xref: /openbmc/linux/arch/x86/kernel/setup.c (revision 9b5db89e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  Copyright (C) 1995  Linus Torvalds
4  *
5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6  *
7  *  Memory region support
8  *	David Parsons <orc@pell.chi.il.us>, July-August 1999
9  *
10  *  Added E820 sanitization routine (removes overlapping memory regions);
11  *  Brian Moyle <bmoyle@mvista.com>, February 2001
12  *
13  * Moved CPU detection code to cpu/${cpu}.c
14  *    Patrick Mochel <mochel@osdl.org>, March 2002
15  *
16  *  Provisions for empty E820 memory regions (reported by certain BIOSes).
17  *  Alex Achenbach <xela@slit.de>, December 2002.
18  *
19  */
20 
21 /*
22  * This file handles the architecture-dependent parts of initialization
23  */
24 
25 #include <linux/sched.h>
26 #include <linux/mm.h>
27 #include <linux/mmzone.h>
28 #include <linux/screen_info.h>
29 #include <linux/ioport.h>
30 #include <linux/acpi.h>
31 #include <linux/sfi.h>
32 #include <linux/apm_bios.h>
33 #include <linux/initrd.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/export.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 #include <xen/xen.h>
54 #include <uapi/linux/mount.h>
55 
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
58 #include <linux/stddef.h>
59 #include <linux/unistd.h>
60 #include <linux/ptrace.h>
61 #include <linux/user.h>
62 #include <linux/delay.h>
63 
64 #include <linux/kallsyms.h>
65 #include <linux/cpufreq.h>
66 #include <linux/dma-mapping.h>
67 #include <linux/ctype.h>
68 #include <linux/uaccess.h>
69 
70 #include <linux/percpu.h>
71 #include <linux/crash_dump.h>
72 #include <linux/tboot.h>
73 #include <linux/jiffies.h>
74 #include <linux/mem_encrypt.h>
75 #include <linux/sizes.h>
76 
77 #include <linux/usb/xhci-dbgp.h>
78 #include <video/edid.h>
79 
80 #include <asm/mtrr.h>
81 #include <asm/apic.h>
82 #include <asm/realmode.h>
83 #include <asm/e820/api.h>
84 #include <asm/mpspec.h>
85 #include <asm/setup.h>
86 #include <asm/efi.h>
87 #include <asm/timer.h>
88 #include <asm/i8259.h>
89 #include <asm/sections.h>
90 #include <asm/io_apic.h>
91 #include <asm/ist.h>
92 #include <asm/setup_arch.h>
93 #include <asm/bios_ebda.h>
94 #include <asm/cacheflush.h>
95 #include <asm/processor.h>
96 #include <asm/bugs.h>
97 #include <asm/kasan.h>
98 
99 #include <asm/vsyscall.h>
100 #include <asm/cpu.h>
101 #include <asm/desc.h>
102 #include <asm/dma.h>
103 #include <asm/iommu.h>
104 #include <asm/gart.h>
105 #include <asm/mmu_context.h>
106 #include <asm/proto.h>
107 
108 #include <asm/paravirt.h>
109 #include <asm/hypervisor.h>
110 #include <asm/olpc_ofw.h>
111 
112 #include <asm/percpu.h>
113 #include <asm/topology.h>
114 #include <asm/apicdef.h>
115 #include <asm/amd_nb.h>
116 #include <asm/mce.h>
117 #include <asm/alternative.h>
118 #include <asm/prom.h>
119 #include <asm/microcode.h>
120 #include <asm/kaslr.h>
121 #include <asm/unwind.h>
122 
123 /*
124  * max_low_pfn_mapped: highest direct mapped pfn under 4GB
125  * max_pfn_mapped:     highest direct mapped pfn over 4GB
126  *
127  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
128  * represented by pfn_mapped
129  */
130 unsigned long max_low_pfn_mapped;
131 unsigned long max_pfn_mapped;
132 
133 #ifdef CONFIG_DMI
134 RESERVE_BRK(dmi_alloc, 65536);
135 #endif
136 
137 
138 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
139 unsigned long _brk_end = (unsigned long)__brk_base;
140 
141 struct boot_params boot_params;
142 
143 /*
144  * Machine setup..
