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/ima.h>
15 #include <linux/init_ohci1394_dma.h>
16 #include <linux/initrd.h>
17 #include <linux/iscsi_ibft.h>
18 #include <linux/memblock.h>
19 #include <linux/panic_notifier.h>
20 #include <linux/pci.h>
21 #include <linux/root_dev.h>
22 #include <linux/hugetlb.h>
23 #include <linux/tboot.h>
24 #include <linux/usb/xhci-dbgp.h>
25 #include <linux/static_call.h>
26 #include <linux/swiotlb.h>
27 #include <linux/random.h>
28
29 #include <uapi/linux/mount.h>
30
31 #include <xen/xen.h>
32
33 #include <asm/apic.h>
34 #include <asm/efi.h>
35 #include <asm/numa.h>
36 #include <asm/bios_ebda.h>
37 #include <asm/bugs.h>
38 #include <asm/cacheinfo.h>
39 #include <asm/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/realmode.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/thermal.h>
56 #include <asm/unwind.h>
57 #include <asm/vsyscall.h>
58 #include <linux/vmalloc.h>
59 #if defined(CONFIG_X86_LOCAL_APIC)
60 #include <asm/nmi.h>
61 #endif
62
63 /*
64 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
65 * max_pfn_mapped: highest directly mapped pfn > 4 GB
66 *
67 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
68 * represented by pfn_mapped[].
69 */
70 unsigned long max_low_pfn_mapped;
71 unsigned long max_pfn_mapped;
72
73 #ifdef CONFIG_DMI
74 RESERVE_BRK(dmi_alloc, 65536);
75 #endif
76
77
78 unsigned long _brk_start = (unsigned long)__brk_base;
79 unsigned long _brk_end = (unsigned long)__brk_base;
80
81 struct boot_params boot_params;
82
83 /*
84 * These are the four main kernel memory regions, we put them into
85 * the resource tree so that kdump tools and other debugging tools
86 * recover it:
87 */
88
89 static struct resource rodata_resource = {
90 .name = "Kernel rodata",
91 .start = 0,
92 .end = 0,
93 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
94 };
95
96 static struct resource data_resource = {
97 .name = "Kernel data",
98 .start = 0,
99 .end = 0,
100 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
101 };
102
103 static struct resource code_resource = {
104 .name = "Kernel code",
105 .start = 0,
106 .end = 0,
107 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
108 };
109
110 static struct resource bss_resource = {
111 .name = "Kernel bss",
112 .start = 0,
113 .end = 0,
114 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
115 };
116
117
118 #ifdef CONFIG_X86_32
119 /* CPU data as detected by the assembly code in head_32.S */
120 struct cpuinfo_x86 new_cpu_data;
121
122 struct apm_info apm_info;
123 EXPORT_SYMBOL(apm_info);
124
125 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
126 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
127 struct ist_info ist_info;
128 EXPORT_SYMBOL(ist_info);
129 #else
130 struct ist_info ist_info;
131 #endif
132
133 #endif
134
135 struct cpuinfo_x86 boot_cpu_data __read_mostly;
136 EXPORT_SYMBOL(boot_cpu_data);
137
138 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
139 __visible unsigned long mmu_cr4_features __ro_after_init;
140 #else
141 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
142 #endif
143
144 #ifdef CONFIG_IMA
145 static phys_addr_t ima_kexec_buffer_phys;
146 static size_t ima_kexec_buffer_size;
147 #endif
148
149 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
150 int bootloader_type, bootloader_version;
151
152 static const struct ctl_table x86_sysctl_table[] = {
153 {
154 .procname = "panic_on_unrecovered_nmi",
155 .data = &panic_on_unrecovered_nmi,
156 .maxlen = sizeof(int),
157 .mode = 0644,
158 .proc_handler = proc_dointvec,
159 },
160 {
161 .procname = "panic_on_io_nmi",
162 .data = &panic_on_io_nmi,
163 .maxlen = sizeof(int),
164 .mode = 0644,
165 .proc_handler = proc_dointvec,
166 },
167 {
168 .procname = "bootloader_type",
169 .data = &bootloader_type,
170 .maxlen = sizeof(int),
171 .mode = 0444,
172 .proc_handler = proc_dointvec,
173 },
174 {
175 .procname = "bootloader_version",
176 .data = &bootloader_version,
177 .maxlen = sizeof(int),
178 .mode = 0444,
179 .proc_handler = proc_dointvec,
180 },
181 {
182 .procname = "io_delay_type",
183 .data = &io_delay_type,
184 .maxlen = sizeof(int),
185 .mode = 0644,
186 .proc_handler = proc_dointvec,
187 },
188 #if defined(CONFIG_X86_LOCAL_APIC)
189 {
