xref: /linux/arch/x86/kernel/setup.c (revision 37a93dd5c49b5fda807fd204edf2547c3493319c)
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/hugetlb.h>
15 #include <linux/ima.h>
16 #include <linux/init_ohci1394_dma.h>
17 #include <linux/initrd.h>
18 #include <linux/iscsi_ibft.h>
19 #include <linux/memblock.h>
20 #include <linux/panic_notifier.h>
21 #include <linux/pci.h>
22 #include <linux/random.h>
23 #include <linux/root_dev.h>
24 #include <linux/static_call.h>
25 #include <linux/sysfb.h>
26 #include <linux/swiotlb.h>
27 #include <linux/tboot.h>
28 #include <linux/usb/xhci-dbgp.h>
29 #include <linux/vmalloc.h>
30 
31 #include <uapi/linux/mount.h>
32 
33 #include <xen/xen.h>
34 
35 #include <asm/apic.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/nmi.h>
51 #include <asm/numa.h>
52 #include <asm/olpc_ofw.h>
53 #include <asm/pci-direct.h>
54 #include <asm/prom.h>
55 #include <asm/proto.h>
56 #include <asm/realmode.h>
57 #include <asm/thermal.h>
58 #include <asm/unwind.h>
59 #include <asm/vsyscall.h>
60 
61 /*
62  * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
63  * max_pfn_mapped:     highest directly mapped pfn > 4 GB
64  *
65  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
66  * represented by pfn_mapped[].
67  */
68 unsigned long max_low_pfn_mapped;
69 unsigned long max_pfn_mapped;
70 
71 #ifdef CONFIG_DMI
72 RESERVE_BRK(dmi_alloc, 65536);
73 #endif
74 
75 
76 unsigned long _brk_start = (unsigned long)__brk_base;
77 unsigned long _brk_end   = (unsigned long)__brk_base;
78 
79 struct boot_params boot_params;
80 
81 /*
82  * These are the four main kernel memory regions, we put them into
83  * the resource tree so that kdump tools and other debugging tools
84  * recover it:
85  */
86 
87 static struct resource rodata_resource = {
88 	.name	= "Kernel rodata",
89 	.start	= 0,
90 	.end	= 0,
91 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
92 };
93 
94 static struct resource data_resource = {
95 	.name	= "Kernel data",
96 	.start	= 0,
97 	.end	= 0,
98 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
99 };
100 
101 static struct resource code_resource = {
102 	.name	= "Kernel code",
103 	.start	= 0,
104 	.end	= 0,
105 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
106 };
107 
108 static struct resource bss_resource = {
109 	.name	= "Kernel bss",
110 	.start	= 0,
111 	.end	= 0,
112 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
113 };
114 
115 
116 #ifdef CONFIG_X86_32
117 /* CPU data as detected by the assembly code in head_32.S */
118 struct cpuinfo_x86 new_cpu_data;
119 
120 struct apm_info apm_info;
121 EXPORT_SYMBOL(apm_info);
122 
123 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
124 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
125 struct ist_info ist_info;
126 EXPORT_SYMBOL(ist_info);
127 #else
128 struct ist_info ist_info;
129 #endif
130 
131 #endif
132 
133 struct cpuinfo_x86 boot_cpu_data __read_mostly;
134 EXPORT_SYMBOL(boot_cpu_data);
135 SYM_PIC_ALIAS(boot_cpu_data);
136 
137 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
138 __visible unsigned long mmu_cr4_features __ro_after_init;
139 #else
140 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
141 #endif
142 
143 #ifdef CONFIG_IMA
144 static phys_addr_t ima_kexec_buffer_phys;
145 static size_t ima_kexec_buffer_size;
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 static const struct ctl_table x86_sysctl_table[] = {
152 	{
153 		.procname       = "unknown_nmi_panic",
154 		.data           = &unknown_nmi_panic,
155 		.maxlen         = sizeof(int),
156 		.mode           = 0644,
157 		.proc_handler   = proc_dointvec,
158 	},
159 	{
160 		.procname	= "panic_on_unrecovered_nmi",
161 		.data		= &panic_on_unrecovered_nmi,
162 		.maxlen		= sizeof(int),
163 		.mode		= 0644,
164 		.proc_handler	= proc_dointvec,
165 	},
166 	{
167 		.procname	= "panic_on_io_nmi",
168 		.data		= &panic_on_io_nmi,
169 		.maxlen		= sizeof(int),
170 		.mode		= 0644,
171 		.proc_handler	= proc_dointvec,
172 	},
173 	{
174 		.procname	= "bootloader_type",
175 		.data		= &bootloader_type,
176 		.maxlen		= sizeof(int),
177 		.mode		= 0444,
178 		.proc_handler	= proc_dointvec,
179 	},
180 	{
181 		.procname	= "bootloader_version",
182 		.data		= &bootloader_version,
183 		.maxlen		= sizeof(int),
184 		.mode		= 0444,
185 		.proc_handler	= proc_dointvec,
186 	},
187 	{
188 		.procname	= "io_delay_type",
189 		.data		= &io_delay_type,
190 		.maxlen		= sizeof(int),
191 		.mode		= 0644,
192 		.proc_handler	= proc_dointvec,
193 	},
