x86: use the correct cpuid method to detect MWAIT support for C states
[linux-2.6/kvm.git] / arch / x86 / kernel / setup_64.c
blob4a3f00b49236aac3f105b13163694dcdc1a0165f
1 /*
2 * Copyright (C) 1995 Linus Torvalds
3 */
5 /*
6 * This file handles the architecture-dependent parts of initialization
7 */
9 #include <linux/errno.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/mm.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/a.out.h>
19 #include <linux/screen_info.h>
20 #include <linux/ioport.h>
21 #include <linux/delay.h>
22 #include <linux/init.h>
23 #include <linux/initrd.h>
24 #include <linux/highmem.h>
25 #include <linux/bootmem.h>
26 #include <linux/module.h>
27 #include <asm/processor.h>
28 #include <linux/console.h>
29 #include <linux/seq_file.h>
30 #include <linux/crash_dump.h>
31 #include <linux/root_dev.h>
32 #include <linux/pci.h>
33 #include <linux/efi.h>
34 #include <linux/acpi.h>
35 #include <linux/kallsyms.h>
36 #include <linux/edd.h>
37 #include <linux/mmzone.h>
38 #include <linux/kexec.h>
39 #include <linux/cpufreq.h>
40 #include <linux/dmi.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/ctype.h>
43 #include <linux/uaccess.h>
45 #include <asm/mtrr.h>
46 #include <asm/uaccess.h>
47 #include <asm/system.h>
48 #include <asm/vsyscall.h>
49 #include <asm/io.h>
50 #include <asm/smp.h>
51 #include <asm/msr.h>
52 #include <asm/desc.h>
53 #include <video/edid.h>
54 #include <asm/e820.h>
55 #include <asm/dma.h>
56 #include <asm/gart.h>
57 #include <asm/mpspec.h>
58 #include <asm/mmu_context.h>
59 #include <asm/proto.h>
60 #include <asm/setup.h>
61 #include <asm/mach_apic.h>
62 #include <asm/numa.h>
63 #include <asm/sections.h>
64 #include <asm/dmi.h>
65 #include <asm/cacheflush.h>
66 #include <asm/mce.h>
67 #include <asm/ds.h>
68 #include <asm/topology.h>
70 #ifdef CONFIG_PARAVIRT
71 #include <asm/paravirt.h>
72 #else
73 #define ARCH_SETUP
74 #endif
77 * Machine setup..
80 struct cpuinfo_x86 boot_cpu_data __read_mostly;
81 EXPORT_SYMBOL(boot_cpu_data);
83 unsigned long mmu_cr4_features;
85 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
86 int bootloader_type;
88 unsigned long saved_video_mode;
90 int force_mwait __cpuinitdata;
93 * Early DMI memory
95 int dmi_alloc_index;
96 char dmi_alloc_data[DMI_MAX_DATA];
99 * Setup options
101 struct screen_info screen_info;
102 EXPORT_SYMBOL(screen_info);
103 struct sys_desc_table_struct {
104 unsigned short length;
105 unsigned char table[0];
108 struct edid_info edid_info;
109 EXPORT_SYMBOL_GPL(edid_info);
111 extern int root_mountflags;
113 char __initdata command_line[COMMAND_LINE_SIZE];
115 struct resource standard_io_resources[] = {
116 { .name = "dma1", .start = 0x00, .end = 0x1f,
117 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
118 { .name = "pic1", .start = 0x20, .end = 0x21,
119 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
120 { .name = "timer0", .start = 0x40, .end = 0x43,
121 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
122 { .name = "timer1", .start = 0x50, .end = 0x53,
123 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
124 { .name = "keyboard", .start = 0x60, .end = 0x6f,
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
126 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
127 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
128 { .name = "pic2", .start = 0xa0, .end = 0xa1,
129 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
130 { .name = "dma2", .start = 0xc0, .end = 0xdf,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
132 { .name = "fpu", .start = 0xf0, .end = 0xff,
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
136 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
138 static struct resource data_resource = {
139 .name = "Kernel data",
140 .start = 0,
141 .end = 0,
142 .flags = IORESOURCE_RAM,
144 static struct resource code_resource = {
145 .name = "Kernel code",
146 .start = 0,
147 .end = 0,
148 .flags = IORESOURCE_RAM,
150 static struct resource bss_resource = {
151 .name = "Kernel bss",
152 .start = 0,
153 .end = 0,
154 .flags = IORESOURCE_RAM,
157 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
159 #ifdef CONFIG_PROC_VMCORE
160 /* elfcorehdr= specifies the location of elf core header
161 * stored by the crashed kernel. This option will be passed
162 * by kexec loader to the capture kernel.
