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/dmi.h>
13 #include <linux/efi.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>
27 #include <uapi/linux/mount.h>
33 #include <asm/bios_ebda.h>
38 #include <asm/hypervisor.h>
39 #include <asm/io_apic.h>
40 #include <asm/kasan.h>
41 #include <asm/kaslr.h>
44 #include <asm/realmode.h>
45 #include <asm/olpc_ofw.h>
46 #include <asm/pci-direct.h>
48 #include <asm/proto.h>
49 #include <asm/thermal.h>
50 #include <asm/unwind.h>
51 #include <asm/vsyscall.h>
52 #include <linux/vmalloc.h>
55 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
56 * max_pfn_mapped: highest directly mapped pfn > 4 GB
58 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
59 * represented by pfn_mapped[].
61 unsigned long max_low_pfn_mapped;
62 unsigned long max_pfn_mapped;
65 RESERVE_BRK(dmi_alloc, 65536);
70 * Range of the BSS area. The size of the BSS area is determined
71 * at link time, with RESERVE_BRK() facility reserving additional
74 unsigned long _brk_start = (unsigned long)__brk_base;
75 unsigned long _brk_end = (unsigned long)__brk_base;
77 struct boot_params boot_params;
80 * These are the four main kernel memory regions, we put them into
81 * the resource tree so that kdump tools and other debugging tools
85 static struct resource rodata_resource = {
86 .name = "Kernel rodata",
89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
92 static struct resource data_resource = {
93 .name = "Kernel data",
96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
99 static struct resource code_resource = {
100 .name = "Kernel code",
103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
106 static struct resource bss_resource = {
107 .name = "Kernel bss",
110 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
115 /* CPU data as detected by the assembly code in head_32.S */
116 struct cpuinfo_x86 new_cpu_data;
118 /* Common CPU data for all CPUs */
119 struct cpuinfo_x86 boot_cpu_data __read_mostly;
120 EXPORT_SYMBOL(boot_cpu_data);
122 unsigned int def_to_bigsmp;
124 struct apm_info apm_info;
125 EXPORT_SYMBOL(apm_info);
127 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
128 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
129 struct ist_info ist_info;
130 EXPORT_SYMBOL(ist_info);
132 struct ist_info ist_info;
136 struct cpuinfo_x86 boot_cpu_data __read_mostly;
137 EXPORT_SYMBOL(boot_cpu_data);
141 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
142 __visible unsigned long mmu_cr4_features __ro_after_init;
144 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
147 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
148 int bootloader_type, bootloader_version;
153 struct screen_info screen_info;
154 EXPORT_SYMBOL(screen_info);
155 struct edid_info edid_info;
156 EXPORT_SYMBOL_GPL(edid_info);
158 extern int root_mountflags;
160 unsigned long saved_video_mode;
162 #define RAMDISK_IMAGE_START_MASK 0x07FF
163 #define RAMDISK_PROMPT_FLAG 0x8000
164 #define RAMDISK_LOAD_FLAG 0x4000
166 static char __initdata command_line[COMMAND_LINE_SIZE];
167 #ifdef CONFIG_CMDLINE_BOOL
168 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
171 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
173 #ifdef CONFIG_EDD_MODULE
177 * copy_edd() - Copy the BIOS EDD information
178 * from boot_params into a safe place.
