xref: /openbmc/linux/arch/x86/kernel/setup.c (revision 3a83e4e6)
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/console.h>
9 #include <linux/crash_dump.h>
10 #include <linux/dmi.h>
11 #include <linux/efi.h>
12 #include <linux/init_ohci1394_dma.h>
13 #include <linux/initrd.h>
14 #include <linux/iscsi_ibft.h>
15 #include <linux/memblock.h>
16 #include <linux/pci.h>
17 #include <linux/root_dev.h>
18 #include <linux/sfi.h>
19 #include <linux/hugetlb.h>
20 #include <linux/tboot.h>
21 #include <linux/usb/xhci-dbgp.h>
22 
23 #include <uapi/linux/mount.h>
24 
25 #include <xen/xen.h>
26 
27 #include <asm/apic.h>
28 #include <asm/numa.h>
29 #include <asm/bios_ebda.h>
30 #include <asm/bugs.h>
31 #include <asm/cpu.h>
32 #include <asm/efi.h>
33 #include <asm/gart.h>
34 #include <asm/hypervisor.h>
35 #include <asm/io_apic.h>
36 #include <asm/kasan.h>
37 #include <asm/kaslr.h>
38 #include <asm/mce.h>
39 #include <asm/mtrr.h>
40 #include <asm/realmode.h>
41 #include <asm/olpc_ofw.h>
42 #include <asm/pci-direct.h>
43 #include <asm/prom.h>
44 #include <asm/proto.h>
45 #include <asm/unwind.h>
46 #include <asm/vsyscall.h>
47 #include <linux/vmalloc.h>
48 
49 /*
50  * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
51  * max_pfn_mapped:     highest directly mapped pfn > 4 GB
52  *
53  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
54  * represented by pfn_mapped[].
55  */
56 unsigned long max_low_pfn_mapped;
57 unsigned long max_pfn_mapped;
58 
59 #ifdef CONFIG_DMI
60 RESERVE_BRK(dmi_alloc, 65536);
61 #endif
62 
63 
64 /*
65  * Range of the BSS area. The size of the BSS area is determined
66  * at link time, with RESERVE_BRK*() facility reserving additional
67  * chunks.
68  */
69 unsigned long _brk_start = (unsigned long)__brk_base;
70 unsigned long _brk_end   = (unsigned long)__brk_base;
71 
72 struct boot_params boot_params;
73 
74 /*
75  * These are the four main kernel memory regions, we put them into
76  * the resource tree so that kdump tools and other debugging tools
77  * recover it:
78  */
79 
80 static struct resource rodata_resource = {
81 	.name	= "Kernel rodata",
82 	.start	= 0,
83 	.end	= 0,
84 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
85 };
86 
87 static struct resource data_resource = {
88 	.name	= "Kernel data",
89 	.start	= 0,
90 	.end	= 0,
91 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
92 };
93 
94 static struct resource code_resource = {
95 	.name	= "Kernel code",
96 	.start	= 0,
97 	.end	= 0,
98 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
99 };
100 
101 static struct resource bss_resource = {
102 	.name	= "Kernel bss",
103 	.start	= 0,
104 	.end	= 0,
105 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
106 };
107 
108 
109 #ifdef CONFIG_X86_32
110 /* CPU data as detected by the assembly code in head_32.S */
111 struct cpuinfo_x86 new_cpu_data;
112 
113 /* Common CPU data for all CPUs */
114 struct cpuinfo_x86 boot_cpu_data __read_mostly;
115 EXPORT_SYMBOL(boot_cpu_data);
116 
117 unsigned int def_to_bigsmp;
118 
119 /* For MCA, but anyone else can use it if they want */
120 unsigned int machine_id;
121 unsigned int machine_submodel_id;
122 unsigned int BIOS_revision;
123 
124 struct apm_info apm_info;
125 EXPORT_SYMBOL(apm_info);
126 
127 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
128 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
129 struct ist_info ist_info;
130 EXPORT_SYMBOL(ist_info);
131 #else
132 struct ist_info ist_info;
133 #endif
134 
135 #else
136 struct cpuinfo_x86 boot_cpu_data __read_mostly;
137 EXPORT_SYMBOL(boot_cpu_data);
138 #endif
139 
140 
141 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
142 __visible unsigned long mmu_cr4_features __ro_after_init;
143 #else
144 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
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 /*
151  * Setup options
152  */
153 struct screen_info screen_info;
154 EXPORT_SYMBOL(screen_info);
155 struct edid_info edid_info;
156 EXPORT_SYMBOL_GPL(edid_info);
157 
158 extern int root_mountflags;
159 
160 unsigned long saved_video_mode;
161 
162 #define RAMDISK_IMAGE_START_MASK	0x07FF
163 #define RAMDISK_PROMPT_FLAG		0x8000
164 #define RAMDISK_LOAD_FLAG		0x4000
165 
166 static char __initdata command_line[COMMAND_LINE_SIZE];
167 #ifdef CONFIG_CMDLINE_BOOL
168 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
169 #endif
170 
171 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
172 struct edd edd;
173 #ifdef CONFIG_EDD_MODULE
174 EXPORT_SYMBOL(edd);
175 #endif
176 /**
177  * copy_edd() - Copy the BIOS EDD information
178  *              from boot_params into a safe place.
