xref: /linux/arch/x86/kernel/setup.c (revision 04303f8ec14269b0ea2553863553bc7eaadca1f8)
1 /*
2  *  Copyright (C) 1995  Linus Torvalds
3  *
4  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5  *
6  *  Memory region support
7  *	David Parsons <orc@pell.chi.il.us>, July-August 1999
8  *
9  *  Added E820 sanitization routine (removes overlapping memory regions);
10  *  Brian Moyle <bmoyle@mvista.com>, February 2001
11  *
12  * Moved CPU detection code to cpu/${cpu}.c
13  *    Patrick Mochel <mochel@osdl.org>, March 2002
14  *
15  *  Provisions for empty E820 memory regions (reported by certain BIOSes).
16  *  Alex Achenbach <xela@slit.de>, December 2002.
17  *
18  */
19 
20 /*
21  * This file handles the architecture-dependent parts of initialization
22  */
23 
24 #include <linux/sched.h>
25 #include <linux/mm.h>
26 #include <linux/mmzone.h>
27 #include <linux/screen_info.h>
28 #include <linux/ioport.h>
29 #include <linux/acpi.h>
30 #include <linux/sfi.h>
31 #include <linux/apm_bios.h>
32 #include <linux/initrd.h>
33 #include <linux/bootmem.h>
34 #include <linux/memblock.h>
35 #include <linux/seq_file.h>
36 #include <linux/console.h>
37 #include <linux/root_dev.h>
38 #include <linux/highmem.h>
39 #include <linux/module.h>
40 #include <linux/efi.h>
41 #include <linux/init.h>
42 #include <linux/edd.h>
43 #include <linux/iscsi_ibft.h>
44 #include <linux/nodemask.h>
45 #include <linux/kexec.h>
46 #include <linux/dmi.h>
47 #include <linux/pfn.h>
48 #include <linux/pci.h>
49 #include <asm/pci-direct.h>
50 #include <linux/init_ohci1394_dma.h>
51 #include <linux/kvm_para.h>
52 #include <linux/dma-contiguous.h>
53 
54 #include <linux/errno.h>
55 #include <linux/kernel.h>
56 #include <linux/stddef.h>
57 #include <linux/unistd.h>
58 #include <linux/ptrace.h>
59 #include <linux/user.h>
60 #include <linux/delay.h>
61 
62 #include <linux/kallsyms.h>
63 #include <linux/cpufreq.h>
64 #include <linux/dma-mapping.h>
65 #include <linux/ctype.h>
66 #include <linux/uaccess.h>
67 
68 #include <linux/percpu.h>
69 #include <linux/crash_dump.h>
70 #include <linux/tboot.h>
71 #include <linux/jiffies.h>
72 
73 #include <video/edid.h>
74 
75 #include <asm/mtrr.h>
76 #include <asm/apic.h>
77 #include <asm/realmode.h>
78 #include <asm/e820.h>
79 #include <asm/mpspec.h>
80 #include <asm/setup.h>
81 #include <asm/efi.h>
82 #include <asm/timer.h>
83 #include <asm/i8259.h>
84 #include <asm/sections.h>
85 #include <asm/io_apic.h>
86 #include <asm/ist.h>
87 #include <asm/setup_arch.h>
88 #include <asm/bios_ebda.h>
89 #include <asm/cacheflush.h>
90 #include <asm/processor.h>
91 #include <asm/bugs.h>
92 
93 #include <asm/vsyscall.h>
94 #include <asm/cpu.h>
95 #include <asm/desc.h>
96 #include <asm/dma.h>
97 #include <asm/iommu.h>
98 #include <asm/gart.h>
99 #include <asm/mmu_context.h>
100 #include <asm/proto.h>
101 
102 #include <asm/paravirt.h>
103 #include <asm/hypervisor.h>
104 #include <asm/olpc_ofw.h>
105 
106 #include <asm/percpu.h>
107 #include <asm/topology.h>
108 #include <asm/apicdef.h>
109 #include <asm/amd_nb.h>
110 #include <asm/mce.h>
111 #include <asm/alternative.h>
112 #include <asm/prom.h>
113 
114 /*
115  * max_low_pfn_mapped: highest direct mapped pfn under 4GB
116  * max_pfn_mapped:     highest direct mapped pfn over 4GB
117  *
118  * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
119  * represented by pfn_mapped
120  */
121 unsigned long max_low_pfn_mapped;
122 unsigned long max_pfn_mapped;
123 
124 #ifdef CONFIG_DMI
125 RESERVE_BRK(dmi_alloc, 65536);
126 #endif
127 
128 
129 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
130 unsigned long _brk_end = (unsigned long)__brk_base;
131 
132 #ifdef CONFIG_X86_64
133 int default_cpu_present_to_apicid(int mps_cpu)
134 {
135 	return __default_cpu_present_to_apicid(mps_cpu);
136 }
137 
138 int default_check_phys_apicid_present(int phys_apicid)
139 {
140 	return __default_check_phys_apicid_present(phys_apicid);
141 }
142 #endif
143 
144 struct boot_params boot_params;
145 
146 /*
147  * Machine setup..
