xref: /linux/arch/x86/kernel/setup.c (revision 06eaeaee5b780106c3578bfb9ff2babc19ccffb7)
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/acpi.h>
9 #include <linux/console.h>
10 #include <linux/cpu.h>
11 #include <linux/crash_dump.h>
12 #include <linux/dma-map-ops.h>
13 #include <linux/efi.h>
14 #include <linux/ima.h>
15 #include <linux/init_ohci1394_dma.h>
16 #include <linux/initrd.h>
17 #include <linux/iscsi_ibft.h>
18 #include <linux/memblock.h>
19 #include <linux/panic_notifier.h>
20 #include <linux/pci.h>
21 #include <linux/root_dev.h>
22 #include <linux/hugetlb.h>
23 #include <linux/tboot.h>
24 #include <linux/usb/xhci-dbgp.h>
25 #include <linux/static_call.h>
26 #include <linux/swiotlb.h>
27 #include <linux/random.h>
28 
29 #include <uapi/linux/mount.h>
30 
31 #include <xen/xen.h>
32 
33 #include <asm/apic.h>
34 #include <asm/efi.h>
35 #include <asm/numa.h>
36 #include <asm/bios_ebda.h>
37 #include <asm/bugs.h>
38 #include <asm/cacheinfo.h>
39 #include <asm/coco.h>
40 #include <asm/cpu.h>
41 #include <asm/efi.h>
42 #include <asm/gart.h>
43 #include <asm/hypervisor.h>
44 #include <asm/io_apic.h>
45 #include <asm/kasan.h>
46 #include <asm/kaslr.h>
47 #include <asm/mce.h>
48 #include <asm/memtype.h>
49 #include <asm/mtrr.h>
50 #include <asm/realmode.h>
51 #include <asm/olpc_ofw.h>
52 #include <asm/pci-direct.h>
53 #include <asm/prom.h>
54 #include <asm/proto.h>
55 #include <asm/thermal.h>
56 #include <asm/unwind.h>
57 #include <asm/vsyscall.h>
58 #include <linux/vmalloc.h>
59 
60 /*
61  * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
62  * max_pfn_mapped:     highest directly mapped pfn > 4 GB
63  *
64  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
65  * represented by pfn_mapped[].
66  */
67 unsigned long max_low_pfn_mapped;
68 unsigned long max_pfn_mapped;
69 
70 #ifdef CONFIG_DMI
71 RESERVE_BRK(dmi_alloc, 65536);
72 #endif
73 
74 
75 unsigned long _brk_start = (unsigned long)__brk_base;
76 unsigned long _brk_end   = (unsigned long)__brk_base;
77 
78 struct boot_params boot_params;
79 
80 /*
81  * These are the four main kernel memory regions, we put them into
82  * the resource tree so that kdump tools and other debugging tools
83  * recover it:
84  */
85 
86 static struct resource rodata_resource = {
87 	.name	= "Kernel rodata",
88 	.start	= 0,
89 	.end	= 0,
90 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
91 };
92 
93 static struct resource data_resource = {
94 	.name	= "Kernel data",
95 	.start	= 0,
96 	.end	= 0,
97 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
98 };
99 
100 static struct resource code_resource = {
101 	.name	= "Kernel code",
102 	.start	= 0,
103 	.end	= 0,
104 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
105 };
106 
107 static struct resource bss_resource = {
108 	.name	= "Kernel bss",
109 	.start	= 0,
110 	.end	= 0,
111 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
112 };
113 
114 
115 #ifdef CONFIG_X86_32
116 /* CPU data as detected by the assembly code in head_32.S */
117 struct cpuinfo_x86 new_cpu_data;
118 
119 struct apm_info apm_info;
120 EXPORT_SYMBOL(apm_info);
121 
122 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
123 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
124 struct ist_info ist_info;
125 EXPORT_SYMBOL(ist_info);
126 #else
127 struct ist_info ist_info;
128 #endif
129 
130 #endif
131 
132 struct cpuinfo_x86 boot_cpu_data __read_mostly;
133 EXPORT_SYMBOL(boot_cpu_data);
134 
135 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
136 __visible unsigned long mmu_cr4_features __ro_after_init;
137 #else
138 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
139 #endif
140 
141 #ifdef CONFIG_IMA
142 static phys_addr_t ima_kexec_buffer_phys;
143 static size_t ima_kexec_buffer_size;
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 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
168 bool builtin_cmdline_added __ro_after_init;
169 #endif
170 
171 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
172 struct edd edd;
173 #ifdef CONFIG_EDD_MODULE
174 EXPORT_SYMBOL(edd);
175 #endif
176 /**
177  * copy_edd() - Copy the BIOS EDD information
178  *              from boot_params into a safe place.