181 static inline void __init copy_edd(void)
183 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
184 sizeof(edd.mbr_signature));
185 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
186 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
187 edd.edd_info_nr = boot_params.eddbuf_entries;
190 static inline void __init copy_edd(void)
195 void * __init extend_brk(size_t size, size_t align)
197 size_t mask = align - 1;
200 BUG_ON(_brk_start == 0);
201 BUG_ON(align & mask);
203 _brk_end = (_brk_end + mask) & ~mask;
204 BUG_ON((char *)(_brk_end + size) > __brk_limit);
206 ret = (void *)_brk_end;
209 memset(ret, 0, size);
215 static void __init cleanup_highmap(void)
220 static void __init reserve_brk(void)
222 if (_brk_end > _brk_start)
223 memblock_reserve(__pa_symbol(_brk_start),
224 _brk_end - _brk_start);
226 /* Mark brk area as locked down and no longer taking any
231 u64 relocated_ramdisk;
233 #ifdef CONFIG_BLK_DEV_INITRD
235 static u64 __init get_ramdisk_image(void)
237 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
239 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
241 if (ramdisk_image == 0)
242 ramdisk_image = phys_initrd_start;
244 return ramdisk_image;
246 static u64 __init get_ramdisk_size(void)
248 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
250 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
252 if (ramdisk_size == 0)
253 ramdisk_size = phys_initrd_size;
258 static void __init relocate_initrd(void)
260 /* Assume only end is not page aligned */
261 u64 ramdisk_image = get_ramdisk_image();
262 u64 ramdisk_size = get_ramdisk_size();
263 u64 area_size = PAGE_ALIGN(ramdisk_size);
265 /* We need to move the initrd down into directly mapped mem */
266 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
267 PFN_PHYS(max_pfn_mapped));
268 if (!relocated_ramdisk)
269 panic("Cannot find place for new RAMDISK of size %lld\n",
272 initrd_start = relocated_ramdisk + PAGE_OFFSET;
273 initrd_end = initrd_start + ramdisk_size;
274 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
275 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
277 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
279 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
280 " [mem %#010llx-%#010llx]\n",
281 ramdisk_image, ramdisk_image + ramdisk_size - 1,
282 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
285 static void __init early_reserve_initrd(void)
287 /* Assume only end is not page aligned */
288 u64 ramdisk_image = get_ramdisk_image();
289 u64 ramdisk_size = get_ramdisk_size();
290 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
292 if (!boot_params.hdr.type_of_loader ||
293 !ramdisk_image || !ramdisk_size)
294 return; /* No initrd provided by bootloader */
296 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
299 static void __init reserve_initrd(void)
301 /* Assume only end is not page aligned */
302 u64 ramdisk_image = get_ramdisk_image();
303 u64 ramdisk_size = get_ramdisk_size();
304 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
306 if (!boot_params.hdr.type_of_loader ||
307 !ramdisk_image || !ramdisk_size)
308 return; /* No initrd provided by bootloader */
312 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
315 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
316 PFN_DOWN(ramdisk_end))) {
317 /* All are mapped, easy case */
318 initrd_start = ramdisk_image + PAGE_OFFSET;
319 initrd_end = initrd_start + ramdisk_size;
325 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
329 static void __init early_reserve_initrd(void)
332 static void __init reserve_initrd(void)
335 #endif /* CONFIG_BLK_DEV_INITRD */
337 static void __init parse_setup_data(void)
339 struct setup_data *data;
340 u64 pa_data, pa_next;
342 pa_data = boot_params.hdr.setup_data;
344 u32 data_len, data_type;
346 data = early_memremap(pa_data, sizeof(*data));
347 data_len = data->len + sizeof(struct setup_data);
348 data_type = data->type;
349 pa_next = data->next;
350 early_memunmap(data, sizeof(*data));
354 e820__memory_setup_extended(pa_data, data_len);
360 parse_efi_setup(pa_data, data_len);
369 static void __init memblock_x86_reserve_range_setup_data(void)
371 struct setup_indirect *indirect;
372 struct setup_data *data;
373 u64 pa_data, pa_next;
376 pa_data = boot_params.hdr.setup_data;
378 data = early_memremap(pa_data, sizeof(*data));
380 pr_warn("setup: failed to memremap setup_data entry\n");
385 pa_next = data->next;
387 memblock_reserve(pa_data, sizeof(*data) + data->len);
389 if (data->type == SETUP_INDIRECT) {
391 early_memunmap(data, sizeof(*data));
392 data = early_memremap(pa_data, len);
394 pr_warn("setup: failed to memremap indirect setup_data\n");
398 indirect = (struct setup_indirect *)data->data;
400 if (indirect->type != SETUP_INDIRECT)
401 memblock_reserve(indirect->addr, indirect->len);
405 early_memunmap(data, len);
410 * --------- Crashkernel reservation ------------------------------
413 #ifdef CONFIG_KEXEC_CORE
415 /* 16M alignment for crash kernel regions */
416 #define CRASH_ALIGN SZ_16M
419 * Keep the crash kernel below this limit.
