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/sfi.h>
20 #include <linux/hugetlb.h>
21 #include <linux/tboot.h>
22 #include <linux/usb/xhci-dbgp.h>
23 #include <linux/static_call.h>
24 #include <linux/swiotlb.h>
25
26 #include <uapi/linux/mount.h>
27
28 #include <xen/xen.h>
29
30 #include <asm/apic.h>
31 #include <asm/numa.h>
32 #include <asm/bios_ebda.h>
33 #include <asm/bugs.h>
34 #include <asm/cpu.h>
35 #include <asm/efi.h>
36 #include <asm/gart.h>
37 #include <asm/hypervisor.h>
38 #include <asm/io_apic.h>
39 #include <asm/kasan.h>
40 #include <asm/kaslr.h>
41 #include <asm/mce.h>
42 #include <asm/mtrr.h>
43 #include <asm/realmode.h>
44 #include <asm/olpc_ofw.h>
45 #include <asm/pci-direct.h>
46 #include <asm/prom.h>
47 #include <asm/proto.h>
48 #include <asm/unwind.h>
49 #include <asm/vsyscall.h>
50 #include <linux/vmalloc.h>
51
52 /*
53 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
54 * max_pfn_mapped: highest directly mapped pfn > 4 GB
55 *
56 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
57 * represented by pfn_mapped[].
58 */
59 unsigned long max_low_pfn_mapped;
60 unsigned long max_pfn_mapped;
61
62 #ifdef CONFIG_DMI
63 RESERVE_BRK(dmi_alloc, 65536);
64 #endif
65
66
67 /*
68 * Range of the BSS area. The size of the BSS area is determined
69 * at link time, with RESERVE_BRK*() facility reserving additional
70 * chunks.
71 */
72 unsigned long _brk_start = (unsigned long)__brk_base;
73 unsigned long _brk_end = (unsigned long)__brk_base;
74
75 struct boot_params boot_params;
76
77 /*
78 * These are the four main kernel memory regions, we put them into
79 * the resource tree so that kdump tools and other debugging tools
80 * recover it:
81 */
82
83 static struct resource rodata_resource = {
84 .name = "Kernel rodata",
85 .start = 0,
86 .end = 0,
87 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
88 };
89
90 static struct resource data_resource = {
91 .name = "Kernel data",
92 .start = 0,
93 .end = 0,
94 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
95 };
96
97 static struct resource code_resource = {
98 .name = "Kernel code",
99 .start = 0,
100 .end = 0,
101 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
102 };
103
104 static struct resource bss_resource = {
105 .name = "Kernel bss",
106 .start = 0,
107 .end = 0,
108 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
109 };
110
111
112 #ifdef CONFIG_X86_32
113 /* CPU data as detected by the assembly code in head_32.S */
114 struct cpuinfo_x86 new_cpu_data;
115
116 /* Common CPU data for all CPUs */
117 struct cpuinfo_x86 boot_cpu_data __read_mostly;
118 EXPORT_SYMBOL(boot_cpu_data);
119
120 unsigned int def_to_bigsmp;
121
122 /* For MCA, but anyone else can use it if they want */
123 unsigned int machine_id;
124 unsigned int machine_submodel_id;
125 unsigned int BIOS_revision;
126
127 struct apm_info apm_info;
128 EXPORT_SYMBOL(apm_info);
129
130 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
131 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
132 struct ist_info ist_info;
133 EXPORT_SYMBOL(ist_info);
134 #else
135 struct ist_info ist_info;
136 #endif
137
138 #else
139 struct cpuinfo_x86 boot_cpu_data __read_mostly;
140 EXPORT_SYMBOL(boot_cpu_data);
141 #endif
142
143
144 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
145 __visible unsigned long mmu_cr4_features __ro_after_init;
146 #else
147 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
148 #endif
149
150 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
151 int bootloader_type, bootloader_version;
152
153 /*
154 * Setup options
155 */
156 struct screen_info screen_info;
157 EXPORT_SYMBOL(screen_info);
158 struct edid_info edid_info;
159 EXPORT_SYMBOL_GPL(edid_info);
160
161 extern int root_mountflags;
162
163 unsigned long saved_video_mode;
164
165 #define RAMDISK_IMAGE_START_MASK 0x07FF
166 #define RAMDISK_PROMPT_FLAG 0x8000
167 #define RAMDISK_LOAD_FLAG 0x4000
168
169 static char __initdata command_line[COMMAND_LINE_SIZE];
170 #ifdef CONFIG_CMDLINE_BOOL
171 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
172 #endif
173
174 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
175 struct edd edd;
176 #ifdef CONFIG_EDD_MODULE
177 EXPORT_SYMBOL(edd);
178 #endif
179 /**
180 * copy_edd() - Copy the BIOS EDD information
181 * from boot_params into a safe place.