179  *
180  */
181 static inline void __init copy_edd(void)
182 {
183      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
184 	    sizeof(edd.mbr_signature));
185      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
186      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
187      edd.edd_info_nr = boot_params.eddbuf_entries;
188 }
189 #else
190 static inline void __init copy_edd(void)
191 {
192 }
193 #endif
194 
195 void * __init extend_brk(size_t size, size_t align)
196 {
197 	size_t mask = align - 1;
198 	void *ret;
199 
200 	BUG_ON(_brk_start == 0);
201 	BUG_ON(align & mask);
202 
203 	_brk_end = (_brk_end + mask) & ~mask;
204 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
205 
206 	ret = (void *)_brk_end;
207 	_brk_end += size;
208 
209 	memset(ret, 0, size);
210 
211 	return ret;
212 }
213 
214 #ifdef CONFIG_X86_32
215 static void __init cleanup_highmap(void)
216 {
217 }
218 #endif
219 
220 static void __init reserve_brk(void)
221 {
222 	if (_brk_end > _brk_start)
223 		memblock_reserve(__pa_symbol(_brk_start),
224 				 _brk_end - _brk_start);
225 
226 	/* Mark brk area as locked down and no longer taking any
227 	   new allocations */
228 	_brk_start = 0;
229 }
230 
231 #ifdef CONFIG_BLK_DEV_INITRD
232 
233 static u64 __init get_ramdisk_image(void)
234 {
235 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
236 
237 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
238 
239 	if (ramdisk_image == 0)
240 		ramdisk_image = phys_initrd_start;
241 
242 	return ramdisk_image;
243 }
244 static u64 __init get_ramdisk_size(void)
245 {
246 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
247 
248 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
249 
250 	if (ramdisk_size == 0)
251 		ramdisk_size = phys_initrd_size;
252 
253 	return ramdisk_size;
254 }
255 
256 static void __init relocate_initrd(void)
257 {
258 	/* Assume only end is not page aligned */
259 	u64 ramdisk_image = get_ramdisk_image();
260 	u64 ramdisk_size  = get_ramdisk_size();
261 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
262 	int ret = 0;
263 
264 	/* We need to move the initrd down into directly mapped mem */
265 	u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
266 						      PFN_PHYS(max_pfn_mapped));
267 	if (!relocated_ramdisk)
268 		panic("Cannot find place for new RAMDISK of size %lld\n",
269 		      ramdisk_size);
270 
271 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
272 	initrd_end   = initrd_start + ramdisk_size;
273 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
274 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
275 
276 	ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
277 	if (ret)
278 		panic("Copy RAMDISK failed\n");
279 
280 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
281 		" [mem %#010llx-%#010llx]\n",
282 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
283 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
284 }
285 
286 static void __init early_reserve_initrd(void)
287 {
288 	/* Assume only end is not page aligned */
289 	u64 ramdisk_image = get_ramdisk_image();
290 	u64 ramdisk_size  = get_ramdisk_size();
291 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
292 
293 	if (!boot_params.hdr.type_of_loader ||
294 	    !ramdisk_image || !ramdisk_size)
295 		return;		/* No initrd provided by bootloader */
296 
297 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
298 }
299 
300 static void __init reserve_initrd(void)
301 {
302 	/* Assume only end is not page aligned */
303 	u64 ramdisk_image = get_ramdisk_image();
304 	u64 ramdisk_size  = get_ramdisk_size();
305 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
306 
307 	if (!boot_params.hdr.type_of_loader ||
308 	    !ramdisk_image || !ramdisk_size)
309 		return;		/* No initrd provided by bootloader */
310 
311 	initrd_start = 0;
312 
313 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
314 			ramdisk_end - 1);
315 
316 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
317 				PFN_DOWN(ramdisk_end))) {
318 		/* All are mapped, easy case */
319 		initrd_start = ramdisk_image + PAGE_OFFSET;
320 		initrd_end = initrd_start + ramdisk_size;
321 		return;
322 	}
323 
324 	relocate_initrd();
325 
326 	memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
327 }
328 
329 #else
330 static void __init early_reserve_initrd(void)
331 {
332 }
333 static void __init reserve_initrd(void)
334 {
335 }
336 #endif /* CONFIG_BLK_DEV_INITRD */
337 
338 static void __init add_early_ima_buffer(u64 phys_addr)
339 {
340 #ifdef CONFIG_IMA
341 	struct ima_setup_data *data;
342 
343 	data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
344 	if (!data) {
345 		pr_warn("setup: failed to memremap ima_setup_data entry\n");
346 		return;
347 	}
348 
349 	if (data->size) {
350 		memblock_reserve(data->addr, data->size);
351 		ima_kexec_buffer_phys = data->addr;
352 		ima_kexec_buffer_size = data->size;
353 	}
354 
355 	early_memunmap(data, sizeof(*data));
356 #else
357 	pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
358 #endif
359 }
360 
361 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
362 int __init ima_free_kexec_buffer(void)
363 {
364 	if (!ima_kexec_buffer_size)
365 		return -ENOENT;
366 
367 	memblock_free_late(ima_kexec_buffer_phys,
368 			   ima_kexec_buffer_size);
369 
370 	ima_kexec_buffer_phys = 0;
371 	ima_kexec_buffer_size = 0;
