xref: /linux/arch/x86/kernel/setup.c (revision 222408cde4d0ab17e54d4db26751c2b5cab9ac2b)
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/crash_dump.h>
11 #include <linux/dma-map-ops.h>
12 #include <linux/efi.h>
13 #include <linux/ima.h>
14 #include <linux/init_ohci1394_dma.h>
15 #include <linux/initrd.h>
16 #include <linux/iscsi_ibft.h>
17 #include <linux/memblock.h>
18 #include <linux/panic_notifier.h>
19 #include <linux/pci.h>
20 #include <linux/root_dev.h>
21 #include <linux/hugetlb.h>
22 #include <linux/tboot.h>
23 #include <linux/usb/xhci-dbgp.h>
24 #include <linux/static_call.h>
25 #include <linux/swiotlb.h>
26 #include <linux/random.h>
27 
28 #include <uapi/linux/mount.h>
29 
30 #include <xen/xen.h>
31 
32 #include <asm/apic.h>
33 #include <asm/efi.h>
34 #include <asm/numa.h>
35 #include <asm/bios_ebda.h>
36 #include <asm/bugs.h>
37 #include <asm/cacheinfo.h>
38 #include <asm/cpu.h>
39 #include <asm/efi.h>
40 #include <asm/gart.h>
41 #include <asm/hypervisor.h>
42 #include <asm/io_apic.h>
43 #include <asm/kasan.h>
44 #include <asm/kaslr.h>
45 #include <asm/mce.h>
46 #include <asm/memtype.h>
47 #include <asm/mtrr.h>
48 #include <asm/realmode.h>
49 #include <asm/olpc_ofw.h>
50 #include <asm/pci-direct.h>
51 #include <asm/prom.h>
52 #include <asm/proto.h>
53 #include <asm/thermal.h>
54 #include <asm/unwind.h>
55 #include <asm/vsyscall.h>
56 #include <linux/vmalloc.h>
57 
58 /*
59  * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
60  * max_pfn_mapped:     highest directly mapped pfn > 4 GB
61  *
62  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
63  * represented by pfn_mapped[].
64  */
65 unsigned long max_low_pfn_mapped;
66 unsigned long max_pfn_mapped;
67 
68 #ifdef CONFIG_DMI
69 RESERVE_BRK(dmi_alloc, 65536);
70 #endif
71 
72 
73 unsigned long _brk_start = (unsigned long)__brk_base;
74 unsigned long _brk_end   = (unsigned long)__brk_base;
75 
76 struct boot_params boot_params;
77 
78 /*
79  * These are the four main kernel memory regions, we put them into
80  * the resource tree so that kdump tools and other debugging tools
81  * recover it:
82  */
83 
84 static struct resource rodata_resource = {
85 	.name	= "Kernel rodata",
86 	.start	= 0,
87 	.end	= 0,
88 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
89 };
90 
91 static struct resource data_resource = {
92 	.name	= "Kernel data",
93 	.start	= 0,
94 	.end	= 0,
95 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
96 };
97 
98 static struct resource code_resource = {
99 	.name	= "Kernel code",
100 	.start	= 0,
101 	.end	= 0,
102 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
103 };
104 
105 static struct resource bss_resource = {
106 	.name	= "Kernel bss",
107 	.start	= 0,
108 	.end	= 0,
109 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
110 };
111 
112 
113 #ifdef CONFIG_X86_32
114 /* CPU data as detected by the assembly code in head_32.S */
115 struct cpuinfo_x86 new_cpu_data;
116 
117 struct apm_info apm_info;
118 EXPORT_SYMBOL(apm_info);
119 
120 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
121 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
122 struct ist_info ist_info;
123 EXPORT_SYMBOL(ist_info);
124 #else
125 struct ist_info ist_info;
126 #endif
127 
128 #endif
129 
130 struct cpuinfo_x86 boot_cpu_data __read_mostly;
131 EXPORT_SYMBOL(boot_cpu_data);
132 
133 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
134 __visible unsigned long mmu_cr4_features __ro_after_init;
135 #else
136 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
137 #endif
138 
139 #ifdef CONFIG_IMA
140 static phys_addr_t ima_kexec_buffer_phys;
141 static size_t ima_kexec_buffer_size;
142 #endif
143 
144 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
145 int bootloader_type, bootloader_version;
146 
147 /*
148  * Setup options
149  */
150 struct screen_info screen_info;
151 EXPORT_SYMBOL(screen_info);
152 struct edid_info edid_info;
153 EXPORT_SYMBOL_GPL(edid_info);
154 
155 extern int root_mountflags;
156 
157 unsigned long saved_video_mode;
158 
159 #define RAMDISK_IMAGE_START_MASK	0x07FF
160 #define RAMDISK_PROMPT_FLAG		0x8000
161 #define RAMDISK_LOAD_FLAG		0x4000
162 
163 static char __initdata command_line[COMMAND_LINE_SIZE];
164 #ifdef CONFIG_CMDLINE_BOOL
165 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
166 #endif
167 
168 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
169 struct edd edd;
170 #ifdef CONFIG_EDD_MODULE
171 EXPORT_SYMBOL(edd);
172 #endif
173 /**
174  * copy_edd() - Copy the BIOS EDD information
175  *              from boot_params into a safe place.