179  *
180  */
181 static inline void __init copy_edd(void)
182 {
183      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
184 	    sizeof(edd.mbr_signature));
185      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
186      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
187      edd.edd_info_nr = boot_params.eddbuf_entries;
188 }
189 #else
190 static inline void __init copy_edd(void)
191 {
192 }
193 #endif
194 
195 void * __init extend_brk(size_t size, size_t align)
196 {
197 	size_t mask = align - 1;
198 	void *ret;
199 
200 	BUG_ON(_brk_start == 0);
201 	BUG_ON(align & mask);
202 
203 	_brk_end = (_brk_end + mask) & ~mask;
204 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
205 
206 	ret = (void *)_brk_end;
207 	_brk_end += size;
208 
209 	memset(ret, 0, size);
210 
211 	return ret;
212 }
213 
214 #ifdef CONFIG_X86_32
215 static void __init cleanup_highmap(void)
216 {
217 }
218 #endif
219 
220 static void __init reserve_brk(void)
221 {
222 	if (_brk_end > _brk_start)
223 		memblock_reserve(__pa_symbol(_brk_start),
224 				 _brk_end - _brk_start);
225 
226 	/* Mark brk area as locked down and no longer taking any
227 	   new allocations */
228 	_brk_start = 0;
229 }
230 
231 u64 relocated_ramdisk;
232 
233 #ifdef CONFIG_BLK_DEV_INITRD
234 
235 static u64 __init get_ramdisk_image(void)
236 {
237 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
238 
239 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
240 
241 	if (ramdisk_image == 0)
242 		ramdisk_image = phys_initrd_start;
243 
244 	return ramdisk_image;
245 }
246 static u64 __init get_ramdisk_size(void)
247 {
248 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
249 
250 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
251 
252 	if (ramdisk_size == 0)
253 		ramdisk_size = phys_initrd_size;
254 
255 	return ramdisk_size;
256 }
257 
258 static void __init relocate_initrd(void)
259 {
260 	/* Assume only end is not page aligned */
261 	u64 ramdisk_image = get_ramdisk_image();
262 	u64 ramdisk_size  = get_ramdisk_size();
263 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
264 
265 	/* We need to move the initrd down into directly mapped mem */
266 	relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
267 						   area_size, PAGE_SIZE);
268 
269 	if (!relocated_ramdisk)
270 		panic("Cannot find place for new RAMDISK of size %lld\n",
271 		      ramdisk_size);
272 
273 	/* Note: this includes all the mem currently occupied by
274 	   the initrd, we rely on that fact to keep the data intact. */
275 	memblock_reserve(relocated_ramdisk, area_size);
276 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
277 	initrd_end   = initrd_start + ramdisk_size;
278 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
279 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
280 
281 	copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
282 
283 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
284 		" [mem %#010llx-%#010llx]\n",
285 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
286 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
287 }
288 
289 static void __init early_reserve_initrd(void)
290 {
291 	/* Assume only end is not page aligned */
292 	u64 ramdisk_image = get_ramdisk_image();
293 	u64 ramdisk_size  = get_ramdisk_size();
294 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
295 
296 	if (!boot_params.hdr.type_of_loader ||
297 	    !ramdisk_image || !ramdisk_size)
298 		return;		/* No initrd provided by bootloader */
299 
300 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
301 }
302 static void __init reserve_initrd(void)
303 {
304 	/* Assume only end is not page aligned */
305 	u64 ramdisk_image = get_ramdisk_image();
306 	u64 ramdisk_size  = get_ramdisk_size();
307 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
308 	u64 mapped_size;
309 
310 	if (!boot_params.hdr.type_of_loader ||
311 	    !ramdisk_image || !ramdisk_size)
312 		return;		/* No initrd provided by bootloader */
313 
314 	initrd_start = 0;
315 
316 	mapped_size = memblock_mem_size(max_pfn_mapped);
317 	if (ramdisk_size >= (mapped_size>>1))
318 		panic("initrd too large to handle, "
319 		       "disabling initrd (%lld needed, %lld available)\n",
320 		       ramdisk_size, mapped_size>>1);
321 
322 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
323 			ramdisk_end - 1);
324 
325 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
326 				PFN_DOWN(ramdisk_end))) {
327 		/* All are mapped, easy case */
328 		initrd_start = ramdisk_image + PAGE_OFFSET;
329 		initrd_end = initrd_start + ramdisk_size;
330 		return;
331 	}
332 
333 	relocate_initrd();
334 
335 	memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
336 }
337 
338 #else
339 static void __init early_reserve_initrd(void)
340 {
341 }
342 static void __init reserve_initrd(void)
343 {
344 }
345 #endif /* CONFIG_BLK_DEV_INITRD */
346 
347 static void __init parse_setup_data(void)
348 {
349 	struct setup_data *data;