145  */
146 static struct resource data_resource = {
147 	.name	= "Kernel data",
148 	.start	= 0,
149 	.end	= 0,
150 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
151 };
152 
153 static struct resource code_resource = {
154 	.name	= "Kernel code",
155 	.start	= 0,
156 	.end	= 0,
157 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
158 };
159 
160 static struct resource bss_resource = {
161 	.name	= "Kernel bss",
162 	.start	= 0,
163 	.end	= 0,
164 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
165 };
166 
167 
168 #ifdef CONFIG_X86_32
169 /* cpu data as detected by the assembly code in head_32.S */
170 struct cpuinfo_x86 new_cpu_data;
171 
172 /* common cpu data for all cpus */
173 struct cpuinfo_x86 boot_cpu_data __read_mostly;
174 EXPORT_SYMBOL(boot_cpu_data);
175 
176 unsigned int def_to_bigsmp;
177 
178 /* for MCA, but anyone else can use it if they want */
179 unsigned int machine_id;
180 unsigned int machine_submodel_id;
181 unsigned int BIOS_revision;
182 
183 struct apm_info apm_info;
184 EXPORT_SYMBOL(apm_info);
185 
186 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
187 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
188 struct ist_info ist_info;
189 EXPORT_SYMBOL(ist_info);
190 #else
191 struct ist_info ist_info;
192 #endif
193 
194 #else
195 struct cpuinfo_x86 boot_cpu_data __read_mostly;
196 EXPORT_SYMBOL(boot_cpu_data);
197 #endif
198 
199 
200 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
201 __visible unsigned long mmu_cr4_features __ro_after_init;
202 #else
203 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
204 #endif
205 
206 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
207 int bootloader_type, bootloader_version;
208 
209 /*
210  * Setup options
211  */
212 struct screen_info screen_info;
213 EXPORT_SYMBOL(screen_info);
214 struct edid_info edid_info;
215 EXPORT_SYMBOL_GPL(edid_info);
216 
217 extern int root_mountflags;
218 
219 unsigned long saved_video_mode;
220 
221 #define RAMDISK_IMAGE_START_MASK	0x07FF
222 #define RAMDISK_PROMPT_FLAG		0x8000
223 #define RAMDISK_LOAD_FLAG		0x4000
224 
225 static char __initdata command_line[COMMAND_LINE_SIZE];
226 #ifdef CONFIG_CMDLINE_BOOL
227 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
228 #endif
229 
230 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
231 struct edd edd;
232 #ifdef CONFIG_EDD_MODULE
233 EXPORT_SYMBOL(edd);
234 #endif
235 /**
236  * copy_edd() - Copy the BIOS EDD information
237  *              from boot_params into a safe place.
238  *
239  */
240 static inline void __init copy_edd(void)
241 {
242      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
243 	    sizeof(edd.mbr_signature));
244      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
245      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
246      edd.edd_info_nr = boot_params.eddbuf_entries;
247 }
248 #else
249 static inline void __init copy_edd(void)
250 {
251 }
252 #endif
253 
254 void * __init extend_brk(size_t size, size_t align)
255 {
256 	size_t mask = align - 1;
257 	void *ret;
258 
259 	BUG_ON(_brk_start == 0);
260 	BUG_ON(align & mask);
261 
262 	_brk_end = (_brk_end + mask) & ~mask;
263 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
264 
265 	ret = (void *)_brk_end;
266 	_brk_end += size;
267 
268 	memset(ret, 0, size);
269 
270 	return ret;
271 }
272 
273 #ifdef CONFIG_X86_32
274 static void __init cleanup_highmap(void)
275 {
276 }
277 #endif
278 
279 static void __init reserve_brk(void)
280 {
281 	if (_brk_end > _brk_start)
282 		memblock_reserve(__pa_symbol(_brk_start),
283 				 _brk_end - _brk_start);
284 
285 	/* Mark brk area as locked down and no longer taking any
286 	   new allocations */
287 	_brk_start = 0;
288 }
289 
290 u64 relocated_ramdisk;
291 
292 #ifdef CONFIG_BLK_DEV_INITRD
293 
294 static u64 __init get_ramdisk_image(void)
295 {
296 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
297 
298 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
299 
300 	return ramdisk_image;
301 }
302 static u64 __init get_ramdisk_size(void)
303 {
304 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
305 
306 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
307 
308 	return ramdisk_size;
309 }
310 
311 static void __init relocate_initrd(void)
312 {
313 	/* Assume only end is not page aligned */
314 	u64 ramdisk_image = get_ramdisk_image();
315 	u64 ramdisk_size  = get_ramdisk_size();
316 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
317 
318 	/* We need to move the initrd down into directly mapped mem */
319 	relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
320 						   area_size, PAGE_SIZE);
321 
322 	if (!relocated_ramdisk)
323 		panic("Cannot find place for new RAMDISK of size %lld\n",
