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/swiotlb.h>
26 #include <linux/tboot.h>
27 #include <linux/usb/xhci-dbgp.h>
28 #include <linux/vmalloc.h>
29
30 #include <uapi/linux/mount.h>
31
32 #include <xen/xen.h>
33
34 #include <asm/apic.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/nmi.h>
50 #include <asm/numa.h>
51 #include <asm/olpc_ofw.h>
52 #include <asm/pci-direct.h>
53 #include <asm/prom.h>
54 #include <asm/proto.h>
55 #include <asm/realmode.h>
56 #include <asm/thermal.h>
57 #include <asm/unwind.h>
58 #include <asm/vsyscall.h>
59
60 /*
61 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
62 * max_pfn_mapped: highest directly mapped pfn > 4 GB
63 *
64 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
65 * represented by pfn_mapped[].
66 */
67 unsigned long max_low_pfn_mapped;
68 unsigned long max_pfn_mapped;
69
70 #ifdef CONFIG_DMI
71 RESERVE_BRK(dmi_alloc, 65536);
72 #endif
73
74
75 unsigned long _brk_start = (unsigned long)__brk_base;
76 unsigned long _brk_end = (unsigned long)__brk_base;
77
78 struct boot_params boot_params;
79
80 /*
81 * These are the four main kernel memory regions, we put them into
82 * the resource tree so that kdump tools and other debugging tools
83 * recover it:
84 */
85
86 static struct resource rodata_resource = {
87 .name = "Kernel rodata",
88 .start = 0,
89 .end = 0,
90 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
91 };
92
93 static struct resource data_resource = {
94 .name = "Kernel data",
95 .start = 0,
96 .end = 0,
97 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
98 };
99
100 static struct resource code_resource = {
101 .name = "Kernel code",
102 .start = 0,
103 .end = 0,
104 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
105 };
106
107 static struct resource bss_resource = {
108 .name = "Kernel bss",
109 .start = 0,
110 .end = 0,
111 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
112 };
113
114
115 #ifdef CONFIG_X86_32
116 /* CPU data as detected by the assembly code in head_32.S */
117 struct cpuinfo_x86 new_cpu_data;
118
119 struct apm_info apm_info;
120 EXPORT_SYMBOL(apm_info);
121
122 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
123 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
124 struct ist_info ist_info;
125 EXPORT_SYMBOL(ist_info);
126 #else
127 struct ist_info ist_info;
128 #endif
129
130 #endif
131
132 struct cpuinfo_x86 boot_cpu_data __read_mostly;
133 EXPORT_SYMBOL(boot_cpu_data);
134 SYM_PIC_ALIAS(boot_cpu_data);
135
136 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
137 __visible unsigned long mmu_cr4_features __ro_after_init;
138 #else
139 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
140 #endif
141
142 #ifdef CONFIG_IMA
143 static phys_addr_t ima_kexec_buffer_phys;
144 static size_t ima_kexec_buffer_size;
145 #endif
146
147 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
148 int bootloader_type, bootloader_version;
149
150 static const struct ctl_table x86_sysctl_table[] = {
151 {
152 .procname = "unknown_nmi_panic",
153 .data = &unknown_nmi_panic,
154 .maxlen = sizeof(int),
155 .mode = 0644,
156 .proc_handler = proc_dointvec,
157 },
158 {
159 .procname = "panic_on_unrecovered_nmi",
160 .data = &panic_on_unrecovered_nmi,
161 .maxlen = sizeof(int),
162 .mode = 0644,
163 .proc_handler = proc_dointvec,
164 },
165 {
166 .procname = "panic_on_io_nmi",
167 .data = &panic_on_io_nmi,
168 .maxlen = sizeof(int),
