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/dmi.h>
13#include <linux/efi.h>
14#include <linux/ima.h>
15#include <linux/init_ohci1394_dma.h>
16#include <linux/initrd.h>
17#include <linux/iscsi_ibft.h>
18#include <linux/memblock.h>
19#include <linux/panic_notifier.h>
20#include <linux/pci.h>
21#include <linux/root_dev.h>
22#include <linux/hugetlb.h>
23#include <linux/tboot.h>
24#include <linux/usb/xhci-dbgp.h>
25#include <linux/static_call.h>
26#include <linux/swiotlb.h>
27#include <linux/random.h>
28
29#include <uapi/linux/mount.h>
30
31#include <xen/xen.h>
32
33#include <asm/apic.h>
34#include <asm/efi.h>
35#include <asm/numa.h>
36#include <asm/bios_ebda.h>
37#include <asm/bugs.h>
38#include <asm/cacheinfo.h>
39#include <asm/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 size: _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
229u64 relocated_ramdisk;
230
231#ifdef CONFIG_BLK_DEV_INITRD
232
233static u64 __init get_ramdisk_image(void)
234{
235 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
236
237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
238
239 if (ramdisk_image == 0)
240 ramdisk_image = phys_initrd_start;
241
242 return ramdisk_image;
243}
244static u64 __init get_ramdisk_size(void)
245{
246 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
247
248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
249
250 if (ramdisk_size == 0)
251 ramdisk_size = phys_initrd_size;
252
253 return ramdisk_size;
254}
255
256static void __init relocate_initrd(void)
257{
258 /* Assume only end is not page aligned */
259 u64 ramdisk_image = get_ramdisk_image();
260 u64 ramdisk_size = get_ramdisk_size();
261 u64 area_size = PAGE_ALIGN(ramdisk_size);
262
263 /* We need to move the initrd down into directly mapped mem */
264 relocated_ramdisk = memblock_phys_alloc_range(size: area_size, PAGE_SIZE, start: 0,
265 PFN_PHYS(max_pfn_mapped));
266 if (!relocated_ramdisk)
267 panic(fmt: "Cannot find place for new RAMDISK of size %lld\n",
268 ramdisk_size);
269
270 initrd_start = relocated_ramdisk + PAGE_OFFSET;
271 initrd_end = initrd_start + ramdisk_size;
272 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
273 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
274
275 copy_from_early_mem(dest: (void *)initrd_start, src: ramdisk_image, size: ramdisk_size);
276
277 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
278 " [mem %#010llx-%#010llx]\n",
279 ramdisk_image, ramdisk_image + ramdisk_size - 1,
280 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
281}
282
283static void __init early_reserve_initrd(void)
284{
285 /* Assume only end is not page aligned */
286 u64 ramdisk_image = get_ramdisk_image();
287 u64 ramdisk_size = get_ramdisk_size();
288 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
289
290 if (!boot_params.hdr.type_of_loader ||
291 !ramdisk_image || !ramdisk_size)
292 return; /* No initrd provided by bootloader */
293
294 memblock_reserve(base: ramdisk_image, size: ramdisk_end - ramdisk_image);
295}
296
297static void __init reserve_initrd(void)
298{
299 /* Assume only end is not page aligned */
300 u64 ramdisk_image = get_ramdisk_image();
301 u64 ramdisk_size = get_ramdisk_size();
302 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
303
304 if (!boot_params.hdr.type_of_loader ||
305 !ramdisk_image || !ramdisk_size)
306 return; /* No initrd provided by bootloader */
307
308 initrd_start = 0;
309
310 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
311 ramdisk_end - 1);
312
313 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
314 PFN_DOWN(ramdisk_end))) {
315 /* All are mapped, easy case */
316 initrd_start = ramdisk_image + PAGE_OFFSET;
317 initrd_end = initrd_start + ramdisk_size;
318 return;
319 }
320
321 relocate_initrd();
322
