xref: /openbmc/linux/arch/x86/kernel/setup.c (revision 45471cd98decae5fced8b38e46c223f54a924814)
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 
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 = {
175 	.wp_works_ok = -1,
176 };
177 /* common cpu data for all cpus */
178 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
179 	.wp_works_ok = -1,
180 };
181 EXPORT_SYMBOL(boot_cpu_data);
182 
183 unsigned int def_to_bigsmp;
184 
185 /* for MCA, but anyone else can use it if they want */
186 unsigned int machine_id;
187 unsigned int machine_submodel_id;
188 unsigned int BIOS_revision;
189 
190 struct apm_info apm_info;
191 EXPORT_SYMBOL(apm_info);
192 
193 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
194 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
195 struct ist_info ist_info;
196 EXPORT_SYMBOL(ist_info);
197 #else
198 struct ist_info ist_info;
199 #endif
200 
201 #else
202 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
203 	.x86_phys_bits = MAX_PHYSMEM_BITS,
204 };
205 EXPORT_SYMBOL(boot_cpu_data);
206 #endif
207 
208 
209 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
210 __visible unsigned long mmu_cr4_features;
211 #else
212 __visible unsigned long mmu_cr4_features = X86_CR4_PAE;
213 #endif
214 
215 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
216 int bootloader_type, bootloader_version;
217 
218 /*
219  * Setup options
220  */
221 struct screen_info screen_info;
222 EXPORT_SYMBOL(screen_info);
223 struct edid_info edid_info;
224 EXPORT_SYMBOL_GPL(edid_info);
225 
226 extern int root_mountflags;
227 
228 unsigned long saved_video_mode;
229 
230 #define RAMDISK_IMAGE_START_MASK	0x07FF
231 #define RAMDISK_PROMPT_FLAG		0x8000
232 #define RAMDISK_LOAD_FLAG		0x4000
233 
234 static char __initdata command_line[COMMAND_LINE_SIZE];
235 #ifdef CONFIG_CMDLINE_BOOL
236 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
237 #endif
238 
239 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
240 struct edd edd;
241 #ifdef CONFIG_EDD_MODULE
242 EXPORT_SYMBOL(edd);
243 #endif
244 /**
245  * copy_edd() - Copy the BIOS EDD information
246  *              from boot_params into a safe place.
247  *
248  */
249 static inline void __init copy_edd(void)
250 {
251      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
252 	    sizeof(edd.mbr_signature));
253      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
254      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
255      edd.edd_info_nr = boot_params.eddbuf_entries;
256 }
257 #else
258 static inline void __init copy_edd(void)
259 {
260 }
261 #endif
262 
263 void * __init extend_brk(size_t size, size_t align)
264 {
265 	size_t mask = align - 1;
266 	void *ret;
267 
268 	BUG_ON(_brk_start == 0);
269 	BUG_ON(align & mask);
270 
271 	_brk_end = (_brk_end + mask) & ~mask;
272 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
273 
274 	ret = (void *)_brk_end;
275 	_brk_end += size;
276 
277 	memset(ret, 0, size);
278 
279 	return ret;
280 }
281 
282 #ifdef CONFIG_X86_32
283 static void __init cleanup_highmap(void)
284 {
285 }
286 #endif
287 
288 static void __init reserve_brk(void)
289 {
290 	if (_brk_end > _brk_start)
291 		memblock_reserve(__pa_symbol(_brk_start),
292 				 _brk_end - _brk_start);
293 
294 	/* Mark brk area as locked down and no longer taking any
295 	   new allocations */
296 	_brk_start = 0;
297 }
298 
299 u64 relocated_ramdisk;
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 	unsigned long slop, clen, mapaddr;
328 	char *p, *q;
329 
330 	/* We need to move the initrd down into directly mapped mem */
331 	relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
332 						   area_size, PAGE_SIZE);
333 
334 	if (!relocated_ramdisk)
335 		panic("Cannot find place for new RAMDISK of size %lld\n",
336 		      ramdisk_size);
337 
338 	/* Note: this includes all the mem currently occupied by
339 	   the initrd, we rely on that fact to keep the data intact. */
340 	memblock_reserve(relocated_ramdisk, area_size);
341 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
342 	initrd_end   = initrd_start + ramdisk_size;
343 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
344 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
345 
346 	q = (char *)initrd_start;
347 
348 	/* Copy the initrd */
349 	while (ramdisk_size) {
350 		slop = ramdisk_image & ~PAGE_MASK;
351 		clen = ramdisk_size;
352 		if (clen > MAX_MAP_CHUNK-slop)
353 			clen = MAX_MAP_CHUNK-slop;
