x86: re-add rdmsrl_safe
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / x86 / kernel / setup_64.c
CommitLineData
1da177e4 1/*
1da177e4 2 * Copyright (C) 1995 Linus Torvalds
1da177e4
LT
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>
894673ee 18#include <linux/screen_info.h>
1da177e4
LT
19#include <linux/ioport.h>
20#include <linux/delay.h>
1da177e4
LT
21#include <linux/init.h>
22#include <linux/initrd.h>
23#include <linux/highmem.h>
24#include <linux/bootmem.h>
25#include <linux/module.h>
26#include <asm/processor.h>
27#include <linux/console.h>
28#include <linux/seq_file.h>
aac04b32 29#include <linux/crash_dump.h>
1da177e4
LT
30#include <linux/root_dev.h>
31#include <linux/pci.h>
5b83683f 32#include <linux/efi.h>
1da177e4
LT
33#include <linux/acpi.h>
34#include <linux/kallsyms.h>
35#include <linux/edd.h>
bbfceef4 36#include <linux/mmzone.h>
5f5609df 37#include <linux/kexec.h>
95235ca2 38#include <linux/cpufreq.h>
e9928674 39#include <linux/dmi.h>
17a941d8 40#include <linux/dma-mapping.h>
681558fd 41#include <linux/ctype.h>
746ef0cd 42#include <linux/uaccess.h>
f212ec4b 43#include <linux/init_ohci1394_dma.h>
bbfceef4 44
1da177e4
LT
45#include <asm/mtrr.h>
46#include <asm/uaccess.h>
47#include <asm/system.h>
e4026440 48#include <asm/vsyscall.h>
1da177e4
LT
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>
aaf23042 56#include <asm/gart.h>
1da177e4
LT
57#include <asm/mpspec.h>
58#include <asm/mmu_context.h>
1da177e4
LT
59#include <asm/proto.h>
60#include <asm/setup.h>
61#include <asm/mach_apic.h>
62#include <asm/numa.h>
2bc0414e 63#include <asm/sections.h>
f2d3efed 64#include <asm/dmi.h>
00bf4098 65#include <asm/cacheflush.h>
af7a78e9 66#include <asm/mce.h>
eee3af4a 67#include <asm/ds.h>
df3825c5 68#include <asm/topology.h>
1da177e4 69
746ef0cd
GOC
70#ifdef CONFIG_PARAVIRT
71#include <asm/paravirt.h>
72#else
73#define ARCH_SETUP
74#endif
75
1da177e4
LT
76/*
77 * Machine setup..
78 */
79
6c231b7b 80struct cpuinfo_x86 boot_cpu_data __read_mostly;
2ee60e17 81EXPORT_SYMBOL(boot_cpu_data);
1da177e4 82
7d851c8d
AK
83__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
84
1da177e4
LT
85unsigned long mmu_cr4_features;
86
1da177e4
LT
87/* Boot loader ID as an integer, for the benefit of proc_dointvec */
88int bootloader_type;
89
90unsigned long saved_video_mode;
91
f039b754
AK
92int force_mwait __cpuinitdata;
93
04e1ba85 94/*
f2d3efed
AK
95 * Early DMI memory
96 */
97int dmi_alloc_index;
98char dmi_alloc_data[DMI_MAX_DATA];
99
1da177e4
LT
100/*
101 * Setup options
102 */
1da177e4 103struct screen_info screen_info;
2ee60e17 104EXPORT_SYMBOL(screen_info);
1da177e4
LT
105struct sys_desc_table_struct {
106 unsigned short length;
107 unsigned char table[0];
108};
109
110struct edid_info edid_info;
ba70710e 111EXPORT_SYMBOL_GPL(edid_info);
1da177e4
LT
112
113extern int root_mountflags;
1da177e4 114
adf48856 115char __initdata command_line[COMMAND_LINE_SIZE];
1da177e4
LT
116
117struct resource standard_io_resources[] = {
118 { .name = "dma1", .start = 0x00, .end = 0x1f,
119 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
120 { .name = "pic1", .start = 0x20, .end = 0x21,
121 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
122 { .name = "timer0", .start = 0x40, .end = 0x43,
123 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
124 { .name = "timer1", .start = 0x50, .end = 0x53,
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
126 { .name = "keyboard", .start = 0x60, .end = 0x6f,
127 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
128 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
129 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
130 { .name = "pic2", .start = 0xa0, .end = 0xa1,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
132 { .name = "dma2", .start = 0xc0, .end = 0xdf,
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
134 { .name = "fpu", .start = 0xf0, .end = 0xff,
135 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
136};
137
1da177e4
LT
138#define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
139
c9cce83d 140static struct resource data_resource = {
1da177e4
LT
141 .name = "Kernel data",
142 .start = 0,
143 .end = 0,
144 .flags = IORESOURCE_RAM,
145};
c9cce83d 146static struct resource code_resource = {
1da177e4
LT
147 .name = "Kernel code",
148 .start = 0,
149 .end = 0,
150 .flags = IORESOURCE_RAM,
151};
c9cce83d 152static struct resource bss_resource = {
00bf4098
BW
153 .name = "Kernel bss",
154 .start = 0,
155 .end = 0,
156 .flags = IORESOURCE_RAM,
157};
1da177e4 158
8c61b900
TG
159static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
160
2c8c0e6b
AK
161#ifdef CONFIG_PROC_VMCORE
162/* elfcorehdr= specifies the location of elf core header
163 * stored by the crashed kernel. This option will be passed
164 * by kexec loader to the capture kernel.