182 *
183 */
copy_edd(void)184 static inline void __init copy_edd(void)
185 {
186 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
187 sizeof(edd.mbr_signature));
188 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
189 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
190 edd.edd_info_nr = boot_params.eddbuf_entries;
191 }
192 #else
copy_edd(void)193 static inline void __init copy_edd(void)
194 {
195 }
196 #endif
197
extend_brk(size_t size,size_t align)198 void * __init extend_brk(size_t size, size_t align)
199 {
200 size_t mask = align - 1;
201 void *ret;
202
203 BUG_ON(_brk_start == 0);
204 BUG_ON(align & mask);
205
206 _brk_end = (_brk_end + mask) & ~mask;
207 BUG_ON((char *)(_brk_end + size) > __brk_limit);
208
209 ret = (void *)_brk_end;
210 _brk_end += size;
211
212 memset(ret, 0, size);
213
214 return ret;
215 }
216
217 #ifdef CONFIG_X86_32
cleanup_highmap(void)218 static void __init cleanup_highmap(void)
219 {
220 }
221 #endif
222
reserve_brk(void)223 static void __init reserve_brk(void)
224 {
225 if (_brk_end > _brk_start)
226 memblock_reserve(__pa_symbol(_brk_start),
227 _brk_end - _brk_start);
228
229 /* Mark brk area as locked down and no longer taking any
230 new allocations */
231 _brk_start = 0;
232 }
233
234 u64 relocated_ramdisk;
235
236 #ifdef CONFIG_BLK_DEV_INITRD
237
get_ramdisk_image(void)238 static u64 __init get_ramdisk_image(void)
239 {
240 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
241
242 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
243
244 if (ramdisk_image == 0)
245 ramdisk_image = phys_initrd_start;
246
247 return ramdisk_image;
248 }
get_ramdisk_size(void)249 static u64 __init get_ramdisk_size(void)
250 {
251 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
252
253 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
254
255 if (ramdisk_size == 0)
256 ramdisk_size = phys_initrd_size;
257
258 return ramdisk_size;
259 }
260
relocate_initrd(void)261 static void __init relocate_initrd(void)
262 {
263 /* Assume only end is not page aligned */
264 u64 ramdisk_image = get_ramdisk_image();
265 u64 ramdisk_size = get_ramdisk_size();
266 u64 area_size = PAGE_ALIGN(ramdisk_size);
267
268 /* We need to move the initrd down into directly mapped mem */
269 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
270 PFN_PHYS(max_pfn_mapped));
271 if (!relocated_ramdisk)
272 panic("Cannot find place for new RAMDISK of size %lld\n",
273 ramdisk_size);
274
275 initrd_start = relocated_ramdisk + PAGE_OFFSET;
276 initrd_end = initrd_start + ramdisk_size;
277 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
278 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
279
280 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
281
282 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
283 " [mem %#010llx-%#010llx]\n",
284 ramdisk_image, ramdisk_image + ramdisk_size - 1,
285 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
286 }
287
early_reserve_initrd(void)288 static void __init early_reserve_initrd(void)
289 {
290 /* Assume only end is not page aligned */
291 u64 ramdisk_image = get_ramdisk_image();
292 u64 ramdisk_size = get_ramdisk_size();
293 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
294
295 if (!boot_params.hdr.type_of_loader ||
296 !ramdisk_image || !ramdisk_size)
297 return; /* No initrd provided by bootloader */
298
299 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
300 }
301
reserve_initrd(void)302 static void __init reserve_initrd(void)
303 {
304 /* Assume only end is not page aligned */
305 u64 ramdisk_image = get_ramdisk_image();
306 u64 ramdisk_size = get_ramdisk_size();
307 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
308
309 if (!boot_params.hdr.type_of_loader ||
310 !ramdisk_image || !ramdisk_size)
311 return; /* No initrd provided by bootloader */
312
313 initrd_start = 0;
314
