLAG 0x4000
232
233static char __initdata command_line[COMMAND_LINE_SIZE];
234#ifdef CONFIG_CMDLINE_BOOL
235static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
236#endif
237
238#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
239struct edd edd;
240#ifdef CONFIG_EDD_MODULE
241EXPORT_SYMBOL(edd);
242#endif
243/**
244 * copy_edd() - Copy the BIOS EDD information
245 * from boot_params into a safe place.
246 *
247 */
248static inline void __init copy_edd(void)
249{
250 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
251 sizeof(edd.mbr_signature));
252 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
253 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
254 edd.edd_info_nr = boot_params.eddbuf_entries;
255}
256#else
257static inline void __init copy_edd(void)
258{
259}
260#endif
261
262void * __init extend_brk(size_t size, size_t align)
263{
264 size_t mask = align - 1;
265 void *ret;
266
267 BUG_ON(_brk_start == 0);
268 BUG_ON(align & mask);
269
270 _brk_end = (_brk_end + mask) & ~mask;
271 BUG_ON((char *)(_brk_end + size) > __brk_limit);
272
273 ret = (void *)_brk_end;
274 _brk_end += size;
275
276 memset(ret, 0, size);
277
278 return ret;
279}
280
281#ifdef CONFIG_X86_32
282static void __init cleanup_highmap(void)
283{
284}
285#endif
286
287static void __init reserve_brk(void)
288{
289 if (_brk_end > _brk_start)
290 memblock_reserve(__pa_symbol(_brk_start),
291 _brk_end - _brk_start);
292
293 /* Mark brk area as locked down and no longer taking any
294 new allocations */
295 _brk_start = 0;
296}
297
298u64 relocated_ramdisk;
299
300#ifdef CONFIG_BLK_DEV_INITRD
301
302static u64 __init get_ramdisk_image(void)
303{
304 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
305
306 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
307
308 return ramdisk_image;
309}
310static u64 __init get_ramdisk_size(void)
311{
312 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
313
314 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
315
316 return ramdisk_size;
317}
318
319#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
320static void __init relocate_initrd(void)
321{
322 /* Assume only end is not page aligned */
323 u64 ramdisk_image = get_ramdisk_image();
324 u64 ramdisk_size = get_ramdisk_size();
325 u64 area_size = PAGE_ALIGN(ramdisk_size);
326 unsigned long slop, clen, mapaddr;
327 char *p, *q;
328
329 /* We need to move the initrd down into directly mapped mem */
330 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
331 area_size, PAGE_SIZE);
332
333 if (!relocated_ramdisk)
334 panic("Cannot find place for new RAMDISK of size %lld\n",
335 ramdisk_size);
336
337 /* Note: this includes all the mem currently occupied by
338 the initrd, we rely on that fact to keep the data intact. */
339 memblock_reserve(relocated_ramdisk, area_size);
340 initrd_start = relocated_ramdisk + PAGE_OFFSET;
341 initrd_end = initrd_start + ramdisk_size;
342 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
343 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
344
345 q = (char *)initrd_start;
346
347 /* Copy the initrd */
348 while (ramdisk_size) {
349 slop = ramdisk_image & ~PAGE_MASK;
350 clen = ramdisk_size;
351 if (clen > MAX_MAP_CHUNK-slop)
352 clen = MAX_MAP_CHUNK-slop;
353 mapaddr = ramdisk_image & PAGE_MASK;
354 p = early_memremap(mapaddr, clen+slop);
355 memcpy(q, p+slop, clen);
356 early_iounmap(p, clen+slop);
357 q += clen;
358 ramdisk_image += clen;
359 ramdisk_size -= clen;
360 }
361
362 ramdisk_image = get_ramdisk_image();
363 ramdisk_size = get_ramdisk_size();
364 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
365 " [mem %#010llx-%#010llx]\n",
366 ramdisk_image, ramdisk_image + ramdisk_size - 1,
367 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
368}
369
370static void __init early_reserve_initrd(void)
371{
372 /* Assume only end is not page aligned */
373 u64 ramdisk_image = get_ramdisk_image();
374 u64 ramdisk_size = get_ramdisk_size();
375 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
376
377 if (!boot_params.hdr.type_of_loader ||
378 !ramdisk_image || !ramdisk_size)
379 return; /* No initrd provided by bootloader */
380