372 
373 	return 0;
374 }
375 
376 int __init ima_get_kexec_buffer(void **addr, size_t *size)
377 {
378 	if (!ima_kexec_buffer_size)
379 		return -ENOENT;
380 
381 	*addr = __va(ima_kexec_buffer_phys);
382 	*size = ima_kexec_buffer_size;
383 
384 	return 0;
385 }
386 #endif
387 
388 static void __init parse_setup_data(void)
389 {
390 	struct setup_data *data;
391 	u64 pa_data, pa_next;
392 
393 	pa_data = boot_params.hdr.setup_data;
394 	while (pa_data) {
395 		u32 data_len, data_type;
396 
397 		data = early_memremap(pa_data, sizeof(*data));
398 		data_len = data->len + sizeof(struct setup_data);
399 		data_type = data->type;
400 		pa_next = data->next;
401 		early_memunmap(data, sizeof(*data));
402 
403 		switch (data_type) {
404 		case SETUP_E820_EXT:
405 			e820__memory_setup_extended(pa_data, data_len);
406 			break;
407 		case SETUP_DTB:
408 			add_dtb(pa_data);
409 			break;
410 		case SETUP_EFI:
411 			parse_efi_setup(pa_data, data_len);
412 			break;
413 		case SETUP_IMA:
414 			add_early_ima_buffer(pa_data);
415 			break;
416 		case SETUP_RNG_SEED:
417 			data = early_memremap(pa_data, data_len);
418 			add_bootloader_randomness(data->data, data->len);
419 			/* Zero seed for forward secrecy. */
420 			memzero_explicit(data->data, data->len);
421 			/* Zero length in case we find ourselves back here by accident. */
422 			memzero_explicit(&data->len, sizeof(data->len));
423 			early_memunmap(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_indirect *indirect;
435 	struct setup_data *data;
436 	u64 pa_data, pa_next;
437 	u32 len;
438 
439 	pa_data = boot_params.hdr.setup_data;
440 	while (pa_data) {
441 		data = early_memremap(pa_data, sizeof(*data));
442 		if (!data) {
443 			pr_warn("setup: failed to memremap setup_data entry\n");
444 			return;
445 		}
446 
447 		len = sizeof(*data);
448 		pa_next = data->next;
449 
450 		memblock_reserve(pa_data, sizeof(*data) + data->len);
451 
452 		if (data->type == SETUP_INDIRECT) {
453 			len += data->len;
454 			early_memunmap(data, sizeof(*data));
455 			data = early_memremap(pa_data, len);
456 			if (!data) {
457 				pr_warn("setup: failed to memremap indirect setup_data\n");
458 				return;
459 			}
460 
461 			indirect = (struct setup_indirect *)data->data;
462 
463 			if (indirect->type != SETUP_INDIRECT)
464 				memblock_reserve(indirect->addr, indirect->len);
465 		}
466 
467 		pa_data = pa_next;
468 		early_memunmap(data, len);
469 	}
470 }
471 
472 static void __init arch_reserve_crashkernel(void)
473 {
474 	unsigned long long crash_base, crash_size, low_size = 0;
475 	bool high = false;
476 	int ret;
477 
478 	if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
479 		return;
480 
481 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
482 				&crash_size, &crash_base,
483 				&low_size, &high);
484 	if (ret)
485 		return;
486 
487 	if (xen_pv_domain()) {
488 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
489 		return;
490 	}
491 
492 	reserve_crashkernel_generic(crash_size, crash_base, low_size, high);
493 }
494 
495 static struct resource standard_io_resources[] = {
496 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
497 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
498 	{ .name = "pic1", .start = 0x20, .end = 0x21,
499 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
500 	{ .name = "timer0", .start = 0x40, .end = 0x43,
501 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
502 	{ .name = "timer1", .start = 0x50, .end = 0x53,
503 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
504 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
505 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
506 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
507 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
508 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
509 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
510 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
511 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
512 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
513 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
514 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
515 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
516 };
517 
518 void __init reserve_standard_io_resources(void)
519 {
520 	int i;
521 
522 	/* request I/O space for devices used on all i[345]86 PCs */
523 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
524 		request_resource(&ioport_resource, &standard_io_resources[i]);
525 
526 }
527 
528 static bool __init snb_gfx_workaround_needed(void)
529 {
530 #ifdef CONFIG_PCI
531 	int i;
532 	u16 vendor, devid;
533 	static const __initconst u16 snb_ids[] = {
534 		0x0102,
535 		0x0112,
536 		0x0122,
537 		0x0106,
538 		0x0116,
539 		0x0126,
540 		0x010a,
541 	};
542 
543 	/* Assume no if something weird is going on with PCI */
544 	if (!early_pci_allowed())
545 		return false;
546 
547 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
548 	if (vendor != 0x8086)
549 		return false;
550 
551 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
552 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
553 		if (devid == snb_ids[i])
554 			return true;
555 #endif
556 
557 	return false;
558 }
559 
560 /*
561  * Sandy Bridge graphics has trouble with certain ranges, exclude
562  * them from allocation.