165 */
166static int __init setup_elfcorehdr(char *arg)
681558fd 167{
2c8c0e6b
AK
168 char *end;
169 if (!arg)
170 return -EINVAL;
171 elfcorehdr_addr = memparse(arg, &end);
172 return end > arg ? 0 : -EINVAL;
681558fd 173}
2c8c0e6b 174early_param("elfcorehdr", setup_elfcorehdr);
e2c03888
AK
175#endif
176
2b97690f 177#ifndef CONFIG_NUMA
bbfceef4
MT
178static void __init
179contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
1da177e4 180{
bbfceef4
MT
181 unsigned long bootmap_size, bootmap;
182
bbfceef4 183 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
24a5da73
YL
184 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
185 PAGE_SIZE);
bbfceef4 186 if (bootmap == -1L)
04e1ba85 187 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
bbfceef4 188 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
5cb248ab
MG
189 e820_register_active_regions(0, start_pfn, end_pfn);
190 free_bootmem_with_active_regions(0, end_pfn);
72a7fe39 191 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
04e1ba85 192}
1da177e4
LT
193#endif
194
1da177e4
LT
195#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
196struct edd edd;
197#ifdef CONFIG_EDD_MODULE
198EXPORT_SYMBOL(edd);
199#endif
200/**
201 * copy_edd() - Copy the BIOS EDD information
202 * from boot_params into a safe place.
203 *
204 */
205static inline void copy_edd(void)
206{
30c82645
PA
207 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
208 sizeof(edd.mbr_signature));
209 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
210 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
211 edd.edd_info_nr = boot_params.eddbuf_entries;
1da177e4
LT
212}
213#else
214static inline void copy_edd(void)
215{
216}
217#endif
218
5c3391f9
BW
219#ifdef CONFIG_KEXEC
220static void __init reserve_crashkernel(void)
221{
18a01a3b 222 unsigned long long total_mem;
5c3391f9
BW
223 unsigned long long crash_size, crash_base;
224 int ret;
225
18a01a3b 226 total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
5c3391f9 227
18a01a3b 228 ret = parse_crashkernel(boot_command_line, total_mem,
5c3391f9
BW
229 &crash_size, &crash_base);
230 if (ret == 0 && crash_size) {
18a01a3b 231 if (crash_base <= 0) {
5c3391f9
BW
232 printk(KERN_INFO "crashkernel reservation failed - "
233 "you have to specify a base address\n");
18a01a3b
BW
234 return;
235 }
236
237 if (reserve_bootmem(crash_base, crash_size,
238 BOOTMEM_EXCLUSIVE) < 0) {
239 printk(KERN_INFO "crashkernel reservation failed - "
240 "memory is in use\n");
241 return;
242 }
243
244 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
245 "for crashkernel (System RAM: %ldMB)\n",
246 (unsigned long)(crash_size >> 20),
247 (unsigned long)(crash_base >> 20),
248 (unsigned long)(total_mem >> 20));
249 crashk_res.start = crash_base;
250 crashk_res.end = crash_base + crash_size - 1;
3def3d6d 251 insert_resource(&iomem_resource, &crashk_res);
5c3391f9
BW
252 }
253}
254#else
255static inline void __init reserve_crashkernel(void)
256{}
257#endif
258
746ef0cd 259/* Overridden in paravirt.c if CONFIG_PARAVIRT */
e3cfac84 260void __attribute__((weak)) __init memory_setup(void)
746ef0cd
GOC
261{
262 machine_specific_memory_setup();
263}
264
f212ec4b
BK
265/*
266 * setup_arch - architecture-specific boot-time initializations
267 *
268 * Note: On x86_64, fixmaps are ready for use even before this is called.
