Loading...
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/coco.h>
39#include <asm/cpu.h>
40#include <asm/efi.h>
41#include <asm/gart.h>
42#include <asm/hypervisor.h>
43#include <asm/io_apic.h>
44#include <asm/kasan.h>
45#include <asm/kaslr.h>
46#include <asm/mce.h>
47#include <asm/memtype.h>
48#include <asm/mtrr.h>
49#include <asm/realmode.h>
50#include <asm/olpc_ofw.h>
51#include <asm/pci-direct.h>
52#include <asm/prom.h>
53#include <asm/proto.h>
54#include <asm/thermal.h>
55#include <asm/unwind.h>
56#include <asm/vsyscall.h>
57#include <linux/vmalloc.h>
58
59/*
60 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
61 * max_pfn_mapped: highest directly mapped pfn > 4 GB
62 *
63 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
64 * represented by pfn_mapped[].
65 */
66unsigned long max_low_pfn_mapped;
67unsigned long max_pfn_mapped;
68
69#ifdef CONFIG_DMI
70RESERVE_BRK(dmi_alloc, 65536);
71#endif
72
73
74unsigned long _brk_start = (unsigned long)__brk_base;
75unsigned long _brk_end = (unsigned long)__brk_base;
76
77struct boot_params boot_params;
78
79/*
80 * These are the four main kernel memory regions, we put them into
81 * the resource tree so that kdump tools and other debugging tools
82 * recover it:
83 */
84
85static struct resource rodata_resource = {
86 .name = "Kernel rodata",
87 .start = 0,
88 .end = 0,
89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
90};
91
92static struct resource data_resource = {
93 .name = "Kernel data",
94 .start = 0,
95 .end = 0,
96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
97};
98
99static struct resource code_resource = {
100 .name = "Kernel code",
101 .start = 0,
102 .end = 0,
103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
104};
105
106static struct resource bss_resource = {
107 .name = "Kernel bss",
108 .start = 0,
109 .end = 0,
110 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
111};
112
113
114#ifdef CONFIG_X86_32
115/* CPU data as detected by the assembly code in head_32.S */
116struct cpuinfo_x86 new_cpu_data;
117
118struct apm_info apm_info;
119EXPORT_SYMBOL(apm_info);
120
121#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
122 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
123struct ist_info ist_info;
124EXPORT_SYMBOL(ist_info);
125#else
126struct ist_info ist_info;
127#endif
128
129#endif
130
131struct cpuinfo_x86 boot_cpu_data __read_mostly;
132EXPORT_SYMBOL(boot_cpu_data);
133
134#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
135__visible unsigned long mmu_cr4_features __ro_after_init;
136#else
137__visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
138#endif
139
140#ifdef CONFIG_IMA
141static phys_addr_t ima_kexec_buffer_phys;
142static size_t ima_kexec_buffer_size;
143#endif
144
145/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
146int bootloader_type, bootloader_version;
147
148/*
149 * Setup options
150 */
151struct screen_info screen_info;
152EXPORT_SYMBOL(screen_info);
153struct edid_info edid_info;
154EXPORT_SYMBOL_GPL(edid_info);
155
156extern int root_mountflags;
157
158unsigned long saved_video_mode;
159
160#define RAMDISK_IMAGE_START_MASK 0x07FF
161#define RAMDISK_PROMPT_FLAG 0x8000
162#define RAMDISK_LOAD_FLAG 0x4000
163
164static char __initdata command_line[COMMAND_LINE_SIZE];
165#ifdef CONFIG_CMDLINE_BOOL
166static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
167#endif
168
169#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
170struct edd edd;
171#ifdef CONFIG_EDD_MODULE
172EXPORT_SYMBOL(edd);
173#endif
174/**
175 * copy_edd() - Copy the BIOS EDD information
176 * from boot_params into a safe place.
177 *
178 */
179static inline void __init copy_edd(void)
180{
181 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
182 sizeof(edd.mbr_signature));
183 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
184 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
185 edd.edd_info_nr = boot_params.eddbuf_entries;
186}
187#else
188static inline void __init copy_edd(void)
189{
190}
191#endif
192
193void * __init extend_brk(size_t size, size_t align)
194{
195 size_t mask = align - 1;
196 void *ret;
197
198 BUG_ON(_brk_start == 0);
199 BUG_ON(align & mask);
200
201 _brk_end = (_brk_end + mask) & ~mask;
202 BUG_ON((char *)(_brk_end + size) > __brk_limit);
203
204 ret = (void *)_brk_end;
205 _brk_end += size;
206
207 memset(ret, 0, size);
208
209 return ret;
210}
211
212#ifdef CONFIG_X86_32
213static void __init cleanup_highmap(void)
214{
215}
216#endif
217
218static void __init reserve_brk(void)
219{
220 if (_brk_end > _brk_start)
221 memblock_reserve(__pa_symbol(_brk_start),
222 _brk_end - _brk_start);
223
224 /* Mark brk area as locked down and no longer taking any
225 new allocations */
226 _brk_start = 0;
227}
228
229#ifdef CONFIG_BLK_DEV_INITRD
230
231static u64 __init get_ramdisk_image(void)
232{
233 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
234
235 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
236
237 if (ramdisk_image == 0)
238 ramdisk_image = phys_initrd_start;
239
240 return ramdisk_image;
241}
242static 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 if (ramdisk_size == 0)
249 ramdisk_size = phys_initrd_size;
250
251 return ramdisk_size;
252}
253
254static void __init relocate_initrd(void)
255{
256 /* Assume only end is not page aligned */
257 u64 ramdisk_image = get_ramdisk_image();
258 u64 ramdisk_size = get_ramdisk_size();
259 u64 area_size = PAGE_ALIGN(ramdisk_size);
260
261 /* We need to move the initrd down into directly mapped mem */
262 u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
263 PFN_PHYS(max_pfn_mapped));
264 if (!relocated_ramdisk)
265 panic("Cannot find place for new RAMDISK of size %lld\n",
266 ramdisk_size);
267
268 initrd_start = relocated_ramdisk + PAGE_OFFSET;
269 initrd_end = initrd_start + ramdisk_size;
270 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
271 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
272
273 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
274
275 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
276 " [mem %#010llx-%#010llx]\n",
277 ramdisk_image, ramdisk_image + ramdisk_size - 1,
278 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
279}
280
281static void __init early_reserve_initrd(void)
282{
283 /* Assume only end is not page aligned */
284 u64 ramdisk_image = get_ramdisk_image();
285 u64 ramdisk_size = get_ramdisk_size();
286 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
287
288 if (!boot_params.hdr.type_of_loader ||
289 !ramdisk_image || !ramdisk_size)
290 return; /* No initrd provided by bootloader */
291
292 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
293}
294
295static 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
302 if (!boot_params.hdr.type_of_loader ||
303 !ramdisk_image || !ramdisk_size)
304 return; /* No initrd provided by bootloader */
305
306 initrd_start = 0;
307
308 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
309 ramdisk_end - 1);
310
311 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
312 PFN_DOWN(ramdisk_end))) {
313 /* All are mapped, easy case */
314 initrd_start = ramdisk_image + PAGE_OFFSET;
315 initrd_end = initrd_start + ramdisk_size;
316 return;
317 }
318
319 relocate_initrd();
320
321 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
322}
323
324#else
325static void __init early_reserve_initrd(void)
326{
327}
328static void __init reserve_initrd(void)
329{
330}
331#endif /* CONFIG_BLK_DEV_INITRD */
332
333static void __init add_early_ima_buffer(u64 phys_addr)
334{
335#ifdef CONFIG_IMA
336 struct ima_setup_data *data;
337
338 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
339 if (!data) {
340 pr_warn("setup: failed to memremap ima_setup_data entry\n");
341 return;
342 }
343
344 if (data->size) {
345 memblock_reserve(data->addr, data->size);
346 ima_kexec_buffer_phys = data->addr;
347 ima_kexec_buffer_size = data->size;
348 }
349
350 early_memunmap(data, sizeof(*data));
351#else
352 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
353#endif
354}
355
356#if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
357int __init ima_free_kexec_buffer(void)
358{
359 if (!ima_kexec_buffer_size)
360 return -ENOENT;
361
362 memblock_free_late(ima_kexec_buffer_phys,
363 ima_kexec_buffer_size);
364
365 ima_kexec_buffer_phys = 0;
366 ima_kexec_buffer_size = 0;
367
368 return 0;
369}
370
371int __init ima_get_kexec_buffer(void **addr, size_t *size)
372{
373 if (!ima_kexec_buffer_size)
374 return -ENOENT;
375
376 *addr = __va(ima_kexec_buffer_phys);
377 *size = ima_kexec_buffer_size;
378
379 return 0;
380}
381#endif
382
383static void __init parse_setup_data(void)
384{
385 struct setup_data *data;
386 u64 pa_data, pa_next;
387
388 pa_data = boot_params.hdr.setup_data;
389 while (pa_data) {
390 u32 data_len, data_type;
391
392 data = early_memremap(pa_data, sizeof(*data));
393 data_len = data->len + sizeof(struct setup_data);
394 data_type = data->type;
395 pa_next = data->next;
396 early_memunmap(data, sizeof(*data));
397
398 switch (data_type) {
399 case SETUP_E820_EXT:
400 e820__memory_setup_extended(pa_data, data_len);
401 break;
402 case SETUP_DTB:
403 add_dtb(pa_data);
404 break;
405 case SETUP_EFI:
406 parse_efi_setup(pa_data, data_len);
407 break;
408 case SETUP_IMA:
409 add_early_ima_buffer(pa_data);