563  */
564 static void __init trim_snb_memory(void)
565 {
566 	static const __initconst unsigned long bad_pages[] = {
567 		0x20050000,
568 		0x20110000,
569 		0x20130000,
570 		0x20138000,
571 		0x40004000,
572 	};
573 	int i;
574 
575 	if (!snb_gfx_workaround_needed())
576 		return;
577 
578 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
579 
580 	/*
581 	 * SandyBridge integrated graphics devices have a bug that prevents
582 	 * them from accessing certain memory ranges, namely anything below
583 	 * 1M and in the pages listed in bad_pages[] above.
584 	 *
585 	 * To avoid these pages being ever accessed by SNB gfx devices reserve
586 	 * bad_pages that have not already been reserved at boot time.
587 	 * All memory below the 1 MB mark is anyway reserved later during
588 	 * setup_arch(), so there is no need to reserve it here.
589 	 */
590 
591 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
592 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
593 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
594 			       bad_pages[i]);
595 	}
596 }
597 
598 static void __init trim_bios_range(void)
599 {
600 	/*
601 	 * A special case is the first 4Kb of memory;
602 	 * This is a BIOS owned area, not kernel ram, but generally
603 	 * not listed as such in the E820 table.
604 	 *
605 	 * This typically reserves additional memory (64KiB by default)
606 	 * since some BIOSes are known to corrupt low memory.  See the
607 	 * Kconfig help text for X86_RESERVE_LOW.
608 	 */
609 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
610 
611 	/*
612 	 * special case: Some BIOSes report the PC BIOS
613 	 * area (640Kb -> 1Mb) as RAM even though it is not.
614 	 * take them out.
615 	 */
616 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
617 
618 	e820__update_table(e820_table);
619 }
620 
621 /* called before trim_bios_range() to spare extra sanitize */
622 static void __init e820_add_kernel_range(void)
623 {
624 	u64 start = __pa_symbol(_text);
625 	u64 size = __pa_symbol(_end) - start;
626 
627 	/*
628 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
629 	 * attempt to fix it by adding the range. We may have a confused BIOS,
630 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
631 	 * exclude kernel range. If we really are running on top non-RAM,
632 	 * we will crash later anyways.
633 	 */
634 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
635 		return;
636 
637 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
638 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
639 	e820__range_add(start, size, E820_TYPE_RAM);
640 }
641 
642 static void __init early_reserve_memory(void)
643 {
644 	/*
645 	 * Reserve the memory occupied by the kernel between _text and
646 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
647 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
648 	 * separate memblock_reserve() or they will be discarded.
649 	 */
650 	memblock_reserve(__pa_symbol(_text),
651 			 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
652 
653 	/*
654 	 * The first 4Kb of memory is a BIOS owned area, but generally it is
655 	 * not listed as such in the E820 table.
656 	 *
657 	 * Reserve the first 64K of memory since some BIOSes are known to
658 	 * corrupt low memory. After the real mode trampoline is allocated the
659 	 * rest of the memory below 640k is reserved.
660 	 *
661 	 * In addition, make sure page 0 is always reserved because on
662 	 * systems with L1TF its contents can be leaked to user processes.
663 	 */
664 	memblock_reserve(0, SZ_64K);
665 
666 	early_reserve_initrd();
667 
668 	memblock_x86_reserve_range_setup_data();
669 
670 	reserve_bios_regions();
671 	trim_snb_memory();
672 }
673 
674 /*
675  * Dump out kernel offset information on panic.