323 memblock_phys_free(base: ramdisk_image, size: ramdisk_end - ramdisk_image);
324}
325
326#else
327static void __init early_reserve_initrd(void)
328{
329}
330static void __init reserve_initrd(void)
331{
332}
333#endif /* CONFIG_BLK_DEV_INITRD */
334
335static void __init add_early_ima_buffer(u64 phys_addr)
336{
337#ifdef CONFIG_IMA
338 struct ima_setup_data *data;
339
340 data = early_memremap(phys_addr: phys_addr + sizeof(struct setup_data), size: sizeof(*data));
341 if (!data) {
342 pr_warn("setup: failed to memremap ima_setup_data entry\n");
343 return;
344 }
345
346 if (data->size) {
347 memblock_reserve(base: data->addr, size: data->size);
348 ima_kexec_buffer_phys = data->addr;
349 ima_kexec_buffer_size = data->size;
350 }
351
352 early_memunmap(addr: data, size: sizeof(*data));
353#else
354 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
355#endif
356}
357
358#if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
359int __init ima_free_kexec_buffer(void)
360{
361 if (!ima_kexec_buffer_size)
362 return -ENOENT;
363
364 memblock_free_late(ima_kexec_buffer_phys,
365 ima_kexec_buffer_size);
366
367 ima_kexec_buffer_phys = 0;
368 ima_kexec_buffer_size = 0;
369
370 return 0;
371}
372
373int __init ima_get_kexec_buffer(void **addr, size_t *size)
374{
375 if (!ima_kexec_buffer_size)
376 return -ENOENT;
377
378 *addr = __va(ima_kexec_buffer_phys);
379 *size = ima_kexec_buffer_size;
380
381 return 0;
382}
383#endif
384
385static void __init parse_setup_data(void)
386{
387 struct setup_data *data;
388 u64 pa_data, pa_next;
389
390 pa_data = boot_params.hdr.setup_data;
391 while (pa_data) {
392 u32 data_len, data_type;
393
394 data = early_memremap(phys_addr: pa_data, size: sizeof(*data));
395 data_len = data->len + sizeof(struct setup_data);
396 data_type = data->type;
397 pa_next = data->next;
398 early_memunmap(addr: data, size: sizeof(*data));
399
400 switch (data_type) {
401 case SETUP_E820_EXT:
402 e820__memory_setup_extended(phys_addr: pa_data, data_len);
403 break;
404 case SETUP_DTB:
405 add_dtb(data: pa_data);
406 break;
407 case SETUP_EFI:
408 parse_efi_setup(phys_addr: pa_data, data_len);
409 break;
410 case SETUP_IMA:
411 add_early_ima_buffer(phys_addr: pa_data);
412 break;
413 case SETUP_RNG_SEED:
414 data = early_memremap(phys_addr: pa_data, size: data_len);
415 add_bootloader_randomness(buf: data->data, len: data->len);
416 /* Zero seed for forward secrecy. */
417 memzero_explicit(s: data->data, count: data->len);
418 /* Zero length in case we find ourselves back here by accident. */
419 memzero_explicit(s: &data->len, count: sizeof(data->len));
420 early_memunmap(addr: data, size: data_len);
421 break;
422 default:
423 break;
424 }
425 pa_data = pa_next;
426 }
427}
428
429static void __init memblock_x86_reserve_range_setup_data(void)
430{
431 struct setup_indirect *indirect;
432 struct setup_data *data;
433 u64 pa_data, pa_next;
434 u32 len;
435
436 pa_data = boot_params.hdr.setup_data;
437 while (pa_data) {
438 data = early_memremap(phys_addr: pa_data, size: sizeof(*data));
439 if (!data) {
440 pr_warn("setup: failed to memremap setup_data entry\n");
441 return;
442 }
443
444 len = sizeof(*data);
445 pa_next = data->next;
446
447 memblock_reserve(base: pa_data, size: sizeof(*data) + data->len);
448
449 if (data->type == SETUP_INDIRECT) {
450 len += data->len;
451 early_memunmap(addr: data, size: sizeof(*data));
452 data = early_memremap(phys_addr: pa_data, size: len);
453 if (!data) {
454 pr_warn("setup: failed to memremap indirect setup_data\n");
455 return;
456 }
457
458 indirect = (struct setup_indirect *)data->data;
459
460 if (indirect->type != SETUP_INDIRECT)
461 memblock_reserve(base: indirect->addr, size: indirect->len);
462 }
463
464 pa_data = pa_next;
465 early_memunmap(addr: data, size: len);