350 	u64 pa_data, pa_next;
351 
352 	pa_data = boot_params.hdr.setup_data;
353 	while (pa_data) {
354 		u32 data_len, data_type;
355 
356 		data = early_memremap(pa_data, sizeof(*data));
357 		data_len = data->len + sizeof(struct setup_data);
358 		data_type = data->type;
359 		pa_next = data->next;
360 		early_memunmap(data, sizeof(*data));
361 
362 		switch (data_type) {
363 		case SETUP_E820_EXT:
364 			e820__memory_setup_extended(pa_data, data_len);
365 			break;
366 		case SETUP_DTB:
367 			add_dtb(pa_data);
368 			break;
369 		case SETUP_EFI:
370 			parse_efi_setup(pa_data, data_len);
371 			break;
372 		default:
373 			break;
374 		}
375 		pa_data = pa_next;
376 	}
377 }
378 
379 static void __init memblock_x86_reserve_range_setup_data(void)
380 {
381 	struct setup_data *data;
382 	u64 pa_data;
383 
384 	pa_data = boot_params.hdr.setup_data;
385 	while (pa_data) {
386 		data = early_memremap(pa_data, sizeof(*data));
387 		memblock_reserve(pa_data, sizeof(*data) + data->len);
388 
389 		if (data->type == SETUP_INDIRECT &&
390 		    ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
391 			memblock_reserve(((struct setup_indirect *)data->data)->addr,
392 					 ((struct setup_indirect *)data->data)->len);
393 
394 		pa_data = data->next;
395 		early_memunmap(data, sizeof(*data));
396 	}
397 }
398 
399 /*
400  * --------- Crashkernel reservation ------------------------------
401  */
402 
403 #ifdef CONFIG_KEXEC_CORE
404 
405 /* 16M alignment for crash kernel regions */
406 #define CRASH_ALIGN		SZ_16M
407 
408 /*
409  * Keep the crash kernel below this limit.
410  *
411  * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
412  * due to mapping restrictions.
413  *
414  * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
415  * the upper limit of system RAM in 4-level paging mode. Since the kdump
416  * jump could be from 5-level paging to 4-level paging, the jump will fail if
417  * the kernel is put above 64 TB, and during the 1st kernel bootup there's
418  * no good way to detect the paging mode of the target kernel which will be
419  * loaded for dumping.
420  */
421 #ifdef CONFIG_X86_32
422 # define CRASH_ADDR_LOW_MAX	SZ_512M
423 # define CRASH_ADDR_HIGH_MAX	SZ_512M
424 #else
425 # define CRASH_ADDR_LOW_MAX	SZ_4G
426 # define CRASH_ADDR_HIGH_MAX	SZ_64T
427 #endif
428 
429 static int __init reserve_crashkernel_low(void)
430 {
431 #ifdef CONFIG_X86_64
432 	unsigned long long base, low_base = 0, low_size = 0;
433 	unsigned long total_low_mem;
434 	int ret;
435 
436 	total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
437 
438 	/* crashkernel=Y,low */
439 	ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
440 	if (ret) {
441 		/*
442 		 * two parts from kernel/dma/swiotlb.c:
443 		 * -swiotlb size: user-specified with swiotlb= or default.
444 		 *
445 		 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
446 		 * to 8M for other buffers that may need to stay low too. Also
447 		 * make sure we allocate enough extra low memory so that we
448 		 * don't run out of DMA buffers for 32-bit devices.
449 		 */
450 		low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
451 	} else {
452 		/* passed with crashkernel=0,low ? */
453 		if (!low_size)
454 			return 0;
455 	}
456 
457 	low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN);
458 	if (!low_base) {
459 		pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
460 		       (unsigned long)(low_size >> 20));
461 		return -ENOMEM;
462 	}
463 
464 	ret = memblock_reserve(low_base, low_size);
465 	if (ret) {
466 		pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
467 		return ret;
468 	}
469 
470 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
471 		(unsigned long)(low_size >> 20),
472 		(unsigned long)(low_base >> 20),
473 		(unsigned long)(total_low_mem >> 20));
474 
475 	crashk_low_res.start = low_base;
476 	crashk_low_res.end   = low_base + low_size - 1;
477 	insert_resource(&iomem_resource, &crashk_low_res);
478 #endif
479 	return 0;
480 }
481 
482 static void __init reserve_crashkernel(void)
483 {
484 	unsigned long long crash_size, crash_base, total_mem;
485 	bool high = false;
486 	int ret;
487 
488 	total_mem = memblock_phys_mem_size();
489 
490 	/* crashkernel=XM */
491 	ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
492 	if (ret != 0 || crash_size <= 0) {
493 		/* crashkernel=X,high */
494 		ret = parse_crashkernel_high(boot_command_line, total_mem,
495 					     &crash_size, &crash_base);
496 		if (ret != 0 || crash_size <= 0)
497 			return;
498 		high = true;
499 	}
500 
501 	if (xen_pv_domain()) {
502 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
503 		return;
504 	}
505 
506 	/* 0 means: find the address automatically */
507 	if (!crash_base) {
508 		/*
509 		 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
510 		 * crashkernel=x,high reserves memory over 4G, also allocates
511 		 * 256M extra low memory for DMA buffers and swiotlb.