676  */
677 static int
678 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
679 {
680 	if (kaslr_enabled()) {
681 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
682 			 kaslr_offset(),
683 			 __START_KERNEL,
684 			 __START_KERNEL_map,
685 			 MODULES_VADDR-1);
686 	} else {
687 		pr_emerg("Kernel Offset: disabled\n");
688 	}
689 
690 	return 0;
691 }
692 
693 void x86_configure_nx(void)
694 {
695 	if (boot_cpu_has(X86_FEATURE_NX))
696 		__supported_pte_mask |= _PAGE_NX;
697 	else
698 		__supported_pte_mask &= ~_PAGE_NX;
699 }
700 
701 static void __init x86_report_nx(void)
702 {
703 	if (!boot_cpu_has(X86_FEATURE_NX)) {
704 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
705 		       "missing in CPU!\n");
706 	} else {
707 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
708 		printk(KERN_INFO "NX (Execute Disable) protection: active\n");
709 #else
710 		/* 32bit non-PAE kernel, NX cannot be used */
711 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
712 		       "cannot be enabled: non-PAE kernel!\n");
713 #endif
714 	}
715 }
716 
717 /*
718  * Determine if we were loaded by an EFI loader.  If so, then we have also been
719  * passed the efi memmap, systab, etc., so we should use these data structures
720  * for initialization.  Note, the efi init code path is determined by the
721  * global efi_enabled. This allows the same kernel image to be used on existing
722  * systems (with a traditional BIOS) as well as on EFI systems.
723  */
724 /*
725  * setup_arch - architecture-specific boot-time initializations
726  *
727  * Note: On x86_64, fixmaps are ready for use even before this is called.
728  */
729 
730 void __init setup_arch(char **cmdline_p)
731 {
732 #ifdef CONFIG_X86_32
733 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
734 
735 	/*
736 	 * copy kernel address range established so far and switch
737 	 * to the proper swapper page table
738 	 */
739 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
740 			initial_page_table + KERNEL_PGD_BOUNDARY,
741 			KERNEL_PGD_PTRS);
742 
743 	load_cr3(swapper_pg_dir);
744 	/*
745 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
746 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
747 	 * will not flush anything because the CPU quirk which clears
748 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
749 	 * load_cr3() above the TLB has been flushed already. The
750 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
751 	 * so proper operation is guaranteed.
752 	 */
753 	__flush_tlb_all();
754 #else
755 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
756 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
757 #endif
758 
759 #ifdef CONFIG_CMDLINE_BOOL
760 #ifdef CONFIG_CMDLINE_OVERRIDE
761 	strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
762 #else
763 	if (builtin_cmdline[0]) {
764 		/* append boot loader cmdline to builtin */
765 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
766 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
767 		strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
768 	}
769 #endif
770 	builtin_cmdline_added = true;
771 #endif
772 
773 	strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
774 	*cmdline_p = command_line;
775 
776 	/*
777 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
778 	 * reserve_top(), so do this before touching the ioremap area.
779 	 */
780 	olpc_ofw_detect();
781 
782 	idt_setup_early_traps();
783 	early_cpu_init();
784 	jump_label_init();
785 	static_call_init();
786 	early_ioremap_init();
787 
788 	setup_olpc_ofw_pgd();
789 
790 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
791 	screen_info = boot_params.screen_info;
792 	edid_info = boot_params.edid_info;
793 #ifdef CONFIG_X86_32
794 	apm_info.bios = boot_params.apm_bios_info;
795 	ist_info = boot_params.ist_info;
796 #endif
797 	saved_video_mode = boot_params.hdr.vid_mode;
798 	bootloader_type = boot_params.hdr.type_of_loader;
799 	if ((bootloader_type >> 4) == 0xe) {
800 		bootloader_type &= 0xf;
801 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
802 	}
803 	bootloader_version  = bootloader_type & 0xf;
804 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
805 
806 #ifdef CONFIG_BLK_DEV_RAM
807 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
808 #endif
809 #ifdef CONFIG_EFI
810 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
811 		     EFI32_LOADER_SIGNATURE, 4)) {
812 		set_bit(EFI_BOOT, &efi.flags);
813 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
814 		     EFI64_LOADER_SIGNATURE, 4)) {
815 		set_bit(EFI_BOOT, &efi.flags);
816 		set_bit(EFI_64BIT, &efi.flags);
817 	}
818 #endif
819 
820 	x86_init.oem.arch_setup();
821 
822 	/*
823 	 * Do some memory reservations *before* memory is added to memblock, so
824 	 * memblock allocations won't overwrite it.
