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/console.h>
9 #include <linux/crash_dump.h>
10 #include <linux/dma-map-ops.h>
11 #include <linux/dmi.h>
12 #include <linux/efi.h>
13 #include <linux/init_ohci1394_dma.h>
14 #include <linux/initrd.h>
15 #include <linux/iscsi_ibft.h>
16 #include <linux/memblock.h>
17 #include <linux/pci.h>
18 #include <linux/root_dev.h>
19 #include <linux/hugetlb.h>
20 #include <linux/tboot.h>
21 #include <linux/usb/xhci-dbgp.h>
22 #include <linux/static_call.h>
23 #include <linux/swiotlb.h>
24
25 #include <uapi/linux/mount.h>
26
27 #include <xen/xen.h>
28
29 #include <asm/apic.h>
30 #include <asm/numa.h>
31 #include <asm/bios_ebda.h>
32 #include <asm/bugs.h>
33 #include <asm/cpu.h>
34 #include <asm/efi.h>
35 #include <asm/gart.h>
36 #include <asm/hypervisor.h>
37 #include <asm/io_apic.h>
38 #include <asm/kasan.h>
39 #include <asm/kaslr.h>
40 #include <asm/mce.h>
41 #include <asm/mtrr.h>
42 #include <asm/realmode.h>
43 #include <asm/olpc_ofw.h>
44 #include <asm/pci-direct.h>
45 #include <asm/prom.h>
46 #include <asm/proto.h>
47 #include <asm/unwind.h>
48 #include <asm/vsyscall.h>
49 #include <linux/vmalloc.h>
50
51 /*
52 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
53 * max_pfn_mapped: highest directly mapped pfn > 4 GB
54 *
55 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
56 * represented by pfn_mapped[].
57 */
58 unsigned long max_low_pfn_mapped;
59 unsigned long max_pfn_mapped;
60
61 #ifdef CONFIG_DMI
62 RESERVE_BRK(dmi_alloc, 65536);
63 #endif
64
65
66 /*
67 * Range of the BSS area. The size of the BSS area is determined
68 * at link time, with RESERVE_BRK() facility reserving additional
69 * chunks.
70 */
71 unsigned long _brk_start = (unsigned long)__brk_base;
72 unsigned long _brk_end = (unsigned long)__brk_base;
73
74 struct boot_params boot_params;
75
76 /*
77 * These are the four main kernel memory regions, we put them into
78 * the resource tree so that kdump tools and other debugging tools
79 * recover it:
80 */
81
82 static struct resource rodata_resource = {
83 .name = "Kernel rodata",
84 .start = 0,
85 .end = 0,
86 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
87 };
88
89 static struct resource data_resource = {
90 .name = "Kernel data",
91 .start = 0,
92 .end = 0,
93 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
94 };
95
96 static struct resource code_resource = {
97 .name = "Kernel code",
98 .start = 0,
99 .end = 0,
100 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
101 };
102
103 static struct resource bss_resource = {
104 .name = "Kernel bss",
105 .start = 0,
106 .end = 0,
107 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
108 };
109
110
111 #ifdef CONFIG_X86_32
112 /* CPU data as detected by the assembly code in head_32.S */
113 struct cpuinfo_x86 new_cpu_data;
114
115 /* Common CPU data for all CPUs */
116 struct cpuinfo_x86 boot_cpu_data __read_mostly;
117 EXPORT_SYMBOL(boot_cpu_data);
118
119 unsigned int def_to_bigsmp;
120
121 struct apm_info apm_info;
122 EXPORT_SYMBOL(apm_info);
123
124 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
125 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
126 struct ist_info ist_info;
127 EXPORT_SYMBOL(ist_info);
128 #else
129 struct ist_info ist_info;
130 #endif
131
132 #else
133 struct cpuinfo_x86 boot_cpu_data __read_mostly;
134 EXPORT_SYMBOL(boot_cpu_data);
135 #endif
136
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 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
145 int bootloader_type, bootloader_version;
146
147 /*
148 * Setup options
149 */
150 struct screen_info screen_info;
151 EXPORT_SYMBOL(screen_info);
152 struct edid_info edid_info;
153 EXPORT_SYMBOL_GPL(edid_info);
154
155 extern int root_mountflags;
156
157 unsigned long saved_video_mode;
158
159 #define RAMDISK_IMAGE_START_MASK 0x07FF
160 #define RAMDISK_PROMPT_FLAG 0x8000
161 #define RAMDISK_LOAD_FLAG 0x4000
162
163 static char __initdata command_line[COMMAND_LINE_SIZE];
164 #ifdef CONFIG_CMDLINE_BOOL
165 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
166 #endif
167
168 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
169 struct edd edd;
170 #ifdef CONFIG_EDD_MODULE
171 EXPORT_SYMBOL(edd);
172 #endif
173 /**
174 * copy_edd() - Copy the BIOS EDD information
175 * from boot_params into a safe place.
