Loading...
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 *
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
8 *
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
11 *
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
14 *
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
17 *
18 */
19
20/*
21 * This file handles the architecture-dependent parts of initialization
22 */
23
24#include <linux/sched.h>
25#include <linux/mm.h>
26#include <linux/mmzone.h>
27#include <linux/screen_info.h>
28#include <linux/ioport.h>
29#include <linux/acpi.h>
30#include <linux/sfi.h>
31#include <linux/apm_bios.h>
32#include <linux/initrd.h>
33#include <linux/bootmem.h>
34#include <linux/memblock.h>
35#include <linux/seq_file.h>
36#include <linux/console.h>
37#include <linux/root_dev.h>
38#include <linux/highmem.h>
39#include <linux/module.h>
40#include <linux/efi.h>
41#include <linux/init.h>
42#include <linux/edd.h>
43#include <linux/iscsi_ibft.h>
44#include <linux/nodemask.h>
45#include <linux/kexec.h>
46#include <linux/dmi.h>
47#include <linux/pfn.h>
48#include <linux/pci.h>
49#include <asm/pci-direct.h>
50#include <linux/init_ohci1394_dma.h>
51#include <linux/kvm_para.h>
52#include <linux/dma-contiguous.h>
53
54#include <linux/errno.h>
55#include <linux/kernel.h>
56#include <linux/stddef.h>
57#include <linux/unistd.h>
58#include <linux/ptrace.h>
59#include <linux/user.h>
60#include <linux/delay.h>
61
62#include <linux/kallsyms.h>
63#include <linux/cpufreq.h>
64#include <linux/dma-mapping.h>
65#include <linux/ctype.h>
66#include <linux/uaccess.h>
67
68#include <linux/percpu.h>
69#include <linux/crash_dump.h>
70#include <linux/tboot.h>
71
72#include <video/edid.h>
73
74#include <asm/mtrr.h>
75#include <asm/apic.h>
76#include <asm/realmode.h>
77#include <asm/e820.h>
78#include <asm/mpspec.h>
79#include <asm/setup.h>
80#include <asm/efi.h>
81#include <asm/timer.h>
82#include <asm/i8259.h>
83#include <asm/sections.h>
84#include <asm/dmi.h>
85#include <asm/io_apic.h>
86#include <asm/ist.h>
87#include <asm/setup_arch.h>
88#include <asm/bios_ebda.h>
89#include <asm/cacheflush.h>
90#include <asm/processor.h>
91#include <asm/bugs.h>
92
93#include <asm/vsyscall.h>
94#include <asm/cpu.h>
95#include <asm/desc.h>
96#include <asm/dma.h>
97#include <asm/iommu.h>
98#include <asm/gart.h>
99#include <asm/mmu_context.h>
100#include <asm/proto.h>
101
102#include <asm/paravirt.h>
103#include <asm/hypervisor.h>
104#include <asm/olpc_ofw.h>
105
106#include <asm/percpu.h>
107#include <asm/topology.h>
108#include <asm/apicdef.h>
109#include <asm/amd_nb.h>
110#ifdef CONFIG_X86_64
111#include <asm/numa_64.h>
112#endif
113#include <asm/mce.h>
114#include <asm/alternative.h>
115#include <asm/prom.h>
116
117/*
118 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
119 * The direct mapping extends to max_pfn_mapped, so that we can directly access
120 * apertures, ACPI and other tables without having to play with fixmaps.
121 */
122unsigned long max_low_pfn_mapped;
123unsigned long max_pfn_mapped;
124
125#ifdef CONFIG_DMI
126RESERVE_BRK(dmi_alloc, 65536);
127#endif
128
129
130static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
131unsigned long _brk_end = (unsigned long)__brk_base;
132
133#ifdef CONFIG_X86_64
134int default_cpu_present_to_apicid(int mps_cpu)
135{
136 return __default_cpu_present_to_apicid(mps_cpu);
137}
138
139int default_check_phys_apicid_present(int phys_apicid)
140{
141 return __default_check_phys_apicid_present(phys_apicid);
142}
143#endif
144
145#ifndef CONFIG_DEBUG_BOOT_PARAMS
146struct boot_params __initdata boot_params;
147#else
148struct boot_params boot_params;
149#endif
150
151/*
152 * Machine setup..
153 */
154static struct resource data_resource = {
155 .name = "Kernel data",
156 .start = 0,
157 .end = 0,
158 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
159};
160
161static struct resource code_resource = {
162 .name = "Kernel code",
163 .start = 0,
164 .end = 0,
165 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
166};
167
168static struct resource bss_resource = {
169 .name = "Kernel bss",
170 .start = 0,
171 .end = 0,
172 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
173};
174
175
176#ifdef CONFIG_X86_32
177/* cpu data as detected by the assembly code in head.S */
178struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
179/* common cpu data for all cpus */
180struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
181EXPORT_SYMBOL(boot_cpu_data);
182
183unsigned int def_to_bigsmp;
184
185/* for MCA, but anyone else can use it if they want */
186unsigned int machine_id;
187unsigned int machine_submodel_id;
188unsigned int BIOS_revision;
189
190struct apm_info apm_info;
191EXPORT_SYMBOL(apm_info);
192
193#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
194 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
195struct ist_info ist_info;
196EXPORT_SYMBOL(ist_info);
197#else
198struct ist_info ist_info;
199#endif
200
201#else
202struct cpuinfo_x86 boot_cpu_data __read_mostly = {
203 .x86_phys_bits = MAX_PHYSMEM_BITS,
204};
205EXPORT_SYMBOL(boot_cpu_data);
206#endif
207
208
209#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
210unsigned long mmu_cr4_features;
211#else
212unsigned long mmu_cr4_features = X86_CR4_PAE;
213#endif
214
215/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
216int bootloader_type, bootloader_version;
217
218/*
219 * Setup options
220 */
221struct screen_info screen_info;
222EXPORT_SYMBOL(screen_info);
223struct edid_info edid_info;
224EXPORT_SYMBOL_GPL(edid_info);
225
226extern int root_mountflags;
227
228unsigned long saved_video_mode;
229
230#define RAMDISK_IMAGE_START_MASK 0x07FF
231#define RAMDISK_PROMPT_FLAG 0x8000
232#define RAMDISK_LOAD_FLAG 0x4000
233
234static char __initdata command_line[COMMAND_LINE_SIZE];
235#ifdef CONFIG_CMDLINE_BOOL
236static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
237#endif
238
239#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
240struct edd edd;
241#ifdef CONFIG_EDD_MODULE
242EXPORT_SYMBOL(edd);
243#endif
244/**
245 * copy_edd() - Copy the BIOS EDD information
246 * from boot_params into a safe place.
