xref: /linux/arch/x86/kernel/setup.c (revision f4cdf7ca9a1fdcca413157df19753f388a5a224e)
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 IS_ENABLED(CONFIG_X86_SPEEDSTEP_SMI)
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 SYM_PIC_ALIAS(boot_cpu_data);
135 
136 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
137 __visible unsigned long mmu_cr4_features __ro_after_init;
138 #else
139 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
140 #endif
141 
142 #ifdef CONFIG_IMA
143 static phys_addr_t ima_kexec_buffer_phys;
144 static size_t ima_kexec_buffer_size;
145 #endif
146 
147 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
148 int bootloader_type, bootloader_version;
149 
150 static const struct ctl_table x86_sysctl_table[] = {
151 	{
152 		.procname       = "unknown_nmi_panic",
153 		.data           = &unknown_nmi_panic,
154 		.maxlen         = sizeof(int),
155 		.mode           = 0644,
156 		.proc_handler   = proc_dointvec,
157 	},
158 	{
159 		.procname	= "panic_on_unrecovered_nmi",
160 		.data		= &panic_on_unrecovered_nmi,
161 		.maxlen		= sizeof(int),
162 		.mode		= 0644,
163 		.proc_handler	= proc_dointvec,
164 	},
165 	{
166 		.procname	= "panic_on_io_nmi",
167 		.data		= &panic_on_io_nmi,
168 		.maxlen		= sizeof(int),
169 		.mode		= 0644,
170 		.proc_handler	= proc_dointvec,
171 	},
172 	{
173 		.procname	= "bootloader_type",
174 		.data		= &bootloader_type,
175 		.maxlen		= sizeof(int),
176 		.mode		= 0444,
177 		.proc_handler	= proc_dointvec,
178 	},
179 	{
180 		.procname	= "bootloader_version",
181 		.data		= &bootloader_version,
182 		.maxlen		= sizeof(int),
183 		.mode		= 0444,
184 		.proc_handler	= proc_dointvec,
185 	},
186 	{
187 		.procname	= "io_delay_type",
188 		.data		= &io_delay_type,
189 		.maxlen		= sizeof(int),
190 		.mode		= 0644,
191 		.proc_handler	= proc_dointvec,
192 	},
193 #if defined(CONFIG_ACPI_SLEEP)
194 	{
195 		.procname	= "acpi_video_flags",
196 		.data		= &acpi_realmode_flags,
197 		.maxlen		= sizeof(unsigned long),
198 		.mode		= 0644,
199 		.proc_handler	= proc_doulongvec_minmax,
200 	},
201 #endif
202 };
203 
204 static int __init init_x86_sysctl(void)
205 {
206 	register_sysctl_init("kernel", x86_sysctl_table);
207 	return 0;
208 }
209 arch_initcall(init_x86_sysctl);
210 
211 /*
212  * Setup options
213  */
214 
215 struct sysfb_display_info sysfb_primary_display;
216 EXPORT_SYMBOL(sysfb_primary_display);
217 
218 extern int root_mountflags;
219 
220 unsigned long saved_video_mode;
221 
222 #define RAMDISK_IMAGE_START_MASK	0x07FF
223 #define RAMDISK_PROMPT_FLAG		0x8000
224 #define RAMDISK_LOAD_FLAG		0x4000
225 
226 static char __initdata command_line[COMMAND_LINE_SIZE];
227 #ifdef CONFIG_CMDLINE_BOOL
228 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
229 bool builtin_cmdline_added __ro_after_init;
230 #endif
231 
232 #if IS_ENABLED(CONFIG_EDD)
233 struct edd edd;
234 #ifdef CONFIG_EDD_MODULE
235 EXPORT_SYMBOL(edd);
236 #endif
237 /**
238  * copy_edd() - Copy the BIOS EDD information
239  *              from boot_params into a safe place.
