1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 1995 Linus Torvalds
5 * This file contains the setup_arch() code, which handles the architecture-dependent
6 * parts of early kernel initialization.
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>
26 #include <uapi/linux/mount.h>
32 #include <asm/bios_ebda.h>
37 #include <asm/hypervisor.h>
38 #include <asm/io_apic.h>
39 #include <asm/kasan.h>
40 #include <asm/kaslr.h>
43 #include <asm/realmode.h>
44 #include <asm/olpc_ofw.h>
45 #include <asm/pci-direct.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>
54 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
55 * max_pfn_mapped: highest directly mapped pfn > 4 GB
57 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
58 * represented by pfn_mapped[].
60 unsigned long max_low_pfn_mapped;
61 unsigned long max_pfn_mapped;
64 RESERVE_BRK(dmi_alloc, 65536);
69 * Range of the BSS area. The size of the BSS area is determined
70 * at link time, with RESERVE_BRK() facility reserving additional
73 unsigned long _brk_start = (unsigned long)__brk_base;
74 unsigned long _brk_end = (unsigned long)__brk_base;
76 struct boot_params boot_params;
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
84 static struct resource rodata_resource = {
85 .name = "Kernel rodata",
88 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
91 static struct resource data_resource = {
92 .name = "Kernel data",
95 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
98 static struct resource code_resource = {
99 .name = "Kernel code",
102 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
105 static struct resource bss_resource = {
106 .name = "Kernel bss",
109 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
114 /* CPU data as detected by the assembly code in head_32.S */
115 struct cpuinfo_x86 new_cpu_data;
117 /* Common CPU data for all CPUs */
118 struct cpuinfo_x86 boot_cpu_data __read_mostly;
119 EXPORT_SYMBOL(boot_cpu_data);
121 unsigned int def_to_bigsmp;
123 struct apm_info apm_info;
124 EXPORT_SYMBOL(apm_info);
126 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
127 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
128 struct ist_info ist_info;
129 EXPORT_SYMBOL(ist_info);
131 struct ist_info ist_info;
135 struct cpuinfo_x86 boot_cpu_data __read_mostly;
136 EXPORT_SYMBOL(boot_cpu_data);
140 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
141 __visible unsigned long mmu_cr4_features __ro_after_init;
143 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
146 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
147 int bootloader_type, bootloader_version;
152 struct screen_info screen_info;
153 EXPORT_SYMBOL(screen_info);
154 struct edid_info edid_info;
155 EXPORT_SYMBOL_GPL(edid_info);
157 extern int root_mountflags;
159 unsigned long saved_video_mode;
161 #define RAMDISK_IMAGE_START_MASK 0x07FF
162 #define RAMDISK_PROMPT_FLAG 0x8000
163 #define RAMDISK_LOAD_FLAG 0x4000
165 static char __initdata command_line[COMMAND_LINE_SIZE];
166 #ifdef CONFIG_CMDLINE_BOOL
167 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
170 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
172 #ifdef CONFIG_EDD_MODULE
176 * copy_edd() - Copy the BIOS EDD information
177 * from boot_params into a safe place.
180 static inline void __init copy_edd(void)
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;
189 static inline void __init copy_edd(void)
194 void * __init extend_brk(size_t size, size_t align)
196 size_t mask = align - 1;
199 BUG_ON(_brk_start == 0);
200 BUG_ON(align & mask);
202 _brk_end = (_brk_end + mask) & ~mask;
203 BUG_ON((char *)(_brk_end + size) > __brk_limit);
205 ret = (void *)_brk_end;
208 memset(ret, 0, size);
214 static void __init cleanup_highmap(void)
219 static void __init reserve_brk(void)
221 if (_brk_end > _brk_start)
222 memblock_reserve(__pa_symbol(_brk_start),
223 _brk_end - _brk_start);
225 /* Mark brk area as locked down and no longer taking any
230 u64 relocated_ramdisk;
232 #ifdef CONFIG_BLK_DEV_INITRD
234 static u64 __init get_ramdisk_image(void)
236 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
238 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
240 if (ramdisk_image == 0)
241 ramdisk_image = phys_initrd_start;
243 return ramdisk_image;
245 static u64 __init get_ramdisk_size(void)
247 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
249 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
251 if (ramdisk_size == 0)
252 ramdisk_size = phys_initrd_size;
257 static void __init relocate_initrd(void)
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);
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",
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);
276 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
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);
284 static void __init early_reserve_initrd(void)
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);
291 if (!boot_params.hdr.type_of_loader ||
292 !ramdisk_image || !ramdisk_size)
293 return; /* No initrd provided by bootloader */
295 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
298 static void __init reserve_initrd(void)
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);
305 if (!boot_params.hdr.type_of_loader ||
306 !ramdisk_image || !ramdisk_size)
307 return; /* No initrd provided by bootloader */
311 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
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;
324 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
328 static void __init early_reserve_initrd(void)
331 static void __init reserve_initrd(void)
334 #endif /* CONFIG_BLK_DEV_INITRD */
336 static void __init parse_setup_data(void)
338 struct setup_data *data;
339 u64 pa_data, pa_next;
341 pa_data = boot_params.hdr.setup_data;
343 u32 data_len, data_type;
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));
353 e820__memory_setup_extended(pa_data, data_len);
359 parse_efi_setup(pa_data, data_len);
368 static void __init memblock_x86_reserve_range_setup_data(void)
370 struct setup_data *data;
373 pa_data = boot_params.hdr.setup_data;
375 data = early_memremap(pa_data, sizeof(*data));
376 memblock_reserve(pa_data, sizeof(*data) + data->len);
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);
383 pa_data = data->next;
384 early_memunmap(data, sizeof(*data));
389 * --------- Crashkernel reservation ------------------------------
392 #ifdef CONFIG_KEXEC_CORE
394 /* 16M alignment for crash kernel regions */
395 #define CRASH_ALIGN SZ_16M
398 * Keep the crash kernel below this limit.