315 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
316 ramdisk_end - 1);
317
318 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
319 PFN_DOWN(ramdisk_end))) {
320 /* All are mapped, easy case */
321 initrd_start = ramdisk_image + PAGE_OFFSET;
322 initrd_end = initrd_start + ramdisk_size;
323 return;
324 }
325
326 relocate_initrd();
327
328 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
329 }
330
331 #else
early_reserve_initrd(void)332 static void __init early_reserve_initrd(void)
333 {
334 }
reserve_initrd(void)335 static void __init reserve_initrd(void)
336 {
337 }
338 #endif /* CONFIG_BLK_DEV_INITRD */
339
parse_setup_data(void)340 static void __init parse_setup_data(void)
341 {
342 struct setup_data *data;
343 u64 pa_data, pa_next;
344
345 pa_data = boot_params.hdr.setup_data;
346 while (pa_data) {
347 u32 data_len, data_type;
348
349 data = early_memremap(pa_data, sizeof(*data));
350 data_len = data->len + sizeof(struct setup_data);
351 data_type = data->type;
352 pa_next = data->next;
353 early_memunmap(data, sizeof(*data));
354
355 switch (data_type) {
356 case SETUP_E820_EXT:
357 e820__memory_setup_extended(pa_data, data_len);
358 break;
359 case SETUP_DTB:
360 add_dtb(pa_data);
361 break;
362 case SETUP_EFI:
363 parse_efi_setup(pa_data, data_len);
364 break;
365 default:
366 break;
367 }
368 pa_data = pa_next;
369 }
370 }
371
memblock_x86_reserve_range_setup_data(void)372 static void __init memblock_x86_reserve_range_setup_data(void)
373 {
374 struct setup_indirect *indirect;
375 struct setup_data *data;
376 u64 pa_data, pa_next;
377 u32 len;
378
379 pa_data = boot_params.hdr.setup_data;
380 while (pa_data) {
381 data = early_memremap(pa_data, sizeof(*data));
382 if (!data) {
383 pr_warn("setup: failed to memremap setup_data entry\n");
384 return;
385 }
386
387 len = sizeof(*data);
388 pa_next = data->next;
389
390 memblock_reserve(pa_data, sizeof(*data) + data->len);
391
392 if (data->type == SETUP_INDIRECT) {
393 len += data->len;
394 early_memunmap(data, sizeof(*data));
395 data = early_memremap(pa_data, len);
396 if (!data) {
397 pr_warn("setup: failed to memremap indirect setup_data\n");
398 return;
399 }
400
401 indirect = (struct setup_indirect *)data->data;
402
403 if (indirect->type != SETUP_INDIRECT)
404 memblock_reserve(indirect->addr, indirect->len);
405 }
406
407 pa_data = pa_next;
408 early_memunmap(data, len);
409 }
410 }
411
412 /*
413 * --------- Crashkernel reservation ------------------------------
414 */
415
416 #ifdef CONFIG_KEXEC_CORE
417
418 /* 16M alignment for crash kernel regions */
419 #define CRASH_ALIGN SZ_16M
420
421 /*
422 * Keep the crash kernel below this limit.
423 *
424 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
425 * due to mapping restrictions.
426 *
427 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
428 * the upper limit of system RAM in 4-level paging mode. Since the kdump
429 * jump could be from 5-level paging to 4-level paging, the jump will fail if
430 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
431 * no good way to detect the paging mode of the target kernel which will be
432 * loaded for dumping.
433 */
434 #ifdef CONFIG_X86_32
435 # define CRASH_ADDR_LOW_MAX SZ_512M
436 # define CRASH_ADDR_HIGH_MAX SZ_512M
437 #else
438 # define CRASH_ADDR_LOW_MAX SZ_4G
439 # define CRASH_ADDR_HIGH_MAX SZ_64T
440 #endif
441
reserve_crashkernel_low(void)442 static int __init reserve_crashkernel_low(void)
443 {
444 #ifdef CONFIG_X86_64
445 unsigned long long base, low_base = 0, low_size = 0;
446 unsigned long low_mem_limit;
447 int ret;
448
449 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
450
451 /* crashkernel=Y,low */
452 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
453 if (ret) {
454 /*
455 * two parts from kernel/dma/swiotlb.c:
456 * -swiotlb size: user-specified with swiotlb= or default.