466 }
467}
468
469static void __init arch_reserve_crashkernel(void)
470{
471 unsigned long long crash_base, crash_size, low_size = 0;
472 char *cmdline = boot_command_line;
473 bool high = false;
474 int ret;
475
476 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
477 return;
478
479 ret = parse_crashkernel(cmdline, system_ram: memblock_phys_mem_size(),
480 crash_size: &crash_size, crash_base: &crash_base,
481 low_size: &low_size, high: &high);
482 if (ret)
483 return;
484
485 if (xen_pv_domain()) {
486 pr_info("Ignoring crashkernel for a Xen PV domain\n");
487 return;
488 }
489
490 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
491 crash_low_size: low_size, high);
492}
493
494static struct resource standard_io_resources[] = {
495 { .name = "dma1", .start = 0x00, .end = 0x1f,
496 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
497 { .name = "pic1", .start = 0x20, .end = 0x21,
498 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
499 { .name = "timer0", .start = 0x40, .end = 0x43,
500 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
501 { .name = "timer1", .start = 0x50, .end = 0x53,
502 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
503 { .name = "keyboard", .start = 0x60, .end = 0x60,
504 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
505 { .name = "keyboard", .start = 0x64, .end = 0x64,
506 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
507 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
508 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
509 { .name = "pic2", .start = 0xa0, .end = 0xa1,
510 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
511 { .name = "dma2", .start = 0xc0, .end = 0xdf,
512 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
513 { .name = "fpu", .start = 0xf0, .end = 0xff,
514 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
515};
516
517void __init reserve_standard_io_resources(void)
518{
519 int i;
520
521 /* request I/O space for devices used on all i[345]86 PCs */
522 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
523 request_resource(root: &ioport_resource, new: &standard_io_resources[i]);
524
525}
526
527static bool __init snb_gfx_workaround_needed(void)
528{
529#ifdef CONFIG_PCI
530 int i;
531 u16 vendor, devid;
532 static const __initconst u16 snb_ids[] = {
533 0x0102,
534 0x0112,
535 0x0122,
536 0x0106,
537 0x0116,
538 0x0126,
539 0x010a,
540 };
541
542 /* Assume no if something weird is going on with PCI */
543 if (!early_pci_allowed())
544 return false;
545
546 vendor = read_pci_config_16(bus: 0, slot: 2, func: 0, PCI_VENDOR_ID);
547 if (vendor != 0x8086)
548 return false;
549
550 devid = read_pci_config_16(bus: 0, slot: 2, func: 0, PCI_DEVICE_ID);
551 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
552 if (devid == snb_ids[i])
553 return true;
554#endif
555
556 return false;
557}
558
559/*
560 * Sandy Bridge graphics has trouble with certain ranges, exclude
561 * them from allocation.
562 */
563static void __init trim_snb_memory(void)
564{
565 static const __initconst unsigned long bad_pages[] = {
566 0x20050000,
567 0x20110000,
568 0x20130000,
569 0x20138000,
570 0x40004000,
571 };
572 int i;
573
574 if (!snb_gfx_workaround_needed())
575 return;
576
577 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
578
579 /*
580 * SandyBridge integrated graphics devices have a bug that prevents
581 * them from accessing certain memory ranges, namely anything below
582 * 1M and in the pages listed in bad_pages[] above.
583 *
584 * To avoid these pages being ever accessed by SNB gfx devices reserve
585 * bad_pages that have not already been reserved at boot time.
586 * All memory below the 1 MB mark is anyway reserved later during
587 * setup_arch(), so there is no need to reserve it here.
