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>
75 #include <asm/mmconfig.h>
77 #include <mach_apic.h>
78 #ifdef CONFIG_PARAVIRT
79 #include <asm/paravirt.h>
88 struct cpuinfo_x86 boot_cpu_data __read_mostly;
89 EXPORT_SYMBOL(boot_cpu_data);
91 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
93 unsigned long mmu_cr4_features;
95 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
98 unsigned long saved_video_mode;
104 char dmi_alloc_data[DMI_MAX_DATA];
109 struct screen_info screen_info;
110 EXPORT_SYMBOL(screen_info);
111 struct sys_desc_table_struct {
112 unsigned short length;
113 unsigned char table[0];
116 struct edid_info edid_info;
117 EXPORT_SYMBOL_GPL(edid_info);
119 extern int root_mountflags;
121 static char __initdata command_line[COMMAND_LINE_SIZE];
123 static struct resource standard_io_resources[] = {
124 { .name = "dma1", .start = 0x00, .end = 0x1f,
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
126 { .name = "pic1", .start = 0x20, .end = 0x21,
127 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
128 { .name = "timer0", .start = 0x40, .end = 0x43,
129 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
130 { .name = "timer1", .start = 0x50, .end = 0x53,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
132 { .name = "keyboard", .start = 0x60, .end = 0x60,
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
134 { .name = "keyboard", .start = 0x64, .end = 0x64,
135 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
136 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
137 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
138 { .name = "pic2", .start = 0xa0, .end = 0xa1,
139 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
140 { .name = "dma2", .start = 0xc0, .end = 0xdf,
141 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
142 { .name = "fpu", .start = 0xf0, .end = 0xff,
143 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
146 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
148 static struct resource data_resource = {
149 .name = "Kernel data",
152 .flags = IORESOURCE_RAM,
154 static struct resource code_resource = {
155 .name = "Kernel code",
158 .flags = IORESOURCE_RAM,
160 static struct resource bss_resource = {
161 .name = "Kernel bss",
164 .flags = IORESOURCE_RAM,
167 static void __init early_cpu_init(void);
168 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
170 #ifdef CONFIG_PROC_VMCORE
171 /* elfcorehdr= specifies the location of elf core header
172 * stored by the crashed kernel. This option will be passed
173 * by kexec loader to the capture kernel.
175 static int __init setup_elfcorehdr(char *arg)
180 elfcorehdr_addr = memparse(arg, &end);
181 return end > arg ? 0 : -EINVAL;
183 early_param("elfcorehdr", setup_elfcorehdr);
188 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
190 unsigned long bootmap_size, bootmap;
192 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
193 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
196 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
197 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
198 e820_register_active_regions(0, start_pfn, end_pfn);
199 free_bootmem_with_active_regions(0, end_pfn);
200 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
201 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
205 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
207 #ifdef CONFIG_EDD_MODULE
211 * copy_edd() - Copy the BIOS EDD information
212 * from boot_params into a safe place.
215 static inline void copy_edd(void)
217 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
218 sizeof(edd.mbr_signature));
219 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
220 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
221 edd.edd_info_nr = boot_params.eddbuf_entries;
224 static inline void copy_edd(void)
229 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
230 void __attribute__((weak)) __init memory_setup(void)
232 machine_specific_memory_setup();
235 /* Current gdt points %fs at the "master" per-cpu area: after this,
236 * it's on the real one. */
237 void switch_to_new_gdt(void)
239 struct desc_ptr gdt_descr;
241 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
242 gdt_descr.size = GDT_SIZE - 1;
243 load_gdt(&gdt_descr);
247 * setup_arch - architecture-specific boot-time initializations
249 * Note: On x86_64, fixmaps are ready for use even before this is called.