169 .mode = 0644,
170 .proc_handler = proc_dointvec,
171 },
172 {
173 .procname = "bootloader_type",
174 .data = &bootloader_type,
175 .maxlen = sizeof(int),
176 .mode = 0444,
177 .proc_handler = proc_dointvec,
178 },
179 {
180 .procname = "bootloader_version",
181 .data = &bootloader_version,
182 .maxlen = sizeof(int),
183 .mode = 0444,
184 .proc_handler = proc_dointvec,
185 },
186 {
187 .procname = "io_delay_type",
188 .data = &io_delay_type,
189 .maxlen = sizeof(int),
190 .mode = 0644,
191 .proc_handler = proc_dointvec,
192 },
193 #if defined(CONFIG_ACPI_SLEEP)
194 {
195 .procname = "acpi_video_flags",
196 .data = &acpi_realmode_flags,
197 .maxlen = sizeof(unsigned long),
198 .mode = 0644,
199 .proc_handler = proc_doulongvec_minmax,
200 },
201 #endif
202 };
203
init_x86_sysctl(void)204 static int __init init_x86_sysctl(void)
205 {
206 register_sysctl_init("kernel", x86_sysctl_table);
207 return 0;
208 }
209 arch_initcall(init_x86_sysctl);
210
211 /*
212 * Setup options
213 */
214 struct screen_info screen_info;
215 EXPORT_SYMBOL(screen_info);
216 struct edid_info edid_info;
217 EXPORT_SYMBOL_GPL(edid_info);
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 if (!ima_kexec_buffer_size)
441 return -ENOENT;
442
443 *addr = __va(ima_kexec_buffer_phys);
444 *size = ima_kexec_buffer_size;
445
446 return 0;
447 }
448 #endif
449
add_kho(u64 phys_addr,u32 data_len)450 static void __init add_kho(u64 phys_addr, u32 data_len)
451 {
452 struct kho_data *kho;
453 u64 addr = phys_addr + sizeof(struct setup_data);
454 u64 size = data_len - sizeof(struct setup_data);
455
456 if (!IS_ENABLED(CONFIG_KEXEC_HANDOVER)) {
457 pr_warn("Passed KHO data, but CONFIG_KEXEC_HANDOVER not set. Ignoring.\n");
458 return;
459 }
460
461 kho = early_memremap(addr, size);
462 if (!kho) {
463 pr_warn("setup: failed to memremap kho data (0x%llx, 0x%llx)\n",
464 addr, size);
465 return;
466 }
467
468 kho_populate(kho->fdt_addr, kho->fdt_size, kho->scratch_addr, kho->scratch_size);
469
470 early_memunmap(kho, size);
471 }
472
parse_setup_data(void)473 static void __init parse_setup_data(void)
474 {
475 struct setup_data *data;
476 u64 pa_data, pa_next;
477
478 pa_data = boot_params.hdr.setup_data;
479 while (pa_data) {
480 u32 data_len, data_type;
481
482 data = early_memremap(pa_data, sizeof(*data));
483 data_len = data->len + sizeof(struct setup_data);
484 data_type = data->type;
485 pa_next = data->next;
486 early_memunmap(data, sizeof(*data));
487
488 switch (data_type) {
489 case SETUP_E820_EXT:
490 e820__memory_setup_extended(pa_data, data_len);
491 break;
492 case SETUP_DTB:
493 add_dtb(pa_data);
494 break;
495 case SETUP_EFI:
496 parse_efi_setup(pa_data, data_len);
497 break;
498 case SETUP_IMA:
499 add_early_ima_buffer(pa_data);
500 break;
501 case SETUP_KEXEC_KHO:
502 add_kho(pa_data, data_len);
503 break;
504 case SETUP_RNG_SEED:
505 data = early_memremap(pa_data, data_len);
506 add_bootloader_randomness(data->data, data->len);
507 /* Zero seed for forward secrecy. */
508 memzero_explicit(data->data, data->len);
509 /* Zero length in case we find ourselves back here by accident. */
510 memzero_explicit(&data->len, sizeof(data->len));
511 early_memunmap(data, data_len);
512 break;
513 default:
514 break;
515 }
516 pa_data = pa_next;
517 }
518 }
519
520 /*
521 * Translate the fields of 'struct boot_param' into global variables
522 * representing these parameters.
