x86: move ioapic code where it belongs
[GitHub/mt8127/android_kernel_alcatel_ttab.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/acpi.h>
34 #include <linux/kallsyms.h>
35 #include <linux/edd.h>
36 #include <linux/mmzone.h>
37 #include <linux/kexec.h>
38 #include <linux/cpufreq.h>
39 #include <linux/dmi.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/ctype.h>
42
43 #include <asm/mtrr.h>
44 #include <asm/uaccess.h>
45 #include <asm/system.h>
46 #include <asm/io.h>
47 #include <asm/smp.h>
48 #include <asm/msr.h>
49 #include <asm/desc.h>
50 #include <video/edid.h>
51 #include <asm/e820.h>
52 #include <asm/dma.h>
53 #include <asm/mpspec.h>
54 #include <asm/mmu_context.h>
55 #include <asm/proto.h>
56 #include <asm/setup.h>
57 #include <asm/mach_apic.h>
58 #include <asm/numa.h>
59 #include <asm/sections.h>
60 #include <asm/dmi.h>
61 #include <asm/cacheflush.h>
62 #include <asm/mce.h>
63
64 /*
65 * Machine setup..
66 */
67
68 struct cpuinfo_x86 boot_cpu_data __read_mostly;
69 EXPORT_SYMBOL(boot_cpu_data);
70
71 unsigned long mmu_cr4_features;
72
73 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
74 int bootloader_type;
75
76 unsigned long saved_video_mode;
77
78 int force_mwait __cpuinitdata;
79
80 /*
81 * Early DMI memory
82 */
83 int dmi_alloc_index;
84 char dmi_alloc_data[DMI_MAX_DATA];
85
86 /*
87 * Setup options
88 */
89 struct screen_info screen_info;
90 EXPORT_SYMBOL(screen_info);
91 struct sys_desc_table_struct {
92 unsigned short length;
93 unsigned char table[0];
94 };
95
96 struct edid_info edid_info;
97 EXPORT_SYMBOL_GPL(edid_info);
98
99 extern int root_mountflags;
100
101 char __initdata command_line[COMMAND_LINE_SIZE];
102
103 struct resource standard_io_resources[] = {
104 { .name = "dma1", .start = 0x00, .end = 0x1f,
105 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
106 { .name = "pic1", .start = 0x20, .end = 0x21,
107 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
108 { .name = "timer0", .start = 0x40, .end = 0x43,
109 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
110 { .name = "timer1", .start = 0x50, .end = 0x53,
111 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
112 { .name = "keyboard", .start = 0x60, .end = 0x6f,
113 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
114 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
115 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
116 { .name = "pic2", .start = 0xa0, .end = 0xa1,
117 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
118 { .name = "dma2", .start = 0xc0, .end = 0xdf,
119 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
120 { .name = "fpu", .start = 0xf0, .end = 0xff,
121 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
122 };
123
124 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
125
126 struct resource data_resource = {
127 .name = "Kernel data",
128 .start = 0,
129 .end = 0,
130 .flags = IORESOURCE_RAM,
131 };
132 struct resource code_resource = {
133 .name = "Kernel code",
134 .start = 0,
135 .end = 0,
136 .flags = IORESOURCE_RAM,
137 };
138 struct resource bss_resource = {
139 .name = "Kernel bss",
140 .start = 0,
141 .end = 0,
142 .flags = IORESOURCE_RAM,
143 };
144
145 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
146
147 #ifdef CONFIG_PROC_VMCORE
148 /* elfcorehdr= specifies the location of elf core header
149 * stored by the crashed kernel. This option will be passed
150 * by kexec loader to the capture kernel.
151 */
152 static int __init setup_elfcorehdr(char *arg)
153 {
154 char *end;
155 if (!arg)
156 return -EINVAL;
157 elfcorehdr_addr = memparse(arg, &end);
158 return end > arg ? 0 : -EINVAL;
159 }
160 early_param("elfcorehdr", setup_elfcorehdr);
161 #endif
162
163 #ifndef CONFIG_NUMA
164 static void __init
165 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
166 {
167 unsigned long bootmap_size, bootmap;
168
169 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
170 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
171 if (bootmap == -1L)
172 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
173 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
174 e820_register_active_regions(0, start_pfn, end_pfn);
175 free_bootmem_with_active_regions(0, end_pfn);
176 reserve_bootmem(bootmap, bootmap_size);
177 }
178 #endif
179
180 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
181 struct edd edd;
182 #ifdef CONFIG_EDD_MODULE
183 EXPORT_SYMBOL(edd);
184 #endif
185 /**
186 * copy_edd() - Copy the BIOS EDD information
187 * from boot_params into a safe place.