176 *
177 */
copy_edd(void)178 static inline void __init copy_edd(void)
179 {
180 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
181 sizeof(edd.mbr_signature));
182 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
183 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
184 edd.edd_info_nr = boot_params.eddbuf_entries;
185 }
186 #else
copy_edd(void)187 static inline void __init copy_edd(void)
188 {
189 }
190 #endif
191
extend_brk(size_t size,size_t align)192 void * __init extend_brk(size_t size, size_t align)
193 {
194 size_t mask = align - 1;
195 void *ret;
196
197 BUG_ON(_brk_start == 0);
198 BUG_ON(align & mask);
199
200 _brk_end = (_brk_end + mask) & ~mask;
201 BUG_ON((char *)(_brk_end + size) > __brk_limit);
202
203 ret = (void *)_brk_end;
204 _brk_end += size;
205
206 memset(ret, 0, size);
207
208 return ret;
209 }
210
211 #ifdef CONFIG_X86_32
cleanup_highmap(void)212 static void __init cleanup_highmap(void)
213 {
214 }
215 #endif
216
reserve_brk(void)217 static void __init reserve_brk(void)
218 {
219 if (_brk_end > _brk_start)
220 memblock_reserve(__pa_symbol(_brk_start),
221 _brk_end - _brk_start);
222
223 /* Mark brk area as locked down and no longer taking any
224 new allocations */
225 _brk_start = 0;
226 }
227
228 u64 relocated_ramdisk;
229
230 #ifdef CONFIG_BLK_DEV_INITRD
231
get_ramdisk_image(void)232 static u64 __init get_ramdisk_image(void)
233 {
234 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
235
236 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
237
238 if (ramdisk_image == 0)
239 ramdisk_image = phys_initrd_start;
240
241 return ramdisk_image;
242 }
get_ramdisk_size(void)243 static u64 __init get_ramdisk_size(void)
244 {
245 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
246
247 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
248
249 if (ramdisk_size == 0)
250 ramdisk_size = phys_initrd_size;
251
252 return ramdisk_size;
253 }
254
relocate_initrd(void)255 static void __init relocate_initrd(void)
256 {
257 /* Assume only end is not page aligned */
258 u64 ramdisk_image = get_ramdisk_image();
259 u64 ramdisk_size = get_ramdisk_size();
260 u64 area_size = PAGE_ALIGN(ramdisk_size);
261
262 /* We need to move the initrd down into directly mapped mem */
263 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
264 PFN_PHYS(max_pfn_mapped));
265 if (!relocated_ramdisk)
266 panic("Cannot find place for new RAMDISK of size %lld\n",
267 ramdisk_size);
268
269 initrd_start = relocated_ramdisk + PAGE_OFFSET;
270 initrd_end = initrd_start + ramdisk_size;
271 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
272 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
273
274 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
275
276 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
277 " [mem %#010llx-%#010llx]\n",
278 ramdisk_image, ramdisk_image + ramdisk_size - 1,
279 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
280 }
281
early_reserve_initrd(void)282 static void __init early_reserve_initrd(void)
283 {
284 /* Assume only end is not page aligned */
285 u64 ramdisk_image = get_ramdisk_image();
286 u64 ramdisk_size = get_ramdisk_size();
287 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
288
289 if (!boot_params.hdr.type_of_loader ||
290 !ramdisk_image || !ramdisk_size)
291 return; /* No initrd provided by bootloader */
292
293 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
294 }
295
reserve_initrd(void)296 static void __init reserve_initrd(void)
297 {
298 /* Assume only end is not page aligned */
299 u64 ramdisk_image = get_ramdisk_image();
300 u64 ramdisk_size = get_ramdisk_size();
301 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
302
303 if (!boot_params.hdr.type_of_loader ||
304 !ramdisk_image || !ramdisk_size)
305 return; /* No initrd provided by bootloader */
306
307 initrd_start = 0;
308
309 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
310 ramdisk_end - 1);
311
312 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
313 PFN_DOWN(ramdisk_end))) {
314 /* All are mapped, easy case */
315 initrd_start = ramdisk_image + PAGE_OFFSET;
316 initrd_end = initrd_start + ramdisk_size;
317 return;
318 }
319
320 relocate_initrd();
321
322 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
323 }
324
325 #else
early_reserve_initrd(void)326 static void __init early_reserve_initrd(void)
327 {
328 }
reserve_initrd(void)329 static void __init reserve_initrd(void)
330 {
331 }
332 #endif /* CONFIG_BLK_DEV_INITRD */
333
parse_setup_data(void)334 static void __init parse_setup_data(void)
335 {
336 struct setup_data *data;
337 u64 pa_data, pa_next;
