xref: /openbmc/linux/arch/x86/kernel/setup.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /*
2  *  Copyright (C) 1995  Linus Torvalds
3  *
4  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5  *
6  *  Memory region support
7  *	David Parsons <orc@pell.chi.il.us>, July-August 1999
8  *
9  *  Added E820 sanitization routine (removes overlapping memory regions);
10  *  Brian Moyle <bmoyle@mvista.com>, February 2001
11  *
12  * Moved CPU detection code to cpu/${cpu}.c
13  *    Patrick Mochel <mochel@osdl.org>, March 2002
14  *
15  *  Provisions for empty E820 memory regions (reported by certain BIOSes).
16  *  Alex Achenbach <xela@slit.de>, December 2002.
17  *
18  */
19 
20 /*
21  * This file handles the architecture-dependent parts of initialization
22  */
23 
24 #include <linux/sched.h>
25 #include <linux/mm.h>
26 #include <linux/mmzone.h>
27 #include <linux/screen_info.h>
28 #include <linux/ioport.h>
29 #include <linux/acpi.h>
30 #include <linux/sfi.h>
31 #include <linux/apm_bios.h>
32 #include <linux/initrd.h>
33 #include <linux/bootmem.h>
34 #include <linux/memblock.h>
35 #include <linux/seq_file.h>
36 #include <linux/console.h>
37 #include <linux/mca.h>
38 #include <linux/root_dev.h>
39 #include <linux/highmem.h>
40 #include <linux/module.h>
41 #include <linux/efi.h>
42 #include <linux/init.h>
43 #include <linux/edd.h>
44 #include <linux/iscsi_ibft.h>
45 #include <linux/nodemask.h>
46 #include <linux/kexec.h>
47 #include <linux/dmi.h>
48 #include <linux/pfn.h>
49 #include <linux/pci.h>
50 #include <asm/pci-direct.h>
51 #include <linux/init_ohci1394_dma.h>
52 #include <linux/kvm_para.h>
53 
54 #include <linux/errno.h>
55 #include <linux/kernel.h>
56 #include <linux/stddef.h>
57 #include <linux/unistd.h>
58 #include <linux/ptrace.h>
59 #include <linux/user.h>
60 #include <linux/delay.h>
61 
62 #include <linux/kallsyms.h>
63 #include <linux/cpufreq.h>
64 #include <linux/dma-mapping.h>
65 #include <linux/ctype.h>
66 #include <linux/uaccess.h>
67 
68 #include <linux/percpu.h>
69 #include <linux/crash_dump.h>
70 #include <linux/tboot.h>
71 
72 #include <video/edid.h>
73 
74 #include <asm/mtrr.h>
75 #include <asm/apic.h>
76 #include <asm/trampoline.h>
77 #include <asm/e820.h>
78 #include <asm/mpspec.h>
79 #include <asm/setup.h>
80 #include <asm/efi.h>
81 #include <asm/timer.h>
82 #include <asm/i8259.h>
83 #include <asm/sections.h>
84 #include <asm/dmi.h>
85 #include <asm/io_apic.h>
86 #include <asm/ist.h>
87 #include <asm/setup_arch.h>
88 #include <asm/bios_ebda.h>
89 #include <asm/cacheflush.h>
90 #include <asm/processor.h>
91 #include <asm/bugs.h>
92 
93 #include <asm/system.h>
94 #include <asm/vsyscall.h>
95 #include <asm/cpu.h>
96 #include <asm/desc.h>
97 #include <asm/dma.h>
98 #include <asm/iommu.h>
99 #include <asm/gart.h>
100 #include <asm/mmu_context.h>
101 #include <asm/proto.h>
102 
103 #include <asm/paravirt.h>
104 #include <asm/hypervisor.h>
105 #include <asm/olpc_ofw.h>
106 
107 #include <asm/percpu.h>
108 #include <asm/topology.h>
109 #include <asm/apicdef.h>
110 #include <asm/amd_nb.h>
111 #ifdef CONFIG_X86_64
112 #include <asm/numa_64.h>
113 #endif
114 #include <asm/mce.h>
115 #include <asm/alternative.h>
116 #include <asm/prom.h>
117 
118 /*
119  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
120  * The direct mapping extends to max_pfn_mapped, so that we can directly access
121  * apertures, ACPI and other tables without having to play with fixmaps.