324 		      ramdisk_size);
325 
326 	/* Note: this includes all the mem currently occupied by
327 	   the initrd, we rely on that fact to keep the data intact. */
328 	memblock_reserve(relocated_ramdisk, area_size);
329 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
330 	initrd_end   = initrd_start + ramdisk_size;
331 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
332 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
333 
334 	copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
335 
336 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
337 		" [mem %#010llx-%#010llx]\n",
338 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
339 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
340 }
341 
342 static void __init early_reserve_initrd(void)
343 {
344 	/* Assume only end is not page aligned */
345 	u64 ramdisk_image = get_ramdisk_image();
346 	u64 ramdisk_size  = get_ramdisk_size();
347 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
348 
349 	if (!boot_params.hdr.type_of_loader ||
350 	    !ramdisk_image || !ramdisk_size)
351 		return;		/* No initrd provided by bootloader */
352 
353 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
354 }
355 static void __init reserve_initrd(void)
356 {
357 	/* Assume only end is not page aligned */
358 	u64 ramdisk_image = get_ramdisk_image();
359 	u64 ramdisk_size  = get_ramdisk_size();
360 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
361 	u64 mapped_size;
362 
363 	if (!boot_params.hdr.type_of_loader ||
364 	    !ramdisk_image || !ramdisk_size)
365 		return;		/* No initrd provided by bootloader */
366 
367 	initrd_start = 0;
368 
369 	mapped_size = memblock_mem_size(max_pfn_mapped);
370 	if (ramdisk_size >= (mapped_size>>1))
371 		panic("initrd too large to handle, "
372 		       "disabling initrd (%lld needed, %lld available)\n",
373 		       ramdisk_size, mapped_size>>1);
374 
375 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
376 			ramdisk_end - 1);
377 
378 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
379 				PFN_DOWN(ramdisk_end))) {
380 		/* All are mapped, easy case */
381 		initrd_start = ramdisk_image + PAGE_OFFSET;
382 		initrd_end = initrd_start + ramdisk_size;
383 		return;
384 	}
385 
386 	relocate_initrd();
387 
388 	memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
389 }
390 
391 #else
392 static void __init early_reserve_initrd(void)
393 {
394 }
395 static void __init reserve_initrd(void)
396 {
397 }
398 #endif /* CONFIG_BLK_DEV_INITRD */
399 
400 static void __init parse_setup_data(void)
401 {
402 	struct setup_data *data;
403 	u64 pa_data, pa_next;
404 
405 	pa_data = boot_params.hdr.setup_data;
406 	while (pa_data) {
407 		u32 data_len, data_type;
408 
409 		data = early_memremap(pa_data, sizeof(*data));
410 		data_len = data->len + sizeof(struct setup_data);
411 		data_type = data->type;
412 		pa_next = data->next;
413 		early_memunmap(data, sizeof(*data));
414 
415 		switch (data_type) {
416 		case SETUP_E820_EXT:
417 			e820__memory_setup_extended(pa_data, data_len);
418 			break;
419 		case SETUP_DTB:
420 			add_dtb(pa_data);
421 			break;
422 		case SETUP_EFI:
423 			parse_efi_setup(pa_data, data_len);
424 			break;
425 		default:
426 			break;
427 		}
428 		pa_data = pa_next;
429 	}
430 }
431 
432 static void __init memblock_x86_reserve_range_setup_data(void)
433 {
434 	struct setup_data *data;
435 	u64 pa_data;
436 
437 	pa_data = boot_params.hdr.setup_data;
438 	while (pa_data) {
439 		data = early_memremap(pa_data, sizeof(*data));
440 		memblock_reserve(pa_data, sizeof(*data) + data->len);
441 		pa_data = data->next;
442 		early_memunmap(data, sizeof(*data));
443 	}
444 }
445 
446 /*
447  * --------- Crashkernel reservation ------------------------------
448  */
449 
450 #ifdef CONFIG_KEXEC_CORE
451 
452 /* 16M alignment for crash kernel regions */
453 #define CRASH_ALIGN		SZ_16M
454 
455 /*
456  * Keep the crash kernel below this limit.  On 32 bits earlier kernels
457  * would limit the kernel to the low 512 MiB due to mapping restrictions.
458  */
459 #ifdef CONFIG_X86_32
460 # define CRASH_ADDR_LOW_MAX	SZ_512M
461 # define CRASH_ADDR_HIGH_MAX	SZ_512M
462 #else
463 # define CRASH_ADDR_LOW_MAX	SZ_4G
464 # define CRASH_ADDR_HIGH_MAX	MAXMEM
465 #endif
466 
467 static int __init reserve_crashkernel_low(void)
468 {
469 #ifdef CONFIG_X86_64
470 	unsigned long long base, low_base = 0, low_size = 0;
471 	unsigned long total_low_mem;
472 	int ret;
473 
474 	total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
475 
476 	/* crashkernel=Y,low */
477 	ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
478 	if (ret) {
479 		/*
480 		 * two parts from lib/swiotlb.c:
481 		 * -swiotlb size: user-specified with swiotlb= or default.