188 *
189 */
190 static inline void copy_edd(void)
191 {
192 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
193 sizeof(edd.mbr_signature));
194 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
195 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
196 edd.edd_info_nr = boot_params.eddbuf_entries;
197 }
198 #else
199 static inline void copy_edd(void)
200 {
201 }
202 #endif
203
204 #ifdef CONFIG_KEXEC
205 static void __init reserve_crashkernel(void)
206 {
207 unsigned long long free_mem;
208 unsigned long long crash_size, crash_base;
209 int ret;
210
211 free_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
212
213 ret = parse_crashkernel(boot_command_line, free_mem,
214 &crash_size, &crash_base);
215 if (ret == 0 && crash_size) {
216 if (crash_base > 0) {
217 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
218 "for crashkernel (System RAM: %ldMB)\n",
219 (unsigned long)(crash_size >> 20),
220 (unsigned long)(crash_base >> 20),
221 (unsigned long)(free_mem >> 20));
222 crashk_res.start = crash_base;
223 crashk_res.end = crash_base + crash_size - 1;
224 reserve_bootmem(crash_base, crash_size);
225 } else
226 printk(KERN_INFO "crashkernel reservation failed - "
227 "you have to specify a base address\n");
228 }
229 }
230 #else
231 static inline void __init reserve_crashkernel(void)
232 {}
233 #endif
234
235 #define EBDA_ADDR_POINTER 0x40E
236
237 unsigned __initdata ebda_addr;
238 unsigned __initdata ebda_size;
239
240 static void discover_ebda(void)
241 {
242 /*
243 * there is a real-mode segmented pointer pointing to the
244 * 4K EBDA area at 0x40E
245 */
246 ebda_addr = *(unsigned short *)__va(EBDA_ADDR_POINTER);
247 ebda_addr <<= 4;
248
249 ebda_size = *(unsigned short *)__va(ebda_addr);
250
251 /* Round EBDA up to pages */
252 if (ebda_size == 0)
253 ebda_size = 1;
254 ebda_size <<= 10;
255 ebda_size = round_up(ebda_size + (ebda_addr & ~PAGE_MASK), PAGE_SIZE);
256 if (ebda_size > 64*1024)
257 ebda_size = 64*1024;
258 }
259
260 void __init setup_arch(char **cmdline_p)
261 {
262 printk(KERN_INFO "Command line: %s\n", boot_command_line);
263
264 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
265 screen_info = boot_params.screen_info;
266 edid_info = boot_params.edid_info;
267 saved_video_mode = boot_params.hdr.vid_mode;
268 bootloader_type = boot_params.hdr.type_of_loader;
269
270 #ifdef CONFIG_BLK_DEV_RAM
271 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
272 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
273 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
274 #endif
275 setup_memory_region();
276 copy_edd();
277
278 if (!boot_params.hdr.root_flags)
279 root_mountflags &= ~MS_RDONLY;
280 init_mm.start_code = (unsigned long) &_text;
281 init_mm.end_code = (unsigned long) &_etext;
282 init_mm.end_data = (unsigned long) &_edata;
283 init_mm.brk = (unsigned long) &_end;
284
285 code_resource.start = virt_to_phys(&_text);
286 code_resource.end = virt_to_phys(&_etext)-1;
287 data_resource.start = virt_to_phys(&_etext);
288 data_resource.end = virt_to_phys(&_edata)-1;
289 bss_resource.start = virt_to_phys(&__bss_start);
290 bss_resource.end = virt_to_phys(&__bss_stop)-1;
291
292 early_identify_cpu(&boot_cpu_data);
293
294 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
295 *cmdline_p = command_line;
296
297 parse_early_param();
298
299 finish_e820_parsing();
300
301 e820_register_active_regions(0, 0, -1UL);
302 /*
303 * partially used pages are not usable - thus
304 * we are rounding upwards:
305 */
306 end_pfn = e820_end_of_ram();
307 num_physpages = end_pfn;
308
309 check_efer();
310
311 discover_ebda();
312
313 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
314
315 dmi_scan_machine();
316
317 io_delay_init();
318
319 #ifdef CONFIG_SMP
320 /* setup to use the static apicid table during kernel startup */
321 x86_cpu_to_apicid_ptr = (void *)&x86_cpu_to_apicid_init;
322 #endif
323
324 #ifdef CONFIG_ACPI
325 /*
326 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
327 * Call this early for SRAT node setup.