410 break;
411 case SETUP_RNG_SEED:
412 data = early_memremap(pa_data, data_len);
413 add_bootloader_randomness(data->data, data->len);
414 /* Zero seed for forward secrecy. */
415 memzero_explicit(data->data, data->len);
416 /* Zero length in case we find ourselves back here by accident. */
417 memzero_explicit(&data->len, sizeof(data->len));
418 early_memunmap(data, data_len);
419 break;
420 default:
421 break;
422 }
423 pa_data = pa_next;
424 }
425}
426
427static void __init memblock_x86_reserve_range_setup_data(void)
428{
429 struct setup_indirect *indirect;
430 struct setup_data *data;
431 u64 pa_data, pa_next;
432 u32 len;
433
434 pa_data = boot_params.hdr.setup_data;
435 while (pa_data) {
436 data = early_memremap(pa_data, sizeof(*data));
437 if (!data) {
438 pr_warn("setup: failed to memremap setup_data entry\n");
439 return;
440 }
441
442 len = sizeof(*data);
443 pa_next = data->next;
444
445 memblock_reserve(pa_data, sizeof(*data) + data->len);
446
447 if (data->type == SETUP_INDIRECT) {
448 len += data->len;
449 early_memunmap(data, sizeof(*data));
450 data = early_memremap(pa_data, len);
451 if (!data) {
452 pr_warn("setup: failed to memremap indirect setup_data\n");
453 return;
454 }
455
456 indirect = (struct setup_indirect *)data->data;
457
458 if (indirect->type != SETUP_INDIRECT)
459 memblock_reserve(indirect->addr, indirect->len);
460 }
461
462 pa_data = pa_next;
463 early_memunmap(data, len);
464 }
465}
466
467static void __init arch_reserve_crashkernel(void)
468{
469 unsigned long long crash_base, crash_size, low_size = 0;
470 char *cmdline = boot_command_line;
471 bool high = false;
472 int ret;
473
474 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
475 return;
476
477 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
478 &crash_size, &crash_base,
479 &low_size, &high);
480 if (ret)
481 return;
482
483 if (xen_pv_domain()) {
484 pr_info("Ignoring crashkernel for a Xen PV domain\n");
485 return;
486 }
487
488 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
489 low_size, high);
490}
491
492static struct resource standard_io_resources[] = {
493 { .name = "dma1", .start = 0x00, .end = 0x1f,
494 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
495 { .name = "pic1", .start = 0x20, .end = 0x21,
496 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
497 { .name = "timer0", .start = 0x40, .end = 0x43,
498 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
499 { .name = "timer1", .start = 0x50, .end = 0x53,
500 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
501 { .name = "keyboard", .start = 0x60, .end = 0x60,
502 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
503 { .name = "keyboard", .start = 0x64, .end = 0x64,
504 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
505 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
506 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
507 { .name = "pic2", .start = 0xa0, .end = 0xa1,
508 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
509 { .name = "dma2", .start = 0xc0, .end = 0xdf,
510 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
511 { .name = "fpu", .start = 0xf0, .end = 0xff,
512 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
513};
514
515void __init reserve_standard_io_resources(void)
516{
517 int i;
518
519 /* request I/O space for devices used on all i[345]86 PCs */
520 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
521 request_resource(&ioport_resource, &standard_io_resources[i]);
522
523}
524
525static bool __init snb_gfx_workaround_needed(void)
526{
527#ifdef CONFIG_PCI
528 int i;
529 u16 vendor, devid;
530 static const __initconst u16 snb_ids[] = {
531 0x0102,
532 0x0112,
533 0x0122,
534 0x0106,
535 0x0116,
536 0x0126,
537 0x010a,
538 };
539
540 /* Assume no if something weird is going on with PCI */
541 if (!early_pci_allowed())
542 return false;
543
544 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
545 if (vendor != 0x8086)
546 return false;
547
548 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
549 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
550 if (devid == snb_ids[i])
551 return true;
552#endif
553
554 return false;
555}
556
557/*
558 * Sandy Bridge graphics has trouble with certain ranges, exclude
559 * them from allocation.
560 */
561static void __init trim_snb_memory(void)
562{
563 static const __initconst unsigned long bad_pages[] = {
564 0x20050000,
565 0x20110000,
566 0x20130000,
567 0x20138000,
568 0x40004000,
569 };
570 int i;
571
572 if (!snb_gfx_workaround_needed())
573 return;
574
575 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
576
577 /*
578 * SandyBridge integrated graphics devices have a bug that prevents
579 * them from accessing certain memory ranges, namely anything below
580 * 1M and in the pages listed in bad_pages[] above.
581 *
582 * To avoid these pages being ever accessed by SNB gfx devices reserve
583 * bad_pages that have not already been reserved at boot time.
584 * All memory below the 1 MB mark is anyway reserved later during
585 * setup_arch(), so there is no need to reserve it here.
586 */
587
588 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
589 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
590 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
591 bad_pages[i]);
592 }
593}
594
595static void __init trim_bios_range(void)
596{
597 /*
598 * A special case is the first 4Kb of memory;
599 * This is a BIOS owned area, not kernel ram, but generally
600 * not listed as such in the E820 table.
601 *
602 * This typically reserves additional memory (64KiB by default)
603 * since some BIOSes are known to corrupt low memory. See the
604 * Kconfig help text for X86_RESERVE_LOW.
605 */
606 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
607
608 /*
609 * special case: Some BIOSes report the PC BIOS
610 * area (640Kb -> 1Mb) as RAM even though it is not.
611 * take them out.
612 */
613 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
614
615 e820__update_table(e820_table);
616}
617
618/* called before trim_bios_range() to spare extra sanitize */
619static void __init e820_add_kernel_range(void)
620{
621 u64 start = __pa_symbol(_text);
622 u64 size = __pa_symbol(_end) - start;
623
624 /*
625 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
626 * attempt to fix it by adding the range. We may have a confused BIOS,
627 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
628 * exclude kernel range. If we really are running on top non-RAM,
629 * we will crash later anyways.
630 */
631 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
632 return;
633
634 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
635 e820__range_remove(start, size, E820_TYPE_RAM, 0);
636 e820__range_add(start, size, E820_TYPE_RAM);
637}
638
639static void __init early_reserve_memory(void)
640{
641 /*
642 * Reserve the memory occupied by the kernel between _text and
643 * __end_of_kernel_reserve symbols. Any kernel sections after the
644 * __end_of_kernel_reserve symbol must be explicitly reserved with a
645 * separate memblock_reserve() or they will be discarded.
646 */
647 memblock_reserve(__pa_symbol(_text),
648 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
649
650 /*
651 * The first 4Kb of memory is a BIOS owned area, but generally it is
652 * not listed as such in the E820 table.
653 *
654 * Reserve the first 64K of memory since some BIOSes are known to
655 * corrupt low memory. After the real mode trampoline is allocated the
656 * rest of the memory below 640k is reserved.
657 *
658 * In addition, make sure page 0 is always reserved because on
659 * systems with L1TF its contents can be leaked to user processes.
660 */
661 memblock_reserve(0, SZ_64K);
662
663 early_reserve_initrd();
664
665 memblock_x86_reserve_range_setup_data();
666
667 reserve_bios_regions();
668 trim_snb_memory();
669}
670
671/*
672 * Dump out kernel offset information on panic.
673 */
674static int
675dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
676{
677 if (kaslr_enabled()) {
678 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
679 kaslr_offset(),
680 __START_KERNEL,
681 __START_KERNEL_map,
682 MODULES_VADDR-1);
683 } else {
684 pr_emerg("Kernel Offset: disabled\n");
685 }
686
687 return 0;
688}
689
690void x86_configure_nx(void)
691{
692 if (boot_cpu_has(X86_FEATURE_NX))
693 __supported_pte_mask |= _PAGE_NX;
694 else
695 __supported_pte_mask &= ~_PAGE_NX;
696}
697
698static void __init x86_report_nx(void)
699{
700 if (!boot_cpu_has(X86_FEATURE_NX)) {
701 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
702 "missing in CPU!\n");
703 } else {
704#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
705 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
706#else
707 /* 32bit non-PAE kernel, NX cannot be used */
708 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
709 "cannot be enabled: non-PAE kernel!\n");
710#endif
711 }
712}
713
714/*
715 * Determine if we were loaded by an EFI loader. If so, then we have also been
716 * passed the efi memmap, systab, etc., so we should use these data structures
717 * for initialization. Note, the efi init code path is determined by the
718 * global efi_enabled. This allows the same kernel image to be used on existing
719 * systems (with a traditional BIOS) as well as on EFI systems.