251 void __init setup_arch(char **cmdline_p)
255 printk(KERN_INFO "Command line: %s\n", boot_command_line);
257 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
258 screen_info = boot_params.screen_info;
259 edid_info = boot_params.edid_info;
260 saved_video_mode = boot_params.hdr.vid_mode;
261 bootloader_type = boot_params.hdr.type_of_loader;
263 #ifdef CONFIG_BLK_DEV_RAM
264 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
265 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
266 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
269 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
281 if (!boot_params.hdr.root_flags)
282 root_mountflags &= ~MS_RDONLY;
283 init_mm.start_code = (unsigned long) &_text;
284 init_mm.end_code = (unsigned long) &_etext;
285 init_mm.end_data = (unsigned long) &_edata;
286 init_mm.brk = (unsigned long) &_end;
288 code_resource.start = virt_to_phys(&_text);
289 code_resource.end = virt_to_phys(&_etext)-1;
290 data_resource.start = virt_to_phys(&_etext);
291 data_resource.end = virt_to_phys(&_edata)-1;
292 bss_resource.start = virt_to_phys(&__bss_start);
293 bss_resource.end = virt_to_phys(&__bss_stop)-1;
296 early_identify_cpu(&boot_cpu_data);
298 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
299 *cmdline_p = command_line;
305 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
306 if (init_ohci1394_dma_early)
307 init_ohci1394_dma_on_all_controllers();
310 finish_e820_parsing();
312 /* after parse_early_param, so could debug it */
313 insert_resource(&iomem_resource, &code_resource);
314 insert_resource(&iomem_resource, &data_resource);
315 insert_resource(&iomem_resource, &bss_resource);
317 early_gart_iommu_check();
319 e820_register_active_regions(0, 0, -1UL);
321 * partially used pages are not usable - thus
322 * we are rounding upwards:
324 end_pfn = e820_end_of_ram();
326 /* pre allocte 4k for mptable mpc */
327 early_reserve_e820_mpc_new();
328 /* update e820 for memory not covered by WB MTRRs */
330 if (mtrr_trim_uncached_memory(end_pfn)) {
331 remove_all_active_ranges();
332 e820_register_active_regions(0, 0, -1UL);
333 end_pfn = e820_end_of_ram();
336 num_physpages = end_pfn;
340 max_pfn_mapped = init_memory_mapping(0, (end_pfn << PAGE_SHIFT));
350 #ifdef CONFIG_KVM_CLOCK
356 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
357 * Call this early for SRAT node setup.
359 acpi_boot_table_init();
362 /* How many end-of-memory variables you have, grandma! */
363 max_low_pfn = end_pfn;
365 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
367 /* Remove active ranges so rediscovery with NUMA-awareness happens */
368 remove_all_active_ranges();
370 #ifdef CONFIG_ACPI_NUMA
372 * Parse SRAT to discover nodes.
378 numa_initmem_init(0, end_pfn);
380 contig_initmem_init(0, end_pfn);
383 dma32_reserve_bootmem();
385 #ifdef CONFIG_ACPI_SLEEP
387 * Reserve low memory region for sleep support.
389 acpi_reserve_bootmem();
392 #ifdef CONFIG_X86_MPPARSE
394 * Find and reserve possible boot-time SMP configuration:
398 #ifdef CONFIG_BLK_DEV_INITRD
399 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
400 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
401 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
402 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
403 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
405 if (ramdisk_end <= end_of_mem) {
407 * don't need to reserve again, already reserved early
408 * in x86_64_start_kernel, and early_res_to_bootmem
409 * convert that to reserved in bootmem
411 initrd_start = ramdisk_image + PAGE_OFFSET;
412 initrd_end = initrd_start+ramdisk_size;
414 free_bootmem(ramdisk_image, ramdisk_size);
415 printk(KERN_ERR "initrd extends beyond end of memory "
416 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
417 ramdisk_end, end_of_mem);
422 reserve_crashkernel();
424 reserve_ibft_region();
433 * Read APIC and some other early information from ACPI tables.
440 #ifdef CONFIG_X86_MPPARSE
442 * get boot-time SMP configuration:
444 if (smp_found_config)
447 init_apic_mappings();
448 ioapic_init_mappings();
453 * We trust e820 completely. No explicit ROM probing in memory.
