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_data *data;
375 u64 pa_data;
376
377 pa_data = boot_params.hdr.setup_data;
378 while (pa_data) {
379 data = early_memremap(pa_data, sizeof(*data));
380 memblock_reserve(pa_data, sizeof(*data) + data->len);
381
382 if (data->type == SETUP_INDIRECT &&
383 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
384 memblock_reserve(((struct setup_indirect *)data->data)->addr,
385 ((struct setup_indirect *)data->data)->len);
386
387 pa_data = data->next;
388 early_memunmap(data, sizeof(*data));
389 }
390 }
391
392 /*
393 * --------- Crashkernel reservation ------------------------------
394 */
395
396 #ifdef CONFIG_KEXEC_CORE
397
398 /* 16M alignment for crash kernel regions */
399 #define CRASH_ALIGN SZ_16M
400
401 /*
402 * Keep the crash kernel below this limit.
403 *
404 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
405 * due to mapping restrictions.
406 *
407 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
408 * the upper limit of system RAM in 4-level paging mode. Since the kdump
409 * jump could be from 5-level paging to 4-level paging, the jump will fail if
410 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
411 * no good way to detect the paging mode of the target kernel which will be
412 * loaded for dumping.
413 */
414 #ifdef CONFIG_X86_32
415 # define CRASH_ADDR_LOW_MAX SZ_512M
416 # define CRASH_ADDR_HIGH_MAX SZ_512M
417 #else
418 # define CRASH_ADDR_LOW_MAX SZ_4G
419 # define CRASH_ADDR_HIGH_MAX SZ_64T
420 #endif
421
reserve_crashkernel_low(void)422 static int __init reserve_crashkernel_low(void)
423 {
424 #ifdef CONFIG_X86_64
425 unsigned long long base, low_base = 0, low_size = 0;
426 unsigned long low_mem_limit;
427 int ret;
428
429 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
430
431 /* crashkernel=Y,low */
432 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
433 if (ret) {
434 /*
435 * two parts from kernel/dma/swiotlb.c:
436 * -swiotlb size: user-specified with swiotlb= or default.
437 *
438 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
439 * to 8M for other buffers that may need to stay low too. Also
440 * make sure we allocate enough extra low memory so that we
441 * don't run out of DMA buffers for 32-bit devices.
442 */
443 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
444 } else {
445 /* passed with crashkernel=0,low ? */
446 if (!low_size)
447 return 0;
448 }
449
450 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
451 if (!low_base) {
452 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
453 (unsigned long)(low_size >> 20));
454 return -ENOMEM;
455 }
456
457 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
458 (unsigned long)(low_size >> 20),
459 (unsigned long)(low_base >> 20),
460 (unsigned long)(low_mem_limit >> 20));
461
462 crashk_low_res.start = low_base;
463 crashk_low_res.end = low_base + low_size - 1;
464 insert_resource(&iomem_resource, &crashk_low_res);
465 #endif
466 return 0;
467 }
468
reserve_crashkernel(void)469 static void __init reserve_crashkernel(void)
470 {
471 unsigned long long crash_size, crash_base, total_mem;
472 bool high = false;
473 int ret;
474
475 total_mem = memblock_phys_mem_size();
476
477 /* crashkernel=XM */
478 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
479 if (ret != 0 || crash_size <= 0) {
480 /* crashkernel=X,high */
481 ret = parse_crashkernel_high(boot_command_line, total_mem,
482 &crash_size, &crash_base);
483 if (ret != 0 || crash_size <= 0)
484 return;
485 high = true;
486 }
487
488 if (xen_pv_domain()) {
489 pr_info("Ignoring crashkernel for a Xen PV domain\n");
490 return;
491 }
492
493 /* 0 means: find the address automatically */
494 if (!crash_base) {
495 /*
496 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
497 * crashkernel=x,high reserves memory over 4G, also allocates
498 * 256M extra low memory for DMA buffers and swiotlb.
499 * But the extra memory is not required for all machines.
