xref: /linux/arch/x86/kernel/setup.c (revision a97756cb)
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  */
copy_edd(void)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
copy_edd(void)190 static inline void __init copy_edd(void)
191 {
192 }
193 #endif
194 
extend_brk(size_t size,size_t align)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
cleanup_highmap(void)215 static void __init cleanup_highmap(void)
216 {
217 }
218 #endif
219 
reserve_brk(void)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 
get_ramdisk_image(void)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 }
get_ramdisk_size(void)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 
relocate_initrd(void)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 
263 	/* We need to move the initrd down into directly mapped mem */
264 	u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
265 						      PFN_PHYS(max_pfn_mapped));
266 	if (!relocated_ramdisk)
267 		panic("Cannot find place for new RAMDISK of size %lld\n",
268 		      ramdisk_size);
269 
270 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
271 	initrd_end   = initrd_start + ramdisk_size;
272 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
273 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
274 
275 	copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
276 
277 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
278 		" [mem %#010llx-%#010llx]\n",
279 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
280 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
281 }
282 
early_reserve_initrd(void)283 static void __init early_reserve_initrd(void)
284 {
285 	/* Assume only end is not page aligned */
286 	u64 ramdisk_image = get_ramdisk_image();
287 	u64 ramdisk_size  = get_ramdisk_size();
288 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
289 
290 	if (!boot_params.hdr.type_of_loader ||
291 	    !ramdisk_image || !ramdisk_size)
292 		return;		/* No initrd provided by bootloader */
293 
294 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
295 }
296 
reserve_initrd(void)297 static void __init reserve_initrd(void)
298 {
299 	/* Assume only end is not page aligned */
300 	u64 ramdisk_image = get_ramdisk_image();
301 	u64 ramdisk_size  = get_ramdisk_size();
302 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
303 
304 	if (!boot_params.hdr.type_of_loader ||
305 	    !ramdisk_image || !ramdisk_size)
306 		return;		/* No initrd provided by bootloader */
307 
308 	initrd_start = 0;
309 
310 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
311 			ramdisk_end - 1);
312 
313 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
314 				PFN_DOWN(ramdisk_end))) {
315 		/* All are mapped, easy case */
316 		initrd_start = ramdisk_image + PAGE_OFFSET;
317 		initrd_end = initrd_start + ramdisk_size;
318 		return;
319 	}
320 
321 	relocate_initrd();
322 
323 	memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
324 }
325 
326 #else
early_reserve_initrd(void)327 static void __init early_reserve_initrd(void)
328 {
329 }
reserve_initrd(void)330 static void __init reserve_initrd(void)
331 {
332 }
333 #endif /* CONFIG_BLK_DEV_INITRD */
334 
add_early_ima_buffer(u64 phys_addr)335 static void __init add_early_ima_buffer(u64 phys_addr)
336 {
337 #ifdef CONFIG_IMA
338 	struct ima_setup_data *data;
339 
340 	data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
341 	if (!data) {
342 		pr_warn("setup: failed to memremap ima_setup_data entry\n");
343 		return;
344 	}
345 
346 	if (data->size) {
347 		memblock_reserve(data->addr, data->size);
348 		ima_kexec_buffer_phys = data->addr;
349 		ima_kexec_buffer_size = data->size;
350 	}
351 
352 	early_memunmap(data, sizeof(*data));
353 #else
354 	pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
355 #endif
356 }
357 
358 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
ima_free_kexec_buffer(void)359 int __init ima_free_kexec_buffer(void)
360 {
361 	if (!ima_kexec_buffer_size)
362 		return -ENOENT;
363 
364 	memblock_free_late(ima_kexec_buffer_phys,
365 			   ima_kexec_buffer_size);
366 
367 	ima_kexec_buffer_phys = 0;
