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/pci.h>
18 #include <linux/root_dev.h>
19 #include <linux/hugetlb.h>
20 #include <linux/tboot.h>
21 #include <linux/usb/xhci-dbgp.h>
22 #include <linux/static_call.h>
23 #include <linux/swiotlb.h>
25 #include <uapi/linux/mount.h>
31 #include <asm/bios_ebda.h>
36 #include <asm/hypervisor.h>
37 #include <asm/io_apic.h>
38 #include <asm/kasan.h>
39 #include <asm/kaslr.h>
42 #include <asm/realmode.h>
43 #include <asm/olpc_ofw.h>
44 #include <asm/pci-direct.h>
46 #include <asm/proto.h>
47 #include <asm/thermal.h>
48 #include <asm/unwind.h>
49 #include <asm/vsyscall.h>
50 #include <linux/vmalloc.h>
53 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
54 * max_pfn_mapped: highest directly mapped pfn > 4 GB
56 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
57 * represented by pfn_mapped[].
59 unsigned long max_low_pfn_mapped;
60 unsigned long max_pfn_mapped;
63 RESERVE_BRK(dmi_alloc, 65536);
68 * Range of the BSS area. The size of the BSS area is determined
69 * at link time, with RESERVE_BRK() facility reserving additional
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;
145 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
146 int bootloader_type, bootloader_version;
151 struct screen_info screen_info;
152 EXPORT_SYMBOL(screen_info);
153 struct edid_info edid_info;
154 EXPORT_SYMBOL_GPL(edid_info);
156 extern int root_mountflags;
158 unsigned long saved_video_mode;
160 #define RAMDISK_IMAGE_START_MASK 0x07FF
161 #define RAMDISK_PROMPT_FLAG 0x8000
162 #define RAMDISK_LOAD_FLAG 0x4000
164 static char __initdata command_line[COMMAND_LINE_SIZE];
165 #ifdef CONFIG_CMDLINE_BOOL
166 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
169 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
171 #ifdef CONFIG_EDD_MODULE
175 * copy_edd() - Copy the BIOS EDD information
176 * from boot_params into a safe place.
179 static inline void __init copy_edd(void)
181 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
182 sizeof(edd.mbr_signature));
183 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
184 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
185 edd.edd_info_nr = boot_params.eddbuf_entries;
188 static inline void __init copy_edd(void)
193 void * __init extend_brk(size_t size, size_t align)
195 size_t mask = align - 1;
198 BUG_ON(_brk_start == 0);
199 BUG_ON(align & mask);
201 _brk_end = (_brk_end + mask) & ~mask;
202 BUG_ON((char *)(_brk_end + size) > __brk_limit);
204 ret = (void *)_brk_end;
207 memset(ret, 0, size);
213 static void __init cleanup_highmap(void)
218 static void __init reserve_brk(void)
220 if (_brk_end > _brk_start)
221 memblock_reserve(__pa_symbol(_brk_start),
222 _brk_end - _brk_start);
224 /* Mark brk area as locked down and no longer taking any
229 u64 relocated_ramdisk;
231 #ifdef CONFIG_BLK_DEV_INITRD
233 static u64 __init get_ramdisk_image(void)
235 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
239 if (ramdisk_image == 0)
240 ramdisk_image = phys_initrd_start;
242 return ramdisk_image;
244 static u64 __init get_ramdisk_size(void)
246 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
250 if (ramdisk_size == 0)
251 ramdisk_size = phys_initrd_size;
256 static void __init relocate_initrd(void)
258 /* Assume only end is not page aligned */
259 u64 ramdisk_image = get_ramdisk_image();
260 u64 ramdisk_size = get_ramdisk_size();
261 u64 area_size = PAGE_ALIGN(ramdisk_size);
263 /* We need to move the initrd down into directly mapped mem */
264 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
265 PFN_PHYS(max_pfn_mapped));
266 if (!relocated_ramdisk)
267 panic("Cannot find place for new RAMDISK of size %lld\n",
270 initrd_start = relocated_ramdisk + PAGE_OFFSET;
271 initrd_end = initrd_start + ramdisk_size;
272 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
273 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
275 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
277 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
278 " [mem %#010llx-%#010llx]\n",
279 ramdisk_image, ramdisk_image + ramdisk_size - 1,
280 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
283 static void __init early_reserve_initrd(void)
285 /* Assume only end is not page aligned */
286 u64 ramdisk_image = get_ramdisk_image();
287 u64 ramdisk_size = get_ramdisk_size();
288 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
290 if (!boot_params.hdr.type_of_loader ||
291 !ramdisk_image || !ramdisk_size)
292 return; /* No initrd provided by bootloader */
294 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
