12948316e6a627adc19a51bbd7455f6f75e5b988
[deliverable/linux.git] / arch / x86 / kernel / setup_64.c
1 /*
2 * Copyright (C) 1995 Linus Torvalds
3 */
4
5 /*
6 * This file handles the architecture-dependent parts of initialization
7 */
8
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>
44
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>
69
70 #ifdef CONFIG_PARAVIRT
71 #include <asm/paravirt.h>
72 #else
73 #define ARCH_SETUP
74 #endif
75
76 /*
77 * Machine setup..
78 */
79
80 struct cpuinfo_x86 boot_cpu_data __read_mostly;
81 EXPORT_SYMBOL(boot_cpu_data);
82
83 unsigned long mmu_cr4_features;
84
85 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
86 int bootloader_type;
87
88 unsigned long saved_video_mode;
89
90 int force_mwait __cpuinitdata;
91
92 /*
93 * Early DMI memory
94 */
95 int dmi_alloc_index;
96 char dmi_alloc_data[DMI_MAX_DATA];
97
98 /*
99 * Setup options
100 */
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];
106 };
107
108 struct edid_info edid_info;
109 EXPORT_SYMBOL_GPL(edid_info);
110
111 extern int root_mountflags;
112
113 char __initdata command_line[COMMAND_LINE_SIZE];
114
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 }
134 };
135
136 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
137
138 static struct resource data_resource = {
139 .name = "Kernel data",
140 .start = 0,
141 .end = 0,
142 .flags = IORESOURCE_RAM,
143 };
144 static struct resource code_resource = {
145 .name = "Kernel code",
146 .start = 0,
147 .end = 0,
148 .flags = IORESOURCE_RAM,
149 };
150 static struct resource bss_resource = {
151 .name = "Kernel bss",
152 .start = 0,
153 .end = 0,
154 .flags = IORESOURCE_RAM,
155 };
156
157 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
158
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.
163 */
164 static int __init setup_elfcorehdr(char *arg)
165 {
166 char *end;
167 if (!arg)
168 return -EINVAL;
169 elfcorehdr_addr = memparse(arg, &end);
170 return end > arg ? 0 : -EINVAL;
171 }
172 early_param("elfcorehdr", setup_elfcorehdr);
173 #endif
174
175 #ifndef CONFIG_NUMA
176 static void __init
177 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
178 {
179 unsigned long bootmap_size, bootmap;
180
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);
189 }
190 #endif
191
192 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
193 struct edd edd;
194 #ifdef CONFIG_EDD_MODULE
195 EXPORT_SYMBOL(edd);
196 #endif
197 /**
198 * copy_edd() - Copy the BIOS EDD information
199 * from boot_params into a safe place.
200 *
201 */
202 static inline void copy_edd(void)
203 {
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;
209 }
210 #else
211 static inline void copy_edd(void)
212 {
213 }
214 #endif
215
216 #ifdef CONFIG_KEXEC
217 static void __init reserve_crashkernel(void)
218 {
219 unsigned long long free_mem;
220 unsigned long long crash_size, crash_base;
221 int ret;
222
223 free_mem =
224 ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
225
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");
241 }
242 }
243 #else
244 static inline void __init reserve_crashkernel(void)
245 {}
246 #endif
247
248 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
249 void __attribute__((weak)) __init memory_setup(void)
250 {
251 machine_specific_memory_setup();
252 }
253
254 void __init setup_arch(char **cmdline_p)
255 {
256 unsigned i;
257
258 printk(KERN_INFO "Command line: %s\n", boot_command_line);
259
260 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
261 screen_info = boot_params.screen_info;
262 edid_info = boot_params.edid_info;
263 saved_video_mode = boot_params.hdr.vid_mode;
264 bootloader_type = boot_params.hdr.type_of_loader;
265
266 #ifdef CONFIG_BLK_DEV_RAM
267 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
268 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
269 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
270 #endif
271 #ifdef CONFIG_EFI
272 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
273 "EL64", 4))
274 efi_enabled = 1;
275 #endif
276