354 		mapaddr = ramdisk_image & PAGE_MASK;
355 		p = early_memremap(mapaddr, clen+slop);
356 		memcpy(q, p+slop, clen);
357 		early_memunmap(p, clen+slop);
358 		q += clen;
359 		ramdisk_image += clen;
360 		ramdisk_size  -= clen;
361 	}
362 
363 	ramdisk_image = get_ramdisk_image();
364 	ramdisk_size  = get_ramdisk_size();
365 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
366 		" [mem %#010llx-%#010llx]\n",
367 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
368 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
369 }
370 
371 static void __init early_reserve_initrd(void)
372 {
373 	/* Assume only end is not page aligned */
374 	u64 ramdisk_image = get_ramdisk_image();
375 	u64 ramdisk_size  = get_ramdisk_size();
376 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
377 
378 	if (!boot_params.hdr.type_of_loader ||
379 	    !ramdisk_image || !ramdisk_size)
380 		return;		/* No initrd provided by bootloader */
381 
382 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
383 }
384 static void __init reserve_initrd(void)
385 {
386 	/* Assume only end is not page aligned */
387 	u64 ramdisk_image = get_ramdisk_image();
388 	u64 ramdisk_size  = get_ramdisk_size();
389 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
390 	u64 mapped_size;
391 
392 	if (!boot_params.hdr.type_of_loader ||
393 	    !ramdisk_image || !ramdisk_size)
394 		return;		/* No initrd provided by bootloader */
395 
396 	initrd_start = 0;
397 
398 	mapped_size = memblock_mem_size(max_pfn_mapped);
399 	if (ramdisk_size >= (mapped_size>>1))
400 		panic("initrd too large to handle, "
401 		       "disabling initrd (%lld needed, %lld available)\n",
402 		       ramdisk_size, mapped_size>>1);
403 
404 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
405 			ramdisk_end - 1);
406 
407 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
408 				PFN_DOWN(ramdisk_end))) {
409 		/* All are mapped, easy case */
410 		initrd_start = ramdisk_image + PAGE_OFFSET;
411 		initrd_end = initrd_start + ramdisk_size;
412 		return;
413 	}
414 
415 	relocate_initrd();
416 
417 	memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
418 }
419 #else
420 static void __init early_reserve_initrd(void)
421 {
422 }
423 static void __init reserve_initrd(void)
424 {
425 }
426 #endif /* CONFIG_BLK_DEV_INITRD */
427 
428 static void __init parse_setup_data(void)
429 {
430 	struct setup_data *data;
431 	u64 pa_data, pa_next;
432 
433 	pa_data = boot_params.hdr.setup_data;
434 	while (pa_data) {
435 		u32 data_len, data_type;
436 
437 		data = early_memremap(pa_data, sizeof(*data));
438 		data_len = data->len + sizeof(struct setup_data);
439 		data_type = data->type;
440 		pa_next = data->next;
441 		early_memunmap(data, sizeof(*data));
442 
443 		switch (data_type) {
444 		case SETUP_E820_EXT:
445 			parse_e820_ext(pa_data, data_len);
446 			break;
447 		case SETUP_DTB:
448 			add_dtb(pa_data);
449 			break;
450 		case SETUP_EFI:
451 			parse_efi_setup(pa_data, data_len);
452 			break;
453 		default:
454 			break;
455 		}
456 		pa_data = pa_next;
457 	}
458 }
459 
460 static void __init e820_reserve_setup_data(void)
461 {
462 	struct setup_data *data;
463 	u64 pa_data;
464 	int found = 0;
465 
466 	pa_data = boot_params.hdr.setup_data;
467 	while (pa_data) {
468 		data = early_memremap(pa_data, sizeof(*data));
469 		e820_update_range(pa_data, sizeof(*data)+data->len,
470 			 E820_RAM, E820_RESERVED_KERN);
471 		found = 1;
472 		pa_data = data->next;
473 		early_memunmap(data, sizeof(*data));
474 	}
475 	if (!found)
476 		return;
477 
478 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
479 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
480 	printk(KERN_INFO "extended physical RAM map:\n");
481 	e820_print_map("reserve setup_data");
482 }
483 
484 static void __init memblock_x86_reserve_range_setup_data(void)
485 {
486 	struct setup_data *data;
487 	u64 pa_data;
488 
489 	pa_data = boot_params.hdr.setup_data;
490 	while (pa_data) {
491 		data = early_memremap(pa_data, sizeof(*data));
492 		memblock_reserve(pa_data, sizeof(*data) + data->len);
493 		pa_data = data->next;
494 		early_memunmap(data, sizeof(*data));
495 	}
496 }
497 
498 /*
499  * --------- Crashkernel reservation ------------------------------
500  */
501 
502 #ifdef CONFIG_KEXEC
503 
504 /*
505  * Keep the crash kernel below this limit.  On 32 bits earlier kernels
506  * would limit the kernel to the low 512 MiB due to mapping restrictions.
