xref: /openbmc/linux/arch/x86/kernel/setup.c (revision 3e30a927)
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/tboot.h>
20 #include <linux/usb/xhci-dbgp.h>
21 
22 #include <uapi/linux/mount.h>
23 
24 #include <xen/xen.h>
25 
26 #include <asm/apic.h>
27 #include <asm/bios_ebda.h>
28 #include <asm/bugs.h>
29 #include <asm/cpu.h>
30 #include <asm/efi.h>
31 #include <asm/gart.h>
32 #include <asm/hypervisor.h>
33 #include <asm/io_apic.h>
34 #include <asm/kasan.h>
35 #include <asm/kaslr.h>
36 #include <asm/mce.h>
37 #include <asm/mtrr.h>
38 #include <asm/realmode.h>
39 #include <asm/olpc_ofw.h>
40 #include <asm/pci-direct.h>
41 #include <asm/prom.h>
42 #include <asm/proto.h>
43 #include <asm/unwind.h>
44 #include <asm/vsyscall.h>
45 #include <linux/vmalloc.h>
46 
47 /*
48  * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
49  * max_pfn_mapped:     highest directly mapped pfn > 4 GB
50  *
51  * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
52  * represented by pfn_mapped[].
53  */
54 unsigned long max_low_pfn_mapped;
55 unsigned long max_pfn_mapped;
56 
57 #ifdef CONFIG_DMI
58 RESERVE_BRK(dmi_alloc, 65536);
59 #endif
60 
61 
62 /*
63  * Range of the BSS area. The size of the BSS area is determined
64  * at link time, with RESERVE_BRK*() facility reserving additional
65  * chunks.
66  */
67 static __initdata
68 unsigned long _brk_start = (unsigned long)__brk_base;
69 unsigned long _brk_end   = (unsigned long)__brk_base;
70 
71 struct boot_params boot_params;
72 
73 /*
74  * These are the four main kernel memory regions, we put them into
75  * the resource tree so that kdump tools and other debugging tools
76  * recover it:
77  */
78 
79 static struct resource rodata_resource = {
80 	.name	= "Kernel rodata",
81 	.start	= 0,
82 	.end	= 0,
83 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
84 };
85 
86 static struct resource data_resource = {
87 	.name	= "Kernel data",
88 	.start	= 0,
89 	.end	= 0,
90 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
91 };
92 
93 static struct resource code_resource = {
94 	.name	= "Kernel code",
95 	.start	= 0,
96 	.end	= 0,
97 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
98 };
99 
100 static struct resource bss_resource = {
101 	.name	= "Kernel bss",
102 	.start	= 0,
103 	.end	= 0,
104 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
105 };
106 
107 
108 #ifdef CONFIG_X86_32
109 /* CPU data as detected by the assembly code in head_32.S */
110 struct cpuinfo_x86 new_cpu_data;
111 
112 /* Common CPU data for all CPUs */
113 struct cpuinfo_x86 boot_cpu_data __read_mostly;
114 EXPORT_SYMBOL(boot_cpu_data);
115 
116 unsigned int def_to_bigsmp;
117 
118 /* For MCA, but anyone else can use it if they want */
119 unsigned int machine_id;
120 unsigned int machine_submodel_id;
121 unsigned int BIOS_revision;
122 
123 struct apm_info apm_info;
124 EXPORT_SYMBOL(apm_info);
125 
126 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
127 	defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
128 struct ist_info ist_info;
129 EXPORT_SYMBOL(ist_info);
130 #else
131 struct ist_info ist_info;
132 #endif
133 
134 #else
135 struct cpuinfo_x86 boot_cpu_data __read_mostly;
136 EXPORT_SYMBOL(boot_cpu_data);
137 #endif
138 
139 
140 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
141 __visible unsigned long mmu_cr4_features __ro_after_init;
142 #else
143 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
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 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
168 #endif
169 
170 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
171 struct edd edd;
172 #ifdef CONFIG_EDD_MODULE
173 EXPORT_SYMBOL(edd);
174 #endif
175 /**
176  * copy_edd() - Copy the BIOS EDD information
177  *              from boot_params into a safe place.