194 #if defined(CONFIG_ACPI_SLEEP)
195 	{
196 		.procname	= "acpi_video_flags",
197 		.data		= &acpi_realmode_flags,
198 		.maxlen		= sizeof(unsigned long),
199 		.mode		= 0644,
200 		.proc_handler	= proc_doulongvec_minmax,
201 	},
202 #endif
203 };
204 
205 static int __init init_x86_sysctl(void)
206 {
207 	register_sysctl_init("kernel", x86_sysctl_table);
208 	return 0;
209 }
210 arch_initcall(init_x86_sysctl);
211 
212 /*
213  * Setup options
214  */
215 
216 struct sysfb_display_info sysfb_primary_display;
217 EXPORT_SYMBOL(sysfb_primary_display);
218 
219 extern int root_mountflags;
220 
221 unsigned long saved_video_mode;
222 
223 #define RAMDISK_IMAGE_START_MASK	0x07FF
224 #define RAMDISK_PROMPT_FLAG		0x8000
225 #define RAMDISK_LOAD_FLAG		0x4000
226 
227 static char __initdata command_line[COMMAND_LINE_SIZE];
228 #ifdef CONFIG_CMDLINE_BOOL
229 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
230 bool builtin_cmdline_added __ro_after_init;
231 #endif
232 
233 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
234 struct edd edd;
235 #ifdef CONFIG_EDD_MODULE
236 EXPORT_SYMBOL(edd);
237 #endif
238 /**
239  * copy_edd() - Copy the BIOS EDD information
240  *              from boot_params into a safe place.
241  *
242  */
243 static inline void __init copy_edd(void)
244 {
245      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
246 	    sizeof(edd.mbr_signature));
247      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
248      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
249      edd.edd_info_nr = boot_params.eddbuf_entries;
250 }
251 #else
252 static inline void __init copy_edd(void)
253 {
254 }
255 #endif
256 
257 void * __init extend_brk(size_t size, size_t align)
258 {
259 	size_t mask = align - 1;
260 	void *ret;
261 
262 	BUG_ON(_brk_start == 0);
263 	BUG_ON(align & mask);
264 
265 	_brk_end = (_brk_end + mask) & ~mask;
266 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
267 
268 	ret = (void *)_brk_end;
269 	_brk_end += size;
270 
271 	memset(ret, 0, size);
272 
273 	return ret;
274 }
275 
276 #ifdef CONFIG_X86_32
277 static void __init cleanup_highmap(void)
278 {
279 }
280 #endif
281 
282 static void __init reserve_brk(void)
283 {
284 	if (_brk_end > _brk_start)
285 		memblock_reserve_kern(__pa_symbol(_brk_start),
286 				      _brk_end - _brk_start);
287 
288 	/* Mark brk area as locked down and no longer taking any
289 	   new allocations */
290 	_brk_start = 0;
291 }
292 
293 #ifdef CONFIG_BLK_DEV_INITRD
294 
295 static u64 __init get_ramdisk_image(void)
296 {
297 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
298 
299 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
300 
301 	if (ramdisk_image == 0)
302 		ramdisk_image = phys_initrd_start;
303 
304 	return ramdisk_image;
305 }
306 static u64 __init get_ramdisk_size(void)
307 {
308 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
309 
310 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
311 
312 	if (ramdisk_size == 0)
313 		ramdisk_size = phys_initrd_size;
314 
315 	return ramdisk_size;
316 }
317 
318 static void __init relocate_initrd(void)
319 {
320 	/* Assume only end is not page aligned */
321 	u64 ramdisk_image = get_ramdisk_image();
322 	u64 ramdisk_size  = get_ramdisk_size();
323 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
324 	int ret = 0;
325 
326 	/* We need to move the initrd down into directly mapped mem */
327 	u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
328 						      PFN_PHYS(max_pfn_mapped));
329 	if (!relocated_ramdisk)
330 		panic("Cannot find place for new RAMDISK of size %lld\n",
331 		      ramdisk_size);
332 
333 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
334 	initrd_end   = initrd_start + ramdisk_size;
335 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
336 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
337 
338 	ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
339 	if (ret)
340 		panic("Copy RAMDISK failed\n");
341 
342 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
343 		" [mem %#010llx-%#010llx]\n",
344 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
345 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
346 }
347 
348 static void __init early_reserve_initrd(void)
349 {
350 	/* Assume only end is not page aligned */
351 	u64 ramdisk_image = get_ramdisk_image();
352 	u64 ramdisk_size  = get_ramdisk_size();
353 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
354 
355 	if (!boot_params.hdr.type_of_loader ||
356 	    !ramdisk_image || !ramdisk_size)