523 */
parse_boot_params(void)524 static void __init parse_boot_params(void)
525 {
526 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
527 screen_info = boot_params.screen_info;
528 edid_info = boot_params.edid_info;
529 #ifdef CONFIG_X86_32
530 apm_info.bios = boot_params.apm_bios_info;
531 ist_info = boot_params.ist_info;
532 #endif
533 saved_video_mode = boot_params.hdr.vid_mode;
534 bootloader_type = boot_params.hdr.type_of_loader;
535 if ((bootloader_type >> 4) == 0xe) {
536 bootloader_type &= 0xf;
537 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
538 }
539 bootloader_version = bootloader_type & 0xf;
540 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
541
542 #ifdef CONFIG_BLK_DEV_RAM
543 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
544 #endif
545 #ifdef CONFIG_EFI
546 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
547 EFI32_LOADER_SIGNATURE, 4)) {
548 set_bit(EFI_BOOT, &efi.flags);
549 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
550 EFI64_LOADER_SIGNATURE, 4)) {
551 set_bit(EFI_BOOT, &efi.flags);
552 set_bit(EFI_64BIT, &efi.flags);
553 }
554 #endif
555
556 if (!boot_params.hdr.root_flags)
557 root_mountflags &= ~MS_RDONLY;
558 }
559
memblock_x86_reserve_range_setup_data(void)560 static void __init memblock_x86_reserve_range_setup_data(void)
561 {
562 struct setup_indirect *indirect;
563 struct setup_data *data;
564 u64 pa_data, pa_next;
565 u32 len;
566
567 pa_data = boot_params.hdr.setup_data;
568 while (pa_data) {
569 data = early_memremap(pa_data, sizeof(*data));
570 if (!data) {
571 pr_warn("setup: failed to memremap setup_data entry\n");
572 return;
573 }
574
575 len = sizeof(*data);
576 pa_next = data->next;
577
578 memblock_reserve_kern(pa_data, sizeof(*data) + data->len);
579
580 if (data->type == SETUP_INDIRECT) {
581 len += data->len;
582 early_memunmap(data, sizeof(*data));
583 data = early_memremap(pa_data, len);
584 if (!data) {
585 pr_warn("setup: failed to memremap indirect setup_data\n");
586 return;
587 }
588
589 indirect = (struct setup_indirect *)data->data;
590
591 if (indirect->type != SETUP_INDIRECT)
592 memblock_reserve_kern(indirect->addr, indirect->len);
593 }
594
595 pa_data = pa_next;
596 early_memunmap(data, len);
597 }
598 }
599
arch_reserve_crashkernel(void)600 static void __init arch_reserve_crashkernel(void)
601 {
602 unsigned long long crash_base, crash_size, low_size = 0;
603 bool high = false;
604 int ret;
605
606 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
607 return;
608
609 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
610 &crash_size, &crash_base,
611 &low_size, &high);
612 if (ret)
613 return;
614
615 if (xen_pv_domain()) {
616 pr_info("Ignoring crashkernel for a Xen PV domain\n");
617 return;
618 }
619
620 reserve_crashkernel_generic(crash_size, crash_base, low_size, high);
621 }
622
623 static struct resource standard_io_resources[] = {
624 { .name = "dma1", .start = 0x00, .end = 0x1f,
625 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
626 { .name = "pic1", .start = 0x20, .end = 0x21,
627 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
628 { .name = "timer0", .start = 0x40, .end = 0x43,
629 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
630 { .name = "timer1", .start = 0x50, .end = 0x53,
631 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
632 { .name = "keyboard", .start = 0x60, .end = 0x60,
633 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
634 { .name = "keyboard", .start = 0x64, .end = 0x64,
635 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
636 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
637 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
638 { .name = "pic2", .start = 0xa0, .end = 0xa1,
639 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
640 { .name = "dma2", .start = 0xc0, .end = 0xdf,
641 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
642 { .name = "fpu", .start = 0xf0, .end = 0xff,
643 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
644 };
645
reserve_standard_io_resources(void)646 void __init reserve_standard_io_resources(void)