190 .procname = "unknown_nmi_panic",
191 .data = &unknown_nmi_panic,
192 .maxlen = sizeof(int),
193 .mode = 0644,
194 .proc_handler = proc_dointvec,
195 },
196 #endif
197 #if defined(CONFIG_ACPI_SLEEP)
198 {
199 .procname = "acpi_video_flags",
200 .data = &acpi_realmode_flags,
201 .maxlen = sizeof(unsigned long),
202 .mode = 0644,
203 .proc_handler = proc_doulongvec_minmax,
204 },
205 #endif
206 };
207
init_x86_sysctl(void)208 static int __init init_x86_sysctl(void)
209 {
210 register_sysctl_init("kernel", x86_sysctl_table);
211 return 0;
212 }
213 arch_initcall(init_x86_sysctl);
214
215 /*
216 * Setup options
217 */
218 struct screen_info screen_info;
219 EXPORT_SYMBOL(screen_info);
220 struct edid_info edid_info;
221 EXPORT_SYMBOL_GPL(edid_info);
222
223 extern int root_mountflags;
224
225 unsigned long saved_video_mode;
226
227 #define RAMDISK_IMAGE_START_MASK 0x07FF
228 #define RAMDISK_PROMPT_FLAG 0x8000
229 #define RAMDISK_LOAD_FLAG 0x4000
230
231 static char __initdata command_line[COMMAND_LINE_SIZE];
232 #ifdef CONFIG_CMDLINE_BOOL
233 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
234 bool builtin_cmdline_added __ro_after_init;
235 #endif
236
237 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
238 struct edd edd;
239 #ifdef CONFIG_EDD_MODULE
240 EXPORT_SYMBOL(edd);
241 #endif
242 /**
243 * copy_edd() - Copy the BIOS EDD information
244 * from boot_params into a safe place.
245 *
246 */
copy_edd(void)247 static inline void __init copy_edd(void)
248 {
249 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
250 sizeof(edd.mbr_signature));
251 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
252 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
253 edd.edd_info_nr = boot_params.eddbuf_entries;
254 }
255 #else
copy_edd(void)256 static inline void __init copy_edd(void)
257 {
258 }
259 #endif
260
extend_brk(size_t size,size_t align)261 void * __init extend_brk(size_t size, size_t align)
262 {
263 size_t mask = align - 1;
264 void *ret;
265
266 BUG_ON(_brk_start == 0);
267 BUG_ON(align & mask);
268
269 _brk_end = (_brk_end + mask) & ~mask;
270 BUG_ON((char *)(_brk_end + size) > __brk_limit);
271
272 ret = (void *)_brk_end;
273 _brk_end += size;
274
275 memset(ret, 0, size);
276
277 return ret;
278 }
279
280 #ifdef CONFIG_X86_32
cleanup_highmap(void)281 static void __init cleanup_highmap(void)
282 {
283 }
284 #endif
285
reserve_brk(void)286 static void __init reserve_brk(void)
287 {
288 if (_brk_end > _brk_start)
289 memblock_reserve(__pa_symbol(_brk_start),
290 _brk_end - _brk_start);
291
292 /* Mark brk area as locked down and no longer taking any
293 new allocations */
294 _brk_start = 0;
295 }
296
297 #ifdef CONFIG_BLK_DEV_INITRD
298
get_ramdisk_image(void)299 static u64 __init get_ramdisk_image(void)
300 {
301 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
302
303 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
304
305 if (ramdisk_image == 0)
306 ramdisk_image = phys_initrd_start;
307
308 return ramdisk_image;
309 }
get_ramdisk_size(void)310 static u64 __init get_ramdisk_size(void)
311 {
312 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
313
314 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
315
316 if (ramdisk_size == 0)
317 ramdisk_size = phys_initrd_size;
318
319 return ramdisk_size;
320 }
321
relocate_initrd(void)322 static void __init relocate_initrd(void)
323 {
324 /* Assume only end is not page aligned */
325 u64 ramdisk_image = get_ramdisk_image();
326 u64 ramdisk_size = get_ramdisk_size();
327 u64 area_size = PAGE_ALIGN(ramdisk_size);
328 int ret = 0;
329
330 /* We need to move the initrd down into directly mapped mem */
331 u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
332 PFN_PHYS(max_pfn_mapped));
333 if (!relocated_ramdisk)
334 panic("Cannot find place for new RAMDISK of size %lld\n",
335 ramdisk_size);
336
337 initrd_start = relocated_ramdisk + PAGE_OFFSET;
338 initrd_end = initrd_start + ramdisk_size;
339 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
340 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
341
342 ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
343 if (ret)
344 panic("Copy RAMDISK failed\n");
345
346 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
