1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 1995 Linus Torvalds
4 *
5 * This file contains the setup_arch() code, which handles the architecture-dependent
6 * parts of early kernel initialization.
7 */
8 #include <linux/acpi.h>
9 #include <linux/console.h>
10 #include <linux/crash_dump.h>
11 #include <linux/dma-map-ops.h>
12 #include <linux/dmi.h>
13 #include <linux/efi.h>
14 #include <linux/init_ohci1394_dma.h>
15 #include <linux/initrd.h>
16 #include <linux/iscsi_ibft.h>
17 #include <linux/memblock.h>
18 #include <linux/panic_notifier.h>
19 #include <linux/pci.h>
20 #include <linux/root_dev.h>
21 #include <linux/hugetlb.h>
22 #include <linux/tboot.h>
23 #include <linux/usb/xhci-dbgp.h>
24 #include <linux/static_call.h>
25 #include <linux/swiotlb.h>
26
27 #include <uapi/linux/mount.h>
28
29 #include <xen/xen.h>
30
31 #include <asm/apic.h>
32 #include <asm/numa.h>
33 #include <asm/bios_ebda.h>
34 #include <asm/bugs.h>
35 #include <asm/cpu.h>
36 #include <asm/efi.h>
37 #include <asm/gart.h>
38 #include <asm/hypervisor.h>
39 #include <asm/io_apic.h>
40 #include <asm/kasan.h>
41 #include <asm/kaslr.h>
42 #include <asm/mce.h>
43 #include <asm/memtype.h>
44 #include <asm/mtrr.h>
45 #include <asm/realmode.h>
46 #include <asm/olpc_ofw.h>
47 #include <asm/pci-direct.h>
48 #include <asm/prom.h>
49 #include <asm/proto.h>
50 #include <asm/thermal.h>
51 #include <asm/unwind.h>
52 #include <asm/vsyscall.h>
53 #include <linux/vmalloc.h>
54
55 /*
56 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
57 * max_pfn_mapped: highest directly mapped pfn > 4 GB
58 *
59 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
60 * represented by pfn_mapped[].
61 */
62 unsigned long max_low_pfn_mapped;
63 unsigned long max_pfn_mapped;
64
65 #ifdef CONFIG_DMI
66 RESERVE_BRK(dmi_alloc, 65536);
67 #endif
68
69
70 unsigned long _brk_start = (unsigned long)__brk_base;
71 unsigned long _brk_end = (unsigned long)__brk_base;
72
73 struct boot_params boot_params;
74
75 /*
76 * These are the four main kernel memory regions, we put them into
77 * the resource tree so that kdump tools and other debugging tools
78 * recover it:
79 */
80
81 static struct resource rodata_resource = {
82 .name = "Kernel rodata",
83 .start = 0,
84 .end = 0,
85 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
86 };
87
88 static struct resource data_resource = {
89 .name = "Kernel data",
90 .start = 0,
91 .end = 0,
92 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
93 };
94
95 static struct resource code_resource = {
96 .name = "Kernel code",
97 .start = 0,
98 .end = 0,
99 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
100 };
101
102 static struct resource bss_resource = {
103 .name = "Kernel bss",
104 .start = 0,
105 .end = 0,
106 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
107 };
108
109
110 #ifdef CONFIG_X86_32
111 /* CPU data as detected by the assembly code in head_32.S */
112 struct cpuinfo_x86 new_cpu_data;
113
114 /* Common CPU data for all CPUs */
115 struct cpuinfo_x86 boot_cpu_data __read_mostly;
116 EXPORT_SYMBOL(boot_cpu_data);
117
118 unsigned int def_to_bigsmp;
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 #else
132 struct cpuinfo_x86 boot_cpu_data __read_mostly;
133 EXPORT_SYMBOL(boot_cpu_data);
134 #endif
135
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 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
144 int bootloader_type, bootloader_version;
145
146 /*
147 * Setup options
148 */
149 struct screen_info screen_info;
150 EXPORT_SYMBOL(screen_info);
151 struct edid_info edid_info;
152 EXPORT_SYMBOL_GPL(edid_info);
153
154 extern int root_mountflags;
155
156 unsigned long saved_video_mode;
157
158 #define RAMDISK_IMAGE_START_MASK 0x07FF
159 #define RAMDISK_PROMPT_FLAG 0x8000
160 #define RAMDISK_LOAD_FLAG 0x4000
161
162 static char __initdata command_line[COMMAND_LINE_SIZE];
163 #ifdef CONFIG_CMDLINE_BOOL
164 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
165 #endif
166
167 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
168 struct edd edd;
169 #ifdef CONFIG_EDD_MODULE
170 EXPORT_SYMBOL(edd);
171 #endif
172 /**
173 * copy_edd() - Copy the BIOS EDD information
174 * from boot_params into a safe place.