297 static void __init reserve_initrd(void)
299 /* Assume only end is not page aligned */
300 u64 ramdisk_image = get_ramdisk_image();
301 u64 ramdisk_size = get_ramdisk_size();
302 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
304 if (!boot_params.hdr.type_of_loader ||
305 !ramdisk_image || !ramdisk_size)
306 return; /* No initrd provided by bootloader */
310 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
313 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
314 PFN_DOWN(ramdisk_end))) {
315 /* All are mapped, easy case */
316 initrd_start = ramdisk_image + PAGE_OFFSET;
317 initrd_end = initrd_start + ramdisk_size;
323 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
327 static void __init early_reserve_initrd(void)
330 static void __init reserve_initrd(void)
333 #endif /* CONFIG_BLK_DEV_INITRD */
335 static void __init parse_setup_data(void)
337 struct setup_data *data;
338 u64 pa_data, pa_next;
340 pa_data = boot_params.hdr.setup_data;
342 u32 data_len, data_type;
344 data = early_memremap(pa_data, sizeof(*data));
345 data_len = data->len + sizeof(struct setup_data);
346 data_type = data->type;
347 pa_next = data->next;
348 early_memunmap(data, sizeof(*data));
352 e820__memory_setup_extended(pa_data, data_len);
358 parse_efi_setup(pa_data, data_len);
367 static void __init memblock_x86_reserve_range_setup_data(void)
369 struct setup_data *data;
372 pa_data = boot_params.hdr.setup_data;
374 data = early_memremap(pa_data, sizeof(*data));
375 memblock_reserve(pa_data, sizeof(*data) + data->len);
377 if (data->type == SETUP_INDIRECT &&
378 ((struct setup_indirect *)data->data)->type != SETUP_INDIRECT)
379 memblock_reserve(((struct setup_indirect *)data->data)->addr,
380 ((struct setup_indirect *)data->data)->len);
382 pa_data = data->next;
383 early_memunmap(data, sizeof(*data));
388 * --------- Crashkernel reservation ------------------------------
391 #ifdef CONFIG_KEXEC_CORE
393 /* 16M alignment for crash kernel regions */
394 #define CRASH_ALIGN SZ_16M
397 * Keep the crash kernel below this limit.
399 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
400 * due to mapping restrictions.
402 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
403 * the upper limit of system RAM in 4-level paging mode. Since the kdump
404 * jump could be from 5-level paging to 4-level paging, the jump will fail if
405 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
406 * no good way to detect the paging mode of the target kernel which will be
407 * loaded for dumping.
410 # define CRASH_ADDR_LOW_MAX SZ_512M
411 # define CRASH_ADDR_HIGH_MAX SZ_512M
413 # define CRASH_ADDR_LOW_MAX SZ_4G
414 # define CRASH_ADDR_HIGH_MAX SZ_64T
417 static int __init reserve_crashkernel_low(void)
420 unsigned long long base, low_base = 0, low_size = 0;
421 unsigned long low_mem_limit;
424 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
426 /* crashkernel=Y,low */
427 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
430 * two parts from kernel/dma/swiotlb.c:
431 * -swiotlb size: user-specified with swiotlb= or default.
433 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
434 * to 8M for other buffers that may need to stay low too. Also
435 * make sure we allocate enough extra low memory so that we
436 * don't run out of DMA buffers for 32-bit devices.
438 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
440 /* passed with crashkernel=0,low ? */
445 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
447 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
448 (unsigned long)(low_size >> 20));
452 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
453 (unsigned long)(low_size >> 20),
454 (unsigned long)(low_base >> 20),
455 (unsigned long)(low_mem_limit >> 20));
457 crashk_low_res.start = low_base;
458 crashk_low_res.end = low_base + low_size - 1;
459 insert_resource(&iomem_resource, &crashk_low_res);
464 static void __init reserve_crashkernel(void)
466 unsigned long long crash_size, crash_base, total_mem;
470 total_mem = memblock_phys_mem_size();
473 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
474 if (ret != 0 || crash_size <= 0) {
475 /* crashkernel=X,high */
476 ret = parse_crashkernel_high(boot_command_line, total_mem,
477 &crash_size, &crash_base);
478 if (ret != 0 || crash_size <= 0)
483 if (xen_pv_domain()) {
484 pr_info("Ignoring crashkernel for a Xen PV domain\n");
488 /* 0 means: find the address automatically */
491 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
492 * crashkernel=x,high reserves memory over 4G, also allocates
493 * 256M extra low memory for DMA buffers and swiotlb.
