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#include <linux/jiffies.h>
72
73#include <video/edid.h>
74
75#include <asm/mtrr.h>
76#include <asm/apic.h>
77#include <asm/realmode.h>
78#include <asm/e820.h>
79#include <asm/mpspec.h>
80#include <asm/setup.h>
81#include <asm/efi.h>
82#include <asm/timer.h>
83#include <asm/i8259.h>
84#include <asm/sections.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#include <asm/kasan.h>
93
94#include <asm/vsyscall.h>
95#include <asm/cpu.h>
96#include <asm/desc.h>
97#include <asm/dma.h>
98#include <asm/iommu.h>
99#include <asm/gart.h>
100#include <asm/mmu_context.h>
101#include <asm/proto.h>
102
103#include <asm/paravirt.h>
104#include <asm/hypervisor.h>
105#include <asm/olpc_ofw.h>
106
107#include <asm/percpu.h>
108#include <asm/topology.h>
109#include <asm/apicdef.h>
110#include <asm/amd_nb.h>
111#include <asm/mce.h>
112#include <asm/alternative.h>
113#include <asm/prom.h>
114#include <asm/microcode.h>
115#include <asm/mmu_context.h>
116
117/*
118 * max_low_pfn_mapped: highest direct mapped pfn under 4GB
119 * max_pfn_mapped: highest direct mapped pfn over 4GB
120 *
121 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
122 * represented by pfn_mapped
123 */
124unsigned long max_low_pfn_mapped;
125unsigned long max_pfn_mapped;
126
127#ifdef CONFIG_DMI
128RESERVE_BRK(dmi_alloc, 65536);
129#endif
130
131
132static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
133unsigned long _brk_end = (unsigned long)__brk_base;
134
135#ifdef CONFIG_X86_64
136int default_cpu_present_to_apicid(int mps_cpu)
137{
138 return __default_cpu_present_to_apicid(mps_cpu);
139}
140
141int default_check_phys_apicid_present(int phys_apicid)
142{
143 return __default_check_phys_apicid_present(phys_apicid);
144}
145#endif
146
147struct boot_params boot_params;
148
149/*
150 * Machine setup..
151 */
152static struct resource data_resource = {
153 .name = "Kernel data",
154 .start = 0,
155 .end = 0,
156 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
157};
158
159static struct resource code_resource = {
160 .name = "Kernel code",
161 .start = 0,
162 .end = 0,
163 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
164};
165
166static struct resource bss_resource = {
167 .name = "Kernel bss",
168 .start = 0,
169 .end = 0,
170 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
171};
172
173
174#ifdef CONFIG_X86_32
175/* cpu data as detected by the assembly code in head.S */
176struct cpuinfo_x86 new_cpu_data = {
177 .wp_works_ok = -1,
178};
179/* common cpu data for all cpus */
180struct cpuinfo_x86 boot_cpu_data __read_mostly = {
181 .wp_works_ok = -1,
182};
183EXPORT_SYMBOL(boot_cpu_data);
184
185unsigned int def_to_bigsmp;
186
187/* for MCA, but anyone else can use it if they want */
188unsigned int machine_id;
189unsigned int machine_submodel_id;
190unsigned int BIOS_revision;
191
192struct apm_info apm_info;
193EXPORT_SYMBOL(apm_info);
194
195#if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
196 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
197struct ist_info ist_info;
198EXPORT_SYMBOL(ist_info);
199#else
200struct ist_info ist_info;
201#endif
202
203#else
204struct cpuinfo_x86 boot_cpu_data __read_mostly = {
205 .x86_phys_bits = MAX_PHYSMEM_BITS,
206};
207EXPORT_SYMBOL(boot_cpu_data);
208#endif
209
210
211#if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
212__visible unsigned long mmu_cr4_features;
213#else
214__visible unsigned long mmu_cr4_features = X86_CR4_PAE;
215#endif
216
217/* Boot loader ID and version as integers, for the benefit of proc_dointvec */
218int bootloader_type, bootloader_version;
219
220/*
221 * Setup options
222 */
223struct screen_info screen_info;
224EXPORT_SYMBOL(screen_info);
225struct edid_info edid_info;
226EXPORT_SYMBOL_GPL(edid_info);
227
228extern int root_mountflags;
229
230unsigned long saved_video_mode;
231
232#define RAMDISK_IMAGE_START_MASK 0x07FF
233#define RAMDISK_PROMPT_FLAG 0x8000
234#define RAMDISK_LOAD_FLAG 0x4000
235
236static char __initdata command_line[COMMAND_LINE_SIZE];
237#ifdef CONFIG_CMDLINE_BOOL
238static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
239#endif
240
241#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
242struct edd edd;
243#ifdef CONFIG_EDD_MODULE
244EXPORT_SYMBOL(edd);
245#endif
246/**
247 * copy_edd() - Copy the BIOS EDD information
248 * from boot_params into a safe place.