338
339 pa_data = boot_params.hdr.setup_data;
340 while (pa_data) {
341 u32 data_len, data_type;
342
343 data = early_memremap(pa_data, sizeof(*data));
344 data_len = data->len + sizeof(struct setup_data);
345 data_type = data->type;
346 pa_next = data->next;
347 early_memunmap(data, sizeof(*data));
348
349 switch (data_type) {
350 case SETUP_E820_EXT:
351 e820__memory_setup_extended(pa_data, data_len);
352 break;
353 case SETUP_DTB:
354 add_dtb(pa_data);
355 break;
356 case SETUP_EFI:
357 parse_efi_setup(pa_data, data_len);
358 break;
359 default:
360 break;
361 }
362 pa_data = pa_next;
363 }
364 }
365
memblock_x86_reserve_range_setup_data(void)366 static void __init memblock_x86_reserve_range_setup_data(void)
367 {
368 struct setup_data *data;
369 u64 pa_data;
370
371 pa_data = boot_params.hdr.setup_data;
372 while (pa_data) {
373 data = early_memremap(pa_data, sizeof(*data));
374 memblock_reserve(pa_data, sizeof(*data) + data->len);
375
376 if (data->type == SETUP_INDIRECT &&
377 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
378 memblock_reserve(((struct setup_indirect *)data->data)->addr,
379 ((struct setup_indirect *)data->data)->len);
380
381 pa_data = data->next;
382 early_memunmap(data, sizeof(*data));
383 }
384 }
385
386 /*
387 * --------- Crashkernel reservation ------------------------------
388 */
389
390 #ifdef CONFIG_KEXEC_CORE
391
392 /* 16M alignment for crash kernel regions */
393 #define CRASH_ALIGN SZ_16M
394
395 /*
396 * Keep the crash kernel below this limit.
397 *
398 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
399 * due to mapping restrictions.
400 *
401 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
402 * the upper limit of system RAM in 4-level paging mode. Since the kdump
403 * jump could be from 5-level paging to 4-level paging, the jump will fail if
404 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
405 * no good way to detect the paging mode of the target kernel which will be
406 * loaded for dumping.
407 */
408 #ifdef CONFIG_X86_32
409 # define CRASH_ADDR_LOW_MAX SZ_512M
410 # define CRASH_ADDR_HIGH_MAX SZ_512M
411 #else
412 # define CRASH_ADDR_LOW_MAX SZ_4G
413 # define CRASH_ADDR_HIGH_MAX SZ_64T
414 #endif
415
reserve_crashkernel_low(void)416 static int __init reserve_crashkernel_low(void)
417 {
418 #ifdef CONFIG_X86_64
419 unsigned long long base, low_base = 0, low_size = 0;
420 unsigned long low_mem_limit;
421 int ret;
422
423 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
424
425 /* crashkernel=Y,low */
426 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
427 if (ret) {
428 /*
429 * two parts from kernel/dma/swiotlb.c:
430 * -swiotlb size: user-specified with swiotlb= or default.
431 *
432 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
433 * to 8M for other buffers that may need to stay low too. Also
434 * make sure we allocate enough extra low memory so that we
435 * don't run out of DMA buffers for 32-bit devices.
436 */
437 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
438 } else {
439 /* passed with crashkernel=0,low ? */
440 if (!low_size)
441 return 0;
442 }
443
444 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
445 if (!low_base) {
446 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
447 (unsigned long)(low_size >> 20));
448 return -ENOMEM;
449 }
450
451 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
452 (unsigned long)(low_size >> 20),
453 (unsigned long)(low_base >> 20),
454 (unsigned long)(low_mem_limit >> 20));
455
456 crashk_low_res.start = low_base;
457 crashk_low_res.end = low_base + low_size - 1;
458 insert_resource(&iomem_resource, &crashk_low_res);
459 #endif
460 return 0;
461 }
462
reserve_crashkernel(void)463 static void __init reserve_crashkernel(void)
464 {
465 unsigned long long crash_size, crash_base, total_mem;
466 bool high = false;
467 int ret;
468
469 total_mem = memblock_phys_mem_size();
470
471 /* crashkernel=XM */
472 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
473 if (ret != 0 || crash_size <= 0) {
474 /* crashkernel=X,high */
475 ret = parse_crashkernel_high(boot_command_line, total_mem,
476 &crash_size, &crash_base);
477 if (ret != 0 || crash_size <= 0)
478 return;
479 high = true;
480 }
481
482 if (xen_pv_domain()) {
483 pr_info("Ignoring crashkernel for a Xen PV domain\n");
484 return;
485 }
486
487 /* 0 means: find the address automatically */
488 if (!crash_base) {
489 /*
490 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
491 * crashkernel=x,high reserves memory over 4G, also allocates
492 * 256M extra low memory for DMA buffers and swiotlb.