455 e820_reserve_resources();
456 e820_mark_nosave_regions(end_pfn);
458 /* request I/O space for devices used on all i[345]86 PCs */
459 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
460 request_resource(&ioport_resource, &standard_io_resources[i]);
465 #if defined(CONFIG_VGA_CONSOLE)
466 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
467 conswitchp = &vga_con;
468 #elif defined(CONFIG_DUMMY_CONSOLE)
469 conswitchp = &dummy_con;
473 /* do this before identify_cpu for boot cpu */
474 check_enable_amd_mmconf_dmi();
477 struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
479 static void __cpuinit default_init(struct cpuinfo_x86 *c)
481 display_cacheinfo(c);
484 static struct cpu_dev __cpuinitdata default_cpu = {
485 .c_init = default_init,
486 .c_vendor = "Unknown",
488 static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
490 int __cpuinit get_model_name(struct cpuinfo_x86 *c)
494 if (c->extended_cpuid_level < 0x80000004)
497 v = (unsigned int *) c->x86_model_id;
498 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
499 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
500 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
501 c->x86_model_id[48] = 0;
506 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
508 unsigned int n, dummy, eax, ebx, ecx, edx;
510 n = c->extended_cpuid_level;
512 if (n >= 0x80000005) {
513 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
514 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
515 "D cache %dK (%d bytes/line)\n",
516 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
517 c->x86_cache_size = (ecx>>24) + (edx>>24);
518 /* On K8 L1 TLB is inclusive, so don't count it */
522 if (n >= 0x80000006) {
523 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
524 ecx = cpuid_ecx(0x80000006);
525 c->x86_cache_size = ecx >> 16;
526 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
528 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
529 c->x86_cache_size, ecx & 0xFF);
531 if (n >= 0x80000008) {
532 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
533 c->x86_virt_bits = (eax >> 8) & 0xff;
534 c->x86_phys_bits = eax & 0xff;
538 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
541 u32 eax, ebx, ecx, edx;
542 int index_msb, core_bits;
544 cpuid(1, &eax, &ebx, &ecx, &edx);
547 if (!cpu_has(c, X86_FEATURE_HT))
549 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
552 smp_num_siblings = (ebx & 0xff0000) >> 16;
554 if (smp_num_siblings == 1) {
555 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
556 } else if (smp_num_siblings > 1) {
558 if (smp_num_siblings > NR_CPUS) {
559 printk(KERN_WARNING "CPU: Unsupported number of "
560 "siblings %d", smp_num_siblings);
561 smp_num_siblings = 1;
565 index_msb = get_count_order(smp_num_siblings);
566 c->phys_proc_id = phys_pkg_id(index_msb);
568 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
570 index_msb = get_count_order(smp_num_siblings);
572 core_bits = get_count_order(c->x86_max_cores);
574 c->cpu_core_id = phys_pkg_id(index_msb) &
575 ((1 << core_bits) - 1);
578 if ((c->x86_max_cores * smp_num_siblings) > 1) {
579 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
581 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
588 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
590 char *v = c->x86_vendor_id;
594 for (i = 0; i < X86_VENDOR_NUM; i++) {
596 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
597 (cpu_devs[i]->c_ident[1] &&
598 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
600 this_cpu = cpu_devs[i];
607 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
608 printk(KERN_ERR "CPU: Your system may be unstable.\n");
610 c->x86_vendor = X86_VENDOR_UNKNOWN;
613 static void __init early_cpu_support_print(void)
616 struct cpu_dev *cpu_devx;
618 printk("KERNEL supported cpus:\n");
619 for (i = 0; i < X86_VENDOR_NUM; i++) {
620 cpu_devx = cpu_devs[i];
623 for (j = 0; j < 2; j++) {
624 if (!cpu_devx->c_ident[j])
626 printk(" %s %s\n", cpu_devx->c_vendor,
627 cpu_devx->c_ident[j]);
632 static void __init early_cpu_init(void)
634 struct cpu_vendor_dev *cvdev;
636 for (cvdev = __x86cpuvendor_start ;
637 cvdev < __x86cpuvendor_end ;
639 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
640 early_cpu_support_print();
643 /* Do some early cpuid on the boot CPU to get some parameter that are
644 needed before check_bugs. Everything advanced is in identify_cpu
646 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