381 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
382}
383static void __init reserve_initrd(void)
384{
385 /* Assume only end is not page aligned */
386 u64 ramdisk_image = get_ramdisk_image();
387 u64 ramdisk_size = get_ramdisk_size();
388 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
389 u64 mapped_size;
390
391 if (!boot_params.hdr.type_of_loader ||
392 !ramdisk_image || !ramdisk_size)
393 return; /* No initrd provided by bootloader */
394
395 initrd_start = 0;
396
397 mapped_size = memblock_mem_size(max_pfn_mapped);
398 if (ramdisk_size >= (mapped_size>>1))
399 panic("initrd too large to handle, "
400 "disabling initrd (%lld needed, %lld available)\n",
401 ramdisk_size, mapped_size>>1);
402
403 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
404 ramdisk_end - 1);
405
406 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
407 PFN_DOWN(ramdisk_end))) {
408 /* All are mapped, easy case */
409 initrd_start = ramdisk_image + PAGE_OFFSET;
410 initrd_end = initrd_start + ramdisk_size;
411 return;
412 }
413
414 relocate_initrd();
415
416 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
417}
418#else
419static void __init early_reserve_initrd(void)
420{
421}
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, pa_next;
431
432 pa_data = boot_params.hdr.setup_data;
433 while (pa_data) {
434 u32 data_len, map_len, data_type;
435
436 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
437 (u64)sizeof(struct setup_data));
438 data = early_memremap(pa_data, map_len);
439 data_len = data->len + sizeof(struct setup_data);
440 data_type = data->type;
441 pa_next = data->next;
442 early_iounmap(data, map_len);
443
444 switch (data_type) {
445 case SETUP_E820_EXT:
446 parse_e820_ext(pa_data, data_len);
447 break;
448 case SETUP_DTB:
449 add_dtb(pa_data);
450 break;
451 case SETUP_EFI:
452 parse_efi_setup(pa_data, data_len);
453 break;
454 default:
455 break;
456 }
457 pa_data = pa_next;
458 }
459}
460
461static void __init e820_reserve_setup_data(void)
462{
463 struct setup_data *data;
464 u64 pa_data;
465 int found = 0;
466
467 pa_data = boot_params.hdr.setup_data;
468 while (pa_data) {
469 data = early_memremap(pa_data, sizeof(*data));
470 e820_update_range(pa_data, sizeof(*data)+data->len,
471 E820_RAM, E820_RESERVED_KERN);
472 found = 1;
473 pa_data = data->next;
474 early_iounmap(data, sizeof(*data));
475 }
476 if (!found)
477 return;
478
479 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
480 memcpy(&e820_saved, &e820, sizeof(struct e820map));
481 printk(KERN_INFO "extended physical RAM map:\n");
482 e820_print_map("reserve setup_data");
483}
484
485static void __init memblock_x86_reserve_range_setup_data(void)
486{
487 struct setup_data *data;
488 u64 pa_data;
489
490 pa_data = boot_params.hdr.setup_data;
491 while (pa_data) {
492 data = early_memremap(pa_data, sizeof(*data));
493 memblock_reserve(pa_data, sizeof(*data) + data->len);
494 pa_data = data->next;
495 early_iounmap(data, sizeof(*data));
496 }
497}
498
499/*
500 * --------- Crashkernel reservation ------------------------------
501 */
502
503#ifdef CONFIG_KEXEC
504
505/*
506 * Keep the crash kernel below this limit. On 32 bits earlier kernels
507 * would limit the kernel to the low 512 MiB due to mapping restrictions.
508 * On 64bit, old kexec-tools need to under 896MiB.
509 */
510#ifdef CONFIG_X86_32
511# define CRASH_KERNEL_ADDR_LOW_MAX (512 << 20)
512# define CRASH_KERNEL_ADDR_HIGH_MAX (512 << 20)
513#else
514# define CRASH_KERNEL_ADDR_LOW_MAX (896UL<<20)
515# define CRASH_KERNEL_ADDR_HIGH_MAX MAXMEM
516#endif
517
518static void __init reserve_crashkernel_low(void)
519{
520#ifdef CONFIG_X86_64
521 const unsigned long long alignment = 16<<20; /* 16M */
522 unsigned long long low_base = 0, low_size = 0;
523 unsigned long total_low_mem;
524 unsigned long long base;
525 bool auto_set = false;
526 int ret;
527
528 total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT));
529 /* crashkernel=Y,low */
530 ret = parse_crashkernel_low(boot_command_line, total_low_mem,
531 &low_size, &base);
532 if (ret != 0) {
533 /*
534 * two parts from lib/swiotlb.c:
535 * swiotlb size: user specified with swiotlb= or default.