175 *
176 */
copy_edd(void)177 static inline void __init copy_edd(void)
178 {
179 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
180 sizeof(edd.mbr_signature));
181 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
182 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
183 edd.edd_info_nr = boot_params.eddbuf_entries;
184 }
185 #else
copy_edd(void)186 static inline void __init copy_edd(void)
187 {
188 }
189 #endif
190
extend_brk(size_t size,size_t align)191 void * __init extend_brk(size_t size, size_t align)
192 {
193 size_t mask = align - 1;
194 void *ret;
195
196 BUG_ON(_brk_start == 0);
197 BUG_ON(align & mask);
198
199 _brk_end = (_brk_end + mask) & ~mask;
200 BUG_ON((char *)(_brk_end + size) > __brk_limit);
201
202 ret = (void *)_brk_end;
203 _brk_end += size;
204
205 memset(ret, 0, size);
206
207 return ret;
208 }
209
210 #ifdef CONFIG_X86_32
cleanup_highmap(void)211 static void __init cleanup_highmap(void)
212 {
213 }
214 #endif
215
reserve_brk(void)216 static void __init reserve_brk(void)
217 {
218 if (_brk_end > _brk_start)
219 memblock_reserve(__pa_symbol(_brk_start),
220 _brk_end - _brk_start);
221
222 /* Mark brk area as locked down and no longer taking any
223 new allocations */
224 _brk_start = 0;
225 }
226
227 u64 relocated_ramdisk;
228
229 #ifdef CONFIG_BLK_DEV_INITRD
230
get_ramdisk_image(void)231 static u64 __init get_ramdisk_image(void)
232 {
233 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
234
235 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
236
237 if (ramdisk_image == 0)
238 ramdisk_image = phys_initrd_start;
239
240 return ramdisk_image;
241 }
get_ramdisk_size(void)242 static u64 __init get_ramdisk_size(void)
243 {
244 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
245
246 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
247
248 if (ramdisk_size == 0)
249 ramdisk_size = phys_initrd_size;
250
251 return ramdisk_size;
252 }
253
relocate_initrd(void)254 static void __init relocate_initrd(void)
255 {
256 /* Assume only end is not page aligned */
257 u64 ramdisk_image = get_ramdisk_image();
258 u64 ramdisk_size = get_ramdisk_size();
259 u64 area_size = PAGE_ALIGN(ramdisk_size);
260
261 /* We need to move the initrd down into directly mapped mem */
262 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
263 PFN_PHYS(max_pfn_mapped));
264 if (!relocated_ramdisk)
265 panic("Cannot find place for new RAMDISK of size %lld\n",
266 ramdisk_size);
267
268 initrd_start = relocated_ramdisk + PAGE_OFFSET;
269 initrd_end = initrd_start + ramdisk_size;
270 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
271 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
272
273 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
274
275 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
276 " [mem %#010llx-%#010llx]\n",
277 ramdisk_image, ramdisk_image + ramdisk_size - 1,
278 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
279 }
280
early_reserve_initrd(void)281 static void __init early_reserve_initrd(void)
282 {
283 /* Assume only end is not page aligned */
284 u64 ramdisk_image = get_ramdisk_image();
285 u64 ramdisk_size = get_ramdisk_size();
286 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
287
288 if (!boot_params.hdr.type_of_loader ||
289 !ramdisk_image || !ramdisk_size)
290 return; /* No initrd provided by bootloader */
291
292 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
293 }
294
reserve_initrd(void)295 static void __init reserve_initrd(void)
296 {
297 /* Assume only end is not page aligned */
298 u64 ramdisk_image = get_ramdisk_image();
299 u64 ramdisk_size = get_ramdisk_size();
300 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
301
302 if (!boot_params.hdr.type_of_loader ||
303 !ramdisk_image || !ramdisk_size)
304 return; /* No initrd provided by bootloader */
305
306 initrd_start = 0;
307
308 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
309 ramdisk_end - 1);
310
311 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
312 PFN_DOWN(ramdisk_end))) {
313 /* All are mapped, easy case */
314 initrd_start = ramdisk_image + PAGE_OFFSET;
315 initrd_end = initrd_start + ramdisk_size;
316 return;
317 }
318
319 relocate_initrd();
320
321 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
322 }
323
324 #else
early_reserve_initrd(void)325 static void __init early_reserve_initrd(void)
326 {
327 }
reserve_initrd(void)328 static void __init reserve_initrd(void)
329 {
330 }
331 #endif /* CONFIG_BLK_DEV_INITRD */
332
parse_setup_data(void)333 static void __init parse_setup_data(void)
334 {
335 struct setup_data *data;
336 u64 pa_data, pa_next;
337
338 pa_data = boot_params.hdr.setup_data;
339 while (pa_data) {
340 u32 data_len, data_type;
341
342 data = early_memremap(pa_data, sizeof(*data));
343 data_len = data->len + sizeof(struct setup_data);