249 *
250 */
251static inline void __init copy_edd(void)
252{
253 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
254 sizeof(edd.mbr_signature));
255 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
256 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
257 edd.edd_info_nr = boot_params.eddbuf_entries;
258}
259#else
260static inline void __init copy_edd(void)
261{
262}
263#endif
264
265void * __init extend_brk(size_t size, size_t align)
266{
267 size_t mask = align - 1;
268 void *ret;
269
270 BUG_ON(_brk_start == 0);
271 BUG_ON(align & mask);
272
273 _brk_end = (_brk_end + mask) & ~mask;
274 BUG_ON((char *)(_brk_end + size) > __brk_limit);
275
276 ret = (void *)_brk_end;
277 _brk_end += size;
278
279 memset(ret, 0, size);
280
281 return ret;
282}
283
284#ifdef CONFIG_X86_32
285static void __init cleanup_highmap(void)
286{
287}
288#endif
289
290static void __init reserve_brk(void)
291{
292 if (_brk_end > _brk_start)
293 memblock_reserve(__pa_symbol(_brk_start),
294 _brk_end - _brk_start);
295
296 /* Mark brk area as locked down and no longer taking any
297 new allocations */
298 _brk_start = 0;
299}
300
301u64 relocated_ramdisk;
302
303#ifdef CONFIG_BLK_DEV_INITRD
304
305static u64 __init get_ramdisk_image(void)
306{
307 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
308
309 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
310
311 return ramdisk_image;
312}
313static u64 __init get_ramdisk_size(void)
314{
315 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
316
317 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
318
319 return ramdisk_size;
320}
321
322static void __init relocate_initrd(void)
323{
324 /* Assume only end is not page aligned */
325 u64 ramdisk_image = get_ramdisk_image();
326 u64 ramdisk_size = get_ramdisk_size();
327 u64 area_size = PAGE_ALIGN(ramdisk_size);
328
329 /* We need to move the initrd down into directly mapped mem */
330 relocated_ramdisk = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
331 area_size, PAGE_SIZE);
332
333 if (!relocated_ramdisk)
334 panic("Cannot find place for new RAMDISK of size %lld\n",
335 ramdisk_size);
336
337 /* Note: this includes all the mem currently occupied by
338 the initrd, we rely on that fact to keep the data intact. */
339 memblock_reserve(relocated_ramdisk, area_size);
340 initrd_start = relocated_ramdisk + PAGE_OFFSET;
341 initrd_end = initrd_start + ramdisk_size;
342 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
343 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
344
345 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
346
347 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
348 " [mem %#010llx-%#010llx]\n",
349 ramdisk_image, ramdisk_image + ramdisk_size - 1,
350 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
351}
352
353static void __init early_reserve_initrd(void)
354{
355 /* Assume only end is not page aligned */
356 u64 ramdisk_image = get_ramdisk_image();
357 u64 ramdisk_size = get_ramdisk_size();
358 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
359
360 if (!boot_params.hdr.type_of_loader ||
361 !ramdisk_image || !ramdisk_size)
362 return; /* No initrd provided by bootloader */
363
364 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
365}
366static void __init reserve_initrd(void)
367{
368 /* Assume only end is not page aligned */
369 u64 ramdisk_image = get_ramdisk_image();
370 u64 ramdisk_size = get_ramdisk_size();
371 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
372 u64 mapped_size;
373
374 if (!boot_params.hdr.type_of_loader ||
375 !ramdisk_image || !ramdisk_size)
376 return; /* No initrd provided by bootloader */
377
378 initrd_start = 0;
379
380 mapped_size = memblock_mem_size(max_pfn_mapped);
381 if (ramdisk_size >= (mapped_size>>1))