269 */
1da177e4
LT
270void __init setup_arch(char **cmdline_p)
271{
04e1ba85
TG
272 unsigned i;
273
adf48856 274 printk(KERN_INFO "Command line: %s\n", boot_command_line);
43c85c9c 275
30c82645
PA
276 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
277 screen_info = boot_params.screen_info;
278 edid_info = boot_params.edid_info;
279 saved_video_mode = boot_params.hdr.vid_mode;
280 bootloader_type = boot_params.hdr.type_of_loader;
1da177e4
LT
281
282#ifdef CONFIG_BLK_DEV_RAM
30c82645
PA
283 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
284 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
285 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
1da177e4 286#endif
5b83683f
HY
287#ifdef CONFIG_EFI
288 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
289 "EL64", 4))
290 efi_enabled = 1;
291#endif
746ef0cd
GOC
292
293 ARCH_SETUP
294
295 memory_setup();
1da177e4
LT
296 copy_edd();
297
30c82645 298 if (!boot_params.hdr.root_flags)
1da177e4
LT
299 root_mountflags &= ~MS_RDONLY;
300 init_mm.start_code = (unsigned long) &_text;
301 init_mm.end_code = (unsigned long) &_etext;
302 init_mm.end_data = (unsigned long) &_edata;
303 init_mm.brk = (unsigned long) &_end;
304
e3ebadd9
LT
305 code_resource.start = virt_to_phys(&_text);
306 code_resource.end = virt_to_phys(&_etext)-1;
307 data_resource.start = virt_to_phys(&_etext);
308 data_resource.end = virt_to_phys(&_edata)-1;
00bf4098
BW
309 bss_resource.start = virt_to_phys(&__bss_start);
310 bss_resource.end = virt_to_phys(&__bss_stop)-1;
1da177e4 311
1da177e4
LT
312 early_identify_cpu(&boot_cpu_data);
313
adf48856 314 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
2c8c0e6b
AK
315 *cmdline_p = command_line;
316
317 parse_early_param();
318
f212ec4b
BK
319#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
320 if (init_ohci1394_dma_early)
321 init_ohci1394_dma_on_all_controllers();
322#endif
323
2c8c0e6b 324 finish_e820_parsing();
9ca33eb6 325
3def3d6d
YL
326 /* after parse_early_param, so could debug it */
327 insert_resource(&iomem_resource, &code_resource);
328 insert_resource(&iomem_resource, &data_resource);
329 insert_resource(&iomem_resource, &bss_resource);
330
aaf23042
YL
331 early_gart_iommu_check();
332
5cb248ab 333 e820_register_active_regions(0, 0, -1UL);
1da177e4
LT
334 /*
335 * partially used pages are not usable - thus
336 * we are rounding upwards:
337 */
338 end_pfn = e820_end_of_ram();
99fc8d42
JB
339 /* update e820 for memory not covered by WB MTRRs */
340 mtrr_bp_init();
341 if (mtrr_trim_uncached_memory(end_pfn)) {
342 e820_register_active_regions(0, 0, -1UL);
343 end_pfn = e820_end_of_ram();
344 }
345
caff0710 346 num_physpages = end_pfn;
1da177e4
LT
347
348 check_efer();
349
cc615032 350 max_pfn_mapped = init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
5b83683f
HY
351 if (efi_enabled)
352 efi_init();
1da177e4 353
2785c8d0 354 vsmp_init();
2785c8d0 355
f2d3efed
AK
356 dmi_scan_machine();
357
b02aae9c
RH
358 io_delay_init();
359
71fff5e6 360#ifdef CONFIG_SMP
df3825c5 361 /* setup to use the early static init tables during kernel startup */
3effef1f
YL
362 x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
363 x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
e8c10ef9 364#ifdef CONFIG_NUMA
3effef1f 365 x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
71fff5e6 366#endif
e8c10ef9 367#endif
71fff5e6 368
888ba6c6 369#ifdef CONFIG_ACPI
1da177e4
LT
370 /*
371 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
372 * Call this early for SRAT node setup.
373 */
374 acpi_boot_table_init();
375#endif
376
caff0710
JB
377 /* How many end-of-memory variables you have, grandma! */
378 max_low_pfn = end_pfn;
379 max_pfn = end_pfn;
380 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
381
5cb248ab
MG
382 /* Remove active ranges so rediscovery with NUMA-awareness happens */
383 remove_all_active_ranges();
384
1da177e4
LT
385#ifdef CONFIG_ACPI_NUMA
386 /*
387 * Parse SRAT to discover nodes.