240  *
241  */
242 static inline void __init copy_edd(void)
243 {
244      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
245 	    sizeof(edd.mbr_signature));
246      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
247      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
248      edd.edd_info_nr = boot_params.eddbuf_entries;
249 }
250 #else
251 static inline void __init copy_edd(void)
252 {
253 }
254 #endif
255 
256 void * __init extend_brk(size_t size, size_t align)
257 {
258 	size_t mask = align - 1;
259 	void *ret;
260 
261 	BUG_ON(_brk_start == 0);
262 	BUG_ON(align & mask);
263 
264 	_brk_end = (_brk_end + mask) & ~mask;
265 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
266 
267 	ret = (void *)_brk_end;
268 	_brk_end += size;
269 
270 	memset(ret, 0, size);
271 
272 	return ret;
273 }
274 
275 #ifdef CONFIG_X86_32
276 static void __init cleanup_highmap(void)
277 {
278 }
279 #endif
280 
281 static void __init reserve_brk(void)
282 {
283 	if (_brk_end > _brk_start)
284 		memblock_reserve_kern(__pa_symbol(_brk_start),
285 				      _brk_end - _brk_start);
286 
287 	/* Mark brk area as locked down and no longer taking any
288 	   new allocations */
289 	_brk_start = 0;
290 }
291 
292 #ifdef CONFIG_BLK_DEV_INITRD
293 
294 static u64 __init get_ramdisk_image(void)
295 {
296 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
297 
298 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
299 
300 	if (ramdisk_image == 0)
301 		ramdisk_image = phys_initrd_start;
302 
303 	return ramdisk_image;
304 }
305 static u64 __init get_ramdisk_size(void)
306 {
307 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
308 
309 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
310 
311 	if (ramdisk_size == 0)
312 		ramdisk_size = phys_initrd_size;
313 
314 	return ramdisk_size;
315 }
316 
317 static void __init relocate_initrd(void)
318 {
319 	/* Assume only end is not page aligned */
320 	u64 ramdisk_image = get_ramdisk_image();
321 	u64 ramdisk_size  = get_ramdisk_size();
322 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
323 	int ret = 0;
324 
325 	/* We need to move the initrd down into directly mapped mem */
326 	u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
327 						      PFN_PHYS(max_pfn_mapped));
328 	if (!relocated_ramdisk)
329 		panic("Cannot find place for new RAMDISK of size %lld\n",
330 		      ramdisk_size);
331 
332 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
333 	initrd_end   = initrd_start + ramdisk_size;
334 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
335 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
336 
337 	ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
338 	if (ret)
339 		panic("Copy RAMDISK failed\n");
340 
341 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
342 		" [mem %#010llx-%#010llx]\n",
343 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
344 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
345 }
346 
347 static void __init early_reserve_initrd(void)
348 {
349 	/* Assume only end is not page aligned */
350 	u64 ramdisk_image = get_ramdisk_image();
351 	u64 ramdisk_size  = get_ramdisk_size();
352 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
353 
354 	if (!boot_params.hdr.type_of_loader ||
355 	    !ramdisk_image || !ramdisk_size)
356 		return;		/* No initrd provided by bootloader */
357 
358 	memblock_reserve_kern(ramdisk_image, ramdisk_end - ramdisk_image);
359 }
360 
361 static void __init reserve_initrd(void)
362 {
363 	/* Assume only end is not page aligned */
364 	u64 ramdisk_image = get_ramdisk_image();
365 	u64 ramdisk_size  = get_ramdisk_size();
366 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
367 
368 	if (!boot_params.hdr.type_of_loader ||
369 	    !ramdisk_image || !ramdisk_size)
370 		return;		/* No initrd provided by bootloader */
371 
372 	initrd_start = 0;
373 
374 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
375 			ramdisk_end - 1);
376 
