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/mtrr.h>
44 #include <asm/realmode.h>
45 #include <asm/olpc_ofw.h>
46 #include <asm/pci-direct.h>
47 #include <asm/prom.h>
48 #include <asm/proto.h>
49 #include <asm/thermal.h>
50 #include <asm/unwind.h>
51 #include <asm/vsyscall.h>
52 #include <linux/vmalloc.h>
53
54 /*
55 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
56 * max_pfn_mapped: highest directly mapped pfn > 4 GB
57 *
58 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
59 * represented by pfn_mapped[].
60 */
61 unsigned long max_low_pfn_mapped;
62 unsigned long max_pfn_mapped;
63
64 #ifdef CONFIG_DMI
65 RESERVE_BRK(dmi_alloc, 65536);
66 #endif
67
68
69 /*
70 * Range of the BSS area. The size of the BSS area is determined
71 * at link time, with RESERVE_BRK() facility reserving additional
72 * chunks.
73 */
74 unsigned long _brk_start = (unsigned long)__brk_base;
75 unsigned long _brk_end = (unsigned long)__brk_base;
76
77 struct boot_params boot_params;
78
79 /*
80 * These are the four main kernel memory regions, we put them into
81 * the resource tree so that kdump tools and other debugging tools
82 * recover it:
83 */
84
85 static struct resource rodata_resource = {
86 .name = "Kernel rodata",
87 .start = 0,
88 .end = 0,
89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
90 };
91
92 static struct resource data_resource = {
93 .name = "Kernel data",
94 .start = 0,
95 .end = 0,
96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
97 };
98
99 static struct resource code_resource = {
100 .name = "Kernel code",
101 .start = 0,
102 .end = 0,
103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
104 };
105
106 static struct resource bss_resource = {
107 .name = "Kernel bss",
108 .start = 0,
109 .end = 0,
110 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
111 };
112
113
114 #ifdef CONFIG_X86_32
115 /* CPU data as detected by the assembly code in head_32.S */
116 struct cpuinfo_x86 new_cpu_data;
117
118 /* Common CPU data for all CPUs */
119 struct cpuinfo_x86 boot_cpu_data __read_mostly;
120 EXPORT_SYMBOL(boot_cpu_data);
121
122 unsigned int def_to_bigsmp;
123
124 struct apm_info apm_info;
125 EXPORT_SYMBOL(apm_info);
126
127 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
128 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
129 struct ist_info ist_info;
130 EXPORT_SYMBOL(ist_info);
131 #else
132 struct ist_info ist_info;
133 #endif
134
135 #else
136 struct cpuinfo_x86 boot_cpu_data __read_mostly;
137 EXPORT_SYMBOL(boot_cpu_data);
138 #endif
139
140
141 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
142 __visible unsigned long mmu_cr4_features __ro_after_init;
143 #else
144 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
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 /*
151 * Setup options
152 */
153 struct screen_info screen_info;
154 EXPORT_SYMBOL(screen_info);
155 struct edid_info edid_info;
156 EXPORT_SYMBOL_GPL(edid_info);
157
158 extern int root_mountflags;
159
160 unsigned long saved_video_mode;
161
162 #define RAMDISK_IMAGE_START_MASK 0x07FF
163 #define RAMDISK_PROMPT_FLAG 0x8000
164 #define RAMDISK_LOAD_FLAG 0x4000
165
166 static char __initdata command_line[COMMAND_LINE_SIZE];
167 #ifdef CONFIG_CMDLINE_BOOL
168 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
169 #endif
170
171 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
172 struct edd edd;
173 #ifdef CONFIG_EDD_MODULE
174 EXPORT_SYMBOL(edd);
175 #endif
176 /**
177 * copy_edd() - Copy the BIOS EDD information
178 * from boot_params into a safe place.