536 * swiotlb overflow buffer: now is hardcoded to 32k.
537 * We round it to 8M for other buffers that
538 * may need to stay low too.
539 */
540 low_size = swiotlb_size_or_default() + (8UL<<20);
541 auto_set = true;
542 } else {
543 /* passed with crashkernel=0,low ? */
544 if (!low_size)
545 return;
546 }
547
548 low_base = memblock_find_in_range(low_size, (1ULL<<32),
549 low_size, alignment);
550
551 if (!low_base) {
552 if (!auto_set)
553 pr_info("crashkernel low reservation failed - No suitable area found.\n");
554
555 return;
556 }
557
558 memblock_reserve(low_base, low_size);
559 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
560 (unsigned long)(low_size >> 20),
561 (unsigned long)(low_base >> 20),
562 (unsigned long)(total_low_mem >> 20));
563 crashk_low_res.start = low_base;
564 crashk_low_res.end = low_base + low_size - 1;
565 insert_resource(&iomem_resource, &crashk_low_res);
566#endif
567}
568
569static void __init reserve_crashkernel(void)
570{
571 const unsigned long long alignment = 16<<20; /* 16M */
572 unsigned long long total_mem;
573 unsigned long long crash_size, crash_base;
574 bool high = false;
575 int ret;
576
577 total_mem = memblock_phys_mem_size();
578
579 /* crashkernel=XM */
580 ret = parse_crashkernel(boot_command_line, total_mem,
581 &crash_size, &crash_base);
582 if (ret != 0 || crash_size <= 0) {
583 /* crashkernel=X,high */
584 ret = parse_crashkernel_high(boot_command_line, total_mem,
585 &crash_size, &crash_base);
586 if (ret != 0 || crash_size <= 0)
587 return;
588 high = true;
589 }
590
591 /* 0 means: find the address automatically */
592 if (crash_base <= 0) {
593 /*
594 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
595 */
596 crash_base = memblock_find_in_range(alignment,
597 high ? CRASH_KERNEL_ADDR_HIGH_MAX :
598 CRASH_KERNEL_ADDR_LOW_MAX,
599 crash_size, alignment);
600
601 if (!crash_base) {
602 pr_info("crashkernel reservation failed - No suitable area found.\n");
603 return;
604 }
605
606 } else {
607 unsigned long long start;
608
609 start = memblock_find_in_range(crash_base,
610 crash_base + crash_size, crash_size, 1<<20);
611 if (start != crash_base) {
612 pr_info("crashkernel reservation failed - memory is in use.\n");
613 return;
614 }
615 }
616 memblock_reserve(crash_base, crash_size);
617
618 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
619 "for crashkernel (System RAM: %ldMB)\n",
620 (unsigned long)(crash_size >> 20),
621 (unsigned long)(crash_base >> 20),
622 (unsigned long)(total_mem >> 20));
623
624 crashk_res.start = crash_base;
625 crashk_res.end = crash_base + crash_size - 1;
626 insert_resource(&iomem_resource, &crashk_res);
627
628 if (crash_base >= (1ULL<<32))
629 reserve_crashkernel_low();
630}
631#else
632static void __init reserve_crashkernel(void)
633{
634}
635#endif
636
637static struct resource standard_io_resources[] = {
638 { .name = "dma1", .start = 0x00, .end = 0x1f,
639 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
640 { .name = "pic1", .start = 0x20, .end = 0x21,
641 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
642 { .name = "timer0", .start = 0x40, .end = 0x43,
643 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
644 { .name = "timer1", .start = 0x50, .end = 0x53,
645 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
646 { .name = "keyboard", .start = 0x60, .end = 0x60,
647 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
648 { .name = "keyboard", .start = 0x64, .end = 0x64,
649 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
650 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
651 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
652 { .name = "pic2", .start = 0xa0, .end = 0xa1,
653 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
654 { .name = "dma2", .start = 0xc0, .end = 0xdf,
655 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
656 { .name = "fpu", .start = 0xf0, .end = 0xff,
657 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
658};
659
660void __init reserve_standard_io_resources(void)
661{
662 int i;
663
664 /* request I/O space for devices used on all i[345]86 PCs */
665 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
666 request_resource(&ioport_resource, &standard_io_resources[i]);
667
668}
669
670static __init void reserve_ibft_region(void)
671{
672 unsigned long addr, size = 0;
673
674 addr = find_ibft_region(&size);
675
676 if (size)
677 memblock_reserve(addr, size);
678}
679
680static bool __init snb_gfx_workaround_needed(void)
681{
682#ifdef CONFIG_PCI
683 int i;
684 u16 vendor, devid;
685 static const __initconst u16 snb_ids[] = {
686 0x0102,
687 0x0112,
688 0x0122,
689 0x0106,
690 0x0116,
691 0x0126,
692 0x010a,
693 };
694
695 /* Assume no if something weird is going on with PCI */
696 if (!early_pci_allowed())
697 return false;
698
699 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
700 if (vendor != 0x8086)
701 return false;
702
703 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
704 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
705 if (devid == snb_ids[i])
706 return true;
707#endif
708
709 return false;
710}
711
712/*
713 * Sandy Bridge graphics has trouble with certain ranges, exclude
714 * them from allocation.