512 		 * But the extra memory is not required for all machines.
513 		 * So try low memory first and fall back to high memory
514 		 * unless "crashkernel=size[KMG],high" is specified.
515 		 */
516 		if (!high)
517 			crash_base = memblock_find_in_range(CRASH_ALIGN,
518 						CRASH_ADDR_LOW_MAX,
519 						crash_size, CRASH_ALIGN);
520 		if (!crash_base)
521 			crash_base = memblock_find_in_range(CRASH_ALIGN,
522 						CRASH_ADDR_HIGH_MAX,
523 						crash_size, CRASH_ALIGN);
524 		if (!crash_base) {
525 			pr_info("crashkernel reservation failed - No suitable area found.\n");
526 			return;
527 		}
528 	} else {
529 		unsigned long long start;
530 
531 		start = memblock_find_in_range(crash_base,
532 					       crash_base + crash_size,
533 					       crash_size, 1 << 20);
534 		if (start != crash_base) {
535 			pr_info("crashkernel reservation failed - memory is in use.\n");
536 			return;
537 		}
538 	}
539 	ret = memblock_reserve(crash_base, crash_size);
540 	if (ret) {
541 		pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
542 		return;
543 	}
544 
545 	if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
546 		memblock_free(crash_base, crash_size);
547 		return;
548 	}
549 
550 	pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
551 		(unsigned long)(crash_size >> 20),
552 		(unsigned long)(crash_base >> 20),
553 		(unsigned long)(total_mem >> 20));
554 
555 	crashk_res.start = crash_base;
556 	crashk_res.end   = crash_base + crash_size - 1;
557 	insert_resource(&iomem_resource, &crashk_res);
558 }
559 #else
560 static void __init reserve_crashkernel(void)
561 {
562 }
563 #endif
564 
565 static struct resource standard_io_resources[] = {
566 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
567 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
568 	{ .name = "pic1", .start = 0x20, .end = 0x21,
569 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
570 	{ .name = "timer0", .start = 0x40, .end = 0x43,
571 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
572 	{ .name = "timer1", .start = 0x50, .end = 0x53,
573 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
574 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
575 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
576 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
577 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
578 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
579 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
580 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
581 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
582 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
583 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
584 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
585 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
586 };
587 
588 void __init reserve_standard_io_resources(void)
589 {
590 	int i;
591 
592 	/* request I/O space for devices used on all i[345]86 PCs */
593 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
594 		request_resource(&ioport_resource, &standard_io_resources[i]);
595 
596 }
597 
598 static __init void reserve_ibft_region(void)
599 {
600 	unsigned long addr, size = 0;
601 
602 	addr = find_ibft_region(&size);
603 
604 	if (size)
605 		memblock_reserve(addr, size);
606 }
607 
608 static bool __init snb_gfx_workaround_needed(void)
609 {
610 #ifdef CONFIG_PCI
611 	int i;
612 	u16 vendor, devid;
613 	static const __initconst u16 snb_ids[] = {
614 		0x0102,
615 		0x0112,
616 		0x0122,
617 		0x0106,
618 		0x0116,
619 		0x0126,
620 		0x010a,
621 	};
622 
623 	/* Assume no if something weird is going on with PCI */
624 	if (!early_pci_allowed())
625 		return false;
626 
627 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
628 	if (vendor != 0x8086)
629 		return false;
630 
631 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
632 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
633 		if (devid == snb_ids[i])
634 			return true;
635 #endif
636 
637 	return false;
638 }
639 
640 /*
641  * Sandy Bridge graphics has trouble with certain ranges, exclude
642  * them from allocation.
643  */
644 static void __init trim_snb_memory(void)
645 {
646 	static const __initconst unsigned long bad_pages[] = {
647 		0x20050000,
648 		0x20110000,
649 		0x20130000,
650 		0x20138000,
651 		0x40004000,
652 	};
653 	int i;
654 
655 	if (!snb_gfx_workaround_needed())
656 		return;
657 
658 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
659 
660 	/*
661 	 * Reserve all memory below the 1 MB mark that has not
662 	 * already been reserved.
