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/acpi.h>
9 #include <linux/console.h>
10 #include <linux/crash_dump.h>
11 #include <linux/dma-map-ops.h>
12 #include <linux/efi.h>
13 #include <linux/ima.h>
14 #include <linux/init_ohci1394_dma.h>
15 #include <linux/initrd.h>
16 #include <linux/iscsi_ibft.h>
17 #include <linux/memblock.h>
18 #include <linux/panic_notifier.h>
19 #include <linux/pci.h>
20 #include <linux/root_dev.h>
21 #include <linux/hugetlb.h>
22 #include <linux/tboot.h>
23 #include <linux/usb/xhci-dbgp.h>
24 #include <linux/static_call.h>
25 #include <linux/swiotlb.h>
26 #include <linux/random.h>
28 #include <uapi/linux/mount.h>
34 #include <asm/bios_ebda.h>
40 #include <asm/hypervisor.h>
41 #include <asm/io_apic.h>
42 #include <asm/kasan.h>
43 #include <asm/kaslr.h>
45 #include <asm/memtype.h>
47 #include <asm/realmode.h>
48 #include <asm/olpc_ofw.h>
49 #include <asm/pci-direct.h>
51 #include <asm/proto.h>
52 #include <asm/thermal.h>
53 #include <asm/unwind.h>
54 #include <asm/vsyscall.h>
55 #include <linux/vmalloc.h>
58 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
59 * max_pfn_mapped: highest directly mapped pfn > 4 GB
61 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
62 * represented by pfn_mapped[].
64 unsigned long max_low_pfn_mapped;
65 unsigned long max_pfn_mapped;
68 RESERVE_BRK(dmi_alloc, 65536);
72 unsigned long _brk_start = (unsigned long)__brk_base;
73 unsigned long _brk_end = (unsigned long)__brk_base;
75 struct boot_params boot_params;
78 * These are the four main kernel memory regions, we put them into
79 * the resource tree so that kdump tools and other debugging tools
83 static struct resource rodata_resource = {
84 .name = "Kernel rodata",
87 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
90 static struct resource data_resource = {
91 .name = "Kernel data",
94 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
97 static struct resource code_resource = {
98 .name = "Kernel code",
101 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
104 static struct resource bss_resource = {
105 .name = "Kernel bss",
108 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
113 /* CPU data as detected by the assembly code in head_32.S */
114 struct cpuinfo_x86 new_cpu_data;
116 /* Common CPU data for all CPUs */
117 struct cpuinfo_x86 boot_cpu_data __read_mostly;
118 EXPORT_SYMBOL(boot_cpu_data);
120 unsigned int def_to_bigsmp;
122 struct apm_info apm_info;
123 EXPORT_SYMBOL(apm_info);
125 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
126 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
127 struct ist_info ist_info;
128 EXPORT_SYMBOL(ist_info);
130 struct ist_info ist_info;
134 struct cpuinfo_x86 boot_cpu_data __read_mostly;
135 EXPORT_SYMBOL(boot_cpu_data);
139 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
140 __visible unsigned long mmu_cr4_features __ro_after_init;
142 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
146 static phys_addr_t ima_kexec_buffer_phys;
147 static size_t ima_kexec_buffer_size;
150 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
151 int bootloader_type, bootloader_version;
156 struct screen_info screen_info;
157 EXPORT_SYMBOL(screen_info);
158 struct edid_info edid_info;
159 EXPORT_SYMBOL_GPL(edid_info);
161 extern int root_mountflags;
163 unsigned long saved_video_mode;
165 #define RAMDISK_IMAGE_START_MASK 0x07FF
166 #define RAMDISK_PROMPT_FLAG 0x8000
167 #define RAMDISK_LOAD_FLAG 0x4000
169 static char __initdata command_line[COMMAND_LINE_SIZE];
170 #ifdef CONFIG_CMDLINE_BOOL
171 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
174 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
176 #ifdef CONFIG_EDD_MODULE
180 * copy_edd() - Copy the BIOS EDD information
181 * from boot_params into a safe place.