715 */
716static void __init trim_snb_memory(void)
717{
718 static const __initconst unsigned long bad_pages[] = {
719 0x20050000,
720 0x20110000,
721 0x20130000,
722 0x20138000,
723 0x40004000,
724 };
725 int i;
726
727 if (!snb_gfx_workaround_needed())
728 return;
729
730 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
731
732 /*
733 * Reserve all memory below the 1 MB mark that has not
734 * already been reserved.
735 */
736 memblock_reserve(0, 1<<20);
737
738 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
739 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
740 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
741 bad_pages[i]);
742 }
743}
744
745/*
746 * Here we put platform-specific memory range workarounds, i.e.
747 * memory known to be corrupt or otherwise in need to be reserved on
748 * specific platforms.
749 *
750 * If this gets used more widely it could use a real dispatch mechanism.
751 */
752static void __init trim_platform_memory_ranges(void)
753{
754 trim_snb_memory();
755}
756
757static void __init trim_bios_range(void)
758{
759 /*
760 * A special case is the first 4Kb of memory;
761 * This is a BIOS owned area, not kernel ram, but generally
762 * not listed as such in the E820 table.
763 *
764 * This typically reserves additional memory (64KiB by default)
765 * since some BIOSes are known to corrupt low memory. See the
766 * Kconfig help text for X86_RESERVE_LOW.
767 */
768 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
769
770 /*
771 * special case: Some BIOSen report the PC BIOS
772 * area (640->1Mb) as ram even though it is not.
773 * take them out.
774 */
775 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
776
777 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
778}
779
780/* called before trim_bios_range() to spare extra sanitize */
781static void __init e820_add_kernel_range(void)
782{
783 u64 start = __pa_symbol(_text);
784 u64 size = __pa_symbol(_end) - start;
785
786 /*
787 * Complain if .text .data and .bss are not marked as E820_RAM and
788 * attempt to fix it by adding the range. We may have a confused BIOS,
789 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
790 * exclude kernel range. If we really are running on top non-RAM,
791 * we will crash later anyways.
792 */
793 if (e820_all_mapped(start, start + size, E820_RAM))
794 return;
795
796 pr_warn(".text .data .bss are not marked as E820_RAM!\n");
797 e820_remove_range(start, size, E820_RAM, 0);
798 e820_add_region(start, size, E820_RAM);
799}
800
801static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
802
803static int __init parse_reservelow(char *p)
804{
805 unsigned long long size;
806
807 if (!p)
808 return -EINVAL;
809
810 size = memparse(p, &p);
811
812 if (size < 4096)
813 size = 4096;
814
815 if (size > 640*1024)
816 size = 640*1024;
817
818 reserve_low = size;
819
820 return 0;
821}
822
823early_param("reservelow", parse_reservelow);
824
825static void __init trim_low_memory_range(void)
826{
827 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
828}
829
830/*
831 * Dump out kernel offset information on panic.