382 panic("initrd too large to handle, "
383 "disabling initrd (%lld needed, %lld available)\n",
384 ramdisk_size, mapped_size>>1);
385
386 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
387 ramdisk_end - 1);
388
389 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
390 PFN_DOWN(ramdisk_end))) {
391 /* All are mapped, easy case */
392 initrd_start = ramdisk_image + PAGE_OFFSET;
393 initrd_end = initrd_start + ramdisk_size;
394 return;
395 }
396
397 relocate_initrd();
398
399 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
400}
401#else
402static void __init early_reserve_initrd(void)
403{
404}
405static void __init reserve_initrd(void)
406{
407}
408#endif /* CONFIG_BLK_DEV_INITRD */
409
410static void __init parse_setup_data(void)
411{
412 struct setup_data *data;
413 u64 pa_data, pa_next;
414
415 pa_data = boot_params.hdr.setup_data;
416 while (pa_data) {
417 u32 data_len, data_type;
418
419 data = early_memremap(pa_data, sizeof(*data));
420 data_len = data->len + sizeof(struct setup_data);
421 data_type = data->type;
422 pa_next = data->next;
423 early_memunmap(data, sizeof(*data));
424
425 switch (data_type) {
426 case SETUP_E820_EXT:
427 parse_e820_ext(pa_data, data_len);
428 break;
429 case SETUP_DTB:
430 add_dtb(pa_data);
431 break;
432 case SETUP_EFI:
433 parse_efi_setup(pa_data, data_len);
434 break;
435 default:
436 break;
437 }
438 pa_data = pa_next;
439 }
440}
441
442static void __init e820_reserve_setup_data(void)
443{
444 struct setup_data *data;
445 u64 pa_data;
446
447 pa_data = boot_params.hdr.setup_data;
448 if (!pa_data)
449 return;
450
451 while (pa_data) {
452 data = early_memremap(pa_data, sizeof(*data));
453 e820_update_range(pa_data, sizeof(*data)+data->len,
454 E820_RAM, E820_RESERVED_KERN);
455 pa_data = data->next;
456 early_memunmap(data, sizeof(*data));
457 }
458
459 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
460 memcpy(&e820_saved, &e820, sizeof(struct e820map));
461 printk(KERN_INFO "extended physical RAM map:\n");
462 e820_print_map("reserve setup_data");
463}
464
465static void __init memblock_x86_reserve_range_setup_data(void)
466{
467 struct setup_data *data;
468 u64 pa_data;
469
470 pa_data = boot_params.hdr.setup_data;
471 while (pa_data) {
472 data = early_memremap(pa_data, sizeof(*data));
473 memblock_reserve(pa_data, sizeof(*data) + data->len);
474 pa_data = data->next;
475 early_memunmap(data, sizeof(*data));
476 }
477}
478
479/*
480 * --------- Crashkernel reservation ------------------------------
481 */
482
483#ifdef CONFIG_KEXEC_CORE
484
485/* 16M alignment for crash kernel regions */
486#define CRASH_ALIGN (16 << 20)
487
488/*
489 * Keep the crash kernel below this limit. On 32 bits earlier kernels
490 * would limit the kernel to the low 512 MiB due to mapping restrictions.
491 * On 64bit, old kexec-tools need to under 896MiB.
492 */
493#ifdef CONFIG_X86_32
494# define CRASH_ADDR_LOW_MAX (512 << 20)
495# define CRASH_ADDR_HIGH_MAX (512 << 20)
496#else
497# define CRASH_ADDR_LOW_MAX (896UL << 20)
498# define CRASH_ADDR_HIGH_MAX MAXMEM
499#endif
500
501static int __init reserve_crashkernel_low(void)
502{
503#ifdef CONFIG_X86_64
504 unsigned long long base, low_base = 0, low_size = 0;
505 unsigned long total_low_mem;
506 int ret;
507
508 total_low_mem = memblock_mem_size(1UL << (32 - PAGE_SHIFT));
509
510 /* crashkernel=Y,low */
511 ret = parse_crashkernel_low(boot_command_line, total_low_mem, &low_size, &base);
512 if (ret) {
513 /*
514 * two parts from lib/swiotlb.c:
515 * -swiotlb size: user-specified with swiotlb= or default.
516 *
517 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
518 * to 8M for other buffers that may need to stay low too. Also
519 * make sure we allocate enough extra low memory so that we
520 * don't run out of DMA buffers for 32-bit devices.