494 * But the extra memory is not required for all machines.
495 * So try low memory first and fall back to high memory
496 * unless "crashkernel=size[KMG],high" is specified.
499 crash_base = memblock_phys_alloc_range(crash_size,
500 CRASH_ALIGN, CRASH_ALIGN,
503 crash_base = memblock_phys_alloc_range(crash_size,
504 CRASH_ALIGN, CRASH_ALIGN,
505 CRASH_ADDR_HIGH_MAX);
507 pr_info("crashkernel reservation failed - No suitable area found.\n");
511 unsigned long long start;
513 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
514 crash_base + crash_size);
515 if (start != crash_base) {
516 pr_info("crashkernel reservation failed - memory is in use.\n");
521 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
522 memblock_free(crash_base, crash_size);
526 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
527 (unsigned long)(crash_size >> 20),
528 (unsigned long)(crash_base >> 20),
529 (unsigned long)(total_mem >> 20));
531 crashk_res.start = crash_base;
532 crashk_res.end = crash_base + crash_size - 1;
533 insert_resource(&iomem_resource, &crashk_res);
536 static void __init reserve_crashkernel(void)
541 static struct resource standard_io_resources[] = {
542 { .name = "dma1", .start = 0x00, .end = 0x1f,
543 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
544 { .name = "pic1", .start = 0x20, .end = 0x21,
545 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
546 { .name = "timer0", .start = 0x40, .end = 0x43,
547 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
548 { .name = "timer1", .start = 0x50, .end = 0x53,
549 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
550 { .name = "keyboard", .start = 0x60, .end = 0x60,
551 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
552 { .name = "keyboard", .start = 0x64, .end = 0x64,
553 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
554 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
555 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
556 { .name = "pic2", .start = 0xa0, .end = 0xa1,
557 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
558 { .name = "dma2", .start = 0xc0, .end = 0xdf,
559 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
560 { .name = "fpu", .start = 0xf0, .end = 0xff,
561 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
564 void __init reserve_standard_io_resources(void)
568 /* request I/O space for devices used on all i[345]86 PCs */
569 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
570 request_resource(&ioport_resource, &standard_io_resources[i]);
574 static bool __init snb_gfx_workaround_needed(void)
579 static const __initconst u16 snb_ids[] = {
589 /* Assume no if something weird is going on with PCI */
590 if (!early_pci_allowed())
593 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
594 if (vendor != 0x8086)
597 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
598 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
599 if (devid == snb_ids[i])
607 * Sandy Bridge graphics has trouble with certain ranges, exclude
608 * them from allocation.
610 static void __init trim_snb_memory(void)
612 static const __initconst unsigned long bad_pages[] = {
621 if (!snb_gfx_workaround_needed())
624 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
627 * SandyBridge integrated graphics devices have a bug that prevents
628 * them from accessing certain memory ranges, namely anything below
629 * 1M and in the pages listed in bad_pages[] above.
631 * To avoid these pages being ever accessed by SNB gfx devices reserve
632 * bad_pages that have not already been reserved at boot time.
633 * All memory below the 1 MB mark is anyway reserved later during
634 * setup_arch(), so there is no need to reserve it here.
637 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
638 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
639 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
644 static void __init trim_bios_range(void)
647 * A special case is the first 4Kb of memory;
648 * This is a BIOS owned area, not kernel ram, but generally
649 * not listed as such in the E820 table.
651 * This typically reserves additional memory (64KiB by default)
652 * since some BIOSes are known to corrupt low memory. See the
653 * Kconfig help text for X86_RESERVE_LOW.
655 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
658 * special case: Some BIOSes report the PC BIOS
659 * area (640Kb -> 1Mb) as RAM even though it is not.
662 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
664 e820__update_table(e820_table);
667 /* called before trim_bios_range() to spare extra sanitize */
668 static void __init e820_add_kernel_range(void)
670 u64 start = __pa_symbol(_text);
671 u64 size = __pa_symbol(_end) - start;
674 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
675 * attempt to fix it by adding the range. We may have a confused BIOS,
676 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
677 * exclude kernel range. If we really are running on top non-RAM,
678 * we will crash later anyways.