493 * But the extra memory is not required for all machines.
494 * So try low memory first and fall back to high memory
495 * unless "crashkernel=size[KMG],high" is specified.
496 */
497 if (!high)
498 crash_base = memblock_phys_alloc_range(crash_size,
499 CRASH_ALIGN, CRASH_ALIGN,
500 CRASH_ADDR_LOW_MAX);
501 if (!crash_base)
502 crash_base = memblock_phys_alloc_range(crash_size,
503 CRASH_ALIGN, CRASH_ALIGN,
504 CRASH_ADDR_HIGH_MAX);
505 if (!crash_base) {
506 pr_info("crashkernel reservation failed - No suitable area found.\n");
507 return;
508 }
509 } else {
510 unsigned long long start;
511
512 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
513 crash_base + crash_size);
514 if (start != crash_base) {
515 pr_info("crashkernel reservation failed - memory is in use.\n");
516 return;
517 }
518 }
519
520 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
521 memblock_free(crash_base, crash_size);
522 return;
523 }
524
525 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
526 (unsigned long)(crash_size >> 20),
527 (unsigned long)(crash_base >> 20),
528 (unsigned long)(total_mem >> 20));
529
530 crashk_res.start = crash_base;
531 crashk_res.end = crash_base + crash_size - 1;
532 insert_resource(&iomem_resource, &crashk_res);
533 }
534 #else
reserve_crashkernel(void)535 static void __init reserve_crashkernel(void)
536 {
537 }
538 #endif
539
540 static struct resource standard_io_resources[] = {
541 { .name = "dma1", .start = 0x00, .end = 0x1f,
542 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
543 { .name = "pic1", .start = 0x20, .end = 0x21,
544 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
545 { .name = "timer0", .start = 0x40, .end = 0x43,
546 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
547 { .name = "timer1", .start = 0x50, .end = 0x53,
548 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
549 { .name = "keyboard", .start = 0x60, .end = 0x60,
550 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
551 { .name = "keyboard", .start = 0x64, .end = 0x64,
552 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
553 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
554 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
555 { .name = "pic2", .start = 0xa0, .end = 0xa1,
556 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
557 { .name = "dma2", .start = 0xc0, .end = 0xdf,
558 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
559 { .name = "fpu", .start = 0xf0, .end = 0xff,
560 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
561 };
562
reserve_standard_io_resources(void)563 void __init reserve_standard_io_resources(void)
564 {
565 int i;
566
567 /* request I/O space for devices used on all i[345]86 PCs */
568 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
569 request_resource(&ioport_resource, &standard_io_resources[i]);
570
571 }
572
reserve_ibft_region(void)573 static __init void reserve_ibft_region(void)
574 {
575 unsigned long addr, size = 0;
576
577 addr = find_ibft_region(&size);
578
579 if (size)
580 memblock_reserve(addr, size);
581 }
582
snb_gfx_workaround_needed(void)583 static bool __init snb_gfx_workaround_needed(void)
584 {
585 #ifdef CONFIG_PCI
586 int i;
587 u16 vendor, devid;
588 static const __initconst u16 snb_ids[] = {
589 0x0102,
590 0x0112,
591 0x0122,
592 0x0106,
593 0x0116,
594 0x0126,
595 0x010a,
596 };
597
598 /* Assume no if something weird is going on with PCI */
599 if (!early_pci_allowed())
600 return false;
601
602 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
603 if (vendor != 0x8086)
604 return false;
605
606 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
607 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
608 if (devid == snb_ids[i])
609 return true;
610 #endif
611
612 return false;
613 }
614
615 /*
616 * Sandy Bridge graphics has trouble with certain ranges, exclude
617 * them from allocation.
618 */
trim_snb_memory(void)619 static void __init trim_snb_memory(void)
620 {
621 static const __initconst unsigned long bad_pages[] = {
622 0x20050000,
623 0x20110000,
624 0x20130000,
625 0x20138000,
626 0x40004000,
627 };
628 int i;
629
630 if (!snb_gfx_workaround_needed())
631 return;
632
633 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
634
635 /*
636 * SandyBridge integrated graphics devices have a bug that prevents
637 * them from accessing certain memory ranges, namely anything below
638 * 1M and in the pages listed in bad_pages[] above.