178  *
179  */
180 static inline void __init copy_edd(void)
181 {
182      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
183 	    sizeof(edd.mbr_signature));
184      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
185      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
186      edd.edd_info_nr = boot_params.eddbuf_entries;
187 }
188 #else
189 static inline void __init copy_edd(void)
190 {
191 }
192 #endif
193 
194 void * __init extend_brk(size_t size, size_t align)
195 {
196 	size_t mask = align - 1;
197 	void *ret;
198 
199 	BUG_ON(_brk_start == 0);
200 	BUG_ON(align & mask);
201 
202 	_brk_end = (_brk_end + mask) & ~mask;
203 	BUG_ON((char *)(_brk_end + size) > __brk_limit);
204 
205 	ret = (void *)_brk_end;
206 	_brk_end += size;
207 
208 	memset(ret, 0, size);
209 
210 	return ret;
211 }
212 
213 #ifdef CONFIG_X86_32
214 static void __init cleanup_highmap(void)
215 {
216 }
217 #endif
218 
219 static void __init reserve_brk(void)
220 {
221 	if (_brk_end > _brk_start)
222 		memblock_reserve(__pa_symbol(_brk_start),
223 				 _brk_end - _brk_start);
224 
225 	/* Mark brk area as locked down and no longer taking any
226 	   new allocations */
227 	_brk_start = 0;
228 }
229 
230 u64 relocated_ramdisk;
231 
232 #ifdef CONFIG_BLK_DEV_INITRD
233 
234 static u64 __init get_ramdisk_image(void)
235 {
236 	u64 ramdisk_image = boot_params.hdr.ramdisk_image;
237 
238 	ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
239 
240 	return ramdisk_image;
241 }
242 static u64 __init get_ramdisk_size(void)
243 {
244 	u64 ramdisk_size = boot_params.hdr.ramdisk_size;
245 
246 	ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
247 
248 	return ramdisk_size;
249 }
250 
251 static void __init relocate_initrd(void)
252 {
253 	/* Assume only end is not page aligned */
254 	u64 ramdisk_image = get_ramdisk_image();
255 	u64 ramdisk_size  = get_ramdisk_size();
256 	u64 area_size     = PAGE_ALIGN(ramdisk_size);
257 
258 	/* We need to move the initrd down into directly mapped mem */
259 	relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
260 						   area_size, PAGE_SIZE);
261 
262 	if (!relocated_ramdisk)
263 		panic("Cannot find place for new RAMDISK of size %lld\n",
264 		      ramdisk_size);
265 
266 	/* Note: this includes all the mem currently occupied by
267 	   the initrd, we rely on that fact to keep the data intact. */
268 	memblock_reserve(relocated_ramdisk, area_size);
269 	initrd_start = relocated_ramdisk + PAGE_OFFSET;
270 	initrd_end   = initrd_start + ramdisk_size;
271 	printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
272 	       relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
273 
274 	copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
275 
276 	printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
277 		" [mem %#010llx-%#010llx]\n",
278 		ramdisk_image, ramdisk_image + ramdisk_size - 1,
279 		relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
280 }
281 
282 static void __init early_reserve_initrd(void)
283 {
284 	/* Assume only end is not page aligned */
285 	u64 ramdisk_image = get_ramdisk_image();
286 	u64 ramdisk_size  = get_ramdisk_size();
287 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
288 
289 	if (!boot_params.hdr.type_of_loader ||
290 	    !ramdisk_image || !ramdisk_size)
291 		return;		/* No initrd provided by bootloader */
292 
293 	memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
294 }
295 static void __init reserve_initrd(void)
296 {
297 	/* Assume only end is not page aligned */
298 	u64 ramdisk_image = get_ramdisk_image();
299 	u64 ramdisk_size  = get_ramdisk_size();
300 	u64 ramdisk_end   = PAGE_ALIGN(ramdisk_image + ramdisk_size);
301 	u64 mapped_size;
302 
303 	if (!boot_params.hdr.type_of_loader ||
304 	    !ramdisk_image || !ramdisk_size)
305 		return;		/* No initrd provided by bootloader */
306 
307 	initrd_start = 0;
308 
309 	mapped_size = memblock_mem_size(max_pfn_mapped);
310 	if (ramdisk_size >= (mapped_size>>1))
311 		panic("initrd too large to handle, "
312 		       "disabling initrd (%lld needed, %lld available)\n",
313 		       ramdisk_size, mapped_size>>1);
314 
315 	printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
316 			ramdisk_end - 1);
317 
318 	if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
319 				PFN_DOWN(ramdisk_end))) {
320 		/* All are mapped, easy case */
321 		initrd_start = ramdisk_image + PAGE_OFFSET;
322 		initrd_end = initrd_start + ramdisk_size;
323 		return;
324 	}
325 
326 	relocate_initrd();
327 
328 	memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
329 }
330 
331 #else
332 static void __init early_reserve_initrd(void)
333 {
334 }
335 static void __init reserve_initrd(void)
336 {
337 }
338 #endif /* CONFIG_BLK_DEV_INITRD */
339 
340 static void __init parse_setup_data(void)
341 {
342 	struct setup_data *data;
343 	u64 pa_data, pa_next;
344 
345 	pa_data = boot_params.hdr.setup_data;
346 	while (pa_data) {
347 		u32 data_len, data_type;
348 
349 		data = early_memremap(pa_data, sizeof(*data));
350 		data_len = data->len + sizeof(struct setup_data);
351 		data_type = data->type;
352 		pa_next = data->next;
353 		early_memunmap(data, sizeof(*data));
354 
355 		switch (data_type) {
356 		case SETUP_E820_EXT:
357 			e820__memory_setup_extended(pa_data, data_len);
358 			break;
359 		case SETUP_DTB:
360 			add_dtb(pa_data);
361 			break;
362 		case SETUP_EFI:
363 			parse_efi_setup(pa_data, data_len);
364 			break;
365 		default:
366 			break;
367 		}
368 		pa_data = pa_next;
369 	}
370 }
371 
372 static void __init memblock_x86_reserve_range_setup_data(void)
373 {
374 	struct setup_data *data;
375 	u64 pa_data;
376 
377 	pa_data = boot_params.hdr.setup_data;
378 	while (pa_data) {
379 		data = early_memremap(pa_data, sizeof(*data));
380 		memblock_reserve(pa_data, sizeof(*data) + data->len);
381 
382 		if (data->type == SETUP_INDIRECT &&
383 		    ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
384 			memblock_reserve(((struct setup_indirect *)data->data)->addr,
385 					 ((struct setup_indirect *)data->data)->len);
386 
387 		pa_data = data->next;
388 		early_memunmap(data, sizeof(*data));
389 	}
390 }
391 
392 /*
393  * --------- Crashkernel reservation ------------------------------
394  */
395 
396 #ifdef CONFIG_KEXEC_CORE
397 
398 /* 16M alignment for crash kernel regions */
399 #define CRASH_ALIGN		SZ_16M
400 
401 /*
402  * Keep the crash kernel below this limit.