521 */
522 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
523 } else {
524 /* passed with crashkernel=0,low ? */
525 if (!low_size)
526 return 0;
527 }
528
529 low_base = memblock_find_in_range(low_size, 1ULL << 32, low_size, CRASH_ALIGN);
530 if (!low_base) {
531 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
532 (unsigned long)(low_size >> 20));
533 return -ENOMEM;
534 }
535
536 ret = memblock_reserve(low_base, low_size);
537 if (ret) {
538 pr_err("%s: Error reserving crashkernel low memblock.\n", __func__);
539 return ret;
540 }
541
542 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
543 (unsigned long)(low_size >> 20),
544 (unsigned long)(low_base >> 20),
545 (unsigned long)(total_low_mem >> 20));
546
547 crashk_low_res.start = low_base;
548 crashk_low_res.end = low_base + low_size - 1;
549 insert_resource(&iomem_resource, &crashk_low_res);
550#endif
551 return 0;
552}
553
554static void __init reserve_crashkernel(void)
555{
556 unsigned long long crash_size, crash_base, total_mem;
557 bool high = false;
558 int ret;
559
560 total_mem = memblock_phys_mem_size();
561
562 /* crashkernel=XM */
563 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
564 if (ret != 0 || crash_size <= 0) {
565 /* crashkernel=X,high */
566 ret = parse_crashkernel_high(boot_command_line, total_mem,
567 &crash_size, &crash_base);
568 if (ret != 0 || crash_size <= 0)
569 return;
570 high = true;
571 }
572
573 /* 0 means: find the address automatically */
574 if (crash_base <= 0) {
575 /*
576 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
577 */
578 crash_base = memblock_find_in_range(CRASH_ALIGN,
579 high ? CRASH_ADDR_HIGH_MAX
580 : CRASH_ADDR_LOW_MAX,
581 crash_size, CRASH_ALIGN);
582 if (!crash_base) {
583 pr_info("crashkernel reservation failed - No suitable area found.\n");
584 return;
585 }
586
587 } else {
588 unsigned long long start;
589
590 start = memblock_find_in_range(crash_base,
591 crash_base + crash_size,
592 crash_size, 1 << 20);
593 if (start != crash_base) {
594 pr_info("crashkernel reservation failed - memory is in use.\n");
595 return;
596 }
597 }
598 ret = memblock_reserve(crash_base, crash_size);
599 if (ret) {
600 pr_err("%s: Error reserving crashkernel memblock.\n", __func__);
601 return;
602 }
603
604 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
605 memblock_free(crash_base, crash_size);
606 return;
607 }
608
609 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
610 (unsigned long)(crash_size >> 20),
611 (unsigned long)(crash_base >> 20),
612 (unsigned long)(total_mem >> 20));
613
614 crashk_res.start = crash_base;
615 crashk_res.end = crash_base + crash_size - 1;
616 insert_resource(&iomem_resource, &crashk_res);
617}
618#else
619static void __init reserve_crashkernel(void)
620{
621}
622#endif
623
624static struct resource standard_io_resources[] = {
625 { .name = "dma1", .start = 0x00, .end = 0x1f,
626 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
627 { .name = "pic1", .start = 0x20, .end = 0x21,
628 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
629 { .name = "timer0", .start = 0x40, .end = 0x43,
630 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
631 { .name = "timer1", .start = 0x50, .end = 0x53,
632 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
633 { .name = "keyboard", .start = 0x60, .end = 0x60,
634 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
635 { .name = "keyboard", .start = 0x64, .end = 0x64,
636 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
637 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
638 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
639 { .name = "pic2", .start = 0xa0, .end = 0xa1,
640 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
641 { .name = "dma2", .start = 0xc0, .end = 0xdf,
642 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
643 { .name = "fpu", .start = 0xf0, .end = 0xff,
644 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
645};
646
647void __init reserve_standard_io_resources(void)
648{
649 int i;
650
651 /* request I/O space for devices used on all i[345]86 PCs */
652 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
653 request_resource(&ioport_resource, &standard_io_resources[i]);
654
655}
656
657static __init void reserve_ibft_region(void)
658{
659 unsigned long addr, size = 0;
660
661 addr = find_ibft_region(&size);
662
663 if (size)
664 memblock_reserve(addr, size);
665}
666
667static bool __init snb_gfx_workaround_needed(void)
668{
669#ifdef CONFIG_PCI
670 int i;
671 u16 vendor, devid;
672 static const __initconst u16 snb_ids[] = {
673 0x0102,
674 0x0112,
675 0x0122,
676 0x0106,
677 0x0116,
678 0x0126,
679 0x010a,
680 };
681
682 /* Assume no if something weird is going on with PCI */
683 if (!early_pci_allowed())
684 return false;
685
686 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
687 if (vendor != 0x8086)
688 return false;
689
690 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
691 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
692 if (devid == snb_ids[i])
693 return true;
694#endif
695
696 return false;
697}
698
699/*
700 * Sandy Bridge graphics has trouble with certain ranges, exclude
701 * them from allocation.