588 */
589
590 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
591 if (memblock_reserve(base: bad_pages[i], PAGE_SIZE))
592 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
593 bad_pages[i]);
594 }
595}
596
597static void __init trim_bios_range(void)
598{
599 /*
600 * A special case is the first 4Kb of memory;
601 * This is a BIOS owned area, not kernel ram, but generally
602 * not listed as such in the E820 table.
603 *
604 * This typically reserves additional memory (64KiB by default)
605 * since some BIOSes are known to corrupt low memory. See the
606 * Kconfig help text for X86_RESERVE_LOW.
607 */
608 e820__range_update(start: 0, PAGE_SIZE, old_type: E820_TYPE_RAM, new_type: E820_TYPE_RESERVED);
609
610 /*
611 * special case: Some BIOSes report the PC BIOS
612 * area (640Kb -> 1Mb) as RAM even though it is not.
613 * take them out.
614 */
615 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, old_type: E820_TYPE_RAM, check_type: 1);
616
617 e820__update_table(table: e820_table);
618}
619
620/* called before trim_bios_range() to spare extra sanitize */
621static void __init e820_add_kernel_range(void)
622{
623 u64 start = __pa_symbol(_text);
624 u64 size = __pa_symbol(_end) - start;
625
626 /*
627 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
628 * attempt to fix it by adding the range. We may have a confused BIOS,
629 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
630 * exclude kernel range. If we really are running on top non-RAM,
631 * we will crash later anyways.
632 */
633 if (e820__mapped_all(start, end: start + size, type: E820_TYPE_RAM))
634 return;
635
636 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
637 e820__range_remove(start, size, old_type: E820_TYPE_RAM, check_type: 0);
638 e820__range_add(start, size, type: E820_TYPE_RAM);
639}
640
641static void __init early_reserve_memory(void)
642{
643 /*
644 * Reserve the memory occupied by the kernel between _text and
645 * __end_of_kernel_reserve symbols. Any kernel sections after the
646 * __end_of_kernel_reserve symbol must be explicitly reserved with a
647 * separate memblock_reserve() or they will be discarded.
648 */
649 memblock_reserve(__pa_symbol(_text),
650 size: (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
651
652 /*
653 * The first 4Kb of memory is a BIOS owned area, but generally it is
654 * not listed as such in the E820 table.
655 *
656 * Reserve the first 64K of memory since some BIOSes are known to
657 * corrupt low memory. After the real mode trampoline is allocated the
658 * rest of the memory below 640k is reserved.
659 *
660 * In addition, make sure page 0 is always reserved because on
661 * systems with L1TF its contents can be leaked to user processes.
662 */
663 memblock_reserve(base: 0, SZ_64K);
664
665 early_reserve_initrd();
666
667 memblock_x86_reserve_range_setup_data();
668
669 reserve_bios_regions();
670 trim_snb_memory();
671}
672
673/*
674 * Dump out kernel offset information on panic.
675 */
676static int
677dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
678{
679 if (kaslr_enabled()) {
680 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
681 kaslr_offset(),
682 __START_KERNEL,
683 __START_KERNEL_map,
684 MODULES_VADDR-1);
685 } else {
686 pr_emerg("Kernel Offset: disabled\n");
687 }
688
689 return 0;
690}
691
692void x86_configure_nx(void)
693{
694 if (boot_cpu_has(X86_FEATURE_NX))
695 __supported_pte_mask |= _PAGE_NX;
696 else
697 __supported_pte_mask &= ~_PAGE_NX;
698}
699
700static void __init x86_report_nx(void)
701{
702 if (!boot_cpu_has(X86_FEATURE_NX)) {
703 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
704 "missing in CPU!\n");
705 } else {
706#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
707 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
708#else
709 /* 32bit non-PAE kernel, NX cannot be used */
710 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
711 "cannot be enabled: non-PAE kernel!\n");
712#endif
713 }
714}
715
716/*
717 * Determine if we were loaded by an EFI loader. If so, then we have also been
718 * passed the efi memmap, systab, etc., so we should use these data structures
719 * for initialization. Note, the efi init code path is determined by the
720 * global efi_enabled. This allows the same kernel image to be used on existing
721 * systems (with a traditional BIOS) as well as on EFI systems.
