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/cpu.h>
11 #include <linux/crash_dump.h>
12 #include <linux/dma-map-ops.h>
13 #include <linux/efi.h>
14 #include <linux/hugetlb.h>
15 #include <linux/ima.h>
16 #include <linux/init_ohci1394_dma.h>
17 #include <linux/initrd.h>
18 #include <linux/iscsi_ibft.h>
19 #include <linux/memblock.h>
20 #include <linux/panic_notifier.h>
21 #include <linux/pci.h>
22 #include <linux/random.h>
23 #include <linux/root_dev.h>
24 #include <linux/static_call.h>
25 #include <linux/sysfb.h>
26 #include <linux/swiotlb.h>
27 #include <linux/tboot.h>
28 #include <linux/usb/xhci-dbgp.h>
29 #include <linux/vmalloc.h>
30
31 #include <uapi/linux/mount.h>
32
33 #include <xen/xen.h>
34
35 #include <asm/apic.h>
36 #include <asm/bios_ebda.h>
37 #include <asm/bugs.h>
38 #include <asm/cacheinfo.h>
39 #include <asm/coco.h>
40 #include <asm/cpu.h>
41 #include <asm/efi.h>
42 #include <asm/gart.h>
43 #include <asm/hypervisor.h>
44 #include <asm/io_apic.h>
45 #include <asm/kasan.h>
46 #include <asm/kaslr.h>
47 #include <asm/mce.h>
48 #include <asm/memtype.h>
49 #include <asm/mtrr.h>
50 #include <asm/nmi.h>
51 #include <asm/numa.h>
52 #include <asm/olpc_ofw.h>
53 #include <asm/pci-direct.h>
54 #include <asm/prom.h>
55 #include <asm/proto.h>
56 #include <asm/realmode.h>
57 #include <asm/thermal.h>
58 #include <asm/unwind.h>
59 #include <asm/vsyscall.h>
60
61 /*
62 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
63 * max_pfn_mapped: highest directly mapped pfn > 4 GB
64 *
65 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
66 * represented by pfn_mapped[].
67 */
68 unsigned long max_low_pfn_mapped;
69 unsigned long max_pfn_mapped;
70
71 #ifdef CONFIG_DMI
72 RESERVE_BRK(dmi_alloc, 65536);
73 #endif
74
75
76 unsigned long _brk_start = (unsigned long)__brk_base;
77 unsigned long _brk_end = (unsigned long)__brk_base;
78
79 struct boot_params boot_params;
80
81 /*
82 * These are the four main kernel memory regions, we put them into
83 * the resource tree so that kdump tools and other debugging tools
84 * recover it:
85 */
86
87 static struct resource rodata_resource = {
88 .name = "Kernel rodata",
89 .start = 0,
90 .end = 0,
91 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
92 };
93
94 static struct resource data_resource = {
95 .name = "Kernel data",
96 .start = 0,
97 .end = 0,
98 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
99 };
100
101 static struct resource code_resource = {
102 .name = "Kernel code",
103 .start = 0,
104 .end = 0,
105 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
106 };
107
108 static struct resource bss_resource = {
109 .name = "Kernel bss",
110 .start = 0,
111 .end = 0,
112 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
113 };
114
115
116 #ifdef CONFIG_X86_32
117 /* CPU data as detected by the assembly code in head_32.S */
118 struct cpuinfo_x86 new_cpu_data;
119
120 struct apm_info apm_info;
121 EXPORT_SYMBOL(apm_info);
122
123 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
124 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
125 struct ist_info ist_info;
126 EXPORT_SYMBOL(ist_info);
127 #else
128 struct ist_info ist_info;
129 #endif
130
131 #endif
132
133 struct cpuinfo_x86 boot_cpu_data __read_mostly;
134 EXPORT_SYMBOL(boot_cpu_data);
135 SYM_PIC_ALIAS(boot_cpu_data);
136
137 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
138 __visible unsigned long mmu_cr4_features __ro_after_init;
139 #else
140 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
141 #endif
142
143 #ifdef CONFIG_IMA
144 static phys_addr_t ima_kexec_buffer_phys;
145 static size_t ima_kexec_buffer_size;
146 #endif
147
148 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
149 int bootloader_type, bootloader_version;
150
151 static const struct ctl_table x86_sysctl_table[] = {
152 {
153 .procname = "unknown_nmi_panic",
154 .data = &unknown_nmi_panic,
155 .maxlen = sizeof(int),
156 .mode = 0644,
157 .proc_handler = proc_dointvec,
158 },
159 {
160 .procname = "panic_on_unrecovered_nmi",
161 .data = &panic_on_unrecovered_nmi,
162 .maxlen = sizeof(int),
163 .mode = 0644,
164 .proc_handler = proc_dointvec,
165 },
166 {
167 .procname = "panic_on_io_nmi",
168 .data = &panic_on_io_nmi,
169 .maxlen = sizeof(int),
170 .mode = 0644,
171 .proc_handler = proc_dointvec,
172 },
173 {
174 .procname = "bootloader_type",
175 .data = &bootloader_type,
176 .maxlen = sizeof(int),
177 .mode = 0444,
178 .proc_handler = proc_dointvec,
179 },
180 {
181 .procname = "bootloader_version",
182 .data = &bootloader_version,
183 .maxlen = sizeof(int),
184 .mode = 0444,
185 .proc_handler = proc_dointvec,
186 },
187 {
188 .procname = "io_delay_type",