377 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
378 				PFN_DOWN(ramdisk_end))) {
379 		/* All are mapped, easy case */
380 		initrd_start = ramdisk_image + PAGE_OFFSET;
381 		initrd_end = initrd_start + ramdisk_size;
382 		return;
383 	}
384 
385 	relocate_initrd();
386 
387 	memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
388 }
389 
390 #else
391 static void __init early_reserve_initrd(void)
392 {
393 }
394 static void __init reserve_initrd(void)
395 {
396 }
397 #endif /* CONFIG_BLK_DEV_INITRD */
398 
399 static void __init add_early_ima_buffer(u64 phys_addr)
400 {
401 #ifdef CONFIG_IMA
402 	struct ima_setup_data *data;
403 
404 	data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
405 	if (!data) {
406 		pr_warn("setup: failed to memremap ima_setup_data entry\n");
407 		return;
408 	}
409 
410 	if (data->size) {
411 		memblock_reserve_kern(data->addr, data->size);
412 		ima_kexec_buffer_phys = data->addr;
413 		ima_kexec_buffer_size = data->size;
414 	}
415 
416 	early_memunmap(data, sizeof(*data));
417 #else
418 	pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
419 #endif
420 }
421 
422 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
423 int __init ima_free_kexec_buffer(void)
424 {
425 	if (!ima_kexec_buffer_size)
426 		return -ENOENT;
427 
428 	memblock_phys_free(ima_kexec_buffer_phys,
429 			   ima_kexec_buffer_size);
430 
431 	ima_kexec_buffer_phys = 0;
432 	ima_kexec_buffer_size = 0;
433 
434 	return 0;
435 }
436 
437 int __init ima_get_kexec_buffer(void **addr, size_t *size)
438 {
439 	int ret;
440 
441 	if (!ima_kexec_buffer_size)
442 		return -ENOENT;
443 
444 	ret = ima_validate_range(ima_kexec_buffer_phys, ima_kexec_buffer_size);
445 	if (ret)
446 		return ret;
447 
448 	*addr = __va(ima_kexec_buffer_phys);
449 	*size = ima_kexec_buffer_size;
450 
451 	return 0;
452 }
453 #endif
454 
455 static void __init add_kho(u64 phys_addr, u32 data_len)
456 {
457 	struct kho_data *kho;
458 	u64 addr = phys_addr + sizeof(struct setup_data);
459 	u64 size = data_len - sizeof(struct setup_data);
460 
461 	if (!IS_ENABLED(CONFIG_KEXEC_HANDOVER)) {
462 		pr_warn("Passed KHO data, but CONFIG_KEXEC_HANDOVER not set. Ignoring.\n");
463 		return;
464 	}
465 
466 	kho = early_memremap(addr, size);
467 	if (!kho) {
468 		pr_warn("setup: failed to memremap kho data (0x%llx, 0x%llx)\n",
469 			addr, size);
470 		return;
471 	}
472 
473 	kho_populate(kho->fdt_addr, kho->fdt_size, kho->scratch_addr, kho->scratch_size);
474 
475 	early_memunmap(kho, size);
476 }
477 
478 static void __init parse_setup_data(void)
479 {
480 	struct setup_data *data;
481 	u64 pa_data, pa_next;
482 
483 	pa_data = boot_params.hdr.setup_data;
484 	while (pa_data) {
485 		u32 data_len, data_type;
486 
487 		data = early_memremap(pa_data, sizeof(*data));
488 		data_len = data->len + sizeof(struct setup_data);
489 		data_type = data->type;
490 		pa_next = data->next;
491 		early_memunmap(data, sizeof(*data));
492 
493 		switch (data_type) {
494 		case SETUP_E820_EXT:
495 			e820__memory_setup_extended(pa_data, data_len);
496 			break;
497 		case SETUP_DTB:
498 			add_dtb(pa_data);
499 			break;
500 		case SETUP_EFI:
501 			parse_efi_setup(pa_data, data_len);
502 			break;
503 		case SETUP_IMA:
504 			add_early_ima_buffer(pa_data);
505 			break;
506 		case SETUP_KEXEC_KHO:
507 			add_kho(pa_data, data_len);
508 			break;
509 		case SETUP_RNG_SEED:
510 			data = early_memremap(pa_data, data_len);
511 			add_bootloader_randomness(data->data, data->len);
512 			/* Zero seed for forward secrecy. */
513 			memzero_explicit(data->data, data->len);
514 			/* Zero length in case we find ourselves back here by accident. */
515 			memzero_explicit(&data->len, sizeof(data->len));
516 			early_memunmap(data, data_len);
517 			break;
518 		default:
519 			break;
520 		}
521 		pa_data = pa_next;
522 	}
523 }
524 
525 /*
526  * Translate the fields of 'struct boot_param' into global variables
527  * representing these parameters.
