xref: /linux/arch/x86/kernel/setup.c (revision c01044cc819160323f3ca4acd44fca487c4432e6)
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 #include <linux/static_call.h>
23 
24 #include <uapi/linux/mount.h>
25 
26 #include <xen/xen.h>
27 
28 #include <asm/apic.h>
29 #include <asm/numa.h>
30 #include <asm/bios_ebda.h>
31 #include <asm/bugs.h>
32 #include <asm/cpu.h>
33 #include <asm/efi.h>
34 #include <asm/gart.h>
35 #include <asm/hypervisor.h>
36 #include <asm/io_apic.h>
37 #include <asm/kasan.h>
38 #include <asm/kaslr.h>
39 #include <asm/mce.h>
40 #include <asm/mtrr.h>
41 #include <asm/realmode.h>
42 #include <asm/olpc_ofw.h>
43 #include <asm/pci-direct.h>
44 #include <asm/prom.h>
45 #include <asm/proto.h>
46 #include <asm/unwind.h>
47 #include <asm/vsyscall.h>
48 #include <linux/vmalloc.h>
49 
50 /*
51  * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
52  * max_pfn_mapped:     highest directly mapped pfn > 4 GB
53  *
54  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
55  * represented by pfn_mapped[].
56  */
57 unsigned long max_low_pfn_mapped;
58 unsigned long max_pfn_mapped;
59 
60 #ifdef CONFIG_DMI
61 RESERVE_BRK(dmi_alloc, 65536);
62 #endif
63 
64 
65 /*
66  * Range of the BSS area. The size of the BSS area is determined
67  * at link time, with RESERVE_BRK*() facility reserving additional
68  * chunks.
69  */
70 unsigned long _brk_start = (unsigned long)__brk_base;
71 unsigned long _brk_end   = (unsigned long)__brk_base;
72 
73 struct boot_params boot_params;
74 
75 /*
76  * These are the four main kernel memory regions, we put them into
77  * the resource tree so that kdump tools and other debugging tools
78  * recover it:
79  */
80 
81 static struct resource rodata_resource = {
82 	.name	= "Kernel rodata",
83 	.start	= 0,
84 	.end	= 0,
85 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
86 };
87 
88 static struct resource data_resource = {
89 	.name	= "Kernel data",
90 	.start	= 0,
91 	.end	= 0,
92 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
93 };
94 
95 static struct resource code_resource = {
96 	.name	= "Kernel code",
97 	.start	= 0,
98 	.end	= 0,
99 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
100 };
101 
102 static struct resource bss_resource = {
103 	.name	= "Kernel bss",
104 	.start	= 0,
105 	.end	= 0,
106 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
107 };
108 
109 
110 #ifdef CONFIG_X86_32
111 /* CPU data as detected by the assembly code in head_32.S */
112 struct cpuinfo_x86 new_cpu_data;
113 
114 /* Common CPU data for all CPUs */
115 struct cpuinfo_x86 boot_cpu_data __read_mostly;
116 EXPORT_SYMBOL(boot_cpu_data);
117 
118 unsigned int def_to_bigsmp;
119 
120 /* For MCA, but anyone else can use it if they want */
121 unsigned int machine_id;
122 unsigned int machine_submodel_id;
123 unsigned int BIOS_revision;
124 
125 struct apm_info apm_info;
126 EXPORT_SYMBOL(apm_info);
127 
128 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
129 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
130 struct ist_info ist_info;
131 EXPORT_SYMBOL(ist_info);
132 #else
133 struct ist_info ist_info;
134 #endif
135 
136 #else
137 struct cpuinfo_x86 boot_cpu_data __read_mostly;
138 EXPORT_SYMBOL(boot_cpu_data);
139 #endif
140 
141 
142 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
143 __visible unsigned long mmu_cr4_features __ro_after_init;
144 #else
145 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
146 #endif
147 
148 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
149 int bootloader_type, bootloader_version;
150 
151 /*
152  * Setup options
153  */
154 struct screen_info screen_info;
155 EXPORT_SYMBOL(screen_info);
156 struct edid_info edid_info;
157 EXPORT_SYMBOL_GPL(edid_info);
158 
159 extern int root_mountflags;
160 
161 unsigned long saved_video_mode;
162 
163 #define RAMDISK_IMAGE_START_MASK	0x07FF
164 #define RAMDISK_PROMPT_FLAG		0x8000
165 #define RAMDISK_LOAD_FLAG		0x4000
166 
167 static char __initdata command_line[COMMAND_LINE_SIZE];
168 #ifdef CONFIG_CMDLINE_BOOL
169 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
170 #endif
171 
172 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
173 struct edd edd;
174 #ifdef CONFIG_EDD_MODULE
175 EXPORT_SYMBOL(edd);
176 #endif
177 /**
178  * copy_edd() - Copy the BIOS EDD information
179  *              from boot_params into a safe place.