357 		return;		/* No initrd provided by bootloader */
358 
359 	memblock_reserve_kern(ramdisk_image, ramdisk_end - ramdisk_image);
360 }
361 
362 static void __init reserve_initrd(void)
363 {
364 	/* Assume only end is not page aligned */
365 	u64 ramdisk_image = get_ramdisk_image();
366 	u64 ramdisk_size  = get_ramdisk_size();
367 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
368 
369 	if (!boot_params.hdr.type_of_loader ||
370 	    !ramdisk_image || !ramdisk_size)
371 		return;		/* No initrd provided by bootloader */
372 
373 	initrd_start = 0;
374 
375 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
376 			ramdisk_end - 1);
377 
378 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
379 				PFN_DOWN(ramdisk_end))) {
380 		/* All are mapped, easy case */
381 		initrd_start = ramdisk_image + PAGE_OFFSET;
382 		initrd_end = initrd_start + ramdisk_size;
383 		return;
384 	}
385 
386 	relocate_initrd();
387 
388 	memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
389 }
390 
391 #else
392 static void __init early_reserve_initrd(void)
393 {
394 }
395 static void __init reserve_initrd(void)
396 {
397 }
398 #endif /* CONFIG_BLK_DEV_INITRD */
399 
400 static void __init add_early_ima_buffer(u64 phys_addr)
401 {
402 #ifdef CONFIG_IMA
403 	struct ima_setup_data *data;
404 
405 	data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
406 	if (!data) {
407 		pr_warn("setup: failed to memremap ima_setup_data entry\n");
408 		return;
409 	}
410 
411 	if (data->size) {
412 		memblock_reserve_kern(data->addr, data->size);
413 		ima_kexec_buffer_phys = data->addr;
414 		ima_kexec_buffer_size = data->size;
415 	}
416 
417 	early_memunmap(data, sizeof(*data));
418 #else
419 	pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
420 #endif
421 }
422 
423 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
424 int __init ima_free_kexec_buffer(void)
425 {
426 	if (!ima_kexec_buffer_size)
427 		return -ENOENT;
428 
429 	memblock_free_late(ima_kexec_buffer_phys,
430 			   ima_kexec_buffer_size);
431 
432 	ima_kexec_buffer_phys = 0;
433 	ima_kexec_buffer_size = 0;
434 
435 	return 0;
436 }
437 
438 int __init ima_get_kexec_buffer(void **addr, size_t *size)
439 {
440 	if (!ima_kexec_buffer_size)
441 		return -ENOENT;
442 
443 	*addr = __va(ima_kexec_buffer_phys);
444 	*size = ima_kexec_buffer_size;
445 
446 	return 0;
447 }
448 #endif
449 
450 static void __init add_kho(u64 phys_addr, u32 data_len)
451 {
452 	struct kho_data *kho;
453 	u64 addr = phys_addr + sizeof(struct setup_data);
454 	u64 size = data_len - sizeof(struct setup_data);
455 
456 	if (!IS_ENABLED(CONFIG_KEXEC_HANDOVER)) {
457 		pr_warn("Passed KHO data, but CONFIG_KEXEC_HANDOVER not set. Ignoring.\n");
458 		return;
459 	}
460 
461 	kho = early_memremap(addr, size);
462 	if (!kho) {
463 		pr_warn("setup: failed to memremap kho data (0x%llx, 0x%llx)\n",
464 			addr, size);
465 		return;
466 	}
467 
468 	kho_populate(kho->fdt_addr, kho->fdt_size, kho->scratch_addr, kho->scratch_size);
469 
470 	early_memunmap(kho, size);
471 }
472 
473 static void __init parse_setup_data(void)
474 {
475 	struct setup_data *data;
476 	u64 pa_data, pa_next;
477 
478 	pa_data = boot_params.hdr.setup_data;
479 	while (pa_data) {
480 		u32 data_len, data_type;
481 
482 		data = early_memremap(pa_data, sizeof(*data));
483 		data_len = data->len + sizeof(struct setup_data);
484 		data_type = data->type;
485 		pa_next = data->next;
486 		early_memunmap(data, sizeof(*data));
487 
488 		switch (data_type) {
489 		case SETUP_E820_EXT:
490 			e820__memory_setup_extended(pa_data, data_len);
491 			break;
492 		case SETUP_DTB:
493 			add_dtb(pa_data);
494 			break;
495 		case SETUP_EFI:
496 			parse_efi_setup(pa_data, data_len);
497 			break;
498 		case SETUP_IMA:
499 			add_early_ima_buffer(pa_data);
500 			break;
501 		case SETUP_KEXEC_KHO:
502 			add_kho(pa_data, data_len);
503 			break;
504 		case SETUP_RNG_SEED:
505 			data = early_memremap(pa_data, data_len);
506 			add_bootloader_randomness(data->data, data->len);
507 			/* Zero seed for forward secrecy. */
508 			memzero_explicit(data->data, data->len);
509 			/* Zero length in case we find ourselves back here by accident. */
510 			memzero_explicit(&data->len, sizeof(data->len));
511 			early_memunmap(data, data_len);
512 			break;
513 		default:
514 			break;
515 		}
516 		pa_data = pa_next;
517 	}
518 }
519 
520 /*
521  * Translate the fields of 'struct boot_param' into global variables
522  * representing these parameters.