482 		 *
483 		 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
484 		 * to 8M for other buffers that may need to stay low too. Also
485 		 * make sure we allocate enough extra low memory so that we
486 		 * don't run out of DMA buffers for 32-bit devices.
487 		 */
488 		low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
489 	} else {
490 		/* passed with crashkernel=0,low ? */
491 		if (!low_size)
492 			return 0;
493 	}
494 
495 	low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN);
496 	if (!low_base) {
497 		pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
498 		       (unsigned long)(low_size >> 20));
499 		return -ENOMEM;
500 	}
501 
502 	ret = memblock_reserve(low_base, low_size);
503 	if (ret) {
504 		pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
505 		return ret;
506 	}
507 
508 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
509 		(unsigned long)(low_size >> 20),
510 		(unsigned long)(low_base >> 20),
511 		(unsigned long)(total_low_mem >> 20));
512 
513 	crashk_low_res.start = low_base;
514 	crashk_low_res.end   = low_base + low_size - 1;
515 	insert_resource(&iomem_resource, &crashk_low_res);
516 #endif
517 	return 0;
518 }
519 
520 static void __init reserve_crashkernel(void)
521 {
522 	unsigned long long crash_size, crash_base, total_mem;
523 	bool high = false;
524 	int ret;
525 
526 	total_mem = memblock_phys_mem_size();
527 
528 	/* crashkernel=XM */
529 	ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
530 	if (ret != 0 || crash_size <= 0) {
531 		/* crashkernel=X,high */
532 		ret = parse_crashkernel_high(boot_command_line, total_mem,
533 					     &crash_size, &crash_base);
534 		if (ret != 0 || crash_size <= 0)
535 			return;
536 		high = true;
537 	}
538 
539 	if (xen_pv_domain()) {
540 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
541 		return;
542 	}
543 
544 	/* 0 means: find the address automatically */
545 	if (!crash_base) {
546 		/*
547 		 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
548 		 * crashkernel=x,high reserves memory over 4G, also allocates
549 		 * 256M extra low memory for DMA buffers and swiotlb.
550 		 * But the extra memory is not required for all machines.
551 		 * So try low memory first and fall back to high memory
552 		 * unless "crashkernel=size[KMG],high" is specified.
553 		 */
554 		if (!high)
555 			crash_base = memblock_find_in_range(CRASH_ALIGN,
556 						CRASH_ADDR_LOW_MAX,
557 						crash_size, CRASH_ALIGN);
558 		if (!crash_base)
559 			crash_base = memblock_find_in_range(CRASH_ALIGN,
560 						CRASH_ADDR_HIGH_MAX,
561 						crash_size, CRASH_ALIGN);
562 		if (!crash_base) {
563 			pr_info("crashkernel reservation failed - No suitable area found.\n");
564 			return;
565 		}
566 	} else {
567 		unsigned long long start;
568 
569 		start = memblock_find_in_range(crash_base,
570 					       crash_base + crash_size,
571 					       crash_size, 1 << 20);
572 		if (start != crash_base) {
573 			pr_info("crashkernel reservation failed - memory is in use.\n");
574 			return;
575 		}
576 	}
577 	ret = memblock_reserve(crash_base, crash_size);
578 	if (ret) {
579 		pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
580 		return;
581 	}
582 
583 	if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
584 		memblock_free(crash_base, crash_size);
585 		return;
586 	}
587 
588 	pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
589 		(unsigned long)(crash_size >> 20),
590 		(unsigned long)(crash_base >> 20),
591 		(unsigned long)(total_mem >> 20));
592 
593 	crashk_res.start = crash_base;
594 	crashk_res.end   = crash_base + crash_size - 1;
595 	insert_resource(&iomem_resource, &crashk_res);
596 }
597 #else
598 static void __init reserve_crashkernel(void)
599 {
600 }
601 #endif
602 
603 static struct resource standard_io_resources[] = {
604 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
605 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
606 	{ .name = "pic1", .start = 0x20, .end = 0x21,
607 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
608 	{ .name = "timer0", .start = 0x40, .end = 0x43,
609 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
610 	{ .name = "timer1", .start = 0x50, .end = 0x53,
611 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
612 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
613 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
614 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
615 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
616 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
617 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
618 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
619 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
620 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
621 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
622 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
623 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
624 };
625 
626 void __init reserve_standard_io_resources(void)
627 {
628 	int i;
629 
630 	/* request I/O space for devices used on all i[345]86 PCs */
631 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
632 		request_resource(&ioport_resource, &standard_io_resources[i]);
633 
634 }
635 
636 static __init void reserve_ibft_region(void)
637 {
638 	unsigned long addr, size = 0;
639 
640 	addr = find_ibft_region(&size);
641 
642 	if (size)
643 		memblock_reserve(addr, size);
644 }
645 
646 static bool __init snb_gfx_workaround_needed(void)
647 {
648 #ifdef CONFIG_PCI
649 	int i;
650 	u16 vendor, devid;
651 	static const __initconst u16 snb_ids[] = {
652 		0x0102,
653 		0x0112,
654 		0x0122,
655 		0x0106,
656 		0x0116,
657 		0x0126,
658 		0x010a,
659 	};
660 
661 	/* Assume no if something weird is going on with PCI */
662 	if (!early_pci_allowed())
663 		return false;
664 
665 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
666 	if (vendor != 0x8086)
667 		return false;
668 
669 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
670 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
671 		if (devid == snb_ids[i])
672 			return true;
673 #endif
674 
675 	return false;
676 }
677 
678 /*
679  * Sandy Bridge graphics has trouble with certain ranges, exclude
680  * them from allocation.