180  *
181  */
182 static inline void __init copy_edd(void)
183 {
184      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
185 	    sizeof(edd.mbr_signature));
186      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
187      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
188      edd.edd_info_nr = boot_params.eddbuf_entries;
189 }
190 #else
191 static inline void __init copy_edd(void)
192 {
193 }
194 #endif
195 
196 void * __init extend_brk(size_t size, size_t align)
197 {
198 	size_t mask = align - 1;
199 	void *ret;
200 
201 	BUG_ON(_brk_start == 0);
202 	BUG_ON(align & mask);
203 
204 	_brk_end = (_brk_end + mask) & ~mask;
205 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
206 
207 	ret = (void *)_brk_end;
208 	_brk_end += size;
209 
210 	memset(ret, 0, size);
211 
212 	return ret;
213 }
214 
215 #ifdef CONFIG_X86_32
216 static void __init cleanup_highmap(void)
217 {
218 }
219 #endif
220 
221 static void __init reserve_brk(void)
222 {
223 	if (_brk_end > _brk_start)
224 		memblock_reserve(__pa_symbol(_brk_start),
225 				 _brk_end - _brk_start);
226 
227 	/* Mark brk area as locked down and no longer taking any
228 	   new allocations */
229 	_brk_start = 0;
230 }
231 
232 u64 relocated_ramdisk;
233 
234 #ifdef CONFIG_BLK_DEV_INITRD
235 
236 static u64 __init get_ramdisk_image(void)
237 {
238 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
239 
240 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
241 
242 	if (ramdisk_image == 0)
243 		ramdisk_image = phys_initrd_start;
244 
245 	return ramdisk_image;
246 }
247 static u64 __init get_ramdisk_size(void)
248 {
249 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
250 
251 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
252 
253 	if (ramdisk_size == 0)
254 		ramdisk_size = phys_initrd_size;
255 
256 	return ramdisk_size;
257 }
258 
259 static void __init relocate_initrd(void)
260 {
261 	/* Assume only end is not page aligned */
262 	u64 ramdisk_image = get_ramdisk_image();
263 	u64 ramdisk_size  = get_ramdisk_size();
264 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
265 
266 	/* We need to move the initrd down into directly mapped mem */
267 	relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
268 						   area_size, PAGE_SIZE);
269 
270 	if (!relocated_ramdisk)
271 		panic("Cannot find place for new RAMDISK of size %lld\n",
272 		      ramdisk_size);
273 
274 	/* Note: this includes all the mem currently occupied by
275 	   the initrd, we rely on that fact to keep the data intact. */
276 	memblock_reserve(relocated_ramdisk, area_size);
277 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
278 	initrd_end   = initrd_start + ramdisk_size;
279 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
280 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
281 
282 	copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
283 
284 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
285 		" [mem %#010llx-%#010llx]\n",
286 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
287 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
288 }
289 
290 static void __init early_reserve_initrd(void)
291 {
292 	/* Assume only end is not page aligned */
293 	u64 ramdisk_image = get_ramdisk_image();
294 	u64 ramdisk_size  = get_ramdisk_size();
295 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
296 
297 	if (!boot_params.hdr.type_of_loader ||
298 	    !ramdisk_image || !ramdisk_size)
299 		return;		/* No initrd provided by bootloader */
300 
301 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
302 }
303 static void __init reserve_initrd(void)
304 {
305 	/* Assume only end is not page aligned */
306 	u64 ramdisk_image = get_ramdisk_image();
307 	u64 ramdisk_size  = get_ramdisk_size();
308 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
309 	u64 mapped_size;
310 
311 	if (!boot_params.hdr.type_of_loader ||
312 	    !ramdisk_image || !ramdisk_size)
313 		return;		/* No initrd provided by bootloader */
314 
315 	initrd_start = 0;
316 
317 	mapped_size = memblock_mem_size(max_pfn_mapped);
318 	if (ramdisk_size >= (mapped_size>>1))
319 		panic("initrd too large to handle, "
320 		       "disabling initrd (%lld needed, %lld available)\n",
321 		       ramdisk_size, mapped_size>>1);
322 
323 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
324 			ramdisk_end - 1);
325 
326 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
327 				PFN_DOWN(ramdisk_end))) {
328 		/* All are mapped, easy case */
329 		initrd_start = ramdisk_image + PAGE_OFFSET;
330 		initrd_end = initrd_start + ramdisk_size;
331 		return;
332 	}
333 
334 	relocate_initrd();
335 
336 	memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
337 }
338 
339 #else
340 static void __init early_reserve_initrd(void)
341 {
342 }
343 static void __init reserve_initrd(void)
344 {
345 }
346 #endif /* CONFIG_BLK_DEV_INITRD */
347 
348 static void __init parse_setup_data(void)
349 {
350 	struct setup_data *data;