720 */
721/*
722 * setup_arch - architecture-specific boot-time initializations
723 *
724 * Note: On x86_64, fixmaps are ready for use even before this is called.
725 */
726
727void __init setup_arch(char **cmdline_p)
728{
729#ifdef CONFIG_X86_32
730 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
731
732 /*
733 * copy kernel address range established so far and switch
734 * to the proper swapper page table
735 */
736 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
737 initial_page_table + KERNEL_PGD_BOUNDARY,
738 KERNEL_PGD_PTRS);
739
740 load_cr3(swapper_pg_dir);
741 /*
742 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
743 * a cr3 based tlb flush, so the following __flush_tlb_all()
744 * will not flush anything because the CPU quirk which clears
745 * X86_FEATURE_PGE has not been invoked yet. Though due to the
746 * load_cr3() above the TLB has been flushed already. The
747 * quirk is invoked before subsequent calls to __flush_tlb_all()
748 * so proper operation is guaranteed.
749 */
750 __flush_tlb_all();
751#else
752 printk(KERN_INFO "Command line: %s\n", boot_command_line);
753 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
754#endif
755
756 /*
757 * If we have OLPC OFW, we might end up relocating the fixmap due to
758 * reserve_top(), so do this before touching the ioremap area.
759 */
760 olpc_ofw_detect();
761
762 idt_setup_early_traps();
763 early_cpu_init();
764 jump_label_init();
765 static_call_init();
766 early_ioremap_init();
767
768 setup_olpc_ofw_pgd();
769
770 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
771 screen_info = boot_params.screen_info;
772 edid_info = boot_params.edid_info;
773#ifdef CONFIG_X86_32
774 apm_info.bios = boot_params.apm_bios_info;
775 ist_info = boot_params.ist_info;
776#endif
777 saved_video_mode = boot_params.hdr.vid_mode;
778 bootloader_type = boot_params.hdr.type_of_loader;
779 if ((bootloader_type >> 4) == 0xe) {
780 bootloader_type &= 0xf;
781 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
782 }
783 bootloader_version = bootloader_type & 0xf;
784 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
785
786#ifdef CONFIG_BLK_DEV_RAM
787 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
788#endif
789#ifdef CONFIG_EFI
790 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
791 EFI32_LOADER_SIGNATURE, 4)) {
792 set_bit(EFI_BOOT, &efi.flags);
793 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
794 EFI64_LOADER_SIGNATURE, 4)) {
795 set_bit(EFI_BOOT, &efi.flags);
796 set_bit(EFI_64BIT, &efi.flags);
797 }
798#endif
799
800 x86_init.oem.arch_setup();
801
802 /*
803 * Do some memory reservations *before* memory is added to memblock, so
804 * memblock allocations won't overwrite it.
805 *
806 * After this point, everything still needed from the boot loader or
807 * firmware or kernel text should be early reserved or marked not RAM in
808 * e820. All other memory is free game.
809 *
810 * This call needs to happen before e820__memory_setup() which calls the
811 * xen_memory_setup() on Xen dom0 which relies on the fact that those
812 * early reservations have happened already.
813 */
814 early_reserve_memory();
815
816 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
817 e820__memory_setup();
818 parse_setup_data();
819
820 copy_edd();
821
822 if (!boot_params.hdr.root_flags)
823 root_mountflags &= ~MS_RDONLY;
824 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
825
826 code_resource.start = __pa_symbol(_text);
827 code_resource.end = __pa_symbol(_etext)-1;
828 rodata_resource.start = __pa_symbol(__start_rodata);
829 rodata_resource.end = __pa_symbol(__end_rodata)-1;
830 data_resource.start = __pa_symbol(_sdata);
831 data_resource.end = __pa_symbol(_edata)-1;
832 bss_resource.start = __pa_symbol(__bss_start);
833 bss_resource.end = __pa_symbol(__bss_stop)-1;
834
835#ifdef CONFIG_CMDLINE_BOOL
836#ifdef CONFIG_CMDLINE_OVERRIDE
837 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
838#else
839 if (builtin_cmdline[0]) {
840 /* append boot loader cmdline to builtin */
841 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
842 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
843 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
844 }
845#endif
846#endif
847
848 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
849 *cmdline_p = command_line;
850
851 /*
852 * x86_configure_nx() is called before parse_early_param() to detect
853 * whether hardware doesn't support NX (so that the early EHCI debug
854 * console setup can safely call set_fixmap()).
855 */
856 x86_configure_nx();
857
858 parse_early_param();
859
860 if (efi_enabled(EFI_BOOT))
861 efi_memblock_x86_reserve_range();
862
863#ifdef CONFIG_MEMORY_HOTPLUG
864 /*
865 * Memory used by the kernel cannot be hot-removed because Linux
866 * cannot migrate the kernel pages. When memory hotplug is
867 * enabled, we should prevent memblock from allocating memory
868 * for the kernel.
869 *
870 * ACPI SRAT records all hotpluggable memory ranges. But before
871 * SRAT is parsed, we don't know about it.
872 *
873 * The kernel image is loaded into memory at very early time. We
874 * cannot prevent this anyway. So on NUMA system, we set any
875 * node the kernel resides in as un-hotpluggable.
876 *
877 * Since on modern servers, one node could have double-digit
878 * gigabytes memory, we can assume the memory around the kernel
879 * image is also un-hotpluggable. So before SRAT is parsed, just
880 * allocate memory near the kernel image to try the best to keep
881 * the kernel away from hotpluggable memory.
882 */
883 if (movable_node_is_enabled())
884 memblock_set_bottom_up(true);
885#endif
886
887 x86_report_nx();
888
889 apic_setup_apic_calls();
890
891 if (acpi_mps_check()) {
892#ifdef CONFIG_X86_LOCAL_APIC
893 apic_is_disabled = true;
894#endif
895 setup_clear_cpu_cap(X86_FEATURE_APIC);
896 }
897
898 e820__reserve_setup_data();
899 e820__finish_early_params();
900
901 if (efi_enabled(EFI_BOOT))
902 efi_init();
903
904 reserve_ibft_region();
905 x86_init.resources.dmi_setup();
906
907 /*
908 * VMware detection requires dmi to be available, so this
909 * needs to be done after dmi_setup(), for the boot CPU.
910 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
911 * called before cache_bp_init() for setting up MTRR state.
912 */
913 init_hypervisor_platform();
914
915 tsc_early_init();
916 x86_init.resources.probe_roms();
917
918 /* after parse_early_param, so could debug it */
919 insert_resource(&iomem_resource, &code_resource);
920 insert_resource(&iomem_resource, &rodata_resource);
921 insert_resource(&iomem_resource, &data_resource);
922 insert_resource(&iomem_resource, &bss_resource);
923
924 e820_add_kernel_range();
925 trim_bios_range();
926#ifdef CONFIG_X86_32
927 if (ppro_with_ram_bug()) {
928 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
929 E820_TYPE_RESERVED);
930 e820__update_table(e820_table);
931 printk(KERN_INFO "fixed physical RAM map:\n");
932 e820__print_table("bad_ppro");
933 }
934#else
935 early_gart_iommu_check();
936#endif
937
938 /*
939 * partially used pages are not usable - thus
940 * we are rounding upwards:
941 */
942 max_pfn = e820__end_of_ram_pfn();
943
944 /* update e820 for memory not covered by WB MTRRs */
945 cache_bp_init();
946 if (mtrr_trim_uncached_memory(max_pfn))
947 max_pfn = e820__end_of_ram_pfn();
948
949 max_possible_pfn = max_pfn;
950
951 /*
952 * Define random base addresses for memory sections after max_pfn is
953 * defined and before each memory section base is used.
954 */
955 kernel_randomize_memory();
956
957#ifdef CONFIG_X86_32
958 /* max_low_pfn get updated here */
959 find_low_pfn_range();
960#else
961 check_x2apic();
962
963 /* How many end-of-memory variables you have, grandma! */
964 /* need this before calling reserve_initrd */
965 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
966 max_low_pfn = e820__end_of_low_ram_pfn();
967 else
968 max_low_pfn = max_pfn;
969
970 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
971#endif
972
973 /* Find and reserve MPTABLE area */
974 x86_init.mpparse.find_mptable();
975
976 early_alloc_pgt_buf();
977
978 /*
979 * Need to conclude brk, before e820__memblock_setup()
980 * it could use memblock_find_in_range, could overlap with
981 * brk area.