681  */
682 static void __init trim_snb_memory(void)
683 {
684 	static const __initconst unsigned long bad_pages[] = {
685 		0x20050000,
686 		0x20110000,
687 		0x20130000,
688 		0x20138000,
689 		0x40004000,
690 	};
691 	int i;
692 
693 	if (!snb_gfx_workaround_needed())
694 		return;
695 
696 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
697 
698 	/*
699 	 * Reserve all memory below the 1 MB mark that has not
700 	 * already been reserved.
701 	 */
702 	memblock_reserve(0, 1<<20);
703 
704 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
705 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
706 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
707 			       bad_pages[i]);
708 	}
709 }
710 
711 /*
712  * Here we put platform-specific memory range workarounds, i.e.
713  * memory known to be corrupt or otherwise in need to be reserved on
714  * specific platforms.
715  *
716  * If this gets used more widely it could use a real dispatch mechanism.
717  */
718 static void __init trim_platform_memory_ranges(void)
719 {
720 	trim_snb_memory();
721 }
722 
723 static void __init trim_bios_range(void)
724 {
725 	/*
726 	 * A special case is the first 4Kb of memory;
727 	 * This is a BIOS owned area, not kernel ram, but generally
728 	 * not listed as such in the E820 table.
729 	 *
730 	 * This typically reserves additional memory (64KiB by default)
731 	 * since some BIOSes are known to corrupt low memory.  See the
732 	 * Kconfig help text for X86_RESERVE_LOW.
733 	 */
734 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
735 
736 	/*
737 	 * special case: Some BIOSen report the PC BIOS
738 	 * area (640->1Mb) as ram even though it is not.
739 	 * take them out.
740 	 */
741 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
742 
743 	e820__update_table(e820_table);
744 }
745 
746 /* called before trim_bios_range() to spare extra sanitize */
747 static void __init e820_add_kernel_range(void)
748 {
749 	u64 start = __pa_symbol(_text);
750 	u64 size = __pa_symbol(_end) - start;
751 
752 	/*
753 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
754 	 * attempt to fix it by adding the range. We may have a confused BIOS,
755 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
756 	 * exclude kernel range. If we really are running on top non-RAM,
757 	 * we will crash later anyways.
758 	 */
759 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
760 		return;
761 
762 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
763 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
764 	e820__range_add(start, size, E820_TYPE_RAM);
765 }
766 
767 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
768 
769 static int __init parse_reservelow(char *p)
770 {
771 	unsigned long long size;
772 
773 	if (!p)
774 		return -EINVAL;
775 
776 	size = memparse(p, &p);
777 
778 	if (size < 4096)
779 		size = 4096;
780 
781 	if (size > 640*1024)
782 		size = 640*1024;
783 
784 	reserve_low = size;
785 
786 	return 0;
787 }
788 
789 early_param("reservelow", parse_reservelow);
790 
791 static void __init trim_low_memory_range(void)
792 {
793 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
794 }
795 
796 /*
797  * Dump out kernel offset information on panic.
798  */
799 static int
800 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
801 {
802 	if (kaslr_enabled()) {
803 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
804 			 kaslr_offset(),
805 			 __START_KERNEL,
806 			 __START_KERNEL_map,
807 			 MODULES_VADDR-1);
808 	} else {
809 		pr_emerg("Kernel Offset: disabled\n");
810 	}
811 
812 	return 0;
813 }
814 
815 /*
816  * Determine if we were loaded by an EFI loader.  If so, then we have also been
817  * passed the efi memmap, systab, etc., so we should use these data structures
818  * for initialization.  Note, the efi init code path is determined by the
819  * global efi_enabled. This allows the same kernel image to be used on existing
820  * systems (with a traditional BIOS) as well as on EFI systems.
821  */
822 /*
823  * setup_arch - architecture-specific boot-time initializations
824  *
825  * Note: On x86_64, fixmaps are ready for use even before this is called.