500 * So try low memory first and fall back to high memory
501 * unless "crashkernel=size[KMG],high" is specified.
502 */
503 if (!high)
504 crash_base = memblock_phys_alloc_range(crash_size,
505 CRASH_ALIGN, CRASH_ALIGN,
506 CRASH_ADDR_LOW_MAX);
507 if (!crash_base)
508 crash_base = memblock_phys_alloc_range(crash_size,
509 CRASH_ALIGN, CRASH_ALIGN,
510 CRASH_ADDR_HIGH_MAX);
511 if (!crash_base) {
512 pr_info("crashkernel reservation failed - No suitable area found.\n");
513 return;
514 }
515 } else {
516 unsigned long long start;
517
518 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
519 crash_base + crash_size);
520 if (start != crash_base) {
521 pr_info("crashkernel reservation failed - memory is in use.\n");
522 return;
523 }
524 }
525
526 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
527 memblock_free(crash_base, crash_size);
528 return;
529 }
530
531 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
532 (unsigned long)(crash_size >> 20),
533 (unsigned long)(crash_base >> 20),
534 (unsigned long)(total_mem >> 20));
535
536 crashk_res.start = crash_base;
537 crashk_res.end = crash_base + crash_size - 1;
538 insert_resource(&iomem_resource, &crashk_res);
539 }
540 #else
reserve_crashkernel(void)541 static void __init reserve_crashkernel(void)
542 {
543 }
544 #endif
545
546 static struct resource standard_io_resources[] = {
547 { .name = "dma1", .start = 0x00, .end = 0x1f,
548 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
549 { .name = "pic1", .start = 0x20, .end = 0x21,
550 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
551 { .name = "timer0", .start = 0x40, .end = 0x43,
552 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
553 { .name = "timer1", .start = 0x50, .end = 0x53,
554 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
555 { .name = "keyboard", .start = 0x60, .end = 0x60,
556 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
557 { .name = "keyboard", .start = 0x64, .end = 0x64,
558 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
559 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
560 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
561 { .name = "pic2", .start = 0xa0, .end = 0xa1,
562 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
563 { .name = "dma2", .start = 0xc0, .end = 0xdf,
564 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
565 { .name = "fpu", .start = 0xf0, .end = 0xff,
566 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
567 };
568
reserve_standard_io_resources(void)569 void __init reserve_standard_io_resources(void)
570 {
571 int i;
572
573 /* request I/O space for devices used on all i[345]86 PCs */
574 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
575 request_resource(&ioport_resource, &standard_io_resources[i]);
576
577 }
578
reserve_ibft_region(void)579 static __init void reserve_ibft_region(void)
580 {
581 unsigned long addr, size = 0;
582
583 addr = find_ibft_region(&size);
584
585 if (size)
586 memblock_reserve(addr, size);
587 }
588
snb_gfx_workaround_needed(void)589 static bool __init snb_gfx_workaround_needed(void)
590 {
591 #ifdef CONFIG_PCI
592 int i;
593 u16 vendor, devid;
594 static const __initconst u16 snb_ids[] = {
595 0x0102,
596 0x0112,
597 0x0122,
598 0x0106,
599 0x0116,
600 0x0126,
601 0x010a,
602 };
603
604 /* Assume no if something weird is going on with PCI */
605 if (!early_pci_allowed())
606 return false;
607
608 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
609 if (vendor != 0x8086)
610 return false;
611
612 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
613 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
614 if (devid == snb_ids[i])
615 return true;
616 #endif
617
618 return false;
619 }
620
621 /*
622 * Sandy Bridge graphics has trouble with certain ranges, exclude
623 * them from allocation.
624 */
trim_snb_memory(void)625 static void __init trim_snb_memory(void)
626 {
627 static const __initconst unsigned long bad_pages[] = {
628 0x20050000,
629 0x20110000,
630 0x20130000,
631 0x20138000,
632 0x40004000,
633 };
634 int i;
635
636 if (!snb_gfx_workaround_needed())
637 return;
638
639 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
640
641 /*
642 * Reserve all memory below the 1 MB mark that has not
643 * already been reserved.