457 *
458 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
459 * to 8M for other buffers that may need to stay low too. Also
460 * make sure we allocate enough extra low memory so that we
461 * don't run out of DMA buffers for 32-bit devices.
462 */
463 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
464 } else {
465 /* passed with crashkernel=0,low ? */
466 if (!low_size)
467 return 0;
468 }
469
470 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
471 if (!low_base) {
472 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
473 (unsigned long)(low_size >> 20));
474 return -ENOMEM;
475 }
476
477 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
478 (unsigned long)(low_size >> 20),
479 (unsigned long)(low_base >> 20),
480 (unsigned long)(low_mem_limit >> 20));
481
482 crashk_low_res.start = low_base;
483 crashk_low_res.end = low_base + low_size - 1;
484 insert_resource(&iomem_resource, &crashk_low_res);
485 #endif
486 return 0;
487 }
488
reserve_crashkernel(void)489 static void __init reserve_crashkernel(void)
490 {
491 unsigned long long crash_size, crash_base, total_mem;
492 bool high = false;
493 int ret;
494
495 total_mem = memblock_phys_mem_size();
496
497 /* crashkernel=XM */
498 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
499 if (ret != 0 || crash_size <= 0) {
500 /* crashkernel=X,high */
501 ret = parse_crashkernel_high(boot_command_line, total_mem,
502 &crash_size, &crash_base);
503 if (ret != 0 || crash_size <= 0)
504 return;
505 high = true;
506 }
507
508 if (xen_pv_domain()) {
509 pr_info("Ignoring crashkernel for a Xen PV domain\n");
510 return;
511 }
512
513 /* 0 means: find the address automatically */
514 if (!crash_base) {
515 /*
516 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
517 * crashkernel=x,high reserves memory over 4G, also allocates
518 * 256M extra low memory for DMA buffers and swiotlb.
519 * But the extra memory is not required for all machines.
520 * So try low memory first and fall back to high memory
521 * unless "crashkernel=size[KMG],high" is specified.
522 */
523 if (!high)
524 crash_base = memblock_phys_alloc_range(crash_size,
525 CRASH_ALIGN, CRASH_ALIGN,
526 CRASH_ADDR_LOW_MAX);
527 if (!crash_base)
528 crash_base = memblock_phys_alloc_range(crash_size,
529 CRASH_ALIGN, CRASH_ALIGN,
530 CRASH_ADDR_HIGH_MAX);
531 if (!crash_base) {
532 pr_info("crashkernel reservation failed - No suitable area found.\n");
533 return;
534 }
535 } else {
536 unsigned long long start;
537
538 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
539 crash_base + crash_size);
540 if (start != crash_base) {
541 pr_info("crashkernel reservation failed - memory is in use.\n");
542 return;
543 }
544 }
545
546 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
547 memblock_free(crash_base, crash_size);
548 return;
549 }
550
551 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
552 (unsigned long)(crash_size >> 20),
553 (unsigned long)(crash_base >> 20),
554 (unsigned long)(total_mem >> 20));
555
556 crashk_res.start = crash_base;
557 crashk_res.end = crash_base + crash_size - 1;
558 insert_resource(&iomem_resource, &crashk_res);
559 }
560 #else
reserve_crashkernel(void)561 static void __init reserve_crashkernel(void)
562 {
563 }
564 #endif
565
566 static struct resource standard_io_resources[] = {
567 { .name = "dma1", .start = 0x00, .end = 0x1f,
568 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
569 { .name = "pic1", .start = 0x20, .end = 0x21,
570 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
571 { .name = "timer0", .start = 0x40, .end = 0x43,
572 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
573 { .name = "timer1", .start = 0x50, .end = 0x53,
574 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
575 { .name = "keyboard", .start = 0x60, .end = 0x60,
576 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
577 { .name = "keyboard", .start = 0x64, .end = 0x64,
578 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
579 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
580 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
581 { .name = "pic2", .start = 0xa0, .end = 0xa1,
582 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
583 { .name = "dma2", .start = 0xc0, .end = 0xdf,
584 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
585 { .name = "fpu", .start = 0xf0, .end = 0xff,
586 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
587 };
588
reserve_standard_io_resources(void)589 void __init reserve_standard_io_resources(void)