832 */
833static int
834dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
835{
836 pr_emerg("Kernel Offset: 0x%lx from 0x%lx "
837 "(relocation range: 0x%lx-0x%lx)\n",
838 (unsigned long)&_text - __START_KERNEL, __START_KERNEL,
839 __START_KERNEL_map, MODULES_VADDR-1);
840
841 return 0;
842}
843
844/*
845 * Determine if we were loaded by an EFI loader. If so, then we have also been
846 * passed the efi memmap, systab, etc., so we should use these data structures
847 * for initialization. Note, the efi init code path is determined by the
848 * global efi_enabled. This allows the same kernel image to be used on existing
849 * systems (with a traditional BIOS) as well as on EFI systems.
850 */
851/*
852 * setup_arch - architecture-specific boot-time initializations
853 *
854 * Note: On x86_64, fixmaps are ready for use even before this is called.
855 */
856
857void __init setup_arch(char **cmdline_p)
858{
859 memblock_reserve(__pa_symbol(_text),
860 (unsigned long)__bss_stop - (unsigned long)_text);
861
862 early_reserve_initrd();
863
864 /*
865 * At this point everything still needed from the boot loader
866 * or BIOS or kernel text should be early reserved or marked not
867 * RAM in e820. All other memory is free game.
868 */
869
870#ifdef CONFIG_X86_32
871 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
872
873 /*
874 * copy kernel address range established so far and switch
875 * to the proper swapper page table
876 */
877 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
878 initial_page_table + KERNEL_PGD_BOUNDARY,
879 KERNEL_PGD_PTRS);
880
881 load_cr3(swapper_pg_dir);
882 __flush_tlb_all();
883#else
884 printk(KERN_INFO "Command line: %s\n", boot_command_line);
885#endif
886
887 /*
888 * If we have OLPC OFW, we might end up relocating the fixmap due to
889 * reserve_top(), so do this before touching the ioremap area.
890 */
891 olpc_ofw_detect();
892
893 early_trap_init();
894 early_cpu_init();
895 early_ioremap_init();
896
897 setup_olpc_ofw_pgd();
898
899 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
900 screen_info = boot_params.screen_info;
901 edid_info = boot_params.edid_info;
902#ifdef CONFIG_X86_32
903 apm_info.bios = boot_params.apm_bios_info;
904 ist_info = boot_params.ist_info;
905 if (boot_params.sys_desc_table.length != 0) {
906 machine_id = boot_params.sys_desc_table.table[0];
907 machine_submodel_id = boot_params.sys_desc_table.table[1];
908 BIOS_revision = boot_params.sys_desc_table.table[2];
909 }
910#endif
911 saved_video_mode = boot_params.hdr.vid_mode;
912 bootloader_type = boot_params.hdr.type_of_loader;
913 if ((bootloader_type >> 4) == 0xe) {
914 bootloader_type &= 0xf;
915 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
916 }
917 bootloader_version = bootloader_type & 0xf;
918 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
919
920#ifdef CONFIG_BLK_DEV_RAM
921 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
922 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
923 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
924#endif
925#ifdef CONFIG_EFI
926 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
927 "EL32", 4)) {
928 set_bit(EFI_BOOT, &efi.flags);
929 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
930 "EL64", 4)) {
931 set_bit(EFI_BOOT, &efi.flags);
932 set_bit(EFI_64BIT, &efi.flags);
933 }
934
935 if (efi_enabled(EFI_BOOT))
936 efi_memblock_x86_reserve_range();
937#endif
938
939 x86_init.oem.arch_setup();
940
941 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
942 setup_memory_map();
943 parse_setup_data();
944
945 copy_edd();
946
947 if (!boot_params.hdr.root_flags)
948 root_mountflags &= ~MS_RDONLY;
949 init_mm.start_code = (unsigned long) _text;
950 init_mm.end_code = (unsigned long) _etext;
951 init_mm.end_data = (unsigned long) _edata;
952 init_mm.brk = _brk_end;
953
954 code_resource.start = __pa_symbol(_text);
955 code_resource.end = __pa_symbol(_etext)-1;
956 data_resource.start = __pa_symbol(_etext);
957 data_resource.end = __pa_symbol(_edata)-1;
958 bss_resource.start = __pa_symbol(__bss_start);
959 bss_resource.end = __pa_symbol(__bss_stop)-1;
960
961#ifdef CONFIG_CMDLINE_BOOL
962#ifdef CONFIG_CMDLINE_OVERRIDE
963 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
964#else
965 if (builtin_cmdline[0]) {
966 /* append boot loader cmdline to builtin */
967 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
968 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
969 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
970 }
971#endif
972#endif
973
974 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
975 *cmdline_p = command_line;
976
977 /*
978 * x86_configure_nx() is called before parse_early_param() to detect
979 * whether hardware doesn't support NX (so that the early EHCI debug
980 * console setup can safely call set_fixmap()). It may then be called
981 * again from within noexec_setup() during parsing early parameters
982 * to honor the respective command line option.