247 *
248 */
249static inline void __init copy_edd(void)
250{
251 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
252 sizeof(edd.mbr_signature));
253 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
254 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
255 edd.edd_info_nr = boot_params.eddbuf_entries;
256}
257#else
258static inline void __init copy_edd(void)
259{
260}
261#endif
262
263void * __init extend_brk(size_t size, size_t align)
264{
265 size_t mask = align - 1;
266 void *ret;
267
268 BUG_ON(_brk_start == 0);
269 BUG_ON(align & mask);
270
271 _brk_end = (_brk_end + mask) & ~mask;
272 BUG_ON((char *)(_brk_end + size) > __brk_limit);
273
274 ret = (void *)_brk_end;
275 _brk_end += size;
276
277 memset(ret, 0, size);
278
279 return ret;
280}
281
282#ifdef CONFIG_X86_64
283static void __init init_gbpages(void)
284{
285 if (direct_gbpages && cpu_has_gbpages)
286 printk(KERN_INFO "Using GB pages for direct mapping\n");
287 else
288 direct_gbpages = 0;
289}
290#else
291static inline void init_gbpages(void)
292{
293}
294static void __init cleanup_highmap(void)
295{
296}
297#endif
298
299static void __init reserve_brk(void)
300{
301 if (_brk_end > _brk_start)
302 memblock_reserve(__pa(_brk_start),
303 __pa(_brk_end) - __pa(_brk_start));
304
305 /* Mark brk area as locked down and no longer taking any
306 new allocations */
307 _brk_start = 0;
308}
309
310#ifdef CONFIG_BLK_DEV_INITRD
311
312#define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
313static void __init relocate_initrd(void)
314{
315 /* Assume only end is not page aligned */
316 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
317 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
318 u64 area_size = PAGE_ALIGN(ramdisk_size);
319 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
320 u64 ramdisk_here;
321 unsigned long slop, clen, mapaddr;
322 char *p, *q;
323
324 /* We need to move the initrd down into lowmem */
325 ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size,
326 PAGE_SIZE);
327
328 if (!ramdisk_here)
329 panic("Cannot find place for new RAMDISK of size %lld\n",
330 ramdisk_size);
331
332 /* Note: this includes all the lowmem currently occupied by
333 the initrd, we rely on that fact to keep the data intact. */
334 memblock_reserve(ramdisk_here, area_size);
335 initrd_start = ramdisk_here + PAGE_OFFSET;
336 initrd_end = initrd_start + ramdisk_size;
337 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
338 ramdisk_here, ramdisk_here + ramdisk_size - 1);
339
340 q = (char *)initrd_start;
341
342 /* Copy any lowmem portion of the initrd */
343 if (ramdisk_image < end_of_lowmem) {
344 clen = end_of_lowmem - ramdisk_image;
345 p = (char *)__va(ramdisk_image);
346 memcpy(q, p, clen);
347 q += clen;
348 ramdisk_image += clen;
349 ramdisk_size -= clen;
350 }
351
352 /* Copy the highmem portion of the initrd */
353 while (ramdisk_size) {
354 slop = ramdisk_image & ~PAGE_MASK;
355 clen = ramdisk_size;
356 if (clen > MAX_MAP_CHUNK-slop)
357 clen = MAX_MAP_CHUNK-slop;
358 mapaddr = ramdisk_image & PAGE_MASK;
359 p = early_memremap(mapaddr, clen+slop);
360 memcpy(q, p+slop, clen);
361 early_iounmap(p, clen+slop);
362 q += clen;
363 ramdisk_image += clen;
364 ramdisk_size -= clen;
365 }
366 /* high pages is not converted by early_res_to_bootmem */
367 ramdisk_image = boot_params.hdr.ramdisk_image;
368 ramdisk_size = boot_params.hdr.ramdisk_size;
369 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
370 " [mem %#010llx-%#010llx]\n",
371 ramdisk_image, ramdisk_image + ramdisk_size - 1,
372 ramdisk_here, ramdisk_here + ramdisk_size - 1);
373}
374
375static void __init reserve_initrd(void)
376{
377 /* Assume only end is not page aligned */
378 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
379 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
380 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
381 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
382
383 if (!boot_params.hdr.type_of_loader ||
384 !ramdisk_image || !ramdisk_size)
385 return; /* No initrd provided by bootloader */
386
387 initrd_start = 0;
388
389 if (ramdisk_size >= (end_of_lowmem>>1)) {
390 panic("initrd too large to handle, "
391 "disabling initrd (%lld needed, %lld available)\n",
392 ramdisk_size, end_of_lowmem>>1);
393 }
394
395 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
396 ramdisk_end - 1);
397
398
399 if (ramdisk_end <= end_of_lowmem) {
400 /* All in lowmem, easy case */
401 /*
402 * don't need to reserve again, already reserved early
403 * in i386_start_kernel
404 */
405 initrd_start = ramdisk_image + PAGE_OFFSET;
406 initrd_end = initrd_start + ramdisk_size;
407 return;
408 }
409
410 relocate_initrd();
411
412 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
413}
414#else
415static void __init reserve_initrd(void)
416{
417}
418#endif /* CONFIG_BLK_DEV_INITRD */
419
420static void __init parse_setup_data(void)
421{
422 struct setup_data *data;
423 u64 pa_data;
424
425 if (boot_params.hdr.version < 0x0209)
426 return;
427 pa_data = boot_params.hdr.setup_data;
428 while (pa_data) {
429 u32 data_len, map_len;
430
431 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
432 (u64)sizeof(struct setup_data));
433 data = early_memremap(pa_data, map_len);
434 data_len = data->len + sizeof(struct setup_data);
435 if (data_len > map_len) {
436 early_iounmap(data, map_len);
437 data = early_memremap(pa_data, data_len);
438 map_len = data_len;
439 }
440
441 switch (data->type) {
442 case SETUP_E820_EXT:
443 parse_e820_ext(data);
444 break;
445 case SETUP_DTB:
446 add_dtb(pa_data);
447 break;
448 default:
449 break;
450 }
451 pa_data = data->next;
452 early_iounmap(data, map_len);
453 }
454}
455
456static void __init e820_reserve_setup_data(void)
457{
458 struct setup_data *data;
459 u64 pa_data;
460 int found = 0;
461
462 if (boot_params.hdr.version < 0x0209)
463 return;
464 pa_data = boot_params.hdr.setup_data;
465 while (pa_data) {
466 data = early_memremap(pa_data, sizeof(*data));
467 e820_update_range(pa_data, sizeof(*data)+data->len,
468 E820_RAM, E820_RESERVED_KERN);
469 found = 1;
470 pa_data = data->next;
471 early_iounmap(data, sizeof(*data));
472 }
473 if (!found)
474 return;
475
476 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
477 memcpy(&e820_saved, &e820, sizeof(struct e820map));
478 printk(KERN_INFO "extended physical RAM map:\n");
479 e820_print_map("reserve setup_data");
480}
481
482static void __init memblock_x86_reserve_range_setup_data(void)
483{
484 struct setup_data *data;
485 u64 pa_data;
486
487 if (boot_params.hdr.version < 0x0209)
488 return;
489 pa_data = boot_params.hdr.setup_data;
490 while (pa_data) {
491 data = early_memremap(pa_data, sizeof(*data));
492 memblock_reserve(pa_data, sizeof(*data) + data->len);
493 pa_data = data->next;
494 early_iounmap(data, sizeof(*data));
495 }
496}
497
498/*
499 * --------- Crashkernel reservation ------------------------------
500 */
501
502#ifdef CONFIG_KEXEC
503
504/*
505 * Keep the crash kernel below this limit. On 32 bits earlier kernels
506 * would limit the kernel to the low 512 MiB due to mapping restrictions.
507 * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this
508 * limit once kexec-tools are fixed.
509 */
510#ifdef CONFIG_X86_32
511# define CRASH_KERNEL_ADDR_MAX (512 << 20)
512#else
513# define CRASH_KERNEL_ADDR_MAX (896 << 20)
514#endif
515
516static void __init reserve_crashkernel(void)
517{
518 unsigned long long total_mem;
519 unsigned long long crash_size, crash_base;
520 int ret;
521
522 total_mem = memblock_phys_mem_size();
523
524 ret = parse_crashkernel(boot_command_line, total_mem,
525 &crash_size, &crash_base);
526 if (ret != 0 || crash_size <= 0)
527 return;
528
529 /* 0 means: find the address automatically */
530 if (crash_base <= 0) {
531 const unsigned long long alignment = 16<<20; /* 16M */
532
533 /*
534 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
535 */
536 crash_base = memblock_find_in_range(alignment,
537 CRASH_KERNEL_ADDR_MAX, crash_size, alignment);
538
539 if (!crash_base) {
540 pr_info("crashkernel reservation failed - No suitable area found.\n");
541 return;
542 }
543 } else {
544 unsigned long long start;
545
546 start = memblock_find_in_range(crash_base,
547 crash_base + crash_size, crash_size, 1<<20);
548 if (start != crash_base) {
549 pr_info("crashkernel reservation failed - memory is in use.\n");
550 return;
551 }
552 }
553 memblock_reserve(crash_base, crash_size);
554
555 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
556 "for crashkernel (System RAM: %ldMB)\n",
557 (unsigned long)(crash_size >> 20),
558 (unsigned long)(crash_base >> 20),
559 (unsigned long)(total_mem >> 20));
560
561 crashk_res.start = crash_base;
562 crashk_res.end = crash_base + crash_size - 1;
563 insert_resource(&iomem_resource, &crashk_res);
564}
565#else
566static void __init reserve_crashkernel(void)
567{
568}
569#endif
570
571static struct resource standard_io_resources[] = {
572 { .name = "dma1", .start = 0x00, .end = 0x1f,
573 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
574 { .name = "pic1", .start = 0x20, .end = 0x21,
575 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
576 { .name = "timer0", .start = 0x40, .end = 0x43,
577 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
578 { .name = "timer1", .start = 0x50, .end = 0x53,
579 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
580 { .name = "keyboard", .start = 0x60, .end = 0x60,
581 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
582 { .name = "keyboard", .start = 0x64, .end = 0x64,
583 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
584 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
585 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
586 { .name = "pic2", .start = 0xa0, .end = 0xa1,
587 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
588 { .name = "dma2", .start = 0xc0, .end = 0xdf,
589 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
590 { .name = "fpu", .start = 0xf0, .end = 0xff,
591 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
592};
593
594void __init reserve_standard_io_resources(void)
595{
596 int i;
597
598 /* request I/O space for devices used on all i[345]86 PCs */
599 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
600 request_resource(&ioport_resource, &standard_io_resources[i]);
601
602}
603
604static __init void reserve_ibft_region(void)
605{
606 unsigned long addr, size = 0;
607
608 addr = find_ibft_region(&size);
609
610 if (size)
611 memblock_reserve(addr, size);
612}
613
614static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
615
616static void __init trim_bios_range(void)
617{
618 /*
619 * A special case is the first 4Kb of memory;
620 * This is a BIOS owned area, not kernel ram, but generally
621 * not listed as such in the E820 table.