663 	 */
664 	memblock_reserve(0, 1<<20);
665 
666 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
667 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
668 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
669 			       bad_pages[i]);
670 	}
671 }
672 
673 /*
674  * Here we put platform-specific memory range workarounds, i.e.
675  * memory known to be corrupt or otherwise in need to be reserved on
676  * specific platforms.
677  *
678  * If this gets used more widely it could use a real dispatch mechanism.
679  */
680 static void __init trim_platform_memory_ranges(void)
681 {
682 	trim_snb_memory();
683 }
684 
685 static void __init trim_bios_range(void)
686 {
687 	/*
688 	 * A special case is the first 4Kb of memory;
689 	 * This is a BIOS owned area, not kernel ram, but generally
690 	 * not listed as such in the E820 table.
691 	 *
692 	 * This typically reserves additional memory (64KiB by default)
693 	 * since some BIOSes are known to corrupt low memory.  See the
694 	 * Kconfig help text for X86_RESERVE_LOW.
695 	 */
696 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
697 
698 	/*
699 	 * special case: Some BIOSes report the PC BIOS
700 	 * area (640Kb -> 1Mb) as RAM even though it is not.
701 	 * take them out.
702 	 */
703 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
704 
705 	e820__update_table(e820_table);
706 }
707 
708 /* called before trim_bios_range() to spare extra sanitize */
709 static void __init e820_add_kernel_range(void)
710 {
711 	u64 start = __pa_symbol(_text);
712 	u64 size = __pa_symbol(_end) - start;
713 
714 	/*
715 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
716 	 * attempt to fix it by adding the range. We may have a confused BIOS,
717 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
718 	 * exclude kernel range. If we really are running on top non-RAM,
719 	 * we will crash later anyways.
720 	 */
721 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
722 		return;
723 
724 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
725 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
726 	e820__range_add(start, size, E820_TYPE_RAM);
727 }
728 
729 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
730 
731 static int __init parse_reservelow(char *p)
732 {
733 	unsigned long long size;
734 
735 	if (!p)
736 		return -EINVAL;
737 
738 	size = memparse(p, &p);
739 
740 	if (size < 4096)
741 		size = 4096;
742 
743 	if (size > 640*1024)
744 		size = 640*1024;
745 
746 	reserve_low = size;
747 
748 	return 0;
749 }
750 
751 early_param("reservelow", parse_reservelow);
752 
753 static void __init trim_low_memory_range(void)
754 {
755 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
756 }
757 
758 /*
759  * Dump out kernel offset information on panic.
760  */
761 static int
762 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
763 {
764 	if (kaslr_enabled()) {
765 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
766 			 kaslr_offset(),
767 			 __START_KERNEL,
768 			 __START_KERNEL_map,
769 			 MODULES_VADDR-1);
770 	} else {
771 		pr_emerg("Kernel Offset: disabled\n");
772 	}
773 
774 	return 0;
775 }
776 
777 /*
778  * Determine if we were loaded by an EFI loader.  If so, then we have also been
779  * passed the efi memmap, systab, etc., so we should use these data structures
780  * for initialization.  Note, the efi init code path is determined by the
781  * global efi_enabled. This allows the same kernel image to be used on existing
782  * systems (with a traditional BIOS) as well as on EFI systems.
783  */
784 /*
785  * setup_arch - architecture-specific boot-time initializations
786  *
787  * Note: On x86_64, fixmaps are ready for use even before this is called.
788  */
789 
790 void __init setup_arch(char **cmdline_p)
791 {
792 	/*
793 	 * Reserve the memory occupied by the kernel between _text and
794 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
795 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
796 	 * separate memblock_reserve() or they will be discarded.
797 	 */
798 	memblock_reserve(__pa_symbol(_text),
799 			 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
800 
801 	/*
802 	 * Make sure page 0 is always reserved because on systems with
803 	 * L1TF its contents can be leaked to user processes.
804 	 */
805 	memblock_reserve(0, PAGE_SIZE);
806 
807 	early_reserve_initrd();
808 
809 	/*
810 	 * At this point everything still needed from the boot loader
811 	 * or BIOS or kernel text should be early reserved or marked not
812 	 * RAM in e820. All other memory is free game.
813 	 */
814 
815 #ifdef CONFIG_X86_32
816 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
817 
818 	/*
819 	 * copy kernel address range established so far and switch
820 	 * to the proper swapper page table
821 	 */
822 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
823 			initial_page_table + KERNEL_PGD_BOUNDARY,
824 			KERNEL_PGD_PTRS);
825 
826 	load_cr3(swapper_pg_dir);
827 	/*
828 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
829 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
830 	 * will not flush anything because the CPU quirk which clears
831 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
832 	 * load_cr3() above the TLB has been flushed already. The
833 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
834 	 * so proper operation is guaranteed.