644 */
645 memblock_reserve(0, 1<<20);
646
647 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
648 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
649 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
650 bad_pages[i]);
651 }
652 }
653
654 /*
655 * Here we put platform-specific memory range workarounds, i.e.
656 * memory known to be corrupt or otherwise in need to be reserved on
657 * specific platforms.
658 *
659 * If this gets used more widely it could use a real dispatch mechanism.
660 */
trim_platform_memory_ranges(void)661 static void __init trim_platform_memory_ranges(void)
662 {
663 trim_snb_memory();
664 }
665
trim_bios_range(void)666 static void __init trim_bios_range(void)
667 {
668 /*
669 * A special case is the first 4Kb of memory;
670 * This is a BIOS owned area, not kernel ram, but generally
671 * not listed as such in the E820 table.
672 *
673 * This typically reserves additional memory (64KiB by default)
674 * since some BIOSes are known to corrupt low memory. See the
675 * Kconfig help text for X86_RESERVE_LOW.
676 */
677 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
678
679 /*
680 * special case: Some BIOSes report the PC BIOS
681 * area (640Kb -> 1Mb) as RAM even though it is not.
682 * take them out.
683 */
684 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
685
686 e820__update_table(e820_table);
687 }
688
689 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)690 static void __init e820_add_kernel_range(void)
691 {
692 u64 start = __pa_symbol(_text);
693 u64 size = __pa_symbol(_end) - start;
694
695 /*
696 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
697 * attempt to fix it by adding the range. We may have a confused BIOS,
698 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
699 * exclude kernel range. If we really are running on top non-RAM,
700 * we will crash later anyways.
701 */
702 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
703 return;
704
705 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
706 e820__range_remove(start, size, E820_TYPE_RAM, 0);
707 e820__range_add(start, size, E820_TYPE_RAM);
708 }
709
710 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
711
parse_reservelow(char * p)712 static int __init parse_reservelow(char *p)
713 {
714 unsigned long long size;
715
716 if (!p)
717 return -EINVAL;
718
719 size = memparse(p, &p);
720
721 if (size < 4096)
722 size = 4096;
723
724 if (size > 640*1024)
725 size = 640*1024;
726
727 reserve_low = size;
728
729 return 0;
730 }
731
732 early_param("reservelow", parse_reservelow);
733
trim_low_memory_range(void)734 static void __init trim_low_memory_range(void)
735 {
736 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
737 }
738
739 /*
740 * Dump out kernel offset information on panic.
741 */
742 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)743 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
744 {
745 if (kaslr_enabled()) {
746 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
747 kaslr_offset(),
748 __START_KERNEL,
749 __START_KERNEL_map,
750 MODULES_VADDR-1);
751 } else {
752 pr_emerg("Kernel Offset: disabled\n");
753 }
754
755 return 0;
756 }
757
758 /*
759 * Determine if we were loaded by an EFI loader. If so, then we have also been
760 * passed the efi memmap, systab, etc., so we should use these data structures
761 * for initialization. Note, the efi init code path is determined by the
762 * global efi_enabled. This allows the same kernel image to be used on existing
763 * systems (with a traditional BIOS) as well as on EFI systems.
764 */
765 /*
766 * setup_arch - architecture-specific boot-time initializations
767 *
768 * Note: On x86_64, fixmaps are ready for use even before this is called.
769 */
770
setup_arch(char ** cmdline_p)771 void __init setup_arch(char **cmdline_p)
772 {
773 /*
774 * Reserve the memory occupied by the kernel between _text and
775 * __end_of_kernel_reserve symbols. Any kernel sections after the
776 * __end_of_kernel_reserve symbol must be explicitly reserved with a
777 * separate memblock_reserve() or they will be discarded.