403  *
404  * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
405  * due to mapping restrictions.
406  *
407  * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
408  * the upper limit of system RAM in 4-level paging mode. Since the kdump
409  * jump could be from 5-level paging to 4-level paging, the jump will fail if
410  * the kernel is put above 64 TB, and during the 1st kernel bootup there's
411  * no good way to detect the paging mode of the target kernel which will be
412  * loaded for dumping.
413  */
414 #ifdef CONFIG_X86_32
415 # define CRASH_ADDR_LOW_MAX	SZ_512M
416 # define CRASH_ADDR_HIGH_MAX	SZ_512M
417 #else
418 # define CRASH_ADDR_LOW_MAX	SZ_4G
419 # define CRASH_ADDR_HIGH_MAX	SZ_64T
420 #endif
421 
422 static int __init reserve_crashkernel_low(void)
423 {
424 #ifdef CONFIG_X86_64
425 	unsigned long long base, low_base = 0, low_size = 0;
426 	unsigned long total_low_mem;
427 	int ret;
428 
429 	total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
430 
431 	/* crashkernel=Y,low */
432 	ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
433 	if (ret) {
434 		/*
435 		 * two parts from kernel/dma/swiotlb.c:
436 		 * -swiotlb size: user-specified with swiotlb= or default.
437 		 *
438 		 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
439 		 * to 8M for other buffers that may need to stay low too. Also
440 		 * make sure we allocate enough extra low memory so that we
441 		 * don't run out of DMA buffers for 32-bit devices.
442 		 */
443 		low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
444 	} else {
445 		/* passed with crashkernel=0,low ? */
446 		if (!low_size)
447 			return 0;
448 	}
449 
450 	low_base = memblock_find_in_range(0, 1ULL << 32, low_size, CRASH_ALIGN);
451 	if (!low_base) {
452 		pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
453 		       (unsigned long)(low_size >> 20));
454 		return -ENOMEM;
455 	}
456 
457 	ret = memblock_reserve(low_base, low_size);
458 	if (ret) {
459 		pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
460 		return ret;
461 	}
462 
463 	pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
464 		(unsigned long)(low_size >> 20),
465 		(unsigned long)(low_base >> 20),
466 		(unsigned long)(total_low_mem >> 20));
467 
468 	crashk_low_res.start = low_base;
469 	crashk_low_res.end   = low_base + low_size - 1;
470 	insert_resource(&iomem_resource, &crashk_low_res);
471 #endif
472 	return 0;
473 }
474 
475 static void __init reserve_crashkernel(void)
476 {
477 	unsigned long long crash_size, crash_base, total_mem;
478 	bool high = false;
479 	int ret;
480 
481 	total_mem = memblock_phys_mem_size();
482 
483 	/* crashkernel=XM */
484 	ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
485 	if (ret != 0 || crash_size <= 0) {
486 		/* crashkernel=X,high */
487 		ret = parse_crashkernel_high(boot_command_line, total_mem,
488 					     &crash_size, &crash_base);
489 		if (ret != 0 || crash_size <= 0)
490 			return;
491 		high = true;
492 	}
493 
494 	if (xen_pv_domain()) {
495 		pr_info("Ignoring crashkernel for a Xen PV domain\n");
496 		return;
497 	}
498 
499 	/* 0 means: find the address automatically */
500 	if (!crash_base) {
501 		/*
502 		 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
503 		 * crashkernel=x,high reserves memory over 4G, also allocates
504 		 * 256M extra low memory for DMA buffers and swiotlb.
505 		 * But the extra memory is not required for all machines.
506 		 * So try low memory first and fall back to high memory
507 		 * unless "crashkernel=size[KMG],high" is specified.