277 ARCH_SETUP
278
279 memory_setup();
280 copy_edd();
281
282 if (!boot_params.hdr.root_flags)
283 root_mountflags &= ~MS_RDONLY;
284 init_mm.start_code = (unsigned long) &_text;
285 init_mm.end_code = (unsigned long) &_etext;
286 init_mm.end_data = (unsigned long) &_edata;
287 init_mm.brk = (unsigned long) &_end;
288
289 code_resource.start = virt_to_phys(&_text);
290 code_resource.end = virt_to_phys(&_etext)-1;
291 data_resource.start = virt_to_phys(&_etext);
292 data_resource.end = virt_to_phys(&_edata)-1;
293 bss_resource.start = virt_to_phys(&__bss_start);
294 bss_resource.end = virt_to_phys(&__bss_stop)-1;
295
296 early_identify_cpu(&boot_cpu_data);
297
298 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
299 *cmdline_p = command_line;
300
301 parse_early_param();
302
303 finish_e820_parsing();
304
305 early_gart_iommu_check();
306
307 e820_register_active_regions(0, 0, -1UL);
308 /*
309 * partially used pages are not usable - thus
310 * we are rounding upwards:
311 */
312 end_pfn = e820_end_of_ram();
313 /* update e820 for memory not covered by WB MTRRs */
314 mtrr_bp_init();
315 if (mtrr_trim_uncached_memory(end_pfn)) {
316 e820_register_active_regions(0, 0, -1UL);
317 end_pfn = e820_end_of_ram();
318 }
319
320 num_physpages = end_pfn;
321
322 check_efer();
323
324 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
325 if (efi_enabled)
326 efi_init();
327
328 dmi_scan_machine();
329
330 io_delay_init();
331
332 #ifdef CONFIG_SMP
333 /* setup to use the early static init tables during kernel startup */
334 x86_cpu_to_apicid_early_ptr = (void *)&x86_cpu_to_apicid_init;
335 #ifdef CONFIG_NUMA
336 x86_cpu_to_node_map_early_ptr = (void *)&x86_cpu_to_node_map_init;
337 #endif
338 x86_bios_cpu_apicid_early_ptr = (void *)&x86_bios_cpu_apicid_init;
339 #endif
340
341 #ifdef CONFIG_ACPI
342 /*
343 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
344 * Call this early for SRAT node setup.
345 */
346 acpi_boot_table_init();
347 #endif
348
349 /* How many end-of-memory variables you have, grandma! */
350 max_low_pfn = end_pfn;
351 max_pfn = end_pfn;
352 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
353
354 /* Remove active ranges so rediscovery with NUMA-awareness happens */
355 remove_all_active_ranges();
356
357 #ifdef CONFIG_ACPI_NUMA
358 /*
359 * Parse SRAT to discover nodes.
360 */
361 acpi_numa_init();
362 #endif
363
364 #ifdef CONFIG_NUMA
365 numa_initmem_init(0, end_pfn);
366 #else
367 contig_initmem_init(0, end_pfn);
368 #endif
369
370 early_res_to_bootmem();
371
372 #ifdef CONFIG_ACPI_SLEEP
373 /*
374 * Reserve low memory region for sleep support.
375 */
376 acpi_reserve_bootmem();
377 #endif
378
379 if (efi_enabled) {
380 efi_map_memmap();
381 efi_reserve_bootmem();
382 }
383
384 /*
385 * Find and reserve possible boot-time SMP configuration:
386 */
387 find_smp_config();
388 #ifdef CONFIG_BLK_DEV_INITRD
389 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
390 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
391 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
392 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
393 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
394
395 if (ramdisk_end <= end_of_mem) {
396 reserve_bootmem_generic(ramdisk_image, ramdisk_size);
397 initrd_start = ramdisk_image + PAGE_OFFSET;
398 initrd_end = initrd_start+ramdisk_size;
399 } else {
400 /* Assumes everything on node 0 */
401 free_bootmem(ramdisk_image, ramdisk_size);
402 printk(KERN_ERR "initrd extends beyond end of memory "
403 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
404 ramdisk_end, end_of_mem);
405 initrd_start = 0;
406 }
407 }
408 #endif
409 reserve_crashkernel();
410 paging_init();
411 map_vsyscall();
412
413 early_quirks();
414
415 /*
416 * set this early, so we dont allocate cpu0
417 * if MADT list doesnt list BSP first
418 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
419 */
420 cpu_set(0, cpu_present_map);
421 #ifdef CONFIG_ACPI
422 /*
423 * Read APIC and some other early information from ACPI tables.