647 {
648 int i;
649
650 /* request I/O space for devices used on all i[345]86 PCs */
651 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
652 request_resource(&ioport_resource, &standard_io_resources[i]);
653
654 }
655
setup_kernel_resources(void)656 static void __init setup_kernel_resources(void)
657 {
658 code_resource.start = __pa_symbol(_text);
659 code_resource.end = __pa_symbol(_etext)-1;
660 rodata_resource.start = __pa_symbol(__start_rodata);
661 rodata_resource.end = __pa_symbol(__end_rodata)-1;
662 data_resource.start = __pa_symbol(_sdata);
663 data_resource.end = __pa_symbol(_edata)-1;
664 bss_resource.start = __pa_symbol(__bss_start);
665 bss_resource.end = __pa_symbol(__bss_stop)-1;
666
667 insert_resource(&iomem_resource, &code_resource);
668 insert_resource(&iomem_resource, &rodata_resource);
669 insert_resource(&iomem_resource, &data_resource);
670 insert_resource(&iomem_resource, &bss_resource);
671 }
672
snb_gfx_workaround_needed(void)673 static bool __init snb_gfx_workaround_needed(void)
674 {
675 #ifdef CONFIG_PCI
676 int i;
677 u16 vendor, devid;
678 static const __initconst u16 snb_ids[] = {
679 0x0102,
680 0x0112,
681 0x0122,
682 0x0106,
683 0x0116,
684 0x0126,
685 0x010a,
686 };
687
688 /* Assume no if something weird is going on with PCI */
689 if (!early_pci_allowed())
690 return false;
691
692 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
693 if (vendor != 0x8086)
694 return false;
695
696 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
697 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
698 if (devid == snb_ids[i])
699 return true;
700 #endif
701
702 return false;
703 }
704
705 /*
706 * Sandy Bridge graphics has trouble with certain ranges, exclude
707 * them from allocation.
708 */
trim_snb_memory(void)709 static void __init trim_snb_memory(void)
710 {
711 static const __initconst unsigned long bad_pages[] = {
712 0x20050000,
713 0x20110000,
714 0x20130000,
715 0x20138000,
716 0x40004000,
717 };
718 int i;
719
720 if (!snb_gfx_workaround_needed())
721 return;
722
723 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
724
725 /*
726 * SandyBridge integrated graphics devices have a bug that prevents
727 * them from accessing certain memory ranges, namely anything below
728 * 1M and in the pages listed in bad_pages[] above.
729 *
730 * To avoid these pages being ever accessed by SNB gfx devices reserve
731 * bad_pages that have not already been reserved at boot time.
732 * All memory below the 1 MB mark is anyway reserved later during
733 * setup_arch(), so there is no need to reserve it here.
734 */
735
736 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
737 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
738 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
739 bad_pages[i]);
740 }
741 }
742
trim_bios_range(void)743 static void __init trim_bios_range(void)
744 {
745 /*
746 * A special case is the first 4Kb of memory;
747 * This is a BIOS owned area, not kernel ram, but generally
748 * not listed as such in the E820 table.
749 *
750 * This typically reserves additional memory (64KiB by default)
751 * since some BIOSes are known to corrupt low memory. See the
752 * Kconfig help text for X86_RESERVE_LOW.
753 */
754 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
755
756 /*
757 * special case: Some BIOSes report the PC BIOS
758 * area (640Kb -> 1Mb) as RAM even though it is not.
759 * take them out.
760 */
761 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
762
763 e820__update_table(e820_table);
764 }
765
766 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)767 static void __init e820_add_kernel_range(void)
768 {
769 u64 start = __pa_symbol(_text);
770 u64 size = __pa_symbol(_end) - start;
771
772 /*
773 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
774 * attempt to fix it by adding the range. We may have a confused BIOS,
775 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
776 * exclude kernel range. If we really are running on top non-RAM,
777 * we will crash later anyways.