639 *
640 * To avoid these pages being ever accessed by SNB gfx devices
641 * reserve all memory below the 1 MB mark and bad_pages that have
642 * not already been reserved at boot time.
643 */
644 memblock_reserve(0, 1<<20);
645
646 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
647 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
648 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
649 bad_pages[i]);
650 }
651 }
652
trim_bios_range(void)653 static void __init trim_bios_range(void)
654 {
655 /*
656 * A special case is the first 4Kb of memory;
657 * This is a BIOS owned area, not kernel ram, but generally
658 * not listed as such in the E820 table.
659 *
660 * This typically reserves additional memory (64KiB by default)
661 * since some BIOSes are known to corrupt low memory. See the
662 * Kconfig help text for X86_RESERVE_LOW.
663 */
664 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
665
666 /*
667 * special case: Some BIOSes report the PC BIOS
668 * area (640Kb -> 1Mb) as RAM even though it is not.
669 * take them out.
670 */
671 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
672
673 e820__update_table(e820_table);
674 }
675
676 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)677 static void __init e820_add_kernel_range(void)
678 {
679 u64 start = __pa_symbol(_text);
680 u64 size = __pa_symbol(_end) - start;
681
682 /*
683 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
684 * attempt to fix it by adding the range. We may have a confused BIOS,
685 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
686 * exclude kernel range. If we really are running on top non-RAM,
687 * we will crash later anyways.
688 */
689 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
690 return;
691
692 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
693 e820__range_remove(start, size, E820_TYPE_RAM, 0);
694 e820__range_add(start, size, E820_TYPE_RAM);
695 }
696
697 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
698
parse_reservelow(char * p)699 static int __init parse_reservelow(char *p)
700 {
701 unsigned long long size;
702
703 if (!p)
704 return -EINVAL;
705
706 size = memparse(p, &p);
707
708 if (size < 4096)
709 size = 4096;
710
711 if (size > 640*1024)
712 size = 640*1024;
713
714 reserve_low = size;
715
716 return 0;
717 }
718
719 early_param("reservelow", parse_reservelow);
720
early_reserve_memory(void)721 static void __init early_reserve_memory(void)
722 {
723 /*
724 * Reserve the memory occupied by the kernel between _text and
725 * __end_of_kernel_reserve symbols. Any kernel sections after the
726 * __end_of_kernel_reserve symbol must be explicitly reserved with a
727 * separate memblock_reserve() or they will be discarded.
728 */
729 memblock_reserve(__pa_symbol(_text),
730 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
731
732 /*
733 * The first 4Kb of memory is a BIOS owned area, but generally it is
734 * not listed as such in the E820 table.
735 *
736 * Reserve the first memory page and typically some additional
737 * memory (64KiB by default) since some BIOSes are known to corrupt
738 * low memory. See the Kconfig help text for X86_RESERVE_LOW.
739 *
740 * In addition, make sure page 0 is always reserved because on
741 * systems with L1TF its contents can be leaked to user processes.
742 */
743 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
744
745 early_reserve_initrd();
746
747 if (efi_enabled(EFI_BOOT))
748 efi_memblock_x86_reserve_range();
749
750 memblock_x86_reserve_range_setup_data();
751
752 reserve_ibft_region();
753 reserve_bios_regions();
754 }
755
756 /*
757 * Dump out kernel offset information on panic.
758 */
759 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)760 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
761 {
762 if (kaslr_enabled()) {
763 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
764 kaslr_offset(),
765 __START_KERNEL,
766 __START_KERNEL_map,
767 MODULES_VADDR-1);
768 } else {
769 pr_emerg("Kernel Offset: disabled\n");
770 }
771
772 return 0;
773 }
774
775 /*
776 * Determine if we were loaded by an EFI loader. If so, then we have also been
777 * passed the efi memmap, systab, etc., so we should use these data structures
778 * for initialization. Note, the efi init code path is determined by the
779 * global efi_enabled. This allows the same kernel image to be used on existing
780 * systems (with a traditional BIOS) as well as on EFI systems.
781 */
782 /*
783 * setup_arch - architecture-specific boot-time initializations
784 *
785 * Note: On x86_64, fixmaps are ready for use even before this is called.