622 *
623 * This typically reserves additional memory (64KiB by default)
624 * since some BIOSes are known to corrupt low memory. See the
625 * Kconfig help text for X86_RESERVE_LOW.
626 */
627 e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE),
628 E820_RAM, E820_RESERVED);
629
630 /*
631 * special case: Some BIOSen report the PC BIOS
632 * area (640->1Mb) as ram even though it is not.
633 * take them out.
634 */
635 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
636 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
637}
638
639static int __init parse_reservelow(char *p)
640{
641 unsigned long long size;
642
643 if (!p)
644 return -EINVAL;
645
646 size = memparse(p, &p);
647
648 if (size < 4096)
649 size = 4096;
650
651 if (size > 640*1024)
652 size = 640*1024;
653
654 reserve_low = size;
655
656 return 0;
657}
658
659early_param("reservelow", parse_reservelow);
660
661/*
662 * Determine if we were loaded by an EFI loader. If so, then we have also been
663 * passed the efi memmap, systab, etc., so we should use these data structures
664 * for initialization. Note, the efi init code path is determined by the
665 * global efi_enabled. This allows the same kernel image to be used on existing
666 * systems (with a traditional BIOS) as well as on EFI systems.
667 */
668/*
669 * setup_arch - architecture-specific boot-time initializations
670 *
671 * Note: On x86_64, fixmaps are ready for use even before this is called.
672 */
673
674void __init setup_arch(char **cmdline_p)
675{
676#ifdef CONFIG_X86_32
677 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
678 visws_early_detect();
679
680 /*
681 * copy kernel address range established so far and switch
682 * to the proper swapper page table
683 */
684 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
685 initial_page_table + KERNEL_PGD_BOUNDARY,
686 KERNEL_PGD_PTRS);
687
688 load_cr3(swapper_pg_dir);
689 __flush_tlb_all();
690#else
691 printk(KERN_INFO "Command line: %s\n", boot_command_line);
692#endif
693
694 /*
695 * If we have OLPC OFW, we might end up relocating the fixmap due to
696 * reserve_top(), so do this before touching the ioremap area.
697 */
698 olpc_ofw_detect();
699
700 early_trap_init();
701 early_cpu_init();
702 early_ioremap_init();
703
704 setup_olpc_ofw_pgd();
705
706 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
707 screen_info = boot_params.screen_info;
708 edid_info = boot_params.edid_info;
709#ifdef CONFIG_X86_32
710 apm_info.bios = boot_params.apm_bios_info;
711 ist_info = boot_params.ist_info;
712 if (boot_params.sys_desc_table.length != 0) {
713 machine_id = boot_params.sys_desc_table.table[0];
714 machine_submodel_id = boot_params.sys_desc_table.table[1];
715 BIOS_revision = boot_params.sys_desc_table.table[2];
716 }
717#endif
718 saved_video_mode = boot_params.hdr.vid_mode;
719 bootloader_type = boot_params.hdr.type_of_loader;
720 if ((bootloader_type >> 4) == 0xe) {
721 bootloader_type &= 0xf;
722 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
723 }
724 bootloader_version = bootloader_type & 0xf;
725 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
726
727#ifdef CONFIG_BLK_DEV_RAM
728 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
729 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
730 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
731#endif
732#ifdef CONFIG_EFI
733 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
734 "EL32", 4)) {
735 efi_enabled = 1;
736 efi_64bit = false;
737 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
738 "EL64", 4)) {
739 efi_enabled = 1;
740 efi_64bit = true;
741 }
742 if (efi_enabled && efi_memblock_x86_reserve_range())
743 efi_enabled = 0;
744#endif
745
746 x86_init.oem.arch_setup();
747
748 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
749 setup_memory_map();
750 parse_setup_data();
751 /* update the e820_saved too */
752 e820_reserve_setup_data();
753
754 copy_edd();
755
756 if (!boot_params.hdr.root_flags)
757 root_mountflags &= ~MS_RDONLY;
758 init_mm.start_code = (unsigned long) _text;
759 init_mm.end_code = (unsigned long) _etext;
760 init_mm.end_data = (unsigned long) _edata;
761 init_mm.brk = _brk_end;
762
763 code_resource.start = virt_to_phys(_text);
764 code_resource.end = virt_to_phys(_etext)-1;
765 data_resource.start = virt_to_phys(_etext);
766 data_resource.end = virt_to_phys(_edata)-1;
767 bss_resource.start = virt_to_phys(&__bss_start);
768 bss_resource.end = virt_to_phys(&__bss_stop)-1;
769
770#ifdef CONFIG_CMDLINE_BOOL
771#ifdef CONFIG_CMDLINE_OVERRIDE
772 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
773#else
774 if (builtin_cmdline[0]) {
775 /* append boot loader cmdline to builtin */
776 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
777 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
778 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
779 }
780#endif
781#endif
782
783 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
784 *cmdline_p = command_line;
785
786 /*
787 * x86_configure_nx() is called before parse_early_param() to detect
788 * whether hardware doesn't support NX (so that the early EHCI debug
789 * console setup can safely call set_fixmap()). It may then be called
790 * again from within noexec_setup() during parsing early parameters
791 * to honor the respective command line option.
792 */
793 x86_configure_nx();
794
795 parse_early_param();
796
797 x86_report_nx();
798
799 /* after early param, so could get panic from serial */
800 memblock_x86_reserve_range_setup_data();
801
802 if (acpi_mps_check()) {
803#ifdef CONFIG_X86_LOCAL_APIC
804 disable_apic = 1;
805#endif
806 setup_clear_cpu_cap(X86_FEATURE_APIC);
807 }
808
809#ifdef CONFIG_PCI
810 if (pci_early_dump_regs)
811 early_dump_pci_devices();
812#endif
813
814 finish_e820_parsing();
815
816 if (efi_enabled)
817 efi_init();
818
819 dmi_scan_machine();
820
821 /*
822 * VMware detection requires dmi to be available, so this
823 * needs to be done after dmi_scan_machine, for the BP.