122  */
123 unsigned long max_low_pfn_mapped;
124 unsigned long max_pfn_mapped;
125 
126 #ifdef CONFIG_DMI
127 RESERVE_BRK(dmi_alloc, 65536);
128 #endif
129 
130 
131 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
132 unsigned long _brk_end = (unsigned long)__brk_base;
133 
134 #ifdef CONFIG_X86_64
135 int default_cpu_present_to_apicid(int mps_cpu)
136 {
137 	return __default_cpu_present_to_apicid(mps_cpu);
138 }
139 
140 int default_check_phys_apicid_present(int phys_apicid)
141 {
142 	return __default_check_phys_apicid_present(phys_apicid);
143 }
144 #endif
145 
146 #ifndef CONFIG_DEBUG_BOOT_PARAMS
147 struct boot_params __initdata boot_params;
148 #else
149 struct boot_params boot_params;
150 #endif
151 
152 /*
153  * Machine setup..
154  */
155 static struct resource data_resource = {
156 	.name	= "Kernel data",
157 	.start	= 0,
158 	.end	= 0,
159 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
160 };
161 
162 static struct resource code_resource = {
163 	.name	= "Kernel code",
164 	.start	= 0,
165 	.end	= 0,
166 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
167 };
168 
169 static struct resource bss_resource = {
170 	.name	= "Kernel bss",
171 	.start	= 0,
172 	.end	= 0,
173 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
174 };
175 
176 
177 #ifdef CONFIG_X86_32
178 /* cpu data as detected by the assembly code in head.S */
179 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
180 /* common cpu data for all cpus */
181 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
182 EXPORT_SYMBOL(boot_cpu_data);
183 static void set_mca_bus(int x)
184 {
185 #ifdef CONFIG_MCA
186 	MCA_bus = x;
187 #endif
188 }
189 
190 unsigned int def_to_bigsmp;
191 
192 /* for MCA, but anyone else can use it if they want */
193 unsigned int machine_id;
194 unsigned int machine_submodel_id;
195 unsigned int BIOS_revision;
196 
197 struct apm_info apm_info;
198 EXPORT_SYMBOL(apm_info);
199 
200 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
201 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
202 struct ist_info ist_info;
203 EXPORT_SYMBOL(ist_info);
204 #else
205 struct ist_info ist_info;
206 #endif
207 
208 #else
209 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
210 	.x86_phys_bits = MAX_PHYSMEM_BITS,
211 };
212 EXPORT_SYMBOL(boot_cpu_data);
213 #endif
214 
215 
216 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
217 unsigned long mmu_cr4_features;
218 #else
219 unsigned long mmu_cr4_features = X86_CR4_PAE;
220 #endif
221 
222 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
223 int bootloader_type, bootloader_version;
224 
225 /*
226  * Setup options
227  */
228 struct screen_info screen_info;
229 EXPORT_SYMBOL(screen_info);
230 struct edid_info edid_info;
231 EXPORT_SYMBOL_GPL(edid_info);
232 
233 extern int root_mountflags;
234 
235 unsigned long saved_video_mode;
236 
237 #define RAMDISK_IMAGE_START_MASK	0x07FF
238 #define RAMDISK_PROMPT_FLAG		0x8000
239 #define RAMDISK_LOAD_FLAG		0x4000
240 
241 static char __initdata command_line[COMMAND_LINE_SIZE];
242 #ifdef CONFIG_CMDLINE_BOOL
243 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
244 #endif
245 
246 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
247 struct edd edd;
248 #ifdef CONFIG_EDD_MODULE
249 EXPORT_SYMBOL(edd);
250 #endif
251 /**
252  * copy_edd() - Copy the BIOS EDD information
253  *              from boot_params into a safe place.