368 	ima_kexec_buffer_size = 0;
369 
370 	return 0;
371 }
372 
ima_get_kexec_buffer(void ** addr,size_t * size)373 int __init ima_get_kexec_buffer(void **addr, size_t *size)
374 {
375 	if (!ima_kexec_buffer_size)
376 		return -ENOENT;
377 
378 	*addr = __va(ima_kexec_buffer_phys);
379 	*size = ima_kexec_buffer_size;
380 
381 	return 0;
382 }
383 #endif
384 
parse_setup_data(void)385 static void __init parse_setup_data(void)
386 {
387 	struct setup_data *data;
388 	u64 pa_data, pa_next;
389 
390 	pa_data = boot_params.hdr.setup_data;
391 	while (pa_data) {
392 		u32 data_len, data_type;
393 
394 		data = early_memremap(pa_data, sizeof(*data));
395 		data_len = data->len + sizeof(struct setup_data);
396 		data_type = data->type;
397 		pa_next = data->next;
398 		early_memunmap(data, sizeof(*data));
399 
400 		switch (data_type) {
401 		case SETUP_E820_EXT:
402 			e820__memory_setup_extended(pa_data, data_len);
403 			break;
404 		case SETUP_DTB:
405 			add_dtb(pa_data);
406 			break;
407 		case SETUP_EFI:
408 			parse_efi_setup(pa_data, data_len);
409 			break;
410 		case SETUP_IMA:
411 			add_early_ima_buffer(pa_data);
412 			break;
413 		case SETUP_RNG_SEED:
414 			data = early_memremap(pa_data, data_len);
415 			add_bootloader_randomness(data->data, data->len);
416 			/* Zero seed for forward secrecy. */
417 			memzero_explicit(data->data, data->len);
418 			/* Zero length in case we find ourselves back here by accident. */
419 			memzero_explicit(&data->len, sizeof(data->len));
420 			early_memunmap(data, data_len);
421 			break;
422 		default:
423 			break;
424 		}
425 		pa_data = pa_next;
426 	}
427 }
428 
memblock_x86_reserve_range_setup_data(void)429 static void __init memblock_x86_reserve_range_setup_data(void)
430 {
431 	struct setup_indirect *indirect;
432 	struct setup_data *data;
433 	u64 pa_data, pa_next;
434 	u32 len;
435 
436 	pa_data = boot_params.hdr.setup_data;
437 	while (pa_data) {
438 		data = early_memremap(pa_data, sizeof(*data));
439 		if (!data) {
440 			pr_warn("setup: failed to memremap setup_data entry\n");
441 			return;
442 		}
443 
444 		len = sizeof(*data);
445 		pa_next = data->next;
446 
447 		memblock_reserve(pa_data, sizeof(*data) + data->len);
448 
449 		if (data->type == SETUP_INDIRECT) {
450 			len += data->len;
451 			early_memunmap(data, sizeof(*data));
452 			data = early_memremap(pa_data, len);
453 			if (!data) {
454 				pr_warn("setup: failed to memremap indirect setup_data\n");
455 				return;
456 			}
457 
458 			indirect = (struct setup_indirect *)data->data;
459 
460 			if (indirect->type != SETUP_INDIRECT)
461 				memblock_reserve(indirect->addr, indirect->len);
462 		}
463 
464 		pa_data = pa_next;
465 		early_memunmap(data, len);
466 	}
467 }
468 
arch_reserve_crashkernel(void)469 static void __init arch_reserve_crashkernel(void)
470 {
471 	unsigned long long crash_base, crash_size, low_size = 0;
472 	char *cmdline = boot_command_line;
473 	bool high = false;
474 	int ret;
475 
476 	if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
477 		return;
478 
479 	ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
480 				&crash_size, &crash_base,
481 				&low_size, &high);
482 	if (ret)
483 		return;
484 
485 	if (xen_pv_domain()) {
486 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
487 		return;
488 	}
489 
490 	reserve_crashkernel_generic(cmdline, crash_size, crash_base,
491 				    low_size, high);
492 }
493 
494 static struct resource standard_io_resources[] = {
495 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
496 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
497 	{ .name = "pic1", .start = 0x20, .end = 0x21,
498 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
499 	{ .name = "timer0", .start = 0x40, .end = 0x43,
500 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
501 	{ .name = "timer1", .start = 0x50, .end = 0x53,
502 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
503 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
504 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
505 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