176  *
177  */
178 static inline void __init copy_edd(void)
179 {
180      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
181 	    sizeof(edd.mbr_signature));
182      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
183      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
184      edd.edd_info_nr = boot_params.eddbuf_entries;
185 }
186 #else
187 static inline void __init copy_edd(void)
188 {
189 }
190 #endif
191 
192 void * __init extend_brk(size_t size, size_t align)
193 {
194 	size_t mask = align - 1;
195 	void *ret;
196 
197 	BUG_ON(_brk_start == 0);
198 	BUG_ON(align & mask);
199 
200 	_brk_end = (_brk_end + mask) & ~mask;
201 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
202 
203 	ret = (void *)_brk_end;
204 	_brk_end += size;
205 
206 	memset(ret, 0, size);
207 
208 	return ret;
209 }
210 
211 #ifdef CONFIG_X86_32
212 static void __init cleanup_highmap(void)
213 {
214 }
215 #endif
216 
217 static void __init reserve_brk(void)
218 {
219 	if (_brk_end > _brk_start)
220 		memblock_reserve(__pa_symbol(_brk_start),
221 				 _brk_end - _brk_start);
222 
223 	/* Mark brk area as locked down and no longer taking any
224 	   new allocations */
225 	_brk_start = 0;
226 }
227 
228 #ifdef CONFIG_BLK_DEV_INITRD
229 
230 static u64 __init get_ramdisk_image(void)
231 {
232 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
233 
234 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
235 
236 	if (ramdisk_image == 0)
237 		ramdisk_image = phys_initrd_start;
238 
239 	return ramdisk_image;
240 }
241 static u64 __init get_ramdisk_size(void)
242 {
243 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
244 
245 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
246 
247 	if (ramdisk_size == 0)
248 		ramdisk_size = phys_initrd_size;
249 
250 	return ramdisk_size;
251 }
252 
253 static void __init relocate_initrd(void)
254 {
255 	/* Assume only end is not page aligned */
256 	u64 ramdisk_image = get_ramdisk_image();
257 	u64 ramdisk_size  = get_ramdisk_size();
258 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
259 
260 	/* We need to move the initrd down into directly mapped mem */
261 	u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
262 						      PFN_PHYS(max_pfn_mapped));
263 	if (!relocated_ramdisk)
264 		panic("Cannot find place for new RAMDISK of size %lld\n",
265 		      ramdisk_size);
266 
267 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
268 	initrd_end   = initrd_start + ramdisk_size;
269 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
270 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
271 
272 	copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
273 
274 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
275 		" [mem %#010llx-%#010llx]\n",
276 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
277 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
278 }
279 
280 static void __init early_reserve_initrd(void)
281 {
282 	/* Assume only end is not page aligned */
283 	u64 ramdisk_image = get_ramdisk_image();
284 	u64 ramdisk_size  = get_ramdisk_size();
285 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
286 
287 	if (!boot_params.hdr.type_of_loader ||
288 	    !ramdisk_image || !ramdisk_size)
289 		return;		/* No initrd provided by bootloader */
290 
291 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
292 }
293 
294 static void __init reserve_initrd(void)
295 {
296 	/* Assume only end is not page aligned */
297 	u64 ramdisk_image = get_ramdisk_image();
298 	u64 ramdisk_size  = get_ramdisk_size();
299 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
300 
301 	if (!boot_params.hdr.type_of_loader ||
302 	    !ramdisk_image || !ramdisk_size)
303 		return;		/* No initrd provided by bootloader */
304 
305 	initrd_start = 0;
306 
307 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
308 			ramdisk_end - 1);
309 
310 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
311 				PFN_DOWN(ramdisk_end))) {
312 		/* All are mapped, easy case */
313 		initrd_start = ramdisk_image + PAGE_OFFSET;
314 		initrd_end = initrd_start + ramdisk_size;
315 		return;
316 	}
317 
318 	relocate_initrd();
319 
320 	memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
321 }
322 
323 #else
324 static void __init early_reserve_initrd(void)
325 {
326 }
327 static void __init reserve_initrd(void)
328 {
329 }
330 #endif /* CONFIG_BLK_DEV_INITRD */
331 
332 static void __init add_early_ima_buffer(u64 phys_addr)
333 {
334 #ifdef CONFIG_IMA
335 	struct ima_setup_data *data;
336 
337 	data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
338 	if (!data) {
339 		pr_warn("setup: failed to memremap ima_setup_data entry\n");
340 		return;