835 	 */
836 	__flush_tlb_all();
837 #else
838 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
839 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
840 #endif
841 
842 	/*
843 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
844 	 * reserve_top(), so do this before touching the ioremap area.
845 	 */
846 	olpc_ofw_detect();
847 
848 	idt_setup_early_traps();
849 	early_cpu_init();
850 	arch_init_ideal_nops();
851 	jump_label_init();
852 	early_ioremap_init();
853 
854 	setup_olpc_ofw_pgd();
855 
856 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
857 	screen_info = boot_params.screen_info;
858 	edid_info = boot_params.edid_info;
859 #ifdef CONFIG_X86_32
860 	apm_info.bios = boot_params.apm_bios_info;
861 	ist_info = boot_params.ist_info;
862 #endif
863 	saved_video_mode = boot_params.hdr.vid_mode;
864 	bootloader_type = boot_params.hdr.type_of_loader;
865 	if ((bootloader_type >> 4) == 0xe) {
866 		bootloader_type &= 0xf;
867 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
868 	}
869 	bootloader_version  = bootloader_type & 0xf;
870 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
871 
872 #ifdef CONFIG_BLK_DEV_RAM
873 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
874 #endif
875 #ifdef CONFIG_EFI
876 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
877 		     EFI32_LOADER_SIGNATURE, 4)) {
878 		set_bit(EFI_BOOT, &efi.flags);
879 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
880 		     EFI64_LOADER_SIGNATURE, 4)) {
881 		set_bit(EFI_BOOT, &efi.flags);
882 		set_bit(EFI_64BIT, &efi.flags);
883 	}
884 #endif
885 
886 	x86_init.oem.arch_setup();
887 
888 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
889 	e820__memory_setup();
890 	parse_setup_data();
891 
892 	copy_edd();
893 
894 	if (!boot_params.hdr.root_flags)
895 		root_mountflags &= ~MS_RDONLY;
896 	init_mm.start_code = (unsigned long) _text;
897 	init_mm.end_code = (unsigned long) _etext;
898 	init_mm.end_data = (unsigned long) _edata;
899 	init_mm.brk = _brk_end;
900 
901 	code_resource.start = __pa_symbol(_text);
902 	code_resource.end = __pa_symbol(_etext)-1;
903 	rodata_resource.start = __pa_symbol(__start_rodata);
904 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
905 	data_resource.start = __pa_symbol(_sdata);
906 	data_resource.end = __pa_symbol(_edata)-1;
907 	bss_resource.start = __pa_symbol(__bss_start);
908 	bss_resource.end = __pa_symbol(__bss_stop)-1;
909 
910 #ifdef CONFIG_CMDLINE_BOOL
911 #ifdef CONFIG_CMDLINE_OVERRIDE
912 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
913 #else
914 	if (builtin_cmdline[0]) {
915 		/* append boot loader cmdline to builtin */
916 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
917 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
918 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
919 	}
920 #endif
921 #endif
922 
923 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
924 	*cmdline_p = command_line;
925 
926 	/*
927 	 * x86_configure_nx() is called before parse_early_param() to detect
928 	 * whether hardware doesn't support NX (so that the early EHCI debug
929 	 * console setup can safely call set_fixmap()). It may then be called
930 	 * again from within noexec_setup() during parsing early parameters
931 	 * to honor the respective command line option.
932 	 */
933 	x86_configure_nx();
934 
935 	parse_early_param();
936 
937 	if (efi_enabled(EFI_BOOT))
938 		efi_memblock_x86_reserve_range();
939 #ifdef CONFIG_MEMORY_HOTPLUG
940 	/*
941 	 * Memory used by the kernel cannot be hot-removed because Linux
942 	 * cannot migrate the kernel pages. When memory hotplug is
943 	 * enabled, we should prevent memblock from allocating memory
944 	 * for the kernel.
945 	 *
946 	 * ACPI SRAT records all hotpluggable memory ranges. But before
947 	 * SRAT is parsed, we don't know about it.
948 	 *
949 	 * The kernel image is loaded into memory at very early time. We
950 	 * cannot prevent this anyway. So on NUMA system, we set any
951 	 * node the kernel resides in as un-hotpluggable.
952 	 *
953 	 * Since on modern servers, one node could have double-digit
954 	 * gigabytes memory, we can assume the memory around the kernel
955 	 * image is also un-hotpluggable. So before SRAT is parsed, just
956 	 * allocate memory near the kernel image to try the best to keep
957 	 * the kernel away from hotpluggable memory.