778 */
779 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
780 return;
781
782 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
783 e820__range_remove(start, size, E820_TYPE_RAM, 0);
784 e820__range_add(start, size, E820_TYPE_RAM);
785 }
786
early_reserve_memory(void)787 static void __init early_reserve_memory(void)
788 {
789 /*
790 * Reserve the memory occupied by the kernel between _text and
791 * __end_of_kernel_reserve symbols. Any kernel sections after the
792 * __end_of_kernel_reserve symbol must be explicitly reserved with a
793 * separate memblock_reserve() or they will be discarded.
794 */
795 memblock_reserve_kern(__pa_symbol(_text),
796 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
797
798 /*
799 * The first 4Kb of memory is a BIOS owned area, but generally it is
800 * not listed as such in the E820 table.
801 *
802 * Reserve the first 64K of memory since some BIOSes are known to
803 * corrupt low memory. After the real mode trampoline is allocated the
804 * rest of the memory below 640k is reserved.
805 *
806 * In addition, make sure page 0 is always reserved because on
807 * systems with L1TF its contents can be leaked to user processes.
808 */
809 memblock_reserve(0, SZ_64K);
810
811 early_reserve_initrd();
812
813 memblock_x86_reserve_range_setup_data();
814
815 reserve_bios_regions();
816 trim_snb_memory();
817 }
818
819 /*
820 * Dump out kernel offset information on panic.
821 */
822 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)823 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
824 {
825 if (kaslr_enabled()) {
826 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
827 kaslr_offset(),
828 __START_KERNEL,
829 __START_KERNEL_map,
830 MODULES_VADDR-1);
831 } else {
832 pr_emerg("Kernel Offset: disabled\n");
833 }
834
835 return 0;
836 }
837
x86_configure_nx(void)838 void x86_configure_nx(void)
839 {
840 if (boot_cpu_has(X86_FEATURE_NX))
841 __supported_pte_mask |= _PAGE_NX;
842 else
843 __supported_pte_mask &= ~_PAGE_NX;
844 }
845
x86_report_nx(void)846 static void __init x86_report_nx(void)
847 {
848 if (!boot_cpu_has(X86_FEATURE_NX)) {
849 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
850 "missing in CPU!\n");
851 } else {
852 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
853 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
854 #else
855 /* 32bit non-PAE kernel, NX cannot be used */
856 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
857 "cannot be enabled: non-PAE kernel!\n");
858 #endif
859 }
860 }
861
862 /*
863 * Determine if we were loaded by an EFI loader. If so, then we have also been
864 * passed the efi memmap, systab, etc., so we should use these data structures
865 * for initialization. Note, the efi init code path is determined by the
866 * global efi_enabled. This allows the same kernel image to be used on existing
867 * systems (with a traditional BIOS) as well as on EFI systems.
868 */
869 /*
870 * setup_arch - architecture-specific boot-time initializations
871 *
872 * Note: On x86_64, fixmaps are ready for use even before this is called.
873 */
874
setup_arch(char ** cmdline_p)875 void __init setup_arch(char **cmdline_p)
876 {
877 #ifdef CONFIG_X86_32
878 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
879
880 /*
881 * copy kernel address range established so far and switch
882 * to the proper swapper page table
883 */
884 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
885 initial_page_table + KERNEL_PGD_BOUNDARY,
886 KERNEL_PGD_PTRS);
887
888 load_cr3(swapper_pg_dir);
889 /*
890 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
891 * a cr3 based tlb flush, so the following __flush_tlb_all()
892 * will not flush anything because the CPU quirk which clears
893 * X86_FEATURE_PGE has not been invoked yet. Though due to the
894 * load_cr3() above the TLB has been flushed already. The
895 * quirk is invoked before subsequent calls to __flush_tlb_all()
896 * so proper operation is guaranteed.
897 */
898 __flush_tlb_all();
899 #else
900 printk(KERN_INFO "Command line: %s\n", boot_command_line);
901 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
902 #endif
903
904 #ifdef CONFIG_CMDLINE_BOOL
905 #ifdef CONFIG_CMDLINE_OVERRIDE
906 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
907 #else
908 if (builtin_cmdline[0]) {
909 /* append boot loader cmdline to builtin */
910 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
911 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
912 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
913 }
914 #endif
915 builtin_cmdline_added = true;
916 #endif
917
918 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
919 *cmdline_p = command_line;
920
921 /*
922 * If we have OLPC OFW, we might end up relocating the fixmap due to
923 * reserve_top(), so do this before touching the ioremap area.