341 	}
342 
343 	if (data->size) {
344 		memblock_reserve(data->addr, data->size);
345 		ima_kexec_buffer_phys = data->addr;
346 		ima_kexec_buffer_size = data->size;
347 	}
348 
349 	early_memunmap(data, sizeof(*data));
350 #else
351 	pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
352 #endif
353 }
354 
355 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
356 int __init ima_free_kexec_buffer(void)
357 {
358 	if (!ima_kexec_buffer_size)
359 		return -ENOENT;
360 
361 	memblock_free_late(ima_kexec_buffer_phys,
362 			   ima_kexec_buffer_size);
363 
364 	ima_kexec_buffer_phys = 0;
365 	ima_kexec_buffer_size = 0;
366 
367 	return 0;
368 }
369 
370 int __init ima_get_kexec_buffer(void **addr, size_t *size)
371 {
372 	if (!ima_kexec_buffer_size)
373 		return -ENOENT;
374 
375 	*addr = __va(ima_kexec_buffer_phys);
376 	*size = ima_kexec_buffer_size;
377 
378 	return 0;
379 }
380 #endif
381 
382 static void __init parse_setup_data(void)
383 {
384 	struct setup_data *data;
385 	u64 pa_data, pa_next;
386 
387 	pa_data = boot_params.hdr.setup_data;
388 	while (pa_data) {
389 		u32 data_len, data_type;
390 
391 		data = early_memremap(pa_data, sizeof(*data));
392 		data_len = data->len + sizeof(struct setup_data);
393 		data_type = data->type;
394 		pa_next = data->next;
395 		early_memunmap(data, sizeof(*data));
396 
397 		switch (data_type) {
398 		case SETUP_E820_EXT:
399 			e820__memory_setup_extended(pa_data, data_len);
400 			break;
401 		case SETUP_DTB:
402 			add_dtb(pa_data);
403 			break;
404 		case SETUP_EFI:
405 			parse_efi_setup(pa_data, data_len);
406 			break;
407 		case SETUP_IMA:
408 			add_early_ima_buffer(pa_data);
409 			break;
410 		case SETUP_RNG_SEED:
411 			data = early_memremap(pa_data, data_len);
412 			add_bootloader_randomness(data->data, data->len);
413 			/* Zero seed for forward secrecy. */
414 			memzero_explicit(data->data, data->len);
415 			/* Zero length in case we find ourselves back here by accident. */
416 			memzero_explicit(&data->len, sizeof(data->len));
417 			early_memunmap(data, data_len);
418 			break;
419 		default:
420 			break;
421 		}
422 		pa_data = pa_next;
423 	}
424 }
425 
426 static void __init memblock_x86_reserve_range_setup_data(void)
427 {
428 	struct setup_indirect *indirect;
429 	struct setup_data *data;
430 	u64 pa_data, pa_next;
431 	u32 len;
432 
433 	pa_data = boot_params.hdr.setup_data;
434 	while (pa_data) {
435 		data = early_memremap(pa_data, sizeof(*data));
436 		if (!data) {
437 			pr_warn("setup: failed to memremap setup_data entry\n");
438 			return;
439 		}
440 
441 		len = sizeof(*data);
442 		pa_next = data->next;
443 
444 		memblock_reserve(pa_data, sizeof(*data) + data->len);
445 
446 		if (data->type == SETUP_INDIRECT) {
447 			len += data->len;
448 			early_memunmap(data, sizeof(*data));
449 			data = early_memremap(pa_data, len);
450 			if (!data) {
451 				pr_warn("setup: failed to memremap indirect setup_data\n");
452 				return;
453 			}
454 
455 			indirect = (struct setup_indirect *)data->data;
456 
457 			if (indirect->type != SETUP_INDIRECT)
458 				memblock_reserve(indirect->addr, indirect->len);
459 		}
460 
461 		pa_data = pa_next;
462 		early_memunmap(data, len);
463 	}
464 }
465 
466 static void __init arch_reserve_crashkernel(void)
467 {
468 	unsigned long long crash_base, crash_size, low_size = 0;
469 	char *cmdline = boot_command_line;
470 	bool high = false;
471 	int ret;
472 
473 	if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
474 		return;
475 
476 	ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
477 				&crash_size, &crash_base,
478 				&low_size, &high);
479 	if (ret)
480 		return;
481 
482 	if (xen_pv_domain()) {
483 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
484 		return;
485 	}
486 
487 	reserve_crashkernel_generic(cmdline, crash_size, crash_base,
488 				    low_size, high);
489 }
490 
491 static struct resource standard_io_resources[] = {
492 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
493 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
494 	{ .name = "pic1", .start = 0x20, .end = 0x21,
495 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
496 	{ .name = "timer0", .start = 0x40, .end = 0x43,
497 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
498 	{ .name = "timer1", .start = 0x50, .end = 0x53,
499 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
500 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
501 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
502 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
503 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
504 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