344 data_type = data->type;
345 pa_next = data->next;
346 early_memunmap(data, sizeof(*data));
347
348 switch (data_type) {
349 case SETUP_E820_EXT:
350 e820__memory_setup_extended(pa_data, data_len);
351 break;
352 case SETUP_DTB:
353 add_dtb(pa_data);
354 break;
355 case SETUP_EFI:
356 parse_efi_setup(pa_data, data_len);
357 break;
358 default:
359 break;
360 }
361 pa_data = pa_next;
362 }
363 }
364
memblock_x86_reserve_range_setup_data(void)365 static void __init memblock_x86_reserve_range_setup_data(void)
366 {
367 struct setup_indirect *indirect;
368 struct setup_data *data;
369 u64 pa_data, pa_next;
370 u32 len;
371
372 pa_data = boot_params.hdr.setup_data;
373 while (pa_data) {
374 data = early_memremap(pa_data, sizeof(*data));
375 if (!data) {
376 pr_warn("setup: failed to memremap setup_data entry\n");
377 return;
378 }
379
380 len = sizeof(*data);
381 pa_next = data->next;
382
383 memblock_reserve(pa_data, sizeof(*data) + data->len);
384
385 if (data->type == SETUP_INDIRECT) {
386 len += data->len;
387 early_memunmap(data, sizeof(*data));
388 data = early_memremap(pa_data, len);
389 if (!data) {
390 pr_warn("setup: failed to memremap indirect setup_data\n");
391 return;
392 }
393
394 indirect = (struct setup_indirect *)data->data;
395
396 if (indirect->type != SETUP_INDIRECT)
397 memblock_reserve(indirect->addr, indirect->len);
398 }
399
400 pa_data = pa_next;
401 early_memunmap(data, len);
402 }
403 }
404
405 /*
406 * --------- Crashkernel reservation ------------------------------
407 */
408
409 /* 16M alignment for crash kernel regions */
410 #define CRASH_ALIGN SZ_16M
411
412 /*
413 * Keep the crash kernel below this limit.
414 *
415 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
416 * due to mapping restrictions.
417 *
418 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
419 * the upper limit of system RAM in 4-level paging mode. Since the kdump
420 * jump could be from 5-level paging to 4-level paging, the jump will fail if
421 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
422 * no good way to detect the paging mode of the target kernel which will be
423 * loaded for dumping.
424 */
425 #ifdef CONFIG_X86_32
426 # define CRASH_ADDR_LOW_MAX SZ_512M
427 # define CRASH_ADDR_HIGH_MAX SZ_512M
428 #else
429 # define CRASH_ADDR_LOW_MAX SZ_4G
430 # define CRASH_ADDR_HIGH_MAX SZ_64T
431 #endif
432
reserve_crashkernel_low(void)433 static int __init reserve_crashkernel_low(void)
434 {
435 #ifdef CONFIG_X86_64
436 unsigned long long base, low_base = 0, low_size = 0;
437 unsigned long low_mem_limit;
438 int ret;
439
440 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
441
442 /* crashkernel=Y,low */
443 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
444 if (ret) {
445 /*
446 * two parts from kernel/dma/swiotlb.c:
447 * -swiotlb size: user-specified with swiotlb= or default.
448 *
449 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
450 * to 8M for other buffers that may need to stay low too. Also
451 * make sure we allocate enough extra low memory so that we
452 * don't run out of DMA buffers for 32-bit devices.
453 */
454 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
455 } else {
456 /* passed with crashkernel=0,low ? */
457 if (!low_size)
458 return 0;
459 }
460
461 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
462 if (!low_base) {
463 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
464 (unsigned long)(low_size >> 20));
465 return -ENOMEM;
466 }
467
468 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
469 (unsigned long)(low_size >> 20),
470 (unsigned long)(low_base >> 20),
471 (unsigned long)(low_mem_limit >> 20));
472
473 crashk_low_res.start = low_base;
474 crashk_low_res.end = low_base + low_size - 1;
475 insert_resource(&iomem_resource, &crashk_low_res);
476 #endif
477 return 0;
478 }
479
reserve_crashkernel(void)480 static void __init reserve_crashkernel(void)
481 {
482 unsigned long long crash_size, crash_base, total_mem;
483 bool high = false;
484 int ret;
485
486 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
487 return;
488
489 total_mem = memblock_phys_mem_size();
490
491 /* crashkernel=XM */
492 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
493 if (ret != 0 || crash_size <= 0) {
494 /* crashkernel=X,high */
495 ret = parse_crashkernel_high(boot_command_line, total_mem,
496 &crash_size, &crash_base);
497 if (ret != 0 || crash_size <= 0)
498 return;
499 high = true;
500 }
501
502 if (xen_pv_domain()) {
503 pr_info("Ignoring crashkernel for a Xen PV domain\n");
504 return;
505 }
506
507 /* 0 means: find the address automatically */
508 if (!crash_base) {
509 /*
510 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
511 * crashkernel=x,high reserves memory over 4G, also allocates
512 * 256M extra low memory for DMA buffers and swiotlb.