347 " [mem %#010llx-%#010llx]\n",
348 ramdisk_image, ramdisk_image + ramdisk_size - 1,
349 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
350 }
351
early_reserve_initrd(void)352 static void __init early_reserve_initrd(void)
353 {
354 /* Assume only end is not page aligned */
355 u64 ramdisk_image = get_ramdisk_image();
356 u64 ramdisk_size = get_ramdisk_size();
357 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
358
359 if (!boot_params.hdr.type_of_loader ||
360 !ramdisk_image || !ramdisk_size)
361 return; /* No initrd provided by bootloader */
362
363 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
364 }
365
reserve_initrd(void)366 static void __init reserve_initrd(void)
367 {
368 /* Assume only end is not page aligned */
369 u64 ramdisk_image = get_ramdisk_image();
370 u64 ramdisk_size = get_ramdisk_size();
371 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
372
373 if (!boot_params.hdr.type_of_loader ||
374 !ramdisk_image || !ramdisk_size)
375 return; /* No initrd provided by bootloader */
376
377 initrd_start = 0;
378
379 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
380 ramdisk_end - 1);
381
382 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
383 PFN_DOWN(ramdisk_end))) {
384 /* All are mapped, easy case */
385 initrd_start = ramdisk_image + PAGE_OFFSET;
386 initrd_end = initrd_start + ramdisk_size;
387 return;
388 }
389
390 relocate_initrd();
391
392 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
393 }
394
395 #else
early_reserve_initrd(void)396 static void __init early_reserve_initrd(void)
397 {
398 }
reserve_initrd(void)399 static void __init reserve_initrd(void)
400 {
401 }
402 #endif /* CONFIG_BLK_DEV_INITRD */
403
add_early_ima_buffer(u64 phys_addr)404 static void __init add_early_ima_buffer(u64 phys_addr)
405 {
406 #ifdef CONFIG_IMA
407 struct ima_setup_data *data;
408
409 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
410 if (!data) {
411 pr_warn("setup: failed to memremap ima_setup_data entry\n");
412 return;
413 }
414
415 if (data->size) {
416 memblock_reserve(data->addr, data->size);
417 ima_kexec_buffer_phys = data->addr;
418 ima_kexec_buffer_size = data->size;
419 }
420
421 early_memunmap(data, sizeof(*data));
422 #else
423 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
424 #endif
425 }
426
427 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
ima_free_kexec_buffer(void)428 int __init ima_free_kexec_buffer(void)
429 {
430 if (!ima_kexec_buffer_size)
431 return -ENOENT;
432
433 memblock_free_late(ima_kexec_buffer_phys,
434 ima_kexec_buffer_size);
435
436 ima_kexec_buffer_phys = 0;
437 ima_kexec_buffer_size = 0;
438
439 return 0;
440 }
441
ima_get_kexec_buffer(void ** addr,size_t * size)442 int __init ima_get_kexec_buffer(void **addr, size_t *size)
443 {
444 if (!ima_kexec_buffer_size)
445 return -ENOENT;
446
447 *addr = __va(ima_kexec_buffer_phys);
448 *size = ima_kexec_buffer_size;
449
450 return 0;
451 }
452 #endif
453
parse_setup_data(void)454 static void __init parse_setup_data(void)
455 {
456 struct setup_data *data;
457 u64 pa_data, pa_next;
458
459 pa_data = boot_params.hdr.setup_data;
460 while (pa_data) {
461 u32 data_len, data_type;
462
463 data = early_memremap(pa_data, sizeof(*data));
464 data_len = data->len + sizeof(struct setup_data);
465 data_type = data->type;
466 pa_next = data->next;
467 early_memunmap(data, sizeof(*data));
468
469 switch (data_type) {
470 case SETUP_E820_EXT:
471 e820__memory_setup_extended(pa_data, data_len);
472 break;
473 case SETUP_DTB:
474 add_dtb(pa_data);
475 break;
476 case SETUP_EFI:
477 parse_efi_setup(pa_data, data_len);
478 break;
479 case SETUP_IMA:
480 add_early_ima_buffer(pa_data);
481 break;
482 case SETUP_RNG_SEED:
483 data = early_memremap(pa_data, data_len);
484 add_bootloader_randomness(data->data, data->len);
485 /* Zero seed for forward secrecy. */
486 memzero_explicit(data->data, data->len);
487 /* Zero length in case we find ourselves back here by accident. */
488 memzero_explicit(&data->len, sizeof(data->len));
489 early_memunmap(data, data_len);
490 break;
491 default:
492 break;
493 }
494 pa_data = pa_next;
495 }
496 }
497
498 /*
499 * Translate the fields of 'struct boot_param' into global variables
500 * representing these parameters.