590 {
591 int i;
592
593 /* request I/O space for devices used on all i[345]86 PCs */
594 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
595 request_resource(&ioport_resource, &standard_io_resources[i]);
596
597 }
598
reserve_ibft_region(void)599 static __init void reserve_ibft_region(void)
600 {
601 unsigned long addr, size = 0;
602
603 addr = find_ibft_region(&size);
604
605 if (size)
606 memblock_reserve(addr, size);
607 }
608
snb_gfx_workaround_needed(void)609 static bool __init snb_gfx_workaround_needed(void)
610 {
611 #ifdef CONFIG_PCI
612 int i;
613 u16 vendor, devid;
614 static const __initconst u16 snb_ids[] = {
615 0x0102,
616 0x0112,
617 0x0122,
618 0x0106,
619 0x0116,
620 0x0126,
621 0x010a,
622 };
623
624 /* Assume no if something weird is going on with PCI */
625 if (!early_pci_allowed())
626 return false;
627
628 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
629 if (vendor != 0x8086)
630 return false;
631
632 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
633 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
634 if (devid == snb_ids[i])
635 return true;
636 #endif
637
638 return false;
639 }
640
641 /*
642 * Sandy Bridge graphics has trouble with certain ranges, exclude
643 * them from allocation.
644 */
trim_snb_memory(void)645 static void __init trim_snb_memory(void)
646 {
647 static const __initconst unsigned long bad_pages[] = {
648 0x20050000,
649 0x20110000,
650 0x20130000,
651 0x20138000,
652 0x40004000,
653 };
654 int i;
655
656 if (!snb_gfx_workaround_needed())
657 return;
658
659 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
660
661 /*
662 * Reserve all memory below the 1 MB mark that has not
663 * already been reserved.
664 */
665 memblock_reserve(0, 1<<20);
666
667 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
668 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
669 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
670 bad_pages[i]);
671 }
672 }
673
674 /*
675 * Here we put platform-specific memory range workarounds, i.e.
676 * memory known to be corrupt or otherwise in need to be reserved on
677 * specific platforms.
678 *
679 * If this gets used more widely it could use a real dispatch mechanism.
680 */
trim_platform_memory_ranges(void)681 static void __init trim_platform_memory_ranges(void)
682 {
683 trim_snb_memory();
684 }
685
trim_bios_range(void)686 static void __init trim_bios_range(void)
687 {
688 /*
689 * A special case is the first 4Kb of memory;
690 * This is a BIOS owned area, not kernel ram, but generally
691 * not listed as such in the E820 table.
692 *
693 * This typically reserves additional memory (64KiB by default)
694 * since some BIOSes are known to corrupt low memory. See the
695 * Kconfig help text for X86_RESERVE_LOW.
696 */
697 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
698
699 /*
700 * special case: Some BIOSes report the PC BIOS
701 * area (640Kb -> 1Mb) as RAM even though it is not.
702 * take them out.
703 */
704 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
705
706 e820__update_table(e820_table);
707 }
708
709 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)710 static void __init e820_add_kernel_range(void)
711 {
712 u64 start = __pa_symbol(_text);
713 u64 size = __pa_symbol(_end) - start;
714
715 /*
716 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
717 * attempt to fix it by adding the range. We may have a confused BIOS,
718 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
719 * exclude kernel range. If we really are running on top non-RAM,
720 * we will crash later anyways.
721 */
722 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
723 return;
724
725 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
726 e820__range_remove(start, size, E820_TYPE_RAM, 0);
727 e820__range_add(start, size, E820_TYPE_RAM);
728 }
729
730 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
731
parse_reservelow(char * p)732 static int __init parse_reservelow(char *p)
733 {
734 unsigned long long size;
735
736 if (!p)
737 return -EINVAL;
738
739 size = memparse(p, &p);
740
741 if (size < 4096)
742 size = 4096;
743
744 if (size > 640*1024)
745 size = 640*1024;
746
747 reserve_low = size;
748
749 return 0;
750 }
751
752 early_param("reservelow", parse_reservelow);
753
trim_low_memory_range(void)754 static void __init trim_low_memory_range(void)
755 {
756 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
757 }
758
759 /*
760 * Dump out kernel offset information on panic.