388 */
389 acpi_numa_init();
390#endif
391
2b97690f 392#ifdef CONFIG_NUMA
04e1ba85 393 numa_initmem_init(0, end_pfn);
1da177e4 394#else
bbfceef4 395 contig_initmem_init(0, end_pfn);
1da177e4
LT
396#endif
397
75175278 398 early_res_to_bootmem();
1da177e4 399
673d5b43 400#ifdef CONFIG_ACPI_SLEEP
1da177e4 401 /*
04e1ba85 402 * Reserve low memory region for sleep support.
1da177e4 403 */
04e1ba85
TG
404 acpi_reserve_bootmem();
405#endif
5b83683f 406
a3828064 407 if (efi_enabled)
5b83683f 408 efi_reserve_bootmem();
5b83683f 409
04e1ba85
TG
410 /*
411 * Find and reserve possible boot-time SMP configuration:
412 */
1da177e4 413 find_smp_config();
1da177e4 414#ifdef CONFIG_BLK_DEV_INITRD
30c82645
PA
415 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
416 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
417 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
418 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
419 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
420
421 if (ramdisk_end <= end_of_mem) {
422 reserve_bootmem_generic(ramdisk_image, ramdisk_size);
423 initrd_start = ramdisk_image + PAGE_OFFSET;
424 initrd_end = initrd_start+ramdisk_size;
425 } else {
75175278
AK
426 /* Assumes everything on node 0 */
427 free_bootmem(ramdisk_image, ramdisk_size);
1da177e4 428 printk(KERN_ERR "initrd extends beyond end of memory "
30c82645
PA
429 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
430 ramdisk_end, end_of_mem);
1da177e4
LT
431 initrd_start = 0;
432 }
433 }
434#endif
5c3391f9 435 reserve_crashkernel();
1da177e4 436 paging_init();
e4026440 437 map_vsyscall();
1da177e4 438
dfa4698c 439 early_quirks();
1da177e4 440
888ba6c6 441#ifdef CONFIG_ACPI
1da177e4
LT
442 /*
443 * Read APIC and some other early information from ACPI tables.
444 */
445 acpi_boot_init();
446#endif
447
05b3cbd8
RT
448 init_cpu_to_node();
449
1da177e4
LT
450 /*
451 * get boot-time SMP configuration:
452 */
453 if (smp_found_config)
454 get_smp_config();
455 init_apic_mappings();
3e35a0e5 456 ioapic_init_mappings();
1da177e4
LT
457
458 /*
fc986db4 459 * We trust e820 completely. No explicit ROM probing in memory.
04e1ba85 460 */
3def3d6d 461 e820_reserve_resources();
e8eff5ac 462 e820_mark_nosave_regions();
1da177e4 463
1da177e4 464 /* request I/O space for devices used on all i[345]86 PCs */
9d0ef4fd 465 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
1da177e4 466 request_resource(&ioport_resource, &standard_io_resources[i]);
1da177e4 467
a1e97782 468 e820_setup_gap();
1da177e4 469
1da177e4
LT
470#ifdef CONFIG_VT
471#if defined(CONFIG_VGA_CONSOLE)
5b83683f
HY
472 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
473 conswitchp = &vga_con;
1da177e4
LT
474#elif defined(CONFIG_DUMMY_CONSOLE)
475 conswitchp = &dummy_con;
476#endif
477#endif
478}
479
e6982c67 480static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
1da177e4
LT
481{
482 unsigned int *v;
483
ebfcaa96 484 if (c->extended_cpuid_level < 0x80000004)
1da177e4
LT
485 return 0;
486
487 v = (unsigned int *) c->x86_model_id;
488 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
489 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
490 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
491 c->x86_model_id[48] = 0;
492 return 1;
493}
494
495
e6982c67 496static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
1da177e4
LT
497{
498 unsigned int n, dummy, eax, ebx, ecx, edx;
499
ebfcaa96 500 n = c->extended_cpuid_level;
1da177e4
LT
501
502 if (n >= 0x80000005) {
503 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
04e1ba85
TG
504 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
505 "D cache %dK (%d bytes/line)\n",
506 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
507 c->x86_cache_size = (ecx>>24) + (edx>>24);
1da177e4
LT
508 /* On K8 L1 TLB is inclusive, so don't count it */
509 c->x86_tlbsize = 0;
510 }
511
512 if (n >= 0x80000006) {
513 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
514 ecx = cpuid_ecx(0x80000006);
515 c->x86_cache_size = ecx >> 16;
516 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
517
518 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
519 c->x86_cache_size, ecx & 0xFF);
520 }
1da177e4 521 if (n >= 0x80000008) {
04e1ba85 522 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
1da177e4
LT
523 c->x86_virt_bits = (eax >> 8) & 0xff;
524 c->x86_phys_bits = eax & 0xff;
525 }
526}
527
3f098c26 528#ifdef CONFIG_NUMA
08acb672 529static int __cpuinit nearby_node(int apicid)
3f098c26 530{
04e1ba85
TG
531 int i, node;
532
3f098c26 533 for (i = apicid - 1; i >= 0; i--) {
04e1ba85 534 node = apicid_to_node[i];
3f098c26
AK
535 if (node != NUMA_NO_NODE && node_online(node))
536 return node;
537 }
538 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
04e1ba85 539 node = apicid_to_node[i];
3f098c26
AK
540 if (node != NUMA_NO_NODE && node_online(node))
541 return node;
542 }
543 return first_node(node_online_map); /* Shouldn't happen */
544}
545#endif
546
63518644
AK
547/*
548 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
549 * Assumes number of cores is a power of two.