164 static int __init setup_elfcorehdr(char *arg)
166 char *end;
167 if (!arg)
168 return -EINVAL;
169 elfcorehdr_addr = memparse(arg, &end);
170 return end > arg ? 0 : -EINVAL;
172 early_param("elfcorehdr", setup_elfcorehdr);
173 #endif
175 #ifndef CONFIG_NUMA
176 static void __init
177 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
179 unsigned long bootmap_size, bootmap;
181 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
182 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
183 if (bootmap == -1L)
184 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
185 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
186 e820_register_active_regions(0, start_pfn, end_pfn);
187 free_bootmem_with_active_regions(0, end_pfn);
188 reserve_bootmem(bootmap, bootmap_size);
190 #endif
192 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
193 struct edd edd;
194 #ifdef CONFIG_EDD_MODULE
195 EXPORT_SYMBOL(edd);
196 #endif
198 * copy_edd() - Copy the BIOS EDD information
199 * from boot_params into a safe place.
202 static inline void copy_edd(void)
204 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
205 sizeof(edd.mbr_signature));
206 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
207 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
208 edd.edd_info_nr = boot_params.eddbuf_entries;
210 #else
211 static inline void copy_edd(void)
214 #endif
216 #ifdef CONFIG_KEXEC
217 static void __init reserve_crashkernel(void)
219 unsigned long long free_mem;
220 unsigned long long crash_size, crash_base;
221 int ret;
223 free_mem =
224 ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
226 ret = parse_crashkernel(boot_command_line, free_mem,
227 &crash_size, &crash_base);
228 if (ret == 0 && crash_size) {
229 if (crash_base > 0) {
230 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
231 "for crashkernel (System RAM: %ldMB)\n",
232 (unsigned long)(crash_size >> 20),
233 (unsigned long)(crash_base >> 20),
234 (unsigned long)(free_mem >> 20));
235 crashk_res.start = crash_base;
236 crashk_res.end = crash_base + crash_size - 1;
237 reserve_bootmem(crash_base, crash_size);
238 } else
239 printk(KERN_INFO "crashkernel reservation failed - "
240 "you have to specify a base address\n");
243 #else
244 static inline void __init reserve_crashkernel(void)
246 #endif
248 #define EBDA_ADDR_POINTER 0x40E
250 unsigned __initdata ebda_addr;
251 unsigned __initdata ebda_size;
253 static void __init discover_ebda(void)
256 * there is a real-mode segmented pointer pointing to the
257 * 4K EBDA area at 0x40E
259 ebda_addr = *(unsigned short *)__va(EBDA_ADDR_POINTER);
261 * There can be some situations, like paravirtualized guests,
262 * in which there is no available ebda information. In such
263 * case, just skip it
265 if (!ebda_addr) {
266 ebda_size = 0;
267 return;
270 ebda_addr <<= 4;
272 ebda_size = *(unsigned short *)__va(ebda_addr);
274 /* Round EBDA up to pages */
275 if (ebda_size == 0)
276 ebda_size = 1;
277 ebda_size <<= 10;
278 ebda_size = round_up(ebda_size + (ebda_addr & ~PAGE_MASK), PAGE_SIZE);
279 if (ebda_size > 64*1024)
280 ebda_size = 64*1024;
283 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
284 void __attribute__((weak)) __init memory_setup(void)
286 machine_specific_memory_setup();
289 void __init setup_arch(char **cmdline_p)
291 unsigned i;
293 printk(KERN_INFO "Command line: %s\n", boot_command_line);
295 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
296 screen_info = boot_params.screen_info;
297 edid_info = boot_params.edid_info;
298 saved_video_mode = boot_params.hdr.vid_mode;
299 bootloader_type = boot_params.hdr.type_of_loader;
301 #ifdef CONFIG_BLK_DEV_RAM
302 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
303 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
304 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
305 #endif
306 #ifdef CONFIG_EFI
307 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
308 "EL64", 4))
309 efi_enabled = 1;
310 #endif
312 ARCH_SETUP
314 memory_setup();
315 copy_edd();
317 if (!boot_params.hdr.root_flags)
318 root_mountflags &= ~MS_RDONLY;
319 init_mm.start_code = (unsigned long) &_text;
320 init_mm.end_code = (unsigned long) &_etext;
321 init_mm.end_data = (unsigned long) &_edata;
322 init_mm.brk = (unsigned long) &_end;
324 code_resource.start = virt_to_phys(&_text);
325 code_resource.end = virt_to_phys(&_etext)-1;
326 data_resource.start = virt_to_phys(&_etext);
327 data_resource.end = virt_to_phys(&_edata)-1;
328 bss_resource.start = virt_to_phys(&__bss_start);
329 bss_resource.end = virt_to_phys(&__bss_stop)-1;
331 early_identify_cpu(&boot_cpu_data);
333 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
334 *cmdline_p = command_line;
336 parse_early_param();
338 finish_e820_parsing();
340 early_gart_iommu_check();
342 e820_register_active_regions(0, 0, -1UL);
344 * partially used pages are not usable - thus
345 * we are rounding upwards:
347 end_pfn = e820_end_of_ram();
348 num_physpages = end_pfn;
350 check_efer();
352 discover_ebda();
354 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
355 if (efi_enabled)
356 efi_init();
358 dmi_scan_machine();
360 io_delay_init();
362 #ifdef CONFIG_SMP
363 /* setup to use the early static init tables during kernel startup */
364 x86_cpu_to_apicid_early_ptr = (void *)&x86_cpu_to_apicid_init;
365 #ifdef CONFIG_NUMA
366 x86_cpu_to_node_map_early_ptr = (void *)&x86_cpu_to_node_map_init;
367 #endif
368 x86_bios_cpu_apicid_early_ptr = (void *)&x86_bios_cpu_apicid_init;
369 #endif
371 #ifdef CONFIG_ACPI
373 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
374 * Call this early for SRAT node setup.