528  */
529 static void __init parse_boot_params(void)
530 {
531 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
532 	sysfb_primary_display.screen = boot_params.screen_info;
533 #if defined(CONFIG_FIRMWARE_EDID)
534 	sysfb_primary_display.edid = boot_params.edid_info;
535 #endif
536 #ifdef CONFIG_X86_32
537 	apm_info.bios = boot_params.apm_bios_info;
538 	ist_info = boot_params.ist_info;
539 #endif
540 	saved_video_mode = boot_params.hdr.vid_mode;
541 	bootloader_type = boot_params.hdr.type_of_loader;
542 	if ((bootloader_type >> 4) == 0xe) {
543 		bootloader_type &= 0xf;
544 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
545 	}
546 	bootloader_version  = bootloader_type & 0xf;
547 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
548 
549 #ifdef CONFIG_BLK_DEV_RAM
550 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
551 #endif
552 #ifdef CONFIG_EFI
553 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
554 		     EFI32_LOADER_SIGNATURE, 4)) {
555 		set_bit(EFI_BOOT, &efi.flags);
556 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
557 		     EFI64_LOADER_SIGNATURE, 4)) {
558 		set_bit(EFI_BOOT, &efi.flags);
559 		set_bit(EFI_64BIT, &efi.flags);
560 	}
561 #endif
562 
563 	if (!boot_params.hdr.root_flags)
564 		root_mountflags &= ~MS_RDONLY;
565 }
566 
567 static void __init memblock_x86_reserve_range_setup_data(void)
568 {
569 	struct setup_indirect *indirect;
570 	struct setup_data *data;
571 	u64 pa_data, pa_next;
572 	u32 len;
573 
574 	pa_data = boot_params.hdr.setup_data;
575 	while (pa_data) {
576 		data = early_memremap(pa_data, sizeof(*data));
577 		if (!data) {
578 			pr_warn("setup: failed to memremap setup_data entry\n");
579 			return;
580 		}
581 
582 		len = sizeof(*data);
583 		pa_next = data->next;
584 
585 		memblock_reserve_kern(pa_data, sizeof(*data) + data->len);
586 
587 		if (data->type == SETUP_INDIRECT) {
588 			len += data->len;
589 			early_memunmap(data, sizeof(*data));
590 			data = early_memremap(pa_data, len);
591 			if (!data) {
592 				pr_warn("setup: failed to memremap indirect setup_data\n");
593 				return;
594 			}
595 
596 			indirect = (struct setup_indirect *)data->data;
597 
598 			if (indirect->type != SETUP_INDIRECT)
599 				memblock_reserve_kern(indirect->addr, indirect->len);
600 		}
601 
602 		pa_data = pa_next;
603 		early_memunmap(data, len);
604 	}
605 }
606 
607 static void __init arch_reserve_crashkernel(void)
608 {
609 	unsigned long long crash_base, crash_size, low_size = 0, cma_size = 0;
610 	bool high = false;
611 	int ret;
612 
613 	if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
614 		return;
615 
616 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
617 				&crash_size, &crash_base,
618 				&low_size, &cma_size, &high);
619 	if (ret)
620 		return;
621 
622 	if (xen_pv_domain()) {
623 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
624 		return;
625 	}
626 
627 	reserve_crashkernel_generic(crash_size, crash_base, low_size, high);
628 	reserve_crashkernel_cma(cma_size);
629 }
630 
631 static struct resource standard_io_resources[] = {
632 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
633 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
634 	{ .name = "pic1", .start = 0x20, .end = 0x21,
635 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
636 	{ .name = "timer0", .start = 0x40, .end = 0x43,
637 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
638 	{ .name = "timer1", .start = 0x50, .end = 0x53,
639 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
640 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
641 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
642 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
643 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
644 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
645 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
646 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
647 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
648 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
649 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
650 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
651 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
652 };
653 
654 void __init reserve_standard_io_resources(void)
655 {
656 	int i;
657 
658 	/* request I/O space for devices used on all i[345]86 PCs */
659 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
660 		request_resource(&ioport_resource, &standard_io_resources[i]);