148  */
149 static struct resource data_resource = {
150 	.name	= "Kernel data",
151 	.start	= 0,
152 	.end	= 0,
153 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
154 };
155 
156 static struct resource code_resource = {
157 	.name	= "Kernel code",
158 	.start	= 0,
159 	.end	= 0,
160 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
161 };
162 
163 static struct resource bss_resource = {
164 	.name	= "Kernel bss",
165 	.start	= 0,
166 	.end	= 0,
167 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
168 };
169 
170 
171 #ifdef CONFIG_X86_32
172 /* cpu data as detected by the assembly code in head.S */
173 struct cpuinfo_x86 new_cpu_data = {
174 	.wp_works_ok = -1,
175 };
176 /* common cpu data for all cpus */
177 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
178 	.wp_works_ok = -1,
179 };
180 EXPORT_SYMBOL(boot_cpu_data);
181 
182 unsigned int def_to_bigsmp;
183 
184 /* for MCA, but anyone else can use it if they want */
185 unsigned int machine_id;
186 unsigned int machine_submodel_id;
187 unsigned int BIOS_revision;
188 
189 struct apm_info apm_info;
190 EXPORT_SYMBOL(apm_info);
191 
192 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
193 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
194 struct ist_info ist_info;
195 EXPORT_SYMBOL(ist_info);
196 #else
197 struct ist_info ist_info;
198 #endif
199 
200 #else
201 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
202 	.x86_phys_bits = MAX_PHYSMEM_BITS,
203 };
204 EXPORT_SYMBOL(boot_cpu_data);
205 #endif
206 
207 
208 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
209 __visible unsigned long mmu_cr4_features;
210 #else
211 __visible unsigned long mmu_cr4_features = X86_CR4_PAE;
212 #endif
213 
214 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
215 int bootloader_type, bootloader_version;
216 
217 /*
218  * Setup options
219  */
220 struct screen_info screen_info;
221 EXPORT_SYMBOL(screen_info);
222 struct edid_info edid_info;
223 EXPORT_SYMBOL_GPL(edid_info);
224 
225 extern int root_mountflags;
226 
227 unsigned long saved_video_mode;
228 
229 #define RAMDISK_IMAGE_START_MASK	0x07FF
230 #define RAMDISK_PROMPT_FLAG		0x8000
231 #define RAMDISK_LOAD_FLAG		0x4000
232 
233 static char __initdata command_line[COMMAND_LINE_SIZE];
234 #ifdef CONFIG_CMDLINE_BOOL
235 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
236 #endif
237 
238 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
239 struct edd edd;
240 #ifdef CONFIG_EDD_MODULE
241 EXPORT_SYMBOL(edd);
242 #endif
243 /**
244  * copy_edd() - Copy the BIOS EDD information
245  *              from boot_params into a safe place.