523  */
524 static void __init parse_boot_params(void)
525 {
526 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
527 	sysfb_primary_display.screen = boot_params.screen_info;
528 #if defined(CONFIG_FIRMWARE_EDID)
529 	sysfb_primary_display.edid = boot_params.edid_info;
530 #endif
531 #ifdef CONFIG_X86_32
532 	apm_info.bios = boot_params.apm_bios_info;
533 	ist_info = boot_params.ist_info;
534 #endif
535 	saved_video_mode = boot_params.hdr.vid_mode;
536 	bootloader_type = boot_params.hdr.type_of_loader;
537 	if ((bootloader_type >> 4) == 0xe) {
538 		bootloader_type &= 0xf;
539 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
540 	}
541 	bootloader_version  = bootloader_type & 0xf;
542 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
543 
544 #ifdef CONFIG_BLK_DEV_RAM
545 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
546 #endif
547 #ifdef CONFIG_EFI
548 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
549 		     EFI32_LOADER_SIGNATURE, 4)) {
550 		set_bit(EFI_BOOT, &efi.flags);
551 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
552 		     EFI64_LOADER_SIGNATURE, 4)) {
553 		set_bit(EFI_BOOT, &efi.flags);
554 		set_bit(EFI_64BIT, &efi.flags);
555 	}
556 #endif
557 
558 	if (!boot_params.hdr.root_flags)
559 		root_mountflags &= ~MS_RDONLY;
560 }
561 
562 static void __init memblock_x86_reserve_range_setup_data(void)
563 {
564 	struct setup_indirect *indirect;
565 	struct setup_data *data;
566 	u64 pa_data, pa_next;
567 	u32 len;
568 
569 	pa_data = boot_params.hdr.setup_data;
570 	while (pa_data) {
571 		data = early_memremap(pa_data, sizeof(*data));
572 		if (!data) {
573 			pr_warn("setup: failed to memremap setup_data entry\n");
574 			return;
575 		}
576 
577 		len = sizeof(*data);
578 		pa_next = data->next;
579 
580 		memblock_reserve_kern(pa_data, sizeof(*data) + data->len);
581 
582 		if (data->type == SETUP_INDIRECT) {
583 			len += data->len;
584 			early_memunmap(data, sizeof(*data));
585 			data = early_memremap(pa_data, len);
586 			if (!data) {
587 				pr_warn("setup: failed to memremap indirect setup_data\n");
588 				return;
589 			}
590 
591 			indirect = (struct setup_indirect *)data->data;
592 
593 			if (indirect->type != SETUP_INDIRECT)
594 				memblock_reserve_kern(indirect->addr, indirect->len);
595 		}
596 
597 		pa_data = pa_next;
598 		early_memunmap(data, len);
599 	}
600 }
601 
602 static void __init arch_reserve_crashkernel(void)
603 {
604 	unsigned long long crash_base, crash_size, low_size = 0, cma_size = 0;
605 	bool high = false;
606 	int ret;
607 
608 	if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
609 		return;
610 
611 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
612 				&crash_size, &crash_base,
613 				&low_size, &cma_size, &high);
614 	if (ret)
615 		return;
616 
617 	if (xen_pv_domain()) {
618 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
619 		return;
620 	}
621 
622 	reserve_crashkernel_generic(crash_size, crash_base, low_size, high);
623 	reserve_crashkernel_cma(cma_size);
624 }
625 
626 static struct resource standard_io_resources[] = {
627 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
628 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
629 	{ .name = "pic1", .start = 0x20, .end = 0x21,
630 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
631 	{ .name = "timer0", .start = 0x40, .end = 0x43,
632 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
633 	{ .name = "timer1", .start = 0x50, .end = 0x53,
634 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
635 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
636 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
637 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
638 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
639 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
640 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
641 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
642 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
643 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
644 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
645 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
646 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
647 };
648 
649 void __init reserve_standard_io_resources(void)
650 {
651 	int i;
652 
653 	/* request I/O space for devices used on all i[345]86 PCs */
654 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
655 		request_resource(&ioport_resource, &standard_io_resources[i]);
656 
657 }
658 
659 static void __init setup_kernel_resources(void)
660 {
661 	code_resource.start = __pa_symbol(_text);
662 	code_resource.end = __pa_symbol(_etext)-1;
663 	rodata_resource.start = __pa_symbol(__start_rodata);
664 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
665 	data_resource.start = __pa_symbol(_sdata);
666 	data_resource.end = __pa_symbol(_edata)-1;
667 	bss_resource.start = __pa_symbol(__bss_start);
668 	bss_resource.end = __pa_symbol(__bss_stop)-1;
669 
670 	insert_resource(&iomem_resource, &code_resource);
671 	insert_resource(&iomem_resource, &rodata_resource);
672 	insert_resource(&iomem_resource, &data_resource);
673 	insert_resource(&iomem_resource, &bss_resource);
674 }
675 
676 static bool __init snb_gfx_workaround_needed(void)
677 {
678 #ifdef CONFIG_PCI
679 	int i;
680 	u16 vendor, devid;
681 	static const __initconst u16 snb_ids[] = {
682 		0x0102,
683 		0x0112,
684 		0x0122,
685 		0x0106,
686 		0x0116,
687 		0x0126,
688 		0x010a,
689 	};
690 
691 	/* Assume no if something weird is going on with PCI */
692 	if (!early_pci_allowed())
693 		return false;
694 
695 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
696 	if (vendor != 0x8086)
697 		return false;
698 
699 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
700 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
701 		if (devid == snb_ids[i])
702 			return true;
703 #endif
704 
705 	return false;
706 }
707 
708 /*
709  * Sandy Bridge graphics has trouble with certain ranges, exclude
710  * them from allocation.