501 */
parse_boot_params(void)502 static void __init parse_boot_params(void)
503 {
504 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
505 screen_info = boot_params.screen_info;
506 edid_info = boot_params.edid_info;
507 #ifdef CONFIG_X86_32
508 apm_info.bios = boot_params.apm_bios_info;
509 ist_info = boot_params.ist_info;
510 #endif
511 saved_video_mode = boot_params.hdr.vid_mode;
512 bootloader_type = boot_params.hdr.type_of_loader;
513 if ((bootloader_type >> 4) == 0xe) {
514 bootloader_type &= 0xf;
515 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
516 }
517 bootloader_version = bootloader_type & 0xf;
518 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
519
520 #ifdef CONFIG_BLK_DEV_RAM
521 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
522 #endif
523 #ifdef CONFIG_EFI
524 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
525 EFI32_LOADER_SIGNATURE, 4)) {
526 set_bit(EFI_BOOT, &efi.flags);
527 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
528 EFI64_LOADER_SIGNATURE, 4)) {
529 set_bit(EFI_BOOT, &efi.flags);
530 set_bit(EFI_64BIT, &efi.flags);
531 }
532 #endif
533
534 if (!boot_params.hdr.root_flags)
535 root_mountflags &= ~MS_RDONLY;
536 }
537
memblock_x86_reserve_range_setup_data(void)538 static void __init memblock_x86_reserve_range_setup_data(void)
539 {
540 struct setup_indirect *indirect;
541 struct setup_data *data;
542 u64 pa_data, pa_next;
543 u32 len;
544
545 pa_data = boot_params.hdr.setup_data;
546 while (pa_data) {
547 data = early_memremap(pa_data, sizeof(*data));
548 if (!data) {
549 pr_warn("setup: failed to memremap setup_data entry\n");
550 return;
551 }
552
553 len = sizeof(*data);
554 pa_next = data->next;
555
556 memblock_reserve(pa_data, sizeof(*data) + data->len);
557
558 if (data->type == SETUP_INDIRECT) {
559 len += data->len;
560 early_memunmap(data, sizeof(*data));
561 data = early_memremap(pa_data, len);
562 if (!data) {
563 pr_warn("setup: failed to memremap indirect setup_data\n");
564 return;
565 }
566
567 indirect = (struct setup_indirect *)data->data;
568
569 if (indirect->type != SETUP_INDIRECT)
570 memblock_reserve(indirect->addr, indirect->len);
571 }
572
573 pa_data = pa_next;
574 early_memunmap(data, len);
575 }
576 }
577
arch_reserve_crashkernel(void)578 static void __init arch_reserve_crashkernel(void)
579 {
580 unsigned long long crash_base, crash_size, low_size = 0;
581 bool high = false;
582 int ret;
583
584 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
585 return;
586
587 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
588 &crash_size, &crash_base,
589 &low_size, &high);
590 if (ret)
591 return;
592
593 if (xen_pv_domain()) {
594 pr_info("Ignoring crashkernel for a Xen PV domain\n");
595 return;
596 }
597
598 reserve_crashkernel_generic(crash_size, crash_base, low_size, high);
599 }
600
601 static struct resource standard_io_resources[] = {
602 { .name = "dma1", .start = 0x00, .end = 0x1f,
603 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
604 { .name = "pic1", .start = 0x20, .end = 0x21,
605 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
606 { .name = "timer0", .start = 0x40, .end = 0x43,
607 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
608 { .name = "timer1", .start = 0x50, .end = 0x53,
609 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
610 { .name = "keyboard", .start = 0x60, .end = 0x60,
611 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
612 { .name = "keyboard", .start = 0x64, .end = 0x64,
613 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
614 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
615 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
616 { .name = "pic2", .start = 0xa0, .end = 0xa1,
617 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
618 { .name = "dma2", .start = 0xc0, .end = 0xdf,
619 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
620 { .name = "fpu", .start = 0xf0, .end = 0xff,
621 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
622 };
623
reserve_standard_io_resources(void)624 void __init reserve_standard_io_resources(void)
625 {
626 int i;
627
628 /* request I/O space for devices used on all i[345]86 PCs */
629 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
630 request_resource(&ioport_resource, &standard_io_resources[i]);
631
632 }
633
setup_kernel_resources(void)634 static void __init setup_kernel_resources(void)
635 {
636 code_resource.start = __pa_symbol(_text);
637 code_resource.end = __pa_symbol(_etext)-1;
638 rodata_resource.start = __pa_symbol(__start_rodata);
639 rodata_resource.end = __pa_symbol(__end_rodata)-1;
640 data_resource.start = __pa_symbol(_sdata);
641 data_resource.end = __pa_symbol(_edata)-1;
642 bss_resource.start = __pa_symbol(__bss_start);
643 bss_resource.end = __pa_symbol(__bss_stop)-1;
644
645 insert_resource(&iomem_resource, &code_resource);
646 insert_resource(&iomem_resource, &rodata_resource);
647 insert_resource(&iomem_resource, &data_resource);
648 insert_resource(&iomem_resource, &bss_resource);
649 }
650
snb_gfx_workaround_needed(void)651 static bool __init snb_gfx_workaround_needed(void)
652 {
653 #ifdef CONFIG_PCI
654 int i;
655 u16 vendor, devid;
656 static const __initconst u16 snb_ids[] = {
657 0x0102,
658 0x0112,
659 0x0122,
660 0x0106,
661 0x0116,
662 0x0126,
663 0x010a,
664 };
665
666 /* Assume no if something weird is going on with PCI */
667 if (!early_pci_allowed())
668 return false;
669
670 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
671 if (vendor != 0x8086)
672 return false;
673
674 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
675 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
676 if (devid == snb_ids[i])
677 return true;
678 #endif
679
680 return false;
681 }
682
683 /*
684 * Sandy Bridge graphics has trouble with certain ranges, exclude
685 * them from allocation.