722 */
723/*
724 * setup_arch - architecture-specific boot-time initializations
725 *
726 * Note: On x86_64, fixmaps are ready for use even before this is called.
727 */
728
729void __init setup_arch(char **cmdline_p)
730{
731#ifdef CONFIG_X86_32
732 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
733
734 /*
735 * copy kernel address range established so far and switch
736 * to the proper swapper page table
737 */
738 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
739 initial_page_table + KERNEL_PGD_BOUNDARY,
740 KERNEL_PGD_PTRS);
741
742 load_cr3(swapper_pg_dir);
743 /*
744 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
745 * a cr3 based tlb flush, so the following __flush_tlb_all()
746 * will not flush anything because the CPU quirk which clears
747 * X86_FEATURE_PGE has not been invoked yet. Though due to the
748 * load_cr3() above the TLB has been flushed already. The
749 * quirk is invoked before subsequent calls to __flush_tlb_all()
750 * so proper operation is guaranteed.
751 */
752 __flush_tlb_all();
753#else
754 printk(KERN_INFO "Command line: %s\n", boot_command_line);
755 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
756#endif
757
758 /*
759 * If we have OLPC OFW, we might end up relocating the fixmap due to
760 * reserve_top(), so do this before touching the ioremap area.
761 */
762 olpc_ofw_detect();
763
764 idt_setup_early_traps();
765 early_cpu_init();
766 jump_label_init();
767 static_call_init();
768 early_ioremap_init();
769
770 setup_olpc_ofw_pgd();
771
772 ROOT_DEV = old_decode_dev(val: boot_params.hdr.root_dev);
773 screen_info = boot_params.screen_info;
774 edid_info = boot_params.edid_info;
775#ifdef CONFIG_X86_32
776 apm_info.bios = boot_params.apm_bios_info;
777 ist_info = boot_params.ist_info;
778#endif
779 saved_video_mode = boot_params.hdr.vid_mode;
780 bootloader_type = boot_params.hdr.type_of_loader;
781 if ((bootloader_type >> 4) == 0xe) {
782 bootloader_type &= 0xf;
783 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
784 }
785 bootloader_version = bootloader_type & 0xf;
786 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
787
788#ifdef CONFIG_BLK_DEV_RAM
789 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
790#endif
791#ifdef CONFIG_EFI
792 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
793 EFI32_LOADER_SIGNATURE, 4)) {
794 set_bit(EFI_BOOT, addr: &efi.flags);
795 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
796 EFI64_LOADER_SIGNATURE, 4)) {
797 set_bit(EFI_BOOT, addr: &efi.flags);
798 set_bit(EFI_64BIT, addr: &efi.flags);
799 }
800#endif
801
802 x86_init.oem.arch_setup();
803
804 /*
805 * Do some memory reservations *before* memory is added to memblock, so
806 * memblock allocations won't overwrite it.
807 *
808 * After this point, everything still needed from the boot loader or
809 * firmware or kernel text should be early reserved or marked not RAM in
810 * e820. All other memory is free game.
811 *
812 * This call needs to happen before e820__memory_setup() which calls the
813 * xen_memory_setup() on Xen dom0 which relies on the fact that those
814 * early reservations have happened already.