189 .data = &io_delay_type,
190 .maxlen = sizeof(int),
191 .mode = 0644,
192 .proc_handler = proc_dointvec,
193 },
194 #if defined(CONFIG_ACPI_SLEEP)
195 {
196 .procname = "acpi_video_flags",
197 .data = &acpi_realmode_flags,
198 .maxlen = sizeof(unsigned long),
199 .mode = 0644,
200 .proc_handler = proc_doulongvec_minmax,
201 },
202 #endif
203 };
204
init_x86_sysctl(void)205 static int __init init_x86_sysctl(void)
206 {
207 register_sysctl_init("kernel", x86_sysctl_table);
208 return 0;
209 }
210 arch_initcall(init_x86_sysctl);
211
212 /*
213 * Setup options
214 */
215
216 struct sysfb_display_info sysfb_primary_display;
217 EXPORT_SYMBOL(sysfb_primary_display);
218
219 extern int root_mountflags;
220
221 unsigned long saved_video_mode;
222
223 #define RAMDISK_IMAGE_START_MASK 0x07FF
224 #define RAMDISK_PROMPT_FLAG 0x8000
225 #define RAMDISK_LOAD_FLAG 0x4000
226
227 static char __initdata command_line[COMMAND_LINE_SIZE];
228 #ifdef CONFIG_CMDLINE_BOOL
229 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
230 bool builtin_cmdline_added __ro_after_init;
231 #endif
232
233 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
234 struct edd edd;
235 #ifdef CONFIG_EDD_MODULE
236 EXPORT_SYMBOL(edd);
237 #endif
238 /**
239 * copy_edd() - Copy the BIOS EDD information
240 * from boot_params into a safe place.
241 *
242 */
copy_edd(void)243 static inline void __init copy_edd(void)
244 {
245 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
246 sizeof(edd.mbr_signature));
247 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
248 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
249 edd.edd_info_nr = boot_params.eddbuf_entries;
250 }
251 #else
copy_edd(void)252 static inline void __init copy_edd(void)
253 {
254 }
255 #endif
256
extend_brk(size_t size,size_t align)257 void * __init extend_brk(size_t size, size_t align)
258 {
259 size_t mask = align - 1;
260 void *ret;
261
262 BUG_ON(_brk_start == 0);
263 BUG_ON(align & mask);
264
265 _brk_end = (_brk_end + mask) & ~mask;
266 BUG_ON((char *)(_brk_end + size) > __brk_limit);
267
268 ret = (void *)_brk_end;
269 _brk_end += size;
270
271 memset(ret, 0, size);
272
273 return ret;
274 }
275
276 #ifdef CONFIG_X86_32
cleanup_highmap(void)277 static void __init cleanup_highmap(void)
278 {
279 }
280 #endif
281
reserve_brk(void)282 static void __init reserve_brk(void)
283 {
284 if (_brk_end > _brk_start)
285 memblock_reserve_kern(__pa_symbol(_brk_start),
286 _brk_end - _brk_start);
287
288 /* Mark brk area as locked down and no longer taking any
289 new allocations */
290 _brk_start = 0;
291 }
292
293 #ifdef CONFIG_BLK_DEV_INITRD
294
get_ramdisk_image(void)295 static u64 __init get_ramdisk_image(void)
296 {
297 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
298
299 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
300
301 if (ramdisk_image == 0)
302 ramdisk_image = phys_initrd_start;
303
304 return ramdisk_image;
305 }
get_ramdisk_size(void)306 static u64 __init get_ramdisk_size(void)
307 {
308 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
309
310 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
311
312 if (ramdisk_size == 0)
313 ramdisk_size = phys_initrd_size;
314
315 return ramdisk_size;
316 }
317
relocate_initrd(void)318 static void __init relocate_initrd(void)
319 {
320 /* Assume only end is not page aligned */
321 u64 ramdisk_image = get_ramdisk_image();
322 u64 ramdisk_size = get_ramdisk_size();
323 u64 area_size = PAGE_ALIGN(ramdisk_size);
324 int ret = 0;
325
326 /* We need to move the initrd down into directly mapped mem */
327 u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
328 PFN_PHYS(max_pfn_mapped));
329 if (!relocated_ramdisk)
330 panic("Cannot find place for new RAMDISK of size %lld\n",
331 ramdisk_size);
332
333 initrd_start = relocated_ramdisk + PAGE_OFFSET;
334 initrd_end = initrd_start + ramdisk_size;
335 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
336 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
337
338 ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
339 if (ret)
340 panic("Copy RAMDISK failed\n");
341
342 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
343 " [mem %#010llx-%#010llx]\n",
344 ramdisk_image, ramdisk_image + ramdisk_size - 1,
345 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
346 }
347
early_reserve_initrd(void)348 static void __init early_reserve_initrd(void)
349 {
350 /* Assume only end is not page aligned */
351 u64 ramdisk_image = get_ramdisk_image();
352 u64 ramdisk_size = get_ramdisk_size();
353 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
354
355 if (!boot_params.hdr.type_of_loader ||