254  *
255  */
256 static inline void __init copy_edd(void)
257 {
258      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
259 	    sizeof(edd.mbr_signature));
260      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
261      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
262      edd.edd_info_nr = boot_params.eddbuf_entries;
263 }
264 #else
265 static inline void __init copy_edd(void)
266 {
267 }
268 #endif
269 
270 void * __init extend_brk(size_t size, size_t align)
271 {
272 	size_t mask = align - 1;
273 	void *ret;
274 
275 	BUG_ON(_brk_start == 0);
276 	BUG_ON(align & mask);
277 
278 	_brk_end = (_brk_end + mask) & ~mask;
279 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
280 
281 	ret = (void *)_brk_end;
282 	_brk_end += size;
283 
284 	memset(ret, 0, size);
285 
286 	return ret;
287 }
288 
289 #ifdef CONFIG_X86_64
290 static void __init init_gbpages(void)
291 {
292 	if (direct_gbpages && cpu_has_gbpages)
293 		printk(KERN_INFO "Using GB pages for direct mapping\n");
294 	else
295 		direct_gbpages = 0;
296 }
297 #else
298 static inline void init_gbpages(void)
299 {
300 }
301 #endif
302 
303 static void __init reserve_brk(void)
304 {
305 	if (_brk_end > _brk_start)
306 		memblock_x86_reserve_range(__pa(_brk_start), __pa(_brk_end), "BRK");
307 
308 	/* Mark brk area as locked down and no longer taking any
309 	   new allocations */
310 	_brk_start = 0;
311 }
312 
313 #ifdef CONFIG_BLK_DEV_INITRD
314 
315 #define MAX_MAP_CHUNK	(NR_FIX_BTMAPS << PAGE_SHIFT)
316 static void __init relocate_initrd(void)
317 {
318 	/* Assume only end is not page aligned */
319 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
320 	u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
321 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
322 	u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
323 	u64 ramdisk_here;
324 	unsigned long slop, clen, mapaddr;
325 	char *p, *q;
326 
327 	/* We need to move the initrd down into lowmem */
328 	ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size,
329 					 PAGE_SIZE);
330 
331 	if (ramdisk_here == MEMBLOCK_ERROR)
332 		panic("Cannot find place for new RAMDISK of size %lld\n",
333 			 ramdisk_size);
334 
335 	/* Note: this includes all the lowmem currently occupied by
336 	   the initrd, we rely on that fact to keep the data intact. */
337 	memblock_x86_reserve_range(ramdisk_here, ramdisk_here + area_size, "NEW RAMDISK");
338 	initrd_start = ramdisk_here + PAGE_OFFSET;
339 	initrd_end   = initrd_start + ramdisk_size;
340 	printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
341 			 ramdisk_here, ramdisk_here + ramdisk_size);
342 
343 	q = (char *)initrd_start;
344 
345 	/* Copy any lowmem portion of the initrd */
346 	if (ramdisk_image < end_of_lowmem) {
347 		clen = end_of_lowmem - ramdisk_image;
348 		p = (char *)__va(ramdisk_image);
349 		memcpy(q, p, clen);
350 		q += clen;
351 		ramdisk_image += clen;
352 		ramdisk_size  -= clen;
353 	}
354 
355 	/* Copy the highmem portion of the initrd */
356 	while (ramdisk_size) {
357 		slop = ramdisk_image & ~PAGE_MASK;
358 		clen = ramdisk_size;
359 		if (clen > MAX_MAP_CHUNK-slop)
360 			clen = MAX_MAP_CHUNK-slop;
361 		mapaddr = ramdisk_image & PAGE_MASK;
362 		p = early_memremap(mapaddr, clen+slop);
363 		memcpy(q, p+slop, clen);
364 		early_iounmap(p, clen+slop);
365 		q += clen;
366 		ramdisk_image += clen;
367 		ramdisk_size  -= clen;
368 	}
369 	/* high pages is not converted by early_res_to_bootmem */
370 	ramdisk_image = boot_params.hdr.ramdisk_image;
371 	ramdisk_size  = boot_params.hdr.ramdisk_size;
372 	printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
373 		" %08llx - %08llx\n",
374 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
375 		ramdisk_here, ramdisk_here + ramdisk_size - 1);
376 }
377 
378 static void __init reserve_initrd(void)
379 {
380 	/* Assume only end is not page aligned */