505 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
506 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
507 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
508 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
509 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
510 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
511 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
512 };
513 
514 void __init reserve_standard_io_resources(void)
515 {
516 	int i;
517 
518 	/* request I/O space for devices used on all i[345]86 PCs */
519 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
520 		request_resource(&ioport_resource, &standard_io_resources[i]);
521 
522 }
523 
524 static bool __init snb_gfx_workaround_needed(void)
525 {
526 #ifdef CONFIG_PCI
527 	int i;
528 	u16 vendor, devid;
529 	static const __initconst u16 snb_ids[] = {
530 		0x0102,
531 		0x0112,
532 		0x0122,
533 		0x0106,
534 		0x0116,
535 		0x0126,
536 		0x010a,
537 	};
538 
539 	/* Assume no if something weird is going on with PCI */
540 	if (!early_pci_allowed())
541 		return false;
542 
543 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
544 	if (vendor != 0x8086)
545 		return false;
546 
547 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
548 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
549 		if (devid == snb_ids[i])
550 			return true;
551 #endif
552 
553 	return false;
554 }
555 
556 /*
557  * Sandy Bridge graphics has trouble with certain ranges, exclude
558  * them from allocation.
559  */
560 static void __init trim_snb_memory(void)
561 {
562 	static const __initconst unsigned long bad_pages[] = {
563 		0x20050000,
564 		0x20110000,
565 		0x20130000,
566 		0x20138000,
567 		0x40004000,
568 	};
569 	int i;
570 
571 	if (!snb_gfx_workaround_needed())
572 		return;
573 
574 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
575 
576 	/*
577 	 * SandyBridge integrated graphics devices have a bug that prevents
578 	 * them from accessing certain memory ranges, namely anything below
579 	 * 1M and in the pages listed in bad_pages[] above.
580 	 *
581 	 * To avoid these pages being ever accessed by SNB gfx devices reserve
582 	 * bad_pages that have not already been reserved at boot time.
583 	 * All memory below the 1 MB mark is anyway reserved later during
584 	 * setup_arch(), so there is no need to reserve it here.
585 	 */
586 
587 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
588 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
589 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
590 			       bad_pages[i]);
591 	}
592 }
593 
594 static void __init trim_bios_range(void)
595 {
596 	/*
597 	 * A special case is the first 4Kb of memory;
598 	 * This is a BIOS owned area, not kernel ram, but generally
599 	 * not listed as such in the E820 table.
600 	 *
601 	 * This typically reserves additional memory (64KiB by default)
602 	 * since some BIOSes are known to corrupt low memory.  See the
603 	 * Kconfig help text for X86_RESERVE_LOW.
604 	 */
605 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
606 
607 	/*
608 	 * special case: Some BIOSes report the PC BIOS
609 	 * area (640Kb -> 1Mb) as RAM even though it is not.
610 	 * take them out.
611 	 */
612 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
613 
614 	e820__update_table(e820_table);
615 }
616 
617 /* called before trim_bios_range() to spare extra sanitize */
618 static void __init e820_add_kernel_range(void)
619 {
620 	u64 start = __pa_symbol(_text);
621 	u64 size = __pa_symbol(_end) - start;
622 
623 	/*
624 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
625 	 * attempt to fix it by adding the range. We may have a confused BIOS,
626 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
627 	 * exclude kernel range. If we really are running on top non-RAM,
628 	 * we will crash later anyways.
629 	 */
630 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
631 		return;
632 
633 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
634 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
635 	e820__range_add(start, size, E820_TYPE_RAM);
636 }
637 
638 static void __init early_reserve_memory(void)
639 {
640 	/*
641 	 * Reserve the memory occupied by the kernel between _text and
642 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
643 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
644 	 * separate memblock_reserve() or they will be discarded.
645 	 */
646 	memblock_reserve(__pa_symbol(_text),
647 			 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
648 
649 	/*
650 	 * The first 4Kb of memory is a BIOS owned area, but generally it is
651 	 * not listed as such in the E820 table.