686 */
trim_snb_memory(void)687 static void __init trim_snb_memory(void)
688 {
689 static const __initconst unsigned long bad_pages[] = {
690 0x20050000,
691 0x20110000,
692 0x20130000,
693 0x20138000,
694 0x40004000,
695 };
696 int i;
697
698 if (!snb_gfx_workaround_needed())
699 return;
700
701 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
702
703 /*
704 * SandyBridge integrated graphics devices have a bug that prevents
705 * them from accessing certain memory ranges, namely anything below
706 * 1M and in the pages listed in bad_pages[] above.
707 *
708 * To avoid these pages being ever accessed by SNB gfx devices reserve
709 * bad_pages that have not already been reserved at boot time.
710 * All memory below the 1 MB mark is anyway reserved later during
711 * setup_arch(), so there is no need to reserve it here.
712 */
713
714 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
715 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
716 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
717 bad_pages[i]);
718 }
719 }
720
trim_bios_range(void)721 static void __init trim_bios_range(void)
722 {
723 /*
724 * A special case is the first 4Kb of memory;
725 * This is a BIOS owned area, not kernel ram, but generally
726 * not listed as such in the E820 table.
727 *
728 * This typically reserves additional memory (64KiB by default)
729 * since some BIOSes are known to corrupt low memory. See the
730 * Kconfig help text for X86_RESERVE_LOW.
731 */
732 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
733
734 /*
735 * special case: Some BIOSes report the PC BIOS
736 * area (640Kb -> 1Mb) as RAM even though it is not.
737 * take them out.
738 */
739 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
740
741 e820__update_table(e820_table);
742 }
743
744 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)745 static void __init e820_add_kernel_range(void)
746 {
747 u64 start = __pa_symbol(_text);
748 u64 size = __pa_symbol(_end) - start;
749
750 /*
751 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
752 * attempt to fix it by adding the range. We may have a confused BIOS,
753 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
754 * exclude kernel range. If we really are running on top non-RAM,
755 * we will crash later anyways.
756 */
757 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
758 return;
759
760 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
761 e820__range_remove(start, size, E820_TYPE_RAM, 0);
762 e820__range_add(start, size, E820_TYPE_RAM);
763 }
764
early_reserve_memory(void)765 static void __init early_reserve_memory(void)
766 {
767 /*
768 * Reserve the memory occupied by the kernel between _text and
769 * __end_of_kernel_reserve symbols. Any kernel sections after the
770 * __end_of_kernel_reserve symbol must be explicitly reserved with a
771 * separate memblock_reserve() or they will be discarded.
772 */
773 memblock_reserve(__pa_symbol(_text),
774 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
775
776 /*
777 * The first 4Kb of memory is a BIOS owned area, but generally it is
778 * not listed as such in the E820 table.
779 *
780 * Reserve the first 64K of memory since some BIOSes are known to
781 * corrupt low memory. After the real mode trampoline is allocated the
782 * rest of the memory below 640k is reserved.
783 *
784 * In addition, make sure page 0 is always reserved because on
785 * systems with L1TF its contents can be leaked to user processes.
786 */
787 memblock_reserve(0, SZ_64K);
788
789 early_reserve_initrd();
790
791 memblock_x86_reserve_range_setup_data();
792
793 reserve_bios_regions();
794 trim_snb_memory();
795 }
796
797 /*
798 * Dump out kernel offset information on panic.