506 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
507 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
508 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
509 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
510 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
511 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
512 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
513 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
514 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
515 };
516 
reserve_standard_io_resources(void)517 void __init reserve_standard_io_resources(void)
518 {
519 	int i;
520 
521 	/* request I/O space for devices used on all i[345]86 PCs */
522 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
523 		request_resource(&ioport_resource, &standard_io_resources[i]);
524 
525 }
526 
snb_gfx_workaround_needed(void)527 static bool __init snb_gfx_workaround_needed(void)
528 {
529 #ifdef CONFIG_PCI
530 	int i;
531 	u16 vendor, devid;
532 	static const __initconst u16 snb_ids[] = {
533 		0x0102,
534 		0x0112,
535 		0x0122,
536 		0x0106,
537 		0x0116,
538 		0x0126,
539 		0x010a,
540 	};
541 
542 	/* Assume no if something weird is going on with PCI */
543 	if (!early_pci_allowed())
544 		return false;
545 
546 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
547 	if (vendor != 0x8086)
548 		return false;
549 
550 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
551 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
552 		if (devid == snb_ids[i])
553 			return true;
554 #endif
555 
556 	return false;
557 }
558 
559 /*
560  * Sandy Bridge graphics has trouble with certain ranges, exclude
561  * them from allocation.
562  */
trim_snb_memory(void)563 static void __init trim_snb_memory(void)
564 {
565 	static const __initconst unsigned long bad_pages[] = {
566 		0x20050000,
567 		0x20110000,
568 		0x20130000,
569 		0x20138000,
570 		0x40004000,
571 	};
572 	int i;
573 
574 	if (!snb_gfx_workaround_needed())
575 		return;
576 
577 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
578 
579 	/*
580 	 * SandyBridge integrated graphics devices have a bug that prevents
581 	 * them from accessing certain memory ranges, namely anything below
582 	 * 1M and in the pages listed in bad_pages[] above.
583 	 *
584 	 * To avoid these pages being ever accessed by SNB gfx devices reserve
585 	 * bad_pages that have not already been reserved at boot time.
586 	 * All memory below the 1 MB mark is anyway reserved later during
587 	 * setup_arch(), so there is no need to reserve it here.
588 	 */
589 
590 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
591 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
592 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
593 			       bad_pages[i]);
594 	}
595 }
596 
trim_bios_range(void)597 static void __init trim_bios_range(void)
598 {
599 	/*
600 	 * A special case is the first 4Kb of memory;
601 	 * This is a BIOS owned area, not kernel ram, but generally
602 	 * not listed as such in the E820 table.
603 	 *
604 	 * This typically reserves additional memory (64KiB by default)
605 	 * since some BIOSes are known to corrupt low memory.  See the
606 	 * Kconfig help text for X86_RESERVE_LOW.
607 	 */
608 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
609 
610 	/*
611 	 * special case: Some BIOSes report the PC BIOS
612 	 * area (640Kb -> 1Mb) as RAM even though it is not.
613 	 * take them out.
614 	 */
615 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
616 
617 	e820__update_table(e820_table);
618 }
619 
620 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)621 static void __init e820_add_kernel_range(void)
622 {
623 	u64 start = __pa_symbol(_text);
624 	u64 size = __pa_symbol(_end) - start;
625 
626 	/*
627 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
628 	 * attempt to fix it by adding the range. We may have a confused BIOS,
629 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
630 	 * exclude kernel range. If we really are running on top non-RAM,
631 	 * we will crash later anyways.