424 */
425 acpi_boot_init();
426 #endif
427
428 init_cpu_to_node();
429
430 /*
431 * get boot-time SMP configuration:
432 */
433 if (smp_found_config)
434 get_smp_config();
435 init_apic_mappings();
436 ioapic_init_mappings();
437
438 /*
439 * We trust e820 completely. No explicit ROM probing in memory.
440 */
441 e820_reserve_resources(&code_resource, &data_resource, &bss_resource);
442 e820_mark_nosave_regions();
443
444 /* request I/O space for devices used on all i[345]86 PCs */
445 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
446 request_resource(&ioport_resource, &standard_io_resources[i]);
447
448 e820_setup_gap();
449
450 #ifdef CONFIG_VT
451 #if defined(CONFIG_VGA_CONSOLE)
452 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
453 conswitchp = &vga_con;
454 #elif defined(CONFIG_DUMMY_CONSOLE)
455 conswitchp = &dummy_con;
456 #endif
457 #endif
458 }
459
460 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
461 {
462 unsigned int *v;
463
464 if (c->extended_cpuid_level < 0x80000004)
465 return 0;
466
467 v = (unsigned int *) c->x86_model_id;
468 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
469 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
470 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
471 c->x86_model_id[48] = 0;
472 return 1;
473 }
474
475
476 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
477 {
478 unsigned int n, dummy, eax, ebx, ecx, edx;
479
480 n = c->extended_cpuid_level;
481
482 if (n >= 0x80000005) {
483 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
484 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
485 "D cache %dK (%d bytes/line)\n",
486 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
487 c->x86_cache_size = (ecx>>24) + (edx>>24);
488 /* On K8 L1 TLB is inclusive, so don't count it */
489 c->x86_tlbsize = 0;
490 }
491
492 if (n >= 0x80000006) {
493 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
494 ecx = cpuid_ecx(0x80000006);
495 c->x86_cache_size = ecx >> 16;
496 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
497
498 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
499 c->x86_cache_size, ecx & 0xFF);
500 }
501 if (n >= 0x80000008) {
502 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
503 c->x86_virt_bits = (eax >> 8) & 0xff;
504 c->x86_phys_bits = eax & 0xff;
505 }
506 }
507
508 #ifdef CONFIG_NUMA
509 static int nearby_node(int apicid)
510 {
511 int i, node;
512
513 for (i = apicid - 1; i >= 0; i--) {
514 node = apicid_to_node[i];
515 if (node != NUMA_NO_NODE && node_online(node))
516 return node;
517 }
518 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
519 node = apicid_to_node[i];
520 if (node != NUMA_NO_NODE && node_online(node))
521 return node;
522 }
523 return first_node(node_online_map); /* Shouldn't happen */
524 }
525 #endif
526
527 /*
528 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
529 * Assumes number of cores is a power of two.
530 */
531 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
532 {
533 #ifdef CONFIG_SMP
534 unsigned bits;
535 #ifdef CONFIG_NUMA
536 int cpu = smp_processor_id();
537 int node = 0;
538 unsigned apicid = hard_smp_processor_id();
539 #endif
540 bits = c->x86_coreid_bits;
541
542 /* Low order bits define the core id (index of core in socket) */
543 c->cpu_core_id = c->phys_proc_id & ((1 << bits)-1);
544 /* Convert the APIC ID into the socket ID */
545 c->phys_proc_id = phys_pkg_id(bits);
546
547 #ifdef CONFIG_NUMA
548 node = c->phys_proc_id;
549 if (apicid_to_node[apicid] != NUMA_NO_NODE)
550 node = apicid_to_node[apicid];
551 if (!node_online(node)) {
552 /* Two possibilities here:
553 - The CPU is missing memory and no node was created.
554 In that case try picking one from a nearby CPU
555 - The APIC IDs differ from the HyperTransport node IDs
556 which the K8 northbridge parsing fills in.
557 Assume they are all increased by a constant offset,
558 but in the same order as the HT nodeids.