507  * On 64bit, old kexec-tools need to under 896MiB.
508  */
509 #ifdef CONFIG_X86_32
510 # define CRASH_KERNEL_ADDR_LOW_MAX	(512 << 20)
511 # define CRASH_KERNEL_ADDR_HIGH_MAX	(512 << 20)
512 #else
513 # define CRASH_KERNEL_ADDR_LOW_MAX	(896UL<<20)
514 # define CRASH_KERNEL_ADDR_HIGH_MAX	MAXMEM
515 #endif
516 
517 static void __init reserve_crashkernel_low(void)
518 {
519 #ifdef CONFIG_X86_64
520 	const unsigned long long alignment = 16<<20;	/* 16M */
521 	unsigned long long low_base = 0, low_size = 0;
522 	unsigned long total_low_mem;
523 	unsigned long long base;
524 	bool auto_set = false;
525 	int ret;
526 
527 	total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT));
528 	/* crashkernel=Y,low */
529 	ret = parse_crashkernel_low(boot_command_line, total_low_mem,
530 						&low_size, &base);
531 	if (ret != 0) {
532 		/*
533 		 * two parts from lib/swiotlb.c:
534 		 * -swiotlb size: user-specified with swiotlb= or default.
535 		 *
536 		 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
537 		 * to 8M for other buffers that may need to stay low too. Also
538 		 * make sure we allocate enough extra low memory so that we
539 		 * don't run out of DMA buffers for 32-bit devices.
540 		 */
541 		low_size = max(swiotlb_size_or_default() + (8UL<<20), 256UL<<20);
542 		auto_set = true;
543 	} else {
544 		/* passed with crashkernel=0,low ? */
545 		if (!low_size)
546 			return;
547 	}
548 
549 	low_base = memblock_find_in_range(low_size, (1ULL<<32),
550 					low_size, alignment);
551 
552 	if (!low_base) {
553 		if (!auto_set)
554 			pr_info("crashkernel low reservation failed - No suitable area found.\n");
555 
556 		return;
557 	}
558 
559 	memblock_reserve(low_base, low_size);
560 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
561 			(unsigned long)(low_size >> 20),
562 			(unsigned long)(low_base >> 20),
563 			(unsigned long)(total_low_mem >> 20));
564 	crashk_low_res.start = low_base;
565 	crashk_low_res.end   = low_base + low_size - 1;
566 	insert_resource(&iomem_resource, &crashk_low_res);
567 #endif
568 }
569 
570 static void __init reserve_crashkernel(void)
571 {
572 	const unsigned long long alignment = 16<<20;	/* 16M */
573 	unsigned long long total_mem;
574 	unsigned long long crash_size, crash_base;
575 	bool high = false;
576 	int ret;
577 
578 	total_mem = memblock_phys_mem_size();
579 
580 	/* crashkernel=XM */
581 	ret = parse_crashkernel(boot_command_line, total_mem,
582 			&crash_size, &crash_base);
583 	if (ret != 0 || crash_size <= 0) {
584 		/* crashkernel=X,high */
585 		ret = parse_crashkernel_high(boot_command_line, total_mem,
586 				&crash_size, &crash_base);
587 		if (ret != 0 || crash_size <= 0)
588 			return;
589 		high = true;
590 	}
591 
592 	/* 0 means: find the address automatically */
593 	if (crash_base <= 0) {
594 		/*
595 		 *  kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
596 		 */
597 		crash_base = memblock_find_in_range(alignment,
598 					high ? CRASH_KERNEL_ADDR_HIGH_MAX :
599 					       CRASH_KERNEL_ADDR_LOW_MAX,
600 					crash_size, alignment);
601 
602 		if (!crash_base) {
603 			pr_info("crashkernel reservation failed - No suitable area found.\n");
604 			return;
605 		}
606 
607 	} else {
608 		unsigned long long start;
609 
610 		start = memblock_find_in_range(crash_base,
611 				 crash_base + crash_size, crash_size, 1<<20);
612 		if (start != crash_base) {
613 			pr_info("crashkernel reservation failed - memory is in use.\n");
614 			return;
615 		}
616 	}
617 	memblock_reserve(crash_base, crash_size);
618 
619 	printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
620 			"for crashkernel (System RAM: %ldMB)\n",
621 			(unsigned long)(crash_size >> 20),