381 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
382 	u64 ramdisk_size  = boot_params.hdr.ramdisk_size;
383 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
384 	u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
385 
386 	if (!boot_params.hdr.type_of_loader ||
387 	    !ramdisk_image || !ramdisk_size)
388 		return;		/* No initrd provided by bootloader */
389 
390 	initrd_start = 0;
391 
392 	if (ramdisk_size >= (end_of_lowmem>>1)) {
393 		memblock_x86_free_range(ramdisk_image, ramdisk_end);
394 		printk(KERN_ERR "initrd too large to handle, "
395 		       "disabling initrd\n");
396 		return;
397 	}
398 
399 	printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
400 			ramdisk_end);
401 
402 
403 	if (ramdisk_end <= end_of_lowmem) {
404 		/* All in lowmem, easy case */
405 		/*
406 		 * don't need to reserve again, already reserved early
407 		 * in i386_start_kernel
408 		 */
409 		initrd_start = ramdisk_image + PAGE_OFFSET;
410 		initrd_end = initrd_start + ramdisk_size;
411 		return;
412 	}
413 
414 	relocate_initrd();
415 
416 	memblock_x86_free_range(ramdisk_image, ramdisk_end);
417 }
418 #else
419 static void __init reserve_initrd(void)
420 {
421 }
422 #endif /* CONFIG_BLK_DEV_INITRD */
423 
424 static void __init parse_setup_data(void)
425 {
426 	struct setup_data *data;
427 	u64 pa_data;
428 
429 	if (boot_params.hdr.version < 0x0209)
430 		return;
431 	pa_data = boot_params.hdr.setup_data;
432 	while (pa_data) {
433 		u32 data_len, map_len;
434 
435 		map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
436 			      (u64)sizeof(struct setup_data));
437 		data = early_memremap(pa_data, map_len);
438 		data_len = data->len + sizeof(struct setup_data);
439 		if (data_len > map_len) {
440 			early_iounmap(data, map_len);
441 			data = early_memremap(pa_data, data_len);
442 			map_len = data_len;
443 		}
444 
445 		switch (data->type) {
446 		case SETUP_E820_EXT:
447 			parse_e820_ext(data);
448 			break;
449 		case SETUP_DTB:
450 			add_dtb(pa_data);
451 			break;
452 		default:
453 			break;
454 		}
455 		pa_data = data->next;
456 		early_iounmap(data, map_len);
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 	if (boot_params.hdr.version < 0x0209)
467 		return;
468 	pa_data = boot_params.hdr.setup_data;
469 	while (pa_data) {
470 		data = early_memremap(pa_data, sizeof(*data));
471 		e820_update_range(pa_data, sizeof(*data)+data->len,
472 			 E820_RAM, E820_RESERVED_KERN);
473 		found = 1;
474 		pa_data = data->next;
475 		early_iounmap(data, sizeof(*data));
476 	}
477 	if (!found)
478 		return;
479 
480 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
481 	memcpy(&e820_saved, &e820, sizeof(struct e820map));
482 	printk(KERN_INFO "extended physical RAM map:\n");
483 	e820_print_map("reserve setup_data");
484 }
485 
486 static void __init memblock_x86_reserve_range_setup_data(void)
487 {
488 	struct setup_data *data;
489 	u64 pa_data;
490 	char buf[32];
491 
492 	if (boot_params.hdr.version < 0x0209)
493 		return;
494 	pa_data = boot_params.hdr.setup_data;
495 	while (pa_data) {
496 		data = early_memremap(pa_data, sizeof(*data));
497 		sprintf(buf, "setup data %x", data->type);
498 		memblock_x86_reserve_range(pa_data, pa_data+sizeof(*data)+data->len, buf);
499 		pa_data = data->next;
500 		early_iounmap(data, sizeof(*data));
501 	}
502 }
503 
504 /*
505  * --------- Crashkernel reservation ------------------------------
506  */
507 
508 #ifdef CONFIG_KEXEC
509 
510 static inline unsigned long long get_total_mem(void)
511 {
512 	unsigned long long total;
513 
514 	total = max_pfn - min_low_pfn;
515 
516 	return total << PAGE_SHIFT;
517 }
518 
519 /*
520  * Keep the crash kernel below this limit.  On 32 bits earlier kernels
521  * would limit the kernel to the low 512 MiB due to mapping restrictions.
522  * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this
523  * limit once kexec-tools are fixed.
