xref: /openbmc/linux/arch/x86/kernel/setup.c (revision 62975d27)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  Copyright (C) 1995  Linus Torvalds
4  *
5  * This file contains the setup_arch() code, which handles the architecture-dependent
6  * parts of early kernel initialization.
7  */
8 #include <linux/console.h>
9 #include <linux/crash_dump.h>
10 #include <linux/dmi.h>
11 #include <linux/efi.h>
12 #include <linux/init_ohci1394_dma.h>
13 #include <linux/initrd.h>
14 #include <linux/iscsi_ibft.h>
15 #include <linux/memblock.h>
16 #include <linux/pci.h>
17 #include <linux/root_dev.h>
18 #include <linux/sfi.h>
19 #include <linux/hugetlb.h>
20 #include <linux/tboot.h>
21 #include <linux/usb/xhci-dbgp.h>
22 
23 #include <uapi/linux/mount.h>
24 
25 #include <xen/xen.h>
26 
27 #include <asm/apic.h>
28 #include <asm/bios_ebda.h>
29 #include <asm/bugs.h>
30 #include <asm/cpu.h>
31 #include <asm/efi.h>
32 #include <asm/gart.h>
33 #include <asm/hypervisor.h>
34 #include <asm/io_apic.h>
35 #include <asm/kasan.h>
36 #include <asm/kaslr.h>
37 #include <asm/mce.h>
38 #include <asm/mtrr.h>
39 #include <asm/realmode.h>
40 #include <asm/olpc_ofw.h>
41 #include <asm/pci-direct.h>
42 #include <asm/prom.h>
43 #include <asm/proto.h>
44 #include <asm/unwind.h>
45 #include <asm/vsyscall.h>
46 #include <linux/vmalloc.h>
47 
48 /*
49  * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
50  * max_pfn_mapped:     highest directly mapped pfn > 4 GB
51  *
52  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
53  * represented by pfn_mapped[].
54  */
55 unsigned long max_low_pfn_mapped;
56 unsigned long max_pfn_mapped;
57 
58 #ifdef CONFIG_DMI
59 RESERVE_BRK(dmi_alloc, 65536);
60 #endif
61 
62 
63 /*
64  * Range of the BSS area. The size of the BSS area is determined
65  * at link time, with RESERVE_BRK*() facility reserving additional
66  * chunks.
67  */
68 unsigned long _brk_start = (unsigned long)__brk_base;
69 unsigned long _brk_end   = (unsigned long)__brk_base;
70 
71 struct boot_params boot_params;
72 
73 /*
74  * These are the four main kernel memory regions, we put them into
75  * the resource tree so that kdump tools and other debugging tools
76  * recover it:
77  */
78 
79 static struct resource rodata_resource = {
80 	.name	= "Kernel rodata",
81 	.start	= 0,
82 	.end	= 0,
83 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
84 };
85 
86 static struct resource data_resource = {
87 	.name	= "Kernel data",
88 	.start	= 0,
89 	.end	= 0,
90 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
91 };
92 
93 static struct resource code_resource = {
94 	.name	= "Kernel code",
95 	.start	= 0,
96 	.end	= 0,
97 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
98 };
99 
100 static struct resource bss_resource = {
101 	.name	= "Kernel bss",
102 	.start	= 0,
103 	.end	= 0,
104 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
105 };
106 
107 
108 #ifdef CONFIG_X86_32
109 /* CPU data as detected by the assembly code in head_32.S */
110 struct cpuinfo_x86 new_cpu_data;
111 
112 /* Common CPU data for all CPUs */
113 struct cpuinfo_x86 boot_cpu_data __read_mostly;
114 EXPORT_SYMBOL(boot_cpu_data);
115 
116 unsigned int def_to_bigsmp;
117 
118 /* For MCA, but anyone else can use it if they want */
119 unsigned int machine_id;
120 unsigned int machine_submodel_id;
121 unsigned int BIOS_revision;
122 
123 struct apm_info apm_info;
124 EXPORT_SYMBOL(apm_info);
125 
126 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
127 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
128 struct ist_info ist_info;
129 EXPORT_SYMBOL(ist_info);
130 #else
131 struct ist_info ist_info;
132 #endif
133 
134 #else
135 struct cpuinfo_x86 boot_cpu_data __read_mostly;
136 EXPORT_SYMBOL(boot_cpu_data);
137 #endif
138 
139 
140 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
141 __visible unsigned long mmu_cr4_features __ro_after_init;
142 #else
143 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
144 #endif
145 
146 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
147 int bootloader_type, bootloader_version;
148 
149 /*
150  * Setup options
151  */
152 struct screen_info screen_info;
153 EXPORT_SYMBOL(screen_info);
154 struct edid_info edid_info;
155 EXPORT_SYMBOL_GPL(edid_info);
156 
157 extern int root_mountflags;
158 
159 unsigned long saved_video_mode;
160 
161 #define RAMDISK_IMAGE_START_MASK	0x07FF
162 #define RAMDISK_PROMPT_FLAG		0x8000
163 #define RAMDISK_LOAD_FLAG		0x4000
164 
165 static char __initdata command_line[COMMAND_LINE_SIZE];
166 #ifdef CONFIG_CMDLINE_BOOL
167 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
168 #endif
169 
170 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
171 struct edd edd;
172 #ifdef CONFIG_EDD_MODULE
173 EXPORT_SYMBOL(edd);
174 #endif
175 /**
176  * copy_edd() - Copy the BIOS EDD information
177  *              from boot_params into a safe place.