508 		 */
509 		if (!high)
510 			crash_base = memblock_find_in_range(CRASH_ALIGN,
511 						CRASH_ADDR_LOW_MAX,
512 						crash_size, CRASH_ALIGN);
513 		if (!crash_base)
514 			crash_base = memblock_find_in_range(CRASH_ALIGN,
515 						CRASH_ADDR_HIGH_MAX,
516 						crash_size, CRASH_ALIGN);
517 		if (!crash_base) {
518 			pr_info("crashkernel reservation failed - No suitable area found.\n");
519 			return;
520 		}
521 	} else {
522 		unsigned long long start;
523 
524 		start = memblock_find_in_range(crash_base,
525 					       crash_base + crash_size,
526 					       crash_size, 1 << 20);
527 		if (start != crash_base) {
528 			pr_info("crashkernel reservation failed - memory is in use.\n");
529 			return;
530 		}
531 	}
532 	ret = memblock_reserve(crash_base, crash_size);
533 	if (ret) {
534 		pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
535 		return;
536 	}
537 
538 	if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
539 		memblock_free(crash_base, crash_size);
540 		return;
541 	}
542 
543 	pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
544 		(unsigned long)(crash_size >> 20),
545 		(unsigned long)(crash_base >> 20),
546 		(unsigned long)(total_mem >> 20));
547 
548 	crashk_res.start = crash_base;
549 	crashk_res.end   = crash_base + crash_size - 1;
550 	insert_resource(&iomem_resource, &crashk_res);
551 }
552 #else
553 static void __init reserve_crashkernel(void)
554 {
555 }
556 #endif
557 
558 static struct resource standard_io_resources[] = {
559 	{ .name = "dma1", .start = 0x00, .end = 0x1f,
560 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
561 	{ .name = "pic1", .start = 0x20, .end = 0x21,
562 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
563 	{ .name = "timer0", .start = 0x40, .end = 0x43,
564 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
565 	{ .name = "timer1", .start = 0x50, .end = 0x53,
566 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
567 	{ .name = "keyboard", .start = 0x60, .end = 0x60,
568 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
569 	{ .name = "keyboard", .start = 0x64, .end = 0x64,
570 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
571 	{ .name = "dma page reg", .start = 0x80, .end = 0x8f,
572 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
573 	{ .name = "pic2", .start = 0xa0, .end = 0xa1,
574 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
575 	{ .name = "dma2", .start = 0xc0, .end = 0xdf,
576 		.flags = IORESOURCE_BUSY | IORESOURCE_IO },
577 	{ .name = "fpu", .start = 0xf0, .end = 0xff,
578 		.flags = IORESOURCE_BUSY | IORESOURCE_IO }
579 };
580 
581 void __init reserve_standard_io_resources(void)
582 {
583 	int i;
584 
585 	/* request I/O space for devices used on all i[345]86 PCs */
586 	for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
587 		request_resource(&ioport_resource, &standard_io_resources[i]);
588 
589 }
590 
591 static __init void reserve_ibft_region(void)
592 {
593 	unsigned long addr, size = 0;
594 
595 	addr = find_ibft_region(&size);
596 
597 	if (size)
598 		memblock_reserve(addr, size);
599 }
600 
601 static bool __init snb_gfx_workaround_needed(void)
602 {
603 #ifdef CONFIG_PCI
604 	int i;
605 	u16 vendor, devid;
606 	static const __initconst u16 snb_ids[] = {
607 		0x0102,
608 		0x0112,
609 		0x0122,
610 		0x0106,
611 		0x0116,
612 		0x0126,
613 		0x010a,
614 	};
615 
616 	/* Assume no if something weird is going on with PCI */
617 	if (!early_pci_allowed())
618 		return false;
619 
620 	vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
621 	if (vendor != 0x8086)
622 		return false;
623 
624 	devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
625 	for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
626 		if (devid == snb_ids[i])
627 			return true;
628 #endif
629 
630 	return false;
631 }
632 
633 /*
634  * Sandy Bridge graphics has trouble with certain ranges, exclude
635  * them from allocation.
636  */
637 static void __init trim_snb_memory(void)
638 {
639 	static const __initconst unsigned long bad_pages[] = {
640 		0x20050000,
641 		0x20110000,
642 		0x20130000,
643 		0x20138000,
644 		0x40004000,
645 	};
646 	int i;
647 
648 	if (!snb_gfx_workaround_needed())
649 		return;
650 
651 	printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
652 
653 	/*
654 	 * Reserve all memory below the 1 MB mark that has not
655 	 * already been reserved.
656 	 */
657 	memblock_reserve(0, 1<<20);
658 
659 	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
660 		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
661 			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
662 			       bad_pages[i]);
663 	}
664 }
665 
666 /*
667  * Here we put platform-specific memory range workarounds, i.e.
668  * memory known to be corrupt or otherwise in need to be reserved on
669  * specific platforms.