761 */
762 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)763 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
764 {
765 if (kaslr_enabled()) {
766 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
767 kaslr_offset(),
768 __START_KERNEL,
769 __START_KERNEL_map,
770 MODULES_VADDR-1);
771 } else {
772 pr_emerg("Kernel Offset: disabled\n");
773 }
774
775 return 0;
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 /*
794 * Reserve the memory occupied by the kernel between _text and
795 * __end_of_kernel_reserve symbols. Any kernel sections after the
796 * __end_of_kernel_reserve symbol must be explicitly reserved with a
797 * separate memblock_reserve() or they will be discarded.
798 */
799 memblock_reserve(__pa_symbol(_text),
800 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
801
802 /*
803 * Make sure page 0 is always reserved because on systems with
804 * L1TF its contents can be leaked to user processes.
805 */
806 memblock_reserve(0, PAGE_SIZE);
807
808 early_reserve_initrd();
809
810 /*
811 * At this point everything still needed from the boot loader
812 * or BIOS or kernel text should be early reserved or marked not
813 * RAM in e820. All other memory is free game.
814 */
815
816 #ifdef CONFIG_X86_32
817 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
818
819 /*
820 * copy kernel address range established so far and switch
821 * to the proper swapper page table
822 */
823 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
824 initial_page_table + KERNEL_PGD_BOUNDARY,
825 KERNEL_PGD_PTRS);
826
827 load_cr3(swapper_pg_dir);
828 /*
829 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
830 * a cr3 based tlb flush, so the following __flush_tlb_all()
831 * will not flush anything because the CPU quirk which clears
832 * X86_FEATURE_PGE has not been invoked yet. Though due to the
833 * load_cr3() above the TLB has been flushed already. The
834 * quirk is invoked before subsequent calls to __flush_tlb_all()
835 * so proper operation is guaranteed.
836 */
837 __flush_tlb_all();
838 #else
839 printk(KERN_INFO "Command line: %s\n", boot_command_line);
840 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
841 #endif
842
843 /*
844 * If we have OLPC OFW, we might end up relocating the fixmap due to
845 * reserve_top(), so do this before touching the ioremap area.
846 */
847 olpc_ofw_detect();
848
849 idt_setup_early_traps();
850 early_cpu_init();
851 arch_init_ideal_nops();
852 jump_label_init();
853 static_call_init();
854 early_ioremap_init();
855
856 setup_olpc_ofw_pgd();
857
858 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
859 screen_info = boot_params.screen_info;
860 edid_info = boot_params.edid_info;
861 #ifdef CONFIG_X86_32
862 apm_info.bios = boot_params.apm_bios_info;
863 ist_info = boot_params.ist_info;
864 #endif
865 saved_video_mode = boot_params.hdr.vid_mode;
866 bootloader_type = boot_params.hdr.type_of_loader;
867 if ((bootloader_type >> 4) == 0xe) {
868 bootloader_type &= 0xf;
869 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
870 }
871 bootloader_version = bootloader_type & 0xf;
872 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
873
874 #ifdef CONFIG_BLK_DEV_RAM
875 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
876 #endif
877 #ifdef CONFIG_EFI
878 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
879 EFI32_LOADER_SIGNATURE, 4)) {
880 set_bit(EFI_BOOT, &efi.flags);
881 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
882 EFI64_LOADER_SIGNATURE, 4)) {
883 set_bit(EFI_BOOT, &efi.flags);
884 set_bit(EFI_64BIT, &efi.flags);
885 }
886 #endif
887
888 x86_init.oem.arch_setup();
889
890 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
891 e820__memory_setup();
892 parse_setup_data();
893
894 copy_edd();
895
896 if (!boot_params.hdr.root_flags)
897 root_mountflags &= ~MS_RDONLY;
898 init_mm.start_code = (unsigned long) _text;
899 init_mm.end_code = (unsigned long) _etext;
900 init_mm.end_data = (unsigned long) _edata;
901 init_mm.brk = _brk_end;
902
903 code_resource.start = __pa_symbol(_text);
904 code_resource.end = __pa_symbol(_etext)-1;
905 rodata_resource.start = __pa_symbol(__start_rodata);
906 rodata_resource.end = __pa_symbol(__end_rodata)-1;
907 data_resource.start = __pa_symbol(_sdata);
908 data_resource.end = __pa_symbol(_edata)-1;
909 bss_resource.start = __pa_symbol(__bss_start);
910 bss_resource.end = __pa_symbol(__bss_stop)-1;
911
912 #ifdef CONFIG_CMDLINE_BOOL
913 #ifdef CONFIG_CMDLINE_OVERRIDE
914 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
915 #else
916 if (builtin_cmdline[0]) {
917 /* append boot loader cmdline to builtin */
918 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
919 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
920 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
921 }
922 #endif
923 #endif
924
925 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
926 *cmdline_p = command_line;
927
928 /*
929 * x86_configure_nx() is called before parse_early_param() to detect
930 * whether hardware doesn't support NX (so that the early EHCI debug
931 * console setup can safely call set_fixmap()). It may then be called
932 * again from within noexec_setup() during parsing early parameters
933 * to honor the respective command line option.