246  *
247  */
248 static inline void __init copy_edd(void)
249 {
250      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
251 	    sizeof(edd.mbr_signature));
252      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
253      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
254      edd.edd_info_nr = boot_params.eddbuf_entries;
255 }
256 #else
257 static inline void __init copy_edd(void)
258 {
259 }
260 #endif
261 
262 void * __init extend_brk(size_t size, size_t align)
263 {
264 	size_t mask = align - 1;
265 	void *ret;
266 
267 	BUG_ON(_brk_start == 0);
268 	BUG_ON(align & mask);
269 
270 	_brk_end = (_brk_end + mask) & ~mask;
271 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
272 
273 	ret = (void *)_brk_end;
274 	_brk_end += size;
275 
276 	memset(ret, 0, size);
277 
278 	return ret;
279 }
280 
281 #ifdef CONFIG_X86_32
282 static void __init cleanup_highmap(void)
283 {
284 }
285 #endif
286 
287 static void __init reserve_brk(void)
288 {
289 	if (_brk_end > _brk_start)
290 		memblock_reserve(__pa_symbol(_brk_start),
291 				 _brk_end - _brk_start);
292 
293 	/* Mark brk area as locked down and no longer taking any
294 	   new allocations */
295 	_brk_start = 0;
296 }
297 
298 u64 relocated_ramdisk;
299 
300 #ifdef CONFIG_BLK_DEV_INITRD
301 
302 static 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 }
310 static 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)
320 static 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 
370 static 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 }
383 static 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
419 static void __init early_reserve_initrd(void)
420 {
421 }
422 static void __init reserve_initrd(void)
423 {
424 }
425 #endif /* CONFIG_BLK_DEV_INITRD */
426 
427 static 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, data_type;
435 
436 		data = early_memremap(pa_data, sizeof(*data));
437 		data_len = data->len + sizeof(struct setup_data);
438 		data_type = data->type;
439 		pa_next = data->next;
440 		early_iounmap(data, sizeof(*data));
441 
442 		switch (data_type) {
443 		case SETUP_E820_EXT:
444 			parse_e820_ext(pa_data, data_len);
445 			break;
446 		case SETUP_DTB:
447 			add_dtb(pa_data);
448 			break;
449 		case SETUP_EFI:
450 			parse_efi_setup(pa_data, data_len);
451 			break;
452 		default:
453 			break;
454 		}
455 		pa_data = pa_next;
456 	}
457 }
458 
459 static void __init e820_reserve_setup_data(void)
460 {
461 	struct setup_data *data;
462 	u64 pa_data;
463 	int found = 0;
464 
465 	pa_data = boot_params.hdr.setup_data;
466 	while (pa_data) {
467 		data = early_memremap(pa_data, sizeof(*data));
468 		e820_update_range(pa_data, sizeof(*data)+data->len,
469 			 E820_RAM, E820_RESERVED_KERN);
470 		found = 1;
471 		pa_data = data->next;
472 		early_iounmap(data, sizeof(*data));
473 	}
474 	if (!found)
475 		return;
476 
477 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
478 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
479 	printk(KERN_INFO "extended physical RAM map:\n");
480 	e820_print_map("reserve setup_data");
481 }
482 
483 static void __init memblock_x86_reserve_range_setup_data(void)
484 {
485 	struct setup_data *data;
486 	u64 pa_data;
487 
488 	pa_data = boot_params.hdr.setup_data;
489 	while (pa_data) {
490 		data = early_memremap(pa_data, sizeof(*data));
491 		memblock_reserve(pa_data, sizeof(*data) + data->len);
492 		pa_data = data->next;
493 		early_iounmap(data, sizeof(*data));
494 	}
495 }
496 
497 /*
498  * --------- Crashkernel reservation ------------------------------
499  */
500 
501 #ifdef CONFIG_KEXEC
502 
503 /*
504  * Keep the crash kernel below this limit.  On 32 bits earlier kernels
505  * would limit the kernel to the low 512 MiB due to mapping restrictions.
506  * On 64bit, old kexec-tools need to under 896MiB.
507  */
508 #ifdef CONFIG_X86_32
509 # define CRASH_KERNEL_ADDR_LOW_MAX	(512 << 20)
510 # define CRASH_KERNEL_ADDR_HIGH_MAX	(512 << 20)
511 #else
512 # define CRASH_KERNEL_ADDR_LOW_MAX	(896UL<<20)
513 # define CRASH_KERNEL_ADDR_HIGH_MAX	MAXMEM
514 #endif
515 
516 static void __init reserve_crashkernel_low(void)
517 {
518 #ifdef CONFIG_X86_64
519 	const unsigned long long alignment = 16<<20;	/* 16M */
520 	unsigned long long low_base = 0, low_size = 0;
521 	unsigned long total_low_mem;
522 	unsigned long long base;
523 	bool auto_set = false;
524 	int ret;
525 
526 	total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT));
527 	/* crashkernel=Y,low */
528 	ret = parse_crashkernel_low(boot_command_line, total_low_mem,
529 						&low_size, &base);
530 	if (ret != 0) {
531 		/*
532 		 * two parts from lib/swiotlb.c:
533 		 *	swiotlb size: user specified with swiotlb= or default.
534 		 *	swiotlb overflow buffer: now is hardcoded to 32k.
535 		 *		We round it to 8M for other buffers that
536 		 *		may need to stay low too.