958 	 */
959 	if (movable_node_is_enabled())
960 		memblock_set_bottom_up(true);
961 #endif
962 
963 	x86_report_nx();
964 
965 	/* after early param, so could get panic from serial */
966 	memblock_x86_reserve_range_setup_data();
967 
968 	if (acpi_mps_check()) {
969 #ifdef CONFIG_X86_LOCAL_APIC
970 		disable_apic = 1;
971 #endif
972 		setup_clear_cpu_cap(X86_FEATURE_APIC);
973 	}
974 
975 	e820__reserve_setup_data();
976 	e820__finish_early_params();
977 
978 	if (efi_enabled(EFI_BOOT))
979 		efi_init();
980 
981 	dmi_setup();
982 
983 	/*
984 	 * VMware detection requires dmi to be available, so this
985 	 * needs to be done after dmi_setup(), for the boot CPU.
986 	 */
987 	init_hypervisor_platform();
988 
989 	tsc_early_init();
990 	x86_init.resources.probe_roms();
991 
992 	/* after parse_early_param, so could debug it */
993 	insert_resource(&iomem_resource, &code_resource);
994 	insert_resource(&iomem_resource, &rodata_resource);
995 	insert_resource(&iomem_resource, &data_resource);
996 	insert_resource(&iomem_resource, &bss_resource);
997 
998 	e820_add_kernel_range();
999 	trim_bios_range();
1000 #ifdef CONFIG_X86_32
1001 	if (ppro_with_ram_bug()) {
1002 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1003 				  E820_TYPE_RESERVED);
1004 		e820__update_table(e820_table);
1005 		printk(KERN_INFO "fixed physical RAM map:\n");
1006 		e820__print_table("bad_ppro");
1007 	}
1008 #else
1009 	early_gart_iommu_check();
1010 #endif
1011 
1012 	/*
1013 	 * partially used pages are not usable - thus
1014 	 * we are rounding upwards:
1015 	 */
1016 	max_pfn = e820__end_of_ram_pfn();
1017 
1018 	/* update e820 for memory not covered by WB MTRRs */
1019 	mtrr_bp_init();
1020 	if (mtrr_trim_uncached_memory(max_pfn))
1021 		max_pfn = e820__end_of_ram_pfn();
1022 
1023 	max_possible_pfn = max_pfn;
1024 
1025 	/*
1026 	 * This call is required when the CPU does not support PAT. If
1027 	 * mtrr_bp_init() invoked it already via pat_init() the call has no
1028 	 * effect.
1029 	 */
1030 	init_cache_modes();
1031 
1032 	/*
1033 	 * Define random base addresses for memory sections after max_pfn is
1034 	 * defined and before each memory section base is used.
1035 	 */
1036 	kernel_randomize_memory();
1037 
1038 #ifdef CONFIG_X86_32
1039 	/* max_low_pfn get updated here */
1040 	find_low_pfn_range();
1041 #else
1042 	check_x2apic();
1043 
1044 	/* How many end-of-memory variables you have, grandma! */
1045 	/* need this before calling reserve_initrd */
1046 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1047 		max_low_pfn = e820__end_of_low_ram_pfn();
1048 	else
1049 		max_low_pfn = max_pfn;
1050 
1051 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1052 #endif
1053 
1054 	/*
1055 	 * Find and reserve possible boot-time SMP configuration:
1056 	 */
1057 	find_smp_config();
1058 
1059 	reserve_ibft_region();
1060 
1061 	early_alloc_pgt_buf();
1062 
1063 	/*
1064 	 * Need to conclude brk, before e820__memblock_setup()
1065 	 *  it could use memblock_find_in_range, could overlap with
1066 	 *  brk area.
1067 	 */
1068 	reserve_brk();
1069 
1070 	cleanup_highmap();
1071 
1072 	memblock_set_current_limit(ISA_END_ADDRESS);
1073 	e820__memblock_setup();
1074 
1075 	reserve_bios_regions();
1076 
1077 	efi_fake_memmap();
1078 	efi_find_mirror();
1079 	efi_esrt_init();
1080 
1081 	/*
1082 	 * The EFI specification says that boot service code won't be
1083 	 * called after ExitBootServices(). This is, in fact, a lie.
1084 	 */
1085 	efi_reserve_boot_services();
1086 
1087 	/* preallocate 4k for mptable mpc */
1088 	e820__memblock_alloc_reserved_mpc_new();
1089 
1090 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1091 	setup_bios_corruption_check();
1092 #endif
1093 
1094 #ifdef CONFIG_X86_32
1095 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1096 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1097 #endif
1098 
1099 	reserve_real_mode();
1100 
1101 	trim_platform_memory_ranges();
1102 	trim_low_memory_range();
1103 
1104 	init_mem_mapping();
1105 
1106 	idt_setup_early_pf();
1107 
1108 	/*
1109 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1110 	 * with the current CR4 value.  This may not be necessary, but
1111 	 * auditing all the early-boot CR4 manipulation would be needed to
1112 	 * rule it out.