711  */
712 static void __init trim_snb_memory(void)
713 {
714 	static const __initconst unsigned long bad_pages[] = {
715 		0x20050000,
716 		0x20110000,
717 		0x20130000,
718 		0x20138000,
719 		0x40004000,
720 	};
721 	int i;
722 
723 	if (!snb_gfx_workaround_needed())
724 		return;
725 
726 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
727 
728 	/*
729 	 * SandyBridge integrated graphics devices have a bug that prevents
730 	 * them from accessing certain memory ranges, namely anything below
731 	 * 1M and in the pages listed in bad_pages[] above.
732 	 *
733 	 * To avoid these pages being ever accessed by SNB gfx devices reserve
734 	 * bad_pages that have not already been reserved at boot time.
735 	 * All memory below the 1 MB mark is anyway reserved later during
736 	 * setup_arch(), so there is no need to reserve it here.
737 	 */
738 
739 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
740 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
741 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
742 			       bad_pages[i]);
743 	}
744 }
745 
746 static void __init trim_bios_range(void)
747 {
748 	/*
749 	 * A special case is the first 4Kb of memory;
750 	 * This is a BIOS owned area, not kernel ram, but generally
751 	 * not listed as such in the E820 table.
752 	 *
753 	 * This typically reserves additional memory (64KiB by default)
754 	 * since some BIOSes are known to corrupt low memory.  See the
755 	 * Kconfig help text for X86_RESERVE_LOW.
756 	 */
757 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
758 
759 	/*
760 	 * special case: Some BIOSes report the PC BIOS
761 	 * area (640Kb -> 1Mb) as RAM even though it is not.
762 	 * take them out.
763 	 */
764 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM);
765 
766 	e820__update_table(e820_table);
767 }
768 
769 /* called before trim_bios_range() to spare extra sanitize */
770 static void __init e820_add_kernel_range(void)
771 {
772 	u64 start = __pa_symbol(_text);
773 	u64 size = __pa_symbol(_end) - start;
774 
775 	/*
776 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
777 	 * attempt to fix it by adding the range. We may have a confused BIOS,
778 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
779 	 * exclude kernel range. If we really are running on top non-RAM,
780 	 * we will crash later anyways.
781 	 */
782 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
783 		return;
784 
785 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
786 	e820__range_remove(start, size, 0);
787 	e820__range_add(start, size, E820_TYPE_RAM);
788 }
789 
790 static void __init early_reserve_memory(void)
791 {
792 	/*
793 	 * Reserve the memory occupied by the kernel between _text and
794 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
795 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
796 	 * separate memblock_reserve() or they will be discarded.
797 	 */
798 	memblock_reserve_kern(__pa_symbol(_text),
799 			      (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
800 
801 	/*
802 	 * The first 4Kb of memory is a BIOS owned area, but generally it is
803 	 * not listed as such in the E820 table.
804 	 *
805 	 * Reserve the first 64K of memory since some BIOSes are known to
806 	 * corrupt low memory. After the real mode trampoline is allocated the
807 	 * rest of the memory below 640k is reserved.
808 	 *
809 	 * In addition, make sure page 0 is always reserved because on
810 	 * systems with L1TF its contents can be leaked to user processes.
811 	 */
812 	memblock_reserve(0, SZ_64K);
813 
814 	early_reserve_initrd();
815 
816 	memblock_x86_reserve_range_setup_data();
817 
818 	reserve_bios_regions();
819 	trim_snb_memory();
820 }
821 
822 /*
823  * Dump out kernel offset information on panic.