178  *
179  */
180 static inline void __init copy_edd(void)
181 {
182      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
183 	    sizeof(edd.mbr_signature));
184      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
185      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
186      edd.edd_info_nr = boot_params.eddbuf_entries;
187 }
188 #else
189 static inline void __init copy_edd(void)
190 {
191 }
192 #endif
193 
194 void * __init extend_brk(size_t size, size_t align)
195 {
196 	size_t mask = align - 1;
197 	void *ret;
198 
199 	BUG_ON(_brk_start == 0);
200 	BUG_ON(align & mask);
201 
202 	_brk_end = (_brk_end + mask) & ~mask;
203 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
204 
205 	ret = (void *)_brk_end;
206 	_brk_end += size;
207 
208 	memset(ret, 0, size);
209 
210 	return ret;
211 }
212 
213 #ifdef CONFIG_X86_32
214 static void __init cleanup_highmap(void)
215 {
216 }
217 #endif
218 
219 static void __init reserve_brk(void)
220 {
221 	if (_brk_end > _brk_start)
222 		memblock_reserve(__pa_symbol(_brk_start),
223 				 _brk_end - _brk_start);
224 
225 	/* Mark brk area as locked down and no longer taking any
226 	   new allocations */
227 	_brk_start = 0;
228 }
229 
230 u64 relocated_ramdisk;
231 
232 #ifdef CONFIG_BLK_DEV_INITRD
233 
234 static u64 __init get_ramdisk_image(void)
235 {
236 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
237 
238 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
239 
240 	if (ramdisk_image == 0)
241 		ramdisk_image = phys_initrd_start;
242 
243 	return ramdisk_image;
244 }
245 static u64 __init get_ramdisk_size(void)
246 {
247 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
248 
249 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
250 
251 	if (ramdisk_size == 0)
252 		ramdisk_size = phys_initrd_size;
253 
254 	return ramdisk_size;
255 }
256 
257 static void __init relocate_initrd(void)
258 {
259 	/* Assume only end is not page aligned */
260 	u64 ramdisk_image = get_ramdisk_image();
261 	u64 ramdisk_size  = get_ramdisk_size();
262 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
263 
264 	/* We need to move the initrd down into directly mapped mem */
265 	relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
266 						   area_size, PAGE_SIZE);
267 
268 	if (!relocated_ramdisk)
269 		panic("Cannot find place for new RAMDISK of size %lld\n",
270 		      ramdisk_size);
271 
272 	/* Note: this includes all the mem currently occupied by
273 	   the initrd, we rely on that fact to keep the data intact. */
274 	memblock_reserve(relocated_ramdisk, area_size);
275 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
276 	initrd_end   = initrd_start + ramdisk_size;
277 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
278 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
279 
280 	copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
281 
282 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
283 		" [mem %#010llx-%#010llx]\n",
284 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
285 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
286 }
287 
288 static void __init early_reserve_initrd(void)
289 {
290 	/* Assume only end is not page aligned */
291 	u64 ramdisk_image = get_ramdisk_image();
292 	u64 ramdisk_size  = get_ramdisk_size();
293 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
294 
295 	if (!boot_params.hdr.type_of_loader ||
296 	    !ramdisk_image || !ramdisk_size)
297 		return;		/* No initrd provided by bootloader */
298 
299 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
300 }
301 static void __init reserve_initrd(void)
302 {
303 	/* Assume only end is not page aligned */
304 	u64 ramdisk_image = get_ramdisk_image();
305 	u64 ramdisk_size  = get_ramdisk_size();
306 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
307 	u64 mapped_size;
308 
309 	if (!boot_params.hdr.type_of_loader ||
310 	    !ramdisk_image || !ramdisk_size)
311 		return;		/* No initrd provided by bootloader */
312 
313 	initrd_start = 0;
314 