934 */
935 x86_configure_nx();
936
937 parse_early_param();
938
939 if (efi_enabled(EFI_BOOT))
940 efi_memblock_x86_reserve_range();
941 #ifdef CONFIG_MEMORY_HOTPLUG
942 /*
943 * Memory used by the kernel cannot be hot-removed because Linux
944 * cannot migrate the kernel pages. When memory hotplug is
945 * enabled, we should prevent memblock from allocating memory
946 * for the kernel.
947 *
948 * ACPI SRAT records all hotpluggable memory ranges. But before
949 * SRAT is parsed, we don't know about it.
950 *
951 * The kernel image is loaded into memory at very early time. We
952 * cannot prevent this anyway. So on NUMA system, we set any
953 * node the kernel resides in as un-hotpluggable.
954 *
955 * Since on modern servers, one node could have double-digit
956 * gigabytes memory, we can assume the memory around the kernel
957 * image is also un-hotpluggable. So before SRAT is parsed, just
958 * allocate memory near the kernel image to try the best to keep
959 * the kernel away from hotpluggable memory.
960 */
961 if (movable_node_is_enabled())
962 memblock_set_bottom_up(true);
963 #endif
964
965 x86_report_nx();
966
967 /* after early param, so could get panic from serial */
968 memblock_x86_reserve_range_setup_data();
969
970 if (acpi_mps_check()) {
971 #ifdef CONFIG_X86_LOCAL_APIC
972 disable_apic = 1;
973 #endif
974 setup_clear_cpu_cap(X86_FEATURE_APIC);
975 }
976
977 e820__reserve_setup_data();
978 e820__finish_early_params();
979
980 if (efi_enabled(EFI_BOOT))
981 efi_init();
982
983 dmi_setup();
984
985 /*
986 * VMware detection requires dmi to be available, so this
987 * needs to be done after dmi_setup(), for the boot CPU.
988 */
989 init_hypervisor_platform();
990
991 tsc_early_init();
992 x86_init.resources.probe_roms();
993
994 /* after parse_early_param, so could debug it */
995 insert_resource(&iomem_resource, &code_resource);
996 insert_resource(&iomem_resource, &rodata_resource);
997 insert_resource(&iomem_resource, &data_resource);
998 insert_resource(&iomem_resource, &bss_resource);
999
1000 e820_add_kernel_range();
1001 trim_bios_range();
1002 #ifdef CONFIG_X86_32
1003 if (ppro_with_ram_bug()) {
1004 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1005 E820_TYPE_RESERVED);
1006 e820__update_table(e820_table);
1007 printk(KERN_INFO "fixed physical RAM map:\n");
1008 e820__print_table("bad_ppro");
1009 }
1010 #else
1011 early_gart_iommu_check();
1012 #endif
1013
1014 /*
1015 * partially used pages are not usable - thus
1016 * we are rounding upwards:
1017 */
1018 max_pfn = e820__end_of_ram_pfn();
1019
1020 /* update e820 for memory not covered by WB MTRRs */
1021 mtrr_bp_init();
1022 if (mtrr_trim_uncached_memory(max_pfn))
1023 max_pfn = e820__end_of_ram_pfn();
1024
1025 max_possible_pfn = max_pfn;
1026
1027 /*
1028 * This call is required when the CPU does not support PAT. If
1029 * mtrr_bp_init() invoked it already via pat_init() the call has no
1030 * effect.
1031 */
1032 init_cache_modes();
1033
1034 /*
1035 * Define random base addresses for memory sections after max_pfn is
1036 * defined and before each memory section base is used.