652 	 *
653 	 * Reserve the first 64K of memory since some BIOSes are known to
654 	 * corrupt low memory. After the real mode trampoline is allocated the
655 	 * rest of the memory below 640k is reserved.
656 	 *
657 	 * In addition, make sure page 0 is always reserved because on
658 	 * systems with L1TF its contents can be leaked to user processes.
659 	 */
660 	memblock_reserve(0, SZ_64K);
661 
662 	early_reserve_initrd();
663 
664 	memblock_x86_reserve_range_setup_data();
665 
666 	reserve_bios_regions();
667 	trim_snb_memory();
668 }
669 
670 /*
671  * Dump out kernel offset information on panic.
672  */
673 static int
674 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
675 {
676 	if (kaslr_enabled()) {
677 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
678 			 kaslr_offset(),
679 			 __START_KERNEL,
680 			 __START_KERNEL_map,
681 			 MODULES_VADDR-1);
682 	} else {
683 		pr_emerg("Kernel Offset: disabled\n");
684 	}
685 
686 	return 0;
687 }
688 
689 void x86_configure_nx(void)
690 {
691 	if (boot_cpu_has(X86_FEATURE_NX))
692 		__supported_pte_mask |= _PAGE_NX;
693 	else
694 		__supported_pte_mask &= ~_PAGE_NX;
695 }
696 
697 static void __init x86_report_nx(void)
698 {
699 	if (!boot_cpu_has(X86_FEATURE_NX)) {
700 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
701 		       "missing in CPU!\n");
702 	} else {
703 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
704 		printk(KERN_INFO "NX (Execute Disable) protection: active\n");
705 #else
706 		/* 32bit non-PAE kernel, NX cannot be used */
707 		printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
708 		       "cannot be enabled: non-PAE kernel!\n");
709 #endif
710 	}
711 }
712 
713 /*
714  * Determine if we were loaded by an EFI loader.  If so, then we have also been
715  * passed the efi memmap, systab, etc., so we should use these data structures
716  * for initialization.  Note, the efi init code path is determined by the
717  * global efi_enabled. This allows the same kernel image to be used on existing
718  * systems (with a traditional BIOS) as well as on EFI systems.
719  */
720 /*
721  * setup_arch - architecture-specific boot-time initializations
722  *
723  * Note: On x86_64, fixmaps are ready for use even before this is called.
724  */
725 
726 void __init setup_arch(char **cmdline_p)
727 {
728 #ifdef CONFIG_X86_32
729 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
730 
731 	/*
732 	 * copy kernel address range established so far and switch
733 	 * to the proper swapper page table
734 	 */
735 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
736 			initial_page_table + KERNEL_PGD_BOUNDARY,
737 			KERNEL_PGD_PTRS);
738 
739 	load_cr3(swapper_pg_dir);
740 	/*
741 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
742 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
743 	 * will not flush anything because the CPU quirk which clears
744 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
745 	 * load_cr3() above the TLB has been flushed already. The
746 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
747 	 * so proper operation is guaranteed.
748 	 */
749 	__flush_tlb_all();
750 #else
751 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
752 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
753 #endif
754 
755 	/*
756 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
757 	 * reserve_top(), so do this before touching the ioremap area.
758 	 */
759 	olpc_ofw_detect();
760 
761 	idt_setup_early_traps();
762 	early_cpu_init();
763 	jump_label_init();
764 	static_call_init();
765 	early_ioremap_init();
766 
767 	setup_olpc_ofw_pgd();
768 
769 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
770 	screen_info = boot_params.screen_info;
771 	edid_info = boot_params.edid_info;
772 #ifdef CONFIG_X86_32
773 	apm_info.bios = boot_params.apm_bios_info;
774 	ist_info = boot_params.ist_info;
775 #endif
776 	saved_video_mode = boot_params.hdr.vid_mode;
777 	bootloader_type = boot_params.hdr.type_of_loader;
778 	if ((bootloader_type >> 4) == 0xe) {
779 		bootloader_type &= 0xf;
780 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
781 	}
782 	bootloader_version  = bootloader_type & 0xf;
783 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
784 
785 #ifdef CONFIG_BLK_DEV_RAM
786 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
787 #endif
788 #ifdef CONFIG_EFI
789 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
790 		     EFI32_LOADER_SIGNATURE, 4)) {
791 		set_bit(EFI_BOOT, &efi.flags);
792 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
793 		     EFI64_LOADER_SIGNATURE, 4)) {
794 		set_bit(EFI_BOOT, &efi.flags);
795 		set_bit(EFI_64BIT, &efi.flags);
796 	}
797 #endif
798 
799 	x86_init.oem.arch_setup();
800 
801 	/*
802 	 * Do some memory reservations *before* memory is added to memblock, so
803 	 * memblock allocations won't overwrite it.