632 	 */
633 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
634 		return;
635 
636 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
637 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
638 	e820__range_add(start, size, E820_TYPE_RAM);
639 }
640 
early_reserve_memory(void)641 static void __init early_reserve_memory(void)
642 {
643 	/*
644 	 * Reserve the memory occupied by the kernel between _text and
645 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
646 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
647 	 * separate memblock_reserve() or they will be discarded.
648 	 */
649 	memblock_reserve(__pa_symbol(_text),
650 			 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
651 
652 	/*
653 	 * The first 4Kb of memory is a BIOS owned area, but generally it is
654 	 * not listed as such in the E820 table.
655 	 *
656 	 * Reserve the first 64K of memory since some BIOSes are known to
657 	 * corrupt low memory. After the real mode trampoline is allocated the
658 	 * rest of the memory below 640k is reserved.
659 	 *
660 	 * In addition, make sure page 0 is always reserved because on
661 	 * systems with L1TF its contents can be leaked to user processes.
662 	 */
663 	memblock_reserve(0, SZ_64K);
664 
665 	early_reserve_initrd();
666 
667 	memblock_x86_reserve_range_setup_data();
668 
669 	reserve_bios_regions();
670 	trim_snb_memory();
671 }
672 
673 /*
674  * Dump out kernel offset information on panic.
675  */
676 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)677 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
678 {
679 	if (kaslr_enabled()) {
680 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
681 			 kaslr_offset(),
682 			 __START_KERNEL,
683 			 __START_KERNEL_map,
684 			 MODULES_VADDR-1);
685 	} else {
686 		pr_emerg("Kernel Offset: disabled\n");
687 	}
688 
689 	return 0;
690 }
691 
x86_configure_nx(void)692 void x86_configure_nx(void)
693 {
694 	if (boot_cpu_has(X86_FEATURE_NX))
695 		__supported_pte_mask |= _PAGE_NX;
696 	else
697 		__supported_pte_mask &= ~_PAGE_NX;
698 }
699 
x86_report_nx(void)700 static void __init x86_report_nx(void)
701 {
702 	if (!boot_cpu_has(X86_FEATURE_NX)) {
703 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
704 		       "missing in CPU!\n");
705 	} else {
706 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
707 		printk(KERN_INFO "NX (Execute Disable) protection: active\n");
708 #else
709 		/* 32bit non-PAE kernel, NX cannot be used */
710 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
711 		       "cannot be enabled: non-PAE kernel!\n");
712 #endif
713 	}
714 }
715 
716 /*
717  * Determine if we were loaded by an EFI loader.  If so, then we have also been
718  * passed the efi memmap, systab, etc., so we should use these data structures
719  * for initialization.  Note, the efi init code path is determined by the
720  * global efi_enabled. This allows the same kernel image to be used on existing
721  * systems (with a traditional BIOS) as well as on EFI systems.
722  */
723 /*
724  * setup_arch - architecture-specific boot-time initializations
725  *
726  * Note: On x86_64, fixmaps are ready for use even before this is called.
727  */
728 
setup_arch(char ** cmdline_p)729 void __init setup_arch(char **cmdline_p)
730 {
731 #ifdef CONFIG_X86_32
732 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
733 
734 	/*
735 	 * copy kernel address range established so far and switch
736 	 * to the proper swapper page table
737 	 */
738 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
739 			initial_page_table + KERNEL_PGD_BOUNDARY,
740 			KERNEL_PGD_PTRS);
741 
742 	load_cr3(swapper_pg_dir);
743 	/*
744 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
745 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
746 	 * will not flush anything because the CPU quirk which clears
747 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
748 	 * load_cr3() above the TLB has been flushed already. The
749 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
750 	 * so proper operation is guaranteed.
751 	 */
752 	__flush_tlb_all();
753 #else
754 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
755 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
756 #endif
757 
758 #ifdef CONFIG_CMDLINE_BOOL
759 #ifdef CONFIG_CMDLINE_OVERRIDE
760 	strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
761 #else
762 	if (builtin_cmdline[0]) {
763 		/* append boot loader cmdline to builtin */
764 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
765 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
766 		strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
767 	}
768 #endif
769 	builtin_cmdline_added = true;
770 #endif
771 
772 	strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
773 	*cmdline_p = command_line;
774 
775 	/*
776 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
777 	 * reserve_top(), so do this before touching the ioremap area.