328 */
329 acpi_boot_table_init();
330 #endif
331
332 /* How many end-of-memory variables you have, grandma! */
333 max_low_pfn = end_pfn;
334 max_pfn = end_pfn;
335 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
336
337 /* Remove active ranges so rediscovery with NUMA-awareness happens */
338 remove_all_active_ranges();
339
340 #ifdef CONFIG_ACPI_NUMA
341 /*
342 * Parse SRAT to discover nodes.
343 */
344 acpi_numa_init();
345 #endif
346
347 #ifdef CONFIG_NUMA
348 numa_initmem_init(0, end_pfn);
349 #else
350 contig_initmem_init(0, end_pfn);
351 #endif
352
353 /* Reserve direct mapping */
354 reserve_bootmem_generic(table_start << PAGE_SHIFT,
355 (table_end - table_start) << PAGE_SHIFT);
356
357 /* reserve kernel */
358 reserve_bootmem_generic(__pa_symbol(&_text),
359 __pa_symbol(&_end) - __pa_symbol(&_text));
360
361 /*
362 * reserve physical page 0 - it's a special BIOS page on many boxes,
363 * enabling clean reboots, SMP operation, laptop functions.
364 */
365 reserve_bootmem_generic(0, PAGE_SIZE);
366
367 /* reserve ebda region */
368 if (ebda_addr)
369 reserve_bootmem_generic(ebda_addr, ebda_size);
370 #ifdef CONFIG_NUMA
371 /* reserve nodemap region */
372 if (nodemap_addr)
373 reserve_bootmem_generic(nodemap_addr, nodemap_size);
374 #endif
375
376 #ifdef CONFIG_SMP
377 /* Reserve SMP trampoline */
378 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, 2*PAGE_SIZE);
379 #endif
380
381 #ifdef CONFIG_ACPI_SLEEP
382 /*
383 * Reserve low memory region for sleep support.
384 */
385 acpi_reserve_bootmem();
386 #endif
387 /*
388 * Find and reserve possible boot-time SMP configuration:
389 */
390 find_smp_config();
391 #ifdef CONFIG_BLK_DEV_INITRD
392 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
393 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
394 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
395 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
396 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
397
398 if (ramdisk_end <= end_of_mem) {
399 reserve_bootmem_generic(ramdisk_image, ramdisk_size);
400 initrd_start = ramdisk_image + PAGE_OFFSET;
401 initrd_end = initrd_start+ramdisk_size;
402 } else {
403 printk(KERN_ERR "initrd extends beyond end of memory "
404 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
405 ramdisk_end, end_of_mem);
406 initrd_start = 0;
407 }
408 }
409 #endif
410 reserve_crashkernel();
411 paging_init();
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();
442 e820_mark_nosave_regions();
443
444 {
445 unsigned i;
446 /* request I/O space for devices used on all i[345]86 PCs */
447 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
448 request_resource(&ioport_resource, &standard_io_resources[i]);
449 }
450
451 e820_setup_gap();
452
453 #ifdef CONFIG_VT
454 #if defined(CONFIG_VGA_CONSOLE)
455 conswitchp = &vga_con;
456 #elif defined(CONFIG_DUMMY_CONSOLE)
457 conswitchp = &dummy_con;
458 #endif
459 #endif
460 }
461
462 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
463 {
464 unsigned int *v;
465
466 if (c->extended_cpuid_level < 0x80000004)
467 return 0;
468
469 v = (unsigned int *) c->x86_model_id;
470 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
471 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
472 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
473 c->x86_model_id[48] = 0;
474 return 1;
475 }
476
477
478 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
479 {
480 unsigned int n, dummy, eax, ebx, ecx, edx;
481
482 n = c->extended_cpuid_level;
483
484 if (n >= 0x80000005) {
485 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
486 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
487 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
488 c->x86_cache_size=(ecx>>24)+(edx>>24);
489 /* On K8 L1 TLB is inclusive, so don't count it */
490 c->x86_tlbsize = 0;
491 }
492
493 if (n >= 0x80000006) {
494 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
495 ecx = cpuid_ecx(0x80000006);
496 c->x86_cache_size = ecx >> 16;
497 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
498
499 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
500 c->x86_cache_size, ecx & 0xFF);
501 }
502
503 if (n >= 0x80000007)
504 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
505 if (n >= 0x80000008) {
506 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
507 c->x86_virt_bits = (eax >> 8) & 0xff;
508 c->x86_phys_bits = eax & 0xff;
509 }
510 }
511
512 #ifdef CONFIG_NUMA
513 static int nearby_node(int apicid)
514 {
515 int i;
516 for (i = apicid - 1; i >= 0; i--) {
517 int node = apicid_to_node[i];
518 if (node != NUMA_NO_NODE && node_online(node))
519 return node;
520 }
521 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
522 int node = apicid_to_node[i];
523 if (node != NUMA_NO_NODE && node_online(node))
524 return node;
525 }
526 return first_node(node_online_map); /* Shouldn't happen */
527 }
528 #endif
529
530 /*
531 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
532 * Assumes number of cores is a power of two.