179 *
180 */
copy_edd(void)181 static inline void __init copy_edd(void)
182 {
183 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
184 sizeof(edd.mbr_signature));
185 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
186 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
187 edd.edd_info_nr = boot_params.eddbuf_entries;
188 }
189 #else
copy_edd(void)190 static inline void __init copy_edd(void)
191 {
192 }
193 #endif
194
extend_brk(size_t size,size_t align)195 void * __init extend_brk(size_t size, size_t align)
196 {
197 size_t mask = align - 1;
198 void *ret;
199
200 BUG_ON(_brk_start == 0);
201 BUG_ON(align & mask);
202
203 _brk_end = (_brk_end + mask) & ~mask;
204 BUG_ON((char *)(_brk_end + size) > __brk_limit);
205
206 ret = (void *)_brk_end;
207 _brk_end += size;
208
209 memset(ret, 0, size);
210
211 return ret;
212 }
213
214 #ifdef CONFIG_X86_32
cleanup_highmap(void)215 static void __init cleanup_highmap(void)
216 {
217 }
218 #endif
219
reserve_brk(void)220 static void __init reserve_brk(void)
221 {
222 if (_brk_end > _brk_start)
223 memblock_reserve(__pa_symbol(_brk_start),
224 _brk_end - _brk_start);
225
226 /* Mark brk area as locked down and no longer taking any
227 new allocations */
228 _brk_start = 0;
229 }
230
231 u64 relocated_ramdisk;
232
233 #ifdef CONFIG_BLK_DEV_INITRD
234
get_ramdisk_image(void)235 static u64 __init get_ramdisk_image(void)
236 {
237 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
238
239 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
240
241 if (ramdisk_image == 0)
242 ramdisk_image = phys_initrd_start;
243
244 return ramdisk_image;
245 }
get_ramdisk_size(void)246 static u64 __init get_ramdisk_size(void)
247 {
248 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
249
250 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
251
252 if (ramdisk_size == 0)
253 ramdisk_size = phys_initrd_size;
254
255 return ramdisk_size;
256 }
257
relocate_initrd(void)258 static void __init relocate_initrd(void)
259 {
260 /* Assume only end is not page aligned */
261 u64 ramdisk_image = get_ramdisk_image();
262 u64 ramdisk_size = get_ramdisk_size();
263 u64 area_size = PAGE_ALIGN(ramdisk_size);
264
265 /* We need to move the initrd down into directly mapped mem */
266 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
267 PFN_PHYS(max_pfn_mapped));
268 if (!relocated_ramdisk)
269 panic("Cannot find place for new RAMDISK of size %lld\n",
270 ramdisk_size);
271
272 initrd_start = relocated_ramdisk + PAGE_OFFSET;
273 initrd_end = initrd_start + ramdisk_size;
274 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
275 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
276
277 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
278
279 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
280 " [mem %#010llx-%#010llx]\n",
281 ramdisk_image, ramdisk_image + ramdisk_size - 1,
282 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
283 }
284
early_reserve_initrd(void)285 static void __init early_reserve_initrd(void)
286 {
287 /* Assume only end is not page aligned */
288 u64 ramdisk_image = get_ramdisk_image();
289 u64 ramdisk_size = get_ramdisk_size();
290 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
291
292 if (!boot_params.hdr.type_of_loader ||
293 !ramdisk_image || !ramdisk_size)
294 return; /* No initrd provided by bootloader */
295
296 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
297 }
298
reserve_initrd(void)299 static void __init reserve_initrd(void)
300 {
301 /* Assume only end is not page aligned */
302 u64 ramdisk_image = get_ramdisk_image();
303 u64 ramdisk_size = get_ramdisk_size();
304 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
305
306 if (!boot_params.hdr.type_of_loader ||
307 !ramdisk_image || !ramdisk_size)
308 return; /* No initrd provided by bootloader */
309
310 initrd_start = 0;
311
312 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
313 ramdisk_end - 1);
314
315 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
316 PFN_DOWN(ramdisk_end))) {
317 /* All are mapped, easy case */
318 initrd_start = ramdisk_image + PAGE_OFFSET;
319 initrd_end = initrd_start + ramdisk_size;
320 return;
321 }
322
323 relocate_initrd();
324
325 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
326 }
327
328 #else
early_reserve_initrd(void)329 static void __init early_reserve_initrd(void)
330 {
331 }
reserve_initrd(void)332 static void __init reserve_initrd(void)
333 {
334 }
335 #endif /* CONFIG_BLK_DEV_INITRD */
336
parse_setup_data(void)337 static void __init parse_setup_data(void)
338 {
339 struct setup_data *data;
340 u64 pa_data, pa_next;
341
342 pa_data = boot_params.hdr.setup_data;
343 while (pa_data) {
344 u32 data_len, data_type;
345
346 data = early_memremap(pa_data, sizeof(*data));
347 data_len = data->len + sizeof(struct setup_data);
348 data_type = data->type;
349 pa_next = data->next;
350 early_memunmap(data, sizeof(*data));
351
352 switch (data_type) {
353 case SETUP_E820_EXT:
354 e820__memory_setup_extended(pa_data, data_len);
355 break;
356 case SETUP_DTB:
357 add_dtb(pa_data);
358 break;
359 case SETUP_EFI:
360 parse_efi_setup(pa_data, data_len);
361 break;
362 default:
363 break;
364 }
365 pa_data = pa_next;
366 }
367 }
368
memblock_x86_reserve_range_setup_data(void)369 static void __init memblock_x86_reserve_range_setup_data(void)
370 {
371 struct setup_data *data;
372 u64 pa_data;
373
374 pa_data = boot_params.hdr.setup_data;
375 while (pa_data) {
376 data = early_memremap(pa_data, sizeof(*data));
377 memblock_reserve(pa_data, sizeof(*data) + data->len);
378
379 if (data->type == SETUP_INDIRECT &&
380 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
381 memblock_reserve(((struct setup_indirect *)data->data)->addr,
382 ((struct setup_indirect *)data->data)->len);
383
384 pa_data = data->next;
385 early_memunmap(data, sizeof(*data));
386 }
387 }
388
389 /*
390 * --------- Crashkernel reservation ------------------------------
391 */
392
393 #ifdef CONFIG_KEXEC_CORE
394
395 /* 16M alignment for crash kernel regions */
396 #define CRASH_ALIGN SZ_16M
397
398 /*
399 * Keep the crash kernel below this limit.