513 * But the extra memory is not required for all machines.
514 * So try low memory first and fall back to high memory
515 * unless "crashkernel=size[KMG],high" is specified.
516 */
517 if (!high)
518 crash_base = memblock_phys_alloc_range(crash_size,
519 CRASH_ALIGN, CRASH_ALIGN,
520 CRASH_ADDR_LOW_MAX);
521 if (!crash_base)
522 crash_base = memblock_phys_alloc_range(crash_size,
523 CRASH_ALIGN, CRASH_ALIGN,
524 CRASH_ADDR_HIGH_MAX);
525 if (!crash_base) {
526 pr_info("crashkernel reservation failed - No suitable area found.\n");
527 return;
528 }
529 } else {
530 unsigned long long start;
531
532 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
533 crash_base + crash_size);
534 if (start != crash_base) {
535 pr_info("crashkernel reservation failed - memory is in use.\n");
536 return;
537 }
538 }
539
540 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
541 memblock_phys_free(crash_base, crash_size);
542 return;
543 }
544
545 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
546 (unsigned long)(crash_size >> 20),
547 (unsigned long)(crash_base >> 20),
548 (unsigned long)(total_mem >> 20));
549
550 crashk_res.start = crash_base;
551 crashk_res.end = crash_base + crash_size - 1;
552 insert_resource(&iomem_resource, &crashk_res);
553 }
554
555 static struct resource standard_io_resources[] = {
556 { .name = "dma1", .start = 0x00, .end = 0x1f,
557 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
558 { .name = "pic1", .start = 0x20, .end = 0x21,
559 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
560 { .name = "timer0", .start = 0x40, .end = 0x43,
561 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
562 { .name = "timer1", .start = 0x50, .end = 0x53,
563 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
564 { .name = "keyboard", .start = 0x60, .end = 0x60,
565 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
566 { .name = "keyboard", .start = 0x64, .end = 0x64,
567 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
568 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
569 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
570 { .name = "pic2", .start = 0xa0, .end = 0xa1,
571 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
572 { .name = "dma2", .start = 0xc0, .end = 0xdf,
573 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
574 { .name = "fpu", .start = 0xf0, .end = 0xff,
575 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
576 };
577
reserve_standard_io_resources(void)578 void __init reserve_standard_io_resources(void)
579 {
580 int i;
581
582 /* request I/O space for devices used on all i[345]86 PCs */
583 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
584 request_resource(&ioport_resource, &standard_io_resources[i]);
585
586 }
587
snb_gfx_workaround_needed(void)588 static bool __init snb_gfx_workaround_needed(void)
589 {
590 #ifdef CONFIG_PCI
591 int i;
592 u16 vendor, devid;
593 static const __initconst u16 snb_ids[] = {
594 0x0102,
595 0x0112,
596 0x0122,
597 0x0106,
598 0x0116,
599 0x0126,
600 0x010a,
601 };
602
603 /* Assume no if something weird is going on with PCI */
604 if (!early_pci_allowed())
605 return false;
606
607 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
608 if (vendor != 0x8086)
609 return false;
610
611 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
612 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
613 if (devid == snb_ids[i])
614 return true;
615 #endif
616
617 return false;
618 }
619
620 /*
621 * Sandy Bridge graphics has trouble with certain ranges, exclude
622 * them from allocation.
623 */
trim_snb_memory(void)624 static void __init trim_snb_memory(void)
625 {
626 static const __initconst unsigned long bad_pages[] = {
627 0x20050000,
628 0x20110000,
629 0x20130000,
630 0x20138000,
631 0x40004000,
632 };
633 int i;
634
635 if (!snb_gfx_workaround_needed())
636 return;
637
638 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
639
640 /*
641 * SandyBridge integrated graphics devices have a bug that prevents
642 * them from accessing certain memory ranges, namely anything below
643 * 1M and in the pages listed in bad_pages[] above.
644 *
645 * To avoid these pages being ever accessed by SNB gfx devices reserve
646 * bad_pages that have not already been reserved at boot time.
647 * All memory below the 1 MB mark is anyway reserved later during
648 * setup_arch(), so there is no need to reserve it here.
649 */
650
651 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
652 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
653 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
654 bad_pages[i]);
655 }
656 }
657
trim_bios_range(void)658 static void __init trim_bios_range(void)
659 {
660 /*
661 * A special case is the first 4Kb of memory;
662 * This is a BIOS owned area, not kernel ram, but generally
663 * not listed as such in the E820 table.