778 	 */
779 	olpc_ofw_detect();
780 
781 	idt_setup_early_traps();
782 	early_cpu_init();
783 	jump_label_init();
784 	static_call_init();
785 	early_ioremap_init();
786 
787 	setup_olpc_ofw_pgd();
788 
789 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
790 	screen_info = boot_params.screen_info;
791 	edid_info = boot_params.edid_info;
792 #ifdef CONFIG_X86_32
793 	apm_info.bios = boot_params.apm_bios_info;
794 	ist_info = boot_params.ist_info;
795 #endif
796 	saved_video_mode = boot_params.hdr.vid_mode;
797 	bootloader_type = boot_params.hdr.type_of_loader;
798 	if ((bootloader_type >> 4) == 0xe) {
799 		bootloader_type &= 0xf;
800 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
801 	}
802 	bootloader_version  = bootloader_type & 0xf;
803 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
804 
805 #ifdef CONFIG_BLK_DEV_RAM
806 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
807 #endif
808 #ifdef CONFIG_EFI
809 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
810 		     EFI32_LOADER_SIGNATURE, 4)) {
811 		set_bit(EFI_BOOT, &efi.flags);
812 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
813 		     EFI64_LOADER_SIGNATURE, 4)) {
814 		set_bit(EFI_BOOT, &efi.flags);
815 		set_bit(EFI_64BIT, &efi.flags);
816 	}
817 #endif
818 
819 	x86_init.oem.arch_setup();
820 
821 	/*
822 	 * Do some memory reservations *before* memory is added to memblock, so
823 	 * memblock allocations won't overwrite it.
824 	 *
825 	 * After this point, everything still needed from the boot loader or
826 	 * firmware or kernel text should be early reserved or marked not RAM in
827 	 * e820. All other memory is free game.
828 	 *
829 	 * This call needs to happen before e820__memory_setup() which calls the
830 	 * xen_memory_setup() on Xen dom0 which relies on the fact that those
831 	 * early reservations have happened already.
832 	 */
833 	early_reserve_memory();
834 
835 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
836 	e820__memory_setup();
837 	parse_setup_data();
838 
839 	copy_edd();
840 
841 	if (!boot_params.hdr.root_flags)
842 		root_mountflags &= ~MS_RDONLY;
843 	setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
844 
845 	code_resource.start = __pa_symbol(_text);
846 	code_resource.end = __pa_symbol(_etext)-1;
847 	rodata_resource.start = __pa_symbol(__start_rodata);
848 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
849 	data_resource.start = __pa_symbol(_sdata);
850 	data_resource.end = __pa_symbol(_edata)-1;
851 	bss_resource.start = __pa_symbol(__bss_start);
852 	bss_resource.end = __pa_symbol(__bss_stop)-1;
853 
854 	/*
855 	 * x86_configure_nx() is called before parse_early_param() to detect
856 	 * whether hardware doesn't support NX (so that the early EHCI debug
857 	 * console setup can safely call set_fixmap()).
858 	 */
859 	x86_configure_nx();
860 
861 	parse_early_param();
862 
863 	if (efi_enabled(EFI_BOOT))
864 		efi_memblock_x86_reserve_range();
865 
866 #ifdef CONFIG_MEMORY_HOTPLUG
867 	/*
868 	 * Memory used by the kernel cannot be hot-removed because Linux
869 	 * cannot migrate the kernel pages. When memory hotplug is
870 	 * enabled, we should prevent memblock from allocating memory
871 	 * for the kernel.
872 	 *
873 	 * ACPI SRAT records all hotpluggable memory ranges. But before
874 	 * SRAT is parsed, we don't know about it.