826  */
827 
828 void __init setup_arch(char **cmdline_p)
829 {
830 	memblock_reserve(__pa_symbol(_text),
831 			 (unsigned long)__bss_stop - (unsigned long)_text);
832 
833 	/*
834 	 * Make sure page 0 is always reserved because on systems with
835 	 * L1TF its contents can be leaked to user processes.
836 	 */
837 	memblock_reserve(0, PAGE_SIZE);
838 
839 	early_reserve_initrd();
840 
841 	/*
842 	 * At this point everything still needed from the boot loader
843 	 * or BIOS or kernel text should be early reserved or marked not
844 	 * RAM in e820. All other memory is free game.
845 	 */
846 
847 #ifdef CONFIG_X86_32
848 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
849 
850 	/*
851 	 * copy kernel address range established so far and switch
852 	 * to the proper swapper page table
853 	 */
854 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
855 			initial_page_table + KERNEL_PGD_BOUNDARY,
856 			KERNEL_PGD_PTRS);
857 
858 	load_cr3(swapper_pg_dir);
859 	/*
860 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
861 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
862 	 * will not flush anything because the cpu quirk which clears
863 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
864 	 * load_cr3() above the TLB has been flushed already. The
865 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
866 	 * so proper operation is guaranteed.
867 	 */
868 	__flush_tlb_all();
869 #else
870 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
871 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
872 #endif
873 
874 	/*
875 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
876 	 * reserve_top(), so do this before touching the ioremap area.
877 	 */
878 	olpc_ofw_detect();
879 
880 	idt_setup_early_traps();
881 	early_cpu_init();
882 	arch_init_ideal_nops();
883 	jump_label_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 #endif
895 	saved_video_mode = boot_params.hdr.vid_mode;
896 	bootloader_type = boot_params.hdr.type_of_loader;
897 	if ((bootloader_type >> 4) == 0xe) {
898 		bootloader_type &= 0xf;
899 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
900 	}
901 	bootloader_version  = bootloader_type & 0xf;
902 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
903 
904 #ifdef CONFIG_BLK_DEV_RAM
905 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
906 	rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
907 	rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
908 #endif
909 #ifdef CONFIG_EFI
910 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
911 		     EFI32_LOADER_SIGNATURE, 4)) {
912 		set_bit(EFI_BOOT, &efi.flags);
913 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
914 		     EFI64_LOADER_SIGNATURE, 4)) {
915 		set_bit(EFI_BOOT, &efi.flags);
916 		set_bit(EFI_64BIT, &efi.flags);
917 	}
918 #endif
919 
920 	x86_init.oem.arch_setup();
921 
922 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
923 	e820__memory_setup();
924 	parse_setup_data();
925 
926 	copy_edd();
927 
928 	if (!boot_params.hdr.root_flags)
929 		root_mountflags &= ~MS_RDONLY;
930 	init_mm.start_code = (unsigned long) _text;
931 	init_mm.end_code = (unsigned long) _etext;
932 	init_mm.end_data = (unsigned long) _edata;
933 	init_mm.brk = _brk_end;
934 
935 	mpx_mm_init(&init_mm);
936 
937 	code_resource.start = __pa_symbol(_text);
938 	code_resource.end = __pa_symbol(_etext)-1;
939 	data_resource.start = __pa_symbol(_etext);
940 	data_resource.end = __pa_symbol(_edata)-1;
941 	bss_resource.start = __pa_symbol(__bss_start);
942 	bss_resource.end = __pa_symbol(__bss_stop)-1;
943 
944 #ifdef CONFIG_CMDLINE_BOOL
945 #ifdef CONFIG_CMDLINE_OVERRIDE
946 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
947 #else
948 	if (builtin_cmdline[0]) {
949 		/* append boot loader cmdline to builtin */
950 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
951 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
952 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
953 	}
954 #endif
955 #endif
956 
957 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
958 	*cmdline_p = command_line;
959 
960 	/*
961 	 * x86_configure_nx() is called before parse_early_param() to detect
962 	 * whether hardware doesn't support NX (so that the early EHCI debug
963 	 * console setup can safely call set_fixmap()). It may then be called
964 	 * again from within noexec_setup() during parsing early parameters
965 	 * to honor the respective command line option.
966 	 */
967 	x86_configure_nx();
968 
969 	parse_early_param();
970 
971 	if (efi_enabled(EFI_BOOT))
972 		efi_memblock_x86_reserve_range();
973 #ifdef CONFIG_MEMORY_HOTPLUG
974 	/*
975 	 * Memory used by the kernel cannot be hot-removed because Linux
976 	 * cannot migrate the kernel pages. When memory hotplug is
977 	 * enabled, we should prevent memblock from allocating memory
978 	 * for the kernel.
