xref: /linux/arch/x86/kernel/setup.c (revision a078ccff5642a8fe792a43b3d973bcc3f6dd733f)
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/dmi.h>
86 #include <asm/io_apic.h>
87 #include <asm/ist.h>
88 #include <asm/setup_arch.h>
89 #include <asm/bios_ebda.h>
90 #include <asm/cacheflush.h>
91 #include <asm/processor.h>
92 #include <asm/bugs.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 
115 /*
116  * max_low_pfn_mapped: highest direct mapped pfn under 4GB
117  * max_pfn_mapped:     highest direct mapped pfn over 4GB
118  *
119  * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
120  * represented by pfn_mapped
121  */
122 unsigned long max_low_pfn_mapped;
123 unsigned long max_pfn_mapped;
124 
125 #ifdef CONFIG_DMI
126 RESERVE_BRK(dmi_alloc, 65536);
127 #endif
128 
129 
130 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
131 unsigned long _brk_end = (unsigned long)__brk_base;
132 
133 #ifdef CONFIG_X86_64
134 int default_cpu_present_to_apicid(int mps_cpu)
135 {
136 	return __default_cpu_present_to_apicid(mps_cpu);
137 }
138 
139 int default_check_phys_apicid_present(int phys_apicid)
140 {
141 	return __default_check_phys_apicid_present(phys_apicid);
142 }
143 #endif
144 
145 struct boot_params boot_params;
146 
147 /*
148  * Machine setup..
149  */
150 static struct resource data_resource = {
151 	.name	= "Kernel data",
152 	.start	= 0,
153 	.end	= 0,
154 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
155 };
156 
157 static struct resource code_resource = {
158 	.name	= "Kernel code",
159 	.start	= 0,
160 	.end	= 0,
161 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
162 };
163 
164 static struct resource bss_resource = {
165 	.name	= "Kernel bss",
166 	.start	= 0,
167 	.end	= 0,
168 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
169 };
170 
171 
172 #ifdef CONFIG_X86_32
173 /* cpu data as detected by the assembly code in head.S */
174 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {
175 	.wp_works_ok = -1,
176 	.fdiv_bug = -1,
177 };
178 /* common cpu data for all cpus */
179 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
180 	.wp_works_ok = -1,
181 	.fdiv_bug = -1,
182 };
183 EXPORT_SYMBOL(boot_cpu_data);
184 
185 unsigned int def_to_bigsmp;
186 
187 /* for MCA, but anyone else can use it if they want */
188 unsigned int machine_id;
189 unsigned int machine_submodel_id;
190 unsigned int BIOS_revision;
191 
192 struct apm_info apm_info;
193 EXPORT_SYMBOL(apm_info);
194 
195 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
196 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
197 struct ist_info ist_info;
198 EXPORT_SYMBOL(ist_info);
199 #else
200 struct ist_info ist_info;
201 #endif
202 
203 #else
204 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
205 	.x86_phys_bits = MAX_PHYSMEM_BITS,
206 };
207 EXPORT_SYMBOL(boot_cpu_data);
208 #endif
209 
210 
211 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
212 unsigned long mmu_cr4_features;
213 #else
214 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 */
218 int bootloader_type, bootloader_version;
219 
220 /*
221  * Setup options
222  */
223 struct screen_info screen_info;
224 EXPORT_SYMBOL(screen_info);
225 struct edid_info edid_info;
226 EXPORT_SYMBOL_GPL(edid_info);
227 
228 extern int root_mountflags;
229 
230 unsigned 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 
236 static char __initdata command_line[COMMAND_LINE_SIZE];
237 #ifdef CONFIG_CMDLINE_BOOL
238 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
239 #endif
240 
241 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
242 struct edd edd;
243 #ifdef CONFIG_EDD_MODULE
244 EXPORT_SYMBOL(edd);
245 #endif
246 /**
247  * copy_edd() - Copy the BIOS EDD information
248  *              from boot_params into a safe place.