799 */
800 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)801 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
802 {
803 if (kaslr_enabled()) {
804 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
805 kaslr_offset(),
806 __START_KERNEL,
807 __START_KERNEL_map,
808 MODULES_VADDR-1);
809 } else {
810 pr_emerg("Kernel Offset: disabled\n");
811 }
812
813 return 0;
814 }
815
x86_configure_nx(void)816 void x86_configure_nx(void)
817 {
818 if (boot_cpu_has(X86_FEATURE_NX))
819 __supported_pte_mask |= _PAGE_NX;
820 else
821 __supported_pte_mask &= ~_PAGE_NX;
822 }
823
x86_report_nx(void)824 static void __init x86_report_nx(void)
825 {
826 if (!boot_cpu_has(X86_FEATURE_NX)) {
827 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
828 "missing in CPU!\n");
829 } else {
830 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
831 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
832 #else
833 /* 32bit non-PAE kernel, NX cannot be used */
834 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
835 "cannot be enabled: non-PAE kernel!\n");
836 #endif
837 }
838 }
839
840 /*
841 * Determine if we were loaded by an EFI loader. If so, then we have also been
842 * passed the efi memmap, systab, etc., so we should use these data structures
843 * for initialization. Note, the efi init code path is determined by the
844 * global efi_enabled. This allows the same kernel image to be used on existing
845 * systems (with a traditional BIOS) as well as on EFI systems.
846 */
847 /*
848 * setup_arch - architecture-specific boot-time initializations
849 *
850 * Note: On x86_64, fixmaps are ready for use even before this is called.
851 */
852
setup_arch(char ** cmdline_p)853 void __init setup_arch(char **cmdline_p)
854 {
855 #ifdef CONFIG_X86_32
856 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
857
858 /*
859 * copy kernel address range established so far and switch
860 * to the proper swapper page table
861 */
862 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
863 initial_page_table + KERNEL_PGD_BOUNDARY,
864 KERNEL_PGD_PTRS);
865
866 load_cr3(swapper_pg_dir);
867 /*
868 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
869 * a cr3 based tlb flush, so the following __flush_tlb_all()
870 * will not flush anything because the CPU quirk which clears
871 * X86_FEATURE_PGE has not been invoked yet. Though due to the
872 * load_cr3() above the TLB has been flushed already. The
873 * quirk is invoked before subsequent calls to __flush_tlb_all()
874 * so proper operation is guaranteed.
875 */
876 __flush_tlb_all();
877 #else
878 printk(KERN_INFO "Command line: %s\n", boot_command_line);
879 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
880 #endif
881
882 #ifdef CONFIG_CMDLINE_BOOL
883 #ifdef CONFIG_CMDLINE_OVERRIDE
884 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
885 #else
886 if (builtin_cmdline[0]) {
887 /* append boot loader cmdline to builtin */
888 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
889 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
890 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
891 }
892 #endif
893 builtin_cmdline_added = true;
894 #endif
895
896 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
897 *cmdline_p = command_line;
898
899 /*
900 * If we have OLPC OFW, we might end up relocating the fixmap due to
901 * reserve_top(), so do this before touching the ioremap area.
902 */
903 olpc_ofw_detect();
904
905 idt_setup_early_traps();
906 early_cpu_init();
907 jump_label_init();
908 static_call_init();
909 early_ioremap_init();
910
911 setup_olpc_ofw_pgd();
912
913 parse_boot_params();
914
915 x86_init.oem.arch_setup();
916
917 /*
918 * Do some memory reservations *before* memory is added to memblock, so
919 * memblock allocations won't overwrite it.
920 *
921 * After this point, everything still needed from the boot loader or
922 * firmware or kernel text should be early reserved or marked not RAM in
923 * e820. All other memory is free game.
924 *
925 * This call needs to happen before e820__memory_setup() which calls the
926 * xen_memory_setup() on Xen dom0 which relies on the fact that those
927 * early reservations have happened already.
928 */
929 early_reserve_memory();
930
931 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
932 e820__memory_setup();
933 parse_setup_data();
934
935 copy_edd();
936
937 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
938
939 /*
940 * x86_configure_nx() is called before parse_early_param() to detect
941 * whether hardware doesn't support NX (so that the early EHCI debug
942 * console setup can safely call set_fixmap()).
943 */
944 x86_configure_nx();
945
946 parse_early_param();
947
948 if (efi_enabled(EFI_BOOT))
949 efi_memblock_x86_reserve_range();
950
951 x86_report_nx();
952
953 apic_setup_apic_calls();
954
955 if (acpi_mps_check()) {
956 #ifdef CONFIG_X86_LOCAL_APIC
957 apic_is_disabled = true;
958 #endif
959 setup_clear_cpu_cap(X86_FEATURE_APIC);
960 }
961
962 e820__finish_early_params();
963
964 if (efi_enabled(EFI_BOOT))
965 efi_init();
966
967 reserve_ibft_region();
968 x86_init.resources.dmi_setup();
969
970 /*
971 * VMware detection requires dmi to be available, so this
972 * needs to be done after dmi_setup(), for the boot CPU.
973 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be
974 * called before cache_bp_init() for setting up MTRR state.