924 */
925 olpc_ofw_detect();
926
927 idt_setup_early_traps();
928 early_cpu_init();
929 jump_label_init();
930 static_call_init();
931 early_ioremap_init();
932
933 setup_olpc_ofw_pgd();
934
935 parse_boot_params();
936
937 x86_init.oem.arch_setup();
938
939 /*
940 * Do some memory reservations *before* memory is added to memblock, so
941 * memblock allocations won't overwrite it.
942 *
943 * After this point, everything still needed from the boot loader or
944 * firmware or kernel text should be early reserved or marked not RAM in
945 * e820. All other memory is free game.
946 *
947 * This call needs to happen before e820__memory_setup() which calls the
948 * xen_memory_setup() on Xen dom0 which relies on the fact that those
949 * early reservations have happened already.
950 */
951 early_reserve_memory();
952
953 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
954 e820__memory_setup();
955 parse_setup_data();
956
957 copy_edd();
958
959 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
960
961 /*
962 * x86_configure_nx() is called before parse_early_param() to detect
963 * whether hardware doesn't support NX (so that the early EHCI debug
964 * console setup can safely call set_fixmap()).
965 */
966 x86_configure_nx();
967
968 parse_early_param();
969
970 if (efi_enabled(EFI_BOOT))
971 efi_memblock_x86_reserve_range();
972
973 x86_report_nx();
974
975 apic_setup_apic_calls();
976
977 if (acpi_mps_check()) {
978 #ifdef CONFIG_X86_LOCAL_APIC
979 apic_is_disabled = true;
980 #endif
981 setup_clear_cpu_cap(X86_FEATURE_APIC);
982 }
983
984 e820__finish_early_params();
985
986 if (efi_enabled(EFI_BOOT))
987 efi_init();
988
989 reserve_ibft_region();
990 x86_init.resources.dmi_setup();
991
992 /*
993 * VMware detection requires dmi to be available, so this
994 * needs to be done after dmi_setup(), for the boot CPU.
995 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
996 * called before cache_bp_init() for setting up MTRR state.
997 */
998 init_hypervisor_platform();
999
1000 tsc_early_init();
1001 x86_init.resources.probe_roms();
1002
1003 /*
1004 * Add resources for kernel text and data to the iomem_resource.
1005 * Do it after parse_early_param, so it can be debugged.
1006 */
1007 setup_kernel_resources();
1008
1009 e820_add_kernel_range();
1010 trim_bios_range();
1011 #ifdef CONFIG_X86_32
1012 if (ppro_with_ram_bug()) {
1013 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1014 E820_TYPE_RESERVED);
1015 e820__update_table(e820_table);
1016 printk(KERN_INFO "fixed physical RAM map:\n");
1017 e820__print_table("bad_ppro");
1018 }
1019 #else
1020 early_gart_iommu_check();
1021 #endif
1022
1023 /*
1024 * partially used pages are not usable - thus
1025 * we are rounding upwards:
1026 */
1027 max_pfn = e820__end_of_ram_pfn();
1028
1029 /* update e820 for memory not covered by WB MTRRs */
1030 cache_bp_init();
1031 if (mtrr_trim_uncached_memory(max_pfn))
1032 max_pfn = e820__end_of_ram_pfn();
1033
1034 max_possible_pfn = max_pfn;
1035
1036 /*
1037 * Define random base addresses for memory sections after max_pfn is
1038 * defined and before each memory section base is used.