249  *
250  */
251 static 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
260 static inline void __init copy_edd(void)
261 {
262 }
263 #endif
264 
265 void * __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
285 static void __init cleanup_highmap(void)
286 {
287 }
288 #endif
289 
290 static 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 
301 #ifdef CONFIG_BLK_DEV_INITRD
302 
303 static u64 __init get_ramdisk_image(void)
304 {
305 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
306 
307 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
308 
309 	return ramdisk_image;
310 }
311 static u64 __init get_ramdisk_size(void)
312 {
313 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
314 
315 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
316 
317 	return ramdisk_size;
318 }
319 
320 #define MAX_MAP_CHUNK	(NR_FIX_BTMAPS << PAGE_SHIFT)
321 static void __init relocate_initrd(void)
322 {
323 	/* Assume only end is not page aligned */
324 	u64 ramdisk_image = get_ramdisk_image();
325 	u64 ramdisk_size  = get_ramdisk_size();
326 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
327 	u64 ramdisk_here;
328 	unsigned long slop, clen, mapaddr;
329 	char *p, *q;
330 
331 	/* We need to move the initrd down into directly mapped mem */
332 	ramdisk_here = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
333 						 area_size, PAGE_SIZE);
334 
335 	if (!ramdisk_here)
336 		panic("Cannot find place for new RAMDISK of size %lld\n",
337 			 ramdisk_size);
338 
339 	/* Note: this includes all the mem currently occupied by
340 	   the initrd, we rely on that fact to keep the data intact. */
341 	memblock_reserve(ramdisk_here, area_size);
342 	initrd_start = ramdisk_here + PAGE_OFFSET;
343 	initrd_end   = initrd_start + ramdisk_size;
344 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
345 			 ramdisk_here, ramdisk_here + ramdisk_size - 1);
346 
347 	q = (char *)initrd_start;
348 
349 	/* Copy the initrd */
350 	while (ramdisk_size) {
351 		slop = ramdisk_image & ~PAGE_MASK;
352 		clen = ramdisk_size;
353 		if (clen > MAX_MAP_CHUNK-slop)
354 			clen = MAX_MAP_CHUNK-slop;
355 		mapaddr = ramdisk_image & PAGE_MASK;
356 		p = early_memremap(mapaddr, clen+slop);
357 		memcpy(q, p+slop, clen);
358 		early_iounmap(p, clen+slop);
359 		q += clen;
360 		ramdisk_image += clen;
361 		ramdisk_size  -= clen;
362 	}
363 
364 	ramdisk_image = get_ramdisk_image();
365 	ramdisk_size  = get_ramdisk_size();
366 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
367 		" [mem %#010llx-%#010llx]\n",
368 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
369 		ramdisk_here, ramdisk_here + ramdisk_size - 1);
370 }
371 
372 static void __init early_reserve_initrd(void)
373 {
374 	/* Assume only end is not page aligned */
375 	u64 ramdisk_image = get_ramdisk_image();
376 	u64 ramdisk_size  = get_ramdisk_size();
377 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
378 
379 	if (!boot_params.hdr.type_of_loader ||
380 	    !ramdisk_image || !ramdisk_size)
381 		return;		/* No initrd provided by bootloader */
382 
383 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
384 }
385 static void __init reserve_initrd(void)
386 {
387 	/* Assume only end is not page aligned */
388 	u64 ramdisk_image = get_ramdisk_image();
389 	u64 ramdisk_size  = get_ramdisk_size();
390 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
391 	u64 mapped_size;
392 
393 	if (!boot_params.hdr.type_of_loader ||
394 	    !ramdisk_image || !ramdisk_size)
395 		return;		/* No initrd provided by bootloader */
396 
397 	initrd_start = 0;
398 
399 	mapped_size = memblock_mem_size(max_pfn_mapped);
400 	if (ramdisk_size >= (mapped_size>>1))
401 		panic("initrd too large to handle, "
402 		       "disabling initrd (%lld needed, %lld available)\n",
403 		       ramdisk_size, mapped_size>>1);
404 
405 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