400 *
401 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
402 * due to mapping restrictions.
403 *
404 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
405 * the upper limit of system RAM in 4-level paging mode. Since the kdump
406 * jump could be from 5-level paging to 4-level paging, the jump will fail if
407 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
408 * no good way to detect the paging mode of the target kernel which will be
409 * loaded for dumping.
410 */
411 #ifdef CONFIG_X86_32
412 # define CRASH_ADDR_LOW_MAX SZ_512M
413 # define CRASH_ADDR_HIGH_MAX SZ_512M
414 #else
415 # define CRASH_ADDR_LOW_MAX SZ_4G
416 # define CRASH_ADDR_HIGH_MAX SZ_64T
417 #endif
418
reserve_crashkernel_low(void)419 static int __init reserve_crashkernel_low(void)
420 {
421 #ifdef CONFIG_X86_64
422 unsigned long long base, low_base = 0, low_size = 0;
423 unsigned long low_mem_limit;
424 int ret;
425
426 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
427
428 /* crashkernel=Y,low */
429 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
430 if (ret) {
431 /*
432 * two parts from kernel/dma/swiotlb.c:
433 * -swiotlb size: user-specified with swiotlb= or default.
434 *
435 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
436 * to 8M for other buffers that may need to stay low too. Also
437 * make sure we allocate enough extra low memory so that we
438 * don't run out of DMA buffers for 32-bit devices.
439 */
440 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
441 } else {
442 /* passed with crashkernel=0,low ? */
443 if (!low_size)
444 return 0;
445 }
446
447 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
448 if (!low_base) {
449 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
450 (unsigned long)(low_size >> 20));
451 return -ENOMEM;
452 }
453
454 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
455 (unsigned long)(low_size >> 20),
456 (unsigned long)(low_base >> 20),
457 (unsigned long)(low_mem_limit >> 20));
458
459 crashk_low_res.start = low_base;
460 crashk_low_res.end = low_base + low_size - 1;
461 insert_resource(&iomem_resource, &crashk_low_res);
462 #endif
463 return 0;
464 }
465
reserve_crashkernel(void)466 static void __init reserve_crashkernel(void)
467 {
468 unsigned long long crash_size, crash_base, total_mem;
469 bool high = false;
470 int ret;
471
472 total_mem = memblock_phys_mem_size();
473
474 /* crashkernel=XM */
475 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
476 if (ret != 0 || crash_size <= 0) {
477 /* crashkernel=X,high */
478 ret = parse_crashkernel_high(boot_command_line, total_mem,
479 &crash_size, &crash_base);
480 if (ret != 0 || crash_size <= 0)
481 return;
482 high = true;
483 }
484
485 if (xen_pv_domain()) {
486 pr_info("Ignoring crashkernel for a Xen PV domain\n");
487 return;
488 }
489
490 /* 0 means: find the address automatically */
491 if (!crash_base) {
492 /*
493 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
494 * crashkernel=x,high reserves memory over 4G, also allocates
495 * 256M extra low memory for DMA buffers and swiotlb.
496 * But the extra memory is not required for all machines.
497 * So try low memory first and fall back to high memory
498 * unless "crashkernel=size[KMG],high" is specified.