778 */
779 memblock_reserve(__pa_symbol(_text),
780 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
781
782 /*
783 * Make sure page 0 is always reserved because on systems with
784 * L1TF its contents can be leaked to user processes.
785 */
786 memblock_reserve(0, PAGE_SIZE);
787
788 early_reserve_initrd();
789
790 /*
791 * At this point everything still needed from the boot loader
792 * or BIOS or kernel text should be early reserved or marked not
793 * RAM in e820. All other memory is free game.
794 */
795
796 #ifdef CONFIG_X86_32
797 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
798
799 /*
800 * copy kernel address range established so far and switch
801 * to the proper swapper page table
802 */
803 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
804 initial_page_table + KERNEL_PGD_BOUNDARY,
805 KERNEL_PGD_PTRS);
806
807 load_cr3(swapper_pg_dir);
808 /*
809 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
810 * a cr3 based tlb flush, so the following __flush_tlb_all()
811 * will not flush anything because the CPU quirk which clears
812 * X86_FEATURE_PGE has not been invoked yet. Though due to the
813 * load_cr3() above the TLB has been flushed already. The
814 * quirk is invoked before subsequent calls to __flush_tlb_all()
815 * so proper operation is guaranteed.
816 */
817 __flush_tlb_all();
818 #else
819 printk(KERN_INFO "Command line: %s\n", boot_command_line);
820 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
821 #endif
822
823 /*
824 * If we have OLPC OFW, we might end up relocating the fixmap due to
825 * reserve_top(), so do this before touching the ioremap area.
826 */
827 olpc_ofw_detect();
828
829 idt_setup_early_traps();
830 early_cpu_init();
831 arch_init_ideal_nops();
832 jump_label_init();
833 static_call_init();
834 early_ioremap_init();
835
836 setup_olpc_ofw_pgd();
837
838 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
839 screen_info = boot_params.screen_info;
840 edid_info = boot_params.edid_info;
841 #ifdef CONFIG_X86_32
842 apm_info.bios = boot_params.apm_bios_info;
843 ist_info = boot_params.ist_info;
844 #endif
845 saved_video_mode = boot_params.hdr.vid_mode;
846 bootloader_type = boot_params.hdr.type_of_loader;
847 if ((bootloader_type >> 4) == 0xe) {
848 bootloader_type &= 0xf;
849 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
850 }
851 bootloader_version = bootloader_type & 0xf;
852 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
853
854 #ifdef CONFIG_BLK_DEV_RAM
855 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
856 #endif
857 #ifdef CONFIG_EFI
858 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
859 EFI32_LOADER_SIGNATURE, 4)) {
860 set_bit(EFI_BOOT, &efi.flags);
861 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
862 EFI64_LOADER_SIGNATURE, 4)) {
863 set_bit(EFI_BOOT, &efi.flags);
864 set_bit(EFI_64BIT, &efi.flags);
865 }
866 #endif
867
868 x86_init.oem.arch_setup();
869
870 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
871 e820__memory_setup();
872 parse_setup_data();
873
874 copy_edd();
875
876 if (!boot_params.hdr.root_flags)
877 root_mountflags &= ~MS_RDONLY;
878 init_mm.start_code = (unsigned long) _text;
879 init_mm.end_code = (unsigned long) _etext;
880 init_mm.end_data = (unsigned long) _edata;
881 init_mm.brk = _brk_end;
882
883 code_resource.start = __pa_symbol(_text);
884 code_resource.end = __pa_symbol(_etext)-1;
885 rodata_resource.start = __pa_symbol(__start_rodata);
886 rodata_resource.end = __pa_symbol(__end_rodata)-1;
887 data_resource.start = __pa_symbol(_sdata);
888 data_resource.end = __pa_symbol(_edata)-1;
889 bss_resource.start = __pa_symbol(__bss_start);
890 bss_resource.end = __pa_symbol(__bss_stop)-1;
891
892 #ifdef CONFIG_CMDLINE_BOOL
893 #ifdef CONFIG_CMDLINE_OVERRIDE
894 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
895 #else
896 if (builtin_cmdline[0]) {
897 /* append boot loader cmdline to builtin */
898 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
899 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
900 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
901 }
902 #endif
903 #endif
904
905 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
906 *cmdline_p = command_line;
907
908 /*
909 * x86_configure_nx() is called before parse_early_param() to detect
910 * whether hardware doesn't support NX (so that the early EHCI debug
911 * console setup can safely call set_fixmap()). It may then be called
912 * again from within noexec_setup() during parsing early parameters
913 * to honor the respective command line option.