406 			ramdisk_end - 1);
407 
408 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
409 				PFN_DOWN(ramdisk_end))) {
410 		/* All are mapped, easy case */
411 		initrd_start = ramdisk_image + PAGE_OFFSET;
412 		initrd_end = initrd_start + ramdisk_size;
413 		return;
414 	}
415 
416 	relocate_initrd();
417 
418 	memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
419 }
420 #else
421 static void __init early_reserve_initrd(void)
422 {
423 }
424 static void __init reserve_initrd(void)
425 {
426 }
427 #endif /* CONFIG_BLK_DEV_INITRD */
428 
429 static void __init parse_setup_data(void)
430 {
431 	struct setup_data *data;
432 	u64 pa_data;
433 
434 	pa_data = boot_params.hdr.setup_data;
435 	while (pa_data) {
436 		u32 data_len, map_len;
437 
438 		map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
439 			      (u64)sizeof(struct setup_data));
440 		data = early_memremap(pa_data, map_len);
441 		data_len = data->len + sizeof(struct setup_data);
442 		if (data_len > map_len) {
443 			early_iounmap(data, map_len);
444 			data = early_memremap(pa_data, data_len);
445 			map_len = data_len;
446 		}
447 
448 		switch (data->type) {
449 		case SETUP_E820_EXT:
450 			parse_e820_ext(data);
451 			break;
452 		case SETUP_DTB:
453 			add_dtb(pa_data);
454 			break;
455 		default:
456 			break;
457 		}
458 		pa_data = data->next;
459 		early_iounmap(data, map_len);
460 	}
461 }
462 
463 static void __init e820_reserve_setup_data(void)
464 {
465 	struct setup_data *data;
466 	u64 pa_data;
467 	int found = 0;
468 
469 	pa_data = boot_params.hdr.setup_data;
470 	while (pa_data) {
471 		data = early_memremap(pa_data, sizeof(*data));
472 		e820_update_range(pa_data, sizeof(*data)+data->len,
473 			 E820_RAM, E820_RESERVED_KERN);
474 		found = 1;
475 		pa_data = data->next;
476 		early_iounmap(data, sizeof(*data));
477 	}
478 	if (!found)
479 		return;
480 
481 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
482 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
483 	printk(KERN_INFO "extended physical RAM map:\n");
484 	e820_print_map("reserve setup_data");
485 }
486 
487 static void __init memblock_x86_reserve_range_setup_data(void)
488 {
489 	struct setup_data *data;
490 	u64 pa_data;
491 
492 	pa_data = boot_params.hdr.setup_data;
493 	while (pa_data) {
494 		data = early_memremap(pa_data, sizeof(*data));
495 		memblock_reserve(pa_data, sizeof(*data) + data->len);
496 		pa_data = data->next;
497 		early_iounmap(data, sizeof(*data));
498 	}
499 }
500 
501 /*
502  * --------- Crashkernel reservation ------------------------------
503  */
504 
505 #ifdef CONFIG_KEXEC
506 
507 /*
508  * Keep the crash kernel below this limit.  On 32 bits earlier kernels
509  * would limit the kernel to the low 512 MiB due to mapping restrictions.
510  * On 64bit, old kexec-tools need to under 896MiB.
511  */
512 #ifdef CONFIG_X86_32
513 # define CRASH_KERNEL_ADDR_LOW_MAX	(512 << 20)
514 # define CRASH_KERNEL_ADDR_HIGH_MAX	(512 << 20)
515 #else
516 # define CRASH_KERNEL_ADDR_LOW_MAX	(896UL<<20)
517 # define CRASH_KERNEL_ADDR_HIGH_MAX	MAXMEM
518 #endif
519 
520 static void __init reserve_crashkernel_low(void)
521 {
522 #ifdef CONFIG_X86_64
523 	const unsigned long long alignment = 16<<20;	/* 16M */
524 	unsigned long long low_base = 0, low_size = 0;
525 	unsigned long total_low_mem;
526 	unsigned long long base;
527 	bool auto_set = false;
528 	int ret;
529 
530 	total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT));
531 	/* crashkernel=Y,low */
532 	ret = parse_crashkernel_low(boot_command_line, total_low_mem,
533 						&low_size, &base);
534 	if (ret != 0) {
535 		/*
536 		 * two parts from lib/swiotlb.c:
537 		 *	swiotlb size: user specified with swiotlb= or default.
538 		 *	swiotlb overflow buffer: now is hardcoded to 32k.
539 		 *		We round it to 8M for other buffers that
540 		 *		may need to stay low too.
