2 * Copyright (C) 1995 Linus Torvalds
6 * This file handles the architecture-dependent parts of initialization
9 #include <linux/errno.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
13 #include <linux/stddef.h>
14 #include <linux/unistd.h>
15 #include <linux/ptrace.h>
16 #include <linux/slab.h>
17 #include <linux/user.h>
18 #include <linux/screen_info.h>
19 #include <linux/ioport.h>
20 #include <linux/delay.h>
21 #include <linux/init.h>
22 #include <linux/initrd.h>
23 #include <linux/highmem.h>
24 #include <linux/bootmem.h>
25 #include <linux/module.h>
26 #include <asm/processor.h>
27 #include <linux/console.h>
28 #include <linux/seq_file.h>
29 #include <linux/crash_dump.h>
30 #include <linux/root_dev.h>
31 #include <linux/pci.h>
32 #include <asm/pci-direct.h>
33 #include <linux/efi.h>
34 #include <linux/acpi.h>
35 #include <linux/kallsyms.h>
36 #include <linux/edd.h>
37 #include <linux/iscsi_ibft.h>
38 #include <linux/mmzone.h>
39 #include <linux/kexec.h>
40 #include <linux/cpufreq.h>
41 #include <linux/dmi.h>
42 #include <linux/dma-mapping.h>
43 #include <linux/ctype.h>
44 #include <linux/sort.h>
45 #include <linux/uaccess.h>
46 #include <linux/init_ohci1394_dma.h>
47 #include <linux/kvm_para.h>
50 #include <asm/uaccess.h>
51 #include <asm/system.h>
52 #include <asm/vsyscall.h>
57 #include <video/edid.h>
59 #include <asm/mpspec.h>
62 #include <asm/mpspec.h>
63 #include <asm/mmu_context.h>
64 #include <asm/proto.h>
65 #include <asm/setup.h>
67 #include <asm/sections.h>
69 #include <asm/cacheflush.h>
72 #include <asm/topology.h>
73 #include <asm/trampoline.h>
76 #include <mach_apic.h>
77 #ifdef CONFIG_PARAVIRT
78 #include <asm/paravirt.h>
87 struct cpuinfo_x86 boot_cpu_data __read_mostly;
88 EXPORT_SYMBOL(boot_cpu_data);
90 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
92 unsigned long mmu_cr4_features;
94 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
97 unsigned long saved_video_mode;
103 char dmi_alloc_data[DMI_MAX_DATA];
108 struct screen_info screen_info;
109 EXPORT_SYMBOL(screen_info);
110 struct sys_desc_table_struct {
111 unsigned short length;
112 unsigned char table[0];
115 struct edid_info edid_info;
116 EXPORT_SYMBOL_GPL(edid_info);
118 extern int root_mountflags;
120 static char __initdata command_line[COMMAND_LINE_SIZE];
122 static struct resource standard_io_resources[] = {
123 { .name = "dma1", .start = 0x00, .end = 0x1f,
124 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
125 { .name = "pic1", .start = 0x20, .end = 0x21,
126 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
127 { .name = "timer0", .start = 0x40, .end = 0x43,
128 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
129 { .name = "timer1", .start = 0x50, .end = 0x53,
130 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
131 { .name = "keyboard", .start = 0x60, .end = 0x60,
132 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
133 { .name = "keyboard", .start = 0x64, .end = 0x64,
134 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
135 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
136 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
137 { .name = "pic2", .start = 0xa0, .end = 0xa1,
138 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
139 { .name = "dma2", .start = 0xc0, .end = 0xdf,
140 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
141 { .name = "fpu", .start = 0xf0, .end = 0xff,
142 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
145 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
147 static struct resource data_resource = {
148 .name = "Kernel data",
151 .flags = IORESOURCE_RAM,
153 static struct resource code_resource = {
154 .name = "Kernel code",
157 .flags = IORESOURCE_RAM,
159 static struct resource bss_resource = {
160 .name = "Kernel bss",
163 .flags = IORESOURCE_RAM,
166 static void __init early_cpu_init(void);
167 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
169 #ifdef CONFIG_PROC_VMCORE
170 /* elfcorehdr= specifies the location of elf core header
171 * stored by the crashed kernel. This option will be passed
172 * by kexec loader to the capture kernel.
174 static int __init setup_elfcorehdr(char *arg)
179 elfcorehdr_addr = memparse(arg, &end);
180 return end > arg ? 0 : -EINVAL;
182 early_param("elfcorehdr", setup_elfcorehdr);
187 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
189 unsigned long bootmap_size, bootmap;
191 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
192 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
195 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
196 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
197 e820_register_active_regions(0, start_pfn, end_pfn);
198 free_bootmem_with_active_regions(0, end_pfn);
199 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
200 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
204 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
206 #ifdef CONFIG_EDD_MODULE
210 * copy_edd() - Copy the BIOS EDD information
211 * from boot_params into a safe place.