533 */
534 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
535 {
536 #ifdef CONFIG_SMP
537 unsigned bits;
538 #ifdef CONFIG_NUMA
539 int cpu = smp_processor_id();
540 int node = 0;
541 unsigned apicid = hard_smp_processor_id();
542 #endif
543 unsigned ecx = cpuid_ecx(0x80000008);
544
545 c->x86_max_cores = (ecx & 0xff) + 1;
546
547 /* CPU telling us the core id bits shift? */
548 bits = (ecx >> 12) & 0xF;
549
550 /* Otherwise recompute */
551 if (bits == 0) {
552 while ((1 << bits) < c->x86_max_cores)
553 bits++;
554 }
555
556 /* Low order bits define the core id (index of core in socket) */
557 c->cpu_core_id = c->phys_proc_id & ((1 << bits)-1);
558 /* Convert the APIC ID into the socket ID */
559 c->phys_proc_id = phys_pkg_id(bits);
560
561 #ifdef CONFIG_NUMA
562 node = c->phys_proc_id;
563 if (apicid_to_node[apicid] != NUMA_NO_NODE)
564 node = apicid_to_node[apicid];
565 if (!node_online(node)) {
566 /* Two possibilities here:
567 - The CPU is missing memory and no node was created.
568 In that case try picking one from a nearby CPU
569 - The APIC IDs differ from the HyperTransport node IDs
570 which the K8 northbridge parsing fills in.
571 Assume they are all increased by a constant offset,
572 but in the same order as the HT nodeids.
573 If that doesn't result in a usable node fall back to the
574 path for the previous case. */
575 int ht_nodeid = apicid - (cpu_data(0).phys_proc_id << bits);
576 if (ht_nodeid >= 0 &&
577 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
578 node = apicid_to_node[ht_nodeid];
579 /* Pick a nearby node */
580 if (!node_online(node))
581 node = nearby_node(apicid);
582 }
583 numa_set_node(cpu, node);
584
585 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
586 #endif
587 #endif
588 }
589
590 #define ENABLE_C1E_MASK 0x18000000
591 #define CPUID_PROCESSOR_SIGNATURE 1
592 #define CPUID_XFAM 0x0ff00000
593 #define CPUID_XFAM_K8 0x00000000
594 #define CPUID_XFAM_10H 0x00100000
595 #define CPUID_XFAM_11H 0x00200000
596 #define CPUID_XMOD 0x000f0000
597 #define CPUID_XMOD_REV_F 0x00040000
598
599 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
600 static __cpuinit int amd_apic_timer_broken(void)
601 {
602 u32 lo, hi;
603 u32 eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
604 switch (eax & CPUID_XFAM) {
605 case CPUID_XFAM_K8:
606 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
607 break;
608 case CPUID_XFAM_10H:
609 case CPUID_XFAM_11H:
610 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
611 if (lo & ENABLE_C1E_MASK)
612 return 1;
613 break;
614 default:
615 /* err on the side of caution */
616 return 1;
617 }
618 return 0;
619 }
620
621 static void __cpuinit init_amd(struct cpuinfo_x86 *c)
622 {
623 unsigned level;
624
625 #ifdef CONFIG_SMP
626 unsigned long value;
627
628 /*
629 * Disable TLB flush filter by setting HWCR.FFDIS on K8
630 * bit 6 of msr C001_0015
631 *
632 * Errata 63 for SH-B3 steppings
633 * Errata 122 for all steppings (F+ have it disabled by default)
634 */
635 if (c->x86 == 15) {
636 rdmsrl(MSR_K8_HWCR, value);
637 value |= 1 << 6;
638 wrmsrl(MSR_K8_HWCR, value);
639 }
640 #endif
641
642 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
643 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
644 clear_bit(0*32+31, &c->x86_capability);
645
646 /* On C+ stepping K8 rep microcode works well for copy/memset */
647 level = cpuid_eax(1);
648 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58))
649 set_bit(X86_FEATURE_REP_GOOD, &c->x86_capability);
650 if (c->x86 == 0x10 || c->x86 == 0x11)
651 set_bit(X86_FEATURE_REP_GOOD, &c->x86_capability);