622 			(unsigned long)(crash_base >> 20),
623 			(unsigned long)(total_mem >> 20));
624 
625 	crashk_res.start = crash_base;
626 	crashk_res.end   = crash_base + crash_size - 1;
627 	insert_resource(&iomem_resource, &crashk_res);
628 
629 	if (crash_base >= (1ULL<<32))
630 		reserve_crashkernel_low();
631 }
632 #else
633 static void __init reserve_crashkernel(void)
634 {
635 }
636 #endif
637 
638 static struct resource standard_io_resources[] = {
639 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
640 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
641 	{ .name = "pic1", .start = 0x20, .end = 0x21,
642 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
643 	{ .name = "timer0", .start = 0x40, .end = 0x43,
644 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
645 	{ .name = "timer1", .start = 0x50, .end = 0x53,
646 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
647 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
648 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
649 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
650 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
651 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
652 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
653 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
654 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
655 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
656 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
657 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
658 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
659 };
660 
661 void __init reserve_standard_io_resources(void)
662 {
663 	int i;
664 
665 	/* request I/O space for devices used on all i[345]86 PCs */
666 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
667 		request_resource(&ioport_resource, &standard_io_resources[i]);
668 
669 }
670 
671 static __init void reserve_ibft_region(void)
672 {
673 	unsigned long addr, size = 0;
674 
675 	addr = find_ibft_region(&size);
676 
677 	if (size)
678 		memblock_reserve(addr, size);
679 }
680 
681 static bool __init snb_gfx_workaround_needed(void)
682 {
683 #ifdef CONFIG_PCI
684 	int i;
685 	u16 vendor, devid;
686 	static const __initconst u16 snb_ids[] = {
687 		0x0102,
688 		0x0112,
689 		0x0122,
690 		0x0106,
691 		0x0116,
692 		0x0126,
693 		0x010a,
694 	};
695 
696 	/* Assume no if something weird is going on with PCI */
697 	if (!early_pci_allowed())
698 		return false;
699 
700 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
701 	if (vendor != 0x8086)
702 		return false;
703 
704 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
705 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
706 		if (devid == snb_ids[i])
707 			return true;
708 #endif
709 
710 	return false;
711 }
712 
713 /*
714  * Sandy Bridge graphics has trouble with certain ranges, exclude
715  * them from allocation.
716  */
717 static void __init trim_snb_memory(void)
718 {
719 	static const __initconst unsigned long bad_pages[] = {
720 		0x20050000,
721 		0x20110000,
722 		0x20130000,
723 		0x20138000,
724 		0x40004000,
725 	};
726 	int i;
727 
728 	if (!snb_gfx_workaround_needed())
729 		return;
730 
731 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
732 
733 	/*
734 	 * Reserve all memory below the 1 MB mark that has not
735 	 * already been reserved.
736 	 */
737 	memblock_reserve(0, 1<<20);
738 
739 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
740 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
741 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
742 			       bad_pages[i]);
743 	}
744 }
745 
746 /*
747  * Here we put platform-specific memory range workarounds, i.e.