537 		 */
538 		low_size = swiotlb_size_or_default() + (8UL<<20);
539 		auto_set = true;
540 	} else {
541 		/* passed with crashkernel=0,low ? */
542 		if (!low_size)
543 			return;
544 	}
545 
546 	low_base = memblock_find_in_range(low_size, (1ULL<<32),
547 					low_size, alignment);
548 
549 	if (!low_base) {
550 		if (!auto_set)
551 			pr_info("crashkernel low reservation failed - No suitable area found.\n");
552 
553 		return;
554 	}
555 
556 	memblock_reserve(low_base, low_size);
557 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
558 			(unsigned long)(low_size >> 20),
559 			(unsigned long)(low_base >> 20),
560 			(unsigned long)(total_low_mem >> 20));
561 	crashk_low_res.start = low_base;
562 	crashk_low_res.end   = low_base + low_size - 1;
563 	insert_resource(&iomem_resource, &crashk_low_res);
564 #endif
565 }
566 
567 static void __init reserve_crashkernel(void)
568 {
569 	const unsigned long long alignment = 16<<20;	/* 16M */
570 	unsigned long long total_mem;
571 	unsigned long long crash_size, crash_base;
572 	bool high = false;
573 	int ret;
574 
575 	total_mem = memblock_phys_mem_size();
576 
577 	/* crashkernel=XM */
578 	ret = parse_crashkernel(boot_command_line, total_mem,
579 			&crash_size, &crash_base);
580 	if (ret != 0 || crash_size <= 0) {
581 		/* crashkernel=X,high */
582 		ret = parse_crashkernel_high(boot_command_line, total_mem,
583 				&crash_size, &crash_base);
584 		if (ret != 0 || crash_size <= 0)
585 			return;
586 		high = true;
587 	}
588 
589 	/* 0 means: find the address automatically */
590 	if (crash_base <= 0) {
591 		/*
592 		 *  kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
593 		 */
594 		crash_base = memblock_find_in_range(alignment,
595 					high ? CRASH_KERNEL_ADDR_HIGH_MAX :
596 					       CRASH_KERNEL_ADDR_LOW_MAX,
597 					crash_size, alignment);
598 
599 		if (!crash_base) {
600 			pr_info("crashkernel reservation failed - No suitable area found.\n");
601 			return;
602 		}
603 
604 	} else {
605 		unsigned long long start;
606 
607 		start = memblock_find_in_range(crash_base,
608 				 crash_base + crash_size, crash_size, 1<<20);
609 		if (start != crash_base) {
610 			pr_info("crashkernel reservation failed - memory is in use.\n");
611 			return;
612 		}
613 	}
614 	memblock_reserve(crash_base, crash_size);
615 
616 	printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
617 			"for crashkernel (System RAM: %ldMB)\n",
618 			(unsigned long)(crash_size >> 20),
619 			(unsigned long)(crash_base >> 20),
620 			(unsigned long)(total_mem >> 20));
621 
622 	crashk_res.start = crash_base;
623 	crashk_res.end   = crash_base + crash_size - 1;
624 	insert_resource(&iomem_resource, &crashk_res);
625 
626 	if (crash_base >= (1ULL<<32))
627 		reserve_crashkernel_low();
628 }
629 #else
630 static void __init reserve_crashkernel(void)
631 {
632 }
633 #endif
634 
635 static struct resource standard_io_resources[] = {
636 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
637 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
638 	{ .name = "pic1", .start = 0x20, .end = 0x21,
639 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
640 	{ .name = "timer0", .start = 0x40, .end = 0x43,
641 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
642 	{ .name = "timer1", .start = 0x50, .end = 0x53,
643 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
644 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
645 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
646 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
647 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
648 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
649 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
650 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
651 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
652 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
653 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
654 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
655 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
656 };
657 
658 void __init reserve_standard_io_resources(void)
659 {
660 	int i;
661 
662 	/* request I/O space for devices used on all i[345]86 PCs */
663 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
664 		request_resource(&ioport_resource, &standard_io_resources[i]);
665 
666 }
667 
668 static __init void reserve_ibft_region(void)
669 {
670 	unsigned long addr, size = 0;
671 
672 	addr = find_ibft_region(&size);
673 
674 	if (size)
675 		memblock_reserve(addr, size);
676 }
677 
678 static bool __init snb_gfx_workaround_needed(void)
679 {
680 #ifdef CONFIG_PCI
681 	int i;
682 	u16 vendor, devid;
683 	static const __initconst u16 snb_ids[] = {
684 		0x0102,
685 		0x0112,
686 		0x0122,
687 		0x0106,
688 		0x0116,
689 		0x0126,
690 		0x010a,
691 	};
692 
693 	/* Assume no if something weird is going on with PCI */
694 	if (!early_pci_allowed())
695 		return false;
696 
697 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
698 	if (vendor != 0x8086)
699 		return false;
700 
701 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
702 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
703 		if (devid == snb_ids[i])
704 			return true;
705 #endif
706 
707 	return false;
708 }
709 
710 /*
711  * Sandy Bridge graphics has trouble with certain ranges, exclude
712  * them from allocation.