356 !ramdisk_image || !ramdisk_size)
357 return; /* No initrd provided by bootloader */
358
359 memblock_reserve_kern(ramdisk_image, ramdisk_end - ramdisk_image);
360 }
361
reserve_initrd(void)362 static void __init reserve_initrd(void)
363 {
364 /* Assume only end is not page aligned */
365 u64 ramdisk_image = get_ramdisk_image();
366 u64 ramdisk_size = get_ramdisk_size();
367 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
368
369 if (!boot_params.hdr.type_of_loader ||
370 !ramdisk_image || !ramdisk_size)
371 return; /* No initrd provided by bootloader */
372
373 initrd_start = 0;
374
375 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
376 ramdisk_end - 1);
377
378 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
379 PFN_DOWN(ramdisk_end))) {
380 /* All are mapped, easy case */
381 initrd_start = ramdisk_image + PAGE_OFFSET;
382 initrd_end = initrd_start + ramdisk_size;
383 return;
384 }
385
386 relocate_initrd();
387
388 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
389 }
390
391 #else
early_reserve_initrd(void)392 static void __init early_reserve_initrd(void)
393 {
394 }
reserve_initrd(void)395 static void __init reserve_initrd(void)
396 {
397 }
398 #endif /* CONFIG_BLK_DEV_INITRD */
399
add_early_ima_buffer(u64 phys_addr)400 static void __init add_early_ima_buffer(u64 phys_addr)
401 {
402 #ifdef CONFIG_IMA
403 struct ima_setup_data *data;
404
405 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
406 if (!data) {
407 pr_warn("setup: failed to memremap ima_setup_data entry\n");
408 return;
409 }
410
411 if (data->size) {
412 memblock_reserve_kern(data->addr, data->size);
413 ima_kexec_buffer_phys = data->addr;
414 ima_kexec_buffer_size = data->size;
415 }
416
417 early_memunmap(data, sizeof(*data));
418 #else
419 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
420 #endif
421 }
422
423 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
ima_free_kexec_buffer(void)424 int __init ima_free_kexec_buffer(void)
425 {
426 if (!ima_kexec_buffer_size)
427 return -ENOENT;
428
429 memblock_free_late(ima_kexec_buffer_phys,
430 ima_kexec_buffer_size);
431
432 ima_kexec_buffer_phys = 0;
433 ima_kexec_buffer_size = 0;
434
435 return 0;
436 }
437
ima_get_kexec_buffer(void ** addr,size_t * size)438 int __init ima_get_kexec_buffer(void **addr, size_t *size)
439 {
440 int ret;
441
442 if (!ima_kexec_buffer_size)
443 return -ENOENT;
444
445 ret = ima_validate_range(ima_kexec_buffer_phys, ima_kexec_buffer_size);
446 if (ret)
447 return ret;
448
449 *addr = __va(ima_kexec_buffer_phys);
450 *size = ima_kexec_buffer_size;
451
452 return 0;
453 }
454 #endif
455
add_kho(u64 phys_addr,u32 data_len)456 static void __init add_kho(u64 phys_addr, u32 data_len)
457 {
458 struct kho_data *kho;
459 u64 addr = phys_addr + sizeof(struct setup_data);
460 u64 size = data_len - sizeof(struct setup_data);
461
462 if (!IS_ENABLED(CONFIG_KEXEC_HANDOVER)) {
463 pr_warn("Passed KHO data, but CONFIG_KEXEC_HANDOVER not set. Ignoring.\n");
464 return;
465 }
466
467 kho = early_memremap(addr, size);
468 if (!kho) {
469 pr_warn("setup: failed to memremap kho data (0x%llx, 0x%llx)\n",
470 addr, size);
471 return;
472 }
473
474 kho_populate(kho->fdt_addr, kho->fdt_size, kho->scratch_addr, kho->scratch_size);
475
476 early_memunmap(kho, size);
477 }
478
parse_setup_data(void)479 static void __init parse_setup_data(void)
480 {
481 struct setup_data *data;
482 u64 pa_data, pa_next;
483
484 pa_data = boot_params.hdr.setup_data;
485 while (pa_data) {
486 u32 data_len, data_type;
487
488 data = early_memremap(pa_data, sizeof(*data));
489 data_len = data->len + sizeof(struct setup_data);
490 data_type = data->type;
491 pa_next = data->next;
492 early_memunmap(data, sizeof(*data));
493
494 switch (data_type) {
495 case SETUP_E820_EXT:
496 e820__memory_setup_extended(pa_data, data_len);
497 break;
498 case SETUP_DTB:
499 add_dtb(pa_data);
500 break;
501 case SETUP_EFI:
502 parse_efi_setup(pa_data, data_len);
503 break;
504 case SETUP_IMA:
505 add_early_ima_buffer(pa_data);
506 break;
507 case SETUP_KEXEC_KHO:
508 add_kho(pa_data, data_len);
509 break;
510 case SETUP_RNG_SEED:
511 data = early_memremap(pa_data, data_len);
512 add_bootloader_randomness(data->data, data->len);
513 /* Zero seed for forward secrecy. */
514 memzero_explicit(data->data, data->len);
515 /* Zero length in case we find ourselves back here by accident. */
516 memzero_explicit(&data->len, sizeof(data->len));
517 early_memunmap(data, data_len);
518 break;
519 default:
520 break;
521 }
522 pa_data = pa_next;
523 }
524 }
525
526 /*
527 * Translate the fields of 'struct boot_param' into global variables
528 * representing these parameters.