559 If that doesn't result in a usable node fall back to the
560 path for the previous case. */
561
562 int ht_nodeid = apicid - (cpu_data(0).phys_proc_id << bits);
563
564 if (ht_nodeid >= 0 &&
565 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
566 node = apicid_to_node[ht_nodeid];
567 /* Pick a nearby node */
568 if (!node_online(node))
569 node = nearby_node(apicid);
570 }
571 numa_set_node(cpu, node);
572
573 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
574 #endif
575 #endif
576 }
577
578 static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
579 {
580 #ifdef CONFIG_SMP
581 unsigned bits, ecx;
582
583 /* Multi core CPU? */
584 if (c->extended_cpuid_level < 0x80000008)
585 return;
586
587 ecx = cpuid_ecx(0x80000008);
588
589 c->x86_max_cores = (ecx & 0xff) + 1;
590
591 /* CPU telling us the core id bits shift? */
592 bits = (ecx >> 12) & 0xF;
593
594 /* Otherwise recompute */
595 if (bits == 0) {
596 while ((1 << bits) < c->x86_max_cores)
597 bits++;
598 }
599
600 c->x86_coreid_bits = bits;
601
602 #endif
603 }
604
605 #define ENABLE_C1E_MASK 0x18000000
606 #define CPUID_PROCESSOR_SIGNATURE 1
607 #define CPUID_XFAM 0x0ff00000
608 #define CPUID_XFAM_K8 0x00000000
609 #define CPUID_XFAM_10H 0x00100000
610 #define CPUID_XFAM_11H 0x00200000
611 #define CPUID_XMOD 0x000f0000
612 #define CPUID_XMOD_REV_F 0x00040000
613
614 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
615 static __cpuinit int amd_apic_timer_broken(void)
616 {
617 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
618
619 switch (eax & CPUID_XFAM) {
620 case CPUID_XFAM_K8:
621 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
622 break;
623 case CPUID_XFAM_10H:
624 case CPUID_XFAM_11H:
625 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
626 if (lo & ENABLE_C1E_MASK)
627 return 1;
628 break;
629 default:
630 /* err on the side of caution */
631 return 1;
632 }
633 return 0;
634 }
635
636 static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
637 {
638 early_init_amd_mc(c);
639
640 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
641 if (c->x86_power & (1<<8))
642 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
643 }
644
645 static void __cpuinit init_amd(struct cpuinfo_x86 *c)
646 {
647 unsigned level;
648
649 #ifdef CONFIG_SMP
650 unsigned long value;
651
652 /*
653 * Disable TLB flush filter by setting HWCR.FFDIS on K8
654 * bit 6 of msr C001_0015
655 *
656 * Errata 63 for SH-B3 steppings
657 * Errata 122 for all steppings (F+ have it disabled by default)
658 */
659 if (c->x86 == 15) {
660 rdmsrl(MSR_K8_HWCR, value);
661 value |= 1 << 6;
662 wrmsrl(MSR_K8_HWCR, value);
663 }
664 #endif
665
666 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
667 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
668 clear_bit(0*32+31, (unsigned long *)&c->x86_capability);
669
670 /* On C+ stepping K8 rep microcode works well for copy/memset */
671 level = cpuid_eax(1);
672 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
673 level >= 0x0f58))
674 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
675 if (c->x86 == 0x10 || c->x86 == 0x11)
676 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
677
678 /* Enable workaround for FXSAVE leak */
679 if (c->x86 >= 6)
680 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
681
682 level = get_model_name(c);
683 if (!level) {
684 switch (c->x86) {
685 case 15:
686 /* Should distinguish Models here, but this is only
687 a fallback anyways. */
688 strcpy(c->x86_model_id, "Hammer");
689 break;
690 }
691 }
692 display_cacheinfo(c);
693
694 /* Multi core CPU? */
695 if (c->extended_cpuid_level >= 0x80000008)
696 amd_detect_cmp(c);
697
698 if (c->extended_cpuid_level >= 0x80000006 &&
699 (cpuid_edx(0x80000006) & 0xf000))
700 num_cache_leaves = 4;
701 else
702 num_cache_leaves = 3;
703
704 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