983 */
984 x86_configure_nx();
985
986 parse_early_param();
987
988 x86_report_nx();
989
990 /* after early param, so could get panic from serial */
991 memblock_x86_reserve_range_setup_data();
992
993 if (acpi_mps_check()) {
994#ifdef CONFIG_X86_LOCAL_APIC
995 disable_apic = 1;
996#endif
997 setup_clear_cpu_cap(X86_FEATURE_APIC);
998 }
999
1000#ifdef CONFIG_PCI
1001 if (pci_early_dump_regs)
1002 early_dump_pci_devices();
1003#endif
1004
1005 /* update the e820_saved too */
1006 e820_reserve_setup_data();
1007 finish_e820_parsing();
1008
1009 if (efi_enabled(EFI_BOOT))
1010 efi_init();
1011
1012 dmi_scan_machine();
1013 dmi_memdev_walk();
1014 dmi_set_dump_stack_arch_desc();
1015
1016 /*
1017 * VMware detection requires dmi to be available, so this
1018 * needs to be done after dmi_scan_machine, for the BP.
1019 */
1020 init_hypervisor_platform();
1021
1022 x86_init.resources.probe_roms();
1023
1024 /* after parse_early_param, so could debug it */
1025 insert_resource(&iomem_resource, &code_resource);
1026 insert_resource(&iomem_resource, &data_resource);
1027 insert_resource(&iomem_resource, &bss_resource);
1028
1029 e820_add_kernel_range();
1030 trim_bios_range();
1031#ifdef CONFIG_X86_32
1032 if (ppro_with_ram_bug()) {
1033 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1034 E820_RESERVED);
1035 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1036 printk(KERN_INFO "fixed physical RAM map:\n");
1037 e820_print_map("bad_ppro");
1038 }
1039#else
1040 early_gart_iommu_check();
1041#endif
1042
1043 /*
1044 * partially used pages are not usable - thus
1045 * we are rounding upwards:
1046 */
1047 max_pfn = e820_end_of_ram_pfn();
1048
1049 /* update e820 for memory not covered by WB MTRRs */
1050 mtrr_bp_init();
1051 if (mtrr_trim_uncached_memory(max_pfn))
1052 max_pfn = e820_end_of_ram_pfn();
1053
1054#ifdef CONFIG_X86_32
1055 /* max_low_pfn get updated here */
1056 find_low_pfn_range();
1057#else
1058 check_x2apic();
1059
1060 /* How many end-of-memory variables you have, grandma! */
1061 /* need this before calling reserve_initrd */
1062 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1063 max_low_pfn = e820_end_of_low_ram_pfn();
1064 else
1065 max_low_pfn = max_pfn;
1066
1067 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1068#endif
1069
1070 /*
1071 * Find and reserve possible boot-time SMP configuration:
1072 */
1073 find_smp_config();
1074
1075 reserve_ibft_region();
1076
1077 early_alloc_pgt_buf();
1078
1079 /*
1080 * Need to conclude brk, before memblock_x86_fill()
1081 * it could use memblock_find_in_range, could overlap with
1082 * brk area.
1083 */
1084 reserve_brk();
1085
1086 cleanup_highmap();
1087
1088 memblock_set_current_limit(ISA_END_ADDRESS);
1089 memblock_x86_fill();
1090
1091 /*
1092 * The EFI specification says that boot service code won't be called
1093 * after ExitBootServices(). This is, in fact, a lie.