315 	mapped_size = memblock_mem_size(max_pfn_mapped);
316 	if (ramdisk_size >= (mapped_size>>1))
317 		panic("initrd too large to handle, "
318 		       "disabling initrd (%lld needed, %lld available)\n",
319 		       ramdisk_size, mapped_size>>1);
320 
321 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
322 			ramdisk_end - 1);
323 
324 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
325 				PFN_DOWN(ramdisk_end))) {
326 		/* All are mapped, easy case */
327 		initrd_start = ramdisk_image + PAGE_OFFSET;
328 		initrd_end = initrd_start + ramdisk_size;
329 		return;
330 	}
331 
332 	relocate_initrd();
333 
334 	memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
335 }
336 
337 #else
338 static void __init early_reserve_initrd(void)
339 {
340 }
341 static void __init reserve_initrd(void)
342 {
343 }
344 #endif /* CONFIG_BLK_DEV_INITRD */
345 
346 static void __init parse_setup_data(void)
347 {
348 	struct setup_data *data;
349 	u64 pa_data, pa_next;
350 
351 	pa_data = boot_params.hdr.setup_data;
352 	while (pa_data) {
353 		u32 data_len, data_type;
354 
355 		data = early_memremap(pa_data, sizeof(*data));
356 		data_len = data->len + sizeof(struct setup_data);
357 		data_type = data->type;
358 		pa_next = data->next;
359 		early_memunmap(data, sizeof(*data));
360 
361 		switch (data_type) {
362 		case SETUP_E820_EXT:
363 			e820__memory_setup_extended(pa_data, data_len);
364 			break;
365 		case SETUP_DTB:
366 			add_dtb(pa_data);
367 			break;
368 		case SETUP_EFI:
369 			parse_efi_setup(pa_data, data_len);
370 			break;
371 		default:
372 			break;
373 		}
374 		pa_data = pa_next;
375 	}
376 }
377 
378 static void __init memblock_x86_reserve_range_setup_data(void)
379 {
380 	struct setup_data *data;
381 	u64 pa_data;
382 
383 	pa_data = boot_params.hdr.setup_data;
384 	while (pa_data) {
385 		data = early_memremap(pa_data, sizeof(*data));
386 		memblock_reserve(pa_data, sizeof(*data) + data->len);
387 
388 		if (data->type == SETUP_INDIRECT &&
389 		    ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
390 			memblock_reserve(((struct setup_indirect *)data->data)->addr,
391 					 ((struct setup_indirect *)data->data)->len);
392 
393 		pa_data = data->next;
394 		early_memunmap(data, sizeof(*data));
395 	}
396 }
397 
398 /*
399  * --------- Crashkernel reservation ------------------------------
400  */
401 
402 #ifdef CONFIG_KEXEC_CORE
403 
404 /* 16M alignment for crash kernel regions */
405 #define CRASH_ALIGN		SZ_16M
406 
407 /*
408  * Keep the crash kernel below this limit.
409  *
410  * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
411  * due to mapping restrictions.
412  *
413  * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
414  * the upper limit of system RAM in 4-level paging mode. Since the kdump
415  * jump could be from 5-level paging to 4-level paging, the jump will fail if
416  * the kernel is put above 64 TB, and during the 1st kernel bootup there's
417  * no good way to detect the paging mode of the target kernel which will be
418  * loaded for dumping.
419  */
420 #ifdef CONFIG_X86_32
421 # define CRASH_ADDR_LOW_MAX	SZ_512M
422 # define CRASH_ADDR_HIGH_MAX	SZ_512M
423 #else
424 # define CRASH_ADDR_LOW_MAX	SZ_4G
425 # define CRASH_ADDR_HIGH_MAX	SZ_64T
426 #endif
427 
428 static int __init reserve_crashkernel_low(void)
429 {
430 #ifdef CONFIG_X86_64
431 	unsigned long long base, low_base = 0, low_size = 0;
432 	unsigned long total_low_mem;
433 	int ret;
434 
435 	total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
436 
437 	/* crashkernel=Y,low */
438 	ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
439 	if (ret) {
440 		/*
441 		 * two parts from kernel/dma/swiotlb.c:
442 		 * -swiotlb size: user-specified with swiotlb= or default.
443 		 *
444 		 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
445 		 * to 8M for other buffers that may need to stay low too. Also
446 		 * make sure we allocate enough extra low memory so that we
447 		 * don't run out of DMA buffers for 32-bit devices.