824 */
825 init_hypervisor_platform();
826
827 x86_init.resources.probe_roms();
828
829 /* after parse_early_param, so could debug it */
830 insert_resource(&iomem_resource, &code_resource);
831 insert_resource(&iomem_resource, &data_resource);
832 insert_resource(&iomem_resource, &bss_resource);
833
834 trim_bios_range();
835#ifdef CONFIG_X86_32
836 if (ppro_with_ram_bug()) {
837 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
838 E820_RESERVED);
839 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
840 printk(KERN_INFO "fixed physical RAM map:\n");
841 e820_print_map("bad_ppro");
842 }
843#else
844 early_gart_iommu_check();
845#endif
846
847 /*
848 * partially used pages are not usable - thus
849 * we are rounding upwards:
850 */
851 max_pfn = e820_end_of_ram_pfn();
852
853 /* update e820 for memory not covered by WB MTRRs */
854 mtrr_bp_init();
855 if (mtrr_trim_uncached_memory(max_pfn))
856 max_pfn = e820_end_of_ram_pfn();
857
858#ifdef CONFIG_X86_32
859 /* max_low_pfn get updated here */
860 find_low_pfn_range();
861#else
862 num_physpages = max_pfn;
863
864 check_x2apic();
865
866 /* How many end-of-memory variables you have, grandma! */
867 /* need this before calling reserve_initrd */
868 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
869 max_low_pfn = e820_end_of_low_ram_pfn();
870 else
871 max_low_pfn = max_pfn;
872
873 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
874#endif
875
876 /*
877 * Find and reserve possible boot-time SMP configuration:
878 */
879 find_smp_config();
880
881 reserve_ibft_region();
882
883 /*
884 * Need to conclude brk, before memblock_x86_fill()
885 * it could use memblock_find_in_range, could overlap with
886 * brk area.
887 */
888 reserve_brk();
889
890 cleanup_highmap();
891
892 memblock.current_limit = get_max_mapped();
893 memblock_x86_fill();
894
895 /*
896 * The EFI specification says that boot service code won't be called
897 * after ExitBootServices(). This is, in fact, a lie.
898 */
899 if (efi_enabled)
900 efi_reserve_boot_services();
901
902 /* preallocate 4k for mptable mpc */
903 early_reserve_e820_mpc_new();
904
905#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
906 setup_bios_corruption_check();
907#endif
908
909 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
910 (max_pfn_mapped<<PAGE_SHIFT) - 1);
911
912 setup_real_mode();
913
914 init_gbpages();
915
916 /* max_pfn_mapped is updated here */
917 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
918 max_pfn_mapped = max_low_pfn_mapped;
919
920#ifdef CONFIG_X86_64
921 if (max_pfn > max_low_pfn) {
922 max_pfn_mapped = init_memory_mapping(1UL<<32,
923 max_pfn<<PAGE_SHIFT);
924 /* can we preseve max_low_pfn ?*/
925 max_low_pfn = max_pfn;
926 }
927#endif
928 memblock.current_limit = get_max_mapped();
929 dma_contiguous_reserve(0);
930
931 /*
932 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
933 */
934
935#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
936 if (init_ohci1394_dma_early)
937 init_ohci1394_dma_on_all_controllers();
938#endif
939 /* Allocate bigger log buffer */
940 setup_log_buf(1);
941
942 reserve_initrd();
943
944 reserve_crashkernel();
945
946 vsmp_init();
947
948 io_delay_init();
949
950 /*
951 * Parse the ACPI tables for possible boot-time SMP configuration.
952 */
953 acpi_boot_table_init();
954
955 early_acpi_boot_init();
956
957 initmem_init();
958 memblock_find_dma_reserve();
959
960#ifdef CONFIG_KVM_CLOCK
961 kvmclock_init();
962#endif
963
964 x86_init.paging.pagetable_setup_start(swapper_pg_dir);
965 paging_init();
966 x86_init.paging.pagetable_setup_done(swapper_pg_dir);
967
968 if (boot_cpu_data.cpuid_level >= 0) {
969 /* A CPU has %cr4 if and only if it has CPUID */
970 mmu_cr4_features = read_cr4();
971 if (trampoline_cr4_features)
972 *trampoline_cr4_features = mmu_cr4_features;
973 }
974
975#ifdef CONFIG_X86_32
976 /* sync back kernel address range */
977 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
978 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
979 KERNEL_PGD_PTRS);
980#endif
981
982 tboot_probe();
983
984#ifdef CONFIG_X86_64
985 map_vsyscall();
986#endif
987
988 generic_apic_probe();
989
990 early_quirks();
991
992 /*
993 * Read APIC and some other early information from ACPI tables.
994 */
995 acpi_boot_init();
996 sfi_init();
997 x86_dtb_init();
998
999 /*
1000 * get boot-time SMP configuration:
1001 */
1002 if (smp_found_config)
1003 get_smp_config();
1004
1005 prefill_possible_map();
1006
1007 init_cpu_to_node();
1008
1009 init_apic_mappings();
1010 if (x86_io_apic_ops.init)
1011 x86_io_apic_ops.init();
1012
1013 kvm_guest_init();
1014
1015 e820_reserve_resources();
1016 e820_mark_nosave_regions(max_low_pfn);
1017
1018 x86_init.resources.reserve_resources();
1019
1020 e820_setup_gap();
1021
1022#ifdef CONFIG_VT
1023#if defined(CONFIG_VGA_CONSOLE)
1024 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1025 conswitchp = &vga_con;
1026#elif defined(CONFIG_DUMMY_CONSOLE)
1027 conswitchp = &dummy_con;
1028#endif
1029#endif
1030 x86_init.oem.banner();
1031
1032 x86_init.timers.wallclock_init();
1033
1034 x86_platform.wallclock_init();
1035
1036 mcheck_init();
1037
1038 arch_init_ideal_nops();
1039}
1040
1041#ifdef CONFIG_X86_32
1042
1043static struct resource video_ram_resource = {
1044 .name = "Video RAM area",
1045 .start = 0xa0000,
1046 .end = 0xbffff,
1047 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1048};
1049
1050void __init i386_reserve_resources(void)
1051{
1052 request_resource(&iomem_resource, &video_ram_resource);
1053 reserve_standard_io_resources();
1054}
1055
1056#endif /* CONFIG_X86_32 */
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/panic_notifier.h>
18#include <linux/pci.h>
19#include <linux/root_dev.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/thermal.h>
49#include <asm/unwind.h>
50#include <asm/vsyscall.h>
51#include <linux/vmalloc.h>
52
53/*
54 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
55 * max_pfn_mapped: highest directly mapped pfn > 4 GB
56 *
57 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
58 * represented by pfn_mapped[].
59 */
60unsigned long max_low_pfn_mapped;
61unsigned long max_pfn_mapped;
62
63#ifdef CONFIG_DMI
64RESERVE_BRK(dmi_alloc, 65536);
65#endif
66
67
68/*
69 * Range of the BSS area. The size of the BSS area is determined
70 * at link time, with RESERVE_BRK() facility reserving additional
71 * chunks.
72 */
73unsigned long _brk_start = (unsigned long)__brk_base;
74unsigned long _brk_end = (unsigned long)__brk_base;
75
76struct boot_params boot_params;
77
78/*
79 * These are the four main kernel memory regions, we put them into
80 * the resource tree so that kdump tools and other debugging tools
81 * recover it:
82 */
83
84static struct resource rodata_resource = {
85 .name = "Kernel rodata",
86 .start = 0,
87 .end = 0,
88 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
89};
90
91static struct resource data_resource = {
92 .name = "Kernel data",
93 .start = 0,
94 .end = 0,
95 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
96};
97
98static struct resource code_resource = {
99 .name = "Kernel code",
100 .start = 0,
101 .end = 0,
102 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
103};
104
105static struct resource bss_resource = {
106 .name = "Kernel bss",
107 .start = 0,
108 .end = 0,
109 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
110};
111
112
113#ifdef CONFIG_X86_32
114/* CPU data as detected by the assembly code in head_32.S */
115struct cpuinfo_x86 new_cpu_data;
116
117/* Common CPU data for all CPUs */
118struct cpuinfo_x86 boot_cpu_data __read_mostly;
119EXPORT_SYMBOL(boot_cpu_data);
120
121unsigned int def_to_bigsmp;
122
123struct apm_info apm_info;
124EXPORT_SYMBOL(apm_info);
125
126#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
127 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
128struct ist_info ist_info;
129EXPORT_SYMBOL(ist_info);
130#else
131struct ist_info ist_info;
132#endif
133
134#else
135struct cpuinfo_x86 boot_cpu_data __read_mostly;
136EXPORT_SYMBOL(boot_cpu_data);
137#endif
138
139
140#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
141__visible unsigned long mmu_cr4_features __ro_after_init;
142#else
143__visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
144#endif
145
146/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
147int bootloader_type, bootloader_version;
148
149/*
150 * Setup options
151 */
152struct screen_info screen_info;
153EXPORT_SYMBOL(screen_info);
154struct edid_info edid_info;
155EXPORT_SYMBOL_GPL(edid_info);
156
157extern int root_mountflags;
158
159unsigned long saved_video_mode;
160
161#define RAMDISK_IMAGE_START_MASK 0x07FF
162#define RAMDISK_PROMPT_FLAG 0x8000
163#define RAMDISK_LOAD_FLAG 0x4000
164
165static char __initdata command_line[COMMAND_LINE_SIZE];
166#ifdef CONFIG_CMDLINE_BOOL
167static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
168#endif
169
170#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
171struct edd edd;
172#ifdef CONFIG_EDD_MODULE
173EXPORT_SYMBOL(edd);
174#endif
175/**
176 * copy_edd() - Copy the BIOS EDD information
177 * from boot_params into a safe place.