825 	 *
826 	 * After this point, everything still needed from the boot loader or
827 	 * firmware or kernel text should be early reserved or marked not RAM in
828 	 * e820. All other memory is free game.
829 	 *
830 	 * This call needs to happen before e820__memory_setup() which calls the
831 	 * xen_memory_setup() on Xen dom0 which relies on the fact that those
832 	 * early reservations have happened already.
833 	 */
834 	early_reserve_memory();
835 
836 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
837 	e820__memory_setup();
838 	parse_setup_data();
839 
840 	copy_edd();
841 
842 	if (!boot_params.hdr.root_flags)
843 		root_mountflags &= ~MS_RDONLY;
844 	setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
845 
846 	code_resource.start = __pa_symbol(_text);
847 	code_resource.end = __pa_symbol(_etext)-1;
848 	rodata_resource.start = __pa_symbol(__start_rodata);
849 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
850 	data_resource.start = __pa_symbol(_sdata);
851 	data_resource.end = __pa_symbol(_edata)-1;
852 	bss_resource.start = __pa_symbol(__bss_start);
853 	bss_resource.end = __pa_symbol(__bss_stop)-1;
854 
855 	/*
856 	 * x86_configure_nx() is called before parse_early_param() to detect
857 	 * whether hardware doesn't support NX (so that the early EHCI debug
858 	 * console setup can safely call set_fixmap()).
859 	 */
860 	x86_configure_nx();
861 
862 	parse_early_param();
863 
864 	if (efi_enabled(EFI_BOOT))
865 		efi_memblock_x86_reserve_range();
866 
867 #ifdef CONFIG_MEMORY_HOTPLUG
868 	/*
869 	 * Memory used by the kernel cannot be hot-removed because Linux
870 	 * cannot migrate the kernel pages. When memory hotplug is
871 	 * enabled, we should prevent memblock from allocating memory
872 	 * for the kernel.
873 	 *
874 	 * ACPI SRAT records all hotpluggable memory ranges. But before
875 	 * SRAT is parsed, we don't know about it.
876 	 *
877 	 * The kernel image is loaded into memory at very early time. We
878 	 * cannot prevent this anyway. So on NUMA system, we set any
879 	 * node the kernel resides in as un-hotpluggable.
880 	 *
881 	 * Since on modern servers, one node could have double-digit
882 	 * gigabytes memory, we can assume the memory around the kernel
883 	 * image is also un-hotpluggable. So before SRAT is parsed, just
884 	 * allocate memory near the kernel image to try the best to keep
885 	 * the kernel away from hotpluggable memory.
886 	 */
887 	if (movable_node_is_enabled())
888 		memblock_set_bottom_up(true);
889 #endif
890 
891 	x86_report_nx();
892 
893 	apic_setup_apic_calls();
894 
895 	if (acpi_mps_check()) {
896 #ifdef CONFIG_X86_LOCAL_APIC
897 		apic_is_disabled = true;
898 #endif
899 		setup_clear_cpu_cap(X86_FEATURE_APIC);
900 	}
901 
902 	e820__reserve_setup_data();
903 	e820__finish_early_params();
904 
905 	if (efi_enabled(EFI_BOOT))
906 		efi_init();
907 
908 	reserve_ibft_region();
909 	x86_init.resources.dmi_setup();
910 
911 	/*
912 	 * VMware detection requires dmi to be available, so this
913 	 * needs to be done after dmi_setup(), for the boot CPU.
914 	 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
915 	 * called before cache_bp_init() for setting up MTRR state.
916 	 */
917 	init_hypervisor_platform();
918 
919 	tsc_early_init();
920 	x86_init.resources.probe_roms();
921 
922 	/* after parse_early_param, so could debug it */
923 	insert_resource(&iomem_resource, &code_resource);
924 	insert_resource(&iomem_resource, &rodata_resource);
925 	insert_resource(&iomem_resource, &data_resource);
926 	insert_resource(&iomem_resource, &bss_resource);
927 
928 	e820_add_kernel_range();
929 	trim_bios_range();
930 #ifdef CONFIG_X86_32
931 	if (ppro_with_ram_bug()) {
932 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
933 				  E820_TYPE_RESERVED);
934 		e820__update_table(e820_table);
935 		printk(KERN_INFO "fixed physical RAM map:\n");
936 		e820__print_table("bad_ppro");
937 	}
938 #else
939 	early_gart_iommu_check();
940 #endif
941 
942 	/*
943 	 * partially used pages are not usable - thus
944 	 * we are rounding upwards:
945 	 */
946 	max_pfn = e820__end_of_ram_pfn();
947 
948 	/* update e820 for memory not covered by WB MTRRs */
949 	cache_bp_init();
950 	if (mtrr_trim_uncached_memory(max_pfn))
951 		max_pfn = e820__end_of_ram_pfn();
952 
953 	max_possible_pfn = max_pfn;
954 
955 	/*
956 	 * Define random base addresses for memory sections after max_pfn is
957 	 * defined and before each memory section base is used.