680 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
683 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
684 e820__range_remove(start, size, E820_TYPE_RAM, 0);
685 e820__range_add(start, size, E820_TYPE_RAM);
688 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
690 static int __init parse_reservelow(char *p)
692 unsigned long long size;
697 size = memparse(p, &p);
710 early_param("reservelow", parse_reservelow);
712 static void __init early_reserve_memory(void)
715 * Reserve the memory occupied by the kernel between _text and
716 * __end_of_kernel_reserve symbols. Any kernel sections after the
717 * __end_of_kernel_reserve symbol must be explicitly reserved with a
718 * separate memblock_reserve() or they will be discarded.
720 memblock_reserve(__pa_symbol(_text),
721 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
724 * The first 4Kb of memory is a BIOS owned area, but generally it is
725 * not listed as such in the E820 table.
727 * Reserve the first 64K of memory since some BIOSes are known to
728 * corrupt low memory. After the real mode trampoline is allocated the
729 * rest of the memory below 640k is reserved.
731 * In addition, make sure page 0 is always reserved because on
732 * systems with L1TF its contents can be leaked to user processes.
734 memblock_reserve(0, SZ_64K);
736 early_reserve_initrd();
738 if (efi_enabled(EFI_BOOT))
739 efi_memblock_x86_reserve_range();
741 memblock_x86_reserve_range_setup_data();
743 reserve_ibft_region();
744 reserve_bios_regions();
749 * Dump out kernel offset information on panic.
752 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
754 if (kaslr_enabled()) {
755 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
761 pr_emerg("Kernel Offset: disabled\n");
768 * Determine if we were loaded by an EFI loader. If so, then we have also been
769 * passed the efi memmap, systab, etc., so we should use these data structures
770 * for initialization. Note, the efi init code path is determined by the
771 * global efi_enabled. This allows the same kernel image to be used on existing
772 * systems (with a traditional BIOS) as well as on EFI systems.
775 * setup_arch - architecture-specific boot-time initializations
777 * Note: On x86_64, fixmaps are ready for use even before this is called.
780 void __init setup_arch(char **cmdline_p)
783 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
786 * copy kernel address range established so far and switch
787 * to the proper swapper page table
789 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
790 initial_page_table + KERNEL_PGD_BOUNDARY,
793 load_cr3(swapper_pg_dir);
795 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
796 * a cr3 based tlb flush, so the following __flush_tlb_all()
797 * will not flush anything because the CPU quirk which clears
798 * X86_FEATURE_PGE has not been invoked yet. Though due to the
799 * load_cr3() above the TLB has been flushed already. The
800 * quirk is invoked before subsequent calls to __flush_tlb_all()
801 * so proper operation is guaranteed.
805 printk(KERN_INFO "Command line: %s\n", boot_command_line);
806 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
810 * If we have OLPC OFW, we might end up relocating the fixmap due to
811 * reserve_top(), so do this before touching the ioremap area.
815 idt_setup_early_traps();
819 early_ioremap_init();
821 setup_olpc_ofw_pgd();
823 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
824 screen_info = boot_params.screen_info;
825 edid_info = boot_params.edid_info;
827 apm_info.bios = boot_params.apm_bios_info;
828 ist_info = boot_params.ist_info;
830 saved_video_mode = boot_params.hdr.vid_mode;
831 bootloader_type = boot_params.hdr.type_of_loader;
832 if ((bootloader_type >> 4) == 0xe) {
833 bootloader_type &= 0xf;
834 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
836 bootloader_version = bootloader_type & 0xf;
837 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
839 #ifdef CONFIG_BLK_DEV_RAM
840 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
843 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
844 EFI32_LOADER_SIGNATURE, 4)) {
845 set_bit(EFI_BOOT, &efi.flags);
846 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
847 EFI64_LOADER_SIGNATURE, 4)) {
848 set_bit(EFI_BOOT, &efi.flags);
849 set_bit(EFI_64BIT, &efi.flags);
853 x86_init.oem.arch_setup();
855 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
856 e820__memory_setup();
861 if (!boot_params.hdr.root_flags)
862 root_mountflags &= ~MS_RDONLY;
863 init_mm.start_code = (unsigned long) _text;
864 init_mm.end_code = (unsigned long) _etext;
865 init_mm.end_data = (unsigned long) _edata;
866 init_mm.brk = _brk_end;
868 code_resource.start = __pa_symbol(_text);
869 code_resource.end = __pa_symbol(_etext)-1;
870 rodata_resource.start = __pa_symbol(__start_rodata);
871 rodata_resource.end = __pa_symbol(__end_rodata)-1;
872 data_resource.start = __pa_symbol(_sdata);
873 data_resource.end = __pa_symbol(_edata)-1;
874 bss_resource.start = __pa_symbol(__bss_start);
875 bss_resource.end = __pa_symbol(__bss_stop)-1;
877 #ifdef CONFIG_CMDLINE_BOOL
878 #ifdef CONFIG_CMDLINE_OVERRIDE
879 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
881 if (builtin_cmdline[0]) {
882 /* append boot loader cmdline to builtin */
883 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
884 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
885 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
890 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
891 *cmdline_p = command_line;
894 * x86_configure_nx() is called before parse_early_param() to detect
895 * whether hardware doesn't support NX (so that the early EHCI debug
896 * console setup can safely call set_fixmap()). It may then be called
897 * again from within noexec_setup() during parsing early parameters
898 * to honor the respective command line option.