351 	u64 pa_data, pa_next;
352 
353 	pa_data = boot_params.hdr.setup_data;
354 	while (pa_data) {
355 		u32 data_len, data_type;
356 
357 		data = early_memremap(pa_data, sizeof(*data));
358 		data_len = data->len + sizeof(struct setup_data);
359 		data_type = data->type;
360 		pa_next = data->next;
361 		early_memunmap(data, sizeof(*data));
362 
363 		switch (data_type) {
364 		case SETUP_E820_EXT:
365 			e820__memory_setup_extended(pa_data, data_len);
366 			break;
367 		case SETUP_DTB:
368 			add_dtb(pa_data);
369 			break;
370 		case SETUP_EFI:
371 			parse_efi_setup(pa_data, data_len);
372 			break;
373 		default:
374 			break;
375 		}
376 		pa_data = pa_next;
377 	}
378 }
379 
380 static void __init memblock_x86_reserve_range_setup_data(void)
381 {
382 	struct setup_data *data;
383 	u64 pa_data;
384 
385 	pa_data = boot_params.hdr.setup_data;
386 	while (pa_data) {
387 		data = early_memremap(pa_data, sizeof(*data));
388 		memblock_reserve(pa_data, sizeof(*data) + data->len);
389 
390 		if (data->type == SETUP_INDIRECT &&
391 		    ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
392 			memblock_reserve(((struct setup_indirect *)data->data)->addr,
393 					 ((struct setup_indirect *)data->data)->len);
394 
395 		pa_data = data->next;
396 		early_memunmap(data, sizeof(*data));
397 	}
398 }
399 
400 /*
401  * --------- Crashkernel reservation ------------------------------
402  */
403 
404 #ifdef CONFIG_KEXEC_CORE
405 
406 /* 16M alignment for crash kernel regions */
407 #define CRASH_ALIGN		SZ_16M
408 
409 /*
410  * Keep the crash kernel below this limit.
411  *
412  * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
413  * due to mapping restrictions.
414  *
415  * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
416  * the upper limit of system RAM in 4-level paging mode. Since the kdump
417  * jump could be from 5-level paging to 4-level paging, the jump will fail if
418  * the kernel is put above 64 TB, and during the 1st kernel bootup there's
419  * no good way to detect the paging mode of the target kernel which will be
420  * loaded for dumping.
421  */
422 #ifdef CONFIG_X86_32
423 # define CRASH_ADDR_LOW_MAX	SZ_512M
424 # define CRASH_ADDR_HIGH_MAX	SZ_512M
425 #else
426 # define CRASH_ADDR_LOW_MAX	SZ_4G
427 # define CRASH_ADDR_HIGH_MAX	SZ_64T
428 #endif
429 
430 static int __init reserve_crashkernel_low(void)
431 {
432 #ifdef CONFIG_X86_64
433 	unsigned long long base, low_base = 0, low_size = 0;
434 	unsigned long total_low_mem;
435 	int ret;
436 
437 	total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
438 
439 	/* crashkernel=Y,low */
440 	ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
441 	if (ret) {
442 		/*
443 		 * two parts from kernel/dma/swiotlb.c:
444 		 * -swiotlb size: user-specified with swiotlb= or default.
445 		 *
446 		 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
447 		 * to 8M for other buffers that may need to stay low too. Also
448 		 * make sure we allocate enough extra low memory so that we
449 		 * don't run out of DMA buffers for 32-bit devices.
450 		 */
451 		low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
452 	} else {
453 		/* passed with crashkernel=0,low ? */
454 		if (!low_size)
455 			return 0;
456 	}
457 
458 	low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN);
459 	if (!low_base) {
460 		pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
461 		       (unsigned long)(low_size >> 20));
462 		return -ENOMEM;
463 	}
464 
465 	ret = memblock_reserve(low_base, low_size);
466 	if (ret) {
467 		pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
468 		return ret;
469 	}
470 
471 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
472 		(unsigned long)(low_size >> 20),
473 		(unsigned long)(low_base >> 20),
474 		(unsigned long)(total_low_mem >> 20));
475 
476 	crashk_low_res.start = low_base;
477 	crashk_low_res.end   = low_base + low_size - 1;
478 	insert_resource(&iomem_resource, &crashk_low_res);
479 #endif
480 	return 0;
481 }
482 
483 static void __init reserve_crashkernel(void)
484 {
485 	unsigned long long crash_size, crash_base, total_mem;
486 	bool high = false;
487 	int ret;
488 
489 	total_mem = memblock_phys_mem_size();
490 
491 	/* crashkernel=XM */
492 	ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
493 	if (ret != 0 || crash_size <= 0) {
494 		/* crashkernel=X,high */
495 		ret = parse_crashkernel_high(boot_command_line, total_mem,
496 					     &crash_size, &crash_base);
497 		if (ret != 0 || crash_size <= 0)
498 			return;
499 		high = true;
500 	}
501 
502 	if (xen_pv_domain()) {
503 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
504 		return;
505 	}
506 
507 	/* 0 means: find the address automatically */
508 	if (!crash_base) {
509 		/*
510 		 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
511 		 * crashkernel=x,high reserves memory over 4G, also allocates
512 		 * 256M extra low memory for DMA buffers and swiotlb.