376 acpi_boot_table_init();
377 #endif
379 /* How many end-of-memory variables you have, grandma! */
380 max_low_pfn = end_pfn;
381 max_pfn = end_pfn;
382 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
384 /* Remove active ranges so rediscovery with NUMA-awareness happens */
385 remove_all_active_ranges();
387 #ifdef CONFIG_ACPI_NUMA
389 * Parse SRAT to discover nodes.
391 acpi_numa_init();
392 #endif
394 #ifdef CONFIG_NUMA
395 numa_initmem_init(0, end_pfn);
396 #else
397 contig_initmem_init(0, end_pfn);
398 #endif
400 /* Reserve direct mapping */
401 reserve_bootmem_generic(table_start << PAGE_SHIFT,
402 (table_end - table_start) << PAGE_SHIFT);
404 /* reserve kernel */
405 reserve_bootmem_generic(__pa_symbol(&_text),
406 __pa_symbol(&_end) - __pa_symbol(&_text));
409 * reserve physical page 0 - it's a special BIOS page on many boxes,
410 * enabling clean reboots, SMP operation, laptop functions.
412 reserve_bootmem_generic(0, PAGE_SIZE);
414 /* reserve ebda region */
415 if (ebda_addr)
416 reserve_bootmem_generic(ebda_addr, ebda_size);
417 #ifdef CONFIG_NUMA
418 /* reserve nodemap region */
419 if (nodemap_addr)
420 reserve_bootmem_generic(nodemap_addr, nodemap_size);
421 #endif
423 #ifdef CONFIG_SMP
424 /* Reserve SMP trampoline */
425 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, 2*PAGE_SIZE);
426 #endif
428 #ifdef CONFIG_ACPI_SLEEP
430 * Reserve low memory region for sleep support.
432 acpi_reserve_bootmem();
433 #endif
435 if (efi_enabled) {
436 efi_map_memmap();
437 efi_reserve_bootmem();
441 * Find and reserve possible boot-time SMP configuration:
443 find_smp_config();
444 #ifdef CONFIG_BLK_DEV_INITRD
445 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
446 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
447 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
448 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
449 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
451 if (ramdisk_end <= end_of_mem) {
452 reserve_bootmem_generic(ramdisk_image, ramdisk_size);
453 initrd_start = ramdisk_image + PAGE_OFFSET;
454 initrd_end = initrd_start+ramdisk_size;
455 } else {
456 printk(KERN_ERR "initrd extends beyond end of memory "
457 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
458 ramdisk_end, end_of_mem);
459 initrd_start = 0;
462 #endif
463 reserve_crashkernel();
464 paging_init();
465 map_vsyscall();
467 early_quirks();
470 * set this early, so we dont allocate cpu0
471 * if MADT list doesnt list BSP first
472 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
474 cpu_set(0, cpu_present_map);
475 #ifdef CONFIG_ACPI
477 * Read APIC and some other early information from ACPI tables.
479 acpi_boot_init();
480 #endif
482 init_cpu_to_node();
485 * get boot-time SMP configuration:
487 if (smp_found_config)
488 get_smp_config();
489 init_apic_mappings();
490 ioapic_init_mappings();
493 * We trust e820 completely. No explicit ROM probing in memory.