975 */
976 init_hypervisor_platform();
977
978 tsc_early_init();
979 x86_init.resources.probe_roms();
980
981 /*
982 * Add resources for kernel text and data to the iomem_resource.
983 * Do it after parse_early_param, so it can be debugged.
984 */
985 setup_kernel_resources();
986
987 e820_add_kernel_range();
988 trim_bios_range();
989 #ifdef CONFIG_X86_32
990 if (ppro_with_ram_bug()) {
991 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
992 E820_TYPE_RESERVED);
993 e820__update_table(e820_table);
994 printk(KERN_INFO "fixed physical RAM map:\n");
995 e820__print_table("bad_ppro");
996 }
997 #else
998 early_gart_iommu_check();
999 #endif
1000
1001 /*
1002 * partially used pages are not usable - thus
1003 * we are rounding upwards:
1004 */
1005 max_pfn = e820__end_of_ram_pfn();
1006
1007 /* update e820 for memory not covered by WB MTRRs */
1008 cache_bp_init();
1009 if (mtrr_trim_uncached_memory(max_pfn))
1010 max_pfn = e820__end_of_ram_pfn();
1011
1012 max_possible_pfn = max_pfn;
1013
1014 /*
1015 * Define random base addresses for memory sections after max_pfn is
1016 * defined and before each memory section base is used.
1017 */
1018 kernel_randomize_memory();
1019
1020 #ifdef CONFIG_X86_32
1021 /* max_low_pfn get updated here */
1022 find_low_pfn_range();
1023 #else
1024 check_x2apic();
1025
1026 /* How many end-of-memory variables you have, grandma! */
1027 /* need this before calling reserve_initrd */
1028 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1029 max_low_pfn = e820__end_of_low_ram_pfn();
1030 else
1031 max_low_pfn = max_pfn;
1032 #endif
1033
1034 /* Find and reserve MPTABLE area */
1035 x86_init.mpparse.find_mptable();
1036
1037 early_alloc_pgt_buf();
1038
1039 /*
1040 * Need to conclude brk, before e820__memblock_setup()
1041 * it could use memblock_find_in_range, could overlap with
1042 * brk area.
1043 */
1044 reserve_brk();
1045
1046 cleanup_highmap();
1047
1048 e820__memblock_setup();
1049
1050 /*
1051 * Needs to run after memblock setup because it needs the physical
1052 * memory size.
1053 */
1054 mem_encrypt_setup_arch();
1055 cc_random_init();
1056
1057 efi_find_mirror();
1058 efi_esrt_init();
1059 efi_mokvar_table_init();
1060
1061 /*
1062 * The EFI specification says that boot service code won't be
1063 * called after ExitBootServices(). This is, in fact, a lie.
1064 */
1065 efi_reserve_boot_services();
1066
1067 /* preallocate 4k for mptable mpc */
1068 e820__memblock_alloc_reserved_mpc_new();
1069
1070 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1071 setup_bios_corruption_check();
1072 #endif
1073
1074 #ifdef CONFIG_X86_32
1075 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1076 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1077 #endif
1078
1079 /*
1080 * Find free memory for the real mode trampoline and place it there. If
1081 * there is not enough free memory under 1M, on EFI-enabled systems
1082 * there will be additional attempt to reclaim the memory for the real
1083 * mode trampoline at efi_free_boot_services().
1084 *
1085 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1086 * are known to corrupt low memory and several hundred kilobytes are not
1087 * worth complex detection what memory gets clobbered. Windows does the
1088 * same thing for very similar reasons.
1089 *
1090 * Moreover, on machines with SandyBridge graphics or in setups that use
1091 * crashkernel the entire 1M is reserved anyway.
1092 *
1093 * Note the host kernel TDX also requires the first 1MB being reserved.
1094 */
1095 x86_platform.realmode_reserve();
1096
1097 init_mem_mapping();
1098
1099 /*
1100 * init_mem_mapping() relies on the early IDT page fault handling.
1101 * Now either enable FRED or install the real page fault handler
1102 * for 64-bit in the IDT.
1103 */
1104 cpu_init_replace_early_idt();
1105
1106 /*
1107 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1108 * with the current CR4 value. This may not be necessary, but
1109 * auditing all the early-boot CR4 manipulation would be needed to
1110 * rule it out.
1111 *
1112 * Mask off features that don't work outside long mode (just
1113 * PCIDE for now).