713  */
714 static void __init trim_snb_memory(void)
715 {
716 	static const __initconst unsigned long bad_pages[] = {
717 		0x20050000,
718 		0x20110000,
719 		0x20130000,
720 		0x20138000,
721 		0x40004000,
722 	};
723 	int i;
724 
725 	if (!snb_gfx_workaround_needed())
726 		return;
727 
728 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
729 
730 	/*
731 	 * Reserve all memory below the 1 MB mark that has not
732 	 * already been reserved.
733 	 */
734 	memblock_reserve(0, 1<<20);
735 
736 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
737 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
738 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
739 			       bad_pages[i]);
740 	}
741 }
742 
743 /*
744  * Here we put platform-specific memory range workarounds, i.e.
745  * memory known to be corrupt or otherwise in need to be reserved on
746  * specific platforms.
747  *
748  * If this gets used more widely it could use a real dispatch mechanism.
749  */
750 static void __init trim_platform_memory_ranges(void)
751 {
752 	trim_snb_memory();
753 }
754 
755 static void __init trim_bios_range(void)
756 {
757 	/*
758 	 * A special case is the first 4Kb of memory;
759 	 * This is a BIOS owned area, not kernel ram, but generally
760 	 * not listed as such in the E820 table.
761 	 *
762 	 * This typically reserves additional memory (64KiB by default)
763 	 * since some BIOSes are known to corrupt low memory.  See the
764 	 * Kconfig help text for X86_RESERVE_LOW.
765 	 */
766 	e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
767 
768 	/*
769 	 * special case: Some BIOSen report the PC BIOS
770 	 * area (640->1Mb) as ram even though it is not.
771 	 * take them out.
772 	 */
773 	e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
774 
775 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
776 }
777 
778 /* called before trim_bios_range() to spare extra sanitize */
779 static void __init e820_add_kernel_range(void)
780 {
781 	u64 start = __pa_symbol(_text);
782 	u64 size = __pa_symbol(_end) - start;
783 
784 	/*
785 	 * Complain if .text .data and .bss are not marked as E820_RAM and
786 	 * attempt to fix it by adding the range. We may have a confused BIOS,
787 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
788 	 * exclude kernel range. If we really are running on top non-RAM,
789 	 * we will crash later anyways.
790 	 */
791 	if (e820_all_mapped(start, start + size, E820_RAM))
792 		return;
793 
794 	pr_warn(".text .data .bss are not marked as E820_RAM!\n");
795 	e820_remove_range(start, size, E820_RAM, 0);
796 	e820_add_region(start, size, E820_RAM);
797 }
798 
799 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
800 
801 static int __init parse_reservelow(char *p)
802 {
803 	unsigned long long size;
804 
805 	if (!p)
806 		return -EINVAL;
807 
808 	size = memparse(p, &p);
809 
810 	if (size < 4096)
811 		size = 4096;
812 
813 	if (size > 640*1024)
814 		size = 640*1024;
815 
816 	reserve_low = size;
817 
818 	return 0;
819 }
820 
821 early_param("reservelow", parse_reservelow);
822 
823 static void __init trim_low_memory_range(void)
824 {
825 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
826 }
827 
828 /*
829  * Dump out kernel offset information on panic.
830  */
831 static int
832 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
833 {
834 	pr_emerg("Kernel Offset: 0x%lx from 0x%lx "
835 		 "(relocation range: 0x%lx-0x%lx)\n",
836 		 (unsigned long)&_text - __START_KERNEL, __START_KERNEL,
837 		 __START_KERNEL_map, MODULES_VADDR-1);
838 
839 	return 0;
840 }
841 
842 /*
843  * Determine if we were loaded by an EFI loader.  If so, then we have also been
844  * passed the efi memmap, systab, etc., so we should use these data structures
845  * for initialization.  Note, the efi init code path is determined by the
846  * global efi_enabled. This allows the same kernel image to be used on existing
847  * systems (with a traditional BIOS) as well as on EFI systems.
848  */
849 /*
850  * setup_arch - architecture-specific boot-time initializations
851  *
852  * Note: On x86_64, fixmaps are ready for use even before this is called.