184 static inline void __init copy_edd(void)
186 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
187 sizeof(edd.mbr_signature));
188 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
189 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
190 edd.edd_info_nr = boot_params.eddbuf_entries;
193 static inline void __init copy_edd(void)
198 void * __init extend_brk(size_t size, size_t align)
200 size_t mask = align - 1;
203 BUG_ON(_brk_start == 0);
204 BUG_ON(align & mask);
206 _brk_end = (_brk_end + mask) & ~mask;
207 BUG_ON((char *)(_brk_end + size) > __brk_limit);
209 ret = (void *)_brk_end;
212 memset(ret, 0, size);
218 static void __init cleanup_highmap(void)
223 static void __init reserve_brk(void)
225 if (_brk_end > _brk_start)
226 memblock_reserve(__pa_symbol(_brk_start),
227 _brk_end - _brk_start);
229 /* Mark brk area as locked down and no longer taking any
234 u64 relocated_ramdisk;
236 #ifdef CONFIG_BLK_DEV_INITRD
238 static u64 __init get_ramdisk_image(void)
240 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
242 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
244 if (ramdisk_image == 0)
245 ramdisk_image = phys_initrd_start;
247 return ramdisk_image;
249 static u64 __init get_ramdisk_size(void)
251 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
253 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
255 if (ramdisk_size == 0)
256 ramdisk_size = phys_initrd_size;
261 static void __init relocate_initrd(void)
263 /* Assume only end is not page aligned */
264 u64 ramdisk_image = get_ramdisk_image();
265 u64 ramdisk_size = get_ramdisk_size();
266 u64 area_size = PAGE_ALIGN(ramdisk_size);
268 /* We need to move the initrd down into directly mapped mem */
269 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
270 PFN_PHYS(max_pfn_mapped));
271 if (!relocated_ramdisk)
272 panic("Cannot find place for new RAMDISK of size %lld\n",
275 initrd_start = relocated_ramdisk + PAGE_OFFSET;
276 initrd_end = initrd_start + ramdisk_size;
277 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
278 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
280 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
282 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
283 " [mem %#010llx-%#010llx]\n",
284 ramdisk_image, ramdisk_image + ramdisk_size - 1,
285 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
288 static void __init early_reserve_initrd(void)
290 /* Assume only end is not page aligned */
291 u64 ramdisk_image = get_ramdisk_image();
292 u64 ramdisk_size = get_ramdisk_size();
293 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
295 if (!boot_params.hdr.type_of_loader ||
296 !ramdisk_image || !ramdisk_size)
297 return; /* No initrd provided by bootloader */
299 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
302 static void __init reserve_initrd(void)
304 /* Assume only end is not page aligned */
305 u64 ramdisk_image = get_ramdisk_image();
306 u64 ramdisk_size = get_ramdisk_size();
307 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
309 if (!boot_params.hdr.type_of_loader ||
310 !ramdisk_image || !ramdisk_size)
311 return; /* No initrd provided by bootloader */
315 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
318 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
319 PFN_DOWN(ramdisk_end))) {
320 /* All are mapped, easy case */
321 initrd_start = ramdisk_image + PAGE_OFFSET;
322 initrd_end = initrd_start + ramdisk_size;
328 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image);
332 static void __init early_reserve_initrd(void)
335 static void __init reserve_initrd(void)
338 #endif /* CONFIG_BLK_DEV_INITRD */
340 static void __init add_early_ima_buffer(u64 phys_addr)
343 struct ima_setup_data *data;
345 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data));
347 pr_warn("setup: failed to memremap ima_setup_data entry\n");
352 memblock_reserve(data->addr, data->size);
353 ima_kexec_buffer_phys = data->addr;
354 ima_kexec_buffer_size = data->size;
357 early_memunmap(data, sizeof(*data));
359 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n");
363 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE)
364 int __init ima_free_kexec_buffer(void)
366 if (!ima_kexec_buffer_size)
369 memblock_free_late(ima_kexec_buffer_phys,
370 ima_kexec_buffer_size);
372 ima_kexec_buffer_phys = 0;
373 ima_kexec_buffer_size = 0;
378 int __init ima_get_kexec_buffer(void **addr, size_t *size)
380 if (!ima_kexec_buffer_size)