748  * memory known to be corrupt or otherwise in need to be reserved on
749  * specific platforms.
750  *
751  * If this gets used more widely it could use a real dispatch mechanism.
752  */
753 static void __init trim_platform_memory_ranges(void)
754 {
755 	trim_snb_memory();
756 }
757 
758 static void __init trim_bios_range(void)
759 {
760 	/*
761 	 * A special case is the first 4Kb of memory;
762 	 * This is a BIOS owned area, not kernel ram, but generally
763 	 * not listed as such in the E820 table.
764 	 *
765 	 * This typically reserves additional memory (64KiB by default)
766 	 * since some BIOSes are known to corrupt low memory.  See the
767 	 * Kconfig help text for X86_RESERVE_LOW.
768 	 */
769 	e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
770 
771 	/*
772 	 * special case: Some BIOSen report the PC BIOS
773 	 * area (640->1Mb) as ram even though it is not.
774 	 * take them out.
775 	 */
776 	e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
777 
778 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
779 }
780 
781 /* called before trim_bios_range() to spare extra sanitize */
782 static void __init e820_add_kernel_range(void)
783 {
784 	u64 start = __pa_symbol(_text);
785 	u64 size = __pa_symbol(_end) - start;
786 
787 	/*
788 	 * Complain if .text .data and .bss are not marked as E820_RAM and
789 	 * attempt to fix it by adding the range. We may have a confused BIOS,
790 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
791 	 * exclude kernel range. If we really are running on top non-RAM,
792 	 * we will crash later anyways.
793 	 */
794 	if (e820_all_mapped(start, start + size, E820_RAM))
795 		return;
796 
797 	pr_warn(".text .data .bss are not marked as E820_RAM!\n");
798 	e820_remove_range(start, size, E820_RAM, 0);
799 	e820_add_region(start, size, E820_RAM);
800 }
801 
802 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
803 
804 static int __init parse_reservelow(char *p)
805 {
806 	unsigned long long size;
807 
808 	if (!p)
809 		return -EINVAL;
810 
811 	size = memparse(p, &p);
812 
813 	if (size < 4096)
814 		size = 4096;
815 
816 	if (size > 640*1024)
817 		size = 640*1024;
818 
819 	reserve_low = size;
820 
821 	return 0;
822 }
823 
824 early_param("reservelow", parse_reservelow);
825 
826 static void __init trim_low_memory_range(void)
827 {
828 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
829 }
830 
831 /*
832  * Dump out kernel offset information on panic.
833  */
834 static int
835 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
836 {
837 	if (kaslr_enabled()) {
838 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
839 			 kaslr_offset(),
840 			 __START_KERNEL,
841 			 __START_KERNEL_map,
842 			 MODULES_VADDR-1);
843 	} else {
844 		pr_emerg("Kernel Offset: disabled\n");
845 	}
846 
847 	return 0;
848 }
849 
850 /*
851  * Determine if we were loaded by an EFI loader.  If so, then we have also been
852  * passed the efi memmap, systab, etc., so we should use these data structures
853  * for initialization.  Note, the efi init code path is determined by the
854  * global efi_enabled. This allows the same kernel image to be used on existing
855  * systems (with a traditional BIOS) as well as on EFI systems.
856  */
857 /*
858  * setup_arch - architecture-specific boot-time initializations
859  *
860  * Note: On x86_64, fixmaps are ready for use even before this is called.
861  */
862 
863 void __init setup_arch(char **cmdline_p)
864 {
865 	memblock_reserve(__pa_symbol(_text),
866 			 (unsigned long)__bss_stop - (unsigned long)_text);
867 
868 	early_reserve_initrd();
869 
870 	/*
871 	 * At this point everything still needed from the boot loader
872 	 * or BIOS or kernel text should be early reserved or marked not
873 	 * RAM in e820. All other memory is free game.
874 	 */
875 
876 #ifdef CONFIG_X86_32
877 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
878 
879 	/*
880 	 * copy kernel address range established so far and switch
881 	 * to the proper swapper page table
882 	 */
883 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
884 			initial_page_table + KERNEL_PGD_BOUNDARY,
885 			KERNEL_PGD_PTRS);
886 
887 	load_cr3(swapper_pg_dir);
888 	/*
889 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
890 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
891 	 * will not flush anything because the cpu quirk which clears
892 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
893 	 * load_cr3() above the TLB has been flushed already. The
894 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
895 	 * so proper operation is guaranteed.