979 	 *
980 	 * ACPI SRAT records all hotpluggable memory ranges. But before
981 	 * SRAT is parsed, we don't know about it.
982 	 *
983 	 * The kernel image is loaded into memory at very early time. We
984 	 * cannot prevent this anyway. So on NUMA system, we set any
985 	 * node the kernel resides in as un-hotpluggable.
986 	 *
987 	 * Since on modern servers, one node could have double-digit
988 	 * gigabytes memory, we can assume the memory around the kernel
989 	 * image is also un-hotpluggable. So before SRAT is parsed, just
990 	 * allocate memory near the kernel image to try the best to keep
991 	 * the kernel away from hotpluggable memory.
992 	 */
993 	if (movable_node_is_enabled())
994 		memblock_set_bottom_up(true);
995 #endif
996 
997 	x86_report_nx();
998 
999 	/* after early param, so could get panic from serial */
1000 	memblock_x86_reserve_range_setup_data();
1001 
1002 	if (acpi_mps_check()) {
1003 #ifdef CONFIG_X86_LOCAL_APIC
1004 		disable_apic = 1;
1005 #endif
1006 		setup_clear_cpu_cap(X86_FEATURE_APIC);
1007 	}
1008 
1009 	e820__reserve_setup_data();
1010 	e820__finish_early_params();
1011 
1012 	if (efi_enabled(EFI_BOOT))
1013 		efi_init();
1014 
1015 	dmi_setup();
1016 
1017 	/*
1018 	 * VMware detection requires dmi to be available, so this
1019 	 * needs to be done after dmi_setup(), for the boot CPU.
1020 	 */
1021 	init_hypervisor_platform();
1022 
1023 	tsc_early_init();
1024 	x86_init.resources.probe_roms();
1025 
1026 	/* after parse_early_param, so could debug it */
1027 	insert_resource(&iomem_resource, &code_resource);
1028 	insert_resource(&iomem_resource, &data_resource);
1029 	insert_resource(&iomem_resource, &bss_resource);
1030 
1031 	e820_add_kernel_range();
1032 	trim_bios_range();
1033 #ifdef CONFIG_X86_32
1034 	if (ppro_with_ram_bug()) {
1035 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1036 				  E820_TYPE_RESERVED);
1037 		e820__update_table(e820_table);
1038 		printk(KERN_INFO "fixed physical RAM map:\n");
1039 		e820__print_table("bad_ppro");
1040 	}
1041 #else
1042 	early_gart_iommu_check();
1043 #endif
1044 
1045 	/*
1046 	 * partially used pages are not usable - thus
1047 	 * we are rounding upwards:
1048 	 */
1049 	max_pfn = e820__end_of_ram_pfn();
1050 
1051 	/* update e820 for memory not covered by WB MTRRs */
1052 	mtrr_bp_init();
1053 	if (mtrr_trim_uncached_memory(max_pfn))
1054 		max_pfn = e820__end_of_ram_pfn();
1055 
1056 	max_possible_pfn = max_pfn;
1057 
1058 	/*
1059 	 * This call is required when the CPU does not support PAT. If
1060 	 * mtrr_bp_init() invoked it already via pat_init() the call has no
1061 	 * effect.
1062 	 */
1063 	init_cache_modes();
1064 
1065 	/*
1066 	 * Define random base addresses for memory sections after max_pfn is
1067 	 * defined and before each memory section base is used.
1068 	 */
1069 	kernel_randomize_memory();
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 e820__memblock_setup()
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 	e820__memblock_setup();
1107 
1108 	reserve_bios_regions();
1109 
1110 	if (efi_enabled(EFI_MEMMAP)) {
1111 		efi_fake_memmap();
1112 		efi_find_mirror();
1113 		efi_esrt_init();
1114 
1115 		/*
1116 		 * The EFI specification says that boot service code won't be
1117 		 * called after ExitBootServices(). This is, in fact, a lie.
1118 		 */
1119 		efi_reserve_boot_services();
1120 	}
1121 
1122 	/* preallocate 4k for mptable mpc */
1123 	e820__memblock_alloc_reserved_mpc_new();
1124 
1125 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1126 	setup_bios_corruption_check();
1127 #endif
1128 
1129 #ifdef CONFIG_X86_32
1130 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1131 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1132 #endif
1133 
1134 	reserve_real_mode();
1135 
1136 	trim_platform_memory_ranges();
1137 	trim_low_memory_range();
1138 
1139 	init_mem_mapping();
1140 
1141 	idt_setup_early_pf();
1142 
1143 	/*
1144 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1145 	 * with the current CR4 value.  This may not be necessary, but
1146 	 * auditing all the early-boot CR4 manipulation would be needed to
1147 	 * rule it out.
1148 	 *
1149 	 * Mask off features that don't work outside long mode (just
1150 	 * PCIDE for now).