705 set_cpu_cap(c, X86_FEATURE_K8);
706
707 /* MFENCE stops RDTSC speculation */
708 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
709
710 if (amd_apic_timer_broken())
711 disable_apic_timer = 1;
712 }
713
714 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
715 {
716 #ifdef CONFIG_SMP
717 u32 eax, ebx, ecx, edx;
718 int index_msb, core_bits;
719
720 cpuid(1, &eax, &ebx, &ecx, &edx);
721
722
723 if (!cpu_has(c, X86_FEATURE_HT))
724 return;
725 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
726 goto out;
727
728 smp_num_siblings = (ebx & 0xff0000) >> 16;
729
730 if (smp_num_siblings == 1) {
731 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
732 } else if (smp_num_siblings > 1) {
733
734 if (smp_num_siblings > NR_CPUS) {
735 printk(KERN_WARNING "CPU: Unsupported number of "
736 "siblings %d", smp_num_siblings);
737 smp_num_siblings = 1;
738 return;
739 }
740
741 index_msb = get_count_order(smp_num_siblings);
742 c->phys_proc_id = phys_pkg_id(index_msb);
743
744 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
745
746 index_msb = get_count_order(smp_num_siblings);
747
748 core_bits = get_count_order(c->x86_max_cores);
749
750 c->cpu_core_id = phys_pkg_id(index_msb) &
751 ((1 << core_bits) - 1);
752 }
753 out:
754 if ((c->x86_max_cores * smp_num_siblings) > 1) {
755 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
756 c->phys_proc_id);
757 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
758 c->cpu_core_id);
759 }
760
761 #endif
762 }
763
764 /*
765 * find out the number of processor cores on the die
766 */
767 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
768 {
769 unsigned int eax, t;
770
771 if (c->cpuid_level < 4)
772 return 1;
773
774 cpuid_count(4, 0, &eax, &t, &t, &t);
775
776 if (eax & 0x1f)
777 return ((eax >> 26) + 1);
778 else
779 return 1;
780 }
781
782 static void srat_detect_node(void)
783 {
784 #ifdef CONFIG_NUMA
785 unsigned node;
786 int cpu = smp_processor_id();
787 int apicid = hard_smp_processor_id();
788
789 /* Don't do the funky fallback heuristics the AMD version employs
790 for now. */
791 node = apicid_to_node[apicid];
792 if (node == NUMA_NO_NODE)
793 node = first_node(node_online_map);
794 numa_set_node(cpu, node);
795
796 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
797 #endif
798 }
799
800 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
801 {
802 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
803 (c->x86 == 0x6 && c->x86_model >= 0x0e))
804 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
805 }
806
807 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
808 {
809 /* Cache sizes */
810 unsigned n;
811
812 init_intel_cacheinfo(c);
813 if (c->cpuid_level > 9) {
814 unsigned eax = cpuid_eax(10);
815 /* Check for version and the number of counters */
816 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
817 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
818 }
819
820 if (cpu_has_ds) {
821 unsigned int l1, l2;
822 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
823 if (!(l1 & (1<<11)))
824 set_cpu_cap(c, X86_FEATURE_BTS);
825 if (!(l1 & (1<<12)))
826 set_cpu_cap(c, X86_FEATURE_PEBS);
827 }
828
829
830 if (cpu_has_bts)
831 ds_init_intel(c);
832
833 n = c->extended_cpuid_level;
834 if (n >= 0x80000008) {
835 unsigned eax = cpuid_eax(0x80000008);
836 c->x86_virt_bits = (eax >> 8) & 0xff;
837 c->x86_phys_bits = eax & 0xff;
838 /* CPUID workaround for Intel 0F34 CPU */
839 if (c->x86_vendor == X86_VENDOR_INTEL &&
840 c->x86 == 0xF && c->x86_model == 0x3 &&
841 c->x86_mask == 0x4)
842 c->x86_phys_bits = 36;
843 }
844
845 if (c->x86 == 15)
846 c->x86_cache_alignment = c->x86_clflush_size * 2;
847 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
848 (c->x86 == 0x6 && c->x86_model >= 0x0e))
849 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
850 if (c->x86 == 6)