524  */
525 #ifdef CONFIG_X86_32
526 # define CRASH_KERNEL_ADDR_MAX	(512 << 20)
527 #else
528 # define CRASH_KERNEL_ADDR_MAX	(896 << 20)
529 #endif
530 
531 static void __init reserve_crashkernel(void)
532 {
533 	unsigned long long total_mem;
534 	unsigned long long crash_size, crash_base;
535 	int ret;
536 
537 	total_mem = get_total_mem();
538 
539 	ret = parse_crashkernel(boot_command_line, total_mem,
540 			&crash_size, &crash_base);
541 	if (ret != 0 || crash_size <= 0)
542 		return;
543 
544 	/* 0 means: find the address automatically */
545 	if (crash_base <= 0) {
546 		const unsigned long long alignment = 16<<20;	/* 16M */
547 
548 		/*
549 		 *  kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
550 		 */
551 		crash_base = memblock_find_in_range(alignment,
552 			       CRASH_KERNEL_ADDR_MAX, crash_size, alignment);
553 
554 		if (crash_base == MEMBLOCK_ERROR) {
555 			pr_info("crashkernel reservation failed - No suitable area found.\n");
556 			return;
557 		}
558 	} else {
559 		unsigned long long start;
560 
561 		start = memblock_find_in_range(crash_base,
562 				 crash_base + crash_size, crash_size, 1<<20);
563 		if (start != crash_base) {
564 			pr_info("crashkernel reservation failed - memory is in use.\n");
565 			return;
566 		}
567 	}
568 	memblock_x86_reserve_range(crash_base, crash_base + crash_size, "CRASH KERNEL");
569 
570 	printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
571 			"for crashkernel (System RAM: %ldMB)\n",
572 			(unsigned long)(crash_size >> 20),
573 			(unsigned long)(crash_base >> 20),
574 			(unsigned long)(total_mem >> 20));
575 
576 	crashk_res.start = crash_base;
577 	crashk_res.end   = crash_base + crash_size - 1;
578 	insert_resource(&iomem_resource, &crashk_res);
579 }
580 #else
581 static void __init reserve_crashkernel(void)
582 {
583 }
584 #endif
585 
586 static struct resource standard_io_resources[] = {
587 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
588 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
589 	{ .name = "pic1", .start = 0x20, .end = 0x21,
590 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
591 	{ .name = "timer0", .start = 0x40, .end = 0x43,
592 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
593 	{ .name = "timer1", .start = 0x50, .end = 0x53,
594 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
595 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
596 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
597 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
598 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
599 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
600 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
601 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
602 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
603 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
604 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
605 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
606 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
607 };
608 
609 void __init reserve_standard_io_resources(void)
610 {
611 	int i;
612 
613 	/* request I/O space for devices used on all i[345]86 PCs */
614 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
615 		request_resource(&ioport_resource, &standard_io_resources[i]);
616 
617 }
618 
619 /*
620  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
621  * is_kdump_kernel() to determine if we are booting after a panic. Hence
622  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
623  */
624 
625 #ifdef CONFIG_CRASH_DUMP
626 /* elfcorehdr= specifies the location of elf core header
627  * stored by the crashed kernel. This option will be passed
628  * by kexec loader to the capture kernel.
629  */
630 static int __init setup_elfcorehdr(char *arg)
631 {
632 	char *end;
633 	if (!arg)
634 		return -EINVAL;
635 	elfcorehdr_addr = memparse(arg, &end);
636 	return end > arg ? 0 : -EINVAL;
637 }
638 early_param("elfcorehdr", setup_elfcorehdr);
639 #endif
640 
641 static __init void reserve_ibft_region(void)
642 {
643 	unsigned long addr, size = 0;
644 
645 	addr = find_ibft_region(&size);
646 
647 	if (size)
648 		memblock_x86_reserve_range(addr, addr + size, "* ibft");
649 }
650 
651 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
652 
653 static void __init trim_bios_range(void)
654 {
655 	/*
656 	 * A special case is the first 4Kb of memory;
657 	 * This is a BIOS owned area, not kernel ram, but generally
658 	 * not listed as such in the E820 table.
659 	 *
660 	 * This typically reserves additional memory (64KiB by default)
661 	 * since some BIOSes are known to corrupt low memory.  See the
662 	 * Kconfig help text for X86_RESERVE_LOW.
663 	 */
664 	e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE),
665 			  E820_RAM, E820_RESERVED);
666 
667 	/*
668 	 * special case: Some BIOSen report the PC BIOS
669 	 * area (640->1Mb) as ram even though it is not.
670 	 * take them out.