495 e820_reserve_resources(&code_resource, &data_resource, &bss_resource);
496 e820_mark_nosave_regions();
498 /* request I/O space for devices used on all i[345]86 PCs */
499 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
500 request_resource(&ioport_resource, &standard_io_resources[i]);
502 e820_setup_gap();
504 #ifdef CONFIG_VT
505 #if defined(CONFIG_VGA_CONSOLE)
506 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
507 conswitchp = &vga_con;
508 #elif defined(CONFIG_DUMMY_CONSOLE)
509 conswitchp = &dummy_con;
510 #endif
511 #endif
514 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
516 unsigned int *v;
518 if (c->extended_cpuid_level < 0x80000004)
519 return 0;
521 v = (unsigned int *) c->x86_model_id;
522 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
523 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
524 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
525 c->x86_model_id[48] = 0;
526 return 1;
530 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
532 unsigned int n, dummy, eax, ebx, ecx, edx;
534 n = c->extended_cpuid_level;
536 if (n >= 0x80000005) {
537 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
538 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
539 "D cache %dK (%d bytes/line)\n",
540 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
541 c->x86_cache_size = (ecx>>24) + (edx>>24);
542 /* On K8 L1 TLB is inclusive, so don't count it */
543 c->x86_tlbsize = 0;
546 if (n >= 0x80000006) {
547 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
548 ecx = cpuid_ecx(0x80000006);
549 c->x86_cache_size = ecx >> 16;
550 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
552 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
553 c->x86_cache_size, ecx & 0xFF);
555 if (n >= 0x80000008) {
556 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
557 c->x86_virt_bits = (eax >> 8) & 0xff;
558 c->x86_phys_bits = eax & 0xff;
562 #ifdef CONFIG_NUMA
563 static int nearby_node(int apicid)
565 int i, node;
567 for (i = apicid - 1; i >= 0; i--) {
568 node = apicid_to_node[i];
569 if (node != NUMA_NO_NODE && node_online(node))
570 return node;
572 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
573 node = apicid_to_node[i];
574 if (node != NUMA_NO_NODE && node_online(node))
575 return node;
577 return first_node(node_online_map); /* Shouldn't happen */
579 #endif
582 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
583 * Assumes number of cores is a power of two.
585 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
587 #ifdef CONFIG_SMP
588 unsigned bits;
589 #ifdef CONFIG_NUMA
590 int cpu = smp_processor_id();
591 int node = 0;
592 unsigned apicid = hard_smp_processor_id();
593 #endif
594 bits = c->x86_coreid_bits;
596 /* Low order bits define the core id (index of core in socket) */
597 c->cpu_core_id = c->phys_proc_id & ((1 << bits)-1);
598 /* Convert the APIC ID into the socket ID */
599 c->phys_proc_id = phys_pkg_id(bits);
601 #ifdef CONFIG_NUMA
602 node = c->phys_proc_id;
603 if (apicid_to_node[apicid] != NUMA_NO_NODE)
604 node = apicid_to_node[apicid];
605 if (!node_online(node)) {
606 /* Two possibilities here:
607 - The CPU is missing memory and no node was created.
608 In that case try picking one from a nearby CPU
609 - The APIC IDs differ from the HyperTransport node IDs
610 which the K8 northbridge parsing fills in.
611 Assume they are all increased by a constant offset,
612 but in the same order as the HT nodeids.