541 		 */
542 		low_size = swiotlb_size_or_default() + (8UL<<20);
543 		auto_set = true;
544 	} else {
545 		/* passed with crashkernel=0,low ? */
546 		if (!low_size)
547 			return;
548 	}
549 
550 	low_base = memblock_find_in_range(low_size, (1ULL<<32),
551 					low_size, alignment);
552 
553 	if (!low_base) {
554 		if (!auto_set)
555 			pr_info("crashkernel low reservation failed - No suitable area found.\n");
556 
557 		return;
558 	}
559 
560 	memblock_reserve(low_base, low_size);
561 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
562 			(unsigned long)(low_size >> 20),
563 			(unsigned long)(low_base >> 20),
564 			(unsigned long)(total_low_mem >> 20));
565 	crashk_low_res.start = low_base;
566 	crashk_low_res.end   = low_base + low_size - 1;
567 	insert_resource(&iomem_resource, &crashk_low_res);
568 #endif
569 }
570 
571 static void __init reserve_crashkernel(void)
572 {
573 	const unsigned long long alignment = 16<<20;	/* 16M */
574 	unsigned long long total_mem;
575 	unsigned long long crash_size, crash_base;
576 	bool high = false;
577 	int ret;
578 
579 	total_mem = memblock_phys_mem_size();
580 
581 	/* crashkernel=XM */
582 	ret = parse_crashkernel(boot_command_line, total_mem,
583 			&crash_size, &crash_base);
584 	if (ret != 0 || crash_size <= 0) {
585 		/* crashkernel=X,high */
586 		ret = parse_crashkernel_high(boot_command_line, total_mem,
587 				&crash_size, &crash_base);
588 		if (ret != 0 || crash_size <= 0)
589 			return;
590 		high = true;
591 	}
592 
593 	/* 0 means: find the address automatically */
594 	if (crash_base <= 0) {
595 		/*
596 		 *  kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
597 		 */
598 		crash_base = memblock_find_in_range(alignment,
599 					high ? CRASH_KERNEL_ADDR_HIGH_MAX :
600 					       CRASH_KERNEL_ADDR_LOW_MAX,
601 					crash_size, alignment);
602 
603 		if (!crash_base) {
604 			pr_info("crashkernel reservation failed - No suitable area found.\n");
605 			return;
606 		}
607 
608 	} else {
609 		unsigned long long start;
610 
611 		start = memblock_find_in_range(crash_base,
612 				 crash_base + crash_size, crash_size, 1<<20);
613 		if (start != crash_base) {
614 			pr_info("crashkernel reservation failed - memory is in use.\n");
615 			return;
616 		}
617 	}
618 	memblock_reserve(crash_base, crash_size);
619 
620 	printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
621 			"for crashkernel (System RAM: %ldMB)\n",
622 			(unsigned long)(crash_size >> 20),
623 			(unsigned long)(crash_base >> 20),
624 			(unsigned long)(total_mem >> 20));
625 
626 	crashk_res.start = crash_base;
627 	crashk_res.end   = crash_base + crash_size - 1;
628 	insert_resource(&iomem_resource, &crashk_res);
629 
630 	if (crash_base >= (1ULL<<32))
631 		reserve_crashkernel_low();
632 }
633 #else
634 static void __init reserve_crashkernel(void)
635 {
636 }
637 #endif
638 
639 static struct resource standard_io_resources[] = {
640 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
641 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
642 	{ .name = "pic1", .start = 0x20, .end = 0x21,
643 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
644 	{ .name = "timer0", .start = 0x40, .end = 0x43,
645 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
646 	{ .name = "timer1", .start = 0x50, .end = 0x53,
647 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
648 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
649 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
650 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
651 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
652 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
653 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
654 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
655 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
656 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
657 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
658 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
659 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
660 };
661 
662 void __init reserve_standard_io_resources(void)
663 {
664 	int i;
665 
666 	/* request I/O space for devices used on all i[345]86 PCs */
667 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
668 		request_resource(&ioport_resource, &standard_io_resources[i]);
669 
670 }
671 
672 static __init void reserve_ibft_region(void)
673 {
674 	unsigned long addr, size = 0;
675 
676 	addr = find_ibft_region(&size);
677 
678 	if (size)
679 		memblock_reserve(addr, size);
680 }
681 
682 static bool __init snb_gfx_workaround_needed(void)
683 {
684 #ifdef CONFIG_PCI
685 	int i;
686 	u16 vendor, devid;
687 	static const __initconst u16 snb_ids[] = {
688 		0x0102,
689 		0x0112,
690 		0x0122,
691 		0x0106,
692 		0x0116,
693 		0x0126,
694 		0x010a,
695 	};
696 
697 	/* Assume no if something weird is going on with PCI */
698 	if (!early_pci_allowed())
699 		return false;
700 
701 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
702 	if (vendor != 0x8086)
703 		return false;
704 
705 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
706 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
707 		if (devid == snb_ids[i])
708 			return true;
709 #endif
710 
711 	return false;
712 }
713 
714 /*
715  * Sandy Bridge graphics has trouble with certain ranges, exclude
716  * them from allocation.