853  */
854 
855 void __init setup_arch(char **cmdline_p)
856 {
857 	memblock_reserve(__pa_symbol(_text),
858 			 (unsigned long)__bss_stop - (unsigned long)_text);
859 
860 	early_reserve_initrd();
861 
862 	/*
863 	 * At this point everything still needed from the boot loader
864 	 * or BIOS or kernel text should be early reserved or marked not
865 	 * RAM in e820. All other memory is free game.
866 	 */
867 
868 #ifdef CONFIG_X86_32
869 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
870 
871 	/*
872 	 * copy kernel address range established so far and switch
873 	 * to the proper swapper page table
874 	 */
875 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
876 			initial_page_table + KERNEL_PGD_BOUNDARY,
877 			KERNEL_PGD_PTRS);
878 
879 	load_cr3(swapper_pg_dir);
880 	/*
881 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
882 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
883 	 * will not flush anything because the cpu quirk which clears
884 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
885 	 * load_cr3() above the TLB has been flushed already. The
886 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
887 	 * so proper operation is guaranteed.
888 	 */
889 	__flush_tlb_all();
890 #else
891 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
892 #endif
893 
894 	/*
895 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
896 	 * reserve_top(), so do this before touching the ioremap area.
897 	 */
898 	olpc_ofw_detect();
899 
900 	early_trap_init();
901 	early_cpu_init();
902 	early_ioremap_init();
903 
904 	setup_olpc_ofw_pgd();
905 
906 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
907 	screen_info = boot_params.screen_info;
908 	edid_info = boot_params.edid_info;
909 #ifdef CONFIG_X86_32
910 	apm_info.bios = boot_params.apm_bios_info;
911 	ist_info = boot_params.ist_info;
912 	if (boot_params.sys_desc_table.length != 0) {
913 		machine_id = boot_params.sys_desc_table.table[0];
914 		machine_submodel_id = boot_params.sys_desc_table.table[1];
915 		BIOS_revision = boot_params.sys_desc_table.table[2];
916 	}
917 #endif
918 	saved_video_mode = boot_params.hdr.vid_mode;
919 	bootloader_type = boot_params.hdr.type_of_loader;
920 	if ((bootloader_type >> 4) == 0xe) {
921 		bootloader_type &= 0xf;
922 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
923 	}
924 	bootloader_version  = bootloader_type & 0xf;
925 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
926 
927 #ifdef CONFIG_BLK_DEV_RAM
928 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
929 	rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
930 	rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
931 #endif
932 #ifdef CONFIG_EFI
933 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
934 		     EFI32_LOADER_SIGNATURE, 4)) {
935 		set_bit(EFI_BOOT, &efi.flags);
936 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
937 		     EFI64_LOADER_SIGNATURE, 4)) {
938 		set_bit(EFI_BOOT, &efi.flags);
939 		set_bit(EFI_64BIT, &efi.flags);
940 	}
941 
942 	if (efi_enabled(EFI_BOOT))
943 		efi_memblock_x86_reserve_range();
944 #endif
945 
946 	x86_init.oem.arch_setup();
947 
948 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
949 	setup_memory_map();
950 	parse_setup_data();
951 
952 	copy_edd();
953 
954 	if (!boot_params.hdr.root_flags)
955 		root_mountflags &= ~MS_RDONLY;
956 	init_mm.start_code = (unsigned long) _text;
957 	init_mm.end_code = (unsigned long) _etext;
958 	init_mm.end_data = (unsigned long) _edata;
959 	init_mm.brk = _brk_end;
960 
961 	mpx_mm_init(&init_mm);
962 
963 	code_resource.start = __pa_symbol(_text);
964 	code_resource.end = __pa_symbol(_etext)-1;
965 	data_resource.start = __pa_symbol(_etext);
966 	data_resource.end = __pa_symbol(_edata)-1;
967 	bss_resource.start = __pa_symbol(__bss_start);
968 	bss_resource.end = __pa_symbol(__bss_stop)-1;
969 
970 #ifdef CONFIG_CMDLINE_BOOL
971 #ifdef CONFIG_CMDLINE_OVERRIDE
972 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
973 #else
974 	if (builtin_cmdline[0]) {
975 		/* append boot loader cmdline to builtin */
976 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
977 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
978 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
979 	}
980 #endif
981 #endif
982 
983 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
984 	*cmdline_p = command_line;
985 
986 	/*
987 	 * x86_configure_nx() is called before parse_early_param() to detect
988 	 * whether hardware doesn't support NX (so that the early EHCI debug
989 	 * console setup can safely call set_fixmap()). It may then be called
990 	 * again from within noexec_setup() during parsing early parameters
991 	 * to honor the respective command line option.