875 	 *
876 	 * The kernel image is loaded into memory at very early time. We
877 	 * cannot prevent this anyway. So on NUMA system, we set any
878 	 * node the kernel resides in as un-hotpluggable.
879 	 *
880 	 * Since on modern servers, one node could have double-digit
881 	 * gigabytes memory, we can assume the memory around the kernel
882 	 * image is also un-hotpluggable. So before SRAT is parsed, just
883 	 * allocate memory near the kernel image to try the best to keep
884 	 * the kernel away from hotpluggable memory.
885 	 */
886 	if (movable_node_is_enabled())
887 		memblock_set_bottom_up(true);
888 #endif
889 
890 	x86_report_nx();
891 
892 	apic_setup_apic_calls();
893 
894 	if (acpi_mps_check()) {
895 #ifdef CONFIG_X86_LOCAL_APIC
896 		apic_is_disabled = true;
897 #endif
898 		setup_clear_cpu_cap(X86_FEATURE_APIC);
899 	}
900 
901 	e820__reserve_setup_data();
902 	e820__finish_early_params();
903 
904 	if (efi_enabled(EFI_BOOT))
905 		efi_init();
906 
907 	reserve_ibft_region();
908 	x86_init.resources.dmi_setup();
909 
910 	/*
911 	 * VMware detection requires dmi to be available, so this
912 	 * needs to be done after dmi_setup(), for the boot CPU.
913 	 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
914 	 * called before cache_bp_init() for setting up MTRR state.
915 	 */
916 	init_hypervisor_platform();
917 
918 	tsc_early_init();
919 	x86_init.resources.probe_roms();
920 
921 	/* after parse_early_param, so could debug it */
922 	insert_resource(&iomem_resource, &code_resource);
923 	insert_resource(&iomem_resource, &rodata_resource);
924 	insert_resource(&iomem_resource, &data_resource);
925 	insert_resource(&iomem_resource, &bss_resource);
926 
927 	e820_add_kernel_range();
928 	trim_bios_range();
929 #ifdef CONFIG_X86_32
930 	if (ppro_with_ram_bug()) {
931 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
932 				  E820_TYPE_RESERVED);
933 		e820__update_table(e820_table);
934 		printk(KERN_INFO "fixed physical RAM map:\n");
935 		e820__print_table("bad_ppro");
936 	}
937 #else
938 	early_gart_iommu_check();
939 #endif
940 
941 	/*
942 	 * partially used pages are not usable - thus
943 	 * we are rounding upwards:
944 	 */
945 	max_pfn = e820__end_of_ram_pfn();
946 
947 	/* update e820 for memory not covered by WB MTRRs */
948 	cache_bp_init();
949 	if (mtrr_trim_uncached_memory(max_pfn))
950 		max_pfn = e820__end_of_ram_pfn();
951 
952 	max_possible_pfn = max_pfn;
953 
954 	/*
955 	 * Define random base addresses for memory sections after max_pfn is
956 	 * defined and before each memory section base is used.
957 	 */
958 	kernel_randomize_memory();
959 
960 #ifdef CONFIG_X86_32
961 	/* max_low_pfn get updated here */
962 	find_low_pfn_range();
963 #else
964 	check_x2apic();
965 
966 	/* How many end-of-memory variables you have, grandma! */
967 	/* need this before calling reserve_initrd */
968 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
969 		max_low_pfn = e820__end_of_low_ram_pfn();
970 	else
971 		max_low_pfn = max_pfn;
972 
973 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
974 #endif
975 
976 	/* Find and reserve MPTABLE area */
977 	x86_init.mpparse.find_mptable();
978 
979 	early_alloc_pgt_buf();
980 
981 	/*
982 	 * Need to conclude brk, before e820__memblock_setup()
983 	 * it could use memblock_find_in_range, could overlap with
984 	 * brk area.
985 	 */
986 	reserve_brk();
987 
988 	cleanup_highmap();
989 
990 	memblock_set_current_limit(ISA_END_ADDRESS);
991 	e820__memblock_setup();
992 
993 	/*
994 	 * Needs to run after memblock setup because it needs the physical
995 	 * memory size.