982 */
983 reserve_brk();
984
985 cleanup_highmap();
986
987 memblock_set_current_limit(ISA_END_ADDRESS);
988 e820__memblock_setup();
989
990 /*
991 * Needs to run after memblock setup because it needs the physical
992 * memory size.
993 */
994 mem_encrypt_setup_arch();
995 cc_random_init();
996
997 efi_fake_memmap();
998 efi_find_mirror();
999 efi_esrt_init();
1000 efi_mokvar_table_init();
1001
1002 /*
1003 * The EFI specification says that boot service code won't be
1004 * called after ExitBootServices(). This is, in fact, a lie.
1005 */
1006 efi_reserve_boot_services();
1007
1008 /* preallocate 4k for mptable mpc */
1009 e820__memblock_alloc_reserved_mpc_new();
1010
1011#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1012 setup_bios_corruption_check();
1013#endif
1014
1015#ifdef CONFIG_X86_32
1016 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1017 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1018#endif
1019
1020 /*
1021 * Find free memory for the real mode trampoline and place it there. If
1022 * there is not enough free memory under 1M, on EFI-enabled systems
1023 * there will be additional attempt to reclaim the memory for the real
1024 * mode trampoline at efi_free_boot_services().
1025 *
1026 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1027 * are known to corrupt low memory and several hundred kilobytes are not
1028 * worth complex detection what memory gets clobbered. Windows does the
1029 * same thing for very similar reasons.
1030 *
1031 * Moreover, on machines with SandyBridge graphics or in setups that use
1032 * crashkernel the entire 1M is reserved anyway.
1033 *
1034 * Note the host kernel TDX also requires the first 1MB being reserved.
1035 */
1036 x86_platform.realmode_reserve();
1037
1038 init_mem_mapping();
1039
1040 idt_setup_early_pf();
1041
1042 /*
1043 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1044 * with the current CR4 value. This may not be necessary, but
1045 * auditing all the early-boot CR4 manipulation would be needed to
1046 * rule it out.
1047 *
1048 * Mask off features that don't work outside long mode (just
1049 * PCIDE for now).
1050 */
1051 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1052
1053 memblock_set_current_limit(get_max_mapped());
1054
1055 /*
1056 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1057 */
1058
1059#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1060 if (init_ohci1394_dma_early)
1061 init_ohci1394_dma_on_all_controllers();
1062#endif
1063 /* Allocate bigger log buffer */
1064 setup_log_buf(1);
1065
1066 if (efi_enabled(EFI_BOOT)) {
1067 switch (boot_params.secure_boot) {
1068 case efi_secureboot_mode_disabled:
1069 pr_info("Secure boot disabled\n");
1070 break;
1071 case efi_secureboot_mode_enabled:
1072 pr_info("Secure boot enabled\n");
1073 break;
1074 default:
1075 pr_info("Secure boot could not be determined\n");
1076 break;
1077 }
1078 }
1079
1080 reserve_initrd();
1081
1082 acpi_table_upgrade();
1083 /* Look for ACPI tables and reserve memory occupied by them. */
1084 acpi_boot_table_init();
1085
1086 vsmp_init();
1087
1088 io_delay_init();
1089
1090 early_platform_quirks();
1091
1092 /* Some platforms need the APIC registered for NUMA configuration */
1093 early_acpi_boot_init();
1094 x86_init.mpparse.early_parse_smp_cfg();
1095
1096 x86_flattree_get_config();
1097
1098 initmem_init();
1099 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1100
1101 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1102 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1103
1104 /*
1105 * Reserve memory for crash kernel after SRAT is parsed so that it
1106 * won't consume hotpluggable memory.
1107 */
1108 arch_reserve_crashkernel();
1109
1110 memblock_find_dma_reserve();
1111
1112 if (!early_xdbc_setup_hardware())
1113 early_xdbc_register_console();
1114
1115 x86_init.paging.pagetable_init();
1116
1117 kasan_init();
1118
1119 /*
1120 * Sync back kernel address range.
1121 *
1122 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1123 * this call?
1124 */
1125 sync_initial_page_table();
1126
1127 tboot_probe();
1128
1129 map_vsyscall();
1130
1131 x86_32_probe_apic();
1132
1133 early_quirks();
1134
1135 topology_apply_cmdline_limits_early();
1136
1137 /*
1138 * Parse SMP configuration. Try ACPI first and then the platform
1139 * specific parser.
1140 */
1141 acpi_boot_init();
1142 x86_init.mpparse.parse_smp_cfg();
1143
1144 /* Last opportunity to detect and map the local APIC */
1145 init_apic_mappings();
1146
1147 topology_init_possible_cpus();
1148
1149 init_cpu_to_node();
1150 init_gi_nodes();
1151
1152 io_apic_init_mappings();
1153
1154 x86_init.hyper.guest_late_init();
1155
1156 e820__reserve_resources();
1157 e820__register_nosave_regions(max_pfn);
1158
1159 x86_init.resources.reserve_resources();
1160
1161 e820__setup_pci_gap();
1162
1163#ifdef CONFIG_VT
1164#if defined(CONFIG_VGA_CONSOLE)
1165 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1166 vgacon_register_screen(&screen_info);
1167#endif
1168#endif
1169 x86_init.oem.banner();
1170
1171 x86_init.timers.wallclock_init();
1172
1173 /*
1174 * This needs to run before setup_local_APIC() which soft-disables the
1175 * local APIC temporarily and that masks the thermal LVT interrupt,
1176 * leading to softlockups on machines which have configured SMI
1177 * interrupt delivery.
1178 */
1179 therm_lvt_init();
1180
1181 mcheck_init();
1182
1183 register_refined_jiffies(CLOCK_TICK_RATE);
1184
1185#ifdef CONFIG_EFI
1186 if (efi_enabled(EFI_BOOT))
1187 efi_apply_memmap_quirks();
1188#endif
1189
1190 unwind_init();
1191}
1192
1193#ifdef CONFIG_X86_32
1194
1195static struct resource video_ram_resource = {
1196 .name = "Video RAM area",
1197 .start = 0xa0000,
1198 .end = 0xbffff,
1199 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1200};
1201
1202void __init i386_reserve_resources(void)
1203{
1204 request_resource(&iomem_resource, &video_ram_resource);
1205 reserve_standard_io_resources();
1206}
1207
1208#endif /* CONFIG_X86_32 */
1209
1210static struct notifier_block kernel_offset_notifier = {
1211 .notifier_call = dump_kernel_offset
1212};
1213
1214static int __init register_kernel_offset_dumper(void)
1215{
1216 atomic_notifier_chain_register(&panic_notifier_list,
1217 &kernel_offset_notifier);
1218 return 0;
1219}
1220__initcall(register_kernel_offset_dumper);
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 *
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
8 *
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
11 *
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
14 *
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
17 *
18 */
19
20/*
21 * This file handles the architecture-dependent parts of initialization
22 */
23
24#include <linux/sched.h>
25#include <linux/mm.h>
26#include <linux/mmzone.h>
27#include <linux/screen_info.h>
28#include <linux/ioport.h>
29#include <linux/acpi.h>
30#include <linux/sfi.h>
31#include <linux/apm_bios.h>
32#include <linux/initrd.h>
33#include <linux/bootmem.h>
34#include <linux/memblock.h>
35#include <linux/seq_file.h>
36#include <linux/console.h>
37#include <linux/root_dev.h>
38#include <linux/highmem.h>
39#include <linux/export.h>
40#include <linux/efi.h>
41#include <linux/init.h>
42#include <linux/edd.h>
43#include <linux/iscsi_ibft.h>
44#include <linux/nodemask.h>
45#include <linux/kexec.h>
46#include <linux/dmi.h>
47#include <linux/pfn.h>
48#include <linux/pci.h>
49#include <asm/pci-direct.h>
50#include <linux/init_ohci1394_dma.h>
51#include <linux/kvm_para.h>
52#include <linux/dma-contiguous.h>
53
54#include <linux/errno.h>
55#include <linux/kernel.h>
56#include <linux/stddef.h>
57#include <linux/unistd.h>
58#include <linux/ptrace.h>
59#include <linux/user.h>
60#include <linux/delay.h>
61
62#include <linux/kallsyms.h>
63#include <linux/cpufreq.h>
64#include <linux/dma-mapping.h>
65#include <linux/ctype.h>
66#include <linux/uaccess.h>
67
68#include <linux/percpu.h>
69#include <linux/crash_dump.h>
70#include <linux/tboot.h>
71#include <linux/jiffies.h>
72
73#include <video/edid.h>
74
75#include <asm/mtrr.h>
76#include <asm/apic.h>
77#include <asm/realmode.h>
78#include <asm/e820.h>
79#include <asm/mpspec.h>
80#include <asm/setup.h>
81#include <asm/efi.h>
82#include <asm/timer.h>
83#include <asm/i8259.h>
84#include <asm/sections.h>
85#include <asm/io_apic.h>
86#include <asm/ist.h>
87#include <asm/setup_arch.h>
88#include <asm/bios_ebda.h>
89#include <asm/cacheflush.h>
90#include <asm/processor.h>
91#include <asm/bugs.h>
92#include <asm/kasan.h>
93
94#include <asm/vsyscall.h>
95#include <asm/cpu.h>
96#include <asm/desc.h>
97#include <asm/dma.h>
98#include <asm/iommu.h>
99#include <asm/gart.h>
100#include <asm/mmu_context.h>
101#include <asm/proto.h>
102
103#include <asm/paravirt.h>
104#include <asm/hypervisor.h>
105#include <asm/olpc_ofw.h>
106
107#include <asm/percpu.h>
108#include <asm/topology.h>
109#include <asm/apicdef.h>
110#include <asm/amd_nb.h>
111#include <asm/mce.h>
112#include <asm/alternative.h>
113#include <asm/prom.h>
114#include <asm/microcode.h>
115#include <asm/mmu_context.h>
116#include <asm/kaslr.h>
117
118/*
119 * max_low_pfn_mapped: highest direct mapped pfn under 4GB
120 * max_pfn_mapped: highest direct mapped pfn over 4GB
121 *
122 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
123 * represented by pfn_mapped
124 */
125unsigned long max_low_pfn_mapped;
126unsigned long max_pfn_mapped;
127
128#ifdef CONFIG_DMI
129RESERVE_BRK(dmi_alloc, 65536);
130#endif
131
132
133static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
134unsigned long _brk_end = (unsigned long)__brk_base;
135
136#ifdef CONFIG_X86_64
137int default_cpu_present_to_apicid(int mps_cpu)
138{
139 return __default_cpu_present_to_apicid(mps_cpu);
140}
141
142int default_check_phys_apicid_present(int phys_apicid)
143{
144 return __default_check_phys_apicid_present(phys_apicid);
145}
146#endif
147
148struct boot_params boot_params;
149
150/*
151 * Machine setup..