1037 */
1038 kernel_randomize_memory();
1039
1040 #ifdef CONFIG_X86_32
1041 /* max_low_pfn get updated here */
1042 find_low_pfn_range();
1043 #else
1044 check_x2apic();
1045
1046 /* How many end-of-memory variables you have, grandma! */
1047 /* need this before calling reserve_initrd */
1048 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1049 max_low_pfn = e820__end_of_low_ram_pfn();
1050 else
1051 max_low_pfn = max_pfn;
1052
1053 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1054 #endif
1055
1056 /*
1057 * Find and reserve possible boot-time SMP configuration:
1058 */
1059 find_smp_config();
1060
1061 reserve_ibft_region();
1062
1063 early_alloc_pgt_buf();
1064
1065 /*
1066 * Need to conclude brk, before e820__memblock_setup()
1067 * it could use memblock_find_in_range, could overlap with
1068 * brk area.
1069 */
1070 reserve_brk();
1071
1072 cleanup_highmap();
1073
1074 memblock_set_current_limit(ISA_END_ADDRESS);
1075 e820__memblock_setup();
1076
1077 reserve_bios_regions();
1078
1079 efi_fake_memmap();
1080 efi_find_mirror();
1081 efi_esrt_init();
1082 efi_mokvar_table_init();
1083
1084 /*
1085 * The EFI specification says that boot service code won't be
1086 * called after ExitBootServices(). This is, in fact, a lie.
1087 */
1088 efi_reserve_boot_services();
1089
1090 /* preallocate 4k for mptable mpc */
1091 e820__memblock_alloc_reserved_mpc_new();
1092
1093 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1094 setup_bios_corruption_check();
1095 #endif
1096
1097 #ifdef CONFIG_X86_32
1098 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1099 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1100 #endif
1101
1102 reserve_real_mode();
1103
1104 trim_platform_memory_ranges();
1105 trim_low_memory_range();
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 sfi_init();
1205 x86_dtb_init();
1206
1207 /*
1208 * get boot-time SMP configuration:
1209 */
1210 get_smp_config();
1211
1212 /*
1213 * Systems w/o ACPI and mptables might not have it mapped the local
1214 * APIC yet, but prefill_possible_map() might need to access it.
1215 */
1216 init_apic_mappings();
1217
1218 prefill_possible_map();
1219
1220 init_cpu_to_node();
1221 init_gi_nodes();
1222
1223 io_apic_init_mappings();
1224
1225 x86_init.hyper.guest_late_init();
1226
1227 e820__reserve_resources();
1228 e820__register_nosave_regions(max_pfn);
1229
1230 x86_init.resources.reserve_resources();
1231
1232 e820__setup_pci_gap();
1233
1234 #ifdef CONFIG_VT
1235 #if defined(CONFIG_VGA_CONSOLE)
1236 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1237 conswitchp = &vga_con;
1238 #endif
1239 #endif
1240 x86_init.oem.banner();
1241
1242 x86_init.timers.wallclock_init();
1243
1244 mcheck_init();
1245
1246 register_refined_jiffies(CLOCK_TICK_RATE);
1247
1248 #ifdef CONFIG_EFI
1249 if (efi_enabled(EFI_BOOT))
1250 efi_apply_memmap_quirks();
1251 #endif
1252
1253 unwind_init();
1254 }
1255
1256 #ifdef CONFIG_X86_32
1257
1258 static struct resource video_ram_resource = {
1259 .name = "Video RAM area",
1260 .start = 0xa0000,
1261 .end = 0xbffff,
1262 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1263 };
1264
i386_reserve_resources(void)1265 void __init i386_reserve_resources(void)
1266 {
1267 request_resource(&iomem_resource, &video_ram_resource);
1268 reserve_standard_io_resources();
1269 }
1270
1271 #endif /* CONFIG_X86_32 */
1272
1273 static struct notifier_block kernel_offset_notifier = {
1274 .notifier_call = dump_kernel_offset
1275 };
1276
register_kernel_offset_dumper(void)1277 static int __init register_kernel_offset_dumper(void)
1278 {
1279 atomic_notifier_chain_register(&panic_notifier_list,
1280 &kernel_offset_notifier);
1281 return 0;
1282 }
1283 __initcall(register_kernel_offset_dumper);
1284