996 	 */
997 	mem_encrypt_setup_arch();
998 	cc_random_init();
999 
1000 	efi_find_mirror();
1001 	efi_esrt_init();
1002 	efi_mokvar_table_init();
1003 
1004 	/*
1005 	 * The EFI specification says that boot service code won't be
1006 	 * called after ExitBootServices(). This is, in fact, a lie.
1007 	 */
1008 	efi_reserve_boot_services();
1009 
1010 	/* preallocate 4k for mptable mpc */
1011 	e820__memblock_alloc_reserved_mpc_new();
1012 
1013 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1014 	setup_bios_corruption_check();
1015 #endif
1016 
1017 #ifdef CONFIG_X86_32
1018 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1019 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1020 #endif
1021 
1022 	/*
1023 	 * Find free memory for the real mode trampoline and place it there. If
1024 	 * there is not enough free memory under 1M, on EFI-enabled systems
1025 	 * there will be additional attempt to reclaim the memory for the real
1026 	 * mode trampoline at efi_free_boot_services().
1027 	 *
1028 	 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1029 	 * are known to corrupt low memory and several hundred kilobytes are not
1030 	 * worth complex detection what memory gets clobbered. Windows does the
1031 	 * same thing for very similar reasons.
1032 	 *
1033 	 * Moreover, on machines with SandyBridge graphics or in setups that use
1034 	 * crashkernel the entire 1M is reserved anyway.
1035 	 *
1036 	 * Note the host kernel TDX also requires the first 1MB being reserved.
1037 	 */
1038 	x86_platform.realmode_reserve();
1039 
1040 	init_mem_mapping();
1041 
1042 	/*
1043 	 * init_mem_mapping() relies on the early IDT page fault handling.
1044 	 * Now either enable FRED or install the real page fault handler
1045 	 * for 64-bit in the IDT.
1046 	 */
1047 	cpu_init_replace_early_idt();
1048 
1049 	/*
1050 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1051 	 * with the current CR4 value.  This may not be necessary, but
1052 	 * auditing all the early-boot CR4 manipulation would be needed to
1053 	 * rule it out.
1054 	 *
1055 	 * Mask off features that don't work outside long mode (just
1056 	 * PCIDE for now).
1057 	 */
1058 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1059 
1060 	memblock_set_current_limit(get_max_mapped());
1061 
1062 	/*
1063 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1064 	 */
1065 
1066 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1067 	if (init_ohci1394_dma_early)
1068 		init_ohci1394_dma_on_all_controllers();
1069 #endif
1070 	/* Allocate bigger log buffer */
1071 	setup_log_buf(1);
1072 
1073 	if (efi_enabled(EFI_BOOT)) {
1074 		switch (boot_params.secure_boot) {
1075 		case efi_secureboot_mode_disabled:
1076 			pr_info("Secure boot disabled\n");
1077 			break;
1078 		case efi_secureboot_mode_enabled:
1079 			pr_info("Secure boot enabled\n");
1080 			break;
1081 		default:
1082 			pr_info("Secure boot could not be determined\n");
1083 			break;
1084 		}
1085 	}
1086 
1087 	reserve_initrd();
1088 
1089 	acpi_table_upgrade();
1090 	/* Look for ACPI tables and reserve memory occupied by them. */
1091 	acpi_boot_table_init();
1092 
1093 	vsmp_init();
1094 
1095 	io_delay_init();
1096 
1097 	early_platform_quirks();
1098 
1099 	/* Some platforms need the APIC registered for NUMA configuration */
1100 	early_acpi_boot_init();
1101 	x86_init.mpparse.early_parse_smp_cfg();
1102 
1103 	x86_flattree_get_config();
1104 
1105 	initmem_init();
1106 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1107 
1108 	if (boot_cpu_has(X86_FEATURE_GBPAGES))
1109 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1110 
1111 	/*
1112 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1113 	 * won't consume hotpluggable memory.