613 If that doesn't result in a usable node fall back to the
614 path for the previous case. */
616 int ht_nodeid = apicid - (cpu_data(0).phys_proc_id << bits);
618 if (ht_nodeid >= 0 &&
619 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
620 node = apicid_to_node[ht_nodeid];
621 /* Pick a nearby node */
622 if (!node_online(node))
623 node = nearby_node(apicid);
625 numa_set_node(cpu, node);
627 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
628 #endif
629 #endif
632 static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
634 #ifdef CONFIG_SMP
635 unsigned bits, ecx;
637 /* Multi core CPU? */
638 if (c->extended_cpuid_level < 0x80000008)
639 return;
641 ecx = cpuid_ecx(0x80000008);
643 c->x86_max_cores = (ecx & 0xff) + 1;
645 /* CPU telling us the core id bits shift? */
646 bits = (ecx >> 12) & 0xF;
648 /* Otherwise recompute */
649 if (bits == 0) {
650 while ((1 << bits) < c->x86_max_cores)
651 bits++;
654 c->x86_coreid_bits = bits;
656 #endif
659 #define ENABLE_C1E_MASK 0x18000000
660 #define CPUID_PROCESSOR_SIGNATURE 1
661 #define CPUID_XFAM 0x0ff00000
662 #define CPUID_XFAM_K8 0x00000000
663 #define CPUID_XFAM_10H 0x00100000
664 #define CPUID_XFAM_11H 0x00200000
665 #define CPUID_XMOD 0x000f0000
666 #define CPUID_XMOD_REV_F 0x00040000
668 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
669 static __cpuinit int amd_apic_timer_broken(void)
671 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
673 switch (eax & CPUID_XFAM) {
674 case CPUID_XFAM_K8:
675 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
676 break;
677 case CPUID_XFAM_10H:
678 case CPUID_XFAM_11H:
679 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
680 if (lo & ENABLE_C1E_MASK)
681 return 1;
682 break;
683 default:
684 /* err on the side of caution */
685 return 1;
687 return 0;
690 static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
692 early_init_amd_mc(c);
694 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
695 if (c->x86_power & (1<<8))
696 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
699 static void __cpuinit init_amd(struct cpuinfo_x86 *c)
701 unsigned level;
703 #ifdef CONFIG_SMP
704 unsigned long value;
707 * Disable TLB flush filter by setting HWCR.FFDIS on K8
708 * bit 6 of msr C001_0015
710 * Errata 63 for SH-B3 steppings
711 * Errata 122 for all steppings (F+ have it disabled by default)
713 if (c->x86 == 15) {
714 rdmsrl(MSR_K8_HWCR, value);
715 value |= 1 << 6;
716 wrmsrl(MSR_K8_HWCR, value);
718 #endif
720 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
721 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
722 clear_bit(0*32+31, (unsigned long *)&c->x86_capability);
724 /* On C+ stepping K8 rep microcode works well for copy/memset */
725 level = cpuid_eax(1);
726 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
727 level >= 0x0f58))
728 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
729 if (c->x86 == 0x10 || c->x86 == 0x11)
730 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
732 /* Enable workaround for FXSAVE leak */
733 if (c->x86 >= 6)
734 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
736 level = get_model_name(c);
737 if (!level) {
738 switch (c->x86) {
739 case 15:
740 /* Should distinguish Models here, but this is only
741 a fallback anyways. */
742 strcpy(c->x86_model_id, "Hammer");
743 break;
746 display_cacheinfo(c);
748 /* Multi core CPU? */
749 if (c->extended_cpuid_level >= 0x80000008)
750 amd_detect_cmp(c);
752 if (c->extended_cpuid_level >= 0x80000006 &&
753 (cpuid_edx(0x80000006) & 0xf000))
754 num_cache_leaves = 4;
755 else
756 num_cache_leaves = 3;
758 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
759 set_cpu_cap(c, X86_FEATURE_K8);
761 /* MFENCE stops RDTSC speculation */
762 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
764 if (amd_apic_timer_broken())
765 disable_apic_timer = 1;
768 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
770 #ifdef CONFIG_SMP
771 u32 eax, ebx, ecx, edx;
772 int index_msb, core_bits;
774 cpuid(1, &eax, &ebx, &ecx, &edx);
777 if (!cpu_has(c, X86_FEATURE_HT))
778 return;
779 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
780 goto out;
782 smp_num_siblings = (ebx & 0xff0000) >> 16;
784 if (smp_num_siblings == 1) {
785 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
786 } else if (smp_num_siblings > 1) {
788 if (smp_num_siblings > NR_CPUS) {
789 printk(KERN_WARNING "CPU: Unsupported number of "
790 "siblings %d", smp_num_siblings);
791 smp_num_siblings = 1;
792 return;
795 index_msb = get_count_order(smp_num_siblings);
796 c->phys_proc_id = phys_pkg_id(index_msb);
798 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
800 index_msb = get_count_order(smp_num_siblings);
802 core_bits = get_count_order(c->x86_max_cores);
804 c->cpu_core_id = phys_pkg_id(index_msb) &
805 ((1 << core_bits) - 1);
807 out:
808 if ((c->x86_max_cores * smp_num_siblings) > 1) {
809 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
810 c->phys_proc_id);
811 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
812 c->cpu_core_id);
815 #endif
819 * find out the number of processor cores on the die
821 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
823 unsigned int eax, t;
825 if (c->cpuid_level < 4)
826 return 1;
828 cpuid_count(4, 0, &eax, &t, &t, &t);