178 *
179 */
180static inline void __init copy_edd(void)
181{
182 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
183 sizeof(edd.mbr_signature));
184 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
185 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
186 edd.edd_info_nr = boot_params.eddbuf_entries;
187}
188#else
189static inline void __init copy_edd(void)
190{
191}
192#endif
193
194void * __init extend_brk(size_t size, size_t align)
195{
196 size_t mask = align - 1;
197 void *ret;
198
199 BUG_ON(_brk_start == 0);
200 BUG_ON(align & mask);
201
202 _brk_end = (_brk_end + mask) & ~mask;
203 BUG_ON((char *)(_brk_end + size) > __brk_limit);
204
205 ret = (void *)_brk_end;
206 _brk_end += size;
207
208 memset(ret, 0, size);
209
210 return ret;
211}
212
213#ifdef CONFIG_X86_32
214static void __init cleanup_highmap(void)
215{
216}
217#endif
218
219static void __init reserve_brk(void)
220{
221 if (_brk_end > _brk_start)
222 memblock_reserve(__pa_symbol(_brk_start),
223 _brk_end - _brk_start);
224
225 /* Mark brk area as locked down and no longer taking any
226 new allocations */
227 _brk_start = 0;
228}
229
230u64 relocated_ramdisk;
231
232#ifdef CONFIG_BLK_DEV_INITRD
233
234static u64 __init get_ramdisk_image(void)
235{
236 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
237
238 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
239
240 if (ramdisk_image == 0)
241 ramdisk_image = phys_initrd_start;
242
243 return ramdisk_image;
244}
245static u64 __init get_ramdisk_size(void)
246{
247 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
248
249 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
250
251 if (ramdisk_size == 0)
252 ramdisk_size = phys_initrd_size;
253
254 return ramdisk_size;
255}
256
257static void __init relocate_initrd(void)
258{
259 /* Assume only end is not page aligned */
260 u64 ramdisk_image = get_ramdisk_image();
261 u64 ramdisk_size = get_ramdisk_size();
262 u64 area_size = PAGE_ALIGN(ramdisk_size);
263
264 /* We need to move the initrd down into directly mapped mem */
265 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
266 PFN_PHYS(max_pfn_mapped));
267 if (!relocated_ramdisk)
268 panic("Cannot find place for new RAMDISK of size %lld\n",
269 ramdisk_size);
270
271 initrd_start = relocated_ramdisk + PAGE_OFFSET;
272 initrd_end = initrd_start + ramdisk_size;
273 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
274 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
275
276 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
277
278 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
279 " [mem %#010llx-%#010llx]\n",
280 ramdisk_image, ramdisk_image + ramdisk_size - 1,
281 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
282}
283
284static void __init early_reserve_initrd(void)
285{
286 /* Assume only end is not page aligned */
287 u64 ramdisk_image = get_ramdisk_image();
288 u64 ramdisk_size = get_ramdisk_size();
289 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
290
291 if (!boot_params.hdr.type_of_loader ||
292 !ramdisk_image || !ramdisk_size)
293 return; /* No initrd provided by bootloader */
294
295 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
296}
297
298static void __init reserve_initrd(void)
299{
300 /* Assume only end is not page aligned */
301 u64 ramdisk_image = get_ramdisk_image();
302 u64 ramdisk_size = get_ramdisk_size();
303 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
304
305 if (!boot_params.hdr.type_of_loader ||
306 !ramdisk_image || !ramdisk_size)
307 return; /* No initrd provided by bootloader */
308
309 initrd_start = 0;
310
311 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
312 ramdisk_end - 1);
313
314 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
315 PFN_DOWN(ramdisk_end))) {
316 /* All are mapped, easy case */
317 initrd_start = ramdisk_image + PAGE_OFFSET;
318 initrd_end = initrd_start + ramdisk_size;
319 return;
320 }
321
322 relocate_initrd();
323
324 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
325}
326
327#else
328static void __init early_reserve_initrd(void)
329{
330}
331static void __init reserve_initrd(void)
332{
333}
334#endif /* CONFIG_BLK_DEV_INITRD */
335
336static void __init parse_setup_data(void)
337{
338 struct setup_data *data;
339 u64 pa_data, pa_next;
340
341 pa_data = boot_params.hdr.setup_data;
342 while (pa_data) {
343 u32 data_len, data_type;
344
345 data = early_memremap(pa_data, sizeof(*data));
346 data_len = data->len + sizeof(struct setup_data);
347 data_type = data->type;
348 pa_next = data->next;
349 early_memunmap(data, sizeof(*data));
350
351 switch (data_type) {
352 case SETUP_E820_EXT:
353 e820__memory_setup_extended(pa_data, data_len);
354 break;
355 case SETUP_DTB:
356 add_dtb(pa_data);
357 break;
358 case SETUP_EFI:
359 parse_efi_setup(pa_data, data_len);
360 break;
361 default:
362 break;
363 }
364 pa_data = pa_next;
365 }
366}
367
368static void __init memblock_x86_reserve_range_setup_data(void)
369{
370 struct setup_data *data;
371 u64 pa_data;
372
373 pa_data = boot_params.hdr.setup_data;
374 while (pa_data) {
375 data = early_memremap(pa_data, sizeof(*data));
376 memblock_reserve(pa_data, sizeof(*data) + data->len);
377
378 if (data->type == SETUP_INDIRECT &&
379 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
380 memblock_reserve(((struct setup_indirect *)data->data)->addr,
381 ((struct setup_indirect *)data->data)->len);
382
383 pa_data = data->next;
384 early_memunmap(data, sizeof(*data));
385 }
386}
387
388/*
389 * --------- Crashkernel reservation ------------------------------
390 */
391
392#ifdef CONFIG_KEXEC_CORE
393
394/* 16M alignment for crash kernel regions */
395#define CRASH_ALIGN SZ_16M
396
397/*
398 * Keep the crash kernel below this limit.
399 *
400 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
401 * due to mapping restrictions.
402 *
403 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
404 * the upper limit of system RAM in 4-level paging mode. Since the kdump
405 * jump could be from 5-level paging to 4-level paging, the jump will fail if
406 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
407 * no good way to detect the paging mode of the target kernel which will be
408 * loaded for dumping.
409 */
410#ifdef CONFIG_X86_32
411# define CRASH_ADDR_LOW_MAX SZ_512M
412# define CRASH_ADDR_HIGH_MAX SZ_512M
413#else
414# define CRASH_ADDR_LOW_MAX SZ_4G
415# define CRASH_ADDR_HIGH_MAX SZ_64T
416#endif
417
418static int __init reserve_crashkernel_low(void)
419{
420#ifdef CONFIG_X86_64
421 unsigned long long base, low_base = 0, low_size = 0;
422 unsigned long low_mem_limit;
423 int ret;
424
425 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
426
427 /* crashkernel=Y,low */
428 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
429 if (ret) {
430 /*
431 * two parts from kernel/dma/swiotlb.c:
432 * -swiotlb size: user-specified with swiotlb= or default.