661 
662 }
663 
664 static void __init setup_kernel_resources(void)
665 {
666 	code_resource.start = __pa_symbol(_text);
667 	code_resource.end = __pa_symbol(_etext)-1;
668 	rodata_resource.start = __pa_symbol(__start_rodata);
669 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
670 	data_resource.start = __pa_symbol(_sdata);
671 	data_resource.end = __pa_symbol(_edata)-1;
672 	bss_resource.start = __pa_symbol(__bss_start);
673 	bss_resource.end = __pa_symbol(__bss_stop)-1;
674 
675 	insert_resource(&iomem_resource, &code_resource);
676 	insert_resource(&iomem_resource, &rodata_resource);
677 	insert_resource(&iomem_resource, &data_resource);
678 	insert_resource(&iomem_resource, &bss_resource);
679 }
680 
681 static bool __init snb_gfx_workaround_needed(void)
682 {
683 #ifdef CONFIG_PCI
684 	int i;
685 	u16 vendor, devid;
686 	static const __initconst u16 snb_ids[] = {
687 		0x0102,
688 		0x0112,
689 		0x0122,
690 		0x0106,
691 		0x0116,
692 		0x0126,
693 		0x010a,
694 	};
695 
696 	/* Assume no if something weird is going on with PCI */
697 	if (!early_pci_allowed())
698 		return false;
699 
700 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
701 	if (vendor != 0x8086)
702 		return false;
703 
704 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
705 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
706 		if (devid == snb_ids[i])
707 			return true;
708 #endif
709 
710 	return false;
711 }
712 
713 /*
714  * Sandy Bridge graphics has trouble with certain ranges, exclude
715  * them from allocation.
716  */
717 static void __init trim_snb_memory(void)
718 {
719 	static const __initconst unsigned long bad_pages[] = {
720 		0x20050000,
721 		0x20110000,
722 		0x20130000,
723 		0x20138000,
724 		0x40004000,
725 	};
726 	int i;
727 
728 	if (!snb_gfx_workaround_needed())
729 		return;
730 
731 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
732 
733 	/*
734 	 * SandyBridge integrated graphics devices have a bug that prevents
735 	 * them from accessing certain memory ranges, namely anything below
736 	 * 1M and in the pages listed in bad_pages[] above.
737 	 *
738 	 * To avoid these pages being ever accessed by SNB gfx devices reserve
739 	 * bad_pages that have not already been reserved at boot time.
740 	 * All memory below the 1 MB mark is anyway reserved later during
741 	 * setup_arch(), so there is no need to reserve it here.
742 	 */
743 
744 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
745 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
746 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
747 			       bad_pages[i]);
748 	}
749 }
750 
751 static void __init trim_bios_range(void)
752 {
753 	/*
754 	 * A special case is the first 4Kb of memory;
755 	 * This is a BIOS owned area, not kernel ram, but generally
756 	 * not listed as such in the E820 table.
757 	 *
758 	 * This typically reserves additional memory (64KiB by default)
759 	 * since some BIOSes are known to corrupt low memory.  See the
760 	 * Kconfig help text for X86_RESERVE_LOW.
761 	 */
762 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
763 
764 	/*
765 	 * special case: Some BIOSes report the PC BIOS
766 	 * area (640Kb -> 1Mb) as RAM even though it is not.
767 	 * take them out.
768 	 */
769 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM);
770 
771 	e820__update_table(e820_table);
772 }
773 
774 /* called before trim_bios_range() to spare extra sanitize */
775 static void __init e820_add_kernel_range(void)
776 {
777 	u64 start = __pa_symbol(_text);
778 	u64 size = __pa_symbol(_end) - start;
779 
780 	/*
781 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
782 	 * attempt to fix it by adding the range. We may have a confused BIOS,
783 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
784 	 * exclude kernel range. If we really are running on top non-RAM,
785 	 * we will crash later anyways.
786 	 */
787 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
788 		return;
789 
790 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
791 	e820__range_remove(start, size, 0);
792 	e820__range_add(start, size, E820_TYPE_RAM);
793 }
794 
795 static void __init early_reserve_memory(void)
796 {
797 	/*
798 	 * Reserve the memory occupied by the kernel between _text and
799 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
800 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
801 	 * separate memblock_reserve() or they will be discarded.