896 	 */
897 	__flush_tlb_all();
898 #else
899 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
900 #endif
901 
902 	/*
903 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
904 	 * reserve_top(), so do this before touching the ioremap area.
905 	 */
906 	olpc_ofw_detect();
907 
908 	early_trap_init();
909 	early_cpu_init();
910 	early_ioremap_init();
911 
912 	setup_olpc_ofw_pgd();
913 
914 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
915 	screen_info = boot_params.screen_info;
916 	edid_info = boot_params.edid_info;
917 #ifdef CONFIG_X86_32
918 	apm_info.bios = boot_params.apm_bios_info;
919 	ist_info = boot_params.ist_info;
920 	if (boot_params.sys_desc_table.length != 0) {
921 		machine_id = boot_params.sys_desc_table.table[0];
922 		machine_submodel_id = boot_params.sys_desc_table.table[1];
923 		BIOS_revision = boot_params.sys_desc_table.table[2];
924 	}
925 #endif
926 	saved_video_mode = boot_params.hdr.vid_mode;
927 	bootloader_type = boot_params.hdr.type_of_loader;
928 	if ((bootloader_type >> 4) == 0xe) {
929 		bootloader_type &= 0xf;
930 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
931 	}
932 	bootloader_version  = bootloader_type & 0xf;
933 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
934 
935 #ifdef CONFIG_BLK_DEV_RAM
936 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
937 	rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
938 	rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
939 #endif
940 #ifdef CONFIG_EFI
941 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
942 		     EFI32_LOADER_SIGNATURE, 4)) {
943 		set_bit(EFI_BOOT, &efi.flags);
944 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
945 		     EFI64_LOADER_SIGNATURE, 4)) {
946 		set_bit(EFI_BOOT, &efi.flags);
947 		set_bit(EFI_64BIT, &efi.flags);
948 	}
949 
950 	if (efi_enabled(EFI_BOOT))
951 		efi_memblock_x86_reserve_range();
952 #endif
953 
954 	x86_init.oem.arch_setup();
955 
956 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
957 	setup_memory_map();
958 	parse_setup_data();
959 
960 	copy_edd();
961 
962 	if (!boot_params.hdr.root_flags)
963 		root_mountflags &= ~MS_RDONLY;
964 	init_mm.start_code = (unsigned long) _text;
965 	init_mm.end_code = (unsigned long) _etext;
966 	init_mm.end_data = (unsigned long) _edata;
967 	init_mm.brk = _brk_end;
968 
969 	mpx_mm_init(&init_mm);
970 
971 	code_resource.start = __pa_symbol(_text);
972 	code_resource.end = __pa_symbol(_etext)-1;
973 	data_resource.start = __pa_symbol(_etext);
974 	data_resource.end = __pa_symbol(_edata)-1;
975 	bss_resource.start = __pa_symbol(__bss_start);
976 	bss_resource.end = __pa_symbol(__bss_stop)-1;
977 
978 #ifdef CONFIG_CMDLINE_BOOL
979 #ifdef CONFIG_CMDLINE_OVERRIDE
980 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
981 #else
982 	if (builtin_cmdline[0]) {
983 		/* append boot loader cmdline to builtin */
984 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
985 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
986 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
987 	}
988 #endif
989 #endif
990 
991 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
992 	*cmdline_p = command_line;
993 
994 	/*
995 	 * x86_configure_nx() is called before parse_early_param() to detect
996 	 * whether hardware doesn't support NX (so that the early EHCI debug
997 	 * console setup can safely call set_fixmap()). It may then be called
998 	 * again from within noexec_setup() during parsing early parameters
999 	 * to honor the respective command line option.