433 *
434 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
435 * to 8M for other buffers that may need to stay low too. Also
436 * make sure we allocate enough extra low memory so that we
437 * don't run out of DMA buffers for 32-bit devices.
438 */
439 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
440 } else {
441 /* passed with crashkernel=0,low ? */
442 if (!low_size)
443 return 0;
444 }
445
446 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
447 if (!low_base) {
448 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
449 (unsigned long)(low_size >> 20));
450 return -ENOMEM;
451 }
452
453 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
454 (unsigned long)(low_size >> 20),
455 (unsigned long)(low_base >> 20),
456 (unsigned long)(low_mem_limit >> 20));
457
458 crashk_low_res.start = low_base;
459 crashk_low_res.end = low_base + low_size - 1;
460 insert_resource(&iomem_resource, &crashk_low_res);
461#endif
462 return 0;
463}
464
465static void __init reserve_crashkernel(void)
466{
467 unsigned long long crash_size, crash_base, total_mem;
468 bool high = false;
469 int ret;
470
471 total_mem = memblock_phys_mem_size();
472
473 /* crashkernel=XM */
474 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
475 if (ret != 0 || crash_size <= 0) {
476 /* crashkernel=X,high */
477 ret = parse_crashkernel_high(boot_command_line, total_mem,
478 &crash_size, &crash_base);
479 if (ret != 0 || crash_size <= 0)
480 return;
481 high = true;
482 }
483
484 if (xen_pv_domain()) {
485 pr_info("Ignoring crashkernel for a Xen PV domain\n");
486 return;
487 }
488
489 /* 0 means: find the address automatically */
490 if (!crash_base) {
491 /*
492 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
493 * crashkernel=x,high reserves memory over 4G, also allocates
494 * 256M extra low memory for DMA buffers and swiotlb.
495 * But the extra memory is not required for all machines.
496 * So try low memory first and fall back to high memory
497 * unless "crashkernel=size[KMG],high" is specified.
498 */
499 if (!high)
500 crash_base = memblock_phys_alloc_range(crash_size,
501 CRASH_ALIGN, CRASH_ALIGN,
502 CRASH_ADDR_LOW_MAX);
503 if (!crash_base)
504 crash_base = memblock_phys_alloc_range(crash_size,
505 CRASH_ALIGN, CRASH_ALIGN,
506 CRASH_ADDR_HIGH_MAX);
507 if (!crash_base) {
508 pr_info("crashkernel reservation failed - No suitable area found.\n");
509 return;
510 }
511 } else {
512 unsigned long long start;
513
514 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
515 crash_base + crash_size);
516 if (start != crash_base) {
517 pr_info("crashkernel reservation failed - memory is in use.\n");
518 return;
519 }
520 }
521
522 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
523 memblock_free(crash_base, crash_size);
524 return;
525 }
526
527 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
528 (unsigned long)(crash_size >> 20),
529 (unsigned long)(crash_base >> 20),
530 (unsigned long)(total_mem >> 20));
531
532 crashk_res.start = crash_base;
533 crashk_res.end = crash_base + crash_size - 1;
534 insert_resource(&iomem_resource, &crashk_res);
535}
536#else
537static void __init reserve_crashkernel(void)
538{
539}
540#endif
541
542static struct resource standard_io_resources[] = {
543 { .name = "dma1", .start = 0x00, .end = 0x1f,
544 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
545 { .name = "pic1", .start = 0x20, .end = 0x21,
546 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
547 { .name = "timer0", .start = 0x40, .end = 0x43,
548 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
549 { .name = "timer1", .start = 0x50, .end = 0x53,
550 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
551 { .name = "keyboard", .start = 0x60, .end = 0x60,
552 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
553 { .name = "keyboard", .start = 0x64, .end = 0x64,
554 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
555 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
556 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
557 { .name = "pic2", .start = 0xa0, .end = 0xa1,
558 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
559 { .name = "dma2", .start = 0xc0, .end = 0xdf,
560 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
561 { .name = "fpu", .start = 0xf0, .end = 0xff,
562 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
563};
564
565void __init reserve_standard_io_resources(void)
566{
567 int i;
568
569 /* request I/O space for devices used on all i[345]86 PCs */
570 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
571 request_resource(&ioport_resource, &standard_io_resources[i]);
572
573}
574
575static __init void reserve_ibft_region(void)
576{
577 unsigned long addr, size = 0;
578
579 addr = find_ibft_region(&size);
580
581 if (size)
582 memblock_reserve(addr, size);
583}
584
585static bool __init snb_gfx_workaround_needed(void)
586{
587#ifdef CONFIG_PCI
588 int i;
589 u16 vendor, devid;
590 static const __initconst u16 snb_ids[] = {
591 0x0102,
592 0x0112,
593 0x0122,
594 0x0106,
595 0x0116,
596 0x0126,
597 0x010a,
598 };
599
600 /* Assume no if something weird is going on with PCI */
601 if (!early_pci_allowed())
602 return false;
603
604 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
605 if (vendor != 0x8086)
606 return false;
607
608 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
609 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
610 if (devid == snb_ids[i])
611 return true;
612#endif
613
614 return false;
615}
616
617/*
618 * Sandy Bridge graphics has trouble with certain ranges, exclude
619 * them from allocation.
620 */
621static void __init trim_snb_memory(void)
622{
623 static const __initconst unsigned long bad_pages[] = {
624 0x20050000,
625 0x20110000,
626 0x20130000,
627 0x20138000,
628 0x40004000,
629 };
630 int i;
631
632 if (!snb_gfx_workaround_needed())
633 return;
634
635 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
636
637 /*
638 * SandyBridge integrated graphics devices have a bug that prevents
639 * them from accessing certain memory ranges, namely anything below
640 * 1M and in the pages listed in bad_pages[] above.
641 *
642 * To avoid these pages being ever accessed by SNB gfx devices reserve
643 * bad_pages that have not already been reserved at boot time.
644 * All memory below the 1 MB mark is anyway reserved later during
645 * setup_arch(), so there is no need to reserve it here.
646 */
647
648 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
649 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
650 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
651 bad_pages[i]);
652 }
653}
654
655static void __init trim_bios_range(void)
656{
657 /*
658 * A special case is the first 4Kb of memory;
659 * This is a BIOS owned area, not kernel ram, but generally
660 * not listed as such in the E820 table.
661 *
662 * This typically reserves additional memory (64KiB by default)
663 * since some BIOSes are known to corrupt low memory. See the
664 * Kconfig help text for X86_RESERVE_LOW.
665 */
666 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
667
668 /*
669 * special case: Some BIOSes report the PC BIOS
670 * area (640Kb -> 1Mb) as RAM even though it is not.
671 * take them out.
672 */
673 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
674
675 e820__update_table(e820_table);
676}
677
678/* called before trim_bios_range() to spare extra sanitize */
679static void __init e820_add_kernel_range(void)
680{
681 u64 start = __pa_symbol(_text);
682 u64 size = __pa_symbol(_end) - start;
683
684 /*
685 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
686 * attempt to fix it by adding the range. We may have a confused BIOS,
687 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
688 * exclude kernel range. If we really are running on top non-RAM,
689 * we will crash later anyways.
690 */
691 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
692 return;
693
694 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
695 e820__range_remove(start, size, E820_TYPE_RAM, 0);
696 e820__range_add(start, size, E820_TYPE_RAM);
697}
698
699static void __init early_reserve_memory(void)
700{
701 /*
702 * Reserve the memory occupied by the kernel between _text and
703 * __end_of_kernel_reserve symbols. Any kernel sections after the
704 * __end_of_kernel_reserve symbol must be explicitly reserved with a
705 * separate memblock_reserve() or they will be discarded.
706 */
707 memblock_reserve(__pa_symbol(_text),
708 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
709
710 /*
711 * The first 4Kb of memory is a BIOS owned area, but generally it is
712 * not listed as such in the E820 table.
713 *
714 * Reserve the first 64K of memory since some BIOSes are known to
715 * corrupt low memory. After the real mode trampoline is allocated the
716 * rest of the memory below 640k is reserved.