529 */
parse_boot_params(void)530 static void __init parse_boot_params(void)
531 {
532 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
533 sysfb_primary_display.screen = boot_params.screen_info;
534 #if defined(CONFIG_FIRMWARE_EDID)
535 sysfb_primary_display.edid = boot_params.edid_info;
536 #endif
537 #ifdef CONFIG_X86_32
538 apm_info.bios = boot_params.apm_bios_info;
539 ist_info = boot_params.ist_info;
540 #endif
541 saved_video_mode = boot_params.hdr.vid_mode;
542 bootloader_type = boot_params.hdr.type_of_loader;
543 if ((bootloader_type >> 4) == 0xe) {
544 bootloader_type &= 0xf;
545 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
546 }
547 bootloader_version = bootloader_type & 0xf;
548 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
549
550 #ifdef CONFIG_BLK_DEV_RAM
551 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
552 #endif
553 #ifdef CONFIG_EFI
554 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
555 EFI32_LOADER_SIGNATURE, 4)) {
556 set_bit(EFI_BOOT, &efi.flags);
557 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
558 EFI64_LOADER_SIGNATURE, 4)) {
559 set_bit(EFI_BOOT, &efi.flags);
560 set_bit(EFI_64BIT, &efi.flags);
561 }
562 #endif
563
564 if (!boot_params.hdr.root_flags)
565 root_mountflags &= ~MS_RDONLY;
566 }
567
memblock_x86_reserve_range_setup_data(void)568 static void __init memblock_x86_reserve_range_setup_data(void)
569 {
570 struct setup_indirect *indirect;
571 struct setup_data *data;
572 u64 pa_data, pa_next;
573 u32 len;
574
575 pa_data = boot_params.hdr.setup_data;
576 while (pa_data) {
577 data = early_memremap(pa_data, sizeof(*data));
578 if (!data) {
579 pr_warn("setup: failed to memremap setup_data entry\n");
580 return;
581 }
582
583 len = sizeof(*data);
584 pa_next = data->next;
585
586 memblock_reserve_kern(pa_data, sizeof(*data) + data->len);
587
588 if (data->type == SETUP_INDIRECT) {
589 len += data->len;
590 early_memunmap(data, sizeof(*data));
591 data = early_memremap(pa_data, len);
592 if (!data) {
593 pr_warn("setup: failed to memremap indirect setup_data\n");
594 return;
595 }
596
597 indirect = (struct setup_indirect *)data->data;
598
599 if (indirect->type != SETUP_INDIRECT)
600 memblock_reserve_kern(indirect->addr, indirect->len);
601 }
602
603 pa_data = pa_next;
604 early_memunmap(data, len);
605 }
606 }
607
arch_reserve_crashkernel(void)608 static void __init arch_reserve_crashkernel(void)
609 {
610 unsigned long long crash_base, crash_size, low_size = 0, cma_size = 0;
611 bool high = false;
612 int ret;
613
614 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
615 return;
616
617 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
618 &crash_size, &crash_base,
619 &low_size, &cma_size, &high);
620 if (ret)
621 return;
622
623 if (xen_pv_domain()) {
624 pr_info("Ignoring crashkernel for a Xen PV domain\n");
625 return;
626 }
627
628 reserve_crashkernel_generic(crash_size, crash_base, low_size, high);
629 reserve_crashkernel_cma(cma_size);
630 }
631
632 static struct resource standard_io_resources[] = {
633 { .name = "dma1", .start = 0x00, .end = 0x1f,
634 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
635 { .name = "pic1", .start = 0x20, .end = 0x21,
636 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
637 { .name = "timer0", .start = 0x40, .end = 0x43,
638 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
639 { .name = "timer1", .start = 0x50, .end = 0x53,
640 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
641 { .name = "keyboard", .start = 0x60, .end = 0x60,
642 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
643 { .name = "keyboard", .start = 0x64, .end = 0x64,
644 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
645 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
646 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
647 { .name = "pic2", .start = 0xa0, .end = 0xa1,
648 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
649 { .name = "dma2", .start = 0xc0, .end = 0xdf,
650 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
651 { .name = "fpu", .start = 0xf0, .end = 0xff,
652 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
653 };
654
reserve_standard_io_resources(void)655 void __init reserve_standard_io_resources(void)
656 {
657 int i;
658
659 /* request I/O space for devices used on all i[345]86 PCs */
660 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
661 request_resource(&ioport_resource, &standard_io_resources[i]);
662
663 }
664
setup_kernel_resources(void)665 static void __init setup_kernel_resources(void)
666 {
667 code_resource.start = __pa_symbol(_text);
668 code_resource.end = __pa_symbol(_etext)-1;
669 rodata_resource.start = __pa_symbol(__start_rodata);
670 rodata_resource.end = __pa_symbol(__end_rodata)-1;