1151 	 */
1152 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1153 
1154 	memblock_set_current_limit(get_max_mapped());
1155 
1156 	/*
1157 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1158 	 */
1159 
1160 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1161 	if (init_ohci1394_dma_early)
1162 		init_ohci1394_dma_on_all_controllers();
1163 #endif
1164 	/* Allocate bigger log buffer */
1165 	setup_log_buf(1);
1166 
1167 	if (efi_enabled(EFI_BOOT)) {
1168 		switch (boot_params.secure_boot) {
1169 		case efi_secureboot_mode_disabled:
1170 			pr_info("Secure boot disabled\n");
1171 			break;
1172 		case efi_secureboot_mode_enabled:
1173 			pr_info("Secure boot enabled\n");
1174 			break;
1175 		default:
1176 			pr_info("Secure boot could not be determined\n");
1177 			break;
1178 		}
1179 	}
1180 
1181 	reserve_initrd();
1182 
1183 	acpi_table_upgrade();
1184 
1185 	vsmp_init();
1186 
1187 	io_delay_init();
1188 
1189 	early_platform_quirks();
1190 
1191 	/*
1192 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1193 	 */
1194 	acpi_boot_table_init();
1195 
1196 	early_acpi_boot_init();
1197 
1198 	initmem_init();
1199 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1200 
1201 	/*
1202 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1203 	 * won't consume hotpluggable memory.
1204 	 */
1205 	reserve_crashkernel();
1206 
1207 	memblock_find_dma_reserve();
1208 
1209 	if (!early_xdbc_setup_hardware())
1210 		early_xdbc_register_console();
1211 
1212 	x86_init.paging.pagetable_init();
1213 
1214 	kasan_init();
1215 
1216 	/*
1217 	 * Sync back kernel address range.
1218 	 *
1219 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1220 	 * this call?
1221 	 */
1222 	sync_initial_page_table();
1223 
1224 	tboot_probe();
1225 
1226 	map_vsyscall();
1227 
1228 	generic_apic_probe();
1229 
1230 	early_quirks();
1231 
1232 	/*
1233 	 * Read APIC and some other early information from ACPI tables.
1234 	 */
1235 	acpi_boot_init();
1236 	sfi_init();
1237 	x86_dtb_init();
1238 
1239 	/*
1240 	 * get boot-time SMP configuration:
1241 	 */
1242 	get_smp_config();
1243 
1244 	/*
1245 	 * Systems w/o ACPI and mptables might not have it mapped the local
1246 	 * APIC yet, but prefill_possible_map() might need to access it.
1247 	 */
1248 	init_apic_mappings();
1249 
1250 	prefill_possible_map();
1251 
1252 	init_cpu_to_node();
1253 
1254 	io_apic_init_mappings();
1255 
1256 	x86_init.hyper.guest_late_init();
1257 
1258 	e820__reserve_resources();
1259 	e820__register_nosave_regions(max_pfn);
1260 
1261 	x86_init.resources.reserve_resources();
1262 
1263 	e820__setup_pci_gap();
1264 
1265 #ifdef CONFIG_VT
1266 #if defined(CONFIG_VGA_CONSOLE)
1267 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1268 		conswitchp = &vga_con;
1269 #elif defined(CONFIG_DUMMY_CONSOLE)
1270 	conswitchp = &dummy_con;
1271 #endif
1272 #endif
1273 	x86_init.oem.banner();
1274 
1275 	x86_init.timers.wallclock_init();
1276 
1277 	mcheck_init();
1278 
1279 	register_refined_jiffies(CLOCK_TICK_RATE);
1280 
1281 #ifdef CONFIG_EFI
1282 	if (efi_enabled(EFI_BOOT))
1283 		efi_apply_memmap_quirks();
1284 #endif
1285 
1286 	unwind_init();
1287 }
1288 
1289 #ifdef CONFIG_X86_32
1290 
1291 static struct resource video_ram_resource = {
1292 	.name	= "Video RAM area",
1293 	.start	= 0xa0000,
1294 	.end	= 0xbffff,
1295 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1296 };
1297 
1298 void __init i386_reserve_resources(void)
1299 {
1300 	request_resource(&iomem_resource, &video_ram_resource);
1301 	reserve_standard_io_resources();
1302 }
1303 
1304 #endif /* CONFIG_X86_32 */
1305 
1306 static struct notifier_block kernel_offset_notifier = {
1307 	.notifier_call = dump_kernel_offset
1308 };
1309 
1310 static int __init register_kernel_offset_dumper(void)
1311 {
1312 	atomic_notifier_chain_register(&panic_notifier_list,
1313 					&kernel_offset_notifier);
1314 	return 0;
1315 }
1316 __initcall(register_kernel_offset_dumper);
1317