513 		 * But the extra memory is not required for all machines.
514 		 * So try low memory first and fall back to high memory
515 		 * unless "crashkernel=size[KMG],high" is specified.
516 		 */
517 		if (!high)
518 			crash_base = memblock_find_in_range(CRASH_ALIGN,
519 						CRASH_ADDR_LOW_MAX,
520 						crash_size, CRASH_ALIGN);
521 		if (!crash_base)
522 			crash_base = memblock_find_in_range(CRASH_ALIGN,
523 						CRASH_ADDR_HIGH_MAX,
524 						crash_size, CRASH_ALIGN);
525 		if (!crash_base) {
526 			pr_info("crashkernel reservation failed - No suitable area found.\n");
527 			return;
528 		}
529 	} else {
530 		unsigned long long start;
531 
532 		start = memblock_find_in_range(crash_base,
533 					       crash_base + crash_size,
534 					       crash_size, 1 << 20);
535 		if (start != crash_base) {
536 			pr_info("crashkernel reservation failed - memory is in use.\n");
537 			return;
538 		}
539 	}
540 	ret = memblock_reserve(crash_base, crash_size);
541 	if (ret) {
542 		pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
543 		return;
544 	}
545 
546 	if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
547 		memblock_free(crash_base, crash_size);
548 		return;
549 	}
550 
551 	pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
552 		(unsigned long)(crash_size >> 20),
553 		(unsigned long)(crash_base >> 20),
554 		(unsigned long)(total_mem >> 20));
555 
556 	crashk_res.start = crash_base;
557 	crashk_res.end   = crash_base + crash_size - 1;
558 	insert_resource(&iomem_resource, &crashk_res);
559 }
560 #else
561 static void __init reserve_crashkernel(void)
562 {
563 }
564 #endif
565 
566 static struct resource standard_io_resources[] = {
567 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
568 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
569 	{ .name = "pic1", .start = 0x20, .end = 0x21,
570 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
571 	{ .name = "timer0", .start = 0x40, .end = 0x43,
572 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
573 	{ .name = "timer1", .start = 0x50, .end = 0x53,
574 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
575 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
576 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
577 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
578 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
579 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
580 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
581 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
582 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
583 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
584 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
585 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
586 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
587 };
588 
589 void __init reserve_standard_io_resources(void)
590 {
591 	int i;
592 
593 	/* request I/O space for devices used on all i[345]86 PCs */
594 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
595 		request_resource(&ioport_resource, &standard_io_resources[i]);
596 
597 }
598 
599 static __init void reserve_ibft_region(void)
600 {
601 	unsigned long addr, size = 0;
602 
603 	addr = find_ibft_region(&size);
604 
605 	if (size)
606 		memblock_reserve(addr, size);
607 }
608 
609 static bool __init snb_gfx_workaround_needed(void)
610 {
611 #ifdef CONFIG_PCI
612 	int i;
613 	u16 vendor, devid;
614 	static const __initconst u16 snb_ids[] = {
615 		0x0102,
616 		0x0112,
617 		0x0122,
618 		0x0106,
619 		0x0116,
620 		0x0126,
621 		0x010a,
622 	};
623 
624 	/* Assume no if something weird is going on with PCI */
625 	if (!early_pci_allowed())
626 		return false;
627 
628 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
629 	if (vendor != 0x8086)
630 		return false;
631 
632 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
633 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
634 		if (devid == snb_ids[i])
635 			return true;
636 #endif
637 
638 	return false;
639 }
640 
641 /*
642  * Sandy Bridge graphics has trouble with certain ranges, exclude
643  * them from allocation.
644  */
645 static void __init trim_snb_memory(void)
646 {
647 	static const __initconst unsigned long bad_pages[] = {
648 		0x20050000,
649 		0x20110000,
650 		0x20130000,
651 		0x20138000,
652 		0x40004000,
653 	};
654 	int i;
655 
656 	if (!snb_gfx_workaround_needed())
657 		return;
658 
659 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
660 
661 	/*
662 	 * Reserve all memory below the 1 MB mark that has not
663 	 * already been reserved.