671 data_resource.start = __pa_symbol(_sdata);
672 data_resource.end = __pa_symbol(_edata)-1;
673 bss_resource.start = __pa_symbol(__bss_start);
674 bss_resource.end = __pa_symbol(__bss_stop)-1;
675
676 insert_resource(&iomem_resource, &code_resource);
677 insert_resource(&iomem_resource, &rodata_resource);
678 insert_resource(&iomem_resource, &data_resource);
679 insert_resource(&iomem_resource, &bss_resource);
680 }
681
snb_gfx_workaround_needed(void)682 static bool __init snb_gfx_workaround_needed(void)
683 {
684 #ifdef CONFIG_PCI
685 int i;
686 u16 vendor, devid;
687 static const __initconst u16 snb_ids[] = {
688 0x0102,
689 0x0112,
690 0x0122,
691 0x0106,
692 0x0116,
693 0x0126,
694 0x010a,
695 };
696
697 /* Assume no if something weird is going on with PCI */
698 if (!early_pci_allowed())
699 return false;
700
701 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
702 if (vendor != 0x8086)
703 return false;
704
705 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
706 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
707 if (devid == snb_ids[i])
708 return true;
709 #endif
710
711 return false;
712 }
713
714 /*
715 * Sandy Bridge graphics has trouble with certain ranges, exclude
716 * them from allocation.
717 */
trim_snb_memory(void)718 static void __init trim_snb_memory(void)
719 {
720 static const __initconst unsigned long bad_pages[] = {
721 0x20050000,
722 0x20110000,
723 0x20130000,
724 0x20138000,
725 0x40004000,
726 };
727 int i;
728
729 if (!snb_gfx_workaround_needed())
730 return;
731
732 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
733
734 /*
735 * SandyBridge integrated graphics devices have a bug that prevents
736 * them from accessing certain memory ranges, namely anything below
737 * 1M and in the pages listed in bad_pages[] above.
738 *
739 * To avoid these pages being ever accessed by SNB gfx devices reserve
740 * bad_pages that have not already been reserved at boot time.
741 * All memory below the 1 MB mark is anyway reserved later during
742 * setup_arch(), so there is no need to reserve it here.
743 */
744
745 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
746 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
747 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
748 bad_pages[i]);
749 }
750 }
751
trim_bios_range(void)752 static void __init trim_bios_range(void)
753 {
754 /*
755 * A special case is the first 4Kb of memory;
756 * This is a BIOS owned area, not kernel ram, but generally
757 * not listed as such in the E820 table.
758 *
759 * This typically reserves additional memory (64KiB by default)
760 * since some BIOSes are known to corrupt low memory. See the
761 * Kconfig help text for X86_RESERVE_LOW.
762 */
763 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
764
765 /*
766 * special case: Some BIOSes report the PC BIOS
767 * area (640Kb -> 1Mb) as RAM even though it is not.
768 * take them out.
769 */
770 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM);
771
772 e820__update_table(e820_table);
773 }
774
775 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)776 static void __init e820_add_kernel_range(void)
777 {
778 u64 start = __pa_symbol(_text);
779 u64 size = __pa_symbol(_end) - start;
780
781 /*
782 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
783 * attempt to fix it by adding the range. We may have a confused BIOS,
784 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
785 * exclude kernel range. If we really are running on top non-RAM,
786 * we will crash later anyways.
787 */
788 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
789 return;
790
791 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
792 e820__range_remove(start, size, 0);
793 e820__range_add(start, size, E820_TYPE_RAM);
794 }
795
early_reserve_memory(void)796 static void __init early_reserve_memory(void)
797 {
798 /*
799 * Reserve the memory occupied by the kernel between _text and
800 * __end_of_kernel_reserve symbols. Any kernel sections after the
801 * __end_of_kernel_reserve symbol must be explicitly reserved with a
802 * separate memblock_reserve() or they will be discarded.
803 */
804 memblock_reserve_kern(__pa_symbol(_text),
805 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
806
807 /*
808 * The first 4Kb of memory is a BIOS owned area, but generally it is
809 * not listed as such in the E820 table.
810 *
811 * Reserve the first 64K of memory since some BIOSes are known to
812 * corrupt low memory. After the real mode trampoline is allocated the
813 * rest of the memory below 640k is reserved.
814 *
815 * In addition, make sure page 0 is always reserved because on
816 * systems with L1TF its contents can be leaked to user processes.