1094 */
1095 if (efi_enabled(EFI_MEMMAP))
1096 efi_reserve_boot_services();
1097
1098 /* preallocate 4k for mptable mpc */
1099 early_reserve_e820_mpc_new();
1100
1101#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1102 setup_bios_corruption_check();
1103#endif
1104
1105#ifdef CONFIG_X86_32
1106 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1107 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1108#endif
1109
1110 reserve_real_mode();
1111
1112 trim_platform_memory_ranges();
1113 trim_low_memory_range();
1114
1115 init_mem_mapping();
1116
1117 early_trap_pf_init();
1118
1119 setup_real_mode();
1120
1121 memblock_set_current_limit(get_max_mapped());
1122 dma_contiguous_reserve(0);
1123
1124 /*
1125 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1126 */
1127
1128#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1129 if (init_ohci1394_dma_early)
1130 init_ohci1394_dma_on_all_controllers();
1131#endif
1132 /* Allocate bigger log buffer */
1133 setup_log_buf(1);
1134
1135 reserve_initrd();
1136
1137#if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1138 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1139#endif
1140
1141 vsmp_init();
1142
1143 io_delay_init();
1144
1145 /*
1146 * Parse the ACPI tables for possible boot-time SMP configuration.
1147 */
1148 acpi_boot_table_init();
1149
1150 early_acpi_boot_init();
1151
1152 initmem_init();
1153
1154 /*
1155 * Reserve memory for crash kernel after SRAT is parsed so that it
1156 * won't consume hotpluggable memory.
1157 */
1158 reserve_crashkernel();
1159
1160 memblock_find_dma_reserve();
1161
1162#ifdef CONFIG_KVM_GUEST
1163 kvmclock_init();
1164#endif
1165
1166 x86_init.paging.pagetable_init();
1167
1168 if (boot_cpu_data.cpuid_level >= 0) {
1169 /* A CPU has %cr4 if and only if it has CPUID */
1170 mmu_cr4_features = read_cr4();
1171 if (trampoline_cr4_features)
1172 *trampoline_cr4_features = mmu_cr4_features;
1173 }
1174
1175#ifdef CONFIG_X86_32
1176 /* sync back kernel address range */
1177 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1178 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1179 KERNEL_PGD_PTRS);
1180#endif
1181
1182 tboot_probe();
1183
1184#ifdef CONFIG_X86_64
1185 map_vsyscall();
1186#endif
1187
1188 generic_apic_probe();
1189
1190 early_quirks();
1191
1192 /*
1193 * Read APIC and some other early information from ACPI tables.
1194 */
1195 acpi_boot_init();
1196 sfi_init();
1197 x86_dtb_init();
1198
1199 /*
1200 * get boot-time SMP configuration:
1201 */
1202 if (smp_found_config)
1203 get_smp_config();
1204
1205 prefill_possible_map();
1206
1207 init_cpu_to_node();
1208
1209 init_apic_mappings();
1210 if (x86_io_apic_ops.init)
1211 x86_io_apic_ops.init();
1212
1213 kvm_guest_init();
1214
1215 e820_reserve_resources();
1216 e820_mark_nosave_regions(max_low_pfn);
1217
1218 x86_init.resources.reserve_resources();
1219
1220 e820_setup_gap();
1221
1222#ifdef CONFIG_VT
1223#if defined(CONFIG_VGA_CONSOLE)
1224 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1225 conswitchp = &vga_con;
1226#elif defined(CONFIG_DUMMY_CONSOLE)
1227 conswitchp = &dummy_con;
1228#endif
1229#endif
1230 x86_init.oem.banner();
1231
1232 x86_init.timers.wallclock_init();
1233
1234 mcheck_init();
1235
1236 arch_init_ideal_nops();
1237
1238 register_refined_jiffies(CLOCK_TICK_RATE);
1239
1240#ifdef CONFIG_EFI
1241 if (efi_enabled(EFI_BOOT))
1242 efi_apply_memmap_quirks();
1243#endif
1244}
1245
1246#ifdef CONFIG_X86_32
1247
1248static struct resource video_ram_resource = {
1249 .name = "Video RAM area",
1250 .start = 0xa0000,
1251 .end = 0xbffff,
1252 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1253};
1254
1255void __init i386_reserve_resources(void)
1256{
1257 request_resource(&iomem_resource, &video_ram_resource);
1258 reserve_standard_io_resources();
1259}
1260
1261#endif /* CONFIG_X86_32 */
1262
1263static struct notifier_block kernel_offset_notifier = {
1264 .notifier_call = dump_kernel_offset
1265};
1266
1267static int __init register_kernel_offset_dumper(void)
1268{
1269 atomic_notifier_chain_register(&panic_notifier_list,
1270 &kernel_offset_notifier);
1271 return 0;
1272}
1273__initcall(register_kernel_offset_dumper);