1039 */
1040 kernel_randomize_memory();
1041
1042 #ifdef CONFIG_X86_32
1043 /* max_low_pfn get updated here */
1044 find_low_pfn_range();
1045 #else
1046 check_x2apic();
1047
1048 /* How many end-of-memory variables you have, grandma! */
1049 /* need this before calling reserve_initrd */
1050 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1051 max_low_pfn = e820__end_of_low_ram_pfn();
1052 else
1053 max_low_pfn = max_pfn;
1054 #endif
1055
1056 /* Find and reserve MPTABLE area */
1057 x86_init.mpparse.find_mptable();
1058
1059 early_alloc_pgt_buf();
1060
1061 /*
1062 * Need to conclude brk, before e820__memblock_setup()
1063 * it could use memblock_find_in_range, could overlap with
1064 * brk area.
1065 */
1066 reserve_brk();
1067
1068 cleanup_highmap();
1069
1070 e820__memblock_setup();
1071
1072 /*
1073 * Needs to run after memblock setup because it needs the physical
1074 * memory size.
1075 */
1076 mem_encrypt_setup_arch();
1077 cc_random_init();
1078
1079 efi_find_mirror();
1080 efi_esrt_init();
1081 efi_mokvar_table_init();
1082
1083 /*
1084 * The EFI specification says that boot service code won't be
1085 * called after ExitBootServices(). This is, in fact, a lie.
1086 */
1087 efi_reserve_boot_services();
1088
1089 /* preallocate 4k for mptable mpc */
1090 e820__memblock_alloc_reserved_mpc_new();
1091
1092 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1093 setup_bios_corruption_check();
1094 #endif
1095
1096 #ifdef CONFIG_X86_32
1097 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1098 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1099 #endif
1100
1101 /*
1102 * Find free memory for the real mode trampoline and place it there. If
1103 * there is not enough free memory under 1M, on EFI-enabled systems
1104 * there will be additional attempt to reclaim the memory for the real
1105 * mode trampoline at efi_free_boot_services().
1106 *
1107 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1108 * are known to corrupt low memory and several hundred kilobytes are not
1109 * worth complex detection what memory gets clobbered. Windows does the
1110 * same thing for very similar reasons.
1111 *
1112 * Moreover, on machines with SandyBridge graphics or in setups that use
1113 * crashkernel the entire 1M is reserved anyway.
1114 *
1115 * Note the host kernel TDX also requires the first 1MB being reserved.
1116 */
1117 x86_platform.realmode_reserve();
1118
1119 init_mem_mapping();
1120
1121 /*
1122 * init_mem_mapping() relies on the early IDT page fault handling.
1123 * Now either enable FRED or install the real page fault handler
1124 * for 64-bit in the IDT.
1125 */
1126 cpu_init_replace_early_idt();
1127
1128 /*
1129 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1130 * with the current CR4 value. This may not be necessary, but
1131 * auditing all the early-boot CR4 manipulation would be needed to
1132 * rule it out.
1133 *
1134 * Mask off features that don't work outside long mode (just
1135 * PCIDE for now).
1136 */
1137 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1138
1139 memblock_set_current_limit(get_max_mapped());
1140
1141 /*
1142 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1143 */
1144
1145 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1146 if (init_ohci1394_dma_early)
1147 init_ohci1394_dma_on_all_controllers();
1148 #endif
1149 /* Allocate bigger log buffer */
1150 setup_log_buf(1);
1151
1152 if (efi_enabled(EFI_BOOT)) {
1153 switch (boot_params.secure_boot) {
1154 case efi_secureboot_mode_disabled:
1155 pr_info("Secure boot disabled\n");
1156 break;
1157 case efi_secureboot_mode_enabled:
1158 pr_info("Secure boot enabled\n");
1159 break;
1160 default:
1161 pr_info("Secure boot could not be determined\n");
1162 break;
1163 }
1164 }
1165
1166 reserve_initrd();
1167
1168 acpi_table_upgrade();
1169 /* Look for ACPI tables and reserve memory occupied by them. */
1170 acpi_boot_table_init();
1171
1172 vsmp_init();
1173
1174 io_delay_init();
1175
1176 early_platform_quirks();
1177
1178 /* Some platforms need the APIC registered for NUMA configuration */
1179 early_acpi_boot_init();
1180 x86_init.mpparse.early_parse_smp_cfg();
1181
1182 x86_flattree_get_config();
1183
1184 initmem_init();
1185 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1186
1187 if (boot_cpu_has(X86_FEATURE_GBPAGES)) {
1188 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1189 hugetlb_bootmem_alloc();
1190 }
1191
1192 /*
1193 * Reserve memory for crash kernel after SRAT is parsed so that it
1194 * won't consume hotpluggable memory.