214 static inline void copy_edd(void)
216 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
217 sizeof(edd.mbr_signature));
218 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
219 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
220 edd.edd_info_nr = boot_params.eddbuf_entries;
223 static inline void copy_edd(void)
228 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
229 void __attribute__((weak)) __init memory_setup(void)
231 machine_specific_memory_setup();
234 /* Current gdt points %fs at the "master" per-cpu area: after this,
235 * it's on the real one. */
236 void switch_to_new_gdt(void)
238 struct desc_ptr gdt_descr;
240 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
241 gdt_descr.size = GDT_SIZE - 1;
242 load_gdt(&gdt_descr);
246 * setup_arch - architecture-specific boot-time initializations
248 * Note: On x86_64, fixmaps are ready for use even before this is called.
250 void __init setup_arch(char **cmdline_p)
254 printk(KERN_INFO "Command line: %s\n", boot_command_line);
256 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
257 screen_info = boot_params.screen_info;
258 edid_info = boot_params.edid_info;
259 saved_video_mode = boot_params.hdr.vid_mode;
260 bootloader_type = boot_params.hdr.type_of_loader;
262 #ifdef CONFIG_BLK_DEV_RAM
263 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
264 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
265 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
268 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
280 if (!boot_params.hdr.root_flags)
281 root_mountflags &= ~MS_RDONLY;
282 init_mm.start_code = (unsigned long) &_text;
283 init_mm.end_code = (unsigned long) &_etext;
284 init_mm.end_data = (unsigned long) &_edata;
285 init_mm.brk = (unsigned long) &_end;
287 code_resource.start = virt_to_phys(&_text);
288 code_resource.end = virt_to_phys(&_etext)-1;
289 data_resource.start = virt_to_phys(&_etext);
290 data_resource.end = virt_to_phys(&_edata)-1;
291 bss_resource.start = virt_to_phys(&__bss_start);
292 bss_resource.end = virt_to_phys(&__bss_stop)-1;
295 early_identify_cpu(&boot_cpu_data);
297 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
298 *cmdline_p = command_line;
304 if (acpi_mps_check()) {
306 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
309 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
310 if (init_ohci1394_dma_early)
311 init_ohci1394_dma_on_all_controllers();
314 finish_e820_parsing();
316 /* after parse_early_param, so could debug it */
317 insert_resource(&iomem_resource, &code_resource);
318 insert_resource(&iomem_resource, &data_resource);
319 insert_resource(&iomem_resource, &bss_resource);
321 early_gart_iommu_check();
323 e820_register_active_regions(0, 0, -1UL);
325 * partially used pages are not usable - thus
326 * we are rounding upwards:
328 end_pfn = e820_end_of_ram();
330 /* pre allocte 4k for mptable mpc */
331 early_reserve_e820_mpc_new();
332 /* update e820 for memory not covered by WB MTRRs */
334 if (mtrr_trim_uncached_memory(end_pfn)) {
335 remove_all_active_ranges();
336 e820_register_active_regions(0, 0, -1UL);
337 end_pfn = e820_end_of_ram();
340 num_physpages = end_pfn;
344 max_pfn_mapped = init_memory_mapping(0, (end_pfn << PAGE_SHIFT));
354 #ifdef CONFIG_KVM_CLOCK
360 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
361 * Call this early for SRAT node setup.
363 acpi_boot_table_init();
366 /* How many end-of-memory variables you have, grandma! */
367 max_low_pfn = end_pfn;
369 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
371 /* Remove active ranges so rediscovery with NUMA-awareness happens */
372 remove_all_active_ranges();
374 #ifdef CONFIG_ACPI_NUMA
376 * Parse SRAT to discover nodes.
382 numa_initmem_init(0, end_pfn);
384 contig_initmem_init(0, end_pfn);
387 dma32_reserve_bootmem();
389 #ifdef CONFIG_ACPI_SLEEP
391 * Reserve low memory region for sleep support.
393 acpi_reserve_bootmem();
396 #ifdef CONFIG_X86_MPPARSE
398 * Find and reserve possible boot-time SMP configuration:
402 #ifdef CONFIG_BLK_DEV_INITRD
403 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
404 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
405 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
406 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
407 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
409 if (ramdisk_end <= end_of_mem) {
411 * don't need to reserve again, already reserved early
412 * in x86_64_start_kernel, and early_res_to_bootmem
413 * convert that to reserved in bootmem
415 initrd_start = ramdisk_image + PAGE_OFFSET;
416 initrd_end = initrd_start+ramdisk_size;
418 free_bootmem(ramdisk_image, ramdisk_size);
419 printk(KERN_ERR "initrd extends beyond end of memory "
420 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
421 ramdisk_end, end_of_mem);
426 reserve_crashkernel();
428 reserve_ibft_region();
437 * Read APIC and some other early information from ACPI tables.