670  *
671  * If this gets used more widely it could use a real dispatch mechanism.
672  */
673 static void __init trim_platform_memory_ranges(void)
674 {
675 	trim_snb_memory();
676 }
677 
678 static void __init trim_bios_range(void)
679 {
680 	/*
681 	 * A special case is the first 4Kb of memory;
682 	 * This is a BIOS owned area, not kernel ram, but generally
683 	 * not listed as such in the E820 table.
684 	 *
685 	 * This typically reserves additional memory (64KiB by default)
686 	 * since some BIOSes are known to corrupt low memory.  See the
687 	 * Kconfig help text for X86_RESERVE_LOW.
688 	 */
689 	e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
690 
691 	/*
692 	 * special case: Some BIOSes report the PC BIOS
693 	 * area (640Kb -> 1Mb) as RAM even though it is not.
694 	 * take them out.
695 	 */
696 	e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
697 
698 	e820__update_table(e820_table);
699 }
700 
701 /* called before trim_bios_range() to spare extra sanitize */
702 static void __init e820_add_kernel_range(void)
703 {
704 	u64 start = __pa_symbol(_text);
705 	u64 size = __pa_symbol(_end) - start;
706 
707 	/*
708 	 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
709 	 * attempt to fix it by adding the range. We may have a confused BIOS,
710 	 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
711 	 * exclude kernel range. If we really are running on top non-RAM,
712 	 * we will crash later anyways.
713 	 */
714 	if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
715 		return;
716 
717 	pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
718 	e820__range_remove(start, size, E820_TYPE_RAM, 0);
719 	e820__range_add(start, size, E820_TYPE_RAM);
720 }
721 
722 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
723 
724 static int __init parse_reservelow(char *p)
725 {
726 	unsigned long long size;
727 
728 	if (!p)
729 		return -EINVAL;
730 
731 	size = memparse(p, &p);
732 
733 	if (size < 4096)
734 		size = 4096;
735 
736 	if (size > 640*1024)
737 		size = 640*1024;
738 
739 	reserve_low = size;
740 
741 	return 0;
742 }
743 
744 early_param("reservelow", parse_reservelow);
745 
746 static void __init trim_low_memory_range(void)
747 {
748 	memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
749 }
750 
751 /*
752  * Dump out kernel offset information on panic.
753  */
754 static int
755 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
756 {
757 	if (kaslr_enabled()) {
758 		pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
759 			 kaslr_offset(),
760 			 __START_KERNEL,
761 			 __START_KERNEL_map,
762 			 MODULES_VADDR-1);
763 	} else {
764 		pr_emerg("Kernel Offset: disabled\n");
765 	}
766 
767 	return 0;
768 }
769 
770 /*
771  * Determine if we were loaded by an EFI loader.  If so, then we have also been
772  * passed the efi memmap, systab, etc., so we should use these data structures
773  * for initialization.  Note, the efi init code path is determined by the
774  * global efi_enabled. This allows the same kernel image to be used on existing
775  * systems (with a traditional BIOS) as well as on EFI systems.
776  */
777 /*
778  * setup_arch - architecture-specific boot-time initializations
779  *
780  * Note: On x86_64, fixmaps are ready for use even before this is called.
781  */
782 
783 void __init setup_arch(char **cmdline_p)
784 {
785 	/*
786 	 * Reserve the memory occupied by the kernel between _text and
787 	 * __end_of_kernel_reserve symbols. Any kernel sections after the
788 	 * __end_of_kernel_reserve symbol must be explicitly reserved with a
789 	 * separate memblock_reserve() or they will be discarded.
790 	 */
791 	memblock_reserve(__pa_symbol(_text),
792 			 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
793 
794 	/*
795 	 * Make sure page 0 is always reserved because on systems with
796 	 * L1TF its contents can be leaked to user processes.
797 	 */
798 	memblock_reserve(0, PAGE_SIZE);
799 
800 	early_reserve_initrd();
801 
802 	/*
803 	 * At this point everything still needed from the boot loader
804 	 * or BIOS or kernel text should be early reserved or marked not
805 	 * RAM in e820. All other memory is free game.
806 	 */
807 
808 #ifdef CONFIG_X86_32
809 	memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
810 
811 	/*
812 	 * copy kernel address range established so far and switch
813 	 * to the proper swapper page table
814 	 */
815 	clone_pgd_range(swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
816 			initial_page_table + KERNEL_PGD_BOUNDARY,
817 			KERNEL_PGD_PTRS);
818 
819 	load_cr3(swapper_pg_dir);
820 	/*
821 	 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
822 	 * a cr3 based tlb flush, so the following __flush_tlb_all()
823 	 * will not flush anything because the CPU quirk which clears
824 	 * X86_FEATURE_PGE has not been invoked yet. Though due to the
825 	 * load_cr3() above the TLB has been flushed already. The
826 	 * quirk is invoked before subsequent calls to __flush_tlb_all()
827 	 * so proper operation is guaranteed.