717 *
718 * In addition, make sure page 0 is always reserved because on
719 * systems with L1TF its contents can be leaked to user processes.
720 */
721 memblock_reserve(0, SZ_64K);
722
723 early_reserve_initrd();
724
725 if (efi_enabled(EFI_BOOT))
726 efi_memblock_x86_reserve_range();
727
728 memblock_x86_reserve_range_setup_data();
729
730 reserve_ibft_region();
731 reserve_bios_regions();
732 trim_snb_memory();
733}
734
735/*
736 * Dump out kernel offset information on panic.
737 */
738static int
739dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
740{
741 if (kaslr_enabled()) {
742 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
743 kaslr_offset(),
744 __START_KERNEL,
745 __START_KERNEL_map,
746 MODULES_VADDR-1);
747 } else {
748 pr_emerg("Kernel Offset: disabled\n");
749 }
750
751 return 0;
752}
753
754/*
755 * Determine if we were loaded by an EFI loader. If so, then we have also been
756 * passed the efi memmap, systab, etc., so we should use these data structures
757 * for initialization. Note, the efi init code path is determined by the
758 * global efi_enabled. This allows the same kernel image to be used on existing
759 * systems (with a traditional BIOS) as well as on EFI systems.
760 */
761/*
762 * setup_arch - architecture-specific boot-time initializations
763 *
764 * Note: On x86_64, fixmaps are ready for use even before this is called.
765 */
766
767void __init setup_arch(char **cmdline_p)
768{
769#ifdef CONFIG_X86_32
770 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
771
772 /*
773 * copy kernel address range established so far and switch
774 * to the proper swapper page table
775 */
776 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
777 initial_page_table + KERNEL_PGD_BOUNDARY,
778 KERNEL_PGD_PTRS);
779
780 load_cr3(swapper_pg_dir);
781 /*
782 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
783 * a cr3 based tlb flush, so the following __flush_tlb_all()
784 * will not flush anything because the CPU quirk which clears
785 * X86_FEATURE_PGE has not been invoked yet. Though due to the
786 * load_cr3() above the TLB has been flushed already. The
787 * quirk is invoked before subsequent calls to __flush_tlb_all()
788 * so proper operation is guaranteed.
789 */
790 __flush_tlb_all();
791#else
792 printk(KERN_INFO "Command line: %s\n", boot_command_line);
793 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
794#endif
795
796 /*
797 * If we have OLPC OFW, we might end up relocating the fixmap due to
798 * reserve_top(), so do this before touching the ioremap area.
799 */
800 olpc_ofw_detect();
801
802 idt_setup_early_traps();
803 early_cpu_init();
804 jump_label_init();
805 static_call_init();
806 early_ioremap_init();
807
808 setup_olpc_ofw_pgd();
809
810 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
811 screen_info = boot_params.screen_info;
812 edid_info = boot_params.edid_info;
813#ifdef CONFIG_X86_32
814 apm_info.bios = boot_params.apm_bios_info;
815 ist_info = boot_params.ist_info;
816#endif
817 saved_video_mode = boot_params.hdr.vid_mode;
818 bootloader_type = boot_params.hdr.type_of_loader;
819 if ((bootloader_type >> 4) == 0xe) {
820 bootloader_type &= 0xf;
821 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
822 }
823 bootloader_version = bootloader_type & 0xf;
824 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
825
826#ifdef CONFIG_BLK_DEV_RAM
827 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
828#endif
829#ifdef CONFIG_EFI
830 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
831 EFI32_LOADER_SIGNATURE, 4)) {
832 set_bit(EFI_BOOT, &efi.flags);
833 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
834 EFI64_LOADER_SIGNATURE, 4)) {
835 set_bit(EFI_BOOT, &efi.flags);
836 set_bit(EFI_64BIT, &efi.flags);
837 }
838#endif
839
840 x86_init.oem.arch_setup();
841
842 /*
843 * Do some memory reservations *before* memory is added to memblock, so
844 * memblock allocations won't overwrite it.
845 *
846 * After this point, everything still needed from the boot loader or
847 * firmware or kernel text should be early reserved or marked not RAM in
848 * e820. All other memory is free game.
849 *
850 * This call needs to happen before e820__memory_setup() which calls the
851 * xen_memory_setup() on Xen dom0 which relies on the fact that those
852 * early reservations have happened already.
853 */
854 early_reserve_memory();
855
856 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
857 e820__memory_setup();
858 parse_setup_data();
859
860 copy_edd();
861
862 if (!boot_params.hdr.root_flags)
863 root_mountflags &= ~MS_RDONLY;
864 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
865
866 code_resource.start = __pa_symbol(_text);
867 code_resource.end = __pa_symbol(_etext)-1;
868 rodata_resource.start = __pa_symbol(__start_rodata);
869 rodata_resource.end = __pa_symbol(__end_rodata)-1;
870 data_resource.start = __pa_symbol(_sdata);
871 data_resource.end = __pa_symbol(_edata)-1;
872 bss_resource.start = __pa_symbol(__bss_start);
873 bss_resource.end = __pa_symbol(__bss_stop)-1;
874
875#ifdef CONFIG_CMDLINE_BOOL
876#ifdef CONFIG_CMDLINE_OVERRIDE
877 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
878#else
879 if (builtin_cmdline[0]) {
880 /* append boot loader cmdline to builtin */
881 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
882 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
883 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
884 }
885#endif
886#endif
887
888 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
889 *cmdline_p = command_line;
890
891 /*
892 * x86_configure_nx() is called before parse_early_param() to detect
893 * whether hardware doesn't support NX (so that the early EHCI debug
894 * console setup can safely call set_fixmap()). It may then be called
895 * again from within noexec_setup() during parsing early parameters
896 * to honor the respective command line option.
897 */
898 x86_configure_nx();
899
900 parse_early_param();
901
902#ifdef CONFIG_MEMORY_HOTPLUG
903 /*
904 * Memory used by the kernel cannot be hot-removed because Linux
905 * cannot migrate the kernel pages. When memory hotplug is
906 * enabled, we should prevent memblock from allocating memory
907 * for the kernel.
908 *
909 * ACPI SRAT records all hotpluggable memory ranges. But before
910 * SRAT is parsed, we don't know about it.
911 *
912 * The kernel image is loaded into memory at very early time. We
913 * cannot prevent this anyway. So on NUMA system, we set any
914 * node the kernel resides in as un-hotpluggable.
915 *
916 * Since on modern servers, one node could have double-digit
917 * gigabytes memory, we can assume the memory around the kernel
918 * image is also un-hotpluggable. So before SRAT is parsed, just
919 * allocate memory near the kernel image to try the best to keep
920 * the kernel away from hotpluggable memory.
921 */
922 if (movable_node_is_enabled())
923 memblock_set_bottom_up(true);
924#endif
925
926 x86_report_nx();
927
928 if (acpi_mps_check()) {
929#ifdef CONFIG_X86_LOCAL_APIC
930 disable_apic = 1;
931#endif
932 setup_clear_cpu_cap(X86_FEATURE_APIC);
933 }
934
935 e820__reserve_setup_data();
936 e820__finish_early_params();
937
938 if (efi_enabled(EFI_BOOT))
939 efi_init();
940
941 dmi_setup();
942
943 /*
944 * VMware detection requires dmi to be available, so this
945 * needs to be done after dmi_setup(), for the boot CPU.
946 */
947 init_hypervisor_platform();
948
949 tsc_early_init();
950 x86_init.resources.probe_roms();
951
952 /* after parse_early_param, so could debug it */
953 insert_resource(&iomem_resource, &code_resource);
954 insert_resource(&iomem_resource, &rodata_resource);
955 insert_resource(&iomem_resource, &data_resource);
956 insert_resource(&iomem_resource, &bss_resource);
957
958 e820_add_kernel_range();
959 trim_bios_range();
960#ifdef CONFIG_X86_32
961 if (ppro_with_ram_bug()) {
962 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
963 E820_TYPE_RESERVED);
964 e820__update_table(e820_table);
965 printk(KERN_INFO "fixed physical RAM map:\n");
966 e820__print_table("bad_ppro");
967 }
968#else
969 early_gart_iommu_check();
970#endif
971
972 /*
973 * partially used pages are not usable - thus
974 * we are rounding upwards:
975 */
976 max_pfn = e820__end_of_ram_pfn();
977
978 /* update e820 for memory not covered by WB MTRRs */
979 mtrr_bp_init();
980 if (mtrr_trim_uncached_memory(max_pfn))
981 max_pfn = e820__end_of_ram_pfn();
982
983 max_possible_pfn = max_pfn;
984
985 /*
986 * This call is required when the CPU does not support PAT. If
987 * mtrr_bp_init() invoked it already via pat_init() the call has no
988 * effect.