550 */
adb8daed 551static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
63518644
AK
552{
553#ifdef CONFIG_SMP
b41e2939 554 unsigned bits;
3f098c26 555#ifdef CONFIG_NUMA
f3fa8ebc 556 int cpu = smp_processor_id();
3f098c26 557 int node = 0;
60c1bc82 558 unsigned apicid = hard_smp_processor_id();
3f098c26 559#endif
a860b63c 560 bits = c->x86_coreid_bits;
b41e2939
AK
561
562 /* Low order bits define the core id (index of core in socket) */
01aaea1a
YL
563 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
564 /* Convert the initial APIC ID into the socket ID */
565 c->phys_proc_id = c->initial_apicid >> bits;
63518644
AK
566
567#ifdef CONFIG_NUMA
04e1ba85
TG
568 node = c->phys_proc_id;
569 if (apicid_to_node[apicid] != NUMA_NO_NODE)
570 node = apicid_to_node[apicid];
571 if (!node_online(node)) {
572 /* Two possibilities here:
573 - The CPU is missing memory and no node was created.
574 In that case try picking one from a nearby CPU
575 - The APIC IDs differ from the HyperTransport node IDs
576 which the K8 northbridge parsing fills in.
577 Assume they are all increased by a constant offset,
578 but in the same order as the HT nodeids.
579 If that doesn't result in a usable node fall back to the
580 path for the previous case. */
581
01aaea1a 582 int ht_nodeid = c->initial_apicid;
04e1ba85
TG
583
584 if (ht_nodeid >= 0 &&
585 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
586 node = apicid_to_node[ht_nodeid];
587 /* Pick a nearby node */
588 if (!node_online(node))
589 node = nearby_node(apicid);
590 }
69d81fcd 591 numa_set_node(cpu, node);
3f098c26 592
e42f9437 593 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
63518644 594#endif
63518644
AK
595#endif
596}
1da177e4 597
2b16a235 598static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
a860b63c
YL
599{
600#ifdef CONFIG_SMP
601 unsigned bits, ecx;
602
603 /* Multi core CPU? */
604 if (c->extended_cpuid_level < 0x80000008)
605 return;
606
607 ecx = cpuid_ecx(0x80000008);
608
609 c->x86_max_cores = (ecx & 0xff) + 1;
610
611 /* CPU telling us the core id bits shift? */
612 bits = (ecx >> 12) & 0xF;
613
614 /* Otherwise recompute */
615 if (bits == 0) {
616 while ((1 << bits) < c->x86_max_cores)
617 bits++;
618 }
619
620 c->x86_coreid_bits = bits;
621
622#endif
623}
624
fb79d22e
TG
625#define ENABLE_C1E_MASK 0x18000000
626#define CPUID_PROCESSOR_SIGNATURE 1
627#define CPUID_XFAM 0x0ff00000
628#define CPUID_XFAM_K8 0x00000000
629#define CPUID_XFAM_10H 0x00100000
630#define CPUID_XFAM_11H 0x00200000
631#define CPUID_XMOD 0x000f0000
632#define CPUID_XMOD_REV_F 0x00040000
633
634/* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
635static __cpuinit int amd_apic_timer_broken(void)
636{
04e1ba85
TG
637 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
638
fb79d22e
TG
639 switch (eax & CPUID_XFAM) {
640 case CPUID_XFAM_K8:
641 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
642 break;
643 case CPUID_XFAM_10H:
644 case CPUID_XFAM_11H:
645 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
646 if (lo & ENABLE_C1E_MASK)
647 return 1;
648 break;
649 default:
650 /* err on the side of caution */
651 return 1;
652 }
653 return 0;
654}
655
2b16a235
AK
656static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
657{
658 early_init_amd_mc(c);
659
660 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
661 if (c->x86_power & (1<<8))
662 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
663}
664
ed77504b 665static void __cpuinit init_amd(struct cpuinfo_x86 *c)