851 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
852 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
853 c->x86_max_cores = intel_num_cpu_cores(c);
854
855 srat_detect_node();
856 }
857
858 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
859 {
860 char *v = c->x86_vendor_id;
861
862 if (!strcmp(v, "AuthenticAMD"))
863 c->x86_vendor = X86_VENDOR_AMD;
864 else if (!strcmp(v, "GenuineIntel"))
865 c->x86_vendor = X86_VENDOR_INTEL;
866 else
867 c->x86_vendor = X86_VENDOR_UNKNOWN;
868 }
869
870 struct cpu_model_info {
871 int vendor;
872 int family;
873 char *model_names[16];
874 };
875
876 /* Do some early cpuid on the boot CPU to get some parameter that are
877 needed before check_bugs. Everything advanced is in identify_cpu
878 below. */
879 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
880 {
881 u32 tfms, xlvl;
882
883 c->loops_per_jiffy = loops_per_jiffy;
884 c->x86_cache_size = -1;
885 c->x86_vendor = X86_VENDOR_UNKNOWN;
886 c->x86_model = c->x86_mask = 0; /* So far unknown... */
887 c->x86_vendor_id[0] = '\0'; /* Unset */
888 c->x86_model_id[0] = '\0'; /* Unset */
889 c->x86_clflush_size = 64;
890 c->x86_cache_alignment = c->x86_clflush_size;
891 c->x86_max_cores = 1;
892 c->x86_coreid_bits = 0;
893 c->extended_cpuid_level = 0;
894 memset(&c->x86_capability, 0, sizeof c->x86_capability);
895
896 /* Get vendor name */
897 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
898 (unsigned int *)&c->x86_vendor_id[0],
899 (unsigned int *)&c->x86_vendor_id[8],
900 (unsigned int *)&c->x86_vendor_id[4]);
901
902 get_cpu_vendor(c);
903
904 /* Initialize the standard set of capabilities */
905 /* Note that the vendor-specific code below might override */
906
907 /* Intel-defined flags: level 0x00000001 */
908 if (c->cpuid_level >= 0x00000001) {
909 __u32 misc;
910 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
911 &c->x86_capability[0]);
912 c->x86 = (tfms >> 8) & 0xf;
913 c->x86_model = (tfms >> 4) & 0xf;
914 c->x86_mask = tfms & 0xf;
915 if (c->x86 == 0xf)
916 c->x86 += (tfms >> 20) & 0xff;
917 if (c->x86 >= 0x6)
918 c->x86_model += ((tfms >> 16) & 0xF) << 4;
919 if (c->x86_capability[0] & (1<<19))
920 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
921 } else {
922 /* Have CPUID level 0 only - unheard of */
923 c->x86 = 4;
924 }
925
926 #ifdef CONFIG_SMP
927 c->phys_proc_id = (cpuid_ebx(1) >> 24) & 0xff;
928 #endif
929 /* AMD-defined flags: level 0x80000001 */
930 xlvl = cpuid_eax(0x80000000);
931 c->extended_cpuid_level = xlvl;
932 if ((xlvl & 0xffff0000) == 0x80000000) {
933 if (xlvl >= 0x80000001) {
934 c->x86_capability[1] = cpuid_edx(0x80000001);
935 c->x86_capability[6] = cpuid_ecx(0x80000001);
936 }
937 if (xlvl >= 0x80000004)
938 get_model_name(c); /* Default name */
939 }
940
941 /* Transmeta-defined flags: level 0x80860001 */
942 xlvl = cpuid_eax(0x80860000);
943 if ((xlvl & 0xffff0000) == 0x80860000) {
944 /* Don't set x86_cpuid_level here for now to not confuse. */
945 if (xlvl >= 0x80860001)
946 c->x86_capability[2] = cpuid_edx(0x80860001);
947 }
948
949 c->extended_cpuid_level = cpuid_eax(0x80000000);
950 if (c->extended_cpuid_level >= 0x80000007)
951 c->x86_power = cpuid_edx(0x80000007);
952
953 switch (c->x86_vendor) {
954 case X86_VENDOR_AMD:
955 early_init_amd(c);
956 break;
957 }
958
959 }
960
961 /*
962 * This does the hard work of actually picking apart the CPU stuff...
963 */
964 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
965 {
966 int i;
967
968 early_identify_cpu(c);
969
970 init_scattered_cpuid_features(c);
971
972 c->apicid = phys_pkg_id(0);
973
974 /*
975 * Vendor-specific initialization. In this section we
976 * canonicalize the feature flags, meaning if there are
977 * features a certain CPU supports which CPUID doesn't
978 * tell us, CPUID claiming incorrect flags, or other bugs,
979 * we handle them here.