828 	 */
829 	__flush_tlb_all();
830 #else
831 	printk(KERN_INFO "Command line: %s\n", boot_command_line);
832 	boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
833 #endif
834 
835 	/*
836 	 * If we have OLPC OFW, we might end up relocating the fixmap due to
837 	 * reserve_top(), so do this before touching the ioremap area.
838 	 */
839 	olpc_ofw_detect();
840 
841 	idt_setup_early_traps();
842 	early_cpu_init();
843 	arch_init_ideal_nops();
844 	jump_label_init();
845 	early_ioremap_init();
846 
847 	setup_olpc_ofw_pgd();
848 
849 	ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
850 	screen_info = boot_params.screen_info;
851 	edid_info = boot_params.edid_info;
852 #ifdef CONFIG_X86_32
853 	apm_info.bios = boot_params.apm_bios_info;
854 	ist_info = boot_params.ist_info;
855 #endif
856 	saved_video_mode = boot_params.hdr.vid_mode;
857 	bootloader_type = boot_params.hdr.type_of_loader;
858 	if ((bootloader_type >> 4) == 0xe) {
859 		bootloader_type &= 0xf;
860 		bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
861 	}
862 	bootloader_version  = bootloader_type & 0xf;
863 	bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
864 
865 #ifdef CONFIG_BLK_DEV_RAM
866 	rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
867 	rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
868 	rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
869 #endif
870 #ifdef CONFIG_EFI
871 	if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
872 		     EFI32_LOADER_SIGNATURE, 4)) {
873 		set_bit(EFI_BOOT, &efi.flags);
874 	} else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
875 		     EFI64_LOADER_SIGNATURE, 4)) {
876 		set_bit(EFI_BOOT, &efi.flags);
877 		set_bit(EFI_64BIT, &efi.flags);
878 	}
879 #endif
880 
881 	x86_init.oem.arch_setup();
882 
883 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
884 	e820__memory_setup();
885 	parse_setup_data();
886 
887 	copy_edd();
888 
889 	if (!boot_params.hdr.root_flags)
890 		root_mountflags &= ~MS_RDONLY;
891 	init_mm.start_code = (unsigned long) _text;
892 	init_mm.end_code = (unsigned long) _etext;
893 	init_mm.end_data = (unsigned long) _edata;
894 	init_mm.brk = _brk_end;
895 
896 	code_resource.start = __pa_symbol(_text);
897 	code_resource.end = __pa_symbol(_etext)-1;
898 	rodata_resource.start = __pa_symbol(__start_rodata);
899 	rodata_resource.end = __pa_symbol(__end_rodata)-1;
900 	data_resource.start = __pa_symbol(_sdata);
901 	data_resource.end = __pa_symbol(_edata)-1;
902 	bss_resource.start = __pa_symbol(__bss_start);
903 	bss_resource.end = __pa_symbol(__bss_stop)-1;
904 
905 #ifdef CONFIG_CMDLINE_BOOL
906 #ifdef CONFIG_CMDLINE_OVERRIDE
907 	strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
908 #else
909 	if (builtin_cmdline[0]) {
910 		/* append boot loader cmdline to builtin */
911 		strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
912 		strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
913 		strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
914 	}
915 #endif
916 #endif
917 
918 	strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
919 	*cmdline_p = command_line;
920 
921 	/*
922 	 * x86_configure_nx() is called before parse_early_param() to detect
923 	 * whether hardware doesn't support NX (so that the early EHCI debug
924 	 * console setup can safely call set_fixmap()). It may then be called
925 	 * again from within noexec_setup() during parsing early parameters
926 	 * to honor the respective command line option.
927 	 */
928 	x86_configure_nx();
929 
930 	parse_early_param();
931 
932 	if (efi_enabled(EFI_BOOT))
933 		efi_memblock_x86_reserve_range();
934 #ifdef CONFIG_MEMORY_HOTPLUG
935 	/*
936 	 * Memory used by the kernel cannot be hot-removed because Linux
937 	 * cannot migrate the kernel pages. When memory hotplug is
938 	 * enabled, we should prevent memblock from allocating memory
939 	 * for the kernel.
940 	 *
941 	 * ACPI SRAT records all hotpluggable memory ranges. But before
942 	 * SRAT is parsed, we don't know about it.
943 	 *
944 	 * The kernel image is loaded into memory at very early time. We
945 	 * cannot prevent this anyway. So on NUMA system, we set any
946 	 * node the kernel resides in as un-hotpluggable.
947 	 *
948 	 * Since on modern servers, one node could have double-digit
949 	 * gigabytes memory, we can assume the memory around the kernel
950 	 * image is also un-hotpluggable. So before SRAT is parsed, just
951 	 * allocate memory near the kernel image to try the best to keep
952 	 * the kernel away from hotpluggable memory.