652
653 /* Enable workaround for FXSAVE leak */
654 if (c->x86 >= 6)
655 set_bit(X86_FEATURE_FXSAVE_LEAK, &c->x86_capability);
656
657 level = get_model_name(c);
658 if (!level) {
659 switch (c->x86) {
660 case 15:
661 /* Should distinguish Models here, but this is only
662 a fallback anyways. */
663 strcpy(c->x86_model_id, "Hammer");
664 break;
665 }
666 }
667 display_cacheinfo(c);
668
669 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
670 if (c->x86_power & (1<<8))
671 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
672
673 /* Multi core CPU? */
674 if (c->extended_cpuid_level >= 0x80000008)
675 amd_detect_cmp(c);
676
677 if (c->extended_cpuid_level >= 0x80000006 &&
678 (cpuid_edx(0x80000006) & 0xf000))
679 num_cache_leaves = 4;
680 else
681 num_cache_leaves = 3;
682
683 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
684 set_bit(X86_FEATURE_K8, &c->x86_capability);
685
686 /* RDTSC can be speculated around */
687 clear_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
688
689 /* Family 10 doesn't support C states in MWAIT so don't use it */
690 if (c->x86 == 0x10 && !force_mwait)
691 clear_bit(X86_FEATURE_MWAIT, &c->x86_capability);
692
693 if (amd_apic_timer_broken())
694 disable_apic_timer = 1;
695 }
696
697 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
698 {
699 #ifdef CONFIG_SMP
700 u32 eax, ebx, ecx, edx;
701 int index_msb, core_bits;
702
703 cpuid(1, &eax, &ebx, &ecx, &edx);
704
705
706 if (!cpu_has(c, X86_FEATURE_HT))
707 return;
708 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
709 goto out;
710
711 smp_num_siblings = (ebx & 0xff0000) >> 16;
712
713 if (smp_num_siblings == 1) {
714 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
715 } else if (smp_num_siblings > 1 ) {
716
717 if (smp_num_siblings > NR_CPUS) {
718 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
719 smp_num_siblings = 1;
720 return;
721 }
722
723 index_msb = get_count_order(smp_num_siblings);
724 c->phys_proc_id = phys_pkg_id(index_msb);
725
726 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
727
728 index_msb = get_count_order(smp_num_siblings) ;
729
730 core_bits = get_count_order(c->x86_max_cores);
731
732 c->cpu_core_id = phys_pkg_id(index_msb) &
733 ((1 << core_bits) - 1);
734 }
735 out:
736 if ((c->x86_max_cores * smp_num_siblings) > 1) {
737 printk(KERN_INFO "CPU: Physical Processor ID: %d\n", c->phys_proc_id);
738 printk(KERN_INFO "CPU: Processor Core ID: %d\n", c->cpu_core_id);
739 }
740
741 #endif
742 }
743
744 /*
745 * find out the number of processor cores on the die
746 */
747 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
748 {
749 unsigned int eax, t;
750
751 if (c->cpuid_level < 4)
752 return 1;
753
754 cpuid_count(4, 0, &eax, &t, &t, &t);
755
756 if (eax & 0x1f)
757 return ((eax >> 26) + 1);
758 else
759 return 1;
760 }
761
762 static void srat_detect_node(void)
763 {
764 #ifdef CONFIG_NUMA
765 unsigned node;
766 int cpu = smp_processor_id();
767 int apicid = hard_smp_processor_id();
768
769 /* Don't do the funky fallback heuristics the AMD version employs
770 for now. */
771 node = apicid_to_node[apicid];
772 if (node == NUMA_NO_NODE)
773 node = first_node(node_online_map);
774 numa_set_node(cpu, node);
775
776 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
777 #endif
778 }
779
780 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
781 {
782 /* Cache sizes */
783 unsigned n;
784
785 init_intel_cacheinfo(c);
786 if (c->cpuid_level > 9 ) {
787 unsigned eax = cpuid_eax(10);
788 /* Check for version and the number of counters */