400 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
401 * due to mapping restrictions.
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.
411 # define CRASH_ADDR_LOW_MAX SZ_512M
412 # define CRASH_ADDR_HIGH_MAX SZ_512M
414 # define CRASH_ADDR_LOW_MAX SZ_4G
415 # define CRASH_ADDR_HIGH_MAX SZ_64T
418 static int __init reserve_crashkernel_low(void)
421 unsigned long long base, low_base = 0, low_size = 0;
422 unsigned long low_mem_limit;
425 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
427 /* crashkernel=Y,low */
428 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
431 * two parts from kernel/dma/swiotlb.c:
432 * -swiotlb size: user-specified with swiotlb= or default.
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.
439 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
441 /* passed with crashkernel=0,low ? */
446 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
448 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
449 (unsigned long)(low_size >> 20));
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));
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);
465 static void __init reserve_crashkernel(void)
467 unsigned long long crash_size, crash_base, total_mem;
471 total_mem = memblock_phys_mem_size();
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)
484 if (xen_pv_domain()) {
485 pr_info("Ignoring crashkernel for a Xen PV domain\n");
489 /* 0 means: find the address automatically */
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.
500 crash_base = memblock_phys_alloc_range(crash_size,
501 CRASH_ALIGN, CRASH_ALIGN,
504 crash_base = memblock_phys_alloc_range(crash_size,
505 CRASH_ALIGN, CRASH_ALIGN,
506 CRASH_ADDR_HIGH_MAX);
508 pr_info("crashkernel reservation failed - No suitable area found.\n");
512 unsigned long long start;
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");
522 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
523 memblock_free(crash_base, crash_size);
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));
532 crashk_res.start = crash_base;
533 crashk_res.end = crash_base + crash_size - 1;
534 insert_resource(&iomem_resource, &crashk_res);
537 static void __init reserve_crashkernel(void)
542 static 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 }
565 void __init reserve_standard_io_resources(void)
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]);
575 static bool __init snb_gfx_workaround_needed(void)
580 static const __initconst u16 snb_ids[] = {
590 /* Assume no if something weird is going on with PCI */
591 if (!early_pci_allowed())
594 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
595 if (vendor != 0x8086)
598 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
599 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
600 if (devid == snb_ids[i])
608 * Sandy Bridge graphics has trouble with certain ranges, exclude
609 * them from allocation.
611 static void __init trim_snb_memory(void)
613 static const __initconst unsigned long bad_pages[] = {
622 if (!snb_gfx_workaround_needed())
625 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
628 * SandyBridge integrated graphics devices have a bug that prevents
629 * them from accessing certain memory ranges, namely anything below
630 * 1M and in the pages listed in bad_pages[] above.
632 * To avoid these pages being ever accessed by SNB gfx devices reserve
633 * bad_pages that have not already been reserved at boot time.
634 * All memory below the 1 MB mark is anyway reserved later during
635 * setup_arch(), so there is no need to reserve it here.
638 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
639 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
640 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
645 static void __init trim_bios_range(void)
648 * A special case is the first 4Kb of memory;
649 * This is a BIOS owned area, not kernel ram, but generally
650 * not listed as such in the E820 table.