802 	 */
803 	memblock_reserve_kern(__pa_symbol(_text),
804 			      (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
805 
806 	/*
807 	 * The first 4Kb of memory is a BIOS owned area, but generally it is
808 	 * not listed as such in the E820 table.
809 	 *
810 	 * Reserve the first 64K of memory since some BIOSes are known to
811 	 * corrupt low memory. After the real mode trampoline is allocated the
812 	 * rest of the memory below 640k is reserved.
813 	 *
814 	 * In addition, make sure page 0 is always reserved because on
815 	 * systems with L1TF its contents can be leaked to user processes.
816 	 */
817 	memblock_reserve(0, SZ_64K);
818 
819 	early_reserve_initrd();
820 
821 	memblock_x86_reserve_range_setup_data();
822 
823 	reserve_bios_regions();
824 	trim_snb_memory();
825 }
826 
827 /*
828  * Dump out kernel offset information on panic.
829  */
830 static int
831 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
832 {
833 	if (kaslr_enabled()) {
834 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
835 			 kaslr_offset(),
836 			 __START_KERNEL,
837 			 __START_KERNEL_map,
838 			 MODULES_VADDR-1);
839 	} else {
840 		pr_emerg("Kernel Offset: disabled\n");
841 	}
842 
843 	return 0;
844 }
845 
846 void x86_configure_nx(void)
847 {
848 	if (boot_cpu_has(X86_FEATURE_NX))
849 		__supported_pte_mask |= _PAGE_NX;
850 	else
851 		__supported_pte_mask &= ~_PAGE_NX;
852 }
853 
854 static void __init x86_report_nx(void)
855 {
856 	if (!boot_cpu_has(X86_FEATURE_NX)) {
857 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
858 		       "missing in CPU!\n");
859 	} else {
860 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
861 		printk(KERN_INFO "NX (Execute Disable) protection: active\n");
862 #else
863 		/* 32bit non-PAE kernel, NX cannot be used */
864 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
865 		       "cannot be enabled: non-PAE kernel!\n");
866 #endif
867 	}
868 }
869 
870 /*
871  * Determine if we were loaded by an EFI loader.  If so, then we have also been
872  * passed the efi memmap, systab, etc., so we should use these data structures
873  * for initialization.  Note, the efi init code path is determined by the
874  * global efi_enabled. This allows the same kernel image to be used on existing
875  * systems (with a traditional BIOS) as well as on EFI systems.
876  */
877 /*
878  * setup_arch - architecture-specific boot-time initializations
879  *
880  * Note: On x86_64, fixmaps are ready for use even before this is called.
881  */
882 
883 void __init setup_arch(char **cmdline_p)
884 {
885 #ifdef CONFIG_X86_32
886 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
887 
888 	/*
889 	 * copy kernel address range established so far and switch
890 	 * to the proper swapper page table
891 	 */
892 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
893 			initial_page_table + KERNEL_PGD_BOUNDARY,
894 			KERNEL_PGD_PTRS);
895 
896 	load_cr3(swapper_pg_dir);
897 	/*
898 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
899 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
900 	 * will not flush anything because the CPU quirk which clears
901 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
902 	 * load_cr3() above the TLB has been flushed already. The
903 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
904 	 * so proper operation is guaranteed.
905 	 */
906 	__flush_tlb_all();
907 #else
908 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
909 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
910 #endif
911 
912 #ifdef CONFIG_CMDLINE_BOOL
913 #ifdef CONFIG_CMDLINE_OVERRIDE
914 	strscpy(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 		strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
921 	}
922 #endif
923 	builtin_cmdline_added = true;
924 #endif
925 
926 	strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
927 	*cmdline_p = command_line;
928 
929 	/*
930 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
931 	 * reserve_top(), so do this before touching the ioremap area.
932 	 */
933 	olpc_ofw_detect();
934 
935 	idt_setup_early_traps();
936 	early_cpu_init();
937 	jump_label_init();
938 	static_call_init();
939 	early_ioremap_init();
940 
941 	setup_olpc_ofw_pgd();
942 
943 	parse_boot_params();
944 
945 	x86_init.oem.arch_setup();
946 
947 	/*
948 	 * Do some memory reservations *before* memory is added to memblock, so
949 	 * memblock allocations won't overwrite it.