815 */
816 early_reserve_memory();
817
818 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
819 e820__memory_setup();
820 parse_setup_data();
821
822 copy_edd();
823
824 if (!boot_params.hdr.root_flags)
825 root_mountflags &= ~MS_RDONLY;
826 setup_initial_init_mm(start_code: _text, end_code: _etext, end_data: _edata, brk: (void *)_brk_end);
827
828 code_resource.start = __pa_symbol(_text);
829 code_resource.end = __pa_symbol(_etext)-1;
830 rodata_resource.start = __pa_symbol(__start_rodata);
831 rodata_resource.end = __pa_symbol(__end_rodata)-1;
832 data_resource.start = __pa_symbol(_sdata);
833 data_resource.end = __pa_symbol(_edata)-1;
834 bss_resource.start = __pa_symbol(__bss_start);
835 bss_resource.end = __pa_symbol(__bss_stop)-1;
836
837#ifdef CONFIG_CMDLINE_BOOL
838#ifdef CONFIG_CMDLINE_OVERRIDE
839 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
840#else
841 if (builtin_cmdline[0]) {
842 /* append boot loader cmdline to builtin */
843 strlcat(p: builtin_cmdline, q: " ", COMMAND_LINE_SIZE);
844 strlcat(p: builtin_cmdline, q: boot_command_line, COMMAND_LINE_SIZE);
845 strscpy(p: boot_command_line, q: builtin_cmdline, COMMAND_LINE_SIZE);
846 }
847#endif
848#endif
849
850 strscpy(p: command_line, q: boot_command_line, COMMAND_LINE_SIZE);
851 *cmdline_p = command_line;
852
853 /*
854 * x86_configure_nx() is called before parse_early_param() to detect
855 * whether hardware doesn't support NX (so that the early EHCI debug
856 * console setup can safely call set_fixmap()).
857 */
858 x86_configure_nx();
859
860 parse_early_param();
861
862 if (efi_enabled(EFI_BOOT))
863 efi_memblock_x86_reserve_range();
864
865#ifdef CONFIG_MEMORY_HOTPLUG
866 /*
867 * Memory used by the kernel cannot be hot-removed because Linux
868 * cannot migrate the kernel pages. When memory hotplug is
869 * enabled, we should prevent memblock from allocating memory
870 * for the kernel.
871 *
872 * ACPI SRAT records all hotpluggable memory ranges. But before
873 * SRAT is parsed, we don't know about it.
874 *
875 * The kernel image is loaded into memory at very early time. We
876 * cannot prevent this anyway. So on NUMA system, we set any
877 * node the kernel resides in as un-hotpluggable.
878 *
879 * Since on modern servers, one node could have double-digit
880 * gigabytes memory, we can assume the memory around the kernel
881 * image is also un-hotpluggable. So before SRAT is parsed, just
882 * allocate memory near the kernel image to try the best to keep
883 * the kernel away from hotpluggable memory.
884 */
885 if (movable_node_is_enabled())
886 memblock_set_bottom_up(enable: true);
887#endif
888
889 x86_report_nx();
890
891 apic_setup_apic_calls();
892
893 if (acpi_mps_check()) {
894#ifdef CONFIG_X86_LOCAL_APIC
895 apic_is_disabled = true;
896#endif
897 setup_clear_cpu_cap(X86_FEATURE_APIC);
898 }
899
900 e820__reserve_setup_data();
901 e820__finish_early_params();
902
903 if (efi_enabled(EFI_BOOT))
904 efi_init();
905
906 reserve_ibft_region();
907 dmi_setup();
908
909 /*
910 * VMware detection requires dmi to be available, so this
911 * needs to be done after dmi_setup(), for the boot CPU.
912 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
913 * called before cache_bp_init() for setting up MTRR state.
914 */
915 init_hypervisor_platform();
916
917 tsc_early_init();
918 x86_init.resources.probe_roms();
919
920 /* after parse_early_param, so could debug it */
921 insert_resource(parent: &iomem_resource, new: &code_resource);
922 insert_resource(parent: &iomem_resource, new: &rodata_resource);
923 insert_resource(parent: &iomem_resource, new: &data_resource);
924 insert_resource(parent: &iomem_resource, new: &bss_resource);
925
926 e820_add_kernel_range();
927 trim_bios_range();
928#ifdef CONFIG_X86_32
929 if (ppro_with_ram_bug()) {
930 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
931 E820_TYPE_RESERVED);
932 e820__update_table(e820_table);
933 printk(KERN_INFO "fixed physical RAM map:\n");
934 e820__print_table("bad_ppro");
935 }
936#else
937 early_gart_iommu_check();
938#endif
939
940 /*
941 * partially used pages are not usable - thus
942 * we are rounding upwards:
943 */
944 max_pfn = e820__end_of_ram_pfn();
945
946 /* update e820 for memory not covered by WB MTRRs */
947 cache_bp_init();
948 if (mtrr_trim_uncached_memory(end_pfn: max_pfn))
949 max_pfn = e820__end_of_ram_pfn();
950
951 max_possible_pfn = max_pfn;
952
953 /*
954 * Define random base addresses for memory sections after max_pfn is
955 * defined and before each memory section base is used.