830 if (eax & 0x1f)
831 return ((eax >> 26) + 1);
832 else
833 return 1;
836 static void srat_detect_node(void)
838 #ifdef CONFIG_NUMA
839 unsigned node;
840 int cpu = smp_processor_id();
841 int apicid = hard_smp_processor_id();
843 /* Don't do the funky fallback heuristics the AMD version employs
844 for now. */
845 node = apicid_to_node[apicid];
846 if (node == NUMA_NO_NODE)
847 node = first_node(node_online_map);
848 numa_set_node(cpu, node);
850 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
851 #endif
854 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
856 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
857 (c->x86 == 0x6 && c->x86_model >= 0x0e))
858 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
861 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
863 /* Cache sizes */
864 unsigned n;
866 init_intel_cacheinfo(c);
867 if (c->cpuid_level > 9) {
868 unsigned eax = cpuid_eax(10);
869 /* Check for version and the number of counters */
870 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
871 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
874 if (cpu_has_ds) {
875 unsigned int l1, l2;
876 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
877 if (!(l1 & (1<<11)))
878 set_cpu_cap(c, X86_FEATURE_BTS);
879 if (!(l1 & (1<<12)))
880 set_cpu_cap(c, X86_FEATURE_PEBS);
884 if (cpu_has_bts)
885 ds_init_intel(c);
887 n = c->extended_cpuid_level;
888 if (n >= 0x80000008) {
889 unsigned eax = cpuid_eax(0x80000008);
890 c->x86_virt_bits = (eax >> 8) & 0xff;
891 c->x86_phys_bits = eax & 0xff;
892 /* CPUID workaround for Intel 0F34 CPU */
893 if (c->x86_vendor == X86_VENDOR_INTEL &&
894 c->x86 == 0xF && c->x86_model == 0x3 &&
895 c->x86_mask == 0x4)
896 c->x86_phys_bits = 36;
899 if (c->x86 == 15)
900 c->x86_cache_alignment = c->x86_clflush_size * 2;
901 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
902 (c->x86 == 0x6 && c->x86_model >= 0x0e))
903 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
904 if (c->x86 == 6)
905 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
906 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
907 c->x86_max_cores = intel_num_cpu_cores(c);
909 srat_detect_node();
912 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
914 char *v = c->x86_vendor_id;
916 if (!strcmp(v, "AuthenticAMD"))
917 c->x86_vendor = X86_VENDOR_AMD;
918 else if (!strcmp(v, "GenuineIntel"))
919 c->x86_vendor = X86_VENDOR_INTEL;
920 else
921 c->x86_vendor = X86_VENDOR_UNKNOWN;
924 struct cpu_model_info {
925 int vendor;
926 int family;
927 char *model_names[16];
930 /* Do some early cpuid on the boot CPU to get some parameter that are
931 needed before check_bugs. Everything advanced is in identify_cpu
932 below. */
933 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
935 u32 tfms, xlvl;
937 c->loops_per_jiffy = loops_per_jiffy;
938 c->x86_cache_size = -1;
939 c->x86_vendor = X86_VENDOR_UNKNOWN;
940 c->x86_model = c->x86_mask = 0; /* So far unknown... */
941 c->x86_vendor_id[0] = '\0'; /* Unset */
942 c->x86_model_id[0] = '\0'; /* Unset */
943 c->x86_clflush_size = 64;
944 c->x86_cache_alignment = c->x86_clflush_size;
945 c->x86_max_cores = 1;
946 c->x86_coreid_bits = 0;
947 c->extended_cpuid_level = 0;
948 memset(&c->x86_capability, 0, sizeof c->x86_capability);
950 /* Get vendor name */
951 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
952 (unsigned int *)&c->x86_vendor_id[0],
953 (unsigned int *)&c->x86_vendor_id[8],
954 (unsigned int *)&c->x86_vendor_id[4]);
956 get_cpu_vendor(c);
958 /* Initialize the standard set of capabilities */
959 /* Note that the vendor-specific code below might override */
961 /* Intel-defined flags: level 0x00000001 */
962 if (c->cpuid_level >= 0x00000001) {
963 __u32 misc;
964 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
965 &c->x86_capability[0]);
966 c->x86 = (tfms >> 8) & 0xf;
967 c->x86_model = (tfms >> 4) & 0xf;
968 c->x86_mask = tfms & 0xf;
969 if (c->x86 == 0xf)
970 c->x86 += (tfms >> 20) & 0xff;
971 if (c->x86 >= 0x6)
972 c->x86_model += ((tfms >> 16) & 0xF) << 4;
973 if (c->x86_capability[0] & (1<<19))
974 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
975 } else {
976 /* Have CPUID level 0 only - unheard of */
977 c->x86 = 4;
980 #ifdef CONFIG_SMP
981 c->phys_proc_id = (cpuid_ebx(1) >> 24) & 0xff;
982 #endif
983 /* AMD-defined flags: level 0x80000001 */
984 xlvl = cpuid_eax(0x80000000);
985 c->extended_cpuid_level = xlvl;
986 if ((xlvl & 0xffff0000) == 0x80000000) {
987 if (xlvl >= 0x80000001) {
988 c->x86_capability[1] = cpuid_edx(0x80000001);
989 c->x86_capability[6] = cpuid_ecx(0x80000001);
991 if (xlvl >= 0x80000004)
992 get_model_name(c); /* Default name */
995 /* Transmeta-defined flags: level 0x80860001 */
996 xlvl = cpuid_eax(0x80860000);
997 if ((xlvl & 0xffff0000) == 0x80860000) {
998 /* Don't set x86_cpuid_level here for now to not confuse. */
999 if (xlvl >= 0x80860001)
1000 c->x86_capability[2] = cpuid_edx(0x80860001);
1003 c->extended_cpuid_level = cpuid_eax(0x80000000);
1004 if (c->extended_cpuid_level >= 0x80000007)
1005 c->x86_power = cpuid_edx(0x80000007);
1007 switch (c->x86_vendor) {
1008 case X86_VENDOR_AMD:
1009 early_init_amd(c);
1010 break;
1016 * This does the hard work of actually picking apart the CPU stuff...