824  */
825 static int
826 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
827 {
828 	if (kaslr_enabled()) {
829 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
830 			 kaslr_offset(),
831 			 __START_KERNEL,
832 			 __START_KERNEL_map,
833 			 MODULES_VADDR-1);
834 	} else {
835 		pr_emerg("Kernel Offset: disabled\n");
836 	}
837 
838 	return 0;
839 }
840 
841 void x86_configure_nx(void)
842 {
843 	if (boot_cpu_has(X86_FEATURE_NX))
844 		__supported_pte_mask |= _PAGE_NX;
845 	else
846 		__supported_pte_mask &= ~_PAGE_NX;
847 }
848 
849 static void __init x86_report_nx(void)
850 {
851 	if (!boot_cpu_has(X86_FEATURE_NX)) {
852 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
853 		       "missing in CPU!\n");
854 	} else {
855 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
856 		printk(KERN_INFO "NX (Execute Disable) protection: active\n");
857 #else
858 		/* 32bit non-PAE kernel, NX cannot be used */
859 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
860 		       "cannot be enabled: non-PAE kernel!\n");
861 #endif
862 	}
863 }
864 
865 /*
866  * Determine if we were loaded by an EFI loader.  If so, then we have also been
867  * passed the efi memmap, systab, etc., so we should use these data structures
868  * for initialization.  Note, the efi init code path is determined by the
869  * global efi_enabled. This allows the same kernel image to be used on existing
870  * systems (with a traditional BIOS) as well as on EFI systems.
871  */
872 /*
873  * setup_arch - architecture-specific boot-time initializations
874  *
875  * Note: On x86_64, fixmaps are ready for use even before this is called.
876  */
877 
878 void __init setup_arch(char **cmdline_p)
879 {
880 #ifdef CONFIG_X86_32
881 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
882 
883 	/*
884 	 * copy kernel address range established so far and switch
885 	 * to the proper swapper page table
886 	 */
887 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
888 			initial_page_table + KERNEL_PGD_BOUNDARY,
889 			KERNEL_PGD_PTRS);
890 
891 	load_cr3(swapper_pg_dir);
892 	/*
893 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
894 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
895 	 * will not flush anything because the CPU quirk which clears
896 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
897 	 * load_cr3() above the TLB has been flushed already. The
898 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
899 	 * so proper operation is guaranteed.
900 	 */
901 	__flush_tlb_all();
902 #else
903 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
904 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
905 #endif
906 
907 #ifdef CONFIG_CMDLINE_BOOL
908 #ifdef CONFIG_CMDLINE_OVERRIDE
909 	strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
910 #else
911 	if (builtin_cmdline[0]) {
912 		/* append boot loader cmdline to builtin */
913 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
914 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
915 		strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
916 	}
917 #endif
918 	builtin_cmdline_added = true;
919 #endif
920 
921 	strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
922 	*cmdline_p = command_line;
923 
924 	/*
925 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
926 	 * reserve_top(), so do this before touching the ioremap area.
927 	 */
928 	olpc_ofw_detect();
929 
930 	idt_setup_early_traps();
931 	early_cpu_init();
932 	jump_label_init();
933 	static_call_init();
934 	early_ioremap_init();
935 
936 	setup_olpc_ofw_pgd();
937 
938 	parse_boot_params();
939 
940 	x86_init.oem.arch_setup();
941 
942 	/*
943 	 * Do some memory reservations *before* memory is added to memblock, so
944 	 * memblock allocations won't overwrite it.
945 	 *
946 	 * After this point, everything still needed from the boot loader or
947 	 * firmware or kernel text should be early reserved or marked not RAM in
948 	 * e820. All other memory is free game.
949 	 *
950 	 * This call needs to happen before e820__memory_setup() which calls the
951 	 * xen_memory_setup() on Xen dom0 which relies on the fact that those
952 	 * early reservations have happened already.
953 	 */
954 	early_reserve_memory();
955 
956 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
957 	e820__memory_setup();
958 	parse_setup_data();
959 
960 	copy_edd();
961 
962 	setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
963 
964 	/*
965 	 * x86_configure_nx() is called before parse_early_param() to detect
966 	 * whether hardware doesn't support NX (so that the early EHCI debug
967 	 * console setup can safely call set_fixmap()).
968 	 */
969 	x86_configure_nx();
970 
971 	parse_early_param();
972 
973 	if (efi_enabled(EFI_BOOT))
974 		efi_memblock_x86_reserve_range();
975 
976 	x86_report_nx();
977 
978 	apic_setup_apic_calls();
979 
980 	if (acpi_mps_check()) {
981 #ifdef CONFIG_X86_LOCAL_APIC
982 		apic_is_disabled = true;
983 #endif
984 		setup_clear_cpu_cap(X86_FEATURE_APIC);
985 	}
986 
987 	e820__finish_early_params();
988 
989 	if (efi_enabled(EFI_BOOT))
990 		efi_init();
991 
992 	reserve_ibft_region();
993 	x86_init.resources.dmi_setup();
994 
995 	/*
996 	 * VMware detection requires dmi to be available, so this
997 	 * needs to be done after dmi_setup(), for the boot CPU.