702 */
703static void __init trim_snb_memory(void)
704{
705 static const __initconst unsigned long bad_pages[] = {
706 0x20050000,
707 0x20110000,
708 0x20130000,
709 0x20138000,
710 0x40004000,
711 };
712 int i;
713
714 if (!snb_gfx_workaround_needed())
715 return;
716
717 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
718
719 /*
720 * Reserve all memory below the 1 MB mark that has not
721 * already been reserved.
722 */
723 memblock_reserve(0, 1<<20);
724
725 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
726 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
727 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
728 bad_pages[i]);
729 }
730}
731
732/*
733 * Here we put platform-specific memory range workarounds, i.e.
734 * memory known to be corrupt or otherwise in need to be reserved on
735 * specific platforms.
736 *
737 * If this gets used more widely it could use a real dispatch mechanism.
738 */
739static void __init trim_platform_memory_ranges(void)
740{
741 trim_snb_memory();
742}
743
744static void __init trim_bios_range(void)
745{
746 /*
747 * A special case is the first 4Kb of memory;
748 * This is a BIOS owned area, not kernel ram, but generally
749 * not listed as such in the E820 table.
750 *
751 * This typically reserves additional memory (64KiB by default)
752 * since some BIOSes are known to corrupt low memory. See the
753 * Kconfig help text for X86_RESERVE_LOW.
754 */
755 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
756
757 /*
758 * special case: Some BIOSen report the PC BIOS
759 * area (640->1Mb) as ram even though it is not.
760 * take them out.
761 */
762 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
763
764 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
765}
766
767/* called before trim_bios_range() to spare extra sanitize */
768static void __init e820_add_kernel_range(void)
769{
770 u64 start = __pa_symbol(_text);
771 u64 size = __pa_symbol(_end) - start;
772
773 /*
774 * Complain if .text .data and .bss are not marked as E820_RAM and
775 * attempt to fix it by adding the range. We may have a confused BIOS,
776 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
777 * exclude kernel range. If we really are running on top non-RAM,
778 * we will crash later anyways.
779 */
780 if (e820_all_mapped(start, start + size, E820_RAM))
781 return;
782
783 pr_warn(".text .data .bss are not marked as E820_RAM!\n");
784 e820_remove_range(start, size, E820_RAM, 0);
785 e820_add_region(start, size, E820_RAM);
786}
787
788static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
789
790static int __init parse_reservelow(char *p)
791{
792 unsigned long long size;
793
794 if (!p)
795 return -EINVAL;
796
797 size = memparse(p, &p);
798
799 if (size < 4096)
800 size = 4096;
801
802 if (size > 640*1024)
803 size = 640*1024;
804
805 reserve_low = size;
806
807 return 0;
808}
809
810early_param("reservelow", parse_reservelow);
811
812static void __init trim_low_memory_range(void)
813{
814 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
815}
816
817/*
818 * Dump out kernel offset information on panic.
819 */
820static int
821dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
822{
823 if (kaslr_enabled()) {
824 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
825 kaslr_offset(),
826 __START_KERNEL,
827 __START_KERNEL_map,
828 MODULES_VADDR-1);
829 } else {
830 pr_emerg("Kernel Offset: disabled\n");
831 }
832
833 return 0;
834}
835
836/*
837 * Determine if we were loaded by an EFI loader. If so, then we have also been
838 * passed the efi memmap, systab, etc., so we should use these data structures
839 * for initialization. Note, the efi init code path is determined by the
840 * global efi_enabled. This allows the same kernel image to be used on existing
841 * systems (with a traditional BIOS) as well as on EFI systems.
842 */
843/*
844 * setup_arch - architecture-specific boot-time initializations
845 *
846 * Note: On x86_64, fixmaps are ready for use even before this is called.