717  */
718 static void __init trim_snb_memory(void)
719 {
720 	static const __initconst unsigned long bad_pages[] = {
721 		0x20050000,
722 		0x20110000,
723 		0x20130000,
724 		0x20138000,
725 		0x40004000,
726 	};
727 	int i;
728 
729 	if (!snb_gfx_workaround_needed())
730 		return;
731 
732 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
733 
734 	/*
735 	 * Reserve all memory below the 1 MB mark that has not
736 	 * already been reserved.
737 	 */
738 	memblock_reserve(0, 1<<20);
739 
740 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
741 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
742 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
743 			       bad_pages[i]);
744 	}
745 }
746 
747 /*
748  * Here we put platform-specific memory range workarounds, i.e.
749  * memory known to be corrupt or otherwise in need to be reserved on
750  * specific platforms.
751  *
752  * If this gets used more widely it could use a real dispatch mechanism.
753  */
754 static void __init trim_platform_memory_ranges(void)
755 {
756 	trim_snb_memory();
757 }
758 
759 static void __init trim_bios_range(void)
760 {
761 	/*
762 	 * A special case is the first 4Kb of memory;
763 	 * This is a BIOS owned area, not kernel ram, but generally
764 	 * not listed as such in the E820 table.
765 	 *
766 	 * This typically reserves additional memory (64KiB by default)
767 	 * since some BIOSes are known to corrupt low memory.  See the
768 	 * Kconfig help text for X86_RESERVE_LOW.
769 	 */
770 	e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
771 
772 	/*
773 	 * special case: Some BIOSen report the PC BIOS
774 	 * area (640->1Mb) as ram even though it is not.
775 	 * take them out.
776 	 */
777 	e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
778 
779 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
780 }
781 
782 /* called before trim_bios_range() to spare extra sanitize */
783 static void __init e820_add_kernel_range(void)
784 {
785 	u64 start = __pa_symbol(_text);
786 	u64 size = __pa_symbol(_end) - start;
787 
788 	/*
789 	 * Complain if .text .data and .bss are not marked as E820_RAM and
790 	 * attempt to fix it by adding the range. We may have a confused BIOS,
791 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
792 	 * exclude kernel range. If we really are running on top non-RAM,
793 	 * we will crash later anyways.
794 	 */
795 	if (e820_all_mapped(start, start + size, E820_RAM))
796 		return;
797 
798 	pr_warn(".text .data .bss are not marked as E820_RAM!\n");
799 	e820_remove_range(start, size, E820_RAM, 0);
800 	e820_add_region(start, size, E820_RAM);
801 }
802 
803 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
804 
805 static int __init parse_reservelow(char *p)
806 {
807 	unsigned long long size;
808 
809 	if (!p)
810 		return -EINVAL;
811 
812 	size = memparse(p, &p);
813 
814 	if (size < 4096)
815 		size = 4096;
816 
817 	if (size > 640*1024)
818 		size = 640*1024;
819 
820 	reserve_low = size;
821 
822 	return 0;
823 }
824 
825 early_param("reservelow", parse_reservelow);
826 
827 static void __init trim_low_memory_range(void)
828 {
829 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
830 }
831 
832 /*
833  * Determine if we were loaded by an EFI loader.  If so, then we have also been
834  * passed the efi memmap, systab, etc., so we should use these data structures
835  * for initialization.  Note, the efi init code path is determined by the
836  * global efi_enabled. This allows the same kernel image to be used on existing
837  * systems (with a traditional BIOS) as well as on EFI systems.
838  */
839 /*
840  * setup_arch - architecture-specific boot-time initializations
841  *
842  * Note: On x86_64, fixmaps are ready for use even before this is called.