905 * Do some memory reservations *before* memory is added to
906 * memblock, so memblock allocations won't overwrite it.
907 * Do it after early param, so we could get (unlikely) panic from
910 * After this point everything still needed from the boot loader or
911 * firmware or kernel text should be early reserved or marked not
912 * RAM in e820. All other memory is free game.
914 early_reserve_memory();
916 #ifdef CONFIG_MEMORY_HOTPLUG
918 * Memory used by the kernel cannot be hot-removed because Linux
919 * cannot migrate the kernel pages. When memory hotplug is
920 * enabled, we should prevent memblock from allocating memory
923 * ACPI SRAT records all hotpluggable memory ranges. But before
924 * SRAT is parsed, we don't know about it.
926 * The kernel image is loaded into memory at very early time. We
927 * cannot prevent this anyway. So on NUMA system, we set any
928 * node the kernel resides in as un-hotpluggable.
930 * Since on modern servers, one node could have double-digit
931 * gigabytes memory, we can assume the memory around the kernel
932 * image is also un-hotpluggable. So before SRAT is parsed, just
933 * allocate memory near the kernel image to try the best to keep
934 * the kernel away from hotpluggable memory.
936 if (movable_node_is_enabled())
937 memblock_set_bottom_up(true);
942 if (acpi_mps_check()) {
943 #ifdef CONFIG_X86_LOCAL_APIC
946 setup_clear_cpu_cap(X86_FEATURE_APIC);
949 e820__reserve_setup_data();
950 e820__finish_early_params();
952 if (efi_enabled(EFI_BOOT))
958 * VMware detection requires dmi to be available, so this
959 * needs to be done after dmi_setup(), for the boot CPU.
961 init_hypervisor_platform();
964 x86_init.resources.probe_roms();
966 /* after parse_early_param, so could debug it */
967 insert_resource(&iomem_resource, &code_resource);
968 insert_resource(&iomem_resource, &rodata_resource);
969 insert_resource(&iomem_resource, &data_resource);
970 insert_resource(&iomem_resource, &bss_resource);
972 e820_add_kernel_range();
975 if (ppro_with_ram_bug()) {
976 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
978 e820__update_table(e820_table);
979 printk(KERN_INFO "fixed physical RAM map:\n");
980 e820__print_table("bad_ppro");
983 early_gart_iommu_check();
987 * partially used pages are not usable - thus
988 * we are rounding upwards:
990 max_pfn = e820__end_of_ram_pfn();
992 /* update e820 for memory not covered by WB MTRRs */
994 if (mtrr_trim_uncached_memory(max_pfn))
995 max_pfn = e820__end_of_ram_pfn();
997 max_possible_pfn = max_pfn;
1000 * This call is required when the CPU does not support PAT. If
1001 * mtrr_bp_init() invoked it already via pat_init() the call has no
1007 * Define random base addresses for memory sections after max_pfn is
1008 * defined and before each memory section base is used.
1010 kernel_randomize_memory();
1012 #ifdef CONFIG_X86_32
1013 /* max_low_pfn get updated here */
1014 find_low_pfn_range();
1018 /* How many end-of-memory variables you have, grandma! */
1019 /* need this before calling reserve_initrd */
1020 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1021 max_low_pfn = e820__end_of_low_ram_pfn();
1023 max_low_pfn = max_pfn;
1025 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1029 * Find and reserve possible boot-time SMP configuration:
1033 early_alloc_pgt_buf();
1036 * Need to conclude brk, before e820__memblock_setup()
1037 * it could use memblock_find_in_range, could overlap with
1044 memblock_set_current_limit(ISA_END_ADDRESS);
1045 e820__memblock_setup();
1048 * Needs to run after memblock setup because it needs the physical
1056 efi_mokvar_table_init();
1059 * The EFI specification says that boot service code won't be
1060 * called after ExitBootServices(). This is, in fact, a lie.