789 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
790 set_bit(X86_FEATURE_ARCH_PERFMON, &c->x86_capability);
791 }
792
793 if (cpu_has_ds) {
794 unsigned int l1, l2;
795 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
796 if (!(l1 & (1<<11)))
797 set_bit(X86_FEATURE_BTS, c->x86_capability);
798 if (!(l1 & (1<<12)))
799 set_bit(X86_FEATURE_PEBS, c->x86_capability);
800 }
801
802 n = c->extended_cpuid_level;
803 if (n >= 0x80000008) {
804 unsigned eax = cpuid_eax(0x80000008);
805 c->x86_virt_bits = (eax >> 8) & 0xff;
806 c->x86_phys_bits = eax & 0xff;
807 /* CPUID workaround for Intel 0F34 CPU */
808 if (c->x86_vendor == X86_VENDOR_INTEL &&
809 c->x86 == 0xF && c->x86_model == 0x3 &&
810 c->x86_mask == 0x4)
811 c->x86_phys_bits = 36;
812 }
813
814 if (c->x86 == 15)
815 c->x86_cache_alignment = c->x86_clflush_size * 2;
816 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
817 (c->x86 == 0x6 && c->x86_model >= 0x0e))
818 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
819 if (c->x86 == 6)
820 set_bit(X86_FEATURE_REP_GOOD, &c->x86_capability);
821 if (c->x86 == 15)
822 set_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
823 else
824 clear_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
825 c->x86_max_cores = intel_num_cpu_cores(c);
826
827 srat_detect_node();
828 }
829
830 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
831 {
832 char *v = c->x86_vendor_id;
833
834 if (!strcmp(v, "AuthenticAMD"))
835 c->x86_vendor = X86_VENDOR_AMD;
836 else if (!strcmp(v, "GenuineIntel"))
837 c->x86_vendor = X86_VENDOR_INTEL;
838 else
839 c->x86_vendor = X86_VENDOR_UNKNOWN;
840 }
841
842 struct cpu_model_info {
843 int vendor;
844 int family;
845 char *model_names[16];
846 };
847
848 /* Do some early cpuid on the boot CPU to get some parameter that are
849 needed before check_bugs. Everything advanced is in identify_cpu
850 below. */
851 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
852 {
853 u32 tfms;
854
855 c->loops_per_jiffy = loops_per_jiffy;
856 c->x86_cache_size = -1;
857 c->x86_vendor = X86_VENDOR_UNKNOWN;
858 c->x86_model = c->x86_mask = 0; /* So far unknown... */
859 c->x86_vendor_id[0] = '\0'; /* Unset */
860 c->x86_model_id[0] = '\0'; /* Unset */
861 c->x86_clflush_size = 64;
862 c->x86_cache_alignment = c->x86_clflush_size;
863 c->x86_max_cores = 1;
864 c->extended_cpuid_level = 0;
865 memset(&c->x86_capability, 0, sizeof c->x86_capability);
866
867 /* Get vendor name */
868 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
869 (unsigned int *)&c->x86_vendor_id[0],
870 (unsigned int *)&c->x86_vendor_id[8],
871 (unsigned int *)&c->x86_vendor_id[4]);
872
873 get_cpu_vendor(c);
874
875 /* Initialize the standard set of capabilities */
876 /* Note that the vendor-specific code below might override */
877
878 /* Intel-defined flags: level 0x00000001 */
879 if (c->cpuid_level >= 0x00000001) {
880 __u32 misc;
881 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
882 &c->x86_capability[0]);
883 c->x86 = (tfms >> 8) & 0xf;
884 c->x86_model = (tfms >> 4) & 0xf;
885 c->x86_mask = tfms & 0xf;
886 if (c->x86 == 0xf)
887 c->x86 += (tfms >> 20) & 0xff;
888 if (c->x86 >= 0x6)
889 c->x86_model += ((tfms >> 16) & 0xF) << 4;
890 if (c->x86_capability[0] & (1<<19))
891 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
892 } else {
893 /* Have CPUID level 0 only - unheard of */
894 c->x86 = 4;
895 }
896
897 #ifdef CONFIG_SMP
898 c->phys_proc_id = (cpuid_ebx(1) >> 24) & 0xff;
899 #endif
900 }
901
902 /*
903 * This does the hard work of actually picking apart the CPU stuff...