914 */
915 x86_configure_nx();
916
917 parse_early_param();
918
919 if (efi_enabled(EFI_BOOT))
920 efi_memblock_x86_reserve_range();
921 #ifdef CONFIG_MEMORY_HOTPLUG
922 /*
923 * Memory used by the kernel cannot be hot-removed because Linux
924 * cannot migrate the kernel pages. When memory hotplug is
925 * enabled, we should prevent memblock from allocating memory
926 * for the kernel.
927 *
928 * ACPI SRAT records all hotpluggable memory ranges. But before
929 * SRAT is parsed, we don't know about it.
930 *
931 * The kernel image is loaded into memory at very early time. We
932 * cannot prevent this anyway. So on NUMA system, we set any
933 * node the kernel resides in as un-hotpluggable.
934 *
935 * Since on modern servers, one node could have double-digit
936 * gigabytes memory, we can assume the memory around the kernel
937 * image is also un-hotpluggable. So before SRAT is parsed, just
938 * allocate memory near the kernel image to try the best to keep
939 * the kernel away from hotpluggable memory.
940 */
941 if (movable_node_is_enabled())
942 memblock_set_bottom_up(true);
943 #endif
944
945 x86_report_nx();
946
947 /* after early param, so could get panic from serial */
948 memblock_x86_reserve_range_setup_data();
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 reserve_ibft_region();
1042
1043 early_alloc_pgt_buf();
1044
1045 /*
1046 * Need to conclude brk, before e820__memblock_setup()
1047 * it could use memblock_find_in_range, could overlap with
1048 * brk area.
1049 */
1050 reserve_brk();
1051
1052 cleanup_highmap();
1053
1054 memblock_set_current_limit(ISA_END_ADDRESS);
1055 e820__memblock_setup();
1056
1057 reserve_bios_regions();
1058
1059 efi_fake_memmap();
1060 efi_find_mirror();
1061 efi_esrt_init();
1062 efi_mokvar_table_init();
1063
1064 /*
1065 * The EFI specification says that boot service code won't be
1066 * called after ExitBootServices(). This is, in fact, a lie.
1067 */
1068 efi_reserve_boot_services();
1069
1070 /* preallocate 4k for mptable mpc */
1071 e820__memblock_alloc_reserved_mpc_new();
1072
1073 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1074 setup_bios_corruption_check();
1075 #endif
1076
1077 #ifdef CONFIG_X86_32
1078 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1079 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1080 #endif
1081
1082 reserve_real_mode();
1083
1084 trim_platform_memory_ranges();
1085 trim_low_memory_range();
1086
1087 init_mem_mapping();
1088
1089 idt_setup_early_pf();
1090
1091 /*
1092 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1093 * with the current CR4 value. This may not be necessary, but
1094 * auditing all the early-boot CR4 manipulation would be needed to
1095 * rule it out.
1096 *
1097 * Mask off features that don't work outside long mode (just
1098 * PCIDE for now).