950 	 *
951 	 * After this point, everything still needed from the boot loader or
952 	 * firmware or kernel text should be early reserved or marked not RAM in
953 	 * e820. All other memory is free game.
954 	 *
955 	 * This call needs to happen before e820__memory_setup() which calls the
956 	 * xen_memory_setup() on Xen dom0 which relies on the fact that those
957 	 * early reservations have happened already.
958 	 */
959 	early_reserve_memory();
960 
961 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
962 	e820__memory_setup();
963 	parse_setup_data();
964 
965 	copy_edd();
966 
967 	setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
968 
969 	/*
970 	 * x86_configure_nx() is called before parse_early_param() to detect
971 	 * whether hardware doesn't support NX (so that the early EHCI debug
972 	 * console setup can safely call set_fixmap()).
973 	 */
974 	x86_configure_nx();
975 
976 	parse_early_param();
977 
978 	if (efi_enabled(EFI_BOOT))
979 		efi_memblock_x86_reserve_range();
980 
981 	x86_report_nx();
982 
983 	apic_setup_apic_calls();
984 
985 	if (acpi_mps_check()) {
986 #ifdef CONFIG_X86_LOCAL_APIC
987 		apic_is_disabled = true;
988 #endif
989 		setup_clear_cpu_cap(X86_FEATURE_APIC);
990 	}
991 
992 	e820__finish_early_params();
993 
994 	if (efi_enabled(EFI_BOOT))
995 		efi_init();
996 
997 	reserve_ibft_region();
998 	x86_init.resources.dmi_setup();
999 
1000 	/*
1001 	 * VMware detection requires dmi to be available, so this
1002 	 * needs to be done after dmi_setup(), for the boot CPU.
1003 	 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
1004 	 * called before cache_bp_init() for setting up MTRR state.
1005 	 */
1006 	init_hypervisor_platform();
1007 
1008 	tsc_early_init();
1009 	x86_init.resources.probe_roms();
1010 
1011 	/*
1012 	 * Add resources for kernel text and data to the iomem_resource.
1013 	 * Do it after parse_early_param, so it can be debugged.
1014 	 */
1015 	setup_kernel_resources();
1016 
1017 	e820_add_kernel_range();
1018 	trim_bios_range();
1019 #ifdef CONFIG_X86_32
1020 	if (ppro_with_ram_bug()) {
1021 		pr_info("Applying PPro RAM bug workaround: punching 256 kB hole at 1.75 GB physical.\n");
1022 		e820__range_update(0x70000000ULL, SZ_256K, E820_TYPE_RAM, E820_TYPE_RESERVED);
1023 		e820__update_table(e820_table);
1024 	}
1025 #else
1026 	early_gart_iommu_check();
1027 #endif
1028 
1029 	/*
1030 	 * partially used pages are not usable - thus
1031 	 * we are rounding upwards:
1032 	 */
1033 	max_pfn = e820__end_of_ram_pfn();
1034 
1035 	/* update e820 for memory not covered by WB MTRRs */
1036 	cache_bp_init();
1037 	if (mtrr_trim_uncached_memory(max_pfn))
1038 		max_pfn = e820__end_of_ram_pfn();
1039 
1040 	max_possible_pfn = max_pfn;
1041 
1042 	/*
1043 	 * Define random base addresses for memory sections after max_pfn is
1044 	 * defined and before each memory section base is used.
1045 	 */
1046 	kernel_randomize_memory();
1047 
1048 #ifdef CONFIG_X86_32
1049 	/* max_low_pfn get updated here */
1050 	find_low_pfn_range();
1051 #else
1052 	check_x2apic();
1053 
1054 	/* How many end-of-memory variables you have, grandma! */
1055 	/* need this before calling reserve_initrd */
1056 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1057 		max_low_pfn = e820__end_of_low_ram_pfn();
1058 	else
1059 		max_low_pfn = max_pfn;
1060 #endif
1061 
1062 	/* Find and reserve MPTABLE area */
1063 	x86_init.mpparse.find_mptable();
1064 
1065 	early_alloc_pgt_buf();
1066 
1067 	/*
1068 	 * Need to conclude brk, before e820__memblock_setup()
1069 	 * it could use memblock_find_in_range, could overlap with
1070 	 * brk area.