421 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
422 * due to mapping restrictions.
424 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
425 * the upper limit of system RAM in 4-level paging mode. Since the kdump
426 * jump could be from 5-level paging to 4-level paging, the jump will fail if
427 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
428 * no good way to detect the paging mode of the target kernel which will be
429 * loaded for dumping.
432 # define CRASH_ADDR_LOW_MAX SZ_512M
433 # define CRASH_ADDR_HIGH_MAX SZ_512M
435 # define CRASH_ADDR_LOW_MAX SZ_4G
436 # define CRASH_ADDR_HIGH_MAX SZ_64T
439 static int __init reserve_crashkernel_low(void)
442 unsigned long long base, low_base = 0, low_size = 0;
443 unsigned long low_mem_limit;
446 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
448 /* crashkernel=Y,low */
449 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
452 * two parts from kernel/dma/swiotlb.c:
453 * -swiotlb size: user-specified with swiotlb= or default.
455 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
456 * to 8M for other buffers that may need to stay low too. Also
457 * make sure we allocate enough extra low memory so that we
458 * don't run out of DMA buffers for 32-bit devices.
460 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
462 /* passed with crashkernel=0,low ? */
467 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
469 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
470 (unsigned long)(low_size >> 20));
474 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
475 (unsigned long)(low_size >> 20),
476 (unsigned long)(low_base >> 20),
477 (unsigned long)(low_mem_limit >> 20));
479 crashk_low_res.start = low_base;
480 crashk_low_res.end = low_base + low_size - 1;
481 insert_resource(&iomem_resource, &crashk_low_res);
486 static void __init reserve_crashkernel(void)
488 unsigned long long crash_size, crash_base, total_mem;
492 total_mem = memblock_phys_mem_size();
495 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
496 if (ret != 0 || crash_size <= 0) {
497 /* crashkernel=X,high */
498 ret = parse_crashkernel_high(boot_command_line, total_mem,
499 &crash_size, &crash_base);
500 if (ret != 0 || crash_size <= 0)
505 if (xen_pv_domain()) {
506 pr_info("Ignoring crashkernel for a Xen PV domain\n");
510 /* 0 means: find the address automatically */
513 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
514 * crashkernel=x,high reserves memory over 4G, also allocates
515 * 256M extra low memory for DMA buffers and swiotlb.
516 * But the extra memory is not required for all machines.
517 * So try low memory first and fall back to high memory
518 * unless "crashkernel=size[KMG],high" is specified.
521 crash_base = memblock_phys_alloc_range(crash_size,
522 CRASH_ALIGN, CRASH_ALIGN,
525 crash_base = memblock_phys_alloc_range(crash_size,
526 CRASH_ALIGN, CRASH_ALIGN,
527 CRASH_ADDR_HIGH_MAX);
529 pr_info("crashkernel reservation failed - No suitable area found.\n");
533 unsigned long long start;
535 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
536 crash_base + crash_size);
537 if (start != crash_base) {
538 pr_info("crashkernel reservation failed - memory is in use.\n");
543 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
544 memblock_free(crash_base, crash_size);
548 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
549 (unsigned long)(crash_size >> 20),
550 (unsigned long)(crash_base >> 20),
551 (unsigned long)(total_mem >> 20));
553 crashk_res.start = crash_base;
554 crashk_res.end = crash_base + crash_size - 1;
555 insert_resource(&iomem_resource, &crashk_res);
558 static void __init reserve_crashkernel(void)
563 static struct resource standard_io_resources[] = {
564 { .name = "dma1", .start = 0x00, .end = 0x1f,
565 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
566 { .name = "pic1", .start = 0x20, .end = 0x21,
567 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
568 { .name = "timer0", .start = 0x40, .end = 0x43,
569 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
570 { .name = "timer1", .start = 0x50, .end = 0x53,
571 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
572 { .name = "keyboard", .start = 0x60, .end = 0x60,
573 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
574 { .name = "keyboard", .start = 0x64, .end = 0x64,
575 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
576 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
577 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
578 { .name = "pic2", .start = 0xa0, .end = 0xa1,
579 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
580 { .name = "dma2", .start = 0xc0, .end = 0xdf,
581 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
582 { .name = "fpu", .start = 0xf0, .end = 0xff,
583 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
586 void __init reserve_standard_io_resources(void)
590 /* request I/O space for devices used on all i[345]86 PCs */
591 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
592 request_resource(&ioport_resource, &standard_io_resources[i]);
596 static bool __init snb_gfx_workaround_needed(void)
601 static const __initconst u16 snb_ids[] = {
611 /* Assume no if something weird is going on with PCI */
612 if (!early_pci_allowed())
615 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
616 if (vendor != 0x8086)
619 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
620 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
621 if (devid == snb_ids[i])
629 * Sandy Bridge graphics has trouble with certain ranges, exclude
630 * them from allocation.