383 *addr = __va(ima_kexec_buffer_phys);
384 *size = ima_kexec_buffer_size;
390 static void __init parse_setup_data(void)
392 struct setup_data *data;
393 u64 pa_data, pa_next;
395 pa_data = boot_params.hdr.setup_data;
397 u32 data_len, data_type;
399 data = early_memremap(pa_data, sizeof(*data));
400 data_len = data->len + sizeof(struct setup_data);
401 data_type = data->type;
402 pa_next = data->next;
403 early_memunmap(data, sizeof(*data));
407 e820__memory_setup_extended(pa_data, data_len);
413 parse_efi_setup(pa_data, data_len);
416 add_early_ima_buffer(pa_data);
419 data = early_memremap(pa_data, data_len);
420 add_bootloader_randomness(data->data, data->len);
421 /* Zero seed for forward secrecy. */
422 memzero_explicit(data->data, data->len);
423 /* Zero length in case we find ourselves back here by accident. */
424 memzero_explicit(&data->len, sizeof(data->len));
425 early_memunmap(data, data_len);
434 static void __init memblock_x86_reserve_range_setup_data(void)
436 struct setup_indirect *indirect;
437 struct setup_data *data;
438 u64 pa_data, pa_next;
441 pa_data = boot_params.hdr.setup_data;
443 data = early_memremap(pa_data, sizeof(*data));
445 pr_warn("setup: failed to memremap setup_data entry\n");
450 pa_next = data->next;
452 memblock_reserve(pa_data, sizeof(*data) + data->len);
454 if (data->type == SETUP_INDIRECT) {
456 early_memunmap(data, sizeof(*data));
457 data = early_memremap(pa_data, len);
459 pr_warn("setup: failed to memremap indirect setup_data\n");
463 indirect = (struct setup_indirect *)data->data;
465 if (indirect->type != SETUP_INDIRECT)
466 memblock_reserve(indirect->addr, indirect->len);
470 early_memunmap(data, len);
475 * --------- Crashkernel reservation ------------------------------
478 /* 16M alignment for crash kernel regions */
479 #define CRASH_ALIGN SZ_16M
482 * Keep the crash kernel below this limit.
484 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
485 * due to mapping restrictions.
487 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
488 * the upper limit of system RAM in 4-level paging mode. Since the kdump
489 * jump could be from 5-level paging to 4-level paging, the jump will fail if
490 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
491 * no good way to detect the paging mode of the target kernel which will be
492 * loaded for dumping.
495 # define CRASH_ADDR_LOW_MAX SZ_512M
496 # define CRASH_ADDR_HIGH_MAX SZ_512M
498 # define CRASH_ADDR_LOW_MAX SZ_4G
499 # define CRASH_ADDR_HIGH_MAX SZ_64T
502 static int __init reserve_crashkernel_low(void)
505 unsigned long long base, low_base = 0, low_size = 0;
506 unsigned long low_mem_limit;
509 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
511 /* crashkernel=Y,low */
512 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
515 * two parts from kernel/dma/swiotlb.c:
516 * -swiotlb size: user-specified with swiotlb= or default.
518 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
519 * to 8M for other buffers that may need to stay low too. Also
520 * make sure we allocate enough extra low memory so that we
521 * don't run out of DMA buffers for 32-bit devices.
523 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
525 /* passed with crashkernel=0,low ? */
530 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
532 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
533 (unsigned long)(low_size >> 20));
537 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
538 (unsigned long)(low_size >> 20),
539 (unsigned long)(low_base >> 20),
540 (unsigned long)(low_mem_limit >> 20));
542 crashk_low_res.start = low_base;
543 crashk_low_res.end = low_base + low_size - 1;
544 insert_resource(&iomem_resource, &crashk_low_res);
549 static void __init reserve_crashkernel(void)
551 unsigned long long crash_size, crash_base, total_mem;
555 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
558 total_mem = memblock_phys_mem_size();
561 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
562 if (ret != 0 || crash_size <= 0) {
563 /* crashkernel=X,high */
564 ret = parse_crashkernel_high(boot_command_line, total_mem,
565 &crash_size, &crash_base);
566 if (ret != 0 || crash_size <= 0)
571 if (xen_pv_domain()) {
572 pr_info("Ignoring crashkernel for a Xen PV domain\n");
576 /* 0 means: find the address automatically */
579 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
580 * crashkernel=x,high reserves memory over 4G, also allocates
581 * 256M extra low memory for DMA buffers and swiotlb.