992 	 */
993 	x86_configure_nx();
994 
995 	parse_early_param();
996 
997 	x86_report_nx();
998 
999 	/* after early param, so could get panic from serial */
1000 	memblock_x86_reserve_range_setup_data();
1001 
1002 	if (acpi_mps_check()) {
1003 #ifdef CONFIG_X86_LOCAL_APIC
1004 		disable_apic = 1;
1005 #endif
1006 		setup_clear_cpu_cap(X86_FEATURE_APIC);
1007 	}
1008 
1009 #ifdef CONFIG_PCI
1010 	if (pci_early_dump_regs)
1011 		early_dump_pci_devices();
1012 #endif
1013 
1014 	/* update the e820_saved too */
1015 	e820_reserve_setup_data();
1016 	finish_e820_parsing();
1017 
1018 	if (efi_enabled(EFI_BOOT))
1019 		efi_init();
1020 
1021 	dmi_scan_machine();
1022 	dmi_memdev_walk();
1023 	dmi_set_dump_stack_arch_desc();
1024 
1025 	/*
1026 	 * VMware detection requires dmi to be available, so this
1027 	 * needs to be done after dmi_scan_machine, for the BP.
1028 	 */
1029 	init_hypervisor_platform();
1030 
1031 	x86_init.resources.probe_roms();
1032 
1033 	/* after parse_early_param, so could debug it */
1034 	insert_resource(&iomem_resource, &code_resource);
1035 	insert_resource(&iomem_resource, &data_resource);
1036 	insert_resource(&iomem_resource, &bss_resource);
1037 
1038 	e820_add_kernel_range();
1039 	trim_bios_range();
1040 #ifdef CONFIG_X86_32
1041 	if (ppro_with_ram_bug()) {
1042 		e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1043 				  E820_RESERVED);
1044 		sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1045 		printk(KERN_INFO "fixed physical RAM map:\n");
1046 		e820_print_map("bad_ppro");
1047 	}
1048 #else
1049 	early_gart_iommu_check();
1050 #endif
1051 
1052 	/*
1053 	 * partially used pages are not usable - thus
1054 	 * we are rounding upwards:
1055 	 */
1056 	max_pfn = e820_end_of_ram_pfn();
1057 
1058 	/* update e820 for memory not covered by WB MTRRs */
1059 	mtrr_bp_init();
1060 	if (mtrr_trim_uncached_memory(max_pfn))
1061 		max_pfn = e820_end_of_ram_pfn();
1062 
1063 #ifdef CONFIG_X86_32
1064 	/* max_low_pfn get updated here */
1065 	find_low_pfn_range();
1066 #else
1067 	check_x2apic();
1068 
1069 	/* How many end-of-memory variables you have, grandma! */
1070 	/* need this before calling reserve_initrd */
1071 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1072 		max_low_pfn = e820_end_of_low_ram_pfn();
1073 	else
1074 		max_low_pfn = max_pfn;
1075 
1076 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1077 #endif
1078 
1079 	/*
1080 	 * Find and reserve possible boot-time SMP configuration:
1081 	 */
1082 	find_smp_config();
1083 
1084 	reserve_ibft_region();
1085 
1086 	early_alloc_pgt_buf();
1087 
1088 	/*
1089 	 * Need to conclude brk, before memblock_x86_fill()
1090 	 *  it could use memblock_find_in_range, could overlap with
1091 	 *  brk area.
1092 	 */
1093 	reserve_brk();
1094 
1095 	cleanup_highmap();
1096 
1097 	memblock_set_current_limit(ISA_END_ADDRESS);
1098 	memblock_x86_fill();
1099 
1100 	/*
1101 	 * The EFI specification says that boot service code won't be called
1102 	 * after ExitBootServices(). This is, in fact, a lie.