843  */
844 
845 void __init setup_arch(char **cmdline_p)
846 {
847 	memblock_reserve(__pa_symbol(_text),
848 			 (unsigned long)__bss_stop - (unsigned long)_text);
849 
850 	early_reserve_initrd();
851 
852 	/*
853 	 * At this point everything still needed from the boot loader
854 	 * or BIOS or kernel text should be early reserved or marked not
855 	 * RAM in e820. All other memory is free game.
856 	 */
857 
858 #ifdef CONFIG_X86_32
859 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
860 	visws_early_detect();
861 
862 	/*
863 	 * copy kernel address range established so far and switch
864 	 * to the proper swapper page table
865 	 */
866 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
867 			initial_page_table + KERNEL_PGD_BOUNDARY,
868 			KERNEL_PGD_PTRS);
869 
870 	load_cr3(swapper_pg_dir);
871 	__flush_tlb_all();
872 #else
873 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
874 #endif
875 
876 	/*
877 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
878 	 * reserve_top(), so do this before touching the ioremap area.
879 	 */
880 	olpc_ofw_detect();
881 
882 	early_trap_init();
883 	early_cpu_init();
884 	early_ioremap_init();
885 
886 	setup_olpc_ofw_pgd();
887 
888 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
889 	screen_info = boot_params.screen_info;
890 	edid_info = boot_params.edid_info;
891 #ifdef CONFIG_X86_32
892 	apm_info.bios = boot_params.apm_bios_info;
893 	ist_info = boot_params.ist_info;
894 	if (boot_params.sys_desc_table.length != 0) {
895 		machine_id = boot_params.sys_desc_table.table[0];
896 		machine_submodel_id = boot_params.sys_desc_table.table[1];
897 		BIOS_revision = boot_params.sys_desc_table.table[2];
898 	}
899 #endif
900 	saved_video_mode = boot_params.hdr.vid_mode;
901 	bootloader_type = boot_params.hdr.type_of_loader;
902 	if ((bootloader_type >> 4) == 0xe) {
903 		bootloader_type &= 0xf;
904 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
905 	}
906 	bootloader_version  = bootloader_type & 0xf;
907 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
908 
909 #ifdef CONFIG_BLK_DEV_RAM
910 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
911 	rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
912 	rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
913 #endif
914 #ifdef CONFIG_EFI
915 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
916 		     "EL32", 4)) {
917 		set_bit(EFI_BOOT, &x86_efi_facility);
918 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
919 		     "EL64", 4)) {
920 		set_bit(EFI_BOOT, &x86_efi_facility);
921 		set_bit(EFI_64BIT, &x86_efi_facility);
922 	}
923 
924 	if (efi_enabled(EFI_BOOT))
925 		efi_memblock_x86_reserve_range();
926 #endif
927 
928 	x86_init.oem.arch_setup();
929 
930 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
931 	setup_memory_map();
932 	parse_setup_data();
933 	/* update the e820_saved too */
934 	e820_reserve_setup_data();
935 
936 	copy_edd();
937 
938 	if (!boot_params.hdr.root_flags)
939 		root_mountflags &= ~MS_RDONLY;
940 	init_mm.start_code = (unsigned long) _text;
941 	init_mm.end_code = (unsigned long) _etext;
942 	init_mm.end_data = (unsigned long) _edata;
943 	init_mm.brk = _brk_end;
944 
945 	code_resource.start = __pa_symbol(_text);
946 	code_resource.end = __pa_symbol(_etext)-1;
947 	data_resource.start = __pa_symbol(_etext);
948 	data_resource.end = __pa_symbol(_edata)-1;
949 	bss_resource.start = __pa_symbol(__bss_start);
950 	bss_resource.end = __pa_symbol(__bss_stop)-1;
951 
952 #ifdef CONFIG_CMDLINE_BOOL
953 #ifdef CONFIG_CMDLINE_OVERRIDE
954 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
955 #else
956 	if (builtin_cmdline[0]) {
957 		/* append boot loader cmdline to builtin */
958 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
959 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
960 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
961 	}
962 #endif
963 #endif
964 
965 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
966 	*cmdline_p = command_line;
967 
968 	/*
969 	 * x86_configure_nx() is called before parse_early_param() to detect
970 	 * whether hardware doesn't support NX (so that the early EHCI debug
971 	 * console setup can safely call set_fixmap()). It may then be called
972 	 * again from within noexec_setup() during parsing early parameters
973 	 * to honor the respective command line option.