652 * This typically reserves additional memory (64KiB by default)
653 * since some BIOSes are known to corrupt low memory. See the
654 * Kconfig help text for X86_RESERVE_LOW.
656 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
659 * special case: Some BIOSes report the PC BIOS
660 * area (640Kb -> 1Mb) as RAM even though it is not.
663 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
665 e820__update_table(e820_table);
668 /* called before trim_bios_range() to spare extra sanitize */
669 static void __init e820_add_kernel_range(void)
671 u64 start = __pa_symbol(_text);
672 u64 size = __pa_symbol(_end) - start;
675 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
676 * attempt to fix it by adding the range. We may have a confused BIOS,
677 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
678 * exclude kernel range. If we really are running on top non-RAM,
679 * we will crash later anyways.
681 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
684 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
685 e820__range_remove(start, size, E820_TYPE_RAM, 0);
686 e820__range_add(start, size, E820_TYPE_RAM);
689 static void __init early_reserve_memory(void)
692 * Reserve the memory occupied by the kernel between _text and
693 * __end_of_kernel_reserve symbols. Any kernel sections after the
694 * __end_of_kernel_reserve symbol must be explicitly reserved with a
695 * separate memblock_reserve() or they will be discarded.
697 memblock_reserve(__pa_symbol(_text),
698 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
701 * The first 4Kb of memory is a BIOS owned area, but generally it is
702 * not listed as such in the E820 table.
704 * Reserve the first 64K of memory since some BIOSes are known to
705 * corrupt low memory. After the real mode trampoline is allocated the
706 * rest of the memory below 640k is reserved.
708 * In addition, make sure page 0 is always reserved because on
709 * systems with L1TF its contents can be leaked to user processes.
711 memblock_reserve(0, SZ_64K);
713 early_reserve_initrd();
715 if (efi_enabled(EFI_BOOT))
716 efi_memblock_x86_reserve_range();
718 memblock_x86_reserve_range_setup_data();
720 reserve_ibft_region();
721 reserve_bios_regions();
726 * Dump out kernel offset information on panic.
729 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
731 if (kaslr_enabled()) {
732 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
738 pr_emerg("Kernel Offset: disabled\n");
745 * Determine if we were loaded by an EFI loader. If so, then we have also been
746 * passed the efi memmap, systab, etc., so we should use these data structures
747 * for initialization. Note, the efi init code path is determined by the
748 * global efi_enabled. This allows the same kernel image to be used on existing
749 * systems (with a traditional BIOS) as well as on EFI systems.
752 * setup_arch - architecture-specific boot-time initializations
754 * Note: On x86_64, fixmaps are ready for use even before this is called.
757 void __init setup_arch(char **cmdline_p)
760 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
763 * copy kernel address range established so far and switch
764 * to the proper swapper page table
766 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
767 initial_page_table + KERNEL_PGD_BOUNDARY,
770 load_cr3(swapper_pg_dir);
772 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
773 * a cr3 based tlb flush, so the following __flush_tlb_all()
774 * will not flush anything because the CPU quirk which clears
775 * X86_FEATURE_PGE has not been invoked yet. Though due to the
776 * load_cr3() above the TLB has been flushed already. The
777 * quirk is invoked before subsequent calls to __flush_tlb_all()
778 * so proper operation is guaranteed.
782 printk(KERN_INFO "Command line: %s\n", boot_command_line);
783 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
787 * If we have OLPC OFW, we might end up relocating the fixmap due to
788 * reserve_top(), so do this before touching the ioremap area.