786 */
787
setup_arch(char ** cmdline_p)788 void __init setup_arch(char **cmdline_p)
789 {
790 #ifdef CONFIG_X86_32
791 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
792
793 /*
794 * copy kernel address range established so far and switch
795 * to the proper swapper page table
796 */
797 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
798 initial_page_table + KERNEL_PGD_BOUNDARY,
799 KERNEL_PGD_PTRS);
800
801 load_cr3(swapper_pg_dir);
802 /*
803 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
804 * a cr3 based tlb flush, so the following __flush_tlb_all()
805 * will not flush anything because the CPU quirk which clears
806 * X86_FEATURE_PGE has not been invoked yet. Though due to the
807 * load_cr3() above the TLB has been flushed already. The
808 * quirk is invoked before subsequent calls to __flush_tlb_all()
809 * so proper operation is guaranteed.
810 */
811 __flush_tlb_all();
812 #else
813 printk(KERN_INFO "Command line: %s\n", boot_command_line);
814 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
815 #endif
816
817 /*
818 * If we have OLPC OFW, we might end up relocating the fixmap due to
819 * reserve_top(), so do this before touching the ioremap area.
820 */
821 olpc_ofw_detect();
822
823 idt_setup_early_traps();
824 early_cpu_init();
825 jump_label_init();
826 static_call_init();
827 early_ioremap_init();
828
829 setup_olpc_ofw_pgd();
830
831 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
832 screen_info = boot_params.screen_info;
833 edid_info = boot_params.edid_info;
834 #ifdef CONFIG_X86_32
835 apm_info.bios = boot_params.apm_bios_info;
836 ist_info = boot_params.ist_info;
837 #endif
838 saved_video_mode = boot_params.hdr.vid_mode;
839 bootloader_type = boot_params.hdr.type_of_loader;
840 if ((bootloader_type >> 4) == 0xe) {
841 bootloader_type &= 0xf;
842 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
843 }
844 bootloader_version = bootloader_type & 0xf;
845 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
846
847 #ifdef CONFIG_BLK_DEV_RAM
848 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
849 #endif
850 #ifdef CONFIG_EFI
851 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
852 EFI32_LOADER_SIGNATURE, 4)) {
853 set_bit(EFI_BOOT, &efi.flags);
854 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
855 EFI64_LOADER_SIGNATURE, 4)) {
856 set_bit(EFI_BOOT, &efi.flags);
857 set_bit(EFI_64BIT, &efi.flags);
858 }
859 #endif
860
861 x86_init.oem.arch_setup();
862
863 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
864 e820__memory_setup();
865 parse_setup_data();
866
867 copy_edd();
868
869 if (!boot_params.hdr.root_flags)
870 root_mountflags &= ~MS_RDONLY;
871 init_mm.start_code = (unsigned long) _text;
872 init_mm.end_code = (unsigned long) _etext;
873 init_mm.end_data = (unsigned long) _edata;
874 init_mm.brk = _brk_end;
875
876 code_resource.start = __pa_symbol(_text);
877 code_resource.end = __pa_symbol(_etext)-1;
878 rodata_resource.start = __pa_symbol(__start_rodata);
879 rodata_resource.end = __pa_symbol(__end_rodata)-1;
880 data_resource.start = __pa_symbol(_sdata);
881 data_resource.end = __pa_symbol(_edata)-1;
882 bss_resource.start = __pa_symbol(__bss_start);
883 bss_resource.end = __pa_symbol(__bss_stop)-1;
884
885 #ifdef CONFIG_CMDLINE_BOOL
886 #ifdef CONFIG_CMDLINE_OVERRIDE
887 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
888 #else
889 if (builtin_cmdline[0]) {
890 /* append boot loader cmdline to builtin */
891 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
892 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
893 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
894 }
895 #endif
896 #endif
897
898 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
899 *cmdline_p = command_line;
900
901 /*
902 * x86_configure_nx() is called before parse_early_param() to detect
903 * whether hardware doesn't support NX (so that the early EHCI debug
904 * console setup can safely call set_fixmap()). It may then be called
905 * again from within noexec_setup() during parsing early parameters
906 * to honor the respective command line option.
907 */
908 x86_configure_nx();
909
910 parse_early_param();
911
912 /*
913 * Do some memory reservations *before* memory is added to
914 * memblock, so memblock allocations won't overwrite it.
915 * Do it after early param, so we could get (unlikely) panic from
916 * serial.
917 *
918 * After this point everything still needed from the boot loader or
919 * firmware or kernel text should be early reserved or marked not
920 * RAM in e820. All other memory is free game.
921 */
922 early_reserve_memory();
923
924 #ifdef CONFIG_MEMORY_HOTPLUG
925 /*
926 * Memory used by the kernel cannot be hot-removed because Linux
927 * cannot migrate the kernel pages. When memory hotplug is
928 * enabled, we should prevent memblock from allocating memory
929 * for the kernel.