1195 */
1196 arch_reserve_crashkernel();
1197
1198 if (!early_xdbc_setup_hardware())
1199 early_xdbc_register_console();
1200
1201 x86_init.paging.pagetable_init();
1202
1203 kasan_init();
1204
1205 /*
1206 * Sync back kernel address range.
1207 *
1208 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1209 * this call?
1210 */
1211 sync_initial_page_table();
1212
1213 tboot_probe();
1214
1215 map_vsyscall();
1216
1217 x86_32_probe_apic();
1218
1219 early_quirks();
1220
1221 topology_apply_cmdline_limits_early();
1222
1223 /*
1224 * Parse SMP configuration. Try ACPI first and then the platform
1225 * specific parser.
1226 */
1227 acpi_boot_init();
1228 x86_init.mpparse.parse_smp_cfg();
1229
1230 /* Last opportunity to detect and map the local APIC */
1231 init_apic_mappings();
1232
1233 topology_init_possible_cpus();
1234
1235 init_cpu_to_node();
1236 init_gi_nodes();
1237
1238 io_apic_init_mappings();
1239
1240 x86_init.hyper.guest_late_init();
1241
1242 e820__reserve_resources();
1243 e820__register_nosave_regions(max_pfn);
1244
1245 x86_init.resources.reserve_resources();
1246
1247 e820__setup_pci_gap();
1248
1249 #ifdef CONFIG_VT
1250 #if defined(CONFIG_VGA_CONSOLE)
1251 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1252 vgacon_register_screen(&screen_info);
1253 #endif
1254 #endif
1255 x86_init.oem.banner();
1256
1257 x86_init.timers.wallclock_init();
1258
1259 /*
1260 * This needs to run before setup_local_APIC() which soft-disables the
1261 * local APIC temporarily and that masks the thermal LVT interrupt,
1262 * leading to softlockups on machines which have configured SMI
1263 * interrupt delivery.
1264 */
1265 therm_lvt_init();
1266
1267 mcheck_init();
1268
1269 register_refined_jiffies(CLOCK_TICK_RATE);
1270
1271 #ifdef CONFIG_EFI
1272 if (efi_enabled(EFI_BOOT))
1273 efi_apply_memmap_quirks();
1274 #endif
1275
1276 unwind_init();
1277 }
1278
1279 #ifdef CONFIG_X86_32
1280
1281 static struct resource video_ram_resource = {
1282 .name = "Video RAM area",
1283 .start = 0xa0000,
1284 .end = 0xbffff,
1285 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1286 };
1287
i386_reserve_resources(void)1288 void __init i386_reserve_resources(void)
1289 {
1290 request_resource(&iomem_resource, &video_ram_resource);
1291 reserve_standard_io_resources();
1292 }
1293
1294 #endif /* CONFIG_X86_32 */
1295
1296 static struct notifier_block kernel_offset_notifier = {
1297 .notifier_call = dump_kernel_offset
1298 };
1299
register_kernel_offset_dumper(void)1300 static int __init register_kernel_offset_dumper(void)
1301 {
1302 atomic_notifier_chain_register(&panic_notifier_list,
1303 &kernel_offset_notifier);
1304 return 0;
1305 }
1306 __initcall(register_kernel_offset_dumper);
1307
1308 #ifdef CONFIG_HOTPLUG_CPU
arch_cpu_is_hotpluggable(int cpu)1309 bool arch_cpu_is_hotpluggable(int cpu)
1310 {
1311 return cpu > 0;
1312 }
1313 #endif /* CONFIG_HOTPLUG_CPU */
1314