632 static void __init trim_snb_memory(void)
634 static const __initconst unsigned long bad_pages[] = {
643 if (!snb_gfx_workaround_needed())
646 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
649 * SandyBridge integrated graphics devices have a bug that prevents
650 * them from accessing certain memory ranges, namely anything below
651 * 1M and in the pages listed in bad_pages[] above.
653 * To avoid these pages being ever accessed by SNB gfx devices reserve
654 * bad_pages that have not already been reserved at boot time.
655 * All memory below the 1 MB mark is anyway reserved later during
656 * setup_arch(), so there is no need to reserve it here.
659 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
660 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
661 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
666 static void __init trim_bios_range(void)
669 * A special case is the first 4Kb of memory;
670 * This is a BIOS owned area, not kernel ram, but generally
671 * not listed as such in the E820 table.
673 * This typically reserves additional memory (64KiB by default)
674 * since some BIOSes are known to corrupt low memory. See the
675 * Kconfig help text for X86_RESERVE_LOW.
677 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
680 * special case: Some BIOSes report the PC BIOS
681 * area (640Kb -> 1Mb) as RAM even though it is not.
684 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
686 e820__update_table(e820_table);
689 /* called before trim_bios_range() to spare extra sanitize */
690 static void __init e820_add_kernel_range(void)
692 u64 start = __pa_symbol(_text);
693 u64 size = __pa_symbol(_end) - start;
696 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
697 * attempt to fix it by adding the range. We may have a confused BIOS,
698 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
699 * exclude kernel range. If we really are running on top non-RAM,
700 * we will crash later anyways.
702 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
705 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
706 e820__range_remove(start, size, E820_TYPE_RAM, 0);
707 e820__range_add(start, size, E820_TYPE_RAM);
710 static void __init early_reserve_memory(void)
713 * Reserve the memory occupied by the kernel between _text and
714 * __end_of_kernel_reserve symbols. Any kernel sections after the
715 * __end_of_kernel_reserve symbol must be explicitly reserved with a
716 * separate memblock_reserve() or they will be discarded.
718 memblock_reserve(__pa_symbol(_text),
719 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
722 * The first 4Kb of memory is a BIOS owned area, but generally it is
723 * not listed as such in the E820 table.
725 * Reserve the first 64K of memory since some BIOSes are known to
726 * corrupt low memory. After the real mode trampoline is allocated the
727 * rest of the memory below 640k is reserved.
729 * In addition, make sure page 0 is always reserved because on
730 * systems with L1TF its contents can be leaked to user processes.
732 memblock_reserve(0, SZ_64K);
734 early_reserve_initrd();
736 memblock_x86_reserve_range_setup_data();
738 reserve_ibft_region();
739 reserve_bios_regions();
744 * Dump out kernel offset information on panic.