804 	 *
805 	 * After this point, everything still needed from the boot loader or
806 	 * firmware or kernel text should be early reserved or marked not RAM in
807 	 * e820. All other memory is free game.
808 	 *
809 	 * This call needs to happen before e820__memory_setup() which calls the
810 	 * xen_memory_setup() on Xen dom0 which relies on the fact that those
811 	 * early reservations have happened already.
812 	 */
813 	early_reserve_memory();
814 
815 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
816 	e820__memory_setup();
817 	parse_setup_data();
818 
819 	copy_edd();
820 
821 	if (!boot_params.hdr.root_flags)
822 		root_mountflags &= ~MS_RDONLY;
823 	setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
824 
825 	code_resource.start = __pa_symbol(_text);
826 	code_resource.end = __pa_symbol(_etext)-1;
827 	rodata_resource.start = __pa_symbol(__start_rodata);
828 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
829 	data_resource.start = __pa_symbol(_sdata);
830 	data_resource.end = __pa_symbol(_edata)-1;
831 	bss_resource.start = __pa_symbol(__bss_start);
832 	bss_resource.end = __pa_symbol(__bss_stop)-1;
833 
834 #ifdef CONFIG_CMDLINE_BOOL
835 #ifdef CONFIG_CMDLINE_OVERRIDE
836 	strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
837 #else
838 	if (builtin_cmdline[0]) {
839 		/* append boot loader cmdline to builtin */
840 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
841 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
842 		strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
843 	}
844 #endif
845 #endif
846 
847 	strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
848 	*cmdline_p = command_line;
849 
850 	/*
851 	 * x86_configure_nx() is called before parse_early_param() to detect
852 	 * whether hardware doesn't support NX (so that the early EHCI debug
853 	 * console setup can safely call set_fixmap()).
854 	 */
855 	x86_configure_nx();
856 
857 	parse_early_param();
858 
859 	if (efi_enabled(EFI_BOOT))
860 		efi_memblock_x86_reserve_range();
861 
862 #ifdef CONFIG_MEMORY_HOTPLUG
863 	/*
864 	 * Memory used by the kernel cannot be hot-removed because Linux
865 	 * cannot migrate the kernel pages. When memory hotplug is
866 	 * enabled, we should prevent memblock from allocating memory
867 	 * for the kernel.
868 	 *
869 	 * ACPI SRAT records all hotpluggable memory ranges. But before
870 	 * SRAT is parsed, we don't know about it.
871 	 *
872 	 * The kernel image is loaded into memory at very early time. We
873 	 * cannot prevent this anyway. So on NUMA system, we set any
874 	 * node the kernel resides in as un-hotpluggable.
875 	 *
876 	 * Since on modern servers, one node could have double-digit
877 	 * gigabytes memory, we can assume the memory around the kernel
878 	 * image is also un-hotpluggable. So before SRAT is parsed, just
879 	 * allocate memory near the kernel image to try the best to keep
880 	 * the kernel away from hotpluggable memory.
881 	 */
882 	if (movable_node_is_enabled())
883 		memblock_set_bottom_up(true);
884 #endif
885 
886 	x86_report_nx();
887 
888 	apic_setup_apic_calls();
889 
890 	if (acpi_mps_check()) {
891 #ifdef CONFIG_X86_LOCAL_APIC
892 		apic_is_disabled = true;
893 #endif
894 		setup_clear_cpu_cap(X86_FEATURE_APIC);
895 	}
896 
897 	e820__reserve_setup_data();
898 	e820__finish_early_params();
899 
900 	if (efi_enabled(EFI_BOOT))
901 		efi_init();
902 
903 	reserve_ibft_region();
904 	x86_init.resources.dmi_setup();
905 
906 	/*
907 	 * VMware detection requires dmi to be available, so this
908 	 * needs to be done after dmi_setup(), for the boot CPU.
909 	 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
910 	 * called before cache_bp_init() for setting up MTRR state.