956 */
957 kernel_randomize_memory();
958
959#ifdef CONFIG_X86_32
960 /* max_low_pfn get updated here */
961 find_low_pfn_range();
962#else
963 check_x2apic();
964
965 /* How many end-of-memory variables you have, grandma! */
966 /* need this before calling reserve_initrd */
967 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
968 max_low_pfn = e820__end_of_low_ram_pfn();
969 else
970 max_low_pfn = max_pfn;
971
972 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
973#endif
974
975 /*
976 * Find and reserve possible boot-time SMP configuration:
977 */
978 find_smp_config();
979
980 early_alloc_pgt_buf();
981
982 /*
983 * Need to conclude brk, before e820__memblock_setup()
984 * it could use memblock_find_in_range, could overlap with
985 * brk area.
986 */
987 reserve_brk();
988
989 cleanup_highmap();
990
991 memblock_set_current_limit(ISA_END_ADDRESS);
992 e820__memblock_setup();
993
994 /*
995 * Needs to run after memblock setup because it needs the physical
996 * memory size.
997 */
998 mem_encrypt_setup_arch();
999
1000 efi_fake_memmap();
1001 efi_find_mirror();
1002 efi_esrt_init();
1003 efi_mokvar_table_init();
1004
1005 /*
1006 * The EFI specification says that boot service code won't be
1007 * called after ExitBootServices(). This is, in fact, a lie.
1008 */
1009 efi_reserve_boot_services();
1010
1011 /* preallocate 4k for mptable mpc */
1012 e820__memblock_alloc_reserved_mpc_new();
1013
1014#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1015 setup_bios_corruption_check();
1016#endif
1017
1018#ifdef CONFIG_X86_32
1019 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1020 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1021#endif
1022
1023 /*
1024 * Find free memory for the real mode trampoline and place it there. If
1025 * there is not enough free memory under 1M, on EFI-enabled systems
1026 * there will be additional attempt to reclaim the memory for the real
1027 * mode trampoline at efi_free_boot_services().
1028 *
1029 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1030 * are known to corrupt low memory and several hundred kilobytes are not
1031 * worth complex detection what memory gets clobbered. Windows does the
1032 * same thing for very similar reasons.
1033 *
1034 * Moreover, on machines with SandyBridge graphics or in setups that use
1035 * crashkernel the entire 1M is reserved anyway.
1036 */
1037 x86_platform.realmode_reserve();
1038
1039 init_mem_mapping();
1040
1041 idt_setup_early_pf();
1042
1043 /*
1044 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1045 * with the current CR4 value. This may not be necessary, but
1046 * auditing all the early-boot CR4 manipulation would be needed to
1047 * rule it out.
1048 *
1049 * Mask off features that don't work outside long mode (just
1050 * PCIDE for now).
1051 */
1052 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1053
1054 memblock_set_current_limit(limit: get_max_mapped());
1055
1056 /*
1057 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1058 */
1059
1060#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1061 if (init_ohci1394_dma_early)
1062 init_ohci1394_dma_on_all_controllers();
1063#endif
1064 /* Allocate bigger log buffer */
1065 setup_log_buf(1);
1066
1067 if (efi_enabled(EFI_BOOT)) {
1068 switch (boot_params.secure_boot) {
1069 case efi_secureboot_mode_disabled:
1070 pr_info("Secure boot disabled\n");
1071 break;
1072 case efi_secureboot_mode_enabled:
1073 pr_info("Secure boot enabled\n");
1074 break;
1075 default:
1076 pr_info("Secure boot could not be determined\n");
1077 break;
1078 }
1079 }
1080
1081 reserve_initrd();
1082
1083 acpi_table_upgrade();
1084 /* Look for ACPI tables and reserve memory occupied by them. */
1085 acpi_boot_table_init();
1086
1087 vsmp_init();
1088
1089 io_delay_init();
1090
1091 early_platform_quirks();
1092
1093 early_acpi_boot_init();
1094
1095 x86_flattree_get_config();
1096
1097 initmem_init();
1098 dma_contiguous_reserve(addr_limit: max_pfn_mapped << PAGE_SHIFT);
1099
1100 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1101 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1102
1103 /*
1104 * Reserve memory for crash kernel after SRAT is parsed so that it
1105 * won't consume hotpluggable memory.