1113 	 *
1114 	 * Mask off features that don't work outside long mode (just
1115 	 * PCIDE for now).
1116 	 */
1117 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1118 
1119 	memblock_set_current_limit(get_max_mapped());
1120 
1121 	/*
1122 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1123 	 */
1124 
1125 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1126 	if (init_ohci1394_dma_early)
1127 		init_ohci1394_dma_on_all_controllers();
1128 #endif
1129 	/* Allocate bigger log buffer */
1130 	setup_log_buf(1);
1131 
1132 	if (efi_enabled(EFI_BOOT)) {
1133 		switch (boot_params.secure_boot) {
1134 		case efi_secureboot_mode_disabled:
1135 			pr_info("Secure boot disabled\n");
1136 			break;
1137 		case efi_secureboot_mode_enabled:
1138 			pr_info("Secure boot enabled\n");
1139 			break;
1140 		default:
1141 			pr_info("Secure boot could not be determined\n");
1142 			break;
1143 		}
1144 	}
1145 
1146 	reserve_initrd();
1147 
1148 	acpi_table_upgrade();
1149 
1150 	vsmp_init();
1151 
1152 	io_delay_init();
1153 
1154 	early_platform_quirks();
1155 
1156 	/*
1157 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1158 	 */
1159 	acpi_boot_table_init();
1160 
1161 	early_acpi_boot_init();
1162 
1163 	initmem_init();
1164 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1165 
1166 	if (boot_cpu_has(X86_FEATURE_GBPAGES))
1167 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1168 
1169 	/*
1170 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1171 	 * won't consume hotpluggable memory.
1172 	 */
1173 	reserve_crashkernel();
1174 
1175 	memblock_find_dma_reserve();
1176 
1177 	if (!early_xdbc_setup_hardware())
1178 		early_xdbc_register_console();
1179 
1180 	x86_init.paging.pagetable_init();
1181 
1182 	kasan_init();
1183 
1184 	/*
1185 	 * Sync back kernel address range.
1186 	 *
1187 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1188 	 * this call?
1189 	 */
1190 	sync_initial_page_table();
1191 
1192 	tboot_probe();
1193 
1194 	map_vsyscall();
1195 
1196 	generic_apic_probe();
1197 
1198 	early_quirks();
1199 
1200 	/*
1201 	 * Read APIC and some other early information from ACPI tables.
1202 	 */
1203 	acpi_boot_init();
1204 	sfi_init();
1205 	x86_dtb_init();
1206 
1207 	/*
1208 	 * get boot-time SMP configuration:
1209 	 */
1210 	get_smp_config();
1211 
1212 	/*
1213 	 * Systems w/o ACPI and mptables might not have it mapped the local
1214 	 * APIC yet, but prefill_possible_map() might need to access it.
1215 	 */
1216 	init_apic_mappings();
1217 
1218 	prefill_possible_map();
1219 
1220 	init_cpu_to_node();
1221 
1222 	io_apic_init_mappings();
1223 
1224 	x86_init.hyper.guest_late_init();
1225 
1226 	e820__reserve_resources();
1227 	e820__register_nosave_regions(max_pfn);
1228 
1229 	x86_init.resources.reserve_resources();
1230 
1231 	e820__setup_pci_gap();
1232 
1233 #ifdef CONFIG_VT
1234 #if defined(CONFIG_VGA_CONSOLE)
1235 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1236 		conswitchp = &vga_con;
1237 #endif
1238 #endif
1239 	x86_init.oem.banner();
1240 
1241 	x86_init.timers.wallclock_init();
1242 
1243 	mcheck_init();
1244 
1245 	register_refined_jiffies(CLOCK_TICK_RATE);
1246 
1247 #ifdef CONFIG_EFI
1248 	if (efi_enabled(EFI_BOOT))
1249 		efi_apply_memmap_quirks();
1250 #endif
1251 
1252 	unwind_init();
1253 }
1254 
1255 #ifdef CONFIG_X86_32
1256 
1257 static struct resource video_ram_resource = {
1258 	.name	= "Video RAM area",
1259 	.start	= 0xa0000,
1260 	.end	= 0xbffff,
1261 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1262 };
1263 
1264 void __init i386_reserve_resources(void)
1265 {
1266 	request_resource(&iomem_resource, &video_ram_resource);
1267 	reserve_standard_io_resources();
1268 }
1269 
1270 #endif /* CONFIG_X86_32 */
1271 
1272 static struct notifier_block kernel_offset_notifier = {
1273 	.notifier_call = dump_kernel_offset
1274 };
1275 
1276 static int __init register_kernel_offset_dumper(void)
1277 {
1278 	atomic_notifier_chain_register(&panic_notifier_list,
1279 					&kernel_offset_notifier);
1280 	return 0;
1281 }
1282 __initcall(register_kernel_offset_dumper);
1283