650 c->loops_per_jiffy = loops_per_jiffy;
651 c->x86_cache_size = -1;
652 c->x86_vendor = X86_VENDOR_UNKNOWN;
653 c->x86_model = c->x86_mask = 0; /* So far unknown... */
654 c->x86_vendor_id[0] = '\0'; /* Unset */
655 c->x86_model_id[0] = '\0'; /* Unset */
656 c->x86_clflush_size = 64;
657 c->x86_cache_alignment = c->x86_clflush_size;
658 c->x86_max_cores = 1;
659 c->x86_coreid_bits = 0;
660 c->extended_cpuid_level = 0;
661 memset(&c->x86_capability, 0, sizeof c->x86_capability);
663 /* Get vendor name */
664 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
665 (unsigned int *)&c->x86_vendor_id[0],
666 (unsigned int *)&c->x86_vendor_id[8],
667 (unsigned int *)&c->x86_vendor_id[4]);
671 /* Initialize the standard set of capabilities */
672 /* Note that the vendor-specific code below might override */
674 /* Intel-defined flags: level 0x00000001 */
675 if (c->cpuid_level >= 0x00000001) {
677 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
678 &c->x86_capability[0]);
679 c->x86 = (tfms >> 8) & 0xf;
680 c->x86_model = (tfms >> 4) & 0xf;
681 c->x86_mask = tfms & 0xf;
683 c->x86 += (tfms >> 20) & 0xff;
685 c->x86_model += ((tfms >> 16) & 0xF) << 4;
686 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
687 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
689 /* Have CPUID level 0 only - unheard of */
693 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
695 c->phys_proc_id = c->initial_apicid;
697 /* AMD-defined flags: level 0x80000001 */
698 xlvl = cpuid_eax(0x80000000);
699 c->extended_cpuid_level = xlvl;
700 if ((xlvl & 0xffff0000) == 0x80000000) {
701 if (xlvl >= 0x80000001) {
702 c->x86_capability[1] = cpuid_edx(0x80000001);
703 c->x86_capability[6] = cpuid_ecx(0x80000001);
705 if (xlvl >= 0x80000004)
706 get_model_name(c); /* Default name */
709 /* Transmeta-defined flags: level 0x80860001 */
710 xlvl = cpuid_eax(0x80860000);
711 if ((xlvl & 0xffff0000) == 0x80860000) {
712 /* Don't set x86_cpuid_level here for now to not confuse. */
713 if (xlvl >= 0x80860001)
714 c->x86_capability[2] = cpuid_edx(0x80860001);
717 c->extended_cpuid_level = cpuid_eax(0x80000000);
718 if (c->extended_cpuid_level >= 0x80000007)
719 c->x86_power = cpuid_edx(0x80000007);
721 if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
722 cpu_devs[c->x86_vendor]->c_early_init)
723 cpu_devs[c->x86_vendor]->c_early_init(c);
725 validate_pat_support(c);
729 * This does the hard work of actually picking apart the CPU stuff...
731 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
735 early_identify_cpu(c);
737 init_scattered_cpuid_features(c);
739 c->apicid = phys_pkg_id(0);
742 * Vendor-specific initialization. In this section we
743 * canonicalize the feature flags, meaning if there are
744 * features a certain CPU supports which CPUID doesn't
745 * tell us, CPUID claiming incorrect flags, or other bugs,
746 * we handle them here.
748 * At the end of this section, c->x86_capability better
749 * indicate the features this CPU genuinely supports!
751 if (this_cpu->c_init)
757 * On SMP, boot_cpu_data holds the common feature set between
758 * all CPUs; so make sure that we indicate which features are
759 * common between the CPUs. The first time this routine gets
760 * executed, c == &boot_cpu_data.
762 if (c != &boot_cpu_data) {
763 /* AND the already accumulated flags with these */
764 for (i = 0; i < NCAPINTS; i++)
765 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
768 /* Clear all flags overriden by options */
769 for (i = 0; i < NCAPINTS; i++)
770 c->x86_capability[i] &= ~cleared_cpu_caps[i];
772 #ifdef CONFIG_X86_MCE
775 select_idle_routine(c);
778 numa_add_cpu(smp_processor_id());
783 void __cpuinit identify_boot_cpu(void)
785 identify_cpu(&boot_cpu_data);
788 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
790 BUG_ON(c == &boot_cpu_data);
795 static __init int setup_noclflush(char *arg)
797 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
800 __setup("noclflush", setup_noclflush);
802 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
804 if (c->x86_model_id[0])
805 printk(KERN_CONT "%s", c->x86_model_id);
807 if (c->x86_mask || c->cpuid_level >= 0)
808 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
810 printk(KERN_CONT "\n");
813 static __init int setup_disablecpuid(char *arg)
816 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
817 setup_clear_cpu_cap(bit);
822 __setup("clearcpuid=", setup_disablecpuid);