664 	 */
665 	memblock_reserve(0, 1<<20);
666 
667 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
668 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
669 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
670 			       bad_pages[i]);
671 	}
672 }
673 
674 /*
675  * Here we put platform-specific memory range workarounds, i.e.
676  * memory known to be corrupt or otherwise in need to be reserved on
677  * specific platforms.
678  *
679  * If this gets used more widely it could use a real dispatch mechanism.
680  */
681 static void __init trim_platform_memory_ranges(void)
682 {
683 	trim_snb_memory();
684 }
685 
686 static void __init trim_bios_range(void)
687 {
688 	/*
689 	 * A special case is the first 4Kb of memory;
690 	 * This is a BIOS owned area, not kernel ram, but generally
691 	 * not listed as such in the E820 table.
692 	 *
693 	 * This typically reserves additional memory (64KiB by default)
694 	 * since some BIOSes are known to corrupt low memory.  See the
695 	 * Kconfig help text for X86_RESERVE_LOW.
696 	 */
697 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
698 
699 	/*
700 	 * special case: Some BIOSes report the PC BIOS
701 	 * area (640Kb -> 1Mb) as RAM even though it is not.
702 	 * take them out.
703 	 */
704 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
705 
706 	e820__update_table(e820_table);
707 }
708 
709 /* called before trim_bios_range() to spare extra sanitize */
710 static void __init e820_add_kernel_range(void)
711 {
712 	u64 start = __pa_symbol(_text);
713 	u64 size = __pa_symbol(_end) - start;
714 
715 	/*
716 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
717 	 * attempt to fix it by adding the range. We may have a confused BIOS,
718 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
719 	 * exclude kernel range. If we really are running on top non-RAM,
720 	 * we will crash later anyways.
721 	 */
722 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
723 		return;
724 
725 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
726 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
727 	e820__range_add(start, size, E820_TYPE_RAM);
728 }
729 
730 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
731 
732 static int __init parse_reservelow(char *p)
733 {
734 	unsigned long long size;
735 
736 	if (!p)
737 		return -EINVAL;
738 
739 	size = memparse(p, &p);
740 
741 	if (size < 4096)
742 		size = 4096;
743 
744 	if (size > 640*1024)
745 		size = 640*1024;
746 
747 	reserve_low = size;
748 
749 	return 0;
750 }
751 
752 early_param("reservelow", parse_reservelow);
753 
754 static void __init trim_low_memory_range(void)
755 {
756 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
757 }
758 
759 /*
760  * Dump out kernel offset information on panic.
761  */
762 static int
763 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
764 {
765 	if (kaslr_enabled()) {
766 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
767 			 kaslr_offset(),
768 			 __START_KERNEL,
769 			 __START_KERNEL_map,
770 			 MODULES_VADDR-1);
771 	} else {
772 		pr_emerg("Kernel Offset: disabled\n");
773 	}
774 
775 	return 0;
776 }
777 
778 /*
779  * Determine if we were loaded by an EFI loader.  If so, then we have also been
780  * passed the efi memmap, systab, etc., so we should use these data structures
781  * for initialization.  Note, the efi init code path is determined by the
782  * global efi_enabled. This allows the same kernel image to be used on existing
783  * systems (with a traditional BIOS) as well as on EFI systems.
784  */
785 /*
786  * setup_arch - architecture-specific boot-time initializations
787  *
788  * Note: On x86_64, fixmaps are ready for use even before this is called.
789  */
790 
791 void __init setup_arch(char **cmdline_p)
792 {
793 	/*
794 	 * Reserve the memory occupied by the kernel between _text and
795 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
796 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
797 	 * separate memblock_reserve() or they will be discarded.
798 	 */
799 	memblock_reserve(__pa_symbol(_text),
800 			 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
801 
802 	/*
803 	 * Make sure page 0 is always reserved because on systems with
804 	 * L1TF its contents can be leaked to user processes.
805 	 */
806 	memblock_reserve(0, PAGE_SIZE);
807 
808 	early_reserve_initrd();
809 
810 	/*
811 	 * At this point everything still needed from the boot loader
812 	 * or BIOS or kernel text should be early reserved or marked not
813 	 * RAM in e820. All other memory is free game.
814 	 */
815 
816 #ifdef CONFIG_X86_32
817 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
818 
819 	/*
820 	 * copy kernel address range established so far and switch
821 	 * to the proper swapper page table
822 	 */
823 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
824 			initial_page_table + KERNEL_PGD_BOUNDARY,
825 			KERNEL_PGD_PTRS);
826 
827 	load_cr3(swapper_pg_dir);
828 	/*
829 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
830 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
831 	 * will not flush anything because the CPU quirk which clears
832 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
833 	 * load_cr3() above the TLB has been flushed already. The
834 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
835 	 * so proper operation is guaranteed.