792 idt_setup_early_traps();
796 early_ioremap_init();
798 setup_olpc_ofw_pgd();
800 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
801 screen_info = boot_params.screen_info;
802 edid_info = boot_params.edid_info;
804 apm_info.bios = boot_params.apm_bios_info;
805 ist_info = boot_params.ist_info;
807 saved_video_mode = boot_params.hdr.vid_mode;
808 bootloader_type = boot_params.hdr.type_of_loader;
809 if ((bootloader_type >> 4) == 0xe) {
810 bootloader_type &= 0xf;
811 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
813 bootloader_version = bootloader_type & 0xf;
814 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
816 #ifdef CONFIG_BLK_DEV_RAM
817 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
820 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
821 EFI32_LOADER_SIGNATURE, 4)) {
822 set_bit(EFI_BOOT, &efi.flags);
823 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
824 EFI64_LOADER_SIGNATURE, 4)) {
825 set_bit(EFI_BOOT, &efi.flags);
826 set_bit(EFI_64BIT, &efi.flags);
830 x86_init.oem.arch_setup();
832 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
833 e820__memory_setup();
838 if (!boot_params.hdr.root_flags)
839 root_mountflags &= ~MS_RDONLY;
840 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
842 code_resource.start = __pa_symbol(_text);
843 code_resource.end = __pa_symbol(_etext)-1;
844 rodata_resource.start = __pa_symbol(__start_rodata);
845 rodata_resource.end = __pa_symbol(__end_rodata)-1;
846 data_resource.start = __pa_symbol(_sdata);
847 data_resource.end = __pa_symbol(_edata)-1;
848 bss_resource.start = __pa_symbol(__bss_start);
849 bss_resource.end = __pa_symbol(__bss_stop)-1;
851 #ifdef CONFIG_CMDLINE_BOOL
852 #ifdef CONFIG_CMDLINE_OVERRIDE
853 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
855 if (builtin_cmdline[0]) {
856 /* append boot loader cmdline to builtin */
857 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
858 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
859 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
864 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
865 *cmdline_p = command_line;
868 * x86_configure_nx() is called before parse_early_param() to detect
869 * whether hardware doesn't support NX (so that the early EHCI debug
870 * console setup can safely call set_fixmap()). It may then be called
871 * again from within noexec_setup() during parsing early parameters
872 * to honor the respective command line option.
879 * Do some memory reservations *before* memory is added to
880 * memblock, so memblock allocations won't overwrite it.
881 * Do it after early param, so we could get (unlikely) panic from
884 * After this point everything still needed from the boot loader or
885 * firmware or kernel text should be early reserved or marked not
886 * RAM in e820. All other memory is free game.
888 early_reserve_memory();
890 #ifdef CONFIG_MEMORY_HOTPLUG
892 * Memory used by the kernel cannot be hot-removed because Linux
893 * cannot migrate the kernel pages. When memory hotplug is
894 * enabled, we should prevent memblock from allocating memory
897 * ACPI SRAT records all hotpluggable memory ranges. But before
898 * SRAT is parsed, we don't know about it.
900 * The kernel image is loaded into memory at very early time. We
901 * cannot prevent this anyway. So on NUMA system, we set any
902 * node the kernel resides in as un-hotpluggable.
904 * Since on modern servers, one node could have double-digit
905 * gigabytes memory, we can assume the memory around the kernel
906 * image is also un-hotpluggable. So before SRAT is parsed, just
907 * allocate memory near the kernel image to try the best to keep
908 * the kernel away from hotpluggable memory.
910 if (movable_node_is_enabled())
911 memblock_set_bottom_up(true);
916 if (acpi_mps_check()) {
917 #ifdef CONFIG_X86_LOCAL_APIC
920 setup_clear_cpu_cap(X86_FEATURE_APIC);
923 e820__reserve_setup_data();
924 e820__finish_early_params();
926 if (efi_enabled(EFI_BOOT))
932 * VMware detection requires dmi to be available, so this
933 * needs to be done after dmi_setup(), for the boot CPU.
935 init_hypervisor_platform();
938 x86_init.resources.probe_roms();
940 /* after parse_early_param, so could debug it */
941 insert_resource(&iomem_resource, &code_resource);
942 insert_resource(&iomem_resource, &rodata_resource);
943 insert_resource(&iomem_resource, &data_resource);
944 insert_resource(&iomem_resource, &bss_resource);
946 e820_add_kernel_range();
949 if (ppro_with_ram_bug()) {
950 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
952 e820__update_table(e820_table);
953 printk(KERN_INFO "fixed physical RAM map:\n");
954 e820__print_table("bad_ppro");
957 early_gart_iommu_check();
961 * partially used pages are not usable - thus
962 * we are rounding upwards:
964 max_pfn = e820__end_of_ram_pfn();
966 /* update e820 for memory not covered by WB MTRRs */
968 if (mtrr_trim_uncached_memory(max_pfn))
969 max_pfn = e820__end_of_ram_pfn();
971 max_possible_pfn = max_pfn;
974 * This call is required when the CPU does not support PAT. If
975 * mtrr_bp_init() invoked it already via pat_init() the call has no
981 * Define random base addresses for memory sections after max_pfn is
982 * defined and before each memory section base is used.
984 kernel_randomize_memory();
987 /* max_low_pfn get updated here */
988 find_low_pfn_range();
992 /* How many end-of-memory variables you have, grandma! */
993 /* need this before calling reserve_initrd */
994 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
995 max_low_pfn = e820__end_of_low_ram_pfn();
997 max_low_pfn = max_pfn;
999 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1003 * Find and reserve possible boot-time SMP configuration:
1007 early_alloc_pgt_buf();
1010 * Need to conclude brk, before e820__memblock_setup()
1011 * it could use memblock_find_in_range, could overlap with
1018 memblock_set_current_limit(ISA_END_ADDRESS);
1019 e820__memblock_setup();
1022 * Needs to run after memblock setup because it needs the physical
1030 efi_mokvar_table_init();
1033 * The EFI specification says that boot service code won't be
1034 * called after ExitBootServices(). This is, in fact, a lie.