847 */
848
849void __init setup_arch(char **cmdline_p)
850{
851 memblock_reserve(__pa_symbol(_text),
852 (unsigned long)__bss_stop - (unsigned long)_text);
853
854 early_reserve_initrd();
855
856 /*
857 * At this point everything still needed from the boot loader
858 * or BIOS or kernel text should be early reserved or marked not
859 * RAM in e820. All other memory is free game.
860 */
861
862#ifdef CONFIG_X86_32
863 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
864
865 /*
866 * copy kernel address range established so far and switch
867 * to the proper swapper page table
868 */
869 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
870 initial_page_table + KERNEL_PGD_BOUNDARY,
871 KERNEL_PGD_PTRS);
872
873 load_cr3(swapper_pg_dir);
874 /*
875 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
876 * a cr3 based tlb flush, so the following __flush_tlb_all()
877 * will not flush anything because the cpu quirk which clears
878 * X86_FEATURE_PGE has not been invoked yet. Though due to the
879 * load_cr3() above the TLB has been flushed already. The
880 * quirk is invoked before subsequent calls to __flush_tlb_all()
881 * so proper operation is guaranteed.
882 */
883 __flush_tlb_all();
884#else
885 printk(KERN_INFO "Command line: %s\n", boot_command_line);
886#endif
887
888 /*
889 * If we have OLPC OFW, we might end up relocating the fixmap due to
890 * reserve_top(), so do this before touching the ioremap area.
891 */
892 olpc_ofw_detect();
893
894 early_trap_init();
895 early_cpu_init();
896 early_ioremap_init();
897
898 setup_olpc_ofw_pgd();
899
900 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
901 screen_info = boot_params.screen_info;
902 edid_info = boot_params.edid_info;
903#ifdef CONFIG_X86_32
904 apm_info.bios = boot_params.apm_bios_info;
905 ist_info = boot_params.ist_info;
906#endif
907 saved_video_mode = boot_params.hdr.vid_mode;
908 bootloader_type = boot_params.hdr.type_of_loader;
909 if ((bootloader_type >> 4) == 0xe) {
910 bootloader_type &= 0xf;
911 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
912 }
913 bootloader_version = bootloader_type & 0xf;
914 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
915
916#ifdef CONFIG_BLK_DEV_RAM
917 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
918 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
919 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
920#endif
921#ifdef CONFIG_EFI
922 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
923 EFI32_LOADER_SIGNATURE, 4)) {
924 set_bit(EFI_BOOT, &efi.flags);
925 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
926 EFI64_LOADER_SIGNATURE, 4)) {
927 set_bit(EFI_BOOT, &efi.flags);
928 set_bit(EFI_64BIT, &efi.flags);
929 }
930
931 if (efi_enabled(EFI_BOOT))
932 efi_memblock_x86_reserve_range();
933#endif
934
935 x86_init.oem.arch_setup();
936
937 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
938 setup_memory_map();
939 parse_setup_data();
940
941 copy_edd();
942
943 if (!boot_params.hdr.root_flags)
944 root_mountflags &= ~MS_RDONLY;
945 init_mm.start_code = (unsigned long) _text;
946 init_mm.end_code = (unsigned long) _etext;
947 init_mm.end_data = (unsigned long) _edata;
948 init_mm.brk = _brk_end;
949
950 mpx_mm_init(&init_mm);
951
952 code_resource.start = __pa_symbol(_text);
953 code_resource.end = __pa_symbol(_etext)-1;
954 data_resource.start = __pa_symbol(_etext);
955 data_resource.end = __pa_symbol(_edata)-1;
956 bss_resource.start = __pa_symbol(__bss_start);
957 bss_resource.end = __pa_symbol(__bss_stop)-1;
958
959#ifdef CONFIG_CMDLINE_BOOL
960#ifdef CONFIG_CMDLINE_OVERRIDE
961 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
962#else
963 if (builtin_cmdline[0]) {
964 /* append boot loader cmdline to builtin */
965 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
966 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
967 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
968 }
969#endif
970#endif
971
972 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
973 *cmdline_p = command_line;
974
975 /*
976 * x86_configure_nx() is called before parse_early_param() to detect
977 * whether hardware doesn't support NX (so that the early EHCI debug
978 * console setup can safely call set_fixmap()). It may then be called
979 * again from within noexec_setup() during parsing early parameters
980 * to honor the respective command line option.