499 */
500 if (!high)
501 crash_base = memblock_phys_alloc_range(crash_size,
502 CRASH_ALIGN, CRASH_ALIGN,
503 CRASH_ADDR_LOW_MAX);
504 if (!crash_base)
505 crash_base = memblock_phys_alloc_range(crash_size,
506 CRASH_ALIGN, CRASH_ALIGN,
507 CRASH_ADDR_HIGH_MAX);
508 if (!crash_base) {
509 pr_info("crashkernel reservation failed - No suitable area found.\n");
510 return;
511 }
512 } else {
513 unsigned long long start;
514
515 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
516 crash_base + crash_size);
517 if (start != crash_base) {
518 pr_info("crashkernel reservation failed - memory is in use.\n");
519 return;
520 }
521 }
522
523 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
524 memblock_phys_free(crash_base, crash_size);
525 return;
526 }
527
528 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
529 (unsigned long)(crash_size >> 20),
530 (unsigned long)(crash_base >> 20),
531 (unsigned long)(total_mem >> 20));
532
533 crashk_res.start = crash_base;
534 crashk_res.end = crash_base + crash_size - 1;
535 insert_resource(&iomem_resource, &crashk_res);
536 }
537 #else
reserve_crashkernel(void)538 static void __init reserve_crashkernel(void)
539 {
540 }
541 #endif
542
543 static struct resource standard_io_resources[] = {
544 { .name = "dma1", .start = 0x00, .end = 0x1f,
545 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
546 { .name = "pic1", .start = 0x20, .end = 0x21,
547 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
548 { .name = "timer0", .start = 0x40, .end = 0x43,
549 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
550 { .name = "timer1", .start = 0x50, .end = 0x53,
551 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
552 { .name = "keyboard", .start = 0x60, .end = 0x60,
553 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
554 { .name = "keyboard", .start = 0x64, .end = 0x64,
555 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
556 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
557 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
558 { .name = "pic2", .start = 0xa0, .end = 0xa1,
559 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
560 { .name = "dma2", .start = 0xc0, .end = 0xdf,
561 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
562 { .name = "fpu", .start = 0xf0, .end = 0xff,
563 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
564 };
565
reserve_standard_io_resources(void)566 void __init reserve_standard_io_resources(void)
567 {
568 int i;
569
570 /* request I/O space for devices used on all i[345]86 PCs */
571 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
572 request_resource(&ioport_resource, &standard_io_resources[i]);
573
574 }
575
snb_gfx_workaround_needed(void)576 static bool __init snb_gfx_workaround_needed(void)
577 {
578 #ifdef CONFIG_PCI
579 int i;
580 u16 vendor, devid;
581 static const __initconst u16 snb_ids[] = {
582 0x0102,
583 0x0112,
584 0x0122,
585 0x0106,
586 0x0116,
587 0x0126,
588 0x010a,
589 };
590
591 /* Assume no if something weird is going on with PCI */
592 if (!early_pci_allowed())
593 return false;
594
595 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
596 if (vendor != 0x8086)
597 return false;
598
599 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
600 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
601 if (devid == snb_ids[i])
602 return true;
603 #endif
604
605 return false;
606 }
607
608 /*
609 * Sandy Bridge graphics has trouble with certain ranges, exclude
610 * them from allocation.
611 */
trim_snb_memory(void)612 static void __init trim_snb_memory(void)
613 {
614 static const __initconst unsigned long bad_pages[] = {
615 0x20050000,
616 0x20110000,
617 0x20130000,
618 0x20138000,
619 0x40004000,
620 };
621 int i;
622
623 if (!snb_gfx_workaround_needed())
624 return;
625
626 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
627
628 /*
629 * SandyBridge integrated graphics devices have a bug that prevents
630 * them from accessing certain memory ranges, namely anything below
631 * 1M and in the pages listed in bad_pages[] above.
632 *
633 * To avoid these pages being ever accessed by SNB gfx devices reserve
634 * bad_pages that have not already been reserved at boot time.
635 * All memory below the 1 MB mark is anyway reserved later during
636 * setup_arch(), so there is no need to reserve it here.
637 */
638
639 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
640 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
641 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
642 bad_pages[i]);
643 }
644 }
645
trim_bios_range(void)646 static void __init trim_bios_range(void)
647 {
648 /*
649 * A special case is the first 4Kb of memory;
650 * This is a BIOS owned area, not kernel ram, but generally
651 * not listed as such in the E820 table.
652 *
653 * This typically reserves additional memory (64KiB by default)
654 * since some BIOSes are known to corrupt low memory. See the
655 * Kconfig help text for X86_RESERVE_LOW.
656 */
657 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
658
659 /*
660 * special case: Some BIOSes report the PC BIOS
661 * area (640Kb -> 1Mb) as RAM even though it is not.