989 */
990 init_cache_modes();
991
992 /*
993 * Define random base addresses for memory sections after max_pfn is
994 * defined and before each memory section base is used.
995 */
996 kernel_randomize_memory();
997
998#ifdef CONFIG_X86_32
999 /* max_low_pfn get updated here */
1000 find_low_pfn_range();
1001#else
1002 check_x2apic();
1003
1004 /* How many end-of-memory variables you have, grandma! */
1005 /* need this before calling reserve_initrd */
1006 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1007 max_low_pfn = e820__end_of_low_ram_pfn();
1008 else
1009 max_low_pfn = max_pfn;
1010
1011 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1012#endif
1013
1014 /*
1015 * Find and reserve possible boot-time SMP configuration:
1016 */
1017 find_smp_config();
1018
1019 early_alloc_pgt_buf();
1020
1021 /*
1022 * Need to conclude brk, before e820__memblock_setup()
1023 * it could use memblock_find_in_range, could overlap with
1024 * brk area.
1025 */
1026 reserve_brk();
1027
1028 cleanup_highmap();
1029
1030 memblock_set_current_limit(ISA_END_ADDRESS);
1031 e820__memblock_setup();
1032
1033 /*
1034 * Needs to run after memblock setup because it needs the physical
1035 * memory size.
1036 */
1037 sev_setup_arch();
1038
1039 efi_fake_memmap();
1040 efi_find_mirror();
1041 efi_esrt_init();
1042 efi_mokvar_table_init();
1043
1044 /*
1045 * The EFI specification says that boot service code won't be
1046 * called after ExitBootServices(). This is, in fact, a lie.
1047 */
1048 efi_reserve_boot_services();
1049
1050 /* preallocate 4k for mptable mpc */
1051 e820__memblock_alloc_reserved_mpc_new();
1052
1053#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1054 setup_bios_corruption_check();
1055#endif
1056
1057#ifdef CONFIG_X86_32
1058 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1059 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1060#endif
1061
1062 /*
1063 * Find free memory for the real mode trampoline and place it there. If
1064 * there is not enough free memory under 1M, on EFI-enabled systems
1065 * there will be additional attempt to reclaim the memory for the real
1066 * mode trampoline at efi_free_boot_services().
1067 *
1068 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1069 * are known to corrupt low memory and several hundred kilobytes are not
1070 * worth complex detection what memory gets clobbered. Windows does the
1071 * same thing for very similar reasons.
1072 *
1073 * Moreover, on machines with SandyBridge graphics or in setups that use
1074 * crashkernel the entire 1M is reserved anyway.
1075 */
1076 reserve_real_mode();
1077
1078 init_mem_mapping();
1079
1080 idt_setup_early_pf();
1081
1082 /*
1083 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1084 * with the current CR4 value. This may not be necessary, but
1085 * auditing all the early-boot CR4 manipulation would be needed to
1086 * rule it out.
1087 *
1088 * Mask off features that don't work outside long mode (just
1089 * PCIDE for now).
1090 */
1091 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1092
1093 memblock_set_current_limit(get_max_mapped());
1094
1095 /*
1096 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1097 */
1098
1099#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1100 if (init_ohci1394_dma_early)
1101 init_ohci1394_dma_on_all_controllers();
1102#endif
1103 /* Allocate bigger log buffer */
1104 setup_log_buf(1);
1105
1106 if (efi_enabled(EFI_BOOT)) {
1107 switch (boot_params.secure_boot) {
1108 case efi_secureboot_mode_disabled:
1109 pr_info("Secure boot disabled\n");
1110 break;
1111 case efi_secureboot_mode_enabled:
1112 pr_info("Secure boot enabled\n");
1113 break;
1114 default:
1115 pr_info("Secure boot could not be determined\n");
1116 break;
1117 }
1118 }
1119
1120 reserve_initrd();
1121
1122 acpi_table_upgrade();
1123 /* Look for ACPI tables and reserve memory occupied by them. */
1124 acpi_boot_table_init();
1125
1126 vsmp_init();
1127
1128 io_delay_init();
1129
1130 early_platform_quirks();
1131
1132 early_acpi_boot_init();
1133
1134 initmem_init();
1135 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1136
1137 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1138 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1139
1140 /*
1141 * Reserve memory for crash kernel after SRAT is parsed so that it
1142 * won't consume hotpluggable memory.
1143 */
1144 reserve_crashkernel();
1145
1146 memblock_find_dma_reserve();
1147
1148 if (!early_xdbc_setup_hardware())
1149 early_xdbc_register_console();
1150
1151 x86_init.paging.pagetable_init();
1152
1153 kasan_init();
1154
1155 /*
1156 * Sync back kernel address range.
1157 *
1158 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1159 * this call?
1160 */
1161 sync_initial_page_table();
1162
1163 tboot_probe();
1164
1165 map_vsyscall();
1166
1167 generic_apic_probe();
1168
1169 early_quirks();
1170
1171 /*
1172 * Read APIC and some other early information from ACPI tables.
1173 */
1174 acpi_boot_init();
1175 x86_dtb_init();
1176
1177 /*
1178 * get boot-time SMP configuration:
1179 */
1180 get_smp_config();
1181
1182 /*
1183 * Systems w/o ACPI and mptables might not have it mapped the local
1184 * APIC yet, but prefill_possible_map() might need to access it.
1185 */
1186 init_apic_mappings();
1187
1188 prefill_possible_map();
1189
1190 init_cpu_to_node();
1191 init_gi_nodes();
1192
1193 io_apic_init_mappings();
1194
1195 x86_init.hyper.guest_late_init();
1196
1197 e820__reserve_resources();
1198 e820__register_nosave_regions(max_pfn);
1199
1200 x86_init.resources.reserve_resources();
1201
1202 e820__setup_pci_gap();
1203
1204#ifdef CONFIG_VT
1205#if defined(CONFIG_VGA_CONSOLE)
1206 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1207 conswitchp = &vga_con;
1208#endif
1209#endif
1210 x86_init.oem.banner();
1211
1212 x86_init.timers.wallclock_init();
1213
1214 /*
1215 * This needs to run before setup_local_APIC() which soft-disables the
1216 * local APIC temporarily and that masks the thermal LVT interrupt,
1217 * leading to softlockups on machines which have configured SMI
1218 * interrupt delivery.
1219 */
1220 therm_lvt_init();
1221
1222 mcheck_init();
1223
1224 register_refined_jiffies(CLOCK_TICK_RATE);
1225
1226#ifdef CONFIG_EFI
1227 if (efi_enabled(EFI_BOOT))
1228 efi_apply_memmap_quirks();
1229#endif
1230
1231 unwind_init();
1232}
1233
1234#ifdef CONFIG_X86_32
1235
1236static struct resource video_ram_resource = {
1237 .name = "Video RAM area",
1238 .start = 0xa0000,
1239 .end = 0xbffff,
1240 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1241};
1242
1243void __init i386_reserve_resources(void)
1244{
1245 request_resource(&iomem_resource, &video_ram_resource);
1246 reserve_standard_io_resources();
1247}
1248
1249#endif /* CONFIG_X86_32 */
1250
1251static struct notifier_block kernel_offset_notifier = {
1252 .notifier_call = dump_kernel_offset
1253};
1254
1255static int __init register_kernel_offset_dumper(void)
1256{
1257 atomic_notifier_chain_register(&panic_notifier_list,
1258 &kernel_offset_notifier);
1259 return 0;
1260}
1261__initcall(register_kernel_offset_dumper);