671 	 */
672 	e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
673 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
674 }
675 
676 static int __init parse_reservelow(char *p)
677 {
678 	unsigned long long size;
679 
680 	if (!p)
681 		return -EINVAL;
682 
683 	size = memparse(p, &p);
684 
685 	if (size < 4096)
686 		size = 4096;
687 
688 	if (size > 640*1024)
689 		size = 640*1024;
690 
691 	reserve_low = size;
692 
693 	return 0;
694 }
695 
696 early_param("reservelow", parse_reservelow);
697 
698 static u64 __init get_max_mapped(void)
699 {
700 	u64 end = max_pfn_mapped;
701 
702 	end <<= PAGE_SHIFT;
703 
704 	return end;
705 }
706 
707 /*
708  * Determine if we were loaded by an EFI loader.  If so, then we have also been
709  * passed the efi memmap, systab, etc., so we should use these data structures
710  * for initialization.  Note, the efi init code path is determined by the
711  * global efi_enabled. This allows the same kernel image to be used on existing
712  * systems (with a traditional BIOS) as well as on EFI systems.
713  */
714 /*
715  * setup_arch - architecture-specific boot-time initializations
716  *
717  * Note: On x86_64, fixmaps are ready for use even before this is called.
718  */
719 
720 void __init setup_arch(char **cmdline_p)
721 {
722 	int acpi = 0;
723 	int amd = 0;
724 	unsigned long flags;
725 
726 #ifdef CONFIG_X86_32
727 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
728 	visws_early_detect();
729 
730 	/*
731 	 * copy kernel address range established so far and switch
732 	 * to the proper swapper page table
733 	 */
734 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
735 			initial_page_table + KERNEL_PGD_BOUNDARY,
736 			KERNEL_PGD_PTRS);
737 
738 	load_cr3(swapper_pg_dir);
739 	__flush_tlb_all();
740 #else
741 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
742 #endif
743 
744 	/*
745 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
746 	 * reserve_top(), so do this before touching the ioremap area.
747 	 */
748 	olpc_ofw_detect();
749 
750 	early_trap_init();
751 	early_cpu_init();
752 	early_ioremap_init();
753 
754 	setup_olpc_ofw_pgd();
755 
756 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
757 	screen_info = boot_params.screen_info;
758 	edid_info = boot_params.edid_info;
759 #ifdef CONFIG_X86_32
760 	apm_info.bios = boot_params.apm_bios_info;
761 	ist_info = boot_params.ist_info;
762 	if (boot_params.sys_desc_table.length != 0) {
763 		set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
764 		machine_id = boot_params.sys_desc_table.table[0];
765 		machine_submodel_id = boot_params.sys_desc_table.table[1];
766 		BIOS_revision = boot_params.sys_desc_table.table[2];
767 	}
768 #endif
769 	saved_video_mode = boot_params.hdr.vid_mode;
770 	bootloader_type = boot_params.hdr.type_of_loader;
771 	if ((bootloader_type >> 4) == 0xe) {
772 		bootloader_type &= 0xf;
773 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
774 	}
775 	bootloader_version  = bootloader_type & 0xf;
776 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
777 
778 #ifdef CONFIG_BLK_DEV_RAM
779 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
780 	rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
781 	rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
782 #endif
783 #ifdef CONFIG_EFI
784 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
785 #ifdef CONFIG_X86_32
786 		     "EL32",
787 #else
788 		     "EL64",
789 #endif
790 	 4)) {
791 		efi_enabled = 1;
792 		efi_memblock_x86_reserve_range();
793 	}
794 #endif
795 
796 	x86_init.oem.arch_setup();
797 
798 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
799 	setup_memory_map();
800 	parse_setup_data();
801 	/* update the e820_saved too */
802 	e820_reserve_setup_data();
803 
804 	copy_edd();
805 
806 	if (!boot_params.hdr.root_flags)
807 		root_mountflags &= ~MS_RDONLY;
808 	init_mm.start_code = (unsigned long) _text;
809 	init_mm.end_code = (unsigned long) _etext;
810 	init_mm.end_data = (unsigned long) _edata;
811 	init_mm.brk = _brk_end;
812 
813 	code_resource.start = virt_to_phys(_text);
814 	code_resource.end = virt_to_phys(_etext)-1;
815 	data_resource.start = virt_to_phys(_etext);
816 	data_resource.end = virt_to_phys(_edata)-1;
817 	bss_resource.start = virt_to_phys(&__bss_start);
818 	bss_resource.end = virt_to_phys(&__bss_stop)-1;
819 
820 #ifdef CONFIG_CMDLINE_BOOL
821 #ifdef CONFIG_CMDLINE_OVERRIDE
822 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
823 #else
824 	if (builtin_cmdline[0]) {
825 		/* append boot loader cmdline to builtin */
826 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
827 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
828 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
829 	}
830 #endif
831 #endif
832 
833 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
834 	*cmdline_p = command_line;
835 
836 	/*
837 	 * x86_configure_nx() is called before parse_early_param() to detect
838 	 * whether hardware doesn't support NX (so that the early EHCI debug
839 	 * console setup can safely call set_fixmap()). It may then be called
840 	 * again from within noexec_setup() during parsing early parameters
841 	 * to honor the respective command line option.