904 */
905 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
906 {
907 int i;
908 u32 xlvl;
909
910 early_identify_cpu(c);
911
912 /* AMD-defined flags: level 0x80000001 */
913 xlvl = cpuid_eax(0x80000000);
914 c->extended_cpuid_level = xlvl;
915 if ((xlvl & 0xffff0000) == 0x80000000) {
916 if (xlvl >= 0x80000001) {
917 c->x86_capability[1] = cpuid_edx(0x80000001);
918 c->x86_capability[6] = cpuid_ecx(0x80000001);
919 }
920 if (xlvl >= 0x80000004)
921 get_model_name(c); /* Default name */
922 }
923
924 /* Transmeta-defined flags: level 0x80860001 */
925 xlvl = cpuid_eax(0x80860000);
926 if ((xlvl & 0xffff0000) == 0x80860000) {
927 /* Don't set x86_cpuid_level here for now to not confuse. */
928 if (xlvl >= 0x80860001)
929 c->x86_capability[2] = cpuid_edx(0x80860001);
930 }
931
932 init_scattered_cpuid_features(c);
933
934 c->apicid = phys_pkg_id(0);
935
936 /*
937 * Vendor-specific initialization. In this section we
938 * canonicalize the feature flags, meaning if there are
939 * features a certain CPU supports which CPUID doesn't
940 * tell us, CPUID claiming incorrect flags, or other bugs,
941 * we handle them here.
942 *
943 * At the end of this section, c->x86_capability better
944 * indicate the features this CPU genuinely supports!
945 */
946 switch (c->x86_vendor) {
947 case X86_VENDOR_AMD:
948 init_amd(c);
949 break;
950
951 case X86_VENDOR_INTEL:
952 init_intel(c);
953 break;
954
955 case X86_VENDOR_UNKNOWN:
956 default:
957 display_cacheinfo(c);
958 break;
959 }
960
961 select_idle_routine(c);
962 detect_ht(c);
963
964 /*
965 * On SMP, boot_cpu_data holds the common feature set between
966 * all CPUs; so make sure that we indicate which features are
967 * common between the CPUs. The first time this routine gets
968 * executed, c == &boot_cpu_data.
969 */
970 if (c != &boot_cpu_data) {
971 /* AND the already accumulated flags with these */
972 for (i = 0 ; i < NCAPINTS ; i++)
973 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
974 }
975
976 #ifdef CONFIG_X86_MCE
977 mcheck_init(c);
978 #endif
979 if (c != &boot_cpu_data)
980 mtrr_ap_init();
981 #ifdef CONFIG_NUMA
982 numa_add_cpu(smp_processor_id());
983 #endif
984 }
985
986
987 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
988 {
989 if (c->x86_model_id[0])
990 printk("%s", c->x86_model_id);
991
992 if (c->x86_mask || c->cpuid_level >= 0)
993 printk(" stepping %02x\n", c->x86_mask);
994 else
995 printk("\n");
996 }
997
998 /*
999 * Get CPU information for use by the procfs.
1000 */
1001
1002 static int show_cpuinfo(struct seq_file *m, void *v)
1003 {
1004 struct cpuinfo_x86 *c = v;
1005 int cpu = 0;
1006
1007 /*
1008 * These flag bits must match the definitions in <asm/cpufeature.h>.
1009 * NULL means this bit is undefined or reserved; either way it doesn't
1010 * have meaning as far as Linux is concerned. Note that it's important
1011 * to realize there is a difference between this table and CPUID -- if
1012 * applications want to get the raw CPUID data, they should access
1013 * /dev/cpu/<cpu_nr>/cpuid instead.