448 		 */
449 		low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
450 	} else {
451 		/* passed with crashkernel=0,low ? */
452 		if (!low_size)
453 			return 0;
454 	}
455 
456 	low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN);
457 	if (!low_base) {
458 		pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
459 		       (unsigned long)(low_size >> 20));
460 		return -ENOMEM;
461 	}
462 
463 	ret = memblock_reserve(low_base, low_size);
464 	if (ret) {
465 		pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
466 		return ret;
467 	}
468 
469 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
470 		(unsigned long)(low_size >> 20),
471 		(unsigned long)(low_base >> 20),
472 		(unsigned long)(total_low_mem >> 20));
473 
474 	crashk_low_res.start = low_base;
475 	crashk_low_res.end   = low_base + low_size - 1;
476 	insert_resource(&iomem_resource, &crashk_low_res);
477 #endif
478 	return 0;
479 }
480 
481 static void __init reserve_crashkernel(void)
482 {
483 	unsigned long long crash_size, crash_base, total_mem;
484 	bool high = false;
485 	int ret;
486 
487 	total_mem = memblock_phys_mem_size();
488 
489 	/* crashkernel=XM */
490 	ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
491 	if (ret != 0 || crash_size <= 0) {
492 		/* crashkernel=X,high */
493 		ret = parse_crashkernel_high(boot_command_line, total_mem,
494 					     &crash_size, &crash_base);
495 		if (ret != 0 || crash_size <= 0)
496 			return;
497 		high = true;
498 	}
499 
500 	if (xen_pv_domain()) {
501 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
502 		return;
503 	}
504 
505 	/* 0 means: find the address automatically */
506 	if (!crash_base) {
507 		/*
508 		 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
509 		 * crashkernel=x,high reserves memory over 4G, also allocates
510 		 * 256M extra low memory for DMA buffers and swiotlb.
511 		 * But the extra memory is not required for all machines.
512 		 * So try low memory first and fall back to high memory
513 		 * unless "crashkernel=size[KMG],high" is specified.
514 		 */
515 		if (!high)
516 			crash_base = memblock_find_in_range(CRASH_ALIGN,
517 						CRASH_ADDR_LOW_MAX,
518 						crash_size, CRASH_ALIGN);
519 		if (!crash_base)
520 			crash_base = memblock_find_in_range(CRASH_ALIGN,
521 						CRASH_ADDR_HIGH_MAX,
522 						crash_size, CRASH_ALIGN);
523 		if (!crash_base) {
524 			pr_info("crashkernel reservation failed - No suitable area found.\n");
525 			return;
526 		}
527 	} else {
528 		unsigned long long start;
529 
530 		start = memblock_find_in_range(crash_base,
531 					       crash_base + crash_size,
532 					       crash_size, 1 << 20);
533 		if (start != crash_base) {
534 			pr_info("crashkernel reservation failed - memory is in use.\n");
535 			return;
536 		}
537 	}
538 	ret = memblock_reserve(crash_base, crash_size);
539 	if (ret) {
540 		pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
541 		return;
542 	}
543 
544 	if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
545 		memblock_free(crash_base, crash_size);
546 		return;
547 	}
548 
549 	pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
550 		(unsigned long)(crash_size >> 20),
551 		(unsigned long)(crash_base >> 20),
552 		(unsigned long)(total_mem >> 20));
553 
554 	crashk_res.start = crash_base;
555 	crashk_res.end   = crash_base + crash_size - 1;
556 	insert_resource(&iomem_resource, &crashk_res);
557 }
558 #else
559 static void __init reserve_crashkernel(void)
560 {
561 }
562 #endif
563 
564 static struct resource standard_io_resources[] = {
565 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
566 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
567 	{ .name = "pic1", .start = 0x20, .end = 0x21,
568 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
569 	{ .name = "timer0", .start = 0x40, .end = 0x43,
570 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
571 	{ .name = "timer1", .start = 0x50, .end = 0x53,
572 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
573 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
574 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
575 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
576 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
577 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
578 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
579 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
580 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
581 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
582 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
583 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
584 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
585 };
586 
587 void __init reserve_standard_io_resources(void)
588 {
589 	int i;
590 
591 	/* request I/O space for devices used on all i[345]86 PCs */
592 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
593 		request_resource(&ioport_resource, &standard_io_resources[i]);
594 
595 }
596 
597 static __init void reserve_ibft_region(void)
598 {
599 	unsigned long addr, size = 0;
600 
601 	addr = find_ibft_region(&size);
602 
603 	if (size)
604 		memblock_reserve(addr, size);
605 }
606 
607 static bool __init snb_gfx_workaround_needed(void)
608 {
609 #ifdef CONFIG_PCI
610 	int i;
611 	u16 vendor, devid;
612 	static const __initconst u16 snb_ids[] = {
613 		0x0102,
614 		0x0112,
615 		0x0122,
616 		0x0106,
617 		0x0116,
618 		0x0126,
619 		0x010a,
620 	};
621 
622 	/* Assume no if something weird is going on with PCI */
623 	if (!early_pci_allowed())
624 		return false;
625 
626 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
627 	if (vendor != 0x8086)
628 		return false;
629 
630 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
631 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
632 		if (devid == snb_ids[i])
633 			return true;
634 #endif
635 
636 	return false;
637 }
638 
639 /*
640  * Sandy Bridge graphics has trouble with certain ranges, exclude
641  * them from allocation.
642  */
643 static void __init trim_snb_memory(void)
644 {
645 	static const __initconst unsigned long bad_pages[] = {
646 		0x20050000,
647 		0x20110000,
648 		0x20130000,
649 		0x20138000,
650 		0x40004000,
651 	};
652 	int i;
653 
654 	if (!snb_gfx_workaround_needed())
655 		return;
656 
657 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
658 
659 	/*
660 	 * Reserve all memory below the 1 MB mark that has not
661 	 * already been reserved.