1114 	 */
1115 	arch_reserve_crashkernel();
1116 
1117 	if (!early_xdbc_setup_hardware())
1118 		early_xdbc_register_console();
1119 
1120 	x86_init.paging.pagetable_init();
1121 
1122 	kasan_init();
1123 
1124 	/*
1125 	 * Sync back kernel address range.
1126 	 *
1127 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1128 	 * this call?
1129 	 */
1130 	sync_initial_page_table();
1131 
1132 	tboot_probe();
1133 
1134 	map_vsyscall();
1135 
1136 	x86_32_probe_apic();
1137 
1138 	early_quirks();
1139 
1140 	topology_apply_cmdline_limits_early();
1141 
1142 	/*
1143 	 * Parse SMP configuration. Try ACPI first and then the platform
1144 	 * specific parser.
1145 	 */
1146 	acpi_boot_init();
1147 	x86_init.mpparse.parse_smp_cfg();
1148 
1149 	/* Last opportunity to detect and map the local APIC */
1150 	init_apic_mappings();
1151 
1152 	topology_init_possible_cpus();
1153 
1154 	init_cpu_to_node();
1155 	init_gi_nodes();
1156 
1157 	io_apic_init_mappings();
1158 
1159 	x86_init.hyper.guest_late_init();
1160 
1161 	e820__reserve_resources();
1162 	e820__register_nosave_regions(max_pfn);
1163 
1164 	x86_init.resources.reserve_resources();
1165 
1166 	e820__setup_pci_gap();
1167 
1168 #ifdef CONFIG_VT
1169 #if defined(CONFIG_VGA_CONSOLE)
1170 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1171 		vgacon_register_screen(&screen_info);
1172 #endif
1173 #endif
1174 	x86_init.oem.banner();
1175 
1176 	x86_init.timers.wallclock_init();
1177 
1178 	/*
1179 	 * This needs to run before setup_local_APIC() which soft-disables the
1180 	 * local APIC temporarily and that masks the thermal LVT interrupt,
1181 	 * leading to softlockups on machines which have configured SMI
1182 	 * interrupt delivery.
1183 	 */
1184 	therm_lvt_init();
1185 
1186 	mcheck_init();
1187 
1188 	register_refined_jiffies(CLOCK_TICK_RATE);
1189 
1190 #ifdef CONFIG_EFI
1191 	if (efi_enabled(EFI_BOOT))
1192 		efi_apply_memmap_quirks();
1193 #endif
1194 
1195 	unwind_init();
1196 }
1197 
1198 #ifdef CONFIG_X86_32
1199 
1200 static struct resource video_ram_resource = {
1201 	.name	= "Video RAM area",
1202 	.start	= 0xa0000,
1203 	.end	= 0xbffff,
1204 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1205 };
1206 
i386_reserve_resources(void)1207 void __init i386_reserve_resources(void)
1208 {
1209 	request_resource(&iomem_resource, &video_ram_resource);
1210 	reserve_standard_io_resources();
1211 }
1212 
1213 #endif /* CONFIG_X86_32 */
1214 
1215 static struct notifier_block kernel_offset_notifier = {
1216 	.notifier_call = dump_kernel_offset
1217 };
1218 
register_kernel_offset_dumper(void)1219 static int __init register_kernel_offset_dumper(void)
1220 {
1221 	atomic_notifier_chain_register(&panic_notifier_list,
1222 					&kernel_offset_notifier);
1223 	return 0;
1224 }
1225 __initcall(register_kernel_offset_dumper);
1226 
1227 #ifdef CONFIG_HOTPLUG_CPU
arch_cpu_is_hotpluggable(int cpu)1228 bool arch_cpu_is_hotpluggable(int cpu)
1229 {
1230 	return cpu > 0;
1231 }
1232 #endif /* CONFIG_HOTPLUG_CPU */
1233