998 	 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
999 	 * called before cache_bp_init() for setting up MTRR state.
1000 	 */
1001 	init_hypervisor_platform();
1002 
1003 	tsc_early_init();
1004 	x86_init.resources.probe_roms();
1005 
1006 	/*
1007 	 * Add resources for kernel text and data to the iomem_resource.
1008 	 * Do it after parse_early_param, so it can be debugged.
1009 	 */
1010 	setup_kernel_resources();
1011 
1012 	e820_add_kernel_range();
1013 	trim_bios_range();
1014 #ifdef CONFIG_X86_32
1015 	if (ppro_with_ram_bug()) {
1016 		pr_info("Applying PPro RAM bug workaround: punching 256 kB hole at 1.75 GB physical.\n");
1017 		e820__range_update(0x70000000ULL, SZ_256K, E820_TYPE_RAM, E820_TYPE_RESERVED);
1018 		e820__update_table(e820_table);
1019 	}
1020 #else
1021 	early_gart_iommu_check();
1022 #endif
1023 
1024 	/*
1025 	 * partially used pages are not usable - thus
1026 	 * we are rounding upwards:
1027 	 */
1028 	max_pfn = e820__end_of_ram_pfn();
1029 
1030 	/* update e820 for memory not covered by WB MTRRs */
1031 	cache_bp_init();
1032 	if (mtrr_trim_uncached_memory(max_pfn))
1033 		max_pfn = e820__end_of_ram_pfn();
1034 
1035 	max_possible_pfn = max_pfn;
1036 
1037 	/*
1038 	 * Define random base addresses for memory sections after max_pfn is
1039 	 * defined and before each memory section base is used.
1040 	 */
1041 	kernel_randomize_memory();
1042 
1043 #ifdef CONFIG_X86_32
1044 	/* max_low_pfn get updated here */
1045 	find_low_pfn_range();
1046 #else
1047 	check_x2apic();
1048 
1049 	/* How many end-of-memory variables you have, grandma! */
1050 	/* need this before calling reserve_initrd */
1051 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1052 		max_low_pfn = e820__end_of_low_ram_pfn();
1053 	else
1054 		max_low_pfn = max_pfn;
1055 #endif
1056 
1057 	/* Find and reserve MPTABLE area */
1058 	x86_init.mpparse.find_mptable();
1059 
1060 	early_alloc_pgt_buf();
1061 
1062 	/*
1063 	 * Need to conclude brk, before e820__memblock_setup()
1064 	 * it could use memblock_find_in_range, could overlap with
1065 	 * brk area.
1066 	 */
1067 	reserve_brk();
1068 
1069 	cleanup_highmap();
1070 
1071 	e820__memblock_setup();
1072 
1073 	/*
1074 	 * Needs to run after memblock setup because it needs the physical
1075 	 * memory size.
1076 	 */
1077 	mem_encrypt_setup_arch();
1078 	cc_random_init();
1079 
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 	/*
1103 	 * Find free memory for the real mode trampoline and place it there. If
1104 	 * there is not enough free memory under 1M, on EFI-enabled systems
1105 	 * there will be additional attempt to reclaim the memory for the real
1106 	 * mode trampoline at efi_free_boot_services().
1107 	 *
1108 	 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1109 	 * are known to corrupt low memory and several hundred kilobytes are not
1110 	 * worth complex detection what memory gets clobbered. Windows does the
1111 	 * same thing for very similar reasons.
1112 	 *
1113 	 * Moreover, on machines with SandyBridge graphics or in setups that use
1114 	 * crashkernel the entire 1M is reserved anyway.
1115 	 *
1116 	 * Note the host kernel TDX also requires the first 1MB being reserved.
1117 	 */
1118 	x86_platform.realmode_reserve();
1119 
1120 	init_mem_mapping();
1121 
1122 	/*
1123 	 * init_mem_mapping() relies on the early IDT page fault handling.
1124 	 * Now either enable FRED or install the real page fault handler
1125 	 * for 64-bit in the IDT.
1126 	 */
1127 	cpu_init_replace_early_idt();
1128 
1129 	/*
1130 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1131 	 * with the current CR4 value.  This may not be necessary, but
1132 	 * auditing all the early-boot CR4 manipulation would be needed to
1133 	 * rule it out.
1134 	 *
1135 	 * Mask off features that don't work outside long mode (just
1136 	 * PCIDE for now).