1000 	 */
1001 	x86_configure_nx();
1002 
1003 	parse_early_param();
1004 
1005 	x86_report_nx();
1006 
1007 	/* after early param, so could get panic from serial */
1008 	memblock_x86_reserve_range_setup_data();
1009 
1010 	if (acpi_mps_check()) {
1011 #ifdef CONFIG_X86_LOCAL_APIC
1012 		disable_apic = 1;
1013 #endif
1014 		setup_clear_cpu_cap(X86_FEATURE_APIC);
1015 	}
1016 
1017 #ifdef CONFIG_PCI
1018 	if (pci_early_dump_regs)
1019 		early_dump_pci_devices();
1020 #endif
1021 
1022 	/* update the e820_saved too */
1023 	e820_reserve_setup_data();
1024 	finish_e820_parsing();
1025 
1026 	if (efi_enabled(EFI_BOOT))
1027 		efi_init();
1028 
1029 	dmi_scan_machine();
1030 	dmi_memdev_walk();
1031 	dmi_set_dump_stack_arch_desc();
1032 
1033 	/*
1034 	 * VMware detection requires dmi to be available, so this
1035 	 * needs to be done after dmi_scan_machine, for the BP.
1036 	 */
1037 	init_hypervisor_platform();
1038 
1039 	x86_init.resources.probe_roms();
1040 
1041 	/* after parse_early_param, so could debug it */
1042 	insert_resource(&iomem_resource, &code_resource);
1043 	insert_resource(&iomem_resource, &data_resource);
1044 	insert_resource(&iomem_resource, &bss_resource);
1045 
1046 	e820_add_kernel_range();
1047 	trim_bios_range();
1048 #ifdef CONFIG_X86_32
1049 	if (ppro_with_ram_bug()) {
1050 		e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1051 				  E820_RESERVED);
1052 		sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1053 		printk(KERN_INFO "fixed physical RAM map:\n");
1054 		e820_print_map("bad_ppro");
1055 	}
1056 #else
1057 	early_gart_iommu_check();
1058 #endif
1059 
1060 	/*
1061 	 * partially used pages are not usable - thus
1062 	 * we are rounding upwards:
1063 	 */
1064 	max_pfn = e820_end_of_ram_pfn();
1065 
1066 	/* update e820 for memory not covered by WB MTRRs */
1067 	mtrr_bp_init();
1068 	if (mtrr_trim_uncached_memory(max_pfn))
1069 		max_pfn = e820_end_of_ram_pfn();
1070 
1071 #ifdef CONFIG_X86_32
1072 	/* max_low_pfn get updated here */
1073 	find_low_pfn_range();
1074 #else
1075 	check_x2apic();
1076 
1077 	/* How many end-of-memory variables you have, grandma! */
1078 	/* need this before calling reserve_initrd */
1079 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1080 		max_low_pfn = e820_end_of_low_ram_pfn();
1081 	else
1082 		max_low_pfn = max_pfn;
1083 
1084 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1085 #endif
1086 
1087 	/*
1088 	 * Find and reserve possible boot-time SMP configuration:
1089 	 */
1090 	find_smp_config();
1091 
1092 	reserve_ibft_region();
1093 
1094 	early_alloc_pgt_buf();
1095 
1096 	/*
1097 	 * Need to conclude brk, before memblock_x86_fill()
1098 	 *  it could use memblock_find_in_range, could overlap with
1099 	 *  brk area.
1100 	 */
1101 	reserve_brk();
1102 
1103 	cleanup_highmap();
1104 
1105 	memblock_set_current_limit(ISA_END_ADDRESS);
1106 	memblock_x86_fill();
1107 
1108 	/*
1109 	 * The EFI specification says that boot service code won't be called
1110 	 * after ExitBootServices(). This is, in fact, a lie.
1111 	 */
1112 	if (efi_enabled(EFI_MEMMAP))
1113 		efi_reserve_boot_services();
1114 
1115 	/* preallocate 4k for mptable mpc */
1116 	early_reserve_e820_mpc_new();
1117 
1118 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1119 	setup_bios_corruption_check();
1120 #endif
1121 
1122 #ifdef CONFIG_X86_32
1123 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1124 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1125 #endif
1126 
1127 	reserve_real_mode();
1128 
1129 	trim_platform_memory_ranges();
1130 	trim_low_memory_range();
1131 
1132 	init_mem_mapping();
1133 
1134 	early_trap_pf_init();
1135 
1136 	setup_real_mode();
1137 
1138 	memblock_set_current_limit(get_max_mapped());
1139 
1140 	/*
1141 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1142 	 */
1143 
1144 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1145 	if (init_ohci1394_dma_early)
1146 		init_ohci1394_dma_on_all_controllers();
1147 #endif
1148 	/* Allocate bigger log buffer */
1149 	setup_log_buf(1);
1150 
1151 	reserve_initrd();
1152 
1153 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1154 	acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1155 #endif
1156 
1157 	vsmp_init();
1158 
1159 	io_delay_init();
1160 
1161 	/*
1162 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1163 	 */
1164 	acpi_boot_table_init();
1165 
1166 	early_acpi_boot_init();
1167 
1168 	initmem_init();
1169 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1170 
1171 	/*
1172 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1173 	 * won't consume hotpluggable memory.