911 	 */
912 	init_hypervisor_platform();
913 
914 	tsc_early_init();
915 	x86_init.resources.probe_roms();
916 
917 	/* after parse_early_param, so could debug it */
918 	insert_resource(&iomem_resource, &code_resource);
919 	insert_resource(&iomem_resource, &rodata_resource);
920 	insert_resource(&iomem_resource, &data_resource);
921 	insert_resource(&iomem_resource, &bss_resource);
922 
923 	e820_add_kernel_range();
924 	trim_bios_range();
925 #ifdef CONFIG_X86_32
926 	if (ppro_with_ram_bug()) {
927 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
928 				  E820_TYPE_RESERVED);
929 		e820__update_table(e820_table);
930 		printk(KERN_INFO "fixed physical RAM map:\n");
931 		e820__print_table("bad_ppro");
932 	}
933 #else
934 	early_gart_iommu_check();
935 #endif
936 
937 	/*
938 	 * partially used pages are not usable - thus
939 	 * we are rounding upwards:
940 	 */
941 	max_pfn = e820__end_of_ram_pfn();
942 
943 	/* update e820 for memory not covered by WB MTRRs */
944 	cache_bp_init();
945 	if (mtrr_trim_uncached_memory(max_pfn))
946 		max_pfn = e820__end_of_ram_pfn();
947 
948 	max_possible_pfn = max_pfn;
949 
950 	/*
951 	 * Define random base addresses for memory sections after max_pfn is
952 	 * defined and before each memory section base is used.
953 	 */
954 	kernel_randomize_memory();
955 
956 #ifdef CONFIG_X86_32
957 	/* max_low_pfn get updated here */
958 	find_low_pfn_range();
959 #else
960 	check_x2apic();
961 
962 	/* How many end-of-memory variables you have, grandma! */
963 	/* need this before calling reserve_initrd */
964 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
965 		max_low_pfn = e820__end_of_low_ram_pfn();
966 	else
967 		max_low_pfn = max_pfn;
968 
969 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
970 #endif
971 
972 	/* Find and reserve MPTABLE area */
973 	x86_init.mpparse.find_mptable();
974 
975 	early_alloc_pgt_buf();
976 
977 	/*
978 	 * Need to conclude brk, before e820__memblock_setup()
979 	 * it could use memblock_find_in_range, could overlap with
980 	 * brk area.
981 	 */
982 	reserve_brk();
983 
984 	cleanup_highmap();
985 
986 	memblock_set_current_limit(ISA_END_ADDRESS);
987 	e820__memblock_setup();
988 
989 	/*
990 	 * Needs to run after memblock setup because it needs the physical
991 	 * memory size.
992 	 */
993 	mem_encrypt_setup_arch();
994 
995 	efi_fake_memmap();
996 	efi_find_mirror();
997 	efi_esrt_init();
998 	efi_mokvar_table_init();
999 
1000 	/*
1001 	 * The EFI specification says that boot service code won't be
1002 	 * called after ExitBootServices(). This is, in fact, a lie.
1003 	 */
1004 	efi_reserve_boot_services();
1005 
1006 	/* preallocate 4k for mptable mpc */
1007 	e820__memblock_alloc_reserved_mpc_new();
1008 
1009 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1010 	setup_bios_corruption_check();
1011 #endif
1012 
1013 #ifdef CONFIG_X86_32
1014 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1015 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1016 #endif
1017 
1018 	/*
1019 	 * Find free memory for the real mode trampoline and place it there. If
1020 	 * there is not enough free memory under 1M, on EFI-enabled systems
1021 	 * there will be additional attempt to reclaim the memory for the real
1022 	 * mode trampoline at efi_free_boot_services().
1023 	 *
1024 	 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1025 	 * are known to corrupt low memory and several hundred kilobytes are not
1026 	 * worth complex detection what memory gets clobbered. Windows does the
1027 	 * same thing for very similar reasons.
1028 	 *
1029 	 * Moreover, on machines with SandyBridge graphics or in setups that use
1030 	 * crashkernel the entire 1M is reserved anyway.
1031 	 *
1032 	 * Note the host kernel TDX also requires the first 1MB being reserved.
1033 	 */
1034 	x86_platform.realmode_reserve();
1035 
1036 	init_mem_mapping();
1037 
1038 	idt_setup_early_pf();
1039 
1040 	/*
1041 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1042 	 * with the current CR4 value.  This may not be necessary, but
1043 	 * auditing all the early-boot CR4 manipulation would be needed to
1044 	 * rule it out.
1045 	 *
1046 	 * Mask off features that don't work outside long mode (just
1047 	 * PCIDE for now).