974 	 */
975 	x86_configure_nx();
976 
977 	parse_early_param();
978 
979 	x86_report_nx();
980 
981 	/* after early param, so could get panic from serial */
982 	memblock_x86_reserve_range_setup_data();
983 
984 	if (acpi_mps_check()) {
985 #ifdef CONFIG_X86_LOCAL_APIC
986 		disable_apic = 1;
987 #endif
988 		setup_clear_cpu_cap(X86_FEATURE_APIC);
989 	}
990 
991 #ifdef CONFIG_PCI
992 	if (pci_early_dump_regs)
993 		early_dump_pci_devices();
994 #endif
995 
996 	finish_e820_parsing();
997 
998 	if (efi_enabled(EFI_BOOT))
999 		efi_init();
1000 
1001 	dmi_scan_machine();
1002 
1003 	/*
1004 	 * VMware detection requires dmi to be available, so this
1005 	 * needs to be done after dmi_scan_machine, for the BP.
1006 	 */
1007 	init_hypervisor_platform();
1008 
1009 	x86_init.resources.probe_roms();
1010 
1011 	/* after parse_early_param, so could debug it */
1012 	insert_resource(&iomem_resource, &code_resource);
1013 	insert_resource(&iomem_resource, &data_resource);
1014 	insert_resource(&iomem_resource, &bss_resource);
1015 
1016 	e820_add_kernel_range();
1017 	trim_bios_range();
1018 #ifdef CONFIG_X86_32
1019 	if (ppro_with_ram_bug()) {
1020 		e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1021 				  E820_RESERVED);
1022 		sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1023 		printk(KERN_INFO "fixed physical RAM map:\n");
1024 		e820_print_map("bad_ppro");
1025 	}
1026 #else
1027 	early_gart_iommu_check();
1028 #endif
1029 
1030 	/*
1031 	 * partially used pages are not usable - thus
1032 	 * we are rounding upwards:
1033 	 */
1034 	max_pfn = e820_end_of_ram_pfn();
1035 
1036 	/* update e820 for memory not covered by WB MTRRs */
1037 	mtrr_bp_init();
1038 	if (mtrr_trim_uncached_memory(max_pfn))
1039 		max_pfn = e820_end_of_ram_pfn();
1040 
1041 #ifdef CONFIG_X86_32
1042 	/* max_low_pfn get updated here */
1043 	find_low_pfn_range();
1044 #else
1045 	num_physpages = max_pfn;
1046 
1047 	check_x2apic();
1048 
1049 	/* How many end-of-memory variables you have, grandma! */
1050 	/* need this before calling reserve_initrd */
1051 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1052 		max_low_pfn = e820_end_of_low_ram_pfn();
1053 	else
1054 		max_low_pfn = max_pfn;
1055 
1056 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1057 #endif
1058 
1059 	/*
1060 	 * Find and reserve possible boot-time SMP configuration:
1061 	 */
1062 	find_smp_config();
1063 
1064 	reserve_ibft_region();
1065 
1066 	early_alloc_pgt_buf();
1067 
1068 	/*
1069 	 * Need to conclude brk, before memblock_x86_fill()
1070 	 *  it could use memblock_find_in_range, could overlap with
1071 	 *  brk area.
1072 	 */
1073 	reserve_brk();
1074 
1075 	cleanup_highmap();
1076 
1077 	memblock.current_limit = ISA_END_ADDRESS;
1078 	memblock_x86_fill();
1079 
1080 	/*
1081 	 * The EFI specification says that boot service code won't be called
1082 	 * after ExitBootServices(). This is, in fact, a lie.