662 * take them out.
663 */
664 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
665
666 e820__update_table(e820_table);
667 }
668
669 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)670 static void __init e820_add_kernel_range(void)
671 {
672 u64 start = __pa_symbol(_text);
673 u64 size = __pa_symbol(_end) - start;
674
675 /*
676 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
677 * attempt to fix it by adding the range. We may have a confused BIOS,
678 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
679 * exclude kernel range. If we really are running on top non-RAM,
680 * we will crash later anyways.
681 */
682 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
683 return;
684
685 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
686 e820__range_remove(start, size, E820_TYPE_RAM, 0);
687 e820__range_add(start, size, E820_TYPE_RAM);
688 }
689
early_reserve_memory(void)690 static void __init early_reserve_memory(void)
691 {
692 /*
693 * Reserve the memory occupied by the kernel between _text and
694 * __end_of_kernel_reserve symbols. Any kernel sections after the
695 * __end_of_kernel_reserve symbol must be explicitly reserved with a
696 * separate memblock_reserve() or they will be discarded.
697 */
698 memblock_reserve(__pa_symbol(_text),
699 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
700
701 /*
702 * The first 4Kb of memory is a BIOS owned area, but generally it is
703 * not listed as such in the E820 table.
704 *
705 * Reserve the first 64K of memory since some BIOSes are known to
706 * corrupt low memory. After the real mode trampoline is allocated the
707 * rest of the memory below 640k is reserved.
708 *
709 * In addition, make sure page 0 is always reserved because on
710 * systems with L1TF its contents can be leaked to user processes.
711 */
712 memblock_reserve(0, SZ_64K);
713
714 early_reserve_initrd();
715
716 memblock_x86_reserve_range_setup_data();
717
718 reserve_ibft_region();
719 reserve_bios_regions();
720 trim_snb_memory();
721 }
722
723 /*
724 * Dump out kernel offset information on panic.
725 */
726 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)727 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
728 {
729 if (kaslr_enabled()) {
730 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
731 kaslr_offset(),
732 __START_KERNEL,
733 __START_KERNEL_map,
734 MODULES_VADDR-1);
735 } else {
736 pr_emerg("Kernel Offset: disabled\n");
737 }
738
739 return 0;
740 }
741
742 /*
743 * Determine if we were loaded by an EFI loader. If so, then we have also been
744 * passed the efi memmap, systab, etc., so we should use these data structures
745 * for initialization. Note, the efi init code path is determined by the
746 * global efi_enabled. This allows the same kernel image to be used on existing
747 * systems (with a traditional BIOS) as well as on EFI systems.
748 */
749 /*
750 * setup_arch - architecture-specific boot-time initializations
751 *
752 * Note: On x86_64, fixmaps are ready for use even before this is called.
753 */
754
setup_arch(char ** cmdline_p)755 void __init setup_arch(char **cmdline_p)
756 {
757 #ifdef CONFIG_X86_32
758 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
759
760 /*
761 * copy kernel address range established so far and switch
762 * to the proper swapper page table
763 */
764 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
765 initial_page_table + KERNEL_PGD_BOUNDARY,
766 KERNEL_PGD_PTRS);
767
768 load_cr3(swapper_pg_dir);
769 /*
770 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
771 * a cr3 based tlb flush, so the following __flush_tlb_all()
772 * will not flush anything because the CPU quirk which clears
773 * X86_FEATURE_PGE has not been invoked yet. Though due to the
774 * load_cr3() above the TLB has been flushed already. The
775 * quirk is invoked before subsequent calls to __flush_tlb_all()
776 * so proper operation is guaranteed.
777 */
778 __flush_tlb_all();
779 #else
780 printk(KERN_INFO "Command line: %s\n", boot_command_line);
781 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
782 #endif
783
784 /*
785 * If we have OLPC OFW, we might end up relocating the fixmap due to
786 * reserve_top(), so do this before touching the ioremap area.