953 	 */
954 	if (movable_node_is_enabled())
955 		memblock_set_bottom_up(true);
956 #endif
957 
958 	x86_report_nx();
959 
960 	/* after early param, so could get panic from serial */
961 	memblock_x86_reserve_range_setup_data();
962 
963 	if (acpi_mps_check()) {
964 #ifdef CONFIG_X86_LOCAL_APIC
965 		disable_apic = 1;
966 #endif
967 		setup_clear_cpu_cap(X86_FEATURE_APIC);
968 	}
969 
970 	e820__reserve_setup_data();
971 	e820__finish_early_params();
972 
973 	if (efi_enabled(EFI_BOOT))
974 		efi_init();
975 
976 	dmi_setup();
977 
978 	/*
979 	 * VMware detection requires dmi to be available, so this
980 	 * needs to be done after dmi_setup(), for the boot CPU.
981 	 */
982 	init_hypervisor_platform();
983 
984 	tsc_early_init();
985 	x86_init.resources.probe_roms();
986 
987 	/* after parse_early_param, so could debug it */
988 	insert_resource(&iomem_resource, &code_resource);
989 	insert_resource(&iomem_resource, &rodata_resource);
990 	insert_resource(&iomem_resource, &data_resource);
991 	insert_resource(&iomem_resource, &bss_resource);
992 
993 	e820_add_kernel_range();
994 	trim_bios_range();
995 #ifdef CONFIG_X86_32
996 	if (ppro_with_ram_bug()) {
997 		e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
998 				  E820_TYPE_RESERVED);
999 		e820__update_table(e820_table);
1000 		printk(KERN_INFO "fixed physical RAM map:\n");
1001 		e820__print_table("bad_ppro");
1002 	}
1003 #else
1004 	early_gart_iommu_check();
1005 #endif
1006 
1007 	/*
1008 	 * partially used pages are not usable - thus
1009 	 * we are rounding upwards:
1010 	 */
1011 	max_pfn = e820__end_of_ram_pfn();
1012 
1013 	/* update e820 for memory not covered by WB MTRRs */
1014 	mtrr_bp_init();
1015 	if (mtrr_trim_uncached_memory(max_pfn))
1016 		max_pfn = e820__end_of_ram_pfn();
1017 
1018 	max_possible_pfn = max_pfn;
1019 
1020 	/*
1021 	 * This call is required when the CPU does not support PAT. If
1022 	 * mtrr_bp_init() invoked it already via pat_init() the call has no
1023 	 * effect.
1024 	 */
1025 	init_cache_modes();
1026 
1027 	/*
1028 	 * Define random base addresses for memory sections after max_pfn is
1029 	 * defined and before each memory section base is used.
1030 	 */
1031 	kernel_randomize_memory();
1032 
1033 #ifdef CONFIG_X86_32
1034 	/* max_low_pfn get updated here */
1035 	find_low_pfn_range();
1036 #else
1037 	check_x2apic();
1038 
1039 	/* How many end-of-memory variables you have, grandma! */
1040 	/* need this before calling reserve_initrd */
1041 	if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1042 		max_low_pfn = e820__end_of_low_ram_pfn();
1043 	else
1044 		max_low_pfn = max_pfn;
1045 
1046 	high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1047 #endif
1048 
1049 	/*
1050 	 * Find and reserve possible boot-time SMP configuration:
1051 	 */
1052 	find_smp_config();
1053 
1054 	reserve_ibft_region();
1055 
1056 	early_alloc_pgt_buf();
1057 
1058 	/*
1059 	 * Need to conclude brk, before e820__memblock_setup()
1060 	 *  it could use memblock_find_in_range, could overlap with
1061 	 *  brk area.
1062 	 */
1063 	reserve_brk();
1064 
1065 	cleanup_highmap();
1066 
1067 	memblock_set_current_limit(ISA_END_ADDRESS);
1068 	e820__memblock_setup();
1069 
1070 	reserve_bios_regions();
1071 
1072 	efi_fake_memmap();
1073 	efi_find_mirror();
1074 	efi_esrt_init();
1075 
1076 	/*
1077 	 * The EFI specification says that boot service code won't be
1078 	 * called after ExitBootServices(). This is, in fact, a lie.
1079 	 */
1080 	efi_reserve_boot_services();
1081 
1082 	/* preallocate 4k for mptable mpc */
1083 	e820__memblock_alloc_reserved_mpc_new();
1084 
1085 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1086 	setup_bios_corruption_check();
1087 #endif
1088 
1089 #ifdef CONFIG_X86_32
1090 	printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1091 			(max_pfn_mapped<<PAGE_SHIFT) - 1);
1092 #endif
1093 
1094 	reserve_real_mode();
1095 
1096 	trim_platform_memory_ranges();
1097 	trim_low_memory_range();
1098 
1099 	init_mem_mapping();
1100 
1101 	idt_setup_early_pf();
1102 
1103 	/*
1104 	 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1105 	 * with the current CR4 value.  This may not be necessary, but
1106 	 * auditing all the early-boot CR4 manipulation would be needed to
1107 	 * rule it out.