582 * But the extra memory is not required for all machines.
583 * So try low memory first and fall back to high memory
584 * unless "crashkernel=size[KMG],high" is specified.
587 crash_base = memblock_phys_alloc_range(crash_size,
588 CRASH_ALIGN, CRASH_ALIGN,
591 crash_base = memblock_phys_alloc_range(crash_size,
592 CRASH_ALIGN, CRASH_ALIGN,
593 CRASH_ADDR_HIGH_MAX);
595 pr_info("crashkernel reservation failed - No suitable area found.\n");
599 unsigned long long start;
601 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
602 crash_base + crash_size);
603 if (start != crash_base) {
604 pr_info("crashkernel reservation failed - memory is in use.\n");
609 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
610 memblock_phys_free(crash_base, crash_size);
614 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
615 (unsigned long)(crash_size >> 20),
616 (unsigned long)(crash_base >> 20),
617 (unsigned long)(total_mem >> 20));
619 crashk_res.start = crash_base;
620 crashk_res.end = crash_base + crash_size - 1;
621 insert_resource(&iomem_resource, &crashk_res);
624 static 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 }
647 void __init reserve_standard_io_resources(void)
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]);
657 static bool __init snb_gfx_workaround_needed(void)
662 static const __initconst u16 snb_ids[] = {
672 /* Assume no if something weird is going on with PCI */
673 if (!early_pci_allowed())
676 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
677 if (vendor != 0x8086)
680 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
681 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
682 if (devid == snb_ids[i])
690 * Sandy Bridge graphics has trouble with certain ranges, exclude
691 * them from allocation.
693 static void __init trim_snb_memory(void)
695 static const __initconst unsigned long bad_pages[] = {
704 if (!snb_gfx_workaround_needed())
707 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
710 * SandyBridge integrated graphics devices have a bug that prevents
711 * them from accessing certain memory ranges, namely anything below
712 * 1M and in the pages listed in bad_pages[] above.
714 * To avoid these pages being ever accessed by SNB gfx devices reserve
715 * bad_pages that have not already been reserved at boot time.
716 * All memory below the 1 MB mark is anyway reserved later during
717 * setup_arch(), so there is no need to reserve it here.
720 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
721 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
722 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
727 static void __init trim_bios_range(void)
730 * A special case is the first 4Kb of memory;
731 * This is a BIOS owned area, not kernel ram, but generally
732 * not listed as such in the E820 table.
734 * This typically reserves additional memory (64KiB by default)
735 * since some BIOSes are known to corrupt low memory. See the
736 * Kconfig help text for X86_RESERVE_LOW.
738 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
741 * special case: Some BIOSes report the PC BIOS
742 * area (640Kb -> 1Mb) as RAM even though it is not.
745 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
747 e820__update_table(e820_table);
750 /* called before trim_bios_range() to spare extra sanitize */
751 static void __init e820_add_kernel_range(void)
753 u64 start = __pa_symbol(_text);
754 u64 size = __pa_symbol(_end) - start;
757 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
758 * attempt to fix it by adding the range. We may have a confused BIOS,
759 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
760 * exclude kernel range. If we really are running on top non-RAM,
761 * we will crash later anyways.
763 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
766 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
767 e820__range_remove(start, size, E820_TYPE_RAM, 0);
768 e820__range_add(start, size, E820_TYPE_RAM);
771 static void __init early_reserve_memory(void)
774 * Reserve the memory occupied by the kernel between _text and
775 * __end_of_kernel_reserve symbols. Any kernel sections after the
776 * __end_of_kernel_reserve symbol must be explicitly reserved with a
777 * separate memblock_reserve() or they will be discarded.
779 memblock_reserve(__pa_symbol(_text),
780 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
783 * The first 4Kb of memory is a BIOS owned area, but generally it is
784 * not listed as such in the E820 table.
786 * Reserve the first 64K of memory since some BIOSes are known to
787 * corrupt low memory. After the real mode trampoline is allocated the
788 * rest of the memory below 640k is reserved.
790 * In addition, make sure page 0 is always reserved because on
791 * systems with L1TF its contents can be leaked to user processes.
793 memblock_reserve(0, SZ_64K);
795 early_reserve_initrd();
797 memblock_x86_reserve_range_setup_data();
799 reserve_ibft_region();
800 reserve_bios_regions();
805 * Dump out kernel offset information on panic.