817 */
818 memblock_reserve(0, SZ_64K);
819
820 early_reserve_initrd();
821
822 memblock_x86_reserve_range_setup_data();
823
824 reserve_bios_regions();
825 trim_snb_memory();
826 }
827
828 /*
829 * Dump out kernel offset information on panic.
830 */
831 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)832 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
833 {
834 if (kaslr_enabled()) {
835 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
836 kaslr_offset(),
837 __START_KERNEL,
838 __START_KERNEL_map,
839 MODULES_VADDR-1);
840 } else {
841 pr_emerg("Kernel Offset: disabled\n");
842 }
843
844 return 0;
845 }
846
x86_configure_nx(void)847 void x86_configure_nx(void)
848 {
849 if (boot_cpu_has(X86_FEATURE_NX))
850 __supported_pte_mask |= _PAGE_NX;
851 else
852 __supported_pte_mask &= ~_PAGE_NX;
853 }
854
x86_report_nx(void)855 static void __init x86_report_nx(void)
856 {
857 if (!boot_cpu_has(X86_FEATURE_NX)) {
858 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
859 "missing in CPU!\n");
860 } else {
861 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
862 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
863 #else
864 /* 32bit non-PAE kernel, NX cannot be used */
865 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
866 "cannot be enabled: non-PAE kernel!\n");
867 #endif
868 }
869 }
870
871 /*
872 * Determine if we were loaded by an EFI loader. If so, then we have also been
873 * passed the efi memmap, systab, etc., so we should use these data structures
874 * for initialization. Note, the efi init code path is determined by the
875 * global efi_enabled. This allows the same kernel image to be used on existing
876 * systems (with a traditional BIOS) as well as on EFI systems.
877 */
878 /*
879 * setup_arch - architecture-specific boot-time initializations
880 *
881 * Note: On x86_64, fixmaps are ready for use even before this is called.
882 */
883
setup_arch(char ** cmdline_p)884 void __init setup_arch(char **cmdline_p)
885 {
886 #ifdef CONFIG_X86_32
887 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
888
889 /*
890 * copy kernel address range established so far and switch
891 * to the proper swapper page table
892 */
893 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
894 initial_page_table + KERNEL_PGD_BOUNDARY,
895 KERNEL_PGD_PTRS);
896
897 load_cr3(swapper_pg_dir);
898 /*
899 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
900 * a cr3 based tlb flush, so the following __flush_tlb_all()
901 * will not flush anything because the CPU quirk which clears
902 * X86_FEATURE_PGE has not been invoked yet. Though due to the
903 * load_cr3() above the TLB has been flushed already. The
904 * quirk is invoked before subsequent calls to __flush_tlb_all()
905 * so proper operation is guaranteed.
906 */
907 __flush_tlb_all();
908 #else
909 printk(KERN_INFO "Command line: %s\n", boot_command_line);
910 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
911 #endif
912
913 #ifdef CONFIG_CMDLINE_BOOL
914 #ifdef CONFIG_CMDLINE_OVERRIDE
915 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
916 #else
917 if (builtin_cmdline[0]) {
918 /* append boot loader cmdline to builtin */
919 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
920 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
921 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
922 }
923 #endif
924 builtin_cmdline_added = true;
925 #endif
926
927 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
928 *cmdline_p = command_line;
929
930 /*
931 * If we have OLPC OFW, we might end up relocating the fixmap due to
932 * reserve_top(), so do this before touching the ioremap area.
933 */
934 olpc_ofw_detect();
935
936 idt_setup_early_traps();
937 early_cpu_init();
938 jump_label_init();
939 static_call_init();
940 early_ioremap_init();
941
942 setup_olpc_ofw_pgd();
943
944 parse_boot_params();
945
946 x86_init.oem.arch_setup();
947
948 /*
949 * Do some memory reservations *before* memory is added to memblock, so
950 * memblock allocations won't overwrite it.
951 *
952 * After this point, everything still needed from the boot loader or
953 * firmware or kernel text should be early reserved or marked not RAM in
954 * e820. All other memory is free game.
955 *
956 * This call needs to happen before e820__memory_setup() which calls the
957 * xen_memory_setup() on Xen dom0 which relies on the fact that those
958 * early reservations have happened already.
959 */
960 early_reserve_memory();
961
962 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
963 e820__memory_setup();
964 parse_setup_data();
965
966 copy_edd();
967
968 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
969
970 /*
971 * x86_configure_nx() is called before parse_early_param() to detect
972 * whether hardware doesn't support NX (so that the early EHCI debug
973 * console setup can safely call set_fixmap()).
974 */
975 x86_configure_nx();
976
977 parse_early_param();
978
979 if (efi_enabled(EFI_BOOT))
980 efi_memblock_x86_reserve_range();
981
982 x86_report_nx();
983
984 apic_setup_apic_calls();
985
986 if (acpi_mps_check()) {
987 #ifdef CONFIG_X86_LOCAL_APIC
988 apic_is_disabled = true;
989 #endif
990 setup_clear_cpu_cap(X86_FEATURE_APIC);
991 }
992
993 e820__finish_early_params();
994
995 if (efi_enabled(EFI_BOOT))
996 efi_init();
997
998 reserve_ibft_region();
999 x86_init.resources.dmi_setup();
1000
1001 /*
1002 * VMware detection requires dmi to be available, so this
1003 * needs to be done after dmi_setup(), for the boot CPU.