664 *
665 * This typically reserves additional memory (64KiB by default)
666 * since some BIOSes are known to corrupt low memory. See the
667 * Kconfig help text for X86_RESERVE_LOW.
668 */
669 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
670
671 /*
672 * special case: Some BIOSes report the PC BIOS
673 * area (640Kb -> 1Mb) as RAM even though it is not.
674 * take them out.
675 */
676 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
677
678 e820__update_table(e820_table);
679 }
680
681 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)682 static void __init e820_add_kernel_range(void)
683 {
684 u64 start = __pa_symbol(_text);
685 u64 size = __pa_symbol(_end) - start;
686
687 /*
688 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
689 * attempt to fix it by adding the range. We may have a confused BIOS,
690 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
691 * exclude kernel range. If we really are running on top non-RAM,
692 * we will crash later anyways.
693 */
694 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
695 return;
696
697 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
698 e820__range_remove(start, size, E820_TYPE_RAM, 0);
699 e820__range_add(start, size, E820_TYPE_RAM);
700 }
701
early_reserve_memory(void)702 static void __init early_reserve_memory(void)
703 {
704 /*
705 * Reserve the memory occupied by the kernel between _text and
706 * __end_of_kernel_reserve symbols. Any kernel sections after the
707 * __end_of_kernel_reserve symbol must be explicitly reserved with a
708 * separate memblock_reserve() or they will be discarded.
709 */
710 memblock_reserve(__pa_symbol(_text),
711 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
712
713 /*
714 * The first 4Kb of memory is a BIOS owned area, but generally it is
715 * not listed as such in the E820 table.
716 *
717 * Reserve the first 64K of memory since some BIOSes are known to
718 * corrupt low memory. After the real mode trampoline is allocated the
719 * rest of the memory below 640k is reserved.
720 *
721 * In addition, make sure page 0 is always reserved because on
722 * systems with L1TF its contents can be leaked to user processes.
723 */
724 memblock_reserve(0, SZ_64K);
725
726 early_reserve_initrd();
727
728 memblock_x86_reserve_range_setup_data();
729
730 reserve_ibft_region();
731 reserve_bios_regions();
732 trim_snb_memory();
733 }
734
735 /*
736 * Dump out kernel offset information on panic.
737 */
738 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)739 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
740 {
741 if (kaslr_enabled()) {
742 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
743 kaslr_offset(),
744 __START_KERNEL,
745 __START_KERNEL_map,
746 MODULES_VADDR-1);
747 } else {
748 pr_emerg("Kernel Offset: disabled\n");
749 }
750
751 return 0;
752 }
753
x86_configure_nx(void)754 void x86_configure_nx(void)
755 {
756 if (boot_cpu_has(X86_FEATURE_NX))
757 __supported_pte_mask |= _PAGE_NX;
758 else
759 __supported_pte_mask &= ~_PAGE_NX;
760 }
761
x86_report_nx(void)762 static void __init x86_report_nx(void)
763 {
764 if (!boot_cpu_has(X86_FEATURE_NX)) {
765 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
766 "missing in CPU!\n");
767 } else {
768 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
769 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
770 #else
771 /* 32bit non-PAE kernel, NX cannot be used */
772 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
773 "cannot be enabled: non-PAE kernel!\n");
774 #endif
775 }
776 }
777
778 /*
779 * Determine if we were loaded by an EFI loader. If so, then we have also been
780 * passed the efi memmap, systab, etc., so we should use these data structures
781 * for initialization. Note, the efi init code path is determined by the
782 * global efi_enabled. This allows the same kernel image to be used on existing
783 * systems (with a traditional BIOS) as well as on EFI systems.
784 */
785 /*
786 * setup_arch - architecture-specific boot-time initializations
787 *
788 * Note: On x86_64, fixmaps are ready for use even before this is called.
789 */
790
setup_arch(char ** cmdline_p)791 void __init setup_arch(char **cmdline_p)
792 {
793 #ifdef CONFIG_X86_32
794 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
795
796 /*
797 * copy kernel address range established so far and switch
798 * to the proper swapper page table
799 */
800 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
801 initial_page_table + KERNEL_PGD_BOUNDARY,
802 KERNEL_PGD_PTRS);
803
804 load_cr3(swapper_pg_dir);
805 /*
806 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
807 * a cr3 based tlb flush, so the following __flush_tlb_all()
808 * will not flush anything because the CPU quirk which clears
809 * X86_FEATURE_PGE has not been invoked yet. Though due to the
810 * load_cr3() above the TLB has been flushed already. The
811 * quirk is invoked before subsequent calls to __flush_tlb_all()
812 * so proper operation is guaranteed.