152 */
153static struct resource data_resource = {
154 .name = "Kernel data",
155 .start = 0,
156 .end = 0,
157 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
158};
159
160static struct resource code_resource = {
161 .name = "Kernel code",
162 .start = 0,
163 .end = 0,
164 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
165};
166
167static struct resource bss_resource = {
168 .name = "Kernel bss",
169 .start = 0,
170 .end = 0,
171 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
172};
173
174
175#ifdef CONFIG_X86_32
176/* cpu data as detected by the assembly code in head.S */
177struct cpuinfo_x86 new_cpu_data = {
178 .wp_works_ok = -1,
179};
180/* common cpu data for all cpus */
181struct cpuinfo_x86 boot_cpu_data __read_mostly = {
182 .wp_works_ok = -1,
183};
184EXPORT_SYMBOL(boot_cpu_data);
185
186unsigned int def_to_bigsmp;
187
188/* for MCA, but anyone else can use it if they want */
189unsigned int machine_id;
190unsigned int machine_submodel_id;
191unsigned int BIOS_revision;
192
193struct apm_info apm_info;
194EXPORT_SYMBOL(apm_info);
195
196#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
197 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
198struct ist_info ist_info;
199EXPORT_SYMBOL(ist_info);
200#else
201struct ist_info ist_info;
202#endif
203
204#else
205struct cpuinfo_x86 boot_cpu_data __read_mostly = {
206 .x86_phys_bits = MAX_PHYSMEM_BITS,
207};
208EXPORT_SYMBOL(boot_cpu_data);
209#endif
210
211
212#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
213__visible unsigned long mmu_cr4_features __ro_after_init;
214#else
215__visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
216#endif
217
218/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
219int bootloader_type, bootloader_version;
220
221/*
222 * Setup options
223 */
224struct screen_info screen_info;
225EXPORT_SYMBOL(screen_info);
226struct edid_info edid_info;
227EXPORT_SYMBOL_GPL(edid_info);
228
229extern int root_mountflags;
230
231unsigned long saved_video_mode;
232
233#define RAMDISK_IMAGE_START_MASK 0x07FF
234#define RAMDISK_PROMPT_FLAG 0x8000
235#define RAMDISK_LOAD_FLAG 0x4000
236
237static char __initdata command_line[COMMAND_LINE_SIZE];
238#ifdef CONFIG_CMDLINE_BOOL
239static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
240#endif
241
242#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
243struct edd edd;
244#ifdef CONFIG_EDD_MODULE
245EXPORT_SYMBOL(edd);
246#endif
247/**
248 * copy_edd() - Copy the BIOS EDD information
249 * from boot_params into a safe place.
250 *
251 */
252static inline void __init copy_edd(void)
253{
254 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
255 sizeof(edd.mbr_signature));
256 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
257 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
258 edd.edd_info_nr = boot_params.eddbuf_entries;
259}
260#else
261static inline void __init copy_edd(void)
262{
263}
264#endif
265
266void * __init extend_brk(size_t size, size_t align)
267{
268 size_t mask = align - 1;
269 void *ret;
270
271 BUG_ON(_brk_start == 0);
272 BUG_ON(align & mask);
273
274 _brk_end = (_brk_end + mask) & ~mask;
275 BUG_ON((char *)(_brk_end + size) > __brk_limit);
276
277 ret = (void *)_brk_end;
278 _brk_end += size;
279
280 memset(ret, 0, size);
281
282 return ret;
283}
284
285#ifdef CONFIG_X86_32
286static void __init cleanup_highmap(void)
287{
288}
289#endif
290
291static void __init reserve_brk(void)
292{
293 if (_brk_end > _brk_start)
294 memblock_reserve(__pa_symbol(_brk_start),
295 _brk_end - _brk_start);
296
297 /* Mark brk area as locked down and no longer taking any
298 new allocations */
299 _brk_start = 0;
300}
301
302u64 relocated_ramdisk;
303
304#ifdef CONFIG_BLK_DEV_INITRD
305
306static u64 __init get_ramdisk_image(void)
307{
308 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
309
310 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
311
312 return ramdisk_image;
313}
314static u64 __init get_ramdisk_size(void)
315{
316 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
317
318 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
319
320 return ramdisk_size;
321}
322
323static void __init relocate_initrd(void)
324{
325 /* Assume only end is not page aligned */
326 u64 ramdisk_image = get_ramdisk_image();
327 u64 ramdisk_size = get_ramdisk_size();
328 u64 area_size = PAGE_ALIGN(ramdisk_size);
329
330 /* We need to move the initrd down into directly mapped mem */
331 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
332 area_size, PAGE_SIZE);
333
334 if (!relocated_ramdisk)
335 panic("Cannot find place for new RAMDISK of size %lld\n",
336 ramdisk_size);
337
338 /* Note: this includes all the mem currently occupied by
339 the initrd, we rely on that fact to keep the data intact. */
340 memblock_reserve(relocated_ramdisk, area_size);
341 initrd_start = relocated_ramdisk + PAGE_OFFSET;
342 initrd_end = initrd_start + ramdisk_size;
343 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
344 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
345
346 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
347
348 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
349 " [mem %#010llx-%#010llx]\n",
350 ramdisk_image, ramdisk_image + ramdisk_size - 1,
351 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
352}
353
354static void __init early_reserve_initrd(void)
355{
356 /* Assume only end is not page aligned */
357 u64 ramdisk_image = get_ramdisk_image();
358 u64 ramdisk_size = get_ramdisk_size();
359 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
360
361 if (!boot_params.hdr.type_of_loader ||
362 !ramdisk_image || !ramdisk_size)
363 return; /* No initrd provided by bootloader */
364
365 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
366}
367static void __init reserve_initrd(void)
368{
369 /* Assume only end is not page aligned */
370 u64 ramdisk_image = get_ramdisk_image();
371 u64 ramdisk_size = get_ramdisk_size();
372 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
373 u64 mapped_size;
374
375 if (!boot_params.hdr.type_of_loader ||
376 !ramdisk_image || !ramdisk_size)
377 return; /* No initrd provided by bootloader */
378
379 initrd_start = 0;
380
381 mapped_size = memblock_mem_size(max_pfn_mapped);
382 if (ramdisk_size >= (mapped_size>>1))
383 panic("initrd too large to handle, "
384 "disabling initrd (%lld needed, %lld available)\n",
385 ramdisk_size, mapped_size>>1);
386
387 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
388 ramdisk_end - 1);
389
390 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
391 PFN_DOWN(ramdisk_end))) {
392 /* All are mapped, easy case */
393 initrd_start = ramdisk_image + PAGE_OFFSET;
394 initrd_end = initrd_start + ramdisk_size;
395 return;
396 }
397
398 relocate_initrd();
399
400 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
401}
402
403#else
404static void __init early_reserve_initrd(void)
405{
406}
407static void __init reserve_initrd(void)
408{
409}
410#endif /* CONFIG_BLK_DEV_INITRD */
411
412static void __init parse_setup_data(void)
413{
414 struct setup_data *data;
415 u64 pa_data, pa_next;
416
417 pa_data = boot_params.hdr.setup_data;
418 while (pa_data) {
419 u32 data_len, data_type;
420
421 data = early_memremap(pa_data, sizeof(*data));
422 data_len = data->len + sizeof(struct setup_data);
423 data_type = data->type;
424 pa_next = data->next;
425 early_memunmap(data, sizeof(*data));
426
427 switch (data_type) {
428 case SETUP_E820_EXT:
429 parse_e820_ext(pa_data, data_len);
430 break;
431 case SETUP_DTB:
432 add_dtb(pa_data);
433 break;
434 case SETUP_EFI:
435 parse_efi_setup(pa_data, data_len);
436 break;
437 default:
438 break;
439 }
440 pa_data = pa_next;
441 }
442}
443
444static void __init e820_reserve_setup_data(void)
445{
446 struct setup_data *data;
447 u64 pa_data;
448