444 #ifdef CONFIG_X86_MPPARSE
446 * get boot-time SMP configuration:
448 if (smp_found_config)
451 init_apic_mappings();
452 ioapic_init_mappings();
457 * We trust e820 completely. No explicit ROM probing in memory.
459 e820_reserve_resources();
460 e820_mark_nosave_regions(end_pfn);
462 /* request I/O space for devices used on all i[345]86 PCs */
463 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
464 request_resource(&ioport_resource, &standard_io_resources[i]);
469 #if defined(CONFIG_VGA_CONSOLE)
470 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
471 conswitchp = &vga_con;
472 #elif defined(CONFIG_DUMMY_CONSOLE)
473 conswitchp = &dummy_con;
478 struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
480 static void __cpuinit default_init(struct cpuinfo_x86 *c)
482 display_cacheinfo(c);
485 static struct cpu_dev __cpuinitdata default_cpu = {
486 .c_init = default_init,
487 .c_vendor = "Unknown",
489 static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
491 int __cpuinit get_model_name(struct cpuinfo_x86 *c)
495 if (c->extended_cpuid_level < 0x80000004)
498 v = (unsigned int *) c->x86_model_id;
499 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
500 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
501 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
502 c->x86_model_id[48] = 0;
507 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
509 unsigned int n, dummy, eax, ebx, ecx, edx;
511 n = c->extended_cpuid_level;
513 if (n >= 0x80000005) {
514 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
515 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
516 "D cache %dK (%d bytes/line)\n",
517 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
518 c->x86_cache_size = (ecx>>24) + (edx>>24);
519 /* On K8 L1 TLB is inclusive, so don't count it */
523 if (n >= 0x80000006) {
524 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
525 ecx = cpuid_ecx(0x80000006);
526 c->x86_cache_size = ecx >> 16;
527 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
529 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
530 c->x86_cache_size, ecx & 0xFF);
532 if (n >= 0x80000008) {
533 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
534 c->x86_virt_bits = (eax >> 8) & 0xff;
535 c->x86_phys_bits = eax & 0xff;
539 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
542 u32 eax, ebx, ecx, edx;
543 int index_msb, core_bits;
545 cpuid(1, &eax, &ebx, &ecx, &edx);
548 if (!cpu_has(c, X86_FEATURE_HT))
550 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
553 smp_num_siblings = (ebx & 0xff0000) >> 16;
555 if (smp_num_siblings == 1) {
556 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
557 } else if (smp_num_siblings > 1) {
559 if (smp_num_siblings > NR_CPUS) {
560 printk(KERN_WARNING "CPU: Unsupported number of "
561 "siblings %d", smp_num_siblings);
562 smp_num_siblings = 1;
566 index_msb = get_count_order(smp_num_siblings);
567 c->phys_proc_id = phys_pkg_id(index_msb);
569 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
571 index_msb = get_count_order(smp_num_siblings);
573 core_bits = get_count_order(c->x86_max_cores);
575 c->cpu_core_id = phys_pkg_id(index_msb) &
576 ((1 << core_bits) - 1);
579 if ((c->x86_max_cores * smp_num_siblings) > 1) {
580 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
582 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
589 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
591 char *v = c->x86_vendor_id;
595 for (i = 0; i < X86_VENDOR_NUM; i++) {
597 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
598 (cpu_devs[i]->c_ident[1] &&
599 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
601 this_cpu = cpu_devs[i];
608 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
609 printk(KERN_ERR "CPU: Your system may be unstable.\n");
611 c->x86_vendor = X86_VENDOR_UNKNOWN;
614 static void __init early_cpu_support_print(void)
617 struct cpu_dev *cpu_devx;
619 printk("KERNEL supported cpus:\n");
620 for (i = 0; i < X86_VENDOR_NUM; i++) {
621 cpu_devx = cpu_devs[i];
624 for (j = 0; j < 2; j++) {
625 if (!cpu_devx->c_ident[j])
627 printk(" %s %s\n", cpu_devx->c_vendor,
628 cpu_devx->c_ident[j]);
633 static void __init early_cpu_init(void)
635 struct cpu_vendor_dev *cvdev;
637 for (cvdev = __x86cpuvendor_start ;
638 cvdev < __x86cpuvendor_end ;
640 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
641 early_cpu_support_print();
644 /* Do some early cpuid on the boot CPU to get some parameter that are
645 needed before check_bugs. Everything advanced is in identify_cpu