1004 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
1005 * called before cache_bp_init() for setting up MTRR state.
1006 */
1007 init_hypervisor_platform();
1008
1009 tsc_early_init();
1010 x86_init.resources.probe_roms();
1011
1012 /*
1013 * Add resources for kernel text and data to the iomem_resource.
1014 * Do it after parse_early_param, so it can be debugged.
1015 */
1016 setup_kernel_resources();
1017
1018 e820_add_kernel_range();
1019 trim_bios_range();
1020 #ifdef CONFIG_X86_32
1021 if (ppro_with_ram_bug()) {
1022 pr_info("Applying PPro RAM bug workaround: punching 256 kB hole at 1.75 GB physical.\n");
1023 e820__range_update(0x70000000ULL, SZ_256K, E820_TYPE_RAM, E820_TYPE_RESERVED);
1024 e820__update_table(e820_table);
1025 }
1026 #else
1027 early_gart_iommu_check();
1028 #endif
1029
1030 /*
1031 * partially used pages are not usable - thus
1032 * we are rounding upwards:
1033 */
1034 max_pfn = e820__end_of_ram_pfn();
1035
1036 /* update e820 for memory not covered by WB MTRRs */
1037 cache_bp_init();
1038 if (mtrr_trim_uncached_memory(max_pfn))
1039 max_pfn = e820__end_of_ram_pfn();
1040
1041 max_possible_pfn = max_pfn;
1042
1043 /*
1044 * Define random base addresses for memory sections after max_pfn is
1045 * defined and before each memory section base is used.
1046 */
1047 kernel_randomize_memory();
1048
1049 #ifdef CONFIG_X86_32
1050 /* max_low_pfn get updated here */
1051 find_low_pfn_range();
1052 #else
1053 check_x2apic();
1054
1055 /* How many end-of-memory variables you have, grandma! */
1056 /* need this before calling reserve_initrd */
1057 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1058 max_low_pfn = e820__end_of_low_ram_pfn();
1059 else
1060 max_low_pfn = max_pfn;
1061 #endif
1062
1063 /* Find and reserve MPTABLE area */
1064 x86_init.mpparse.find_mptable();
1065
1066 early_alloc_pgt_buf();
1067
1068 /*
1069 * Need to conclude brk, before e820__memblock_setup()
1070 * it could use memblock_find_in_range, could overlap with
1071 * brk area.
1072 */
1073 reserve_brk();
1074
1075 cleanup_highmap();
1076
1077 e820__memblock_setup();
1078
1079 /*
1080 * Needs to run after memblock setup because it needs the physical
1081 * memory size.
1082 */
1083 mem_encrypt_setup_arch();
1084 cc_random_init();
1085
1086 efi_find_mirror();
1087 efi_esrt_init();
1088 efi_mokvar_table_init();
1089
1090 /*
1091 * The EFI specification says that boot service code won't be
1092 * called after ExitBootServices(). This is, in fact, a lie.
1093 */
1094 efi_reserve_boot_services();
1095
1096 /* preallocate 4k for mptable mpc */
1097 e820__memblock_alloc_reserved_mpc_new();
1098
1099 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1100 setup_bios_corruption_check();
1101 #endif
1102
1103 #ifdef CONFIG_X86_32
1104 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1105 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1106 #endif
1107
1108 /*
1109 * Find free memory for the real mode trampoline and place it there. If
1110 * there is not enough free memory under 1M, on EFI-enabled systems
1111 * there will be additional attempt to reclaim the memory for the real
1112 * mode trampoline at efi_free_boot_services().
1113 *
1114 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1115 * are known to corrupt low memory and several hundred kilobytes are not
1116 * worth complex detection what memory gets clobbered. Windows does the
1117 * same thing for very similar reasons.
1118 *
1119 * Moreover, on machines with SandyBridge graphics or in setups that use
1120 * crashkernel the entire 1M is reserved anyway.
1121 *
1122 * Note the host kernel TDX also requires the first 1MB being reserved.
1123 */
1124 x86_platform.realmode_reserve();
1125
1126 init_mem_mapping();
1127
1128 /*
1129 * init_mem_mapping() relies on the early IDT page fault handling.
1130 * Now either enable FRED or install the real page fault handler
1131 * for 64-bit in the IDT.
1132 */
1133 cpu_init_replace_early_idt();
1134
1135 /*
1136 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1137 * with the current CR4 value. This may not be necessary, but
1138 * auditing all the early-boot CR4 manipulation would be needed to
1139 * rule it out.
1140 *
1141 * Mask off features that don't work outside long mode (just
1142 * PCIDE for now).