449 pa_data = boot_params.hdr.setup_data;
450 if (!pa_data)
451 return;
452
453 while (pa_data) {
454 data = early_memremap(pa_data, sizeof(*data));
455 e820_update_range(pa_data, sizeof(*data)+data->len,
456 E820_RAM, E820_RESERVED_KERN);
457 pa_data = data->next;
458 early_memunmap(data, sizeof(*data));
459 }
460
461 sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
462 memcpy(e820_saved, e820, sizeof(struct e820map));
463 printk(KERN_INFO "extended physical RAM map:\n");
464 e820_print_map("reserve setup_data");
465}
466
467static void __init memblock_x86_reserve_range_setup_data(void)
468{
469 struct setup_data *data;
470 u64 pa_data;
471
472 pa_data = boot_params.hdr.setup_data;
473 while (pa_data) {
474 data = early_memremap(pa_data, sizeof(*data));
475 memblock_reserve(pa_data, sizeof(*data) + data->len);
476 pa_data = data->next;
477 early_memunmap(data, sizeof(*data));
478 }
479}
480
481/*
482 * --------- Crashkernel reservation ------------------------------
483 */
484
485#ifdef CONFIG_KEXEC_CORE
486
487/* 16M alignment for crash kernel regions */
488#define CRASH_ALIGN (16 << 20)
489
490/*
491 * Keep the crash kernel below this limit. On 32 bits earlier kernels
492 * would limit the kernel to the low 512 MiB due to mapping restrictions.
493 * On 64bit, old kexec-tools need to under 896MiB.
494 */
495#ifdef CONFIG_X86_32
496# define CRASH_ADDR_LOW_MAX (512 << 20)
497# define CRASH_ADDR_HIGH_MAX (512 << 20)
498#else
499# define CRASH_ADDR_LOW_MAX (896UL << 20)
500# define CRASH_ADDR_HIGH_MAX MAXMEM
501#endif
502
503static int __init reserve_crashkernel_low(void)
504{
505#ifdef CONFIG_X86_64
506 unsigned long long base, low_base = 0, low_size = 0;
507 unsigned long total_low_mem;
508 int ret;
509
510 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
511
512 /* crashkernel=Y,low */
513 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
514 if (ret) {
515 /*
516 * two parts from lib/swiotlb.c:
517 * -swiotlb size: user-specified with swiotlb= or default.
518 *
519 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
520 * to 8M for other buffers that may need to stay low too. Also
521 * make sure we allocate enough extra low memory so that we
522 * don't run out of DMA buffers for 32-bit devices.
523 */
524 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
525 } else {
526 /* passed with crashkernel=0,low ? */
527 if (!low_size)
528 return 0;
529 }
530
531 low_base = memblock_find_in_range(low_size, 1ULL << 32, low_size, CRASH_ALIGN);
532 if (!low_base) {
533 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
534 (unsigned long)(low_size >> 20));
535 return -ENOMEM;
536 }
537
538 ret = memblock_reserve(low_base, low_size);
539 if (ret) {
540 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
541 return ret;
542 }
543
544 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
545 (unsigned long)(low_size >> 20),
546 (unsigned long)(low_base >> 20),
547 (unsigned long)(total_low_mem >> 20));
548
549 crashk_low_res.start = low_base;
550 crashk_low_res.end = low_base + low_size - 1;
551 insert_resource(&iomem_resource, &crashk_low_res);
552#endif
553 return 0;
554}
555
556static void __init reserve_crashkernel(void)
557{
558 unsigned long long crash_size, crash_base, total_mem;
559 bool high = false;
560 int ret;
561
562 total_mem = memblock_phys_mem_size();
563
564 /* crashkernel=XM */
565 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
566 if (ret != 0 || crash_size <= 0) {
567 /* crashkernel=X,high */
568 ret = parse_crashkernel_high(boot_command_line, total_mem,
569 &crash_size, &crash_base);
570 if (ret != 0 || crash_size <= 0)
571 return;
572 high = true;
573 }
574
575 /* 0 means: find the address automatically */
576 if (crash_base <= 0) {
577 /*
578 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
579 */
580 crash_base = memblock_find_in_range(CRASH_ALIGN,
581 high ? CRASH_ADDR_HIGH_MAX
582 : CRASH_ADDR_LOW_MAX,
583 crash_size, CRASH_ALIGN);
584 if (!crash_base) {
585 pr_info("crashkernel reservation failed - No suitable area found.\n");
586 return;
587 }
588
589 } else {
590 unsigned long long start;
591
592 start = memblock_find_in_range(crash_base,
593 crash_base + crash_size,
594 crash_size, 1 << 20);
595 if (start != crash_base) {
596 pr_info("crashkernel reservation failed - memory is in use.\n");
597 return;
598 }
599 }
600 ret = memblock_reserve(crash_base, crash_size);
601 if (ret) {
602 pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
603 return;
604 }
605
606 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
607 memblock_free(crash_base, crash_size);
608 return;
609 }
610
611 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
612 (unsigned long)(crash_size >> 20),
613 (unsigned long)(crash_base >> 20),
614 (unsigned long)(total_mem >> 20));
615
616 crashk_res.start = crash_base;
617 crashk_res.end = crash_base + crash_size - 1;
618 insert_resource(&iomem_resource, &crashk_res);
619}
620#else
621static void __init reserve_crashkernel(void)
622{
623}
624#endif
625
626static struct resource standard_io_resources[] = {
627 { .name = "dma1", .start = 0x00, .end = 0x1f,
628 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
629 { .name = "pic1", .start = 0x20, .end = 0x21,
630 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
631 { .name = "timer0", .start = 0x40, .end = 0x43,
632 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
633 { .name = "timer1", .start = 0x50, .end = 0x53,
634 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
635 { .name = "keyboard", .start = 0x60, .end = 0x60,
636 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
637 { .name = "keyboard", .start = 0x64, .end = 0x64,
638 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
639 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
640 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
641 { .name = "pic2", .start = 0xa0, .end = 0xa1,
642 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
643 { .name = "dma2", .start = 0xc0, .end = 0xdf,
644 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
645 { .name = "fpu", .start = 0xf0, .end = 0xff,
646 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
647};
648
649void __init reserve_standard_io_resources(void)
650{
651 int i;
652
653 /* request I/O space for devices used on all i[345]86 PCs */
654 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
655 request_resource(&ioport_resource, &standard_io_resources[i]);
656
657}
658
659static __init void reserve_ibft_region(void)
660{
661 unsigned long addr, size = 0;
662
663 addr = find_ibft_region(&size);
664
665 if (size)
666 memblock_reserve(addr, size);
667}
668
669static bool __init snb_gfx_workaround_needed(void)
670{
671#ifdef CONFIG_PCI
672 int i;
673 u16 vendor, devid;
674 static const __initconst u16 snb_ids[] = {
675 0x0102,
676 0x0112,
677 0x0122,
678 0x0106,
679 0x0116,
680 0x0126,
681 0x010a,
682 };
683
684 /* Assume no if something weird is going on with PCI */
685 if (!early_pci_allowed())
686 return false;
687
688 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
689 if (vendor != 0x8086)
690 return false;
691
692 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
693 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
694 if (devid == snb_ids[i])
695 return true;
696#endif
697
698 return false;
699}
700
701/*
702 * Sandy Bridge graphics has trouble with certain ranges, exclude
703 * them from allocation.
704 */
705static void __init trim_snb_memory(void)
706{
707 static const __initconst unsigned long bad_pages[] = {
708 0x20050000,
709 0x20110000,
710 0x20130000,
711 0x20138000,
712 0x40004000,
713 };
714 int i;
715
716 if (!snb_gfx_workaround_needed())
717 return;
718
719 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
720
721 /*
722 * Reserve all memory below the 1 MB mark that has not
723 * already been reserved.