1174 	 */
1175 	reserve_crashkernel();
1176 
1177 	memblock_find_dma_reserve();
1178 
1179 #ifdef CONFIG_KVM_GUEST
1180 	kvmclock_init();
1181 #endif
1182 
1183 	x86_init.paging.pagetable_init();
1184 
1185 	kasan_init();
1186 
1187 	if (boot_cpu_data.cpuid_level >= 0) {
1188 		/* A CPU has %cr4 if and only if it has CPUID */
1189 		mmu_cr4_features = __read_cr4();
1190 		if (trampoline_cr4_features)
1191 			*trampoline_cr4_features = mmu_cr4_features;
1192 	}
1193 
1194 #ifdef CONFIG_X86_32
1195 	/* sync back kernel address range */
1196 	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1197 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
1198 			KERNEL_PGD_PTRS);
1199 #endif
1200 
1201 	tboot_probe();
1202 
1203 	map_vsyscall();
1204 
1205 	generic_apic_probe();
1206 
1207 	early_quirks();
1208 
1209 	/*
1210 	 * Read APIC and some other early information from ACPI tables.
1211 	 */
1212 	acpi_boot_init();
1213 	sfi_init();
1214 	x86_dtb_init();
1215 
1216 	/*
1217 	 * get boot-time SMP configuration:
1218 	 */
1219 	if (smp_found_config)
1220 		get_smp_config();
1221 
1222 	prefill_possible_map();
1223 
1224 	init_cpu_to_node();
1225 
1226 	init_apic_mappings();
1227 	io_apic_init_mappings();
1228 
1229 	kvm_guest_init();
1230 
1231 	e820_reserve_resources();
1232 	e820_mark_nosave_regions(max_low_pfn);
1233 
1234 	x86_init.resources.reserve_resources();
1235 
1236 	e820_setup_gap();
1237 
1238 #ifdef CONFIG_VT
1239 #if defined(CONFIG_VGA_CONSOLE)
1240 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1241 		conswitchp = &vga_con;
1242 #elif defined(CONFIG_DUMMY_CONSOLE)
1243 	conswitchp = &dummy_con;
1244 #endif
1245 #endif
1246 	x86_init.oem.banner();
1247 
1248 	x86_init.timers.wallclock_init();
1249 
1250 	mcheck_init();
1251 
1252 	arch_init_ideal_nops();
1253 
1254 	register_refined_jiffies(CLOCK_TICK_RATE);
1255 
1256 #ifdef CONFIG_EFI
1257 	if (efi_enabled(EFI_BOOT))
1258 		efi_apply_memmap_quirks();
1259 #endif
1260 }
1261 
1262 #ifdef CONFIG_X86_32
1263 
1264 static struct resource video_ram_resource = {
1265 	.name	= "Video RAM area",
1266 	.start	= 0xa0000,
1267 	.end	= 0xbffff,
1268 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1269 };
1270 
1271 void __init i386_reserve_resources(void)
1272 {
1273 	request_resource(&iomem_resource, &video_ram_resource);
1274 	reserve_standard_io_resources();
1275 }
1276 
1277 #endif /* CONFIG_X86_32 */
1278 
1279 static struct notifier_block kernel_offset_notifier = {
1280 	.notifier_call = dump_kernel_offset
1281 };
1282 
1283 static int __init register_kernel_offset_dumper(void)
1284 {
1285 	atomic_notifier_chain_register(&panic_notifier_list,
1286 					&kernel_offset_notifier);
1287 	return 0;
1288 }
1289 __initcall(register_kernel_offset_dumper);
1290