980 *
981 * At the end of this section, c->x86_capability better
982 * indicate the features this CPU genuinely supports!
983 */
984 switch (c->x86_vendor) {
985 case X86_VENDOR_AMD:
986 init_amd(c);
987 break;
988
989 case X86_VENDOR_INTEL:
990 init_intel(c);
991 break;
992
993 case X86_VENDOR_UNKNOWN:
994 default:
995 display_cacheinfo(c);
996 break;
997 }
998
999 select_idle_routine(c);
1000 detect_ht(c);
1001
1002 /*
1003 * On SMP, boot_cpu_data holds the common feature set between
1004 * all CPUs; so make sure that we indicate which features are
1005 * common between the CPUs. The first time this routine gets
1006 * executed, c == &boot_cpu_data.
1007 */
1008 if (c != &boot_cpu_data) {
1009 /* AND the already accumulated flags with these */
1010 for (i = 0; i < NCAPINTS; i++)
1011 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1012 }
1013
1014 #ifdef CONFIG_X86_MCE
1015 mcheck_init(c);
1016 #endif
1017 if (c != &boot_cpu_data)
1018 mtrr_ap_init();
1019 #ifdef CONFIG_NUMA
1020 numa_add_cpu(smp_processor_id());
1021 #endif
1022
1023 switch (c->x86_vendor) {
1024 case X86_VENDOR_AMD:
1025 early_init_amd(c);
1026 break;
1027 case X86_VENDOR_INTEL:
1028 early_init_intel(c);
1029 break;
1030 }
1031 }
1032
1033 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1034 {
1035 if (c->x86_model_id[0])
1036 printk(KERN_INFO "%s", c->x86_model_id);
1037
1038 if (c->x86_mask || c->cpuid_level >= 0)
1039 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1040 else
1041 printk(KERN_CONT "\n");
1042 }
1043
1044 /*
1045 * Get CPU information for use by the procfs.
1046 */
1047
1048 static int show_cpuinfo(struct seq_file *m, void *v)
1049 {
1050 struct cpuinfo_x86 *c = v;
1051 int cpu = 0, i;
1052
1053 /*
1054 * These flag bits must match the definitions in <asm/cpufeature.h>.
1055 * NULL means this bit is undefined or reserved; either way it doesn't
1056 * have meaning as far as Linux is concerned. Note that it's important
1057 * to realize there is a difference between this table and CPUID -- if
1058 * applications want to get the raw CPUID data, they should access
1059 * /dev/cpu/<cpu_nr>/cpuid instead.
1060 */
1061 static const char *const x86_cap_flags[] = {
1062 /* Intel-defined */
1063 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1064 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1065 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1066 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
1067
1068 /* AMD-defined */
1069 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1070 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1071 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1072 NULL, "fxsr_opt", "pdpe1gb", "rdtscp", NULL, "lm",
1073 "3dnowext", "3dnow",
1074
1075 /* Transmeta-defined */
1076 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1077 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1078 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1079 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1080
1081 /* Other (Linux-defined) */
1082 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
1083 NULL, NULL, NULL, NULL,
1084 "constant_tsc", "up", NULL, "arch_perfmon",
1085 "pebs", "bts", NULL, "sync_rdtsc",
1086 "rep_good", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1087 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1088
1089 /* Intel-defined (#2) */
1090 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
1091 "tm2", "ssse3", "cid", NULL, NULL, "cx16", "xtpr", NULL,
1092 NULL, NULL, "dca", "sse4_1", "sse4_2", NULL, NULL, "popcnt",
1093 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1094
1095 /* VIA/Cyrix/Centaur-defined */
1096 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1097 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL, NULL,
1098 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1099 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1100
1101 /* AMD-defined (#2) */
1102 "lahf_lm", "cmp_legacy", "svm", "extapic",