836 	 */
837 	__flush_tlb_all();
838 #else
839 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
840 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
841 #endif
842 
843 	/*
844 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
845 	 * reserve_top(), so do this before touching the ioremap area.
846 	 */
847 	olpc_ofw_detect();
848 
849 	idt_setup_early_traps();
850 	early_cpu_init();
851 	arch_init_ideal_nops();
852 	jump_label_init();
853 	static_call_init();
854 	early_ioremap_init();
855 
856 	setup_olpc_ofw_pgd();
857 
858 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
859 	screen_info = boot_params.screen_info;
860 	edid_info = boot_params.edid_info;
861 #ifdef CONFIG_X86_32
862 	apm_info.bios = boot_params.apm_bios_info;
863 	ist_info = boot_params.ist_info;
864 #endif
865 	saved_video_mode = boot_params.hdr.vid_mode;
866 	bootloader_type = boot_params.hdr.type_of_loader;
867 	if ((bootloader_type >> 4) == 0xe) {
868 		bootloader_type &= 0xf;
869 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
870 	}
871 	bootloader_version  = bootloader_type & 0xf;
872 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
873 
874 #ifdef CONFIG_BLK_DEV_RAM
875 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
876 #endif
877 #ifdef CONFIG_EFI
878 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
879 		     EFI32_LOADER_SIGNATURE, 4)) {
880 		set_bit(EFI_BOOT, &efi.flags);
881 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
882 		     EFI64_LOADER_SIGNATURE, 4)) {
883 		set_bit(EFI_BOOT, &efi.flags);
884 		set_bit(EFI_64BIT, &efi.flags);
885 	}
886 #endif
887 
888 	x86_init.oem.arch_setup();
889 
890 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
891 	e820__memory_setup();
892 	parse_setup_data();
893 
894 	copy_edd();
895 
896 	if (!boot_params.hdr.root_flags)
897 		root_mountflags &= ~MS_RDONLY;
898 	init_mm.start_code = (unsigned long) _text;
899 	init_mm.end_code = (unsigned long) _etext;
900 	init_mm.end_data = (unsigned long) _edata;
901 	init_mm.brk = _brk_end;
902 
903 	code_resource.start = __pa_symbol(_text);
904 	code_resource.end = __pa_symbol(_etext)-1;
905 	rodata_resource.start = __pa_symbol(__start_rodata);
906 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
907 	data_resource.start = __pa_symbol(_sdata);
908 	data_resource.end = __pa_symbol(_edata)-1;
909 	bss_resource.start = __pa_symbol(__bss_start);
910 	bss_resource.end = __pa_symbol(__bss_stop)-1;
911 
912 #ifdef CONFIG_CMDLINE_BOOL
913 #ifdef CONFIG_CMDLINE_OVERRIDE
914 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
915 #else
916 	if (builtin_cmdline[0]) {
917 		/* append boot loader cmdline to builtin */
918 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
919 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
920 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
921 	}
922 #endif
923 #endif
924 
925 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
926 	*cmdline_p = command_line;
927 
928 	/*
929 	 * x86_configure_nx() is called before parse_early_param() to detect
930 	 * whether hardware doesn't support NX (so that the early EHCI debug
931 	 * console setup can safely call set_fixmap()). It may then be called
932 	 * again from within noexec_setup() during parsing early parameters
933 	 * to honor the respective command line option.
934 	 */
935 	x86_configure_nx();
936 
937 	parse_early_param();
938 
939 	if (efi_enabled(EFI_BOOT))
940 		efi_memblock_x86_reserve_range();
941 #ifdef CONFIG_MEMORY_HOTPLUG
942 	/*
943 	 * Memory used by the kernel cannot be hot-removed because Linux
944 	 * cannot migrate the kernel pages. When memory hotplug is
945 	 * enabled, we should prevent memblock from allocating memory
946 	 * for the kernel.
947 	 *
948 	 * ACPI SRAT records all hotpluggable memory ranges. But before
949 	 * SRAT is parsed, we don't know about it.
950 	 *
951 	 * The kernel image is loaded into memory at very early time. We
952 	 * cannot prevent this anyway. So on NUMA system, we set any
953 	 * node the kernel resides in as un-hotpluggable.
954 	 *
955 	 * Since on modern servers, one node could have double-digit
956 	 * gigabytes memory, we can assume the memory around the kernel
957 	 * image is also un-hotpluggable. So before SRAT is parsed, just
958 	 * allocate memory near the kernel image to try the best to keep
959 	 * the kernel away from hotpluggable memory.