724 */
725 memblock_reserve(0, 1<<20);
726
727 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
728 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
729 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
730 bad_pages[i]);
731 }
732}
733
734/*
735 * Here we put platform-specific memory range workarounds, i.e.
736 * memory known to be corrupt or otherwise in need to be reserved on
737 * specific platforms.
738 *
739 * If this gets used more widely it could use a real dispatch mechanism.
740 */
741static void __init trim_platform_memory_ranges(void)
742{
743 trim_snb_memory();
744}
745
746static void __init trim_bios_range(void)
747{
748 /*
749 * A special case is the first 4Kb of memory;
750 * This is a BIOS owned area, not kernel ram, but generally
751 * not listed as such in the E820 table.
752 *
753 * This typically reserves additional memory (64KiB by default)
754 * since some BIOSes are known to corrupt low memory. See the
755 * Kconfig help text for X86_RESERVE_LOW.
756 */
757 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
758
759 /*
760 * special case: Some BIOSen report the PC BIOS
761 * area (640->1Mb) as ram even though it is not.
762 * take them out.
763 */
764 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
765
766 sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
767}
768
769/* called before trim_bios_range() to spare extra sanitize */
770static void __init e820_add_kernel_range(void)
771{
772 u64 start = __pa_symbol(_text);
773 u64 size = __pa_symbol(_end) - start;
774
775 /*
776 * Complain if .text .data and .bss are not marked as E820_RAM and
777 * attempt to fix it by adding the range. We may have a confused BIOS,
778 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
779 * exclude kernel range. If we really are running on top non-RAM,
780 * we will crash later anyways.
781 */
782 if (e820_all_mapped(start, start + size, E820_RAM))
783 return;
784
785 pr_warn(".text .data .bss are not marked as E820_RAM!\n");
786 e820_remove_range(start, size, E820_RAM, 0);
787 e820_add_region(start, size, E820_RAM);
788}
789
790static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
791
792static int __init parse_reservelow(char *p)
793{
794 unsigned long long size;
795
796 if (!p)
797 return -EINVAL;
798
799 size = memparse(p, &p);
800
801 if (size < 4096)
802 size = 4096;
803
804 if (size > 640*1024)
805 size = 640*1024;
806
807 reserve_low = size;
808
809 return 0;
810}
811
812early_param("reservelow", parse_reservelow);
813
814static void __init trim_low_memory_range(void)
815{
816 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
817}
818
819/*
820 * Dump out kernel offset information on panic.
821 */
822static int
823dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
824{
825 if (kaslr_enabled()) {
826 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
827 kaslr_offset(),
828 __START_KERNEL,
829 __START_KERNEL_map,
830 MODULES_VADDR-1);
831 } else {
832 pr_emerg("Kernel Offset: disabled\n");
833 }
834
835 return 0;
836}
837
838/*
839 * Determine if we were loaded by an EFI loader. If so, then we have also been
840 * passed the efi memmap, systab, etc., so we should use these data structures
841 * for initialization. Note, the efi init code path is determined by the
842 * global efi_enabled. This allows the same kernel image to be used on existing
843 * systems (with a traditional BIOS) as well as on EFI systems.
844 */
845/*
846 * setup_arch - architecture-specific boot-time initializations
847 *
848 * Note: On x86_64, fixmaps are ready for use even before this is called.
849 */
850
851void __init setup_arch(char **cmdline_p)
852{
853 memblock_reserve(__pa_symbol(_text),
854 (unsigned long)__bss_stop - (unsigned long)_text);
855
856 early_reserve_initrd();
857
858 /*
859 * At this point everything still needed from the boot loader
860 * or BIOS or kernel text should be early reserved or marked not
861 * RAM in e820. All other memory is free game.
862 */
863
864#ifdef CONFIG_X86_32
865 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
866
867 /*
868 * copy kernel address range established so far and switch
869 * to the proper swapper page table
870 */
871 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
872 initial_page_table + KERNEL_PGD_BOUNDARY,
873 KERNEL_PGD_PTRS);
874
875 load_cr3(swapper_pg_dir);
876 /*
877 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
878 * a cr3 based tlb flush, so the following __flush_tlb_all()
879 * will not flush anything because the cpu quirk which clears
880 * X86_FEATURE_PGE has not been invoked yet. Though due to the
881 * load_cr3() above the TLB has been flushed already. The
882 * quirk is invoked before subsequent calls to __flush_tlb_all()
883 * so proper operation is guaranteed.
884 */
885 __flush_tlb_all();
886#else
887 printk(KERN_INFO "Command line: %s\n", boot_command_line);
888#endif
889
890 /*
891 * If we have OLPC OFW, we might end up relocating the fixmap due to
892 * reserve_top(), so do this before touching the ioremap area.
893 */
894 olpc_ofw_detect();
895
896 early_trap_init();
897 early_cpu_init();
898 early_ioremap_init();
899
900 setup_olpc_ofw_pgd();
901
902 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
903 screen_info = boot_params.screen_info;
904 edid_info = boot_params.edid_info;
905#ifdef CONFIG_X86_32
906 apm_info.bios = boot_params.apm_bios_info;
907 ist_info = boot_params.ist_info;
908#endif
909 saved_video_mode = boot_params.hdr.vid_mode;
910 bootloader_type = boot_params.hdr.type_of_loader;
911 if ((bootloader_type >> 4) == 0xe) {
912 bootloader_type &= 0xf;
913 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
914 }
915 bootloader_version = bootloader_type & 0xf;
916 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
917
918#ifdef CONFIG_BLK_DEV_RAM
919 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
920 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
921 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
922#endif
923#ifdef CONFIG_EFI
924 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
925 EFI32_LOADER_SIGNATURE, 4)) {
926 set_bit(EFI_BOOT, &efi.flags);
927 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
928 EFI64_LOADER_SIGNATURE, 4)) {
929 set_bit(EFI_BOOT, &efi.flags);
930 set_bit(EFI_64BIT, &efi.flags);
931 }
932
933 if (efi_enabled(EFI_BOOT))
934 efi_memblock_x86_reserve_range();
935#endif
936
937 x86_init.oem.arch_setup();
938
939 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
940 setup_memory_map();
941 parse_setup_data();
942
943 copy_edd();
944
945 if (!boot_params.hdr.root_flags)
946 root_mountflags &= ~MS_RDONLY;
947 init_mm.start_code = (unsigned long) _text;
948 init_mm.end_code = (unsigned long) _etext;
949 init_mm.end_data = (unsigned long) _edata;
950 init_mm.brk = _brk_end;
951
952 mpx_mm_init(&init_mm);
953
954 code_resource.start = __pa_symbol(_text);
955 code_resource.end = __pa_symbol(_etext)-1;
956 data_resource.start = __pa_symbol(_etext);
957 data_resource.end = __pa_symbol(_edata)-1;
958 bss_resource.start = __pa_symbol(__bss_start);
959 bss_resource.end = __pa_symbol(__bss_stop)-1;
960
961#ifdef CONFIG_CMDLINE_BOOL
962#ifdef CONFIG_CMDLINE_OVERRIDE
963 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
964#else
965 if (builtin_cmdline[0]) {
966 /* append boot loader cmdline to builtin */
967 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
968 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
969 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
970 }
971#endif
972#endif
973
974 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
975 *cmdline_p = command_line;
976
977 /*
978 * x86_configure_nx() is called before parse_early_param() to detect
979 * whether hardware doesn't support NX (so that the early EHCI debug
980 * console setup can safely call set_fixmap()). It may then be called
981 * again from within noexec_setup() during parsing early parameters
982 * to honor the respective command line option.
983 */
984 x86_configure_nx();
985
986 parse_early_param();
987
988#ifdef CONFIG_MEMORY_HOTPLUG
989 /*
990 * Memory used by the kernel cannot be hot-removed because Linux
991 * cannot migrate the kernel pages. When memory hotplug is
992 * enabled, we should prevent memblock from allocating memory
993 * for the kernel.
994 *
995 * ACPI SRAT records all hotpluggable memory ranges. But before
996 * SRAT is parsed, we don't know about it.
997 *
998 * The kernel image is loaded into memory at very early time. We
999 * cannot prevent this anyway. So on NUMA system, we set any
1000 * node the kernel resides in as un-hotpluggable.
1001 *
1002 * Since on modern servers, one node could have double-digit
1003 * gigabytes memory, we can assume the memory around the kernel
1004 * image is also un-hotpluggable. So before SRAT is parsed, just
1005 * allocate memory near the kernel image to try the best to keep
1006 * the kernel away from hotpluggable memory.