1062 efi_reserve_boot_services();
1064 /* preallocate 4k for mptable mpc */
1065 e820__memblock_alloc_reserved_mpc_new();
1067 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1068 setup_bios_corruption_check();
1071 #ifdef CONFIG_X86_32
1072 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1073 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1077 * Find free memory for the real mode trampoline and place it
1079 * If there is not enough free memory under 1M, on EFI-enabled
1080 * systems there will be additional attempt to reclaim the memory
1081 * for the real mode trampoline at efi_free_boot_services().
1083 * Unconditionally reserve the entire first 1M of RAM because
1084 * BIOSes are know to corrupt low memory and several
1085 * hundred kilobytes are not worth complex detection what memory gets
1086 * clobbered. Moreover, on machines with SandyBridge graphics or in
1087 * setups that use crashkernel the entire 1M is reserved anyway.
1089 reserve_real_mode();
1093 idt_setup_early_pf();
1096 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1097 * with the current CR4 value. This may not be necessary, but
1098 * auditing all the early-boot CR4 manipulation would be needed to
1101 * Mask off features that don't work outside long mode (just
1104 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1106 memblock_set_current_limit(get_max_mapped());
1109 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1112 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1113 if (init_ohci1394_dma_early)
1114 init_ohci1394_dma_on_all_controllers();
1116 /* Allocate bigger log buffer */
1119 if (efi_enabled(EFI_BOOT)) {
1120 switch (boot_params.secure_boot) {
1121 case efi_secureboot_mode_disabled:
1122 pr_info("Secure boot disabled\n");
1124 case efi_secureboot_mode_enabled:
1125 pr_info("Secure boot enabled\n");
1128 pr_info("Secure boot could not be determined\n");
1135 acpi_table_upgrade();
1136 /* Look for ACPI tables and reserve memory occupied by them. */
1137 acpi_boot_table_init();
1143 early_platform_quirks();
1145 early_acpi_boot_init();
1148 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1150 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1151 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1154 * Reserve memory for crash kernel after SRAT is parsed so that it
1155 * won't consume hotpluggable memory.
1157 reserve_crashkernel();
1159 memblock_find_dma_reserve();
1161 if (!early_xdbc_setup_hardware())
1162 early_xdbc_register_console();
1164 x86_init.paging.pagetable_init();
1169 * Sync back kernel address range.
1171 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1174 sync_initial_page_table();
1180 generic_apic_probe();
1185 * Read APIC and some other early information from ACPI tables.
1191 * get boot-time SMP configuration:
1196 * Systems w/o ACPI and mptables might not have it mapped the local
1197 * APIC yet, but prefill_possible_map() might need to access it.
1199 init_apic_mappings();
1201 prefill_possible_map();
1206 io_apic_init_mappings();
1208 x86_init.hyper.guest_late_init();
1210 e820__reserve_resources();
1211 e820__register_nosave_regions(max_pfn);
1213 x86_init.resources.reserve_resources();
1215 e820__setup_pci_gap();
1218 #if defined(CONFIG_VGA_CONSOLE)
1219 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1220 conswitchp = &vga_con;
1223 x86_init.oem.banner();
1225 x86_init.timers.wallclock_init();
1228 * This needs to run before setup_local_APIC() which soft-disables the
1229 * local APIC temporarily and that masks the thermal LVT interrupt,
1230 * leading to softlockups on machines which have configured SMI
1231 * interrupt delivery.
1237 register_refined_jiffies(CLOCK_TICK_RATE);
1240 if (efi_enabled(EFI_BOOT))
1241 efi_apply_memmap_quirks();
1247 #ifdef CONFIG_X86_32
1249 static struct resource video_ram_resource = {
1250 .name = "Video RAM area",
1253 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1256 void __init i386_reserve_resources(void)
1258 request_resource(&iomem_resource, &video_ram_resource);
1259 reserve_standard_io_resources();
1262 #endif /* CONFIG_X86_32 */
1264 static struct notifier_block kernel_offset_notifier = {
1265 .notifier_call = dump_kernel_offset
1268 static int __init register_kernel_offset_dumper(void)
1270 atomic_notifier_chain_register(&panic_notifier_list,
1271 &kernel_offset_notifier);
1274 __initcall(register_kernel_offset_dumper);