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 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
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(size: area_size, PAGE_SIZE, start: 0,
263 PFN_PHYS(max_pfn_mapped));
264 if (!relocated_ramdisk)
265 panic(fmt: "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(dest: (void *)initrd_start, src: ramdisk_image, size: 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(base: ramdisk_image, size: 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(base: ramdisk_image, size: 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: phys_addr + sizeof(struct setup_data), size: 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(base: data->addr, size: data->size);
346 ima_kexec_buffer_phys = data->addr;
347 ima_kexec_buffer_size = data->size;
348 }
349
350 early_memunmap(addr: data, size: 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(phys_addr: pa_data, size: sizeof(*data));
393 data_len = data->len + sizeof(struct setup_data);
394 data_type = data->type;
395 pa_next = data->next;
396 early_memunmap(addr: data, size: sizeof(*data));
397
398 switch (data_type) {
399 case SETUP_E820_EXT:
400 e820__memory_setup_extended(phys_addr: pa_data, data_len);
401 break;
402 case SETUP_DTB:
403 add_dtb(data: pa_data);
404 break;
405 case SETUP_EFI:
406 parse_efi_setup(phys_addr: pa_data, data_len);
407 break;
408 case SETUP_IMA:
409 add_early_ima_buffer(phys_addr: pa_data);
410 break;
411 case SETUP_RNG_SEED:
412 data = early_memremap(phys_addr: pa_data, size: data_len);
413 add_bootloader_randomness(buf: data->data, len: data->len);
414 /* Zero seed for forward secrecy. */
415 memzero_explicit(s: data->data, count: data->len);
416 /* Zero length in case we find ourselves back here by accident. */
417 memzero_explicit(s: &data->len, count: sizeof(data->len));
418 early_memunmap(addr: data, size: 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(phys_addr: pa_data, size: 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(base: pa_data, size: sizeof(*data) + data->len);
446
447 if (data->type == SETUP_INDIRECT) {
448 len += data->len;
449 early_memunmap(addr: data, size: sizeof(*data));
450 data = early_memremap(phys_addr: pa_data, size: 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(base: indirect->addr, size: indirect->len);
460 }
461
462 pa_data = pa_next;
463 early_memunmap(addr: data, size: 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, system_ram: memblock_phys_mem_size(),
478 crash_size: &crash_size, crash_base: &crash_base,
479 low_size: &low_size, high: &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 crash_low_size: 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(root: &ioport_resource, new: &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(bus: 0, slot: 2, func: 0, PCI_VENDOR_ID);
545 if (vendor != 0x8086)
546 return false;
547
548 devid = read_pci_config_16(bus: 0, slot: 2, func: 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(base: 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(start: 0, PAGE_SIZE, old_type: E820_TYPE_RAM, new_type: 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, old_type: E820_TYPE_RAM, check_type: 1);
614
615 e820__update_table(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, end: start + size, type: 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, old_type: E820_TYPE_RAM, check_type: 0);
636 e820__range_add(start, size, type: 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 size: (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(base: 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(val: 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, addr: &efi.flags);
793 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
794 EFI64_LOADER_SIGNATURE, 4)) {
795 set_bit(EFI_BOOT, addr: &efi.flags);
796 set_bit(EFI_64BIT, addr: &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(start_code: _text, end_code: _etext, end_data: _edata, brk: (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(p: builtin_cmdline, q: " ", COMMAND_LINE_SIZE);
842 strlcat(p: builtin_cmdline, q: 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(enable: 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(parent: &iomem_resource, new: &code_resource);
920 insert_resource(parent: &iomem_resource, new: &rodata_resource);
921 insert_resource(parent: &iomem_resource, new: &data_resource);
922 insert_resource(parent: &iomem_resource, new: &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(end_pfn: 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(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(addr_limit: 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(limit_pfn: 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(phys_addr: 0xa0000) != EFI_CONVENTIONAL_MEMORY))
1166 vgacon_register_screen(si: &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(nh: &panic_notifier_list,
1217 nb: &kernel_offset_notifier);
1218 return 0;
1219}
1220__initcall(register_kernel_offset_dumper);
1221

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