958 	 */
959 	kernel_randomize_memory();
960 
961 #ifdef CONFIG_X86_32
962 	/* max_low_pfn get updated here */
963 	find_low_pfn_range();
964 #else
965 	check_x2apic();
966 
967 	/* How many end-of-memory variables you have, grandma! */
968 	/* need this before calling reserve_initrd */
969 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
970 		max_low_pfn = e820__end_of_low_ram_pfn();
971 	else
972 		max_low_pfn = max_pfn;
973 
974 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
975 #endif
976 
977 	/* Find and reserve MPTABLE area */
978 	x86_init.mpparse.find_mptable();
979 
980 	early_alloc_pgt_buf();
981 
982 	/*
983 	 * Need to conclude brk, before e820__memblock_setup()
984 	 * it could use memblock_find_in_range, could overlap with
985 	 * brk area.
986 	 */
987 	reserve_brk();
988 
989 	cleanup_highmap();
990 
991 	memblock_set_current_limit(ISA_END_ADDRESS);
992 	e820__memblock_setup();
993 
994 	/*
995 	 * Needs to run after memblock setup because it needs the physical
996 	 * memory size.
997 	 */
998 	mem_encrypt_setup_arch();
999 	cc_random_init();
1000 
1001 	efi_find_mirror();
1002 	efi_esrt_init();
1003 	efi_mokvar_table_init();
1004 
1005 	/*
1006 	 * The EFI specification says that boot service code won't be
1007 	 * called after ExitBootServices(). This is, in fact, a lie.
1008 	 */
1009 	efi_reserve_boot_services();
1010 
1011 	/* preallocate 4k for mptable mpc */
1012 	e820__memblock_alloc_reserved_mpc_new();
1013 
1014 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1015 	setup_bios_corruption_check();
1016 #endif
1017 
1018 #ifdef CONFIG_X86_32
1019 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1020 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1021 #endif
1022 
1023 	/*
1024 	 * Find free memory for the real mode trampoline and place it there. If
1025 	 * there is not enough free memory under 1M, on EFI-enabled systems
1026 	 * there will be additional attempt to reclaim the memory for the real
1027 	 * mode trampoline at efi_free_boot_services().
1028 	 *
1029 	 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1030 	 * are known to corrupt low memory and several hundred kilobytes are not
1031 	 * worth complex detection what memory gets clobbered. Windows does the
1032 	 * same thing for very similar reasons.
1033 	 *
1034 	 * Moreover, on machines with SandyBridge graphics or in setups that use
1035 	 * crashkernel the entire 1M is reserved anyway.
1036 	 *
1037 	 * Note the host kernel TDX also requires the first 1MB being reserved.
1038 	 */
1039 	x86_platform.realmode_reserve();
1040 
1041 	init_mem_mapping();
1042 
1043 	/*
1044 	 * init_mem_mapping() relies on the early IDT page fault handling.
1045 	 * Now either enable FRED or install the real page fault handler
1046 	 * for 64-bit in the IDT.
1047 	 */
1048 	cpu_init_replace_early_idt();
1049 
1050 	/*
1051 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1052 	 * with the current CR4 value.  This may not be necessary, but
1053 	 * auditing all the early-boot CR4 manipulation would be needed to
1054 	 * rule it out.
1055 	 *
1056 	 * Mask off features that don't work outside long mode (just
1057 	 * PCIDE for now).