1007 */
1008 if (movable_node_is_enabled())
1009 memblock_set_bottom_up(true);
1010#endif
1011
1012 x86_report_nx();
1013
1014 /* after early param, so could get panic from serial */
1015 memblock_x86_reserve_range_setup_data();
1016
1017 if (acpi_mps_check()) {
1018#ifdef CONFIG_X86_LOCAL_APIC
1019 disable_apic = 1;
1020#endif
1021 setup_clear_cpu_cap(X86_FEATURE_APIC);
1022 }
1023
1024#ifdef CONFIG_PCI
1025 if (pci_early_dump_regs)
1026 early_dump_pci_devices();
1027#endif
1028
1029 /* update the e820_saved too */
1030 e820_reserve_setup_data();
1031 finish_e820_parsing();
1032
1033 if (efi_enabled(EFI_BOOT))
1034 efi_init();
1035
1036 dmi_scan_machine();
1037 dmi_memdev_walk();
1038 dmi_set_dump_stack_arch_desc();
1039
1040 /*
1041 * VMware detection requires dmi to be available, so this
1042 * needs to be done after dmi_scan_machine, for the BP.
1043 */
1044 init_hypervisor_platform();
1045
1046 x86_init.resources.probe_roms();
1047
1048 /* after parse_early_param, so could debug it */
1049 insert_resource(&iomem_resource, &code_resource);
1050 insert_resource(&iomem_resource, &data_resource);
1051 insert_resource(&iomem_resource, &bss_resource);
1052
1053 e820_add_kernel_range();
1054 trim_bios_range();
1055#ifdef CONFIG_X86_32
1056 if (ppro_with_ram_bug()) {
1057 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1058 E820_RESERVED);
1059 sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
1060 printk(KERN_INFO "fixed physical RAM map:\n");
1061 e820_print_map("bad_ppro");
1062 }
1063#else
1064 early_gart_iommu_check();
1065#endif
1066
1067 /*
1068 * partially used pages are not usable - thus
1069 * we are rounding upwards:
1070 */
1071 max_pfn = e820_end_of_ram_pfn();
1072
1073 /* update e820 for memory not covered by WB MTRRs */
1074 mtrr_bp_init();
1075 if (mtrr_trim_uncached_memory(max_pfn))
1076 max_pfn = e820_end_of_ram_pfn();
1077
1078 max_possible_pfn = max_pfn;
1079
1080 /*
1081 * Define random base addresses for memory sections after max_pfn is
1082 * defined and before each memory section base is used.
1083 */
1084 kernel_randomize_memory();
1085
1086#ifdef CONFIG_X86_32
1087 /* max_low_pfn get updated here */
1088 find_low_pfn_range();
1089#else
1090 check_x2apic();
1091
1092 /* How many end-of-memory variables you have, grandma! */
1093 /* need this before calling reserve_initrd */
1094 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1095 max_low_pfn = e820_end_of_low_ram_pfn();
1096 else
1097 max_low_pfn = max_pfn;
1098
1099 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1100#endif
1101
1102 /*
1103 * Find and reserve possible boot-time SMP configuration:
1104 */
1105 find_smp_config();
1106
1107 reserve_ibft_region();
1108
1109 early_alloc_pgt_buf();
1110
1111 /*
1112 * Need to conclude brk, before memblock_x86_fill()
1113 * it could use memblock_find_in_range, could overlap with
1114 * brk area.
1115 */
1116 reserve_brk();
1117
1118 cleanup_highmap();
1119
1120 memblock_set_current_limit(ISA_END_ADDRESS);
1121 memblock_x86_fill();
1122
1123 reserve_bios_regions();
1124
1125 if (efi_enabled(EFI_MEMMAP)) {
1126 efi_fake_memmap();
1127 efi_find_mirror();
1128 efi_esrt_init();
1129
1130 /*
1131 * The EFI specification says that boot service code won't be
1132 * called after ExitBootServices(). This is, in fact, a lie.
1133 */
1134 efi_reserve_boot_services();
1135 }
1136
1137 /* preallocate 4k for mptable mpc */
1138 early_reserve_e820_mpc_new();
1139
1140#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1141 setup_bios_corruption_check();
1142#endif
1143
1144#ifdef CONFIG_X86_32
1145 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1146 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1147#endif
1148
1149 reserve_real_mode();
1150
1151 trim_platform_memory_ranges();
1152 trim_low_memory_range();
1153
1154 init_mem_mapping();
1155
1156 early_trap_pf_init();
1157
1158 /*
1159 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1160 * with the current CR4 value. This may not be necessary, but
1161 * auditing all the early-boot CR4 manipulation would be needed to
1162 * rule it out.
1163 */
1164 mmu_cr4_features = __read_cr4();
1165
1166 memblock_set_current_limit(get_max_mapped());
1167
1168 /*
1169 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1170 */
1171
1172#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1173 if (init_ohci1394_dma_early)
1174 init_ohci1394_dma_on_all_controllers();
1175#endif
1176 /* Allocate bigger log buffer */
1177 setup_log_buf(1);
1178
1179 reserve_initrd();
1180
1181 acpi_table_upgrade();
1182
1183 vsmp_init();
1184
1185 io_delay_init();
1186
1187 /*
1188 * Parse the ACPI tables for possible boot-time SMP configuration.
1189 */
1190 acpi_boot_table_init();
1191
1192 early_acpi_boot_init();
1193
1194 initmem_init();
1195 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1196
1197 /*
1198 * Reserve memory for crash kernel after SRAT is parsed so that it
1199 * won't consume hotpluggable memory.
1200 */
1201 reserve_crashkernel();
1202
1203 memblock_find_dma_reserve();
1204
1205#ifdef CONFIG_KVM_GUEST
1206 kvmclock_init();
1207#endif
1208
1209 x86_init.paging.pagetable_init();
1210
1211 kasan_init();
1212
1213#ifdef CONFIG_X86_32
1214 /* sync back kernel address range */
1215 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1216 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1217 KERNEL_PGD_PTRS);
1218
1219 /*
1220 * sync back low identity map too. It is used for example
1221 * in the 32-bit EFI stub.
1222 */
1223 clone_pgd_range(initial_page_table,
1224 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1225 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1226#endif
1227
1228 tboot_probe();
1229
1230 map_vsyscall();
1231
1232 generic_apic_probe();
1233
1234 early_quirks();
1235
1236 /*
1237 * Read APIC and some other early information from ACPI tables.
1238 */
1239 acpi_boot_init();
1240 sfi_init();
1241 x86_dtb_init();
1242
1243 /*
1244 * get boot-time SMP configuration:
1245 */
1246 get_smp_config();
1247
1248 /*
1249 * Systems w/o ACPI and mptables might not have it mapped the local
1250 * APIC yet, but prefill_possible_map() might need to access it.
1251 */
1252 init_apic_mappings();
1253
1254 prefill_possible_map();
1255
1256 init_cpu_to_node();
1257
1258 io_apic_init_mappings();
1259
1260 kvm_guest_init();
1261
1262 e820_reserve_resources();
1263 e820_mark_nosave_regions(max_low_pfn);
1264
1265 x86_init.resources.reserve_resources();
1266
1267 e820_setup_gap();
1268
1269#ifdef CONFIG_VT
1270#if defined(CONFIG_VGA_CONSOLE)
1271 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1272 conswitchp = &vga_con;
1273#elif defined(CONFIG_DUMMY_CONSOLE)
1274 conswitchp = &dummy_con;
1275#endif
1276#endif
1277 x86_init.oem.banner();
1278
1279 x86_init.timers.wallclock_init();
1280
1281 mcheck_init();
1282
1283 arch_init_ideal_nops();
1284
1285 register_refined_jiffies(CLOCK_TICK_RATE);
1286
1287#ifdef CONFIG_EFI
1288 if (efi_enabled(EFI_BOOT))
1289 efi_apply_memmap_quirks();
1290#endif
1291}
1292
1293#ifdef CONFIG_X86_32
1294
1295static struct resource video_ram_resource = {
1296 .name = "Video RAM area",
1297 .start = 0xa0000,
1298 .end = 0xbffff,
1299 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1300};
1301
1302void __init i386_reserve_resources(void)
1303{
1304 request_resource(&iomem_resource, &video_ram_resource);
1305 reserve_standard_io_resources();
1306}
1307
1308#endif /* CONFIG_X86_32 */
1309
1310static struct notifier_block kernel_offset_notifier = {
1311 .notifier_call = dump_kernel_offset
1312};
1313
1314static int __init register_kernel_offset_dumper(void)
1315{
1316 atomic_notifier_chain_register(&panic_notifier_list,
1317 &kernel_offset_notifier);
1318 return 0;
1319}
1320__initcall(register_kernel_offset_dumper);
1321
1322void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
1323{
1324 if (!boot_cpu_has(X86_FEATURE_OSPKE))
1325 return;
1326
1327 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
1328}