960 	 */
961 	if (movable_node_is_enabled())
962 		memblock_set_bottom_up(true);
963 #endif
964 
965 	x86_report_nx();
966 
967 	/* after early param, so could get panic from serial */
968 	memblock_x86_reserve_range_setup_data();
969 
970 	if (acpi_mps_check()) {
971 #ifdef CONFIG_X86_LOCAL_APIC
972 		disable_apic = 1;
973 #endif
974 		setup_clear_cpu_cap(X86_FEATURE_APIC);
975 	}
976 
977 	e820__reserve_setup_data();
978 	e820__finish_early_params();
979 
980 	if (efi_enabled(EFI_BOOT))
981 		efi_init();
982 
983 	dmi_setup();
984 
985 	/*
986 	 * VMware detection requires dmi to be available, so this
987 	 * needs to be done after dmi_setup(), for the boot CPU.
988 	 */
989 	init_hypervisor_platform();
990 
991 	tsc_early_init();
992 	x86_init.resources.probe_roms();
993 
994 	/* after parse_early_param, so could debug it */
995 	insert_resource(&iomem_resource, &code_resource);
996 	insert_resource(&iomem_resource, &rodata_resource);
997 	insert_resource(&iomem_resource, &data_resource);
998 	insert_resource(&iomem_resource, &bss_resource);
999 
1000 	e820_add_kernel_range();
1001 	trim_bios_range();
1002 #ifdef CONFIG_X86_32
1003 	if (ppro_with_ram_bug()) {
1004 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1005 				  E820_TYPE_RESERVED);
1006 		e820__update_table(e820_table);
1007 		printk(KERN_INFO "fixed physical RAM map:\n");
1008 		e820__print_table("bad_ppro");
1009 	}
1010 #else
1011 	early_gart_iommu_check();
1012 #endif
1013 
1014 	/*
1015 	 * partially used pages are not usable - thus
1016 	 * we are rounding upwards:
1017 	 */
1018 	max_pfn = e820__end_of_ram_pfn();
1019 
1020 	/* update e820 for memory not covered by WB MTRRs */
1021 	mtrr_bp_init();
1022 	if (mtrr_trim_uncached_memory(max_pfn))
1023 		max_pfn = e820__end_of_ram_pfn();
1024 
1025 	max_possible_pfn = max_pfn;
1026 
1027 	/*
1028 	 * This call is required when the CPU does not support PAT. If
1029 	 * mtrr_bp_init() invoked it already via pat_init() the call has no
1030 	 * effect.
1031 	 */
1032 	init_cache_modes();
1033 
1034 	/*
1035 	 * Define random base addresses for memory sections after max_pfn is
1036 	 * defined and before each memory section base is used.
1037 	 */
1038 	kernel_randomize_memory();
1039 
1040 #ifdef CONFIG_X86_32
1041 	/* max_low_pfn get updated here */
1042 	find_low_pfn_range();
1043 #else
1044 	check_x2apic();
1045 
1046 	/* How many end-of-memory variables you have, grandma! */
1047 	/* need this before calling reserve_initrd */
1048 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1049 		max_low_pfn = e820__end_of_low_ram_pfn();
1050 	else
1051 		max_low_pfn = max_pfn;
1052 
1053 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1054 #endif
1055 
1056 	/*
1057 	 * Find and reserve possible boot-time SMP configuration:
1058 	 */
1059 	find_smp_config();
1060 
1061 	reserve_ibft_region();
1062 
1063 	early_alloc_pgt_buf();
1064 
1065 	/*
1066 	 * Need to conclude brk, before e820__memblock_setup()
1067 	 *  it could use memblock_find_in_range, could overlap with
1068 	 *  brk area.
1069 	 */
1070 	reserve_brk();
1071 
1072 	cleanup_highmap();
1073 
1074 	memblock_set_current_limit(ISA_END_ADDRESS);
1075 	e820__memblock_setup();
1076 
1077 	reserve_bios_regions();
1078 
1079 	efi_fake_memmap();
1080 	efi_find_mirror();
1081 	efi_esrt_init();
1082 	efi_mokvar_table_init();
1083 
1084 	/*
1085 	 * The EFI specification says that boot service code won't be
1086 	 * called after ExitBootServices(). This is, in fact, a lie.
1087 	 */
1088 	efi_reserve_boot_services();
1089 
1090 	/* preallocate 4k for mptable mpc */
1091 	e820__memblock_alloc_reserved_mpc_new();
1092 
1093 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1094 	setup_bios_corruption_check();
1095 #endif
1096 
1097 #ifdef CONFIG_X86_32
1098 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1099 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1100 #endif
1101 
1102 	reserve_real_mode();
1103 
1104 	trim_platform_memory_ranges();
1105 	trim_low_memory_range();
1106 
1107 	init_mem_mapping();
1108 
1109 	idt_setup_early_pf();
1110 
1111 	/*
1112 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1113 	 * with the current CR4 value.  This may not be necessary, but
1114 	 * auditing all the early-boot CR4 manipulation would be needed to
1115 	 * rule it out.