1114 */
1115 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1116
1117 memblock_set_current_limit(get_max_mapped());
1118
1119 /*
1120 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1121 */
1122
1123 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1124 if (init_ohci1394_dma_early)
1125 init_ohci1394_dma_on_all_controllers();
1126 #endif
1127 /* Allocate bigger log buffer */
1128 setup_log_buf(1);
1129
1130 if (efi_enabled(EFI_BOOT)) {
1131 switch (boot_params.secure_boot) {
1132 case efi_secureboot_mode_disabled:
1133 pr_info("Secure boot disabled\n");
1134 break;
1135 case efi_secureboot_mode_enabled:
1136 pr_info("Secure boot enabled\n");
1137 break;
1138 default:
1139 pr_info("Secure boot could not be determined\n");
1140 break;
1141 }
1142 }
1143
1144 reserve_initrd();
1145
1146 acpi_table_upgrade();
1147 /* Look for ACPI tables and reserve memory occupied by them. */
1148 acpi_boot_table_init();
1149
1150 vsmp_init();
1151
1152 io_delay_init();
1153
1154 early_platform_quirks();
1155
1156 /* Some platforms need the APIC registered for NUMA configuration */
1157 early_acpi_boot_init();
1158 x86_init.mpparse.early_parse_smp_cfg();
1159
1160 x86_flattree_get_config();
1161
1162 initmem_init();
1163 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1164
1165 if (boot_cpu_has(X86_FEATURE_GBPAGES)) {
1166 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1167 hugetlb_bootmem_alloc();
1168 }
1169
1170 /*
1171 * Reserve memory for crash kernel after SRAT is parsed so that it
1172 * won't consume hotpluggable memory.
1173 */
1174 arch_reserve_crashkernel();
1175
1176 if (!early_xdbc_setup_hardware())
1177 early_xdbc_register_console();
1178
1179 x86_init.paging.pagetable_init();
1180
1181 kasan_init();
1182
1183 /*
1184 * Sync back kernel address range.
1185 *
1186 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1187 * this call?
1188 */
1189 sync_initial_page_table();
1190
1191 tboot_probe();
1192
1193 map_vsyscall();
1194
1195 x86_32_probe_apic();
1196
1197 early_quirks();
1198
1199 topology_apply_cmdline_limits_early();
1200
1201 /*
1202 * Parse SMP configuration. Try ACPI first and then the platform
1203 * specific parser.
1204 */
1205 acpi_boot_init();
1206 x86_init.mpparse.parse_smp_cfg();
1207
1208 /* Last opportunity to detect and map the local APIC */
1209 init_apic_mappings();
1210
1211 topology_init_possible_cpus();
1212
1213 init_cpu_to_node();
1214 init_gi_nodes();
1215
1216 io_apic_init_mappings();
1217
1218 x86_init.hyper.guest_late_init();
1219
1220 e820__reserve_resources();
1221 e820__register_nosave_regions(max_pfn);
1222
1223 x86_init.resources.reserve_resources();
1224
1225 e820__setup_pci_gap();
1226
1227 #ifdef CONFIG_VT
1228 #if defined(CONFIG_VGA_CONSOLE)
1229 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1230 vgacon_register_screen(&screen_info);
1231 #endif
1232 #endif
1233 x86_init.oem.banner();
1234
1235 x86_init.timers.wallclock_init();
1236
1237 /*
1238 * This needs to run before setup_local_APIC() which soft-disables the
1239 * local APIC temporarily and that masks the thermal LVT interrupt,
1240 * leading to softlockups on machines which have configured SMI
1241 * interrupt delivery.
1242 */
1243 therm_lvt_init();
1244
1245 mcheck_init();
1246
1247 register_refined_jiffies(CLOCK_TICK_RATE);
1248
1249 #ifdef CONFIG_EFI
1250 if (efi_enabled(EFI_BOOT))
1251 efi_apply_memmap_quirks();
1252 #endif
1253
1254 unwind_init();
1255 }
1256
1257 #ifdef CONFIG_X86_32
1258
1259 static struct resource video_ram_resource = {
1260 .name = "Video RAM area",
1261 .start = 0xa0000,
1262 .end = 0xbffff,
1263 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1264 };
1265
i386_reserve_resources(void)1266 void __init i386_reserve_resources(void)
1267 {
1268 request_resource(&iomem_resource, &video_ram_resource);
1269 reserve_standard_io_resources();
1270 }
1271
1272 #endif /* CONFIG_X86_32 */
1273
1274 static struct notifier_block kernel_offset_notifier = {
1275 .notifier_call = dump_kernel_offset
1276 };
1277
register_kernel_offset_dumper(void)1278 static int __init register_kernel_offset_dumper(void)
1279 {
1280 atomic_notifier_chain_register(&panic_notifier_list,
1281 &kernel_offset_notifier);
1282 return 0;
1283 }
1284 __initcall(register_kernel_offset_dumper);
1285
1286 #ifdef CONFIG_HOTPLUG_CPU
arch_cpu_is_hotpluggable(int cpu)1287 bool arch_cpu_is_hotpluggable(int cpu)
1288 {
1289 return cpu > 0;
1290 }
1291 #endif /* CONFIG_HOTPLUG_CPU */
1292