1036 efi_reserve_boot_services();
1038 /* preallocate 4k for mptable mpc */
1039 e820__memblock_alloc_reserved_mpc_new();
1041 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1042 setup_bios_corruption_check();
1045 #ifdef CONFIG_X86_32
1046 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1047 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1051 * Find free memory for the real mode trampoline and place it there. If
1052 * there is not enough free memory under 1M, on EFI-enabled systems
1053 * there will be additional attempt to reclaim the memory for the real
1054 * mode trampoline at efi_free_boot_services().
1056 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1057 * are known to corrupt low memory and several hundred kilobytes are not
1058 * worth complex detection what memory gets clobbered. Windows does the
1059 * same thing for very similar reasons.
1061 * Moreover, on machines with SandyBridge graphics or in setups that use
1062 * crashkernel the entire 1M is reserved anyway.
1064 reserve_real_mode();
1068 idt_setup_early_pf();
1071 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1072 * with the current CR4 value. This may not be necessary, but
1073 * auditing all the early-boot CR4 manipulation would be needed to
1076 * Mask off features that don't work outside long mode (just
1079 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1081 memblock_set_current_limit(get_max_mapped());
1084 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1087 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1088 if (init_ohci1394_dma_early)
1089 init_ohci1394_dma_on_all_controllers();
1091 /* Allocate bigger log buffer */
1094 if (efi_enabled(EFI_BOOT)) {
1095 switch (boot_params.secure_boot) {
1096 case efi_secureboot_mode_disabled:
1097 pr_info("Secure boot disabled\n");
1099 case efi_secureboot_mode_enabled:
1100 pr_info("Secure boot enabled\n");
1103 pr_info("Secure boot could not be determined\n");
1110 acpi_table_upgrade();
1111 /* Look for ACPI tables and reserve memory occupied by them. */
1112 acpi_boot_table_init();
1118 early_platform_quirks();
1120 early_acpi_boot_init();
1123 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1125 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1126 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1129 * Reserve memory for crash kernel after SRAT is parsed so that it
1130 * won't consume hotpluggable memory.
1132 reserve_crashkernel();
1134 memblock_find_dma_reserve();
1136 if (!early_xdbc_setup_hardware())
1137 early_xdbc_register_console();
1139 x86_init.paging.pagetable_init();
1144 * Sync back kernel address range.
1146 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1149 sync_initial_page_table();
1155 generic_apic_probe();
1160 * Read APIC and some other early information from ACPI tables.
1166 * get boot-time SMP configuration:
1171 * Systems w/o ACPI and mptables might not have it mapped the local
1172 * APIC yet, but prefill_possible_map() might need to access it.
1174 init_apic_mappings();
1176 prefill_possible_map();
1181 io_apic_init_mappings();
1183 x86_init.hyper.guest_late_init();
1185 e820__reserve_resources();
1186 e820__register_nosave_regions(max_pfn);
1188 x86_init.resources.reserve_resources();
1190 e820__setup_pci_gap();
1193 #if defined(CONFIG_VGA_CONSOLE)
1194 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1195 conswitchp = &vga_con;
1198 x86_init.oem.banner();
1200 x86_init.timers.wallclock_init();
1203 * This needs to run before setup_local_APIC() which soft-disables the
1204 * local APIC temporarily and that masks the thermal LVT interrupt,
1205 * leading to softlockups on machines which have configured SMI
1206 * interrupt delivery.
1212 register_refined_jiffies(CLOCK_TICK_RATE);
1215 if (efi_enabled(EFI_BOOT))
1216 efi_apply_memmap_quirks();
1222 #ifdef CONFIG_X86_32
1224 static struct resource video_ram_resource = {
1225 .name = "Video RAM area",
1228 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1231 void __init i386_reserve_resources(void)
1233 request_resource(&iomem_resource, &video_ram_resource);
1234 reserve_standard_io_resources();
1237 #endif /* CONFIG_X86_32 */
1239 static struct notifier_block kernel_offset_notifier = {
1240 .notifier_call = dump_kernel_offset
1243 static int __init register_kernel_offset_dumper(void)
1245 atomic_notifier_chain_register(&panic_notifier_list,
1246 &kernel_offset_notifier);
1249 __initcall(register_kernel_offset_dumper);