662 	 */
663 	memblock_reserve(0, 1<<20);
664 
665 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
666 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
667 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
668 			       bad_pages[i]);
669 	}
670 }
671 
672 /*
673  * Here we put platform-specific memory range workarounds, i.e.
674  * memory known to be corrupt or otherwise in need to be reserved on
675  * specific platforms.
676  *
677  * If this gets used more widely it could use a real dispatch mechanism.
678  */
679 static void __init trim_platform_memory_ranges(void)
680 {
681 	trim_snb_memory();
682 }
683 
684 static void __init trim_bios_range(void)
685 {
686 	/*
687 	 * A special case is the first 4Kb of memory;
688 	 * This is a BIOS owned area, not kernel ram, but generally
689 	 * not listed as such in the E820 table.
690 	 *
691 	 * This typically reserves additional memory (64KiB by default)
692 	 * since some BIOSes are known to corrupt low memory.  See the
693 	 * Kconfig help text for X86_RESERVE_LOW.
694 	 */
695 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
696 
697 	/*
698 	 * special case: Some BIOSes report the PC BIOS
699 	 * area (640Kb -> 1Mb) as RAM even though it is not.
700 	 * take them out.
701 	 */
702 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
703 
704 	e820__update_table(e820_table);
705 }
706 
707 /* called before trim_bios_range() to spare extra sanitize */
708 static void __init e820_add_kernel_range(void)
709 {
710 	u64 start = __pa_symbol(_text);
711 	u64 size = __pa_symbol(_end) - start;
712 
713 	/*
714 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
715 	 * attempt to fix it by adding the range. We may have a confused BIOS,
716 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
717 	 * exclude kernel range. If we really are running on top non-RAM,
718 	 * we will crash later anyways.
719 	 */
720 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
721 		return;
722 
723 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
724 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
725 	e820__range_add(start, size, E820_TYPE_RAM);
726 }
727 
728 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
729 
730 static int __init parse_reservelow(char *p)
731 {
732 	unsigned long long size;
733 
734 	if (!p)
735 		return -EINVAL;
736 
737 	size = memparse(p, &p);
738 
739 	if (size < 4096)
740 		size = 4096;
741 
742 	if (size > 640*1024)
743 		size = 640*1024;
744 
745 	reserve_low = size;
746 
747 	return 0;
748 }
749 
750 early_param("reservelow", parse_reservelow);
751 
752 static void __init trim_low_memory_range(void)
753 {
754 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
755 }
756 
757 /*
758  * Dump out kernel offset information on panic.
759  */
760 static int
761 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
762 {
763 	if (kaslr_enabled()) {
764 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
765 			 kaslr_offset(),
766 			 __START_KERNEL,
767 			 __START_KERNEL_map,
768 			 MODULES_VADDR-1);
769 	} else {
770 		pr_emerg("Kernel Offset: disabled\n");
771 	}
772 
773 	return 0;
774 }
775 
776 /*
777  * Determine if we were loaded by an EFI loader.  If so, then we have also been
778  * passed the efi memmap, systab, etc., so we should use these data structures
779  * for initialization.  Note, the efi init code path is determined by the
780  * global efi_enabled. This allows the same kernel image to be used on existing
781  * systems (with a traditional BIOS) as well as on EFI systems.
782  */
783 /*
784  * setup_arch - architecture-specific boot-time initializations
785  *
786  * Note: On x86_64, fixmaps are ready for use even before this is called.
787  */
788 
789 void __init setup_arch(char **cmdline_p)
790 {
791 	/*
792 	 * Reserve the memory occupied by the kernel between _text and
793 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
794 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
795 	 * separate memblock_reserve() or they will be discarded.
796 	 */
797 	memblock_reserve(__pa_symbol(_text),
798 			 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
799 
800 	/*
801 	 * Make sure page 0 is always reserved because on systems with
802 	 * L1TF its contents can be leaked to user processes.
803 	 */
804 	memblock_reserve(0, PAGE_SIZE);
805 
806 	early_reserve_initrd();
807 
808 	/*
809 	 * At this point everything still needed from the boot loader
810 	 * or BIOS or kernel text should be early reserved or marked not
811 	 * RAM in e820. All other memory is free game.
812 	 */
813 
814 #ifdef CONFIG_X86_32
815 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
816 
817 	/*
818 	 * copy kernel address range established so far and switch
819 	 * to the proper swapper page table
820 	 */
821 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
822 			initial_page_table + KERNEL_PGD_BOUNDARY,
823 			KERNEL_PGD_PTRS);
824 
825 	load_cr3(swapper_pg_dir);
826 	/*
827 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
828 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
829 	 * will not flush anything because the CPU quirk which clears
830 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
831 	 * load_cr3() above the TLB has been flushed already. The
832 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
833 	 * so proper operation is guaranteed.