842 	 */
843 	x86_configure_nx();
844 
845 	parse_early_param();
846 
847 	x86_report_nx();
848 
849 	/* after early param, so could get panic from serial */
850 	memblock_x86_reserve_range_setup_data();
851 
852 	if (acpi_mps_check()) {
853 #ifdef CONFIG_X86_LOCAL_APIC
854 		disable_apic = 1;
855 #endif
856 		setup_clear_cpu_cap(X86_FEATURE_APIC);
857 	}
858 
859 #ifdef CONFIG_PCI
860 	if (pci_early_dump_regs)
861 		early_dump_pci_devices();
862 #endif
863 
864 	finish_e820_parsing();
865 
866 	if (efi_enabled)
867 		efi_init();
868 
869 	dmi_scan_machine();
870 
871 	/*
872 	 * VMware detection requires dmi to be available, so this
873 	 * needs to be done after dmi_scan_machine, for the BP.
874 	 */
875 	init_hypervisor_platform();
876 
877 	x86_init.resources.probe_roms();
878 
879 	/* after parse_early_param, so could debug it */
880 	insert_resource(&iomem_resource, &code_resource);
881 	insert_resource(&iomem_resource, &data_resource);
882 	insert_resource(&iomem_resource, &bss_resource);
883 
884 	trim_bios_range();
885 #ifdef CONFIG_X86_32
886 	if (ppro_with_ram_bug()) {
887 		e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
888 				  E820_RESERVED);
889 		sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
890 		printk(KERN_INFO "fixed physical RAM map:\n");
891 		e820_print_map("bad_ppro");
892 	}
893 #else
894 	early_gart_iommu_check();
895 #endif
896 
897 	/*
898 	 * partially used pages are not usable - thus
899 	 * we are rounding upwards:
900 	 */
901 	max_pfn = e820_end_of_ram_pfn();
902 
903 	/* update e820 for memory not covered by WB MTRRs */
904 	mtrr_bp_init();
905 	if (mtrr_trim_uncached_memory(max_pfn))
906 		max_pfn = e820_end_of_ram_pfn();
907 
908 #ifdef CONFIG_X86_32
909 	/* max_low_pfn get updated here */
910 	find_low_pfn_range();
911 #else
912 	num_physpages = max_pfn;
913 
914 	check_x2apic();
915 
916 	/* How many end-of-memory variables you have, grandma! */
917 	/* need this before calling reserve_initrd */
918 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
919 		max_low_pfn = e820_end_of_low_ram_pfn();
920 	else
921 		max_low_pfn = max_pfn;
922 
923 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
924 #endif
925 
926 	/*
927 	 * Find and reserve possible boot-time SMP configuration:
928 	 */
929 	find_smp_config();
930 
931 	reserve_ibft_region();
932 
933 	/*
934 	 * Need to conclude brk, before memblock_x86_fill()
935 	 *  it could use memblock_find_in_range, could overlap with
936 	 *  brk area.
937 	 */
938 	reserve_brk();
939 
940 	memblock.current_limit = get_max_mapped();
941 	memblock_x86_fill();
942 
943 	/* preallocate 4k for mptable mpc */
944 	early_reserve_e820_mpc_new();
945 
946 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
947 	setup_bios_corruption_check();
948 #endif
949 
950 	printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
951 			max_pfn_mapped<<PAGE_SHIFT);
952 
953 	reserve_trampoline_memory();
954 
955 #ifdef CONFIG_ACPI_SLEEP
956 	/*
957 	 * Reserve low memory region for sleep support.