1018 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1020 int i;
1022 early_identify_cpu(c);
1024 init_scattered_cpuid_features(c);
1026 c->apicid = phys_pkg_id(0);
1029 * Vendor-specific initialization. In this section we
1030 * canonicalize the feature flags, meaning if there are
1031 * features a certain CPU supports which CPUID doesn't
1032 * tell us, CPUID claiming incorrect flags, or other bugs,
1033 * we handle them here.
1035 * At the end of this section, c->x86_capability better
1036 * indicate the features this CPU genuinely supports!
1038 switch (c->x86_vendor) {
1039 case X86_VENDOR_AMD:
1040 init_amd(c);
1041 break;
1043 case X86_VENDOR_INTEL:
1044 init_intel(c);
1045 break;
1047 case X86_VENDOR_UNKNOWN:
1048 default:
1049 display_cacheinfo(c);
1050 break;
1053 select_idle_routine(c);
1054 detect_ht(c);
1057 * On SMP, boot_cpu_data holds the common feature set between
1058 * all CPUs; so make sure that we indicate which features are
1059 * common between the CPUs. The first time this routine gets
1060 * executed, c == &boot_cpu_data.
1062 if (c != &boot_cpu_data) {
1063 /* AND the already accumulated flags with these */
1064 for (i = 0; i < NCAPINTS; i++)
1065 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1068 #ifdef CONFIG_X86_MCE
1069 mcheck_init(c);
1070 #endif
1071 if (c != &boot_cpu_data)
1072 mtrr_ap_init();
1073 #ifdef CONFIG_NUMA
1074 numa_add_cpu(smp_processor_id());
1075 #endif
1077 switch (c->x86_vendor) {
1078 case X86_VENDOR_AMD:
1079 early_init_amd(c);
1080 break;
1081 case X86_VENDOR_INTEL:
1082 early_init_intel(c);
1083 break;
1087 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1089 if (c->x86_model_id[0])
1090 printk(KERN_INFO "%s", c->x86_model_id);
1092 if (c->x86_mask || c->cpuid_level >= 0)
1093 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1094 else
1095 printk(KERN_CONT "\n");
1099 * Get CPU information for use by the procfs.
1102 static int show_cpuinfo(struct seq_file *m, void *v)
1104 struct cpuinfo_x86 *c = v;
1105 int cpu = 0, i;
1108 * These flag bits must match the definitions in <asm/cpufeature.h>.
1109 * NULL means this bit is undefined or reserved; either way it doesn't
1110 * have meaning as far as Linux is concerned. Note that it's important
1111 * to realize there is a difference between this table and CPUID -- if
1112 * applications want to get the raw CPUID data, they should access
1113 * /dev/cpu/<cpu_nr>/cpuid instead.