647 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
651 c->loops_per_jiffy = loops_per_jiffy;
652 c->x86_cache_size = -1;
653 c->x86_vendor = X86_VENDOR_UNKNOWN;
654 c->x86_model = c->x86_mask = 0; /* So far unknown... */
655 c->x86_vendor_id[0] = '\0'; /* Unset */
656 c->x86_model_id[0] = '\0'; /* Unset */
657 c->x86_clflush_size = 64;
658 c->x86_cache_alignment = c->x86_clflush_size;
659 c->x86_max_cores = 1;
660 c->x86_coreid_bits = 0;
661 c->extended_cpuid_level = 0;
662 memset(&c->x86_capability, 0, sizeof c->x86_capability);
664 /* Get vendor name */
665 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
666 (unsigned int *)&c->x86_vendor_id[0],
667 (unsigned int *)&c->x86_vendor_id[8],
668 (unsigned int *)&c->x86_vendor_id[4]);
672 /* Initialize the standard set of capabilities */
673 /* Note that the vendor-specific code below might override */
675 /* Intel-defined flags: level 0x00000001 */
676 if (c->cpuid_level >= 0x00000001) {
678 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
679 &c->x86_capability[0]);
680 c->x86 = (tfms >> 8) & 0xf;
681 c->x86_model = (tfms >> 4) & 0xf;
682 c->x86_mask = tfms & 0xf;
684 c->x86 += (tfms >> 20) & 0xff;
686 c->x86_model += ((tfms >> 16) & 0xF) << 4;
687 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
688 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
690 /* Have CPUID level 0 only - unheard of */
694 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
696 c->phys_proc_id = c->initial_apicid;
698 /* AMD-defined flags: level 0x80000001 */
699 xlvl = cpuid_eax(0x80000000);
700 c->extended_cpuid_level = xlvl;
701 if ((xlvl & 0xffff0000) == 0x80000000) {
702 if (xlvl >= 0x80000001) {
703 c->x86_capability[1] = cpuid_edx(0x80000001);
704 c->x86_capability[6] = cpuid_ecx(0x80000001);
706 if (xlvl >= 0x80000004)
707 get_model_name(c); /* Default name */
710 /* Transmeta-defined flags: level 0x80860001 */
711 xlvl = cpuid_eax(0x80860000);
712 if ((xlvl & 0xffff0000) == 0x80860000) {
713 /* Don't set x86_cpuid_level here for now to not confuse. */
714 if (xlvl >= 0x80860001)
715 c->x86_capability[2] = cpuid_edx(0x80860001);
718 c->extended_cpuid_level = cpuid_eax(0x80000000);
719 if (c->extended_cpuid_level >= 0x80000007)
720 c->x86_power = cpuid_edx(0x80000007);
722 if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
723 cpu_devs[c->x86_vendor]->c_early_init)
724 cpu_devs[c->x86_vendor]->c_early_init(c);
726 validate_pat_support(c);
728 /* early_param could clear that, but recall get it set again */
730 clear_cpu_cap(c, X86_FEATURE_APIC);
734 * This does the hard work of actually picking apart the CPU stuff...
736 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
740 early_identify_cpu(c);
742 init_scattered_cpuid_features(c);
744 c->apicid = phys_pkg_id(0);
747 * Vendor-specific initialization. In this section we
748 * canonicalize the feature flags, meaning if there are
749 * features a certain CPU supports which CPUID doesn't
750 * tell us, CPUID claiming incorrect flags, or other bugs,
751 * we handle them here.
753 * At the end of this section, c->x86_capability better
754 * indicate the features this CPU genuinely supports!
756 if (this_cpu->c_init)
762 * On SMP, boot_cpu_data holds the common feature set between
763 * all CPUs; so make sure that we indicate which features are
764 * common between the CPUs. The first time this routine gets
765 * executed, c == &boot_cpu_data.
767 if (c != &boot_cpu_data) {
768 /* AND the already accumulated flags with these */
769 for (i = 0; i < NCAPINTS; i++)
770 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
773 /* Clear all flags overriden by options */
774 for (i = 0; i < NCAPINTS; i++)
775 c->x86_capability[i] &= ~cleared_cpu_caps[i];
777 #ifdef CONFIG_X86_MCE
780 select_idle_routine(c);
783 numa_add_cpu(smp_processor_id());
788 void __cpuinit identify_boot_cpu(void)
790 identify_cpu(&boot_cpu_data);
793 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
795 BUG_ON(c == &boot_cpu_data);
800 static __init int setup_noclflush(char *arg)
802 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
805 __setup("noclflush", setup_noclflush);
807 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
809 if (c->x86_model_id[0])
810 printk(KERN_CONT "%s", c->x86_model_id);
812 if (c->x86_mask || c->cpuid_level >= 0)
813 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
815 printk(KERN_CONT "\n");
818 static __init int setup_disablecpuid(char *arg)
821 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
822 setup_clear_cpu_cap(bit);
827 __setup("clearcpuid=", setup_disablecpuid);