1071 	 */
1072 	reserve_brk();
1073 
1074 	cleanup_highmap();
1075 
1076 	e820__memblock_setup();
1077 
1078 	/*
1079 	 * Needs to run after memblock setup because it needs the physical
1080 	 * memory size.
1081 	 */
1082 	mem_encrypt_setup_arch();
1083 	cc_random_init();
1084 
1085 	efi_find_mirror();
1086 	efi_esrt_init();
1087 	efi_mokvar_table_init();
1088 
1089 	/*
1090 	 * The EFI specification says that boot service code won't be
1091 	 * called after ExitBootServices(). This is, in fact, a lie.
1092 	 */
1093 	efi_reserve_boot_services();
1094 
1095 	/* preallocate 4k for mptable mpc */
1096 	e820__memblock_alloc_reserved_mpc_new();
1097 
1098 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1099 	setup_bios_corruption_check();
1100 #endif
1101 
1102 #ifdef CONFIG_X86_32
1103 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1104 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1105 #endif
1106 
1107 	/*
1108 	 * Find free memory for the real mode trampoline and place it there. If
1109 	 * there is not enough free memory under 1M, on EFI-enabled systems
1110 	 * there will be additional attempt to reclaim the memory for the real
1111 	 * mode trampoline at efi_free_boot_services().
1112 	 *
1113 	 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1114 	 * are known to corrupt low memory and several hundred kilobytes are not
1115 	 * worth complex detection what memory gets clobbered. Windows does the
1116 	 * same thing for very similar reasons.
1117 	 *
1118 	 * Moreover, on machines with SandyBridge graphics or in setups that use
1119 	 * crashkernel the entire 1M is reserved anyway.
1120 	 *
1121 	 * Note the host kernel TDX also requires the first 1MB being reserved.
1122 	 */
1123 	x86_platform.realmode_reserve();
1124 
1125 	init_mem_mapping();
1126 
1127 	/*
1128 	 * init_mem_mapping() relies on the early IDT page fault handling.
1129 	 * Now either enable FRED or install the real page fault handler
1130 	 * for 64-bit in the IDT.
1131 	 */
1132 	cpu_init_replace_early_idt();
1133 
1134 	/*
1135 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1136 	 * with the current CR4 value.  This may not be necessary, but
1137 	 * auditing all the early-boot CR4 manipulation would be needed to
1138 	 * rule it out.
1139 	 *
1140 	 * Mask off features that don't work outside long mode (just
1141 	 * PCIDE for now).
1142 	 */
1143 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1144 
1145 	memblock_set_current_limit(get_max_mapped());
1146 
1147 	/*
1148 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1149 	 */
1150 
1151 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1152 	if (init_ohci1394_dma_early)
1153 		init_ohci1394_dma_on_all_controllers();
1154 #endif
1155 	/* Allocate bigger log buffer */
1156 	setup_log_buf(1);
1157 
1158 	if (efi_enabled(EFI_BOOT)) {
1159 		switch (boot_params.secure_boot) {
1160 		case efi_secureboot_mode_disabled:
1161 			pr_info("Secure boot disabled\n");
1162 			break;
1163 		case efi_secureboot_mode_enabled:
1164 			pr_info("Secure boot enabled\n");
1165 			break;
1166 		default:
1167 			pr_info("Secure boot could not be determined\n");
1168 			break;
1169 		}
1170 	}
1171 
1172 	reserve_initrd();
1173 
1174 	acpi_table_upgrade();
1175 	/* Look for ACPI tables and reserve memory occupied by them. */
1176 	acpi_boot_table_init();
1177 
1178 	vsmp_init();
1179 
1180 	io_delay_init();
1181 
1182 	early_platform_quirks();
1183 
1184 	/* Some platforms need the APIC registered for NUMA configuration */
1185 	early_acpi_boot_init();
1186 	x86_init.mpparse.early_parse_smp_cfg();
1187 
1188 	x86_flattree_get_config();
1189 
1190 	initmem_init();
1191 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
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(PIT_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