834 	 */
835 	__flush_tlb_all();
836 #else
837 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
838 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
839 #endif
840 
841 	/*
842 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
843 	 * reserve_top(), so do this before touching the ioremap area.
844 	 */
845 	olpc_ofw_detect();
846 
847 	idt_setup_early_traps();
848 	early_cpu_init();
849 	arch_init_ideal_nops();
850 	jump_label_init();
851 	early_ioremap_init();
852 
853 	setup_olpc_ofw_pgd();
854 
855 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
856 	screen_info = boot_params.screen_info;
857 	edid_info = boot_params.edid_info;
858 #ifdef CONFIG_X86_32
859 	apm_info.bios = boot_params.apm_bios_info;
860 	ist_info = boot_params.ist_info;
861 #endif
862 	saved_video_mode = boot_params.hdr.vid_mode;
863 	bootloader_type = boot_params.hdr.type_of_loader;
864 	if ((bootloader_type >> 4) == 0xe) {
865 		bootloader_type &= 0xf;
866 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
867 	}
868 	bootloader_version  = bootloader_type & 0xf;
869 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
870 
871 #ifdef CONFIG_BLK_DEV_RAM
872 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
873 	rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
874 	rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
875 #endif
876 #ifdef CONFIG_EFI
877 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
878 		     EFI32_LOADER_SIGNATURE, 4)) {
879 		set_bit(EFI_BOOT, &efi.flags);
880 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
881 		     EFI64_LOADER_SIGNATURE, 4)) {
882 		set_bit(EFI_BOOT, &efi.flags);
883 		set_bit(EFI_64BIT, &efi.flags);
884 	}
885 #endif
886 
887 	x86_init.oem.arch_setup();
888 
889 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
890 	e820__memory_setup();
891 	parse_setup_data();
892 
893 	copy_edd();
894 
895 	if (!boot_params.hdr.root_flags)
896 		root_mountflags &= ~MS_RDONLY;
897 	init_mm.start_code = (unsigned long) _text;
898 	init_mm.end_code = (unsigned long) _etext;
899 	init_mm.end_data = (unsigned long) _edata;
900 	init_mm.brk = _brk_end;
901 
902 	code_resource.start = __pa_symbol(_text);
903 	code_resource.end = __pa_symbol(_etext)-1;
904 	rodata_resource.start = __pa_symbol(__start_rodata);
905 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
906 	data_resource.start = __pa_symbol(_sdata);
907 	data_resource.end = __pa_symbol(_edata)-1;
908 	bss_resource.start = __pa_symbol(__bss_start);
909 	bss_resource.end = __pa_symbol(__bss_stop)-1;
910 
911 #ifdef CONFIG_CMDLINE_BOOL
912 #ifdef CONFIG_CMDLINE_OVERRIDE
913 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
914 #else
915 	if (builtin_cmdline[0]) {
916 		/* append boot loader cmdline to builtin */
917 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
918 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
919 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
920 	}
921 #endif
922 #endif
923 
924 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
925 	*cmdline_p = command_line;
926 
927 	/*
928 	 * x86_configure_nx() is called before parse_early_param() to detect
929 	 * whether hardware doesn't support NX (so that the early EHCI debug
930 	 * console setup can safely call set_fixmap()). It may then be called
931 	 * again from within noexec_setup() during parsing early parameters
932 	 * to honor the respective command line option.
933 	 */
934 	x86_configure_nx();
935 
936 	parse_early_param();
937 
938 	if (efi_enabled(EFI_BOOT))
939 		efi_memblock_x86_reserve_range();
940 #ifdef CONFIG_MEMORY_HOTPLUG
941 	/*
942 	 * Memory used by the kernel cannot be hot-removed because Linux
943 	 * cannot migrate the kernel pages. When memory hotplug is
944 	 * enabled, we should prevent memblock from allocating memory
945 	 * for the kernel.
946 	 *
947 	 * ACPI SRAT records all hotpluggable memory ranges. But before
948 	 * SRAT is parsed, we don't know about it.
949 	 *
950 	 * The kernel image is loaded into memory at very early time. We
951 	 * cannot prevent this anyway. So on NUMA system, we set any
952 	 * node the kernel resides in as un-hotpluggable.
953 	 *
954 	 * Since on modern servers, one node could have double-digit
955 	 * gigabytes memory, we can assume the memory around the kernel
956 	 * image is also un-hotpluggable. So before SRAT is parsed, just
957 	 * allocate memory near the kernel image to try the best to keep
958 	 * the kernel away from hotpluggable memory.