1116 	 *
1117 	 * Mask off features that don't work outside long mode (just
1118 	 * PCIDE for now).
1119 	 */
1120 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1121 
1122 	memblock_set_current_limit(get_max_mapped());
1123 
1124 	/*
1125 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1126 	 */
1127 
1128 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1129 	if (init_ohci1394_dma_early)
1130 		init_ohci1394_dma_on_all_controllers();
1131 #endif
1132 	/* Allocate bigger log buffer */
1133 	setup_log_buf(1);
1134 
1135 	if (efi_enabled(EFI_BOOT)) {
1136 		switch (boot_params.secure_boot) {
1137 		case efi_secureboot_mode_disabled:
1138 			pr_info("Secure boot disabled\n");
1139 			break;
1140 		case efi_secureboot_mode_enabled:
1141 			pr_info("Secure boot enabled\n");
1142 			break;
1143 		default:
1144 			pr_info("Secure boot could not be determined\n");
1145 			break;
1146 		}
1147 	}
1148 
1149 	reserve_initrd();
1150 
1151 	acpi_table_upgrade();
1152 
1153 	vsmp_init();
1154 
1155 	io_delay_init();
1156 
1157 	early_platform_quirks();
1158 
1159 	/*
1160 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1161 	 */
1162 	acpi_boot_table_init();
1163 
1164 	early_acpi_boot_init();
1165 
1166 	initmem_init();
1167 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1168 
1169 	if (boot_cpu_has(X86_FEATURE_GBPAGES))
1170 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1171 
1172 	/*
1173 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1174 	 * won't consume hotpluggable memory.
1175 	 */
1176 	reserve_crashkernel();
1177 
1178 	memblock_find_dma_reserve();
1179 
1180 	if (!early_xdbc_setup_hardware())
1181 		early_xdbc_register_console();
1182 
1183 	x86_init.paging.pagetable_init();
1184 
1185 	kasan_init();
1186 
1187 	/*
1188 	 * Sync back kernel address range.
1189 	 *
1190 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1191 	 * this call?
1192 	 */
1193 	sync_initial_page_table();
1194 
1195 	tboot_probe();
1196 
1197 	map_vsyscall();
1198 
1199 	generic_apic_probe();
1200 
1201 	early_quirks();
1202 
1203 	/*
1204 	 * Read APIC and some other early information from ACPI tables.
1205 	 */
1206 	acpi_boot_init();
1207 	sfi_init();
1208 	x86_dtb_init();
1209 
1210 	/*
1211 	 * get boot-time SMP configuration:
1212 	 */
1213 	get_smp_config();
1214 
1215 	/*
1216 	 * Systems w/o ACPI and mptables might not have it mapped the local
1217 	 * APIC yet, but prefill_possible_map() might need to access it.
1218 	 */
1219 	init_apic_mappings();
1220 
1221 	prefill_possible_map();
1222 
1223 	init_cpu_to_node();
1224 
1225 	io_apic_init_mappings();
1226 
1227 	x86_init.hyper.guest_late_init();
1228 
1229 	e820__reserve_resources();
1230 	e820__register_nosave_regions(max_pfn);
1231 
1232 	x86_init.resources.reserve_resources();
1233 
1234 	e820__setup_pci_gap();
1235 
1236 #ifdef CONFIG_VT
1237 #if defined(CONFIG_VGA_CONSOLE)
1238 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1239 		conswitchp = &vga_con;
1240 #endif
1241 #endif
1242 	x86_init.oem.banner();
1243 
1244 	x86_init.timers.wallclock_init();
1245 
1246 	mcheck_init();
1247 
1248 	register_refined_jiffies(CLOCK_TICK_RATE);
1249 
1250 #ifdef CONFIG_EFI
1251 	if (efi_enabled(EFI_BOOT))
1252 		efi_apply_memmap_quirks();
1253 #endif
1254 
1255 	unwind_init();
1256 }
1257 
1258 #ifdef CONFIG_X86_32
1259 
1260 static struct resource video_ram_resource = {
1261 	.name	= "Video RAM area",
1262 	.start	= 0xa0000,
1263 	.end	= 0xbffff,
1264 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1265 };
1266 
1267 void __init i386_reserve_resources(void)
1268 {
1269 	request_resource(&iomem_resource, &video_ram_resource);
1270 	reserve_standard_io_resources();
1271 }
1272 
1273 #endif /* CONFIG_X86_32 */
1274 
1275 static struct notifier_block kernel_offset_notifier = {
1276 	.notifier_call = dump_kernel_offset
1277 };
1278 
1279 static int __init register_kernel_offset_dumper(void)
1280 {
1281 	atomic_notifier_chain_register(&panic_notifier_list,
1282 					&kernel_offset_notifier);
1283 	return 0;
1284 }
1285 __initcall(register_kernel_offset_dumper);
1286