747 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
749 if (kaslr_enabled()) {
750 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
756 pr_emerg("Kernel Offset: disabled\n");
763 * Determine if we were loaded by an EFI loader. If so, then we have also been
764 * passed the efi memmap, systab, etc., so we should use these data structures
765 * for initialization. Note, the efi init code path is determined by the
766 * global efi_enabled. This allows the same kernel image to be used on existing
767 * systems (with a traditional BIOS) as well as on EFI systems.
770 * setup_arch - architecture-specific boot-time initializations
772 * Note: On x86_64, fixmaps are ready for use even before this is called.
775 void __init setup_arch(char **cmdline_p)
778 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
781 * copy kernel address range established so far and switch
782 * to the proper swapper page table
784 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
785 initial_page_table + KERNEL_PGD_BOUNDARY,
788 load_cr3(swapper_pg_dir);
790 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
791 * a cr3 based tlb flush, so the following __flush_tlb_all()
792 * will not flush anything because the CPU quirk which clears
793 * X86_FEATURE_PGE has not been invoked yet. Though due to the
794 * load_cr3() above the TLB has been flushed already. The
795 * quirk is invoked before subsequent calls to __flush_tlb_all()
796 * so proper operation is guaranteed.
800 printk(KERN_INFO "Command line: %s\n", boot_command_line);
801 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
805 * If we have OLPC OFW, we might end up relocating the fixmap due to
806 * reserve_top(), so do this before touching the ioremap area.
810 idt_setup_early_traps();
814 early_ioremap_init();
816 setup_olpc_ofw_pgd();
818 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
819 screen_info = boot_params.screen_info;
820 edid_info = boot_params.edid_info;
822 apm_info.bios = boot_params.apm_bios_info;
823 ist_info = boot_params.ist_info;
825 saved_video_mode = boot_params.hdr.vid_mode;
826 bootloader_type = boot_params.hdr.type_of_loader;
827 if ((bootloader_type >> 4) == 0xe) {
828 bootloader_type &= 0xf;
829 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
831 bootloader_version = bootloader_type & 0xf;
832 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
834 #ifdef CONFIG_BLK_DEV_RAM
835 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
838 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
839 EFI32_LOADER_SIGNATURE, 4)) {
840 set_bit(EFI_BOOT, &efi.flags);
841 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
842 EFI64_LOADER_SIGNATURE, 4)) {
843 set_bit(EFI_BOOT, &efi.flags);
844 set_bit(EFI_64BIT, &efi.flags);
848 x86_init.oem.arch_setup();
851 * Do some memory reservations *before* memory is added to memblock, so
852 * memblock allocations won't overwrite it.
854 * After this point, everything still needed from the boot loader or
855 * firmware or kernel text should be early reserved or marked not RAM in
856 * e820. All other memory is free game.
858 * This call needs to happen before e820__memory_setup() which calls the
859 * xen_memory_setup() on Xen dom0 which relies on the fact that those
860 * early reservations have happened already.
862 early_reserve_memory();
864 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
865 e820__memory_setup();
870 if (!boot_params.hdr.root_flags)
871 root_mountflags &= ~MS_RDONLY;
872 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end);
874 code_resource.start = __pa_symbol(_text);
875 code_resource.end = __pa_symbol(_etext)-1;
876 rodata_resource.start = __pa_symbol(__start_rodata);
877 rodata_resource.end = __pa_symbol(__end_rodata)-1;
878 data_resource.start = __pa_symbol(_sdata);
879 data_resource.end = __pa_symbol(_edata)-1;
880 bss_resource.start = __pa_symbol(__bss_start);
881 bss_resource.end = __pa_symbol(__bss_stop)-1;
883 #ifdef CONFIG_CMDLINE_BOOL
884 #ifdef CONFIG_CMDLINE_OVERRIDE
885 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
887 if (builtin_cmdline[0]) {
888 /* append boot loader cmdline to builtin */
889 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
890 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
891 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
896 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
897 *cmdline_p = command_line;
900 * x86_configure_nx() is called before parse_early_param() to detect
901 * whether hardware doesn't support NX (so that the early EHCI debug
902 * console setup can safely call set_fixmap()). It may then be called
903 * again from within noexec_setup() during parsing early parameters
904 * to honor the respective command line option.