958 	 * even before init_memory_mapping
959 	 */
960 	acpi_reserve_wakeup_memory();
961 #endif
962 	init_gbpages();
963 
964 	/* max_pfn_mapped is updated here */
965 	max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
966 	max_pfn_mapped = max_low_pfn_mapped;
967 
968 #ifdef CONFIG_X86_64
969 	if (max_pfn > max_low_pfn) {
970 		max_pfn_mapped = init_memory_mapping(1UL<<32,
971 						     max_pfn<<PAGE_SHIFT);
972 		/* can we preseve max_low_pfn ?*/
973 		max_low_pfn = max_pfn;
974 	}
975 #endif
976 	memblock.current_limit = get_max_mapped();
977 
978 	/*
979 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
980 	 */
981 
982 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
983 	if (init_ohci1394_dma_early)
984 		init_ohci1394_dma_on_all_controllers();
985 #endif
986 
987 	reserve_initrd();
988 
989 	reserve_crashkernel();
990 
991 	vsmp_init();
992 
993 	io_delay_init();
994 
995 	/*
996 	 * Parse the ACPI tables for possible boot-time SMP configuration.
997 	 */
998 	acpi_boot_table_init();
999 
1000 	early_acpi_boot_init();
1001 
1002 #ifdef CONFIG_ACPI_NUMA
1003 	/*
1004 	 * Parse SRAT to discover nodes.
1005 	 */
1006 	acpi = acpi_numa_init();
1007 #endif
1008 
1009 #ifdef CONFIG_AMD_NUMA
1010 	if (!acpi)
1011 		amd = !amd_numa_init(0, max_pfn);
1012 #endif
1013 
1014 	initmem_init(0, max_pfn, acpi, amd);
1015 	memblock_find_dma_reserve();
1016 	dma32_reserve_bootmem();
1017 
1018 #ifdef CONFIG_KVM_CLOCK
1019 	kvmclock_init();
1020 #endif
1021 
1022 	x86_init.paging.pagetable_setup_start(swapper_pg_dir);
1023 	paging_init();
1024 	x86_init.paging.pagetable_setup_done(swapper_pg_dir);
1025 
1026 #ifdef CONFIG_X86_32
1027 	/* sync back kernel address range */
1028 	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1029 			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
1030 			KERNEL_PGD_PTRS);
1031 #endif
1032 
1033 	tboot_probe();
1034 
1035 #ifdef CONFIG_X86_64
1036 	map_vsyscall();
1037 #endif
1038 
1039 	generic_apic_probe();
1040 
1041 	early_quirks();
1042 
1043 	/*
1044 	 * Read APIC and some other early information from ACPI tables.
1045 	 */
1046 	acpi_boot_init();
1047 
1048 	sfi_init();
1049 
1050 	/*
1051 	 * get boot-time SMP configuration:
1052 	 */
1053 	if (smp_found_config)
1054 		get_smp_config();
1055 
1056 	prefill_possible_map();
1057 
1058 #ifdef CONFIG_X86_64
1059 	init_cpu_to_node();
1060 #endif
1061 
1062 	init_apic_mappings();
1063 	ioapic_and_gsi_init();
1064 
1065 	kvm_guest_init();
1066 
1067 	e820_reserve_resources();
1068 	e820_mark_nosave_regions(max_low_pfn);
1069 
1070 	x86_init.resources.reserve_resources();
1071 
1072 	e820_setup_gap();
1073 
1074 #ifdef CONFIG_VT
1075 #if defined(CONFIG_VGA_CONSOLE)
1076 	if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1077 		conswitchp = &vga_con;
1078 #elif defined(CONFIG_DUMMY_CONSOLE)
1079 	conswitchp = &dummy_con;
1080 #endif
1081 #endif
1082 	x86_init.oem.banner();
1083 
1084 	x86_init.timers.wallclock_init();
1085 
1086 	mcheck_init();
1087 
1088 	local_irq_save(flags);
1089 	arch_init_ideal_nop5();
1090 	local_irq_restore(flags);
1091 }
1092 
1093 #ifdef CONFIG_X86_32
1094 
1095 static struct resource video_ram_resource = {
1096 	.name	= "Video RAM area",
1097 	.start	= 0xa0000,
1098 	.end	= 0xbffff,
1099 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1100 };
1101 
1102 void __init i386_reserve_resources(void)
1103 {
1104 	request_resource(&iomem_resource, &video_ram_resource);
1105 	reserve_standard_io_resources();
1106 }
1107 
1108 #endif /* CONFIG_X86_32 */
1109