930 *
931 * ACPI SRAT records all hotpluggable memory ranges. But before
932 * SRAT is parsed, we don't know about it.
933 *
934 * The kernel image is loaded into memory at very early time. We
935 * cannot prevent this anyway. So on NUMA system, we set any
936 * node the kernel resides in as un-hotpluggable.
937 *
938 * Since on modern servers, one node could have double-digit
939 * gigabytes memory, we can assume the memory around the kernel
940 * image is also un-hotpluggable. So before SRAT is parsed, just
941 * allocate memory near the kernel image to try the best to keep
942 * the kernel away from hotpluggable memory.
943 */
944 if (movable_node_is_enabled())
945 memblock_set_bottom_up(true);
946 #endif
947
948 x86_report_nx();
949
950 if (acpi_mps_check()) {
951 #ifdef CONFIG_X86_LOCAL_APIC
952 disable_apic = 1;
953 #endif
954 setup_clear_cpu_cap(X86_FEATURE_APIC);
955 }
956
957 e820__reserve_setup_data();
958 e820__finish_early_params();
959
960 if (efi_enabled(EFI_BOOT))
961 efi_init();
962
963 dmi_setup();
964
965 /*
966 * VMware detection requires dmi to be available, so this
967 * needs to be done after dmi_setup(), for the boot CPU.
968 */
969 init_hypervisor_platform();
970
971 tsc_early_init();
972 x86_init.resources.probe_roms();
973
974 /* after parse_early_param, so could debug it */
975 insert_resource(&iomem_resource, &code_resource);
976 insert_resource(&iomem_resource, &rodata_resource);
977 insert_resource(&iomem_resource, &data_resource);
978 insert_resource(&iomem_resource, &bss_resource);
979
980 e820_add_kernel_range();
981 trim_bios_range();
982 #ifdef CONFIG_X86_32
983 if (ppro_with_ram_bug()) {
984 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
985 E820_TYPE_RESERVED);
986 e820__update_table(e820_table);
987 printk(KERN_INFO "fixed physical RAM map:\n");
988 e820__print_table("bad_ppro");
989 }
990 #else
991 early_gart_iommu_check();
992 #endif
993
994 /*
995 * partially used pages are not usable - thus
996 * we are rounding upwards:
997 */
998 max_pfn = e820__end_of_ram_pfn();
999
1000 /* update e820 for memory not covered by WB MTRRs */
1001 mtrr_bp_init();
1002 if (mtrr_trim_uncached_memory(max_pfn))
1003 max_pfn = e820__end_of_ram_pfn();
1004
1005 max_possible_pfn = max_pfn;
1006
1007 /*
1008 * This call is required when the CPU does not support PAT. If
1009 * mtrr_bp_init() invoked it already via pat_init() the call has no
1010 * effect.
1011 */
1012 init_cache_modes();
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
1033 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1034 #endif
1035
1036 /*
1037 * Find and reserve possible boot-time SMP configuration:
1038 */
1039 find_smp_config();
1040
1041 early_alloc_pgt_buf();
1042
1043 /*
1044 * Need to conclude brk, before e820__memblock_setup()
1045 * it could use memblock_find_in_range, could overlap with
1046 * brk area.
1047 */
1048 reserve_brk();
1049
1050 cleanup_highmap();
1051
1052 memblock_set_current_limit(ISA_END_ADDRESS);
1053 e820__memblock_setup();
1054
1055 /*
1056 * Needs to run after memblock setup because it needs the physical
1057 * memory size.
1058 */
1059 sev_setup_arch();
1060
1061 efi_fake_memmap();
1062 efi_find_mirror();
1063 efi_esrt_init();
1064 efi_mokvar_table_init();
1065
1066 /*
1067 * The EFI specification says that boot service code won't be
1068 * called after ExitBootServices(). This is, in fact, a lie.
1069 */
1070 efi_reserve_boot_services();
1071
1072 /* preallocate 4k for mptable mpc */
1073 e820__memblock_alloc_reserved_mpc_new();
1074
1075 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1076 setup_bios_corruption_check();
1077 #endif
1078
1079 #ifdef CONFIG_X86_32
1080 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1081 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1082 #endif
1083
1084 reserve_real_mode();
1085
1086 /*
1087 * Reserving memory causing GPU hangs on Sandy Bridge integrated
1088 * graphics devices should be done after we allocated memory under
1089 * 1M for the real mode trampoline.
1090 */
1091 trim_snb_memory();
1092
1093 init_mem_mapping();
1094
1095 idt_setup_early_pf();
1096
1097 /*
1098 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1099 * with the current CR4 value. This may not be necessary, but
1100 * auditing all the early-boot CR4 manipulation would be needed to
1101 * rule it out.