808 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
810 if (kaslr_enabled()) {
811 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
817 pr_emerg("Kernel Offset: disabled\n");
823 void x86_configure_nx(void)
825 if (boot_cpu_has(X86_FEATURE_NX))
826 __supported_pte_mask |= _PAGE_NX;
828 __supported_pte_mask &= ~_PAGE_NX;
831 static void __init x86_report_nx(void)
833 if (!boot_cpu_has(X86_FEATURE_NX)) {
834 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
835 "missing in CPU!\n");
837 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
838 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
840 /* 32bit non-PAE kernel, NX cannot be used */
841 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection "
842 "cannot be enabled: non-PAE kernel!\n");
848 * Determine if we were loaded by an EFI loader. If so, then we have also been
849 * passed the efi memmap, systab, etc., so we should use these data structures
850 * for initialization. Note, the efi init code path is determined by the
851 * global efi_enabled. This allows the same kernel image to be used on existing
852 * systems (with a traditional BIOS) as well as on EFI systems.
855 * setup_arch - architecture-specific boot-time initializations
857 * Note: On x86_64, fixmaps are ready for use even before this is called.
860 void __init setup_arch(char **cmdline_p)
863 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
866 * copy kernel address range established so far and switch
867 * to the proper swapper page table
869 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
870 initial_page_table + KERNEL_PGD_BOUNDARY,
873 load_cr3(swapper_pg_dir);
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.
885 printk(KERN_INFO "Command line: %s\n", boot_command_line);
886 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
890 * If we have OLPC OFW, we might end up relocating the fixmap due to
891 * reserve_top(), so do this before touching the ioremap area.
895 idt_setup_early_traps();
899 early_ioremap_init();
901 setup_olpc_ofw_pgd();
903 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
904 screen_info = boot_params.screen_info;
905 edid_info = boot_params.edid_info;
907 apm_info.bios = boot_params.apm_bios_info;
908 ist_info = boot_params.ist_info;
910 saved_video_mode = boot_params.hdr.vid_mode;
911 bootloader_type = boot_params.hdr.type_of_loader;
912 if ((bootloader_type >> 4) == 0xe) {
913 bootloader_type &= 0xf;
914 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
916 bootloader_version = bootloader_type & 0xf;
917 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
919 #ifdef CONFIG_BLK_DEV_RAM
920 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
923 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
924 EFI32_LOADER_SIGNATURE, 4)) {
925 set_bit(EFI_BOOT, &efi.flags);
926 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
927 EFI64_LOADER_SIGNATURE, 4)) {
928 set_bit(EFI_BOOT, &efi.flags);
929 set_bit(EFI_64BIT, &efi.flags);
933 x86_init.oem.arch_setup();
936 * Do some memory reservations *before* memory is added to memblock, so
937 * memblock allocations won't overwrite it.
939 * After this point, everything still needed from the boot loader or
940 * firmware or kernel text should be early reserved or marked not RAM in
941 * e820. All other memory is free game.
943 * This call needs to happen before e820__memory_setup() which calls the
944 * xen_memory_setup() on Xen dom0 which relies on the fact that those
945 * early reservations have happened already.
947 early_reserve_memory();
949 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
950 e820__memory_setup();
955 if (!boot_params.hdr.root_flags)
956 root_mountflags &= ~MS_RDONLY;
957 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
959 code_resource.start = __pa_symbol(_text);
960 code_resource.end = __pa_symbol(_etext)-1;
961 rodata_resource.start = __pa_symbol(__start_rodata);
962 rodata_resource.end = __pa_symbol(__end_rodata)-1;
963 data_resource.start = __pa_symbol(_sdata);
964 data_resource.end = __pa_symbol(_edata)-1;
965 bss_resource.start = __pa_symbol(__bss_start);
966 bss_resource.end = __pa_symbol(__bss_stop)-1;
968 #ifdef CONFIG_CMDLINE_BOOL
969 #ifdef CONFIG_CMDLINE_OVERRIDE
970 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
972 if (builtin_cmdline[0]) {
973 /* append boot loader cmdline to builtin */
974 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
975 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
976 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
981 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
982 *cmdline_p = command_line;
985 * x86_configure_nx() is called before parse_early_param() to detect
986 * whether hardware doesn't support NX (so that the early EHCI debug
987 * console setup can safely call set_fixmap()).