1106 */
1107 arch_reserve_crashkernel();
1108
1109 memblock_find_dma_reserve();
1110
1111 if (!early_xdbc_setup_hardware())
1112 early_xdbc_register_console();
1113
1114 x86_init.paging.pagetable_init();
1115
1116 kasan_init();
1117
1118 /*
1119 * Sync back kernel address range.
1120 *
1121 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1122 * this call?
1123 */
1124 sync_initial_page_table();
1125
1126 tboot_probe();
1127
1128 map_vsyscall();
1129
1130 x86_32_probe_apic();
1131
1132 early_quirks();
1133
1134 /*
1135 * Read APIC and some other early information from ACPI tables.
1136 */
1137 acpi_boot_init();
1138 x86_dtb_init();
1139
1140 /*
1141 * get boot-time SMP configuration:
1142 */
1143 get_smp_config();
1144
1145 /*
1146 * Systems w/o ACPI and mptables might not have it mapped the local
1147 * APIC yet, but prefill_possible_map() might need to access it.
1148 */
1149 init_apic_mappings();
1150
1151 prefill_possible_map();
1152
1153 init_cpu_to_node();
1154 init_gi_nodes();
1155
1156 io_apic_init_mappings();
1157
1158 x86_init.hyper.guest_late_init();
1159
1160 e820__reserve_resources();
1161 e820__register_nosave_regions(limit_pfn: max_pfn);
1162
1163 x86_init.resources.reserve_resources();
1164
1165 e820__setup_pci_gap();
1166
1167#ifdef CONFIG_VT
1168#if defined(CONFIG_VGA_CONSOLE)
1169 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(phys_addr: 0xa0000) != EFI_CONVENTIONAL_MEMORY))
1170 vgacon_register_screen(si: &screen_info);
1171#endif
1172#endif
1173 x86_init.oem.banner();
1174
1175 x86_init.timers.wallclock_init();
1176
1177 /*
1178 * This needs to run before setup_local_APIC() which soft-disables the
1179 * local APIC temporarily and that masks the thermal LVT interrupt,
1180 * leading to softlockups on machines which have configured SMI
1181 * interrupt delivery.
1182 */
1183 therm_lvt_init();
1184
1185 mcheck_init();
1186
1187 register_refined_jiffies(CLOCK_TICK_RATE);
1188
1189#ifdef CONFIG_EFI
1190 if (efi_enabled(EFI_BOOT))
1191 efi_apply_memmap_quirks();
1192#endif
1193
1194 unwind_init();
1195}
1196
1197#ifdef CONFIG_X86_32
1198
1199static struct resource video_ram_resource = {
1200 .name = "Video RAM area",
1201 .start = 0xa0000,
1202 .end = 0xbffff,
1203 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1204};
1205
1206void __init i386_reserve_resources(void)
1207{
1208 request_resource(&iomem_resource, &video_ram_resource);
1209 reserve_standard_io_resources();
1210}
1211
1212#endif /* CONFIG_X86_32 */
1213
1214static struct notifier_block kernel_offset_notifier = {
1215 .notifier_call = dump_kernel_offset
1216};
1217
1218static int __init register_kernel_offset_dumper(void)
1219{
1220 atomic_notifier_chain_register(nh: &panic_notifier_list,
1221 nb: &kernel_offset_notifier);
1222 return 0;
1223}
1224__initcall(register_kernel_offset_dumper);
1225

source code of linux/arch/x86/kernel/setup.c