981 */
982 x86_configure_nx();
983
984 parse_early_param();
985
986 x86_report_nx();
987
988 /* after early param, so could get panic from serial */
989 memblock_x86_reserve_range_setup_data();
990
991 if (acpi_mps_check()) {
992#ifdef CONFIG_X86_LOCAL_APIC
993 disable_apic = 1;
994#endif
995 setup_clear_cpu_cap(X86_FEATURE_APIC);
996 }
997
998#ifdef CONFIG_PCI
999 if (pci_early_dump_regs)
1000 early_dump_pci_devices();
1001#endif
1002
1003 /* update the e820_saved too */
1004 e820_reserve_setup_data();
1005 finish_e820_parsing();
1006
1007 if (efi_enabled(EFI_BOOT))
1008 efi_init();
1009
1010 dmi_scan_machine();
1011 dmi_memdev_walk();
1012 dmi_set_dump_stack_arch_desc();
1013
1014 /*
1015 * VMware detection requires dmi to be available, so this
1016 * needs to be done after dmi_scan_machine, for the BP.
1017 */
1018 init_hypervisor_platform();
1019
1020 x86_init.resources.probe_roms();
1021
1022 /* after parse_early_param, so could debug it */
1023 insert_resource(&iomem_resource, &code_resource);
1024 insert_resource(&iomem_resource, &data_resource);
1025 insert_resource(&iomem_resource, &bss_resource);
1026
1027 e820_add_kernel_range();
1028 trim_bios_range();
1029#ifdef CONFIG_X86_32
1030 if (ppro_with_ram_bug()) {
1031 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1032 E820_RESERVED);
1033 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1034 printk(KERN_INFO "fixed physical RAM map:\n");
1035 e820_print_map("bad_ppro");
1036 }
1037#else
1038 early_gart_iommu_check();
1039#endif
1040
1041 /*
1042 * partially used pages are not usable - thus
1043 * we are rounding upwards:
1044 */
1045 max_pfn = e820_end_of_ram_pfn();
1046
1047 /* update e820 for memory not covered by WB MTRRs */
1048 mtrr_bp_init();
1049 if (mtrr_trim_uncached_memory(max_pfn))
1050 max_pfn = e820_end_of_ram_pfn();
1051
1052 max_possible_pfn = max_pfn;
1053
1054#ifdef CONFIG_X86_32
1055 /* max_low_pfn get updated here */
1056 find_low_pfn_range();
1057#else
1058 check_x2apic();
1059
1060 /* How many end-of-memory variables you have, grandma! */
1061 /* need this before calling reserve_initrd */
1062 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1063 max_low_pfn = e820_end_of_low_ram_pfn();
1064 else
1065 max_low_pfn = max_pfn;
1066
1067 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1068#endif
1069
1070 /*
1071 * Find and reserve possible boot-time SMP configuration:
1072 */
1073 find_smp_config();
1074
1075 reserve_ibft_region();
1076
1077 early_alloc_pgt_buf();
1078
1079 /*
1080 * Need to conclude brk, before memblock_x86_fill()
1081 * it could use memblock_find_in_range, could overlap with
1082 * brk area.
1083 */
1084 reserve_brk();
1085
1086 cleanup_highmap();
1087
1088 memblock_set_current_limit(ISA_END_ADDRESS);
1089 memblock_x86_fill();
1090
1091 if (efi_enabled(EFI_BOOT)) {
1092 efi_fake_memmap();
1093 efi_find_mirror();
1094 }
1095
1096 /*
1097 * The EFI specification says that boot service code won't be called
1098 * after ExitBootServices(). This is, in fact, a lie.
1099 */
1100 if (efi_enabled(EFI_MEMMAP))
1101 efi_reserve_boot_services();
1102
1103 /* preallocate 4k for mptable mpc */
1104 early_reserve_e820_mpc_new();
1105
1106#ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1107 setup_bios_corruption_check();
1108#endif
1109
1110#ifdef CONFIG_X86_32
1111 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1112 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1113#endif
1114
1115 reserve_real_mode();
1116
1117 trim_platform_memory_ranges();
1118 trim_low_memory_range();
1119
1120 init_mem_mapping();
1121
1122 early_trap_pf_init();
1123
1124 setup_real_mode();
1125
1126 memblock_set_current_limit(get_max_mapped());
1127
1128 /*
1129 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1130 */
1131
1132#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1133 if (init_ohci1394_dma_early)
1134 init_ohci1394_dma_on_all_controllers();
1135#endif
1136 /* Allocate bigger log buffer */
1137 setup_log_buf(1);
1138
1139 reserve_initrd();
1140
1141#if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1142 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1143#endif
1144
1145 vsmp_init();
1146
1147 io_delay_init();
1148
1149 /*
1150 * Parse the ACPI tables for possible boot-time SMP configuration.