959 	 */
960 	if (movable_node_is_enabled())
961 		memblock_set_bottom_up(true);
962 #endif
963 
964 	x86_report_nx();
965 
966 	/* after early param, so could get panic from serial */
967 	memblock_x86_reserve_range_setup_data();
968 
969 	if (acpi_mps_check()) {
970 #ifdef CONFIG_X86_LOCAL_APIC
971 		disable_apic = 1;
972 #endif
973 		setup_clear_cpu_cap(X86_FEATURE_APIC);
974 	}
975 
976 	e820__reserve_setup_data();
977 	e820__finish_early_params();
978 
979 	if (efi_enabled(EFI_BOOT))
980 		efi_init();
981 
982 	dmi_setup();
983 
984 	/*
985 	 * VMware detection requires dmi to be available, so this
986 	 * needs to be done after dmi_setup(), for the boot CPU.
987 	 */
988 	init_hypervisor_platform();
989 
990 	tsc_early_init();
991 	x86_init.resources.probe_roms();
992 
993 	/* after parse_early_param, so could debug it */
994 	insert_resource(&iomem_resource, &code_resource);
995 	insert_resource(&iomem_resource, &rodata_resource);
996 	insert_resource(&iomem_resource, &data_resource);
997 	insert_resource(&iomem_resource, &bss_resource);
998 
999 	e820_add_kernel_range();
1000 	trim_bios_range();
1001 #ifdef CONFIG_X86_32
1002 	if (ppro_with_ram_bug()) {
1003 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1004 				  E820_TYPE_RESERVED);
1005 		e820__update_table(e820_table);
1006 		printk(KERN_INFO "fixed physical RAM map:\n");
1007 		e820__print_table("bad_ppro");
1008 	}
1009 #else
1010 	early_gart_iommu_check();
1011 #endif
1012 
1013 	/*
1014 	 * partially used pages are not usable - thus
1015 	 * we are rounding upwards:
1016 	 */
1017 	max_pfn = e820__end_of_ram_pfn();
1018 
1019 	/* update e820 for memory not covered by WB MTRRs */
1020 	mtrr_bp_init();
1021 	if (mtrr_trim_uncached_memory(max_pfn))
1022 		max_pfn = e820__end_of_ram_pfn();
1023 
1024 	max_possible_pfn = max_pfn;
1025 
1026 	/*
1027 	 * This call is required when the CPU does not support PAT. If
1028 	 * mtrr_bp_init() invoked it already via pat_init() the call has no
1029 	 * effect.
1030 	 */
1031 	init_cache_modes();
1032 
1033 	/*
1034 	 * Define random base addresses for memory sections after max_pfn is
1035 	 * defined and before each memory section base is used.
1036 	 */
1037 	kernel_randomize_memory();
1038 
1039 #ifdef CONFIG_X86_32
1040 	/* max_low_pfn get updated here */
1041 	find_low_pfn_range();
1042 #else
1043 	check_x2apic();
1044 
1045 	/* How many end-of-memory variables you have, grandma! */
1046 	/* need this before calling reserve_initrd */
1047 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1048 		max_low_pfn = e820__end_of_low_ram_pfn();
1049 	else
1050 		max_low_pfn = max_pfn;
1051 
1052 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1053 #endif
1054 
1055 	/*
1056 	 * Find and reserve possible boot-time SMP configuration:
1057 	 */
1058 	find_smp_config();
1059 
1060 	reserve_ibft_region();
1061 
1062 	early_alloc_pgt_buf();
1063 
1064 	/*
1065 	 * Need to conclude brk, before e820__memblock_setup()
1066 	 *  it could use memblock_find_in_range, could overlap with
1067 	 *  brk area.
1068 	 */
1069 	reserve_brk();
1070 
1071 	cleanup_highmap();
1072 
1073 	memblock_set_current_limit(ISA_END_ADDRESS);
1074 	e820__memblock_setup();
1075 
1076 	reserve_bios_regions();
1077 
1078 	efi_fake_memmap();
1079 	efi_find_mirror();
1080 	efi_esrt_init();
1081 
1082 	/*
1083 	 * The EFI specification says that boot service code won't be
1084 	 * called after ExitBootServices(). This is, in fact, a lie.
1085 	 */
1086 	efi_reserve_boot_services();
1087 
1088 	/* preallocate 4k for mptable mpc */
1089 	e820__memblock_alloc_reserved_mpc_new();
1090 
1091 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1092 	setup_bios_corruption_check();
1093 #endif
1094 
1095 #ifdef CONFIG_X86_32
1096 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1097 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1098 #endif
1099 
1100 	reserve_real_mode();
1101 
1102 	trim_platform_memory_ranges();
1103 	trim_low_memory_range();
1104 
1105 	init_mem_mapping();
1106 
1107 	idt_setup_early_pf();
1108 
1109 	/*
1110 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1111 	 * with the current CR4 value.  This may not be necessary, but
1112 	 * auditing all the early-boot CR4 manipulation would be needed to
1113 	 * rule it out.