813 */
814 __flush_tlb_all();
815 #else
816 printk(KERN_INFO "Command line: %s\n", boot_command_line);
817 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
818 #endif
819
820 /*
821 * If we have OLPC OFW, we might end up relocating the fixmap due to
822 * reserve_top(), so do this before touching the ioremap area.
823 */
824 olpc_ofw_detect();
825
826 idt_setup_early_traps();
827 early_cpu_init();
828 jump_label_init();
829 static_call_init();
830 early_ioremap_init();
831
832 setup_olpc_ofw_pgd();
833
834 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
835 screen_info = boot_params.screen_info;
836 edid_info = boot_params.edid_info;
837 #ifdef CONFIG_X86_32
838 apm_info.bios = boot_params.apm_bios_info;
839 ist_info = boot_params.ist_info;
840 #endif
841 saved_video_mode = boot_params.hdr.vid_mode;
842 bootloader_type = boot_params.hdr.type_of_loader;
843 if ((bootloader_type >> 4) == 0xe) {
844 bootloader_type &= 0xf;
845 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
846 }
847 bootloader_version = bootloader_type & 0xf;
848 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
849
850 #ifdef CONFIG_BLK_DEV_RAM
851 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
852 #endif
853 #ifdef CONFIG_EFI
854 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
855 EFI32_LOADER_SIGNATURE, 4)) {
856 set_bit(EFI_BOOT, &efi.flags);
857 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
858 EFI64_LOADER_SIGNATURE, 4)) {
859 set_bit(EFI_BOOT, &efi.flags);
860 set_bit(EFI_64BIT, &efi.flags);
861 }
862 #endif
863
864 x86_init.oem.arch_setup();
865
866 /*
867 * Do some memory reservations *before* memory is added to memblock, so
868 * memblock allocations won't overwrite it.
869 *
870 * After this point, everything still needed from the boot loader or
871 * firmware or kernel text should be early reserved or marked not RAM in
872 * e820. All other memory is free game.
873 *
874 * This call needs to happen before e820__memory_setup() which calls the
875 * xen_memory_setup() on Xen dom0 which relies on the fact that those
876 * early reservations have happened already.
877 */
878 early_reserve_memory();
879
880 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
881 e820__memory_setup();
882 parse_setup_data();
883
884 copy_edd();
885
886 if (!boot_params.hdr.root_flags)
887 root_mountflags &= ~MS_RDONLY;
888 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
889
890 code_resource.start = __pa_symbol(_text);
891 code_resource.end = __pa_symbol(_etext)-1;
892 rodata_resource.start = __pa_symbol(__start_rodata);
893 rodata_resource.end = __pa_symbol(__end_rodata)-1;
894 data_resource.start = __pa_symbol(_sdata);
895 data_resource.end = __pa_symbol(_edata)-1;
896 bss_resource.start = __pa_symbol(__bss_start);
897 bss_resource.end = __pa_symbol(__bss_stop)-1;
898
899 #ifdef CONFIG_CMDLINE_BOOL
900 #ifdef CONFIG_CMDLINE_OVERRIDE
901 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
902 #else
903 if (builtin_cmdline[0]) {
904 /* append boot loader cmdline to builtin */
905 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
906 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
907 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
908 }
909 #endif
910 #endif
911
912 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
913 *cmdline_p = command_line;
914
915 /*
916 * x86_configure_nx() is called before parse_early_param() to detect
917 * whether hardware doesn't support NX (so that the early EHCI debug
918 * console setup can safely call set_fixmap()).
919 */
920 x86_configure_nx();
921
922 parse_early_param();
923
924 if (efi_enabled(EFI_BOOT))
925 efi_memblock_x86_reserve_range();
926
927 #ifdef CONFIG_MEMORY_HOTPLUG
928 /*
929 * Memory used by the kernel cannot be hot-removed because Linux
930 * cannot migrate the kernel pages. When memory hotplug is
931 * enabled, we should prevent memblock from allocating memory
932 * for the kernel.
933 *
934 * ACPI SRAT records all hotpluggable memory ranges. But before
935 * SRAT is parsed, we don't know about it.
936 *
937 * The kernel image is loaded into memory at very early time. We
938 * cannot prevent this anyway. So on NUMA system, we set any
939 * node the kernel resides in as un-hotpluggable.
940 *
941 * Since on modern servers, one node could have double-digit
942 * gigabytes memory, we can assume the memory around the kernel
943 * image is also un-hotpluggable. So before SRAT is parsed, just
944 * allocate memory near the kernel image to try the best to keep
945 * the kernel away from hotpluggable memory.