787 */
788 olpc_ofw_detect();
789
790 idt_setup_early_traps();
791 early_cpu_init();
792 jump_label_init();
793 static_call_init();
794 early_ioremap_init();
795
796 setup_olpc_ofw_pgd();
797
798 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
799 screen_info = boot_params.screen_info;
800 edid_info = boot_params.edid_info;
801 #ifdef CONFIG_X86_32
802 apm_info.bios = boot_params.apm_bios_info;
803 ist_info = boot_params.ist_info;
804 #endif
805 saved_video_mode = boot_params.hdr.vid_mode;
806 bootloader_type = boot_params.hdr.type_of_loader;
807 if ((bootloader_type >> 4) == 0xe) {
808 bootloader_type &= 0xf;
809 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
810 }
811 bootloader_version = bootloader_type & 0xf;
812 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
813
814 #ifdef CONFIG_BLK_DEV_RAM
815 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
816 #endif
817 #ifdef CONFIG_EFI
818 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
819 EFI32_LOADER_SIGNATURE, 4)) {
820 set_bit(EFI_BOOT, &efi.flags);
821 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
822 EFI64_LOADER_SIGNATURE, 4)) {
823 set_bit(EFI_BOOT, &efi.flags);
824 set_bit(EFI_64BIT, &efi.flags);
825 }
826 #endif
827
828 x86_init.oem.arch_setup();
829
830 /*
831 * Do some memory reservations *before* memory is added to memblock, so
832 * memblock allocations won't overwrite it.
833 *
834 * After this point, everything still needed from the boot loader or
835 * firmware or kernel text should be early reserved or marked not RAM in
836 * e820. All other memory is free game.
837 *
838 * This call needs to happen before e820__memory_setup() which calls the
839 * xen_memory_setup() on Xen dom0 which relies on the fact that those
840 * early reservations have happened already.
841 */
842 early_reserve_memory();
843
844 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
845 e820__memory_setup();
846 parse_setup_data();
847
848 copy_edd();
849
850 if (!boot_params.hdr.root_flags)
851 root_mountflags &= ~MS_RDONLY;
852 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
853
854 code_resource.start = __pa_symbol(_text);
855 code_resource.end = __pa_symbol(_etext)-1;
856 rodata_resource.start = __pa_symbol(__start_rodata);
857 rodata_resource.end = __pa_symbol(__end_rodata)-1;
858 data_resource.start = __pa_symbol(_sdata);
859 data_resource.end = __pa_symbol(_edata)-1;
860 bss_resource.start = __pa_symbol(__bss_start);
861 bss_resource.end = __pa_symbol(__bss_stop)-1;
862
863 #ifdef CONFIG_CMDLINE_BOOL
864 #ifdef CONFIG_CMDLINE_OVERRIDE
865 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
866 #else
867 if (builtin_cmdline[0]) {
868 /* append boot loader cmdline to builtin */
869 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
870 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
871 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
872 }
873 #endif
874 #endif
875
876 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
877 *cmdline_p = command_line;
878
879 /*
880 * x86_configure_nx() is called before parse_early_param() to detect
881 * whether hardware doesn't support NX (so that the early EHCI debug
882 * console setup can safely call set_fixmap()). It may then be called
883 * again from within noexec_setup() during parsing early parameters
884 * to honor the respective command line option.
885 */
886 x86_configure_nx();
887
888 parse_early_param();
889
890 if (efi_enabled(EFI_BOOT))
891 efi_memblock_x86_reserve_range();
892
893 #ifdef CONFIG_MEMORY_HOTPLUG
894 /*
895 * Memory used by the kernel cannot be hot-removed because Linux
896 * cannot migrate the kernel pages. When memory hotplug is
897 * enabled, we should prevent memblock from allocating memory
898 * for the kernel.
899 *
900 * ACPI SRAT records all hotpluggable memory ranges. But before
901 * SRAT is parsed, we don't know about it.
902 *
903 * The kernel image is loaded into memory at very early time. We
904 * cannot prevent this anyway. So on NUMA system, we set any
905 * node the kernel resides in as un-hotpluggable.
906 *
907 * Since on modern servers, one node could have double-digit
908 * gigabytes memory, we can assume the memory around the kernel
909 * image is also un-hotpluggable. So before SRAT is parsed, just
910 * allocate memory near the kernel image to try the best to keep
911 * the kernel away from hotpluggable memory.
912 */
913 if (movable_node_is_enabled())
914 memblock_set_bottom_up(true);
915 #endif
916
917 x86_report_nx();
918
919 if (acpi_mps_check()) {
920 #ifdef CONFIG_X86_LOCAL_APIC
921 disable_apic = 1;
922 #endif
923 setup_clear_cpu_cap(X86_FEATURE_APIC);
924 }
925
926 e820__reserve_setup_data();
927 e820__finish_early_params();
928
929 if (efi_enabled(EFI_BOOT))
930 efi_init();
931
932 dmi_setup();
933
934 /*
935 * VMware detection requires dmi to be available, so this
936 * needs to be done after dmi_setup(), for the boot CPU.