1014 */
1015 static const char *const x86_cap_flags[] = {
1016 /* Intel-defined */
1017 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1018 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1019 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1020 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
1021
1022 /* AMD-defined */
1023 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1024 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1025 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1026 NULL, "fxsr_opt", "pdpe1gb", "rdtscp", NULL, "lm",
1027 "3dnowext", "3dnow",
1028
1029 /* Transmeta-defined */
1030 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1031 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1032 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1033 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1034
1035 /* Other (Linux-defined) */
1036 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
1037 NULL, NULL, NULL, NULL,
1038 "constant_tsc", "up", NULL, "arch_perfmon",
1039 "pebs", "bts", NULL, "sync_rdtsc",
1040 "rep_good", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1041 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1042
1043 /* Intel-defined (#2) */
1044 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
1045 "tm2", "ssse3", "cid", NULL, NULL, "cx16", "xtpr", NULL,
1046 NULL, NULL, "dca", "sse4_1", "sse4_2", NULL, NULL, "popcnt",
1047 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1048
1049 /* VIA/Cyrix/Centaur-defined */
1050 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1051 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL, NULL,
1052 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1053 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1054
1055 /* AMD-defined (#2) */
1056 "lahf_lm", "cmp_legacy", "svm", "extapic",
1057 "cr8_legacy", "abm", "sse4a", "misalignsse",
1058 "3dnowprefetch", "osvw", "ibs", "sse5",
1059 "skinit", "wdt", NULL, NULL,
1060 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1061 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1062
1063 /* Auxiliary (Linux-defined) */
1064 "ida", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1065 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1066 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1067 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1068 };
1069 static const char *const x86_power_flags[] = {
1070 "ts", /* temperature sensor */
1071 "fid", /* frequency id control */
1072 "vid", /* voltage id control */
1073 "ttp", /* thermal trip */
1074 "tm",
1075 "stc",
1076 "100mhzsteps",
1077 "hwpstate",
1078 "", /* tsc invariant mapped to constant_tsc */
1079 /* nothing */
1080 };
1081
1082
1083 #ifdef CONFIG_SMP
1084 cpu = c->cpu_index;
1085 #endif
1086
1087 seq_printf(m,"processor\t: %u\n"
1088 "vendor_id\t: %s\n"
1089 "cpu family\t: %d\n"
1090 "model\t\t: %d\n"
1091 "model name\t: %s\n",
1092 (unsigned)cpu,
1093 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1094 c->x86,
1095 (int)c->x86_model,
1096 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1097
1098 if (c->x86_mask || c->cpuid_level >= 0)
1099 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1100 else
1101 seq_printf(m, "stepping\t: unknown\n");
1102
1103 if (cpu_has(c,X86_FEATURE_TSC)) {
1104 unsigned int freq = cpufreq_quick_get((unsigned)cpu);
1105 if (!freq)
1106 freq = cpu_khz;
1107 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1108 freq / 1000, (freq % 1000));
1109 }
1110
1111 /* Cache size */
1112 if (c->x86_cache_size >= 0)
1113 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1114
1115 #ifdef CONFIG_SMP
1116 if (smp_num_siblings * c->x86_max_cores > 1) {
1117 seq_printf(m, "physical id\t: %d\n", c->phys_proc_id);
1118 seq_printf(m, "siblings\t: %d\n",
1119 cpus_weight(per_cpu(cpu_core_map, cpu)));
1120 seq_printf(m, "core id\t\t: %d\n", c->cpu_core_id);
1121 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1122 }
1123 #endif
1124
1125 seq_printf(m,
1126 "fpu\t\t: yes\n"
1127 "fpu_exception\t: yes\n"
1128 "cpuid level\t: %d\n"
1129 "wp\t\t: yes\n"
1130 "flags\t\t:",
1131 c->cpuid_level);
1132
1133 {
1134 int i;
1135 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1136 if (cpu_has(c, i) && x86_cap_flags[i] != NULL)
1137 seq_printf(m, " %s", x86_cap_flags[i]);
1138 }
1139
1140 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1141 c->loops_per_jiffy/(500000/HZ),
1142 (c->loops_per_jiffy/(5000/HZ)) % 100);
1143
1144 if (c->x86_tlbsize > 0)
1145 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1146 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1147 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1148
1149 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1150 c->x86_phys_bits, c->x86_virt_bits);
1151
1152 seq_printf(m, "power management:");
1153 {
1154 unsigned i;
1155 for (i = 0; i < 32; i++)
1156 if (c->x86_power & (1 << i)) {
1157 if (i < ARRAY_SIZE(x86_power_flags) &&
1158 x86_power_flags[i])
1159 seq_printf(m, "%s%s",
1160 x86_power_flags[i][0]?" ":"",
1161 x86_power_flags[i]);
1162 else
1163 seq_printf(m, " [%d]", i);
1164 }
1165 }
1166
1167 seq_printf(m, "\n\n");
1168
1169 return 0;
1170 }
1171
1172 static void *c_start(struct seq_file *m, loff_t *pos)
1173 {
1174 if (*pos == 0) /* just in case, cpu 0 is not the first */
1175 *pos = first_cpu(cpu_online_map);
1176 if ((*pos) < NR_CPUS && cpu_online(*pos))
1177 return &cpu_data(*pos);
1178 return NULL;
1179 }
1180
1181 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1182 {
1183 *pos = next_cpu(*pos, cpu_online_map);
1184 return c_start(m, pos);
1185 }
1186
1187 static void c_stop(struct seq_file *m, void *v)
1188 {
1189 }
1190
1191 struct seq_operations cpuinfo_op = {
1192 .start =c_start,
1193 .next = c_next,
1194 .stop = c_stop,
1195 .show = show_cpuinfo,
1196 };