946 */
947 if (movable_node_is_enabled())
948 memblock_set_bottom_up(true);
949 #endif
950
951 x86_report_nx();
952
953 if (acpi_mps_check()) {
954 #ifdef CONFIG_X86_LOCAL_APIC
955 disable_apic = 1;
956 #endif
957 setup_clear_cpu_cap(X86_FEATURE_APIC);
958 }
959
960 e820__reserve_setup_data();
961 e820__finish_early_params();
962
963 if (efi_enabled(EFI_BOOT))
964 efi_init();
965
966 dmi_setup();
967
968 /*
969 * VMware detection requires dmi to be available, so this
970 * needs to be done after dmi_setup(), for the boot CPU.
971 */
972 init_hypervisor_platform();
973
974 tsc_early_init();
975 x86_init.resources.probe_roms();
976
977 /* after parse_early_param, so could debug it */
978 insert_resource(&iomem_resource, &code_resource);
979 insert_resource(&iomem_resource, &rodata_resource);
980 insert_resource(&iomem_resource, &data_resource);
981 insert_resource(&iomem_resource, &bss_resource);
982
983 e820_add_kernel_range();
984 trim_bios_range();
985 #ifdef CONFIG_X86_32
986 if (ppro_with_ram_bug()) {
987 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
988 E820_TYPE_RESERVED);
989 e820__update_table(e820_table);
990 printk(KERN_INFO "fixed physical RAM map:\n");
991 e820__print_table("bad_ppro");
992 }
993 #else
994 early_gart_iommu_check();
995 #endif
996
997 /*
998 * partially used pages are not usable - thus
999 * we are rounding upwards:
1000 */
1001 max_pfn = e820__end_of_ram_pfn();
1002
1003 /* update e820 for memory not covered by WB MTRRs */
1004 if (IS_ENABLED(CONFIG_MTRR))
1005 mtrr_bp_init();
1006 else
1007 pat_disable("PAT support disabled because CONFIG_MTRR is disabled in the kernel.");
1008
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 * This call is required when the CPU does not support PAT. If
1016 * mtrr_bp_init() invoked it already via pat_init() the call has no
1017 * effect.
1018 */
1019 init_cache_modes();
1020
1021 /*
1022 * Define random base addresses for memory sections after max_pfn is
1023 * defined and before each memory section base is used.
1024 */
1025 kernel_randomize_memory();
1026
1027 #ifdef CONFIG_X86_32
1028 /* max_low_pfn get updated here */
1029 find_low_pfn_range();
1030 #else
1031 check_x2apic();
1032
1033 /* How many end-of-memory variables you have, grandma! */
1034 /* need this before calling reserve_initrd */
1035 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1036 max_low_pfn = e820__end_of_low_ram_pfn();
1037 else
1038 max_low_pfn = max_pfn;
1039
1040 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1041 #endif
1042
1043 /*
1044 * Find and reserve possible boot-time SMP configuration:
1045 */
1046 find_smp_config();
1047
1048 early_alloc_pgt_buf();
1049
1050 /*
1051 * Need to conclude brk, before e820__memblock_setup()
1052 * it could use memblock_find_in_range, could overlap with
1053 * brk area.
1054 */
1055 reserve_brk();
1056
1057 cleanup_highmap();
1058
1059 memblock_set_current_limit(ISA_END_ADDRESS);
1060 e820__memblock_setup();
1061
1062 /*
1063 * Needs to run after memblock setup because it needs the physical
1064 * memory size.
1065 */
1066 sev_setup_arch();
1067
1068 efi_fake_memmap();
1069 efi_find_mirror();
1070 efi_esrt_init();
1071 efi_mokvar_table_init();
1072
1073 /*
1074 * The EFI specification says that boot service code won't be
1075 * called after ExitBootServices(). This is, in fact, a lie.
1076 */
1077 efi_reserve_boot_services();
1078
1079 /* preallocate 4k for mptable mpc */
1080 e820__memblock_alloc_reserved_mpc_new();
1081
1082 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1083 setup_bios_corruption_check();
1084 #endif
1085
1086 #ifdef CONFIG_X86_32
1087 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1088 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1089 #endif
1090
1091 /*
1092 * Find free memory for the real mode trampoline and place it there. If
1093 * there is not enough free memory under 1M, on EFI-enabled systems
1094 * there will be additional attempt to reclaim the memory for the real
1095 * mode trampoline at efi_free_boot_services().
1096 *
1097 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1098 * are known to corrupt low memory and several hundred kilobytes are not
1099 * worth complex detection what memory gets clobbered. Windows does the
1100 * same thing for very similar reasons.
1101 *
1102 * Moreover, on machines with SandyBridge graphics or in setups that use
1103 * crashkernel the entire 1M is reserved anyway.
1104 */
1105 reserve_real_mode();
1106
1107 init_mem_mapping();
1108
1109 idt_setup_early_pf();
1110
1111 /*
1112 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1113 * with the current CR4 value. This may not be necessary, but
1114 * auditing all the early-boot CR4 manipulation would be needed to
1115 * rule it out.