993 if (efi_enabled(EFI_BOOT))
994 efi_memblock_x86_reserve_range();
996 #ifdef CONFIG_MEMORY_HOTPLUG
998 * Memory used by the kernel cannot be hot-removed because Linux
999 * cannot migrate the kernel pages. When memory hotplug is
1000 * enabled, we should prevent memblock from allocating memory
1003 * ACPI SRAT records all hotpluggable memory ranges. But before
1004 * SRAT is parsed, we don't know about it.
1006 * The kernel image is loaded into memory at very early time. We
1007 * cannot prevent this anyway. So on NUMA system, we set any
1008 * node the kernel resides in as un-hotpluggable.
1010 * Since on modern servers, one node could have double-digit
1011 * gigabytes memory, we can assume the memory around the kernel
1012 * image is also un-hotpluggable. So before SRAT is parsed, just
1013 * allocate memory near the kernel image to try the best to keep
1014 * the kernel away from hotpluggable memory.
1016 if (movable_node_is_enabled())
1017 memblock_set_bottom_up(true);
1022 if (acpi_mps_check()) {
1023 #ifdef CONFIG_X86_LOCAL_APIC
1026 setup_clear_cpu_cap(X86_FEATURE_APIC);
1029 e820__reserve_setup_data();
1030 e820__finish_early_params();
1032 if (efi_enabled(EFI_BOOT))
1035 x86_init.resources.dmi_setup();
1038 * VMware detection requires dmi to be available, so this
1039 * needs to be done after dmi_setup(), for the boot CPU.
1041 init_hypervisor_platform();
1044 x86_init.resources.probe_roms();
1046 /* after parse_early_param, so could debug it */
1047 insert_resource(&iomem_resource, &code_resource);
1048 insert_resource(&iomem_resource, &rodata_resource);
1049 insert_resource(&iomem_resource, &data_resource);
1050 insert_resource(&iomem_resource, &bss_resource);
1052 e820_add_kernel_range();
1054 #ifdef CONFIG_X86_32
1055 if (ppro_with_ram_bug()) {
1056 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1057 E820_TYPE_RESERVED);
1058 e820__update_table(e820_table);
1059 printk(KERN_INFO "fixed physical RAM map:\n");
1060 e820__print_table("bad_ppro");
1063 early_gart_iommu_check();
1067 * partially used pages are not usable - thus
1068 * we are rounding upwards:
1070 max_pfn = e820__end_of_ram_pfn();
1072 /* update e820 for memory not covered by WB MTRRs */
1073 if (IS_ENABLED(CONFIG_MTRR))
1076 pat_disable("PAT support disabled because CONFIG_MTRR is disabled in the kernel.");
1078 if (mtrr_trim_uncached_memory(max_pfn))
1079 max_pfn = e820__end_of_ram_pfn();
1081 max_possible_pfn = max_pfn;
1084 * This call is required when the CPU does not support PAT. If
1085 * mtrr_bp_init() invoked it already via pat_init() the call has no
1091 * Define random base addresses for memory sections after max_pfn is
1092 * defined and before each memory section base is used.
1094 kernel_randomize_memory();
1096 #ifdef CONFIG_X86_32
1097 /* max_low_pfn get updated here */
1098 find_low_pfn_range();
1102 /* How many end-of-memory variables you have, grandma! */
1103 /* need this before calling reserve_initrd */
1104 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1105 max_low_pfn = e820__end_of_low_ram_pfn();
1107 max_low_pfn = max_pfn;
1109 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1113 * Find and reserve possible boot-time SMP configuration:
1117 early_alloc_pgt_buf();
1120 * Need to conclude brk, before e820__memblock_setup()
1121 * it could use memblock_find_in_range, could overlap with
1128 memblock_set_current_limit(ISA_END_ADDRESS);
1129 e820__memblock_setup();
1132 * Needs to run after memblock setup because it needs the physical
1141 efi_mokvar_table_init();
1144 * The EFI specification says that boot service code won't be
1145 * called after ExitBootServices(). This is, in fact, a lie.