910 if (efi_enabled(EFI_BOOT))
911 efi_memblock_x86_reserve_range();
913 #ifdef CONFIG_MEMORY_HOTPLUG
915 * Memory used by the kernel cannot be hot-removed because Linux
916 * cannot migrate the kernel pages. When memory hotplug is
917 * enabled, we should prevent memblock from allocating memory
920 * ACPI SRAT records all hotpluggable memory ranges. But before
921 * SRAT is parsed, we don't know about it.
923 * The kernel image is loaded into memory at very early time. We
924 * cannot prevent this anyway. So on NUMA system, we set any
925 * node the kernel resides in as un-hotpluggable.
927 * Since on modern servers, one node could have double-digit
928 * gigabytes memory, we can assume the memory around the kernel
929 * image is also un-hotpluggable. So before SRAT is parsed, just
930 * allocate memory near the kernel image to try the best to keep
931 * the kernel away from hotpluggable memory.
933 if (movable_node_is_enabled())
934 memblock_set_bottom_up(true);
939 if (acpi_mps_check()) {
940 #ifdef CONFIG_X86_LOCAL_APIC
943 setup_clear_cpu_cap(X86_FEATURE_APIC);
946 e820__reserve_setup_data();
947 e820__finish_early_params();
949 if (efi_enabled(EFI_BOOT))
955 * VMware detection requires dmi to be available, so this
956 * needs to be done after dmi_setup(), for the boot CPU.
958 init_hypervisor_platform();
961 x86_init.resources.probe_roms();
963 /* after parse_early_param, so could debug it */
964 insert_resource(&iomem_resource, &code_resource);
965 insert_resource(&iomem_resource, &rodata_resource);
966 insert_resource(&iomem_resource, &data_resource);
967 insert_resource(&iomem_resource, &bss_resource);
969 e820_add_kernel_range();
972 if (ppro_with_ram_bug()) {
973 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
975 e820__update_table(e820_table);
976 printk(KERN_INFO "fixed physical RAM map:\n");
977 e820__print_table("bad_ppro");
980 early_gart_iommu_check();
984 * partially used pages are not usable - thus
985 * we are rounding upwards:
987 max_pfn = e820__end_of_ram_pfn();
989 /* update e820 for memory not covered by WB MTRRs */
991 if (mtrr_trim_uncached_memory(max_pfn))
992 max_pfn = e820__end_of_ram_pfn();
994 max_possible_pfn = max_pfn;
997 * This call is required when the CPU does not support PAT. If
998 * mtrr_bp_init() invoked it already via pat_init() the call has no
1004 * Define random base addresses for memory sections after max_pfn is
1005 * defined and before each memory section base is used.
1007 kernel_randomize_memory();
1009 #ifdef CONFIG_X86_32
1010 /* max_low_pfn get updated here */
1011 find_low_pfn_range();
1015 /* How many end-of-memory variables you have, grandma! */
1016 /* need this before calling reserve_initrd */
1017 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1018 max_low_pfn = e820__end_of_low_ram_pfn();
1020 max_low_pfn = max_pfn;
1022 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1026 * Find and reserve possible boot-time SMP configuration:
1030 early_alloc_pgt_buf();
1033 * Need to conclude brk, before e820__memblock_setup()
1034 * it could use memblock_find_in_range, could overlap with
1041 memblock_set_current_limit(ISA_END_ADDRESS);
1042 e820__memblock_setup();
1045 * Needs to run after memblock setup because it needs the physical
1053 efi_mokvar_table_init();
1056 * The EFI specification says that boot service code won't be
1057 * called after ExitBootServices(). This is, in fact, a lie.
1059 efi_reserve_boot_services();
1061 /* preallocate 4k for mptable mpc */
1062 e820__memblock_alloc_reserved_mpc_new();
1064 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1065 setup_bios_corruption_check();
1068 #ifdef CONFIG_X86_32
1069 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1070 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1074 * Find free memory for the real mode trampoline and place it there. If
1075 * there is not enough free memory under 1M, on EFI-enabled systems
1076 * there will be additional attempt to reclaim the memory for the real
1077 * mode trampoline at efi_free_boot_services().