1116 *
1117 * Mask off features that don't work outside long mode (just
1118 * PCIDE for now).
1119 */
1120 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1121
1122 memblock_set_current_limit(get_max_mapped());
1123
1124 /*
1125 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1126 */
1127
1128 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1129 if (init_ohci1394_dma_early)
1130 init_ohci1394_dma_on_all_controllers();
1131 #endif
1132 /* Allocate bigger log buffer */
1133 setup_log_buf(1);
1134
1135 if (efi_enabled(EFI_BOOT)) {
1136 switch (boot_params.secure_boot) {
1137 case efi_secureboot_mode_disabled:
1138 pr_info("Secure boot disabled\n");
1139 break;
1140 case efi_secureboot_mode_enabled:
1141 pr_info("Secure boot enabled\n");
1142 break;
1143 default:
1144 pr_info("Secure boot could not be determined\n");
1145 break;
1146 }
1147 }
1148
1149 reserve_initrd();
1150
1151 acpi_table_upgrade();
1152 /* Look for ACPI tables and reserve memory occupied by them. */
1153 acpi_boot_table_init();
1154
1155 vsmp_init();
1156
1157 io_delay_init();
1158
1159 early_platform_quirks();
1160
1161 early_acpi_boot_init();
1162
1163 initmem_init();
1164 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1165
1166 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1167 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1168
1169 /*
1170 * Reserve memory for crash kernel after SRAT is parsed so that it
1171 * won't consume hotpluggable memory.
1172 */
1173 reserve_crashkernel();
1174
1175 memblock_find_dma_reserve();
1176
1177 if (!early_xdbc_setup_hardware())
1178 early_xdbc_register_console();
1179
1180 x86_init.paging.pagetable_init();
1181
1182 kasan_init();
1183
1184 /*
1185 * Sync back kernel address range.
1186 *
1187 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1188 * this call?
1189 */
1190 sync_initial_page_table();
1191
1192 tboot_probe();
1193
1194 map_vsyscall();
1195
1196 generic_apic_probe();
1197
1198 early_quirks();
1199
1200 /*
1201 * Read APIC and some other early information from ACPI tables.
1202 */
1203 acpi_boot_init();
1204 x86_dtb_init();
1205
1206 /*
1207 * get boot-time SMP configuration:
1208 */
1209 get_smp_config();
1210
1211 /*
1212 * Systems w/o ACPI and mptables might not have it mapped the local
1213 * APIC yet, but prefill_possible_map() might need to access it.
1214 */
1215 init_apic_mappings();
1216
1217 prefill_possible_map();
1218
1219 init_cpu_to_node();
1220 init_gi_nodes();
1221
1222 io_apic_init_mappings();
1223
1224 x86_init.hyper.guest_late_init();
1225
1226 e820__reserve_resources();
1227 e820__register_nosave_regions(max_pfn);
1228
1229 x86_init.resources.reserve_resources();
1230
1231 e820__setup_pci_gap();
1232
1233 #ifdef CONFIG_VT
1234 #if defined(CONFIG_VGA_CONSOLE)
1235 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1236 conswitchp = &vga_con;
1237 #endif
1238 #endif
1239 x86_init.oem.banner();
1240
1241 x86_init.timers.wallclock_init();
1242
1243 /*
1244 * This needs to run before setup_local_APIC() which soft-disables the
1245 * local APIC temporarily and that masks the thermal LVT interrupt,
1246 * leading to softlockups on machines which have configured SMI
1247 * interrupt delivery.
1248 */
1249 therm_lvt_init();
1250
1251 mcheck_init();
1252
1253 register_refined_jiffies(CLOCK_TICK_RATE);
1254
1255 #ifdef CONFIG_EFI
1256 if (efi_enabled(EFI_BOOT))
1257 efi_apply_memmap_quirks();
1258 #endif
1259
1260 unwind_init();
1261 }
1262
1263 #ifdef CONFIG_X86_32
1264
1265 static struct resource video_ram_resource = {
1266 .name = "Video RAM area",
1267 .start = 0xa0000,
1268 .end = 0xbffff,
1269 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1270 };
1271
i386_reserve_resources(void)1272 void __init i386_reserve_resources(void)
1273 {
1274 request_resource(&iomem_resource, &video_ram_resource);
1275 reserve_standard_io_resources();
1276 }
1277
1278 #endif /* CONFIG_X86_32 */
1279
1280 static struct notifier_block kernel_offset_notifier = {
1281 .notifier_call = dump_kernel_offset
1282 };
1283
register_kernel_offset_dumper(void)1284 static int __init register_kernel_offset_dumper(void)
1285 {
1286 atomic_notifier_chain_register(&panic_notifier_list,
1287 &kernel_offset_notifier);
1288 return 0;
1289 }
1290 __initcall(register_kernel_offset_dumper);
1291