1099 */
1100 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1101
1102 memblock_set_current_limit(get_max_mapped());
1103
1104 /*
1105 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1106 */
1107
1108 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1109 if (init_ohci1394_dma_early)
1110 init_ohci1394_dma_on_all_controllers();
1111 #endif
1112 /* Allocate bigger log buffer */
1113 setup_log_buf(1);
1114
1115 if (efi_enabled(EFI_BOOT)) {
1116 switch (boot_params.secure_boot) {
1117 case efi_secureboot_mode_disabled:
1118 pr_info("Secure boot disabled\n");
1119 break;
1120 case efi_secureboot_mode_enabled:
1121 pr_info("Secure boot enabled\n");
1122 break;
1123 default:
1124 pr_info("Secure boot could not be determined\n");
1125 break;
1126 }
1127 }
1128
1129 reserve_initrd();
1130
1131 acpi_table_upgrade();
1132 /* Look for ACPI tables and reserve memory occupied by them. */
1133 acpi_boot_table_init();
1134
1135 vsmp_init();
1136
1137 io_delay_init();
1138
1139 early_platform_quirks();
1140
1141 early_acpi_boot_init();
1142
1143 initmem_init();
1144 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1145
1146 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1147 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1148
1149 /*
1150 * Reserve memory for crash kernel after SRAT is parsed so that it
1151 * won't consume hotpluggable memory.
1152 */
1153 reserve_crashkernel();
1154
1155 memblock_find_dma_reserve();
1156
1157 if (!early_xdbc_setup_hardware())
1158 early_xdbc_register_console();
1159
1160 x86_init.paging.pagetable_init();
1161
1162 kasan_init();
1163
1164 /*
1165 * Sync back kernel address range.
1166 *
1167 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1168 * this call?
1169 */
1170 sync_initial_page_table();
1171
1172 tboot_probe();
1173
1174 map_vsyscall();
1175
1176 generic_apic_probe();
1177
1178 early_quirks();
1179
1180 /*
1181 * Read APIC and some other early information from ACPI tables.
1182 */
1183 acpi_boot_init();
1184 sfi_init();
1185 x86_dtb_init();
1186
1187 /*
1188 * get boot-time SMP configuration:
1189 */
1190 get_smp_config();
1191
1192 /*
1193 * Systems w/o ACPI and mptables might not have it mapped the local
1194 * APIC yet, but prefill_possible_map() might need to access it.
1195 */
1196 init_apic_mappings();
1197
1198 prefill_possible_map();
1199
1200 init_cpu_to_node();
1201 init_gi_nodes();
1202
1203 io_apic_init_mappings();
1204
1205 x86_init.hyper.guest_late_init();
1206
1207 e820__reserve_resources();
1208 e820__register_nosave_regions(max_pfn);
1209
1210 x86_init.resources.reserve_resources();
1211
1212 e820__setup_pci_gap();
1213
1214 #ifdef CONFIG_VT
1215 #if defined(CONFIG_VGA_CONSOLE)
1216 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1217 conswitchp = &vga_con;
1218 #endif
1219 #endif
1220 x86_init.oem.banner();
1221
1222 x86_init.timers.wallclock_init();
1223
1224 mcheck_init();
1225
1226 register_refined_jiffies(CLOCK_TICK_RATE);
1227
1228 #ifdef CONFIG_EFI
1229 if (efi_enabled(EFI_BOOT))
1230 efi_apply_memmap_quirks();
1231 #endif
1232
1233 unwind_init();
1234 }
1235
1236 #ifdef CONFIG_X86_32
1237
1238 static struct resource video_ram_resource = {
1239 .name = "Video RAM area",
1240 .start = 0xa0000,
1241 .end = 0xbffff,
1242 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1243 };
1244
i386_reserve_resources(void)1245 void __init i386_reserve_resources(void)
1246 {
1247 request_resource(&iomem_resource, &video_ram_resource);
1248 reserve_standard_io_resources();
1249 }
1250
1251 #endif /* CONFIG_X86_32 */
1252
1253 static struct notifier_block kernel_offset_notifier = {
1254 .notifier_call = dump_kernel_offset
1255 };
1256
register_kernel_offset_dumper(void)1257 static int __init register_kernel_offset_dumper(void)
1258 {
1259 atomic_notifier_chain_register(&panic_notifier_list,
1260 &kernel_offset_notifier);
1261 return 0;
1262 }
1263 __initcall(register_kernel_offset_dumper);
1264