1103 	 */
1104 	if (efi_enabled(EFI_MEMMAP))
1105 		efi_reserve_boot_services();
1106 
1107 	/* preallocate 4k for mptable mpc */
1108 	early_reserve_e820_mpc_new();
1109 
1110 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1111 	setup_bios_corruption_check();
1112 #endif
1113 
1114 #ifdef CONFIG_X86_32
1115 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1116 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1117 #endif
1118 
1119 	reserve_real_mode();
1120 
1121 	trim_platform_memory_ranges();
1122 	trim_low_memory_range();
1123 
1124 	init_mem_mapping();
1125 
1126 	early_trap_pf_init();
1127 
1128 	setup_real_mode();
1129 
1130 	memblock_set_current_limit(get_max_mapped());
1131 
1132 	/*
1133 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1134 	 */
1135 
1136 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1137 	if (init_ohci1394_dma_early)
1138 		init_ohci1394_dma_on_all_controllers();
1139 #endif
1140 	/* Allocate bigger log buffer */
1141 	setup_log_buf(1);
1142 
1143 	reserve_initrd();
1144 
1145 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1146 	acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1147 #endif
1148 
1149 	vsmp_init();
1150 
1151 	io_delay_init();
1152 
1153 	/*
1154 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1155 	 */
1156 	acpi_boot_table_init();
1157 
1158 	early_acpi_boot_init();
1159 
1160 	initmem_init();
1161 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1162 
1163 	/*
1164 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1165 	 * won't consume hotpluggable memory.
1166 	 */
1167 	reserve_crashkernel();
1168 
1169 	memblock_find_dma_reserve();
1170 
1171 #ifdef CONFIG_KVM_GUEST
1172 	kvmclock_init();
1173 #endif
1174 
1175 	x86_init.paging.pagetable_init();
1176 
1177 	if (boot_cpu_data.cpuid_level >= 0) {
1178 		/* A CPU has %cr4 if and only if it has CPUID */
1179 		mmu_cr4_features = read_cr4();
1180 		if (trampoline_cr4_features)
1181 			*trampoline_cr4_features = mmu_cr4_features;
1182 	}
1183 
1184 #ifdef CONFIG_X86_32
1185 	/* sync back kernel address range */
1186 	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1187 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
1188 			KERNEL_PGD_PTRS);
1189 #endif
1190 
1191 	tboot_probe();
1192 
1193 	map_vsyscall();
1194 
1195 	generic_apic_probe();
1196 
1197 	early_quirks();
1198 
1199 	/*
1200 	 * Read APIC and some other early information from ACPI tables.
1201 	 */
1202 	acpi_boot_init();
1203 	sfi_init();
1204 	x86_dtb_init();
1205 
1206 	/*
1207 	 * get boot-time SMP configuration:
1208 	 */
1209 	if (smp_found_config)
1210 		get_smp_config();
1211 
1212 	prefill_possible_map();
1213 
1214 	init_cpu_to_node();
1215 
1216 	init_apic_mappings();
1217 	if (x86_io_apic_ops.init)
1218 		x86_io_apic_ops.init();
1219 
1220 	kvm_guest_init();
1221 
1222 	e820_reserve_resources();
1223 	e820_mark_nosave_regions(max_low_pfn);
1224 
1225 	x86_init.resources.reserve_resources();
1226 
1227 	e820_setup_gap();
1228 
1229 #ifdef CONFIG_VT
1230 #if defined(CONFIG_VGA_CONSOLE)
1231 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1232 		conswitchp = &vga_con;
1233 #elif defined(CONFIG_DUMMY_CONSOLE)
1234 	conswitchp = &dummy_con;
1235 #endif
1236 #endif
1237 	x86_init.oem.banner();
1238 
1239 	x86_init.timers.wallclock_init();
1240 
1241 	mcheck_init();
1242 
1243 	arch_init_ideal_nops();
1244 
1245 	register_refined_jiffies(CLOCK_TICK_RATE);
1246 
1247 #ifdef CONFIG_EFI
1248 	if (efi_enabled(EFI_BOOT))
1249 		efi_apply_memmap_quirks();
1250 #endif
1251 }
1252 
1253 #ifdef CONFIG_X86_32
1254 
1255 static struct resource video_ram_resource = {
1256 	.name	= "Video RAM area",
1257 	.start	= 0xa0000,
1258 	.end	= 0xbffff,
1259 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1260 };
1261 
1262 void __init i386_reserve_resources(void)
1263 {
1264 	request_resource(&iomem_resource, &video_ram_resource);
1265 	reserve_standard_io_resources();
1266 }
1267 
1268 #endif /* CONFIG_X86_32 */
1269 
1270 static struct notifier_block kernel_offset_notifier = {
1271 	.notifier_call = dump_kernel_offset
1272 };
1273 
1274 static int __init register_kernel_offset_dumper(void)
1275 {
1276 	atomic_notifier_chain_register(&panic_notifier_list,
1277 					&kernel_offset_notifier);
1278 	return 0;
1279 }
1280 __initcall(register_kernel_offset_dumper);
1281