1da177e4 666{
7bcd3f34 667 unsigned level;
1da177e4 668
bc5e8fdf
LT
669#ifdef CONFIG_SMP
670 unsigned long value;
671
7d318d77
AK
672 /*
673 * Disable TLB flush filter by setting HWCR.FFDIS on K8
674 * bit 6 of msr C001_0015
04e1ba85 675 *
7d318d77
AK
676 * Errata 63 for SH-B3 steppings
677 * Errata 122 for all steppings (F+ have it disabled by default)
678 */
679 if (c->x86 == 15) {
680 rdmsrl(MSR_K8_HWCR, value);
681 value |= 1 << 6;
682 wrmsrl(MSR_K8_HWCR, value);
683 }
bc5e8fdf
LT
684#endif
685
1da177e4
LT
686 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
687 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
9716951e 688 clear_cpu_cap(c, 0*32+31);
04e1ba85 689
7bcd3f34
AK
690 /* On C+ stepping K8 rep microcode works well for copy/memset */
691 level = cpuid_eax(1);
04e1ba85
TG
692 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
693 level >= 0x0f58))
53756d37 694 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
99741faa 695 if (c->x86 == 0x10 || c->x86 == 0x11)
53756d37 696 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
7bcd3f34 697
18bd057b
AK
698 /* Enable workaround for FXSAVE leak */
699 if (c->x86 >= 6)
53756d37 700 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
18bd057b 701
e42f9437
RS
702 level = get_model_name(c);
703 if (!level) {
04e1ba85 704 switch (c->x86) {
1da177e4
LT
705 case 15:
706 /* Should distinguish Models here, but this is only
707 a fallback anyways. */
708 strcpy(c->x86_model_id, "Hammer");
04e1ba85
TG
709 break;
710 }
711 }
1da177e4
LT
712 display_cacheinfo(c);
713
faee9a5d
AK
714 /* Multi core CPU? */
715 if (c->extended_cpuid_level >= 0x80000008)
63518644 716 amd_detect_cmp(c);
1da177e4 717
67cddd94
AK
718 if (c->extended_cpuid_level >= 0x80000006 &&
719 (cpuid_edx(0x80000006) & 0xf000))
720 num_cache_leaves = 4;
721 else
722 num_cache_leaves = 3;
2049336f 723
0bd8acd1 724 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
53756d37 725 set_cpu_cap(c, X86_FEATURE_K8);
0bd8acd1 726
de421863
AK
727 /* MFENCE stops RDTSC speculation */
728 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
f039b754 729
fb79d22e
TG
730 if (amd_apic_timer_broken())
731 disable_apic_timer = 1;
1da177e4
LT
732}
733
1a53905a 734void __cpuinit detect_ht(struct cpuinfo_x86 *c)
1da177e4
LT
735{
736#ifdef CONFIG_SMP
04e1ba85
TG
737 u32 eax, ebx, ecx, edx;
738 int index_msb, core_bits;
94605eff
SS
739
740 cpuid(1, &eax, &ebx, &ecx, &edx);
741
94605eff 742
e42f9437 743 if (!cpu_has(c, X86_FEATURE_HT))
1da177e4 744 return;
04e1ba85 745 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
e42f9437 746 goto out;
1da177e4 747
1da177e4 748 smp_num_siblings = (ebx & 0xff0000) >> 16;
94605eff 749
1da177e4
LT
750 if (smp_num_siblings == 1) {
751 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
04e1ba85 752 } else if (smp_num_siblings > 1) {
94605eff 753
1da177e4 754 if (smp_num_siblings > NR_CPUS) {
04e1ba85
TG
755 printk(KERN_WARNING "CPU: Unsupported number of "
756 "siblings %d", smp_num_siblings);
1da177e4
LT
757 smp_num_siblings = 1;
758 return;
759 }
94605eff
SS
760
761 index_msb = get_count_order(smp_num_siblings);
f3fa8ebc 762 c->phys_proc_id = phys_pkg_id(index_msb);
3dd9d514 763
94605eff 764 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
3dd9d514 765
04e1ba85 766 index_msb = get_count_order(smp_num_siblings);
94605eff
SS
767
768 core_bits = get_count_order(c->x86_max_cores);
3dd9d514 769
f3fa8ebc 770 c->cpu_core_id = phys_pkg_id(index_msb) &