1058 	 */
1059 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1060 
1061 	memblock_set_current_limit(get_max_mapped());
1062 
1063 	/*
1064 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1065 	 */
1066 
1067 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1068 	if (init_ohci1394_dma_early)
1069 		init_ohci1394_dma_on_all_controllers();
1070 #endif
1071 	/* Allocate bigger log buffer */
1072 	setup_log_buf(1);
1073 
1074 	if (efi_enabled(EFI_BOOT)) {
1075 		switch (boot_params.secure_boot) {
1076 		case efi_secureboot_mode_disabled:
1077 			pr_info("Secure boot disabled\n");
1078 			break;
1079 		case efi_secureboot_mode_enabled:
1080 			pr_info("Secure boot enabled\n");
1081 			break;
1082 		default:
1083 			pr_info("Secure boot could not be determined\n");
1084 			break;
1085 		}
1086 	}
1087 
1088 	reserve_initrd();
1089 
1090 	acpi_table_upgrade();
1091 	/* Look for ACPI tables and reserve memory occupied by them. */
1092 	acpi_boot_table_init();
1093 
1094 	vsmp_init();
1095 
1096 	io_delay_init();
1097 
1098 	early_platform_quirks();
1099 
1100 	/* Some platforms need the APIC registered for NUMA configuration */
1101 	early_acpi_boot_init();
1102 	x86_init.mpparse.early_parse_smp_cfg();
1103 
1104 	x86_flattree_get_config();
1105 
1106 	initmem_init();
1107 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1108 
1109 	if (boot_cpu_has(X86_FEATURE_GBPAGES))
1110 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1111 
1112 	/*
1113 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1114 	 * won't consume hotpluggable memory.
1115 	 */
1116 	arch_reserve_crashkernel();
1117 
1118 	if (!early_xdbc_setup_hardware())
1119 		early_xdbc_register_console();
1120 
1121 	x86_init.paging.pagetable_init();
1122 
1123 	kasan_init();
1124 
1125 	/*
1126 	 * Sync back kernel address range.
1127 	 *
1128 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1129 	 * this call?
1130 	 */
1131 	sync_initial_page_table();
1132 
1133 	tboot_probe();
1134 
1135 	map_vsyscall();
1136 
1137 	x86_32_probe_apic();
1138 
1139 	early_quirks();
1140 
1141 	topology_apply_cmdline_limits_early();
1142 
1143 	/*
1144 	 * Parse SMP configuration. Try ACPI first and then the platform
1145 	 * specific parser.
1146 	 */
1147 	acpi_boot_init();
1148 	x86_init.mpparse.parse_smp_cfg();
1149 
1150 	/* Last opportunity to detect and map the local APIC */
1151 	init_apic_mappings();
1152 
1153 	topology_init_possible_cpus();
1154 
1155 	init_cpu_to_node();
1156 	init_gi_nodes();
1157 
1158 	io_apic_init_mappings();
1159 
1160 	x86_init.hyper.guest_late_init();
1161 
1162 	e820__reserve_resources();
1163 	e820__register_nosave_regions(max_pfn);
1164 
1165 	x86_init.resources.reserve_resources();
1166 
1167 	e820__setup_pci_gap();
1168 
1169 #ifdef CONFIG_VT
1170 #if defined(CONFIG_VGA_CONSOLE)
1171 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1172 		vgacon_register_screen(&screen_info);
1173 #endif
1174 #endif
1175 	x86_init.oem.banner();
1176 
1177 	x86_init.timers.wallclock_init();
1178 
1179 	/*
1180 	 * This needs to run before setup_local_APIC() which soft-disables the
1181 	 * local APIC temporarily and that masks the thermal LVT interrupt,
1182 	 * leading to softlockups on machines which have configured SMI
1183 	 * interrupt delivery.
1184 	 */
1185 	therm_lvt_init();
1186 
1187 	mcheck_init();
1188 
1189 	register_refined_jiffies(CLOCK_TICK_RATE);
1190 
1191 #ifdef CONFIG_EFI
1192 	if (efi_enabled(EFI_BOOT))
1193 		efi_apply_memmap_quirks();
1194 #endif
1195 
1196 	unwind_init();
1197 }
1198 
1199 #ifdef CONFIG_X86_32
1200 
1201 static struct resource video_ram_resource = {
1202 	.name	= "Video RAM area",
1203 	.start	= 0xa0000,
1204 	.end	= 0xbffff,
1205 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1206 };
1207 
1208 void __init i386_reserve_resources(void)
1209 {
1210 	request_resource(&iomem_resource, &video_ram_resource);
1211 	reserve_standard_io_resources();
1212 }
1213 
1214 #endif /* CONFIG_X86_32 */
1215 
1216 static struct notifier_block kernel_offset_notifier = {
1217 	.notifier_call = dump_kernel_offset
1218 };
1219 
1220 static int __init register_kernel_offset_dumper(void)
1221 {
1222 	atomic_notifier_chain_register(&panic_notifier_list,
1223 					&kernel_offset_notifier);
1224 	return 0;
1225 }
1226 __initcall(register_kernel_offset_dumper);
1227 
1228 #ifdef CONFIG_HOTPLUG_CPU
1229 bool arch_cpu_is_hotpluggable(int cpu)
1230 {
1231 	return cpu > 0;
1232 }
1233 #endif /* CONFIG_HOTPLUG_CPU */
1234