1103 "cr8_legacy", "abm", "sse4a", "misalignsse",
1104 "3dnowprefetch", "osvw", "ibs", "sse5",
1105 "skinit", "wdt", NULL, NULL,
1106 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1107 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1108
1109 /* Auxiliary (Linux-defined) */
1110 "ida", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1111 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1112 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1113 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1114 };
1115 static const char *const x86_power_flags[] = {
1116 "ts", /* temperature sensor */
1117 "fid", /* frequency id control */
1118 "vid", /* voltage id control */
1119 "ttp", /* thermal trip */
1120 "tm",
1121 "stc",
1122 "100mhzsteps",
1123 "hwpstate",
1124 "", /* tsc invariant mapped to constant_tsc */
1125 /* nothing */
1126 };
1127
1128
1129 #ifdef CONFIG_SMP
1130 cpu = c->cpu_index;
1131 #endif
1132
1133 seq_printf(m, "processor\t: %u\n"
1134 "vendor_id\t: %s\n"
1135 "cpu family\t: %d\n"
1136 "model\t\t: %d\n"
1137 "model name\t: %s\n",
1138 (unsigned)cpu,
1139 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1140 c->x86,
1141 (int)c->x86_model,
1142 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1143
1144 if (c->x86_mask || c->cpuid_level >= 0)
1145 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1146 else
1147 seq_printf(m, "stepping\t: unknown\n");
1148
1149 if (cpu_has(c, X86_FEATURE_TSC)) {
1150 unsigned int freq = cpufreq_quick_get((unsigned)cpu);
1151
1152 if (!freq)
1153 freq = cpu_khz;
1154 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1155 freq / 1000, (freq % 1000));
1156 }
1157
1158 /* Cache size */
1159 if (c->x86_cache_size >= 0)
1160 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1161
1162 #ifdef CONFIG_SMP
1163 if (smp_num_siblings * c->x86_max_cores > 1) {
1164 seq_printf(m, "physical id\t: %d\n", c->phys_proc_id);
1165 seq_printf(m, "siblings\t: %d\n",
1166 cpus_weight(per_cpu(cpu_core_map, cpu)));
1167 seq_printf(m, "core id\t\t: %d\n", c->cpu_core_id);
1168 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1169 }
1170 #endif
1171
1172 seq_printf(m,
1173 "fpu\t\t: yes\n"
1174 "fpu_exception\t: yes\n"
1175 "cpuid level\t: %d\n"
1176 "wp\t\t: yes\n"
1177 "flags\t\t:",
1178 c->cpuid_level);
1179
1180 for (i = 0; i < 32*NCAPINTS; i++)
1181 if (cpu_has(c, i) && x86_cap_flags[i] != NULL)
1182 seq_printf(m, " %s", x86_cap_flags[i]);
1183
1184 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1185 c->loops_per_jiffy/(500000/HZ),
1186 (c->loops_per_jiffy/(5000/HZ)) % 100);
1187
1188 if (c->x86_tlbsize > 0)
1189 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1190 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1191 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1192
1193 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1194 c->x86_phys_bits, c->x86_virt_bits);
1195
1196 seq_printf(m, "power management:");
1197 for (i = 0; i < 32; i++) {
1198 if (c->x86_power & (1 << i)) {
1199 if (i < ARRAY_SIZE(x86_power_flags) &&
1200 x86_power_flags[i])
1201 seq_printf(m, "%s%s",
1202 x86_power_flags[i][0]?" ":"",
1203 x86_power_flags[i]);
1204 else
1205 seq_printf(m, " [%d]", i);
1206 }
1207 }
1208
1209 seq_printf(m, "\n\n");
1210
1211 return 0;
1212 }
1213
1214 static void *c_start(struct seq_file *m, loff_t *pos)
1215 {
1216 if (*pos == 0) /* just in case, cpu 0 is not the first */
1217 *pos = first_cpu(cpu_online_map);
1218 if ((*pos) < NR_CPUS && cpu_online(*pos))
1219 return &cpu_data(*pos);
1220 return NULL;
1221 }
1222
1223 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1224 {
1225 *pos = next_cpu(*pos, cpu_online_map);
1226 return c_start(m, pos);
1227 }
1228
1229 static void c_stop(struct seq_file *m, void *v)
1230 {
1231 }
1232
1233 struct seq_operations cpuinfo_op = {
1234 .start = c_start,
1235 .next = c_next,
1236 .stop = c_stop,
1237 .show = show_cpuinfo,
1238 };
This page took 0.054169 seconds and 4 git commands to generate.