1151 */
1152 acpi_boot_table_init();
1153
1154 early_acpi_boot_init();
1155
1156 initmem_init();
1157 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1158
1159 /*
1160 * Reserve memory for crash kernel after SRAT is parsed so that it
1161 * won't consume hotpluggable memory.
1162 */
1163 reserve_crashkernel();
1164
1165 memblock_find_dma_reserve();
1166
1167#ifdef CONFIG_KVM_GUEST
1168 kvmclock_init();
1169#endif
1170
1171 x86_init.paging.pagetable_init();
1172
1173 kasan_init();
1174
1175 if (boot_cpu_data.cpuid_level >= 0) {
1176 /* A CPU has %cr4 if and only if it has CPUID */
1177 mmu_cr4_features = __read_cr4();
1178 if (trampoline_cr4_features)
1179 *trampoline_cr4_features = mmu_cr4_features;
1180 }
1181
1182#ifdef CONFIG_X86_32
1183 /* sync back kernel address range */
1184 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1185 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1186 KERNEL_PGD_PTRS);
1187
1188 /*
1189 * sync back low identity map too. It is used for example
1190 * in the 32-bit EFI stub.
1191 */
1192 clone_pgd_range(initial_page_table,
1193 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1194 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
1195#endif
1196
1197 tboot_probe();
1198
1199 map_vsyscall();
1200
1201 generic_apic_probe();
1202
1203 early_quirks();
1204
1205 /*
1206 * Read APIC and some other early information from ACPI tables.
1207 */
1208 acpi_boot_init();
1209 sfi_init();
1210 x86_dtb_init();
1211
1212 /*
1213 * get boot-time SMP configuration:
1214 */
1215 if (smp_found_config)
1216 get_smp_config();
1217
1218 prefill_possible_map();
1219
1220 init_cpu_to_node();
1221
1222 init_apic_mappings();
1223 io_apic_init_mappings();
1224
1225 kvm_guest_init();
1226
1227 e820_reserve_resources();
1228 e820_mark_nosave_regions(max_low_pfn);
1229
1230 x86_init.resources.reserve_resources();
1231
1232 e820_setup_gap();
1233
1234#ifdef CONFIG_VT
1235#if defined(CONFIG_VGA_CONSOLE)
1236 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1237 conswitchp = &vga_con;
1238#elif defined(CONFIG_DUMMY_CONSOLE)
1239 conswitchp = &dummy_con;
1240#endif
1241#endif
1242 x86_init.oem.banner();
1243
1244 x86_init.timers.wallclock_init();
1245
1246 mcheck_init();
1247
1248 arch_init_ideal_nops();
1249
1250 register_refined_jiffies(CLOCK_TICK_RATE);
1251
1252#ifdef CONFIG_EFI
1253 if (efi_enabled(EFI_BOOT))
1254 efi_apply_memmap_quirks();
1255#endif
1256}
1257
1258#ifdef CONFIG_X86_32
1259
1260static struct resource video_ram_resource = {
1261 .name = "Video RAM area",
1262 .start = 0xa0000,
1263 .end = 0xbffff,
1264 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1265};
1266
1267void __init i386_reserve_resources(void)
1268{
1269 request_resource(&iomem_resource, &video_ram_resource);
1270 reserve_standard_io_resources();
1271}
1272
1273#endif /* CONFIG_X86_32 */
1274
1275static struct notifier_block kernel_offset_notifier = {
1276 .notifier_call = dump_kernel_offset
1277};
1278
1279static int __init register_kernel_offset_dumper(void)
1280{
1281 atomic_notifier_chain_register(&panic_notifier_list,
1282 &kernel_offset_notifier);
1283 return 0;
1284}
1285__initcall(register_kernel_offset_dumper);
1286
1287void arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
1288{
1289 if (!boot_cpu_has(X86_FEATURE_OSPKE))
1290 return;
1291
1292 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
1293}
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);