1114 	 *
1115 	 * Mask off features that don't work outside long mode (just
1116 	 * PCIDE for now).
1117 	 */
1118 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1119 
1120 	memblock_set_current_limit(get_max_mapped());
1121 
1122 	/*
1123 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1124 	 */
1125 
1126 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1127 	if (init_ohci1394_dma_early)
1128 		init_ohci1394_dma_on_all_controllers();
1129 #endif
1130 	/* Allocate bigger log buffer */
1131 	setup_log_buf(1);
1132 
1133 	if (efi_enabled(EFI_BOOT)) {
1134 		switch (boot_params.secure_boot) {
1135 		case efi_secureboot_mode_disabled:
1136 			pr_info("Secure boot disabled\n");
1137 			break;
1138 		case efi_secureboot_mode_enabled:
1139 			pr_info("Secure boot enabled\n");
1140 			break;
1141 		default:
1142 			pr_info("Secure boot could not be determined\n");
1143 			break;
1144 		}
1145 	}
1146 
1147 	reserve_initrd();
1148 
1149 	acpi_table_upgrade();
1150 
1151 	vsmp_init();
1152 
1153 	io_delay_init();
1154 
1155 	early_platform_quirks();
1156 
1157 	/*
1158 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1159 	 */
1160 	acpi_boot_table_init();
1161 
1162 	early_acpi_boot_init();
1163 
1164 	initmem_init();
1165 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1166 
1167 	if (boot_cpu_has(X86_FEATURE_GBPAGES))
1168 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1169 
1170 	/*
1171 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1172 	 * won't consume hotpluggable memory.
1173 	 */
1174 	reserve_crashkernel();
1175 
1176 	memblock_find_dma_reserve();
1177 
1178 	if (!early_xdbc_setup_hardware())
1179 		early_xdbc_register_console();
1180 
1181 	x86_init.paging.pagetable_init();
1182 
1183 	kasan_init();
1184 
1185 	/*
1186 	 * Sync back kernel address range.
1187 	 *
1188 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1189 	 * this call?
1190 	 */
1191 	sync_initial_page_table();
1192 
1193 	tboot_probe();
1194 
1195 	map_vsyscall();
1196 
1197 	generic_apic_probe();
1198 
1199 	early_quirks();
1200 
1201 	/*
1202 	 * Read APIC and some other early information from ACPI tables.
1203 	 */
1204 	acpi_boot_init();
1205 	sfi_init();
1206 	x86_dtb_init();
1207 
1208 	/*
1209 	 * get boot-time SMP configuration:
1210 	 */
1211 	get_smp_config();
1212 
1213 	/*
1214 	 * Systems w/o ACPI and mptables might not have it mapped the local
1215 	 * APIC yet, but prefill_possible_map() might need to access it.
1216 	 */
1217 	init_apic_mappings();
1218 
1219 	prefill_possible_map();
1220 
1221 	init_cpu_to_node();
1222 
1223 	io_apic_init_mappings();
1224 
1225 	x86_init.hyper.guest_late_init();
1226 
1227 	e820__reserve_resources();
1228 	e820__register_nosave_regions(max_pfn);
1229 
1230 	x86_init.resources.reserve_resources();
1231 
1232 	e820__setup_pci_gap();
1233 
1234 #ifdef CONFIG_VT
1235 #if defined(CONFIG_VGA_CONSOLE)
1236 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1237 		conswitchp = &vga_con;
1238 #endif
1239 #endif
1240 	x86_init.oem.banner();
1241 
1242 	x86_init.timers.wallclock_init();
1243 
1244 	mcheck_init();
1245 
1246 	register_refined_jiffies(CLOCK_TICK_RATE);
1247 
1248 #ifdef CONFIG_EFI
1249 	if (efi_enabled(EFI_BOOT))
1250 		efi_apply_memmap_quirks();
1251 #endif
1252 
1253 	unwind_init();
1254 }
1255 
1256 #ifdef CONFIG_X86_32
1257 
1258 static struct resource video_ram_resource = {
1259 	.name	= "Video RAM area",
1260 	.start	= 0xa0000,
1261 	.end	= 0xbffff,
1262 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1263 };
1264 
1265 void __init i386_reserve_resources(void)
1266 {
1267 	request_resource(&iomem_resource, &video_ram_resource);
1268 	reserve_standard_io_resources();
1269 }
1270 
1271 #endif /* CONFIG_X86_32 */
1272 
1273 static struct notifier_block kernel_offset_notifier = {
1274 	.notifier_call = dump_kernel_offset
1275 };
1276 
1277 static int __init register_kernel_offset_dumper(void)
1278 {
1279 	atomic_notifier_chain_register(&panic_notifier_list,
1280 					&kernel_offset_notifier);
1281 	return 0;
1282 }
1283 __initcall(register_kernel_offset_dumper);
1284