1147 efi_reserve_boot_services();
1149 /* preallocate 4k for mptable mpc */
1150 e820__memblock_alloc_reserved_mpc_new();
1152 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1153 setup_bios_corruption_check();
1156 #ifdef CONFIG_X86_32
1157 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1158 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1162 * Find free memory for the real mode trampoline and place it there. If
1163 * there is not enough free memory under 1M, on EFI-enabled systems
1164 * there will be additional attempt to reclaim the memory for the real
1165 * mode trampoline at efi_free_boot_services().
1167 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1168 * are known to corrupt low memory and several hundred kilobytes are not
1169 * worth complex detection what memory gets clobbered. Windows does the
1170 * same thing for very similar reasons.
1172 * Moreover, on machines with SandyBridge graphics or in setups that use
1173 * crashkernel the entire 1M is reserved anyway.
1175 x86_platform.realmode_reserve();
1179 idt_setup_early_pf();
1182 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1183 * with the current CR4 value. This may not be necessary, but
1184 * auditing all the early-boot CR4 manipulation would be needed to
1187 * Mask off features that don't work outside long mode (just
1190 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1192 memblock_set_current_limit(get_max_mapped());
1195 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1198 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1199 if (init_ohci1394_dma_early)
1200 init_ohci1394_dma_on_all_controllers();
1202 /* Allocate bigger log buffer */
1205 if (efi_enabled(EFI_BOOT)) {
1206 switch (boot_params.secure_boot) {
1207 case efi_secureboot_mode_disabled:
1208 pr_info("Secure boot disabled\n");
1210 case efi_secureboot_mode_enabled:
1211 pr_info("Secure boot enabled\n");
1214 pr_info("Secure boot could not be determined\n");
1221 acpi_table_upgrade();
1222 /* Look for ACPI tables and reserve memory occupied by them. */
1223 acpi_boot_table_init();
1229 early_platform_quirks();
1231 early_acpi_boot_init();
1234 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1236 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1237 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1240 * Reserve memory for crash kernel after SRAT is parsed so that it
1241 * won't consume hotpluggable memory.
1243 reserve_crashkernel();
1245 memblock_find_dma_reserve();
1247 if (!early_xdbc_setup_hardware())
1248 early_xdbc_register_console();
1250 x86_init.paging.pagetable_init();
1255 * Sync back kernel address range.
1257 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1260 sync_initial_page_table();
1266 generic_apic_probe();
1271 * Read APIC and some other early information from ACPI tables.
1277 * get boot-time SMP configuration:
1282 * Systems w/o ACPI and mptables might not have it mapped the local
1283 * APIC yet, but prefill_possible_map() might need to access it.
1285 init_apic_mappings();
1287 prefill_possible_map();
1292 io_apic_init_mappings();
1294 x86_init.hyper.guest_late_init();
1296 e820__reserve_resources();
1297 e820__register_nosave_regions(max_pfn);
1299 x86_init.resources.reserve_resources();
1301 e820__setup_pci_gap();
1304 #if defined(CONFIG_VGA_CONSOLE)
1305 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1306 conswitchp = &vga_con;
1309 x86_init.oem.banner();
1311 x86_init.timers.wallclock_init();
1314 * This needs to run before setup_local_APIC() which soft-disables the
1315 * local APIC temporarily and that masks the thermal LVT interrupt,
1316 * leading to softlockups on machines which have configured SMI
1317 * interrupt delivery.
1323 register_refined_jiffies(CLOCK_TICK_RATE);
1326 if (efi_enabled(EFI_BOOT))
1327 efi_apply_memmap_quirks();
1333 #ifdef CONFIG_X86_32
1335 static struct resource video_ram_resource = {
1336 .name = "Video RAM area",
1339 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1342 void __init i386_reserve_resources(void)
1344 request_resource(&iomem_resource, &video_ram_resource);
1345 reserve_standard_io_resources();
1348 #endif /* CONFIG_X86_32 */
1350 static struct notifier_block kernel_offset_notifier = {
1351 .notifier_call = dump_kernel_offset
1354 static int __init register_kernel_offset_dumper(void)
1356 atomic_notifier_chain_register(&panic_notifier_list,
1357 &kernel_offset_notifier);
1360 __initcall(register_kernel_offset_dumper);