1048 	 */
1049 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1050 
1051 	memblock_set_current_limit(get_max_mapped());
1052 
1053 	/*
1054 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1055 	 */
1056 
1057 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1058 	if (init_ohci1394_dma_early)
1059 		init_ohci1394_dma_on_all_controllers();
1060 #endif
1061 	/* Allocate bigger log buffer */
1062 	setup_log_buf(1);
1063 
1064 	if (efi_enabled(EFI_BOOT)) {
1065 		switch (boot_params.secure_boot) {
1066 		case efi_secureboot_mode_disabled:
1067 			pr_info("Secure boot disabled\n");
1068 			break;
1069 		case efi_secureboot_mode_enabled:
1070 			pr_info("Secure boot enabled\n");
1071 			break;
1072 		default:
1073 			pr_info("Secure boot could not be determined\n");
1074 			break;
1075 		}
1076 	}
1077 
1078 	reserve_initrd();
1079 
1080 	acpi_table_upgrade();
1081 	/* Look for ACPI tables and reserve memory occupied by them. */
1082 	acpi_boot_table_init();
1083 
1084 	vsmp_init();
1085 
1086 	io_delay_init();
1087 
1088 	early_platform_quirks();
1089 
1090 	/* Some platforms need the APIC registered for NUMA configuration */
1091 	early_acpi_boot_init();
1092 	x86_init.mpparse.early_parse_smp_cfg();
1093 
1094 	x86_flattree_get_config();
1095 
1096 	initmem_init();
1097 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1098 
1099 	if (boot_cpu_has(X86_FEATURE_GBPAGES))
1100 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1101 
1102 	/*
1103 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1104 	 * won't consume hotpluggable memory.
1105 	 */
1106 	arch_reserve_crashkernel();
1107 
1108 	memblock_find_dma_reserve();
1109 
1110 	if (!early_xdbc_setup_hardware())
1111 		early_xdbc_register_console();
1112 
1113 	x86_init.paging.pagetable_init();
1114 
1115 	kasan_init();
1116 
1117 	/*
1118 	 * Sync back kernel address range.
1119 	 *
1120 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1121 	 * this call?
1122 	 */
1123 	sync_initial_page_table();
1124 
1125 	tboot_probe();
1126 
1127 	map_vsyscall();
1128 
1129 	x86_32_probe_apic();
1130 
1131 	early_quirks();
1132 
1133 	topology_apply_cmdline_limits_early();
1134 
1135 	/*
1136 	 * Parse SMP configuration. Try ACPI first and then the platform
1137 	 * specific parser.
1138 	 */
1139 	acpi_boot_init();
1140 	x86_init.mpparse.parse_smp_cfg();
1141 
1142 	/* Last opportunity to detect and map the local APIC */
1143 	init_apic_mappings();
1144 
1145 	topology_init_possible_cpus();
1146 
1147 	init_cpu_to_node();
1148 	init_gi_nodes();
1149 
1150 	io_apic_init_mappings();
1151 
1152 	x86_init.hyper.guest_late_init();
1153 
1154 	e820__reserve_resources();
1155 	e820__register_nosave_regions(max_pfn);
1156 
1157 	x86_init.resources.reserve_resources();
1158 
1159 	e820__setup_pci_gap();
1160 
1161 #ifdef CONFIG_VT
1162 #if defined(CONFIG_VGA_CONSOLE)
1163 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1164 		vgacon_register_screen(&screen_info);
1165 #endif
1166 #endif
1167 	x86_init.oem.banner();
1168 
1169 	x86_init.timers.wallclock_init();
1170 
1171 	/*
1172 	 * This needs to run before setup_local_APIC() which soft-disables the
1173 	 * local APIC temporarily and that masks the thermal LVT interrupt,
1174 	 * leading to softlockups on machines which have configured SMI
1175 	 * interrupt delivery.
1176 	 */
1177 	therm_lvt_init();
1178 
1179 	mcheck_init();
1180 
1181 	register_refined_jiffies(CLOCK_TICK_RATE);
1182 
1183 #ifdef CONFIG_EFI
1184 	if (efi_enabled(EFI_BOOT))
1185 		efi_apply_memmap_quirks();
1186 #endif
1187 
1188 	unwind_init();
1189 }
1190 
1191 #ifdef CONFIG_X86_32
1192 
1193 static struct resource video_ram_resource = {
1194 	.name	= "Video RAM area",
1195 	.start	= 0xa0000,
1196 	.end	= 0xbffff,
1197 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1198 };
1199 
1200 void __init i386_reserve_resources(void)
1201 {
1202 	request_resource(&iomem_resource, &video_ram_resource);
1203 	reserve_standard_io_resources();
1204 }
1205 
1206 #endif /* CONFIG_X86_32 */
1207 
1208 static struct notifier_block kernel_offset_notifier = {
1209 	.notifier_call = dump_kernel_offset
1210 };
1211 
1212 static int __init register_kernel_offset_dumper(void)
1213 {
1214 	atomic_notifier_chain_register(&panic_notifier_list,
1215 					&kernel_offset_notifier);
1216 	return 0;
1217 }
1218 __initcall(register_kernel_offset_dumper);
1219