937 */
938 init_hypervisor_platform();
939
940 tsc_early_init();
941 x86_init.resources.probe_roms();
942
943 /* after parse_early_param, so could debug it */
944 insert_resource(&iomem_resource, &code_resource);
945 insert_resource(&iomem_resource, &rodata_resource);
946 insert_resource(&iomem_resource, &data_resource);
947 insert_resource(&iomem_resource, &bss_resource);
948
949 e820_add_kernel_range();
950 trim_bios_range();
951 #ifdef CONFIG_X86_32
952 if (ppro_with_ram_bug()) {
953 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
954 E820_TYPE_RESERVED);
955 e820__update_table(e820_table);
956 printk(KERN_INFO "fixed physical RAM map:\n");
957 e820__print_table("bad_ppro");
958 }
959 #else
960 early_gart_iommu_check();
961 #endif
962
963 /*
964 * partially used pages are not usable - thus
965 * we are rounding upwards:
966 */
967 max_pfn = e820__end_of_ram_pfn();
968
969 /* update e820 for memory not covered by WB MTRRs */
970 mtrr_bp_init();
971 if (mtrr_trim_uncached_memory(max_pfn))
972 max_pfn = e820__end_of_ram_pfn();
973
974 max_possible_pfn = max_pfn;
975
976 /*
977 * This call is required when the CPU does not support PAT. If
978 * mtrr_bp_init() invoked it already via pat_init() the call has no
979 * effect.
980 */
981 init_cache_modes();
982
983 /*
984 * Define random base addresses for memory sections after max_pfn is
985 * defined and before each memory section base is used.
986 */
987 kernel_randomize_memory();
988
989 #ifdef CONFIG_X86_32
990 /* max_low_pfn get updated here */
991 find_low_pfn_range();
992 #else
993 check_x2apic();
994
995 /* How many end-of-memory variables you have, grandma! */
996 /* need this before calling reserve_initrd */
997 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
998 max_low_pfn = e820__end_of_low_ram_pfn();
999 else
1000 max_low_pfn = max_pfn;
1001
1002 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1003 #endif
1004
1005 /*
1006 * Find and reserve possible boot-time SMP configuration:
1007 */
1008 find_smp_config();
1009
1010 early_alloc_pgt_buf();
1011
1012 /*
1013 * Need to conclude brk, before e820__memblock_setup()
1014 * it could use memblock_find_in_range, could overlap with
1015 * brk area.
1016 */
1017 reserve_brk();
1018
1019 cleanup_highmap();
1020
1021 memblock_set_current_limit(ISA_END_ADDRESS);
1022 e820__memblock_setup();
1023
1024 /*
1025 * Needs to run after memblock setup because it needs the physical
1026 * memory size.
1027 */
1028 sev_setup_arch();
1029
1030 efi_fake_memmap();
1031 efi_find_mirror();
1032 efi_esrt_init();
1033 efi_mokvar_table_init();
1034
1035 /*
1036 * The EFI specification says that boot service code won't be
1037 * called after ExitBootServices(). This is, in fact, a lie.
1038 */
1039 efi_reserve_boot_services();
1040
1041 /* preallocate 4k for mptable mpc */
1042 e820__memblock_alloc_reserved_mpc_new();
1043
1044 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1045 setup_bios_corruption_check();
1046 #endif
1047
1048 #ifdef CONFIG_X86_32
1049 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1050 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1051 #endif
1052
1053 /*
1054 * Find free memory for the real mode trampoline and place it there. If
1055 * there is not enough free memory under 1M, on EFI-enabled systems
1056 * there will be additional attempt to reclaim the memory for the real
1057 * mode trampoline at efi_free_boot_services().
1058 *
1059 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1060 * are known to corrupt low memory and several hundred kilobytes are not
1061 * worth complex detection what memory gets clobbered. Windows does the
1062 * same thing for very similar reasons.
1063 *
1064 * Moreover, on machines with SandyBridge graphics or in setups that use
1065 * crashkernel the entire 1M is reserved anyway.
1066 */
1067 reserve_real_mode();
1068
1069 init_mem_mapping();
1070
1071 idt_setup_early_pf();
1072
1073 /*
1074 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1075 * with the current CR4 value. This may not be necessary, but
1076 * auditing all the early-boot CR4 manipulation would be needed to
1077 * rule it out.
1078 *
1079 * Mask off features that don't work outside long mode (just
1080 * PCIDE for now).