94605eff 771 ((1 << core_bits) - 1);
1da177e4 772 }
e42f9437
RS
773out:
774 if ((c->x86_max_cores * smp_num_siblings) > 1) {
04e1ba85
TG
775 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
776 c->phys_proc_id);
777 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
778 c->cpu_core_id);
e42f9437
RS
779 }
780
1da177e4
LT
781#endif
782}
783
3dd9d514
AK
784/*
785 * find out the number of processor cores on the die
786 */
e6982c67 787static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
3dd9d514 788{
2bbc419f 789 unsigned int eax, t;
3dd9d514
AK
790
791 if (c->cpuid_level < 4)
792 return 1;
793
2bbc419f 794 cpuid_count(4, 0, &eax, &t, &t, &t);
3dd9d514
AK
795
796 if (eax & 0x1f)
797 return ((eax >> 26) + 1);
798 else
799 return 1;
800}
801
04d733bd 802static void __cpuinit srat_detect_node(void)
df0cc26b
AK
803{
804#ifdef CONFIG_NUMA
ddea7be0 805 unsigned node;
df0cc26b 806 int cpu = smp_processor_id();
e42f9437 807 int apicid = hard_smp_processor_id();
df0cc26b
AK
808
809 /* Don't do the funky fallback heuristics the AMD version employs
810 for now. */
e42f9437 811 node = apicid_to_node[apicid];
475613b9 812 if (node == NUMA_NO_NODE || !node_online(node))
0d015324 813 node = first_node(node_online_map);
69d81fcd 814 numa_set_node(cpu, node);
df0cc26b 815
c31fbb1a 816 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
df0cc26b
AK
817#endif
818}
819
2b16a235
AK
820static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
821{
822 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
823 (c->x86 == 0x6 && c->x86_model >= 0x0e))
9716951e 824 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
2b16a235
AK
825}
826
e6982c67 827static void __cpuinit init_intel(struct cpuinfo_x86 *c)
1da177e4
LT
828{
829 /* Cache sizes */
830 unsigned n;
831
832 init_intel_cacheinfo(c);
04e1ba85 833 if (c->cpuid_level > 9) {
0080e667
VP
834 unsigned eax = cpuid_eax(10);
835 /* Check for version and the number of counters */
836 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
53756d37 837 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
0080e667
VP
838 }
839
36b2a8d5
SE
840 if (cpu_has_ds) {
841 unsigned int l1, l2;
842 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
ee58fad5 843 if (!(l1 & (1<<11)))
53756d37 844 set_cpu_cap(c, X86_FEATURE_BTS);
36b2a8d5 845 if (!(l1 & (1<<12)))
53756d37 846 set_cpu_cap(c, X86_FEATURE_PEBS);
36b2a8d5
SE
847 }
848
eee3af4a
MM
849
850 if (cpu_has_bts)
851 ds_init_intel(c);
852
ebfcaa96 853 n = c->extended_cpuid_level;
1da177e4
LT
854 if (n >= 0x80000008) {
855 unsigned eax = cpuid_eax(0x80000008);
856 c->x86_virt_bits = (eax >> 8) & 0xff;
857 c->x86_phys_bits = eax & 0xff;
af9c142d
SL
858 /* CPUID workaround for Intel 0F34 CPU */
859 if (c->x86_vendor == X86_VENDOR_INTEL &&
860 c->x86 == 0xF && c->x86_model == 0x3 &&
861 c->x86_mask == 0x4)
862 c->x86_phys_bits = 36;
1da177e4
LT
863 }
864
865 if (c->x86 == 15)
866 c->x86_cache_alignment = c->x86_clflush_size * 2;
27fbe5b2 867 if (c->x86 == 6)
53756d37 868 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
707fa8ed 869 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
04e1ba85 870 c->x86_max_cores = intel_num_cpu_cores(c);
df0cc26b
AK
871
872 srat_detect_node();
1da177e4
LT
873}
874
672289e9 875static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1da177e4
LT
876{
877 char *v = c->x86_vendor_id;
878
879 if (!strcmp(v, "AuthenticAMD"))
880 c->x86_vendor = X86_VENDOR_AMD;
881 else if (!strcmp(v, "GenuineIntel"))