1108 	 *
1109 	 * Mask off features that don't work outside long mode (just
1110 	 * PCIDE for now).
1111 	 */
1112 	mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1113 
1114 	memblock_set_current_limit(get_max_mapped());
1115 
1116 	/*
1117 	 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1118 	 */
1119 
1120 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1121 	if (init_ohci1394_dma_early)
1122 		init_ohci1394_dma_on_all_controllers();
1123 #endif
1124 	/* Allocate bigger log buffer */
1125 	setup_log_buf(1);
1126 
1127 	if (efi_enabled(EFI_BOOT)) {
1128 		switch (boot_params.secure_boot) {
1129 		case efi_secureboot_mode_disabled:
1130 			pr_info("Secure boot disabled\n");
1131 			break;
1132 		case efi_secureboot_mode_enabled:
1133 			pr_info("Secure boot enabled\n");
1134 			break;
1135 		default:
1136 			pr_info("Secure boot could not be determined\n");
1137 			break;
1138 		}
1139 	}
1140 
1141 	reserve_initrd();
1142 
1143 	acpi_table_upgrade();
1144 
1145 	vsmp_init();
1146 
1147 	io_delay_init();
1148 
1149 	early_platform_quirks();
1150 
1151 	/*
1152 	 * Parse the ACPI tables for possible boot-time SMP configuration.
1153 	 */
1154 	acpi_boot_table_init();
1155 
1156 	early_acpi_boot_init();
1157 
1158 	initmem_init();
1159 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1160 
1161 	/*
1162 	 * Reserve memory for crash kernel after SRAT is parsed so that it
1163 	 * won't consume hotpluggable memory.
1164 	 */
1165 	reserve_crashkernel();
1166 
1167 	memblock_find_dma_reserve();
1168 
1169 	if (!early_xdbc_setup_hardware())
1170 		early_xdbc_register_console();
1171 
1172 	x86_init.paging.pagetable_init();
1173 
1174 	kasan_init();
1175 
1176 	/*
1177 	 * Sync back kernel address range.
1178 	 *
1179 	 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1180 	 * this call?
1181 	 */
1182 	sync_initial_page_table();
1183 
1184 	tboot_probe();
1185 
1186 	map_vsyscall();
1187 
1188 	generic_apic_probe();
1189 
1190 	early_quirks();
1191 
1192 	/*
1193 	 * Read APIC and some other early information from ACPI tables.
1194 	 */
1195 	acpi_boot_init();
1196 	sfi_init();
1197 	x86_dtb_init();
1198 
1199 	/*
1200 	 * get boot-time SMP configuration:
1201 	 */
1202 	get_smp_config();
1203 
1204 	/*
1205 	 * Systems w/o ACPI and mptables might not have it mapped the local
1206 	 * APIC yet, but prefill_possible_map() might need to access it.
1207 	 */
1208 	init_apic_mappings();
1209 
1210 	prefill_possible_map();
1211 
1212 	init_cpu_to_node();
1213 
1214 	io_apic_init_mappings();
1215 
1216 	x86_init.hyper.guest_late_init();
1217 
1218 	e820__reserve_resources();
1219 	e820__register_nosave_regions(max_pfn);
1220 
1221 	x86_init.resources.reserve_resources();
1222 
1223 	e820__setup_pci_gap();
1224 
1225 #ifdef CONFIG_VT
1226 #if defined(CONFIG_VGA_CONSOLE)
1227 	if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1228 		conswitchp = &vga_con;
1229 #endif
1230 #endif
1231 	x86_init.oem.banner();
1232 
1233 	x86_init.timers.wallclock_init();
1234 
1235 	mcheck_init();
1236 
1237 	register_refined_jiffies(CLOCK_TICK_RATE);
1238 
1239 #ifdef CONFIG_EFI
1240 	if (efi_enabled(EFI_BOOT))
1241 		efi_apply_memmap_quirks();
1242 #endif
1243 
1244 	unwind_init();
1245 }
1246 
1247 #ifdef CONFIG_X86_32
1248 
1249 static struct resource video_ram_resource = {
1250 	.name	= "Video RAM area",
1251 	.start	= 0xa0000,
1252 	.end	= 0xbffff,
1253 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
1254 };
1255 
1256 void __init i386_reserve_resources(void)
1257 {
1258 	request_resource(&iomem_resource, &video_ram_resource);
1259 	reserve_standard_io_resources();
1260 }
1261 
1262 #endif /* CONFIG_X86_32 */
1263 
1264 static struct notifier_block kernel_offset_notifier = {
1265 	.notifier_call = dump_kernel_offset
1266 };
1267 
1268 static int __init register_kernel_offset_dumper(void)
1269 {
1270 	atomic_notifier_chain_register(&panic_notifier_list,
1271 					&kernel_offset_notifier);
1272 	return 0;
1273 }
1274 __initcall(register_kernel_offset_dumper);
1275