1081 */
1082 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1083
1084 memblock_set_current_limit(get_max_mapped());
1085
1086 /*
1087 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1088 */
1089
1090 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1091 if (init_ohci1394_dma_early)
1092 init_ohci1394_dma_on_all_controllers();
1093 #endif
1094 /* Allocate bigger log buffer */
1095 setup_log_buf(1);
1096
1097 if (efi_enabled(EFI_BOOT)) {
1098 switch (boot_params.secure_boot) {
1099 case efi_secureboot_mode_disabled:
1100 pr_info("Secure boot disabled\n");
1101 break;
1102 case efi_secureboot_mode_enabled:
1103 pr_info("Secure boot enabled\n");
1104 break;
1105 default:
1106 pr_info("Secure boot could not be determined\n");
1107 break;
1108 }
1109 }
1110
1111 reserve_initrd();
1112
1113 acpi_table_upgrade();
1114 /* Look for ACPI tables and reserve memory occupied by them. */
1115 acpi_boot_table_init();
1116
1117 vsmp_init();
1118
1119 io_delay_init();
1120
1121 early_platform_quirks();
1122
1123 early_acpi_boot_init();
1124
1125 initmem_init();
1126 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1127
1128 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1129 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1130
1131 /*
1132 * Reserve memory for crash kernel after SRAT is parsed so that it
1133 * won't consume hotpluggable memory.
1134 */
1135 reserve_crashkernel();
1136
1137 memblock_find_dma_reserve();
1138
1139 if (!early_xdbc_setup_hardware())
1140 early_xdbc_register_console();
1141
1142 x86_init.paging.pagetable_init();
1143
1144 kasan_init();
1145
1146 /*
1147 * Sync back kernel address range.
1148 *
1149 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1150 * this call?
1151 */
1152 sync_initial_page_table();
1153
1154 tboot_probe();
1155
1156 map_vsyscall();
1157
1158 generic_apic_probe();
1159
1160 early_quirks();
1161
1162 /*
1163 * Read APIC and some other early information from ACPI tables.
1164 */
1165 acpi_boot_init();
1166 x86_dtb_init();
1167
1168 /*
1169 * get boot-time SMP configuration:
1170 */
1171 get_smp_config();
1172
1173 /*
1174 * Systems w/o ACPI and mptables might not have it mapped the local
1175 * APIC yet, but prefill_possible_map() might need to access it.
1176 */
1177 init_apic_mappings();
1178
1179 prefill_possible_map();
1180
1181 init_cpu_to_node();
1182 init_gi_nodes();
1183
1184 io_apic_init_mappings();
1185
1186 x86_init.hyper.guest_late_init();
1187
1188 e820__reserve_resources();
1189 e820__register_nosave_regions(max_pfn);
1190
1191 x86_init.resources.reserve_resources();
1192
1193 e820__setup_pci_gap();
1194
1195 #ifdef CONFIG_VT
1196 #if defined(CONFIG_VGA_CONSOLE)
1197 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1198 conswitchp = &vga_con;
1199 #endif
1200 #endif
1201 x86_init.oem.banner();
1202
1203 x86_init.timers.wallclock_init();
1204
1205 /*
1206 * This needs to run before setup_local_APIC() which soft-disables the
1207 * local APIC temporarily and that masks the thermal LVT interrupt,
1208 * leading to softlockups on machines which have configured SMI
1209 * interrupt delivery.
1210 */
1211 therm_lvt_init();
1212
1213 mcheck_init();
1214
1215 register_refined_jiffies(CLOCK_TICK_RATE);
1216
1217 #ifdef CONFIG_EFI
1218 if (efi_enabled(EFI_BOOT))
1219 efi_apply_memmap_quirks();
1220 #endif
1221
1222 unwind_init();
1223 }
1224
1225 #ifdef CONFIG_X86_32
1226
1227 static struct resource video_ram_resource = {
1228 .name = "Video RAM area",
1229 .start = 0xa0000,
1230 .end = 0xbffff,
1231 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1232 };
1233
i386_reserve_resources(void)1234 void __init i386_reserve_resources(void)
1235 {
1236 request_resource(&iomem_resource, &video_ram_resource);
1237 reserve_standard_io_resources();
1238 }
1239
1240 #endif /* CONFIG_X86_32 */
1241
1242 static struct notifier_block kernel_offset_notifier = {
1243 .notifier_call = dump_kernel_offset
1244 };
1245
register_kernel_offset_dumper(void)1246 static int __init register_kernel_offset_dumper(void)
1247 {
1248 atomic_notifier_chain_register(&panic_notifier_list,
1249 &kernel_offset_notifier);
1250 return 0;
1251 }
1252 __initcall(register_kernel_offset_dumper);
1253