1079 * Unconditionally reserve the entire first 1M of RAM because BIOSes
1080 * are known to corrupt low memory and several hundred kilobytes are not
1081 * worth complex detection what memory gets clobbered. Windows does the
1082 * same thing for very similar reasons.
1084 * Moreover, on machines with SandyBridge graphics or in setups that use
1085 * crashkernel the entire 1M is reserved anyway.
1087 reserve_real_mode();
1091 idt_setup_early_pf();
1094 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1095 * with the current CR4 value. This may not be necessary, but
1096 * auditing all the early-boot CR4 manipulation would be needed to
1099 * Mask off features that don't work outside long mode (just
1102 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1104 memblock_set_current_limit(get_max_mapped());
1107 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1110 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1111 if (init_ohci1394_dma_early)
1112 init_ohci1394_dma_on_all_controllers();
1114 /* Allocate bigger log buffer */
1117 if (efi_enabled(EFI_BOOT)) {
1118 switch (boot_params.secure_boot) {
1119 case efi_secureboot_mode_disabled:
1120 pr_info("Secure boot disabled\n");
1122 case efi_secureboot_mode_enabled:
1123 pr_info("Secure boot enabled\n");
1126 pr_info("Secure boot could not be determined\n");
1133 acpi_table_upgrade();
1134 /* Look for ACPI tables and reserve memory occupied by them. */
1135 acpi_boot_table_init();
1141 early_platform_quirks();
1143 early_acpi_boot_init();
1146 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1148 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1149 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1152 * Reserve memory for crash kernel after SRAT is parsed so that it
1153 * won't consume hotpluggable memory.
1155 reserve_crashkernel();
1157 memblock_find_dma_reserve();
1159 if (!early_xdbc_setup_hardware())
1160 early_xdbc_register_console();
1162 x86_init.paging.pagetable_init();
1167 * Sync back kernel address range.
1169 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1172 sync_initial_page_table();
1178 generic_apic_probe();
1183 * Read APIC and some other early information from ACPI tables.
1189 * get boot-time SMP configuration:
1194 * Systems w/o ACPI and mptables might not have it mapped the local
1195 * APIC yet, but prefill_possible_map() might need to access it.
1197 init_apic_mappings();
1199 prefill_possible_map();
1204 io_apic_init_mappings();
1206 x86_init.hyper.guest_late_init();
1208 e820__reserve_resources();
1209 e820__register_nosave_regions(max_pfn);
1211 x86_init.resources.reserve_resources();
1213 e820__setup_pci_gap();
1216 #if defined(CONFIG_VGA_CONSOLE)
1217 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1218 conswitchp = &vga_con;
1221 x86_init.oem.banner();
1223 x86_init.timers.wallclock_init();
1226 * This needs to run before setup_local_APIC() which soft-disables the
1227 * local APIC temporarily and that masks the thermal LVT interrupt,
1228 * leading to softlockups on machines which have configured SMI
1229 * interrupt delivery.
1235 register_refined_jiffies(CLOCK_TICK_RATE);
1238 if (efi_enabled(EFI_BOOT))
1239 efi_apply_memmap_quirks();
1245 #ifdef CONFIG_X86_32
1247 static struct resource video_ram_resource = {
1248 .name = "Video RAM area",
1251 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1254 void __init i386_reserve_resources(void)
1256 request_resource(&iomem_resource, &video_ram_resource);
1257 reserve_standard_io_resources();
1260 #endif /* CONFIG_X86_32 */
1262 static struct notifier_block kernel_offset_notifier = {
1263 .notifier_call = dump_kernel_offset
1266 static int __init register_kernel_offset_dumper(void)
1268 atomic_notifier_chain_register(&panic_notifier_list,
1269 &kernel_offset_notifier);
1272 __initcall(register_kernel_offset_dumper);