1137 	 */
1138 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1139 
1140 	memblock_set_current_limit(get_max_mapped());
1141 
1142 	/*
1143 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1144 	 */
1145 
1146 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1147 	if (init_ohci1394_dma_early)
1148 		init_ohci1394_dma_on_all_controllers();
1149 #endif
1150 	/* Allocate bigger log buffer */
1151 	setup_log_buf(1);
1152 
1153 	if (efi_enabled(EFI_BOOT)) {
1154 		switch (boot_params.secure_boot) {
1155 		case efi_secureboot_mode_disabled:
1156 			pr_info("Secure boot disabled\n");
1157 			break;
1158 		case efi_secureboot_mode_enabled:
1159 			pr_info("Secure boot enabled\n");
1160 			break;
1161 		default:
1162 			pr_info("Secure boot could not be determined\n");
1163 			break;
1164 		}
1165 	}
1166 
1167 	reserve_initrd();
1168 
1169 	acpi_table_upgrade();
1170 	/* Look for ACPI tables and reserve memory occupied by them. */
1171 	acpi_boot_table_init();
1172 
1173 	vsmp_init();
1174 
1175 	io_delay_init();
1176 
1177 	early_platform_quirks();
1178 
1179 	/* Some platforms need the APIC registered for NUMA configuration */
1180 	early_acpi_boot_init();
1181 	x86_init.mpparse.early_parse_smp_cfg();
1182 
1183 	x86_flattree_get_config();
1184 
1185 	initmem_init();
1186 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1187 
1188 	if (boot_cpu_has(X86_FEATURE_GBPAGES)) {
1189 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1190 		hugetlb_bootmem_alloc();
1191 	}
1192 
1193 	/*
1194 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1195 	 * won't consume hotpluggable memory.
1196 	 */
1197 	arch_reserve_crashkernel();
1198 
1199 	if (!early_xdbc_setup_hardware())
1200 		early_xdbc_register_console();
1201 
1202 	x86_init.paging.pagetable_init();
1203 
1204 	kasan_init();
1205 
1206 	/*
1207 	 * Sync back kernel address range.
1208 	 *
1209 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1210 	 * this call?
1211 	 */
1212 	sync_initial_page_table();
1213 
1214 	tboot_probe();
1215 
1216 	map_vsyscall();
1217 
1218 	x86_32_probe_apic();
1219 
1220 	early_quirks();
1221 
1222 	topology_apply_cmdline_limits_early();
1223 
1224 	/*
1225 	 * Parse SMP configuration. Try ACPI first and then the platform
1226 	 * specific parser.
1227 	 */
1228 	acpi_boot_init();
1229 	x86_init.mpparse.parse_smp_cfg();
1230 
1231 	/* Last opportunity to detect and map the local APIC */
1232 	init_apic_mappings();
1233 
1234 	topology_init_possible_cpus();
1235 
1236 	init_cpu_to_node();
1237 	init_gi_nodes();
1238 
1239 	io_apic_init_mappings();
1240 
1241 	x86_init.hyper.guest_late_init();
1242 
1243 	e820__reserve_resources();
1244 	e820__register_nosave_regions(max_pfn);
1245 
1246 	x86_init.resources.reserve_resources();
1247 
1248 	e820__setup_pci_gap();
1249 
1250 #ifdef CONFIG_VT
1251 #if defined(CONFIG_VGA_CONSOLE)
1252 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1253 		vgacon_register_screen(&sysfb_primary_display.screen);
1254 #endif
1255 #endif
1256 	x86_init.oem.banner();
1257 
1258 	x86_init.timers.wallclock_init();
1259 
1260 	/*
1261 	 * This needs to run before setup_local_APIC() which soft-disables the
1262 	 * local APIC temporarily and that masks the thermal LVT interrupt,
1263 	 * leading to softlockups on machines which have configured SMI
1264 	 * interrupt delivery.
1265 	 */
1266 	therm_lvt_init();
1267 
1268 	mcheck_init();
1269 
1270 	register_refined_jiffies(CLOCK_TICK_RATE);
1271 
1272 #ifdef CONFIG_EFI
1273 	if (efi_enabled(EFI_BOOT))
1274 		efi_apply_memmap_quirks();
1275 #endif
1276 
1277 	unwind_init();
1278 }
1279 
1280 #ifdef CONFIG_X86_32
1281 
1282 static struct resource video_ram_resource = {
1283 	.name	= "Video RAM area",
1284 	.start	= 0xa0000,
1285 	.end	= 0xbffff,
1286 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1287 };
1288 
1289 void __init i386_reserve_resources(void)
1290 {
1291 	request_resource(&iomem_resource, &video_ram_resource);
1292 	reserve_standard_io_resources();
1293 }
1294 
1295 #endif /* CONFIG_X86_32 */
1296 
1297 static struct notifier_block kernel_offset_notifier = {
1298 	.notifier_call = dump_kernel_offset
1299 };
1300 
1301 static int __init register_kernel_offset_dumper(void)
1302 {
1303 	atomic_notifier_chain_register(&panic_notifier_list,
1304 					&kernel_offset_notifier);
1305 	return 0;
1306 }
1307 __initcall(register_kernel_offset_dumper);
1308 
1309 #ifdef CONFIG_HOTPLUG_CPU
1310 bool arch_cpu_is_hotpluggable(int cpu)
1311 {
1312 	return cpu > 0;
1313 }
1314 #endif /* CONFIG_HOTPLUG_CPU */
1315