1102 *
1103 * Mask off features that don't work outside long mode (just
1104 * PCIDE for now).
1105 */
1106 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1107
1108 memblock_set_current_limit(get_max_mapped());
1109
1110 /*
1111 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1112 */
1113
1114 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1115 if (init_ohci1394_dma_early)
1116 init_ohci1394_dma_on_all_controllers();
1117 #endif
1118 /* Allocate bigger log buffer */
1119 setup_log_buf(1);
1120
1121 if (efi_enabled(EFI_BOOT)) {
1122 switch (boot_params.secure_boot) {
1123 case efi_secureboot_mode_disabled:
1124 pr_info("Secure boot disabled\n");
1125 break;
1126 case efi_secureboot_mode_enabled:
1127 pr_info("Secure boot enabled\n");
1128 break;
1129 default:
1130 pr_info("Secure boot could not be determined\n");
1131 break;
1132 }
1133 }
1134
1135 reserve_initrd();
1136
1137 acpi_table_upgrade();
1138 /* Look for ACPI tables and reserve memory occupied by them. */
1139 acpi_boot_table_init();
1140
1141 vsmp_init();
1142
1143 io_delay_init();
1144
1145 early_platform_quirks();
1146
1147 early_acpi_boot_init();
1148
1149 initmem_init();
1150 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1151
1152 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1153 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1154
1155 /*
1156 * Reserve memory for crash kernel after SRAT is parsed so that it
1157 * won't consume hotpluggable memory.
1158 */
1159 reserve_crashkernel();
1160
1161 memblock_find_dma_reserve();
1162
1163 if (!early_xdbc_setup_hardware())
1164 early_xdbc_register_console();
1165
1166 x86_init.paging.pagetable_init();
1167
1168 kasan_init();
1169
1170 /*
1171 * Sync back kernel address range.
1172 *
1173 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1174 * this call?
1175 */
1176 sync_initial_page_table();
1177
1178 tboot_probe();
1179
1180 map_vsyscall();
1181
1182 generic_apic_probe();
1183
1184 early_quirks();
1185
1186 /*
1187 * Read APIC and some other early information from ACPI tables.
1188 */
1189 acpi_boot_init();
1190 x86_dtb_init();
1191
1192 /*
1193 * get boot-time SMP configuration:
1194 */
1195 get_smp_config();
1196
1197 /*
1198 * Systems w/o ACPI and mptables might not have it mapped the local
1199 * APIC yet, but prefill_possible_map() might need to access it.
1200 */
1201 init_apic_mappings();
1202
1203 prefill_possible_map();
1204
1205 init_cpu_to_node();
1206 init_gi_nodes();
1207
1208 io_apic_init_mappings();
1209
1210 x86_init.hyper.guest_late_init();
1211
1212 e820__reserve_resources();
1213 e820__register_nosave_regions(max_pfn);
1214
1215 x86_init.resources.reserve_resources();
1216
1217 e820__setup_pci_gap();
1218
1219 #ifdef CONFIG_VT
1220 #if defined(CONFIG_VGA_CONSOLE)
1221 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1222 conswitchp = &vga_con;
1223 #endif
1224 #endif
1225 x86_init.oem.banner();
1226
1227 x86_init.timers.wallclock_init();
1228
1229 mcheck_init();
1230
1231 register_refined_jiffies(CLOCK_TICK_RATE);
1232
1233 #ifdef CONFIG_EFI
1234 if (efi_enabled(EFI_BOOT))
1235 efi_apply_memmap_quirks();
1236 #endif
1237
1238 unwind_init();
1239 }
1240
1241 #ifdef CONFIG_X86_32
1242
1243 static struct resource video_ram_resource = {
1244 .name = "Video RAM area",
1245 .start = 0xa0000,
1246 .end = 0xbffff,
1247 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1248 };
1249
i386_reserve_resources(void)1250 void __init i386_reserve_resources(void)
1251 {
1252 request_resource(&iomem_resource, &video_ram_resource);
1253 reserve_standard_io_resources();
1254 }
1255
1256 #endif /* CONFIG_X86_32 */
1257
1258 static struct notifier_block kernel_offset_notifier = {
1259 .notifier_call = dump_kernel_offset
1260 };
1261
register_kernel_offset_dumper(void)1262 static int __init register_kernel_offset_dumper(void)
1263 {
1264 atomic_notifier_chain_register(&panic_notifier_list,
1265 &kernel_offset_notifier);
1266 return 0;
1267 }
1268 __initcall(register_kernel_offset_dumper);
1269