1143 */
1144 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1145
1146 memblock_set_current_limit(get_max_mapped());
1147
1148 /*
1149 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1150 */
1151
1152 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1153 if (init_ohci1394_dma_early)
1154 init_ohci1394_dma_on_all_controllers();
1155 #endif
1156 /* Allocate bigger log buffer */
1157 setup_log_buf(1);
1158
1159 if (efi_enabled(EFI_BOOT)) {
1160 switch (boot_params.secure_boot) {
1161 case efi_secureboot_mode_disabled:
1162 pr_info("Secure boot disabled\n");
1163 break;
1164 case efi_secureboot_mode_enabled:
1165 pr_info("Secure boot enabled\n");
1166 break;
1167 default:
1168 pr_info("Secure boot could not be determined\n");
1169 break;
1170 }
1171 }
1172
1173 reserve_initrd();
1174
1175 acpi_table_upgrade();
1176 /* Look for ACPI tables and reserve memory occupied by them. */
1177 acpi_boot_table_init();
1178
1179 vsmp_init();
1180
1181 io_delay_init();
1182
1183 early_platform_quirks();
1184
1185 /* Some platforms need the APIC registered for NUMA configuration */
1186 early_acpi_boot_init();
1187 x86_init.mpparse.early_parse_smp_cfg();
1188
1189 x86_flattree_get_config();
1190
1191 initmem_init();
1192 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1193
1194 /*
1195 * Reserve memory for crash kernel after SRAT is parsed so that it
1196 * won't consume hotpluggable memory.
1197 */
1198 arch_reserve_crashkernel();
1199
1200 if (!early_xdbc_setup_hardware())
1201 early_xdbc_register_console();
1202
1203 x86_init.paging.pagetable_init();
1204
1205 kasan_init();
1206
1207 /*
1208 * Sync back kernel address range.
1209 *
1210 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1211 * this call?
1212 */
1213 sync_initial_page_table();
1214
1215 tboot_probe();
1216
1217 map_vsyscall();
1218
1219 x86_32_probe_apic();
1220
1221 early_quirks();
1222
1223 topology_apply_cmdline_limits_early();
1224
1225 /*
1226 * Parse SMP configuration. Try ACPI first and then the platform
1227 * specific parser.
1228 */
1229 acpi_boot_init();
1230 x86_init.mpparse.parse_smp_cfg();
1231
1232 /* Last opportunity to detect and map the local APIC */
1233 init_apic_mappings();
1234
1235 topology_init_possible_cpus();
1236
1237 init_cpu_to_node();
1238 init_gi_nodes();
1239
1240 io_apic_init_mappings();
1241
1242 x86_init.hyper.guest_late_init();
1243
1244 e820__reserve_resources();
1245 e820__register_nosave_regions(max_pfn);
1246
1247 x86_init.resources.reserve_resources();
1248
1249 e820__setup_pci_gap();
1250
1251 #ifdef CONFIG_VT
1252 #if defined(CONFIG_VGA_CONSOLE)
1253 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1254 vgacon_register_screen(&sysfb_primary_display.screen);
1255 #endif
1256 #endif
1257 x86_init.oem.banner();
1258
1259 x86_init.timers.wallclock_init();
1260
1261 /*
1262 * This needs to run before setup_local_APIC() which soft-disables the
1263 * local APIC temporarily and that masks the thermal LVT interrupt,
1264 * leading to softlockups on machines which have configured SMI
1265 * interrupt delivery.
1266 */
1267 therm_lvt_init();
1268
1269 mcheck_init();
1270
1271 register_refined_jiffies(CLOCK_TICK_RATE);
1272
1273 #ifdef CONFIG_EFI
1274 if (efi_enabled(EFI_BOOT))
1275 efi_apply_memmap_quirks();
1276 #endif
1277
1278 unwind_init();
1279 }
1280
1281 #ifdef CONFIG_X86_32
1282
1283 static struct resource video_ram_resource = {
1284 .name = "Video RAM area",
1285 .start = 0xa0000,
1286 .end = 0xbffff,
1287 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1288 };
1289
i386_reserve_resources(void)1290 void __init i386_reserve_resources(void)
1291 {
1292 request_resource(&iomem_resource, &video_ram_resource);
1293 reserve_standard_io_resources();
1294 }
1295
1296 #endif /* CONFIG_X86_32 */
1297
1298 static struct notifier_block kernel_offset_notifier = {
1299 .notifier_call = dump_kernel_offset
1300 };
1301
register_kernel_offset_dumper(void)1302 static int __init register_kernel_offset_dumper(void)
1303 {
1304 atomic_notifier_chain_register(&panic_notifier_list,
1305 &kernel_offset_notifier);
1306 return 0;
1307 }
1308 __initcall(register_kernel_offset_dumper);
1309
1310 #ifdef CONFIG_HOTPLUG_CPU
arch_cpu_is_hotpluggable(int cpu)1311 bool arch_cpu_is_hotpluggable(int cpu)
1312 {
1313 return cpu > 0;
1314 }
1315 #endif /* CONFIG_HOTPLUG_CPU */
1316