1083 	 */
1084 	if (efi_enabled(EFI_MEMMAP))
1085 		efi_reserve_boot_services();
1086 
1087 	/* preallocate 4k for mptable mpc */
1088 	early_reserve_e820_mpc_new();
1089 
1090 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1091 	setup_bios_corruption_check();
1092 #endif
1093 
1094 #ifdef CONFIG_X86_32
1095 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1096 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1097 #endif
1098 
1099 	reserve_real_mode();
1100 
1101 	trim_platform_memory_ranges();
1102 	trim_low_memory_range();
1103 
1104 	init_mem_mapping();
1105 
1106 	early_trap_pf_init();
1107 
1108 	setup_real_mode();
1109 
1110 	memblock.current_limit = get_max_mapped();
1111 	dma_contiguous_reserve(0);
1112 
1113 	/*
1114 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1115 	 */
1116 
1117 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1118 	if (init_ohci1394_dma_early)
1119 		init_ohci1394_dma_on_all_controllers();
1120 #endif
1121 	/* Allocate bigger log buffer */
1122 	setup_log_buf(1);
1123 
1124 	reserve_initrd();
1125 
1126 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1127 	acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1128 #endif
1129 
1130 	reserve_crashkernel();
1131 
1132 	vsmp_init();
1133 
1134 	io_delay_init();
1135 
1136 	/*
1137 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1138 	 */
1139 	acpi_boot_table_init();
1140 
1141 	early_acpi_boot_init();
1142 
1143 	initmem_init();
1144 	memblock_find_dma_reserve();
1145 
1146 #ifdef CONFIG_KVM_GUEST
1147 	kvmclock_init();
1148 #endif
1149 
1150 	x86_init.paging.pagetable_init();
1151 
1152 	if (boot_cpu_data.cpuid_level >= 0) {
1153 		/* A CPU has %cr4 if and only if it has CPUID */
1154 		mmu_cr4_features = read_cr4();
1155 		if (trampoline_cr4_features)
1156 			*trampoline_cr4_features = mmu_cr4_features;
1157 	}
1158 
1159 #ifdef CONFIG_X86_32
1160 	/* sync back kernel address range */
1161 	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1162 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
1163 			KERNEL_PGD_PTRS);
1164 #endif
1165 
1166 	tboot_probe();
1167 
1168 #ifdef CONFIG_X86_64
1169 	map_vsyscall();
1170 #endif
1171 
1172 	generic_apic_probe();
1173 
1174 	early_quirks();
1175 
1176 	/*
1177 	 * Read APIC and some other early information from ACPI tables.
1178 	 */
1179 	acpi_boot_init();
1180 	sfi_init();
1181 	x86_dtb_init();
1182 
1183 	/*
1184 	 * get boot-time SMP configuration:
1185 	 */
1186 	if (smp_found_config)
1187 		get_smp_config();
1188 
1189 	prefill_possible_map();
1190 
1191 	init_cpu_to_node();
1192 
1193 	init_apic_mappings();
1194 	if (x86_io_apic_ops.init)
1195 		x86_io_apic_ops.init();
1196 
1197 	kvm_guest_init();
1198 
1199 	e820_reserve_resources();
1200 	e820_mark_nosave_regions(max_low_pfn);
1201 
1202 	x86_init.resources.reserve_resources();
1203 
1204 	e820_setup_gap();
1205 
1206 #ifdef CONFIG_VT
1207 #if defined(CONFIG_VGA_CONSOLE)
1208 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1209 		conswitchp = &vga_con;
1210 #elif defined(CONFIG_DUMMY_CONSOLE)
1211 	conswitchp = &dummy_con;
1212 #endif
1213 #endif
1214 	x86_init.oem.banner();
1215 
1216 	x86_init.timers.wallclock_init();
1217 
1218 	mcheck_init();
1219 
1220 	arch_init_ideal_nops();
1221 
1222 	register_refined_jiffies(CLOCK_TICK_RATE);
1223 
1224 #ifdef CONFIG_EFI
1225 	/* Once setup is done above, unmap the EFI memory map on
1226 	 * mismatched firmware/kernel archtectures since there is no
1227 	 * support for runtime services.
1228 	 */
1229 	if (efi_enabled(EFI_BOOT) && !efi_is_native()) {
1230 		pr_info("efi: Setup done, disabling due to 32/64-bit mismatch\n");
1231 		efi_unmap_memmap();
1232 	}
1233 #endif
1234 }
1235 
1236 #ifdef CONFIG_X86_32
1237 
1238 static struct resource video_ram_resource = {
1239 	.name	= "Video RAM area",
1240 	.start	= 0xa0000,
1241 	.end	= 0xbffff,
1242 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1243 };
1244 
1245 void __init i386_reserve_resources(void)
1246 {
1247 	request_resource(&iomem_resource, &video_ram_resource);
1248 	reserve_standard_io_resources();
1249 }
1250 
1251 #endif /* CONFIG_X86_32 */
1252