1115 static const char *const x86_cap_flags[] = {
1116 /* Intel-defined */
1117 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1118 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1119 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1120 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
1122 /* AMD-defined */
1123 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1124 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1125 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1126 NULL, "fxsr_opt", "pdpe1gb", "rdtscp", NULL, "lm",
1127 "3dnowext", "3dnow",
1129 /* Transmeta-defined */
1130 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1131 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1132 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1133 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1135 /* Other (Linux-defined) */
1136 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
1137 NULL, NULL, NULL, NULL,
1138 "constant_tsc", "up", NULL, "arch_perfmon",
1139 "pebs", "bts", NULL, "sync_rdtsc",
1140 "rep_good", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1141 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1143 /* Intel-defined (#2) */
1144 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
1145 "tm2", "ssse3", "cid", NULL, NULL, "cx16", "xtpr", NULL,
1146 NULL, NULL, "dca", "sse4_1", "sse4_2", NULL, NULL, "popcnt",
1147 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1149 /* VIA/Cyrix/Centaur-defined */
1150 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1151 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL, NULL,
1152 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1153 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1155 /* AMD-defined (#2) */
1156 "lahf_lm", "cmp_legacy", "svm", "extapic",
1157 "cr8_legacy", "abm", "sse4a", "misalignsse",
1158 "3dnowprefetch", "osvw", "ibs", "sse5",
1159 "skinit", "wdt", NULL, NULL,
1160 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1161 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1163 /* Auxiliary (Linux-defined) */
1164 "ida", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1165 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1166 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1167 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1169 static const char *const x86_power_flags[] = {
1170 "ts", /* temperature sensor */
1171 "fid", /* frequency id control */
1172 "vid", /* voltage id control */
1173 "ttp", /* thermal trip */
1174 "tm",
1175 "stc",
1176 "100mhzsteps",
1177 "hwpstate",
1178 "", /* tsc invariant mapped to constant_tsc */
1179 /* nothing */
1183 #ifdef CONFIG_SMP
1184 cpu = c->cpu_index;
1185 #endif
1187 seq_printf(m, "processor\t: %u\n"
1188 "vendor_id\t: %s\n"
1189 "cpu family\t: %d\n"
1190 "model\t\t: %d\n"
1191 "model name\t: %s\n",
1192 (unsigned)cpu,
1193 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1194 c->x86,
1195 (int)c->x86_model,
1196 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1198 if (c->x86_mask || c->cpuid_level >= 0)
1199 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1200 else
1201 seq_printf(m, "stepping\t: unknown\n");
1203 if (cpu_has(c, X86_FEATURE_TSC)) {
1204 unsigned int freq = cpufreq_quick_get((unsigned)cpu);
1206 if (!freq)
1207 freq = cpu_khz;
1208 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1209 freq / 1000, (freq % 1000));
1212 /* Cache size */
1213 if (c->x86_cache_size >= 0)
1214 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1216 #ifdef CONFIG_SMP
1217 if (smp_num_siblings * c->x86_max_cores > 1) {
1218 seq_printf(m, "physical id\t: %d\n", c->phys_proc_id);
1219 seq_printf(m, "siblings\t: %d\n",
1220 cpus_weight(per_cpu(cpu_core_map, cpu)));
1221 seq_printf(m, "core id\t\t: %d\n", c->cpu_core_id);
1222 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1224 #endif
1226 seq_printf(m,
1227 "fpu\t\t: yes\n"
1228 "fpu_exception\t: yes\n"
1229 "cpuid level\t: %d\n"
1230 "wp\t\t: yes\n"
1231 "flags\t\t:",
1232 c->cpuid_level);
1234 for (i = 0; i < 32*NCAPINTS; i++)
1235 if (cpu_has(c, i) && x86_cap_flags[i] != NULL)
1236 seq_printf(m, " %s", x86_cap_flags[i]);
1238 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1239 c->loops_per_jiffy/(500000/HZ),
1240 (c->loops_per_jiffy/(5000/HZ)) % 100);
1242 if (c->x86_tlbsize > 0)
1243 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1244 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1245 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1247 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1248 c->x86_phys_bits, c->x86_virt_bits);
1250 seq_printf(m, "power management:");
1251 for (i = 0; i < 32; i++) {
1252 if (c->x86_power & (1 << i)) {
1253 if (i < ARRAY_SIZE(x86_power_flags) &&
1254 x86_power_flags[i])
1255 seq_printf(m, "%s%s",
1256 x86_power_flags[i][0]?" ":"",
1257 x86_power_flags[i]);
1258 else
1259 seq_printf(m, " [%d]", i);
1263 seq_printf(m, "\n\n");
1265 return 0;
1268 static void *c_start(struct seq_file *m, loff_t *pos)
1270 if (*pos == 0) /* just in case, cpu 0 is not the first */
1271 *pos = first_cpu(cpu_online_map);
1272 if ((*pos) < NR_CPUS && cpu_online(*pos))
1273 return &cpu_data(*pos);
1274 return NULL;
1277 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1279 *pos = next_cpu(*pos, cpu_online_map);
1280 return c_start(m, pos);
1283 static void c_stop(struct seq_file *m, void *v)
1287 struct seq_operations cpuinfo_op = {
1288 .start = c_start,
1289 .next = c_next,
1290 .stop = c_stop,
1291 .show = show_cpuinfo,