882 c->x86_vendor = X86_VENDOR_INTEL;
883 else
884 c->x86_vendor = X86_VENDOR_UNKNOWN;
885}
886
1da177e4
LT
887/* Do some early cpuid on the boot CPU to get some parameter that are
888 needed before check_bugs. Everything advanced is in identify_cpu
889 below. */
8c61b900 890static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1da177e4 891{
a860b63c 892 u32 tfms, xlvl;
1da177e4
LT
893
894 c->loops_per_jiffy = loops_per_jiffy;
895 c->x86_cache_size = -1;
896 c->x86_vendor = X86_VENDOR_UNKNOWN;
897 c->x86_model = c->x86_mask = 0; /* So far unknown... */
898 c->x86_vendor_id[0] = '\0'; /* Unset */
899 c->x86_model_id[0] = '\0'; /* Unset */
900 c->x86_clflush_size = 64;
901 c->x86_cache_alignment = c->x86_clflush_size;
94605eff 902 c->x86_max_cores = 1;
a860b63c 903 c->x86_coreid_bits = 0;
ebfcaa96 904 c->extended_cpuid_level = 0;
1da177e4
LT
905 memset(&c->x86_capability, 0, sizeof c->x86_capability);
906
907 /* Get vendor name */
908 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
909 (unsigned int *)&c->x86_vendor_id[0],
910 (unsigned int *)&c->x86_vendor_id[8],
911 (unsigned int *)&c->x86_vendor_id[4]);
04e1ba85 912
1da177e4
LT
913 get_cpu_vendor(c);
914
915 /* Initialize the standard set of capabilities */
916 /* Note that the vendor-specific code below might override */
917
918 /* Intel-defined flags: level 0x00000001 */
919 if (c->cpuid_level >= 0x00000001) {
920 __u32 misc;
921 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
922 &c->x86_capability[0]);
923 c->x86 = (tfms >> 8) & 0xf;
924 c->x86_model = (tfms >> 4) & 0xf;
925 c->x86_mask = tfms & 0xf;
f5f786d0 926 if (c->x86 == 0xf)
1da177e4 927 c->x86 += (tfms >> 20) & 0xff;
f5f786d0 928 if (c->x86 >= 0x6)
1da177e4 929 c->x86_model += ((tfms >> 16) & 0xF) << 4;
9716951e 930 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
1da177e4 931 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1da177e4
LT
932 } else {
933 /* Have CPUID level 0 only - unheard of */
934 c->x86 = 4;
935 }
a158608b 936
01aaea1a 937 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
a158608b 938#ifdef CONFIG_SMP
01aaea1a 939 c->phys_proc_id = c->initial_apicid;
a158608b 940#endif
1da177e4
LT
941 /* AMD-defined flags: level 0x80000001 */
942 xlvl = cpuid_eax(0x80000000);
ebfcaa96 943 c->extended_cpuid_level = xlvl;
1da177e4
LT
944 if ((xlvl & 0xffff0000) == 0x80000000) {
945 if (xlvl >= 0x80000001) {
946 c->x86_capability[1] = cpuid_edx(0x80000001);
5b7abc6f 947 c->x86_capability[6] = cpuid_ecx(0x80000001);
1da177e4
LT
948 }
949 if (xlvl >= 0x80000004)
950 get_model_name(c); /* Default name */
951 }
952
953 /* Transmeta-defined flags: level 0x80860001 */
954 xlvl = cpuid_eax(0x80860000);
955 if ((xlvl & 0xffff0000) == 0x80860000) {
956 /* Don't set x86_cpuid_level here for now to not confuse. */
957 if (xlvl >= 0x80860001)
958 c->x86_capability[2] = cpuid_edx(0x80860001);
959 }
960
9566e91d
AH
961 c->extended_cpuid_level = cpuid_eax(0x80000000);
962 if (c->extended_cpuid_level >= 0x80000007)
963 c->x86_power = cpuid_edx(0x80000007);
964
9307caca
YL
965
966 clear_cpu_cap(c, X86_FEATURE_PAT);
967
a860b63c
YL
968 switch (c->x86_vendor) {
969 case X86_VENDOR_AMD:
970 early_init_amd(c);
9307caca
YL
971 if (c->x86 >= 0xf && c->x86 <= 0x11)
972 set_cpu_cap(c, X86_FEATURE_PAT);
a860b63c 973 break;
71617bf1
YL
974 case X86_VENDOR_INTEL:
975 early_init_intel(c);
9307caca
YL
976 if (c->x86 == 0xF || (c->x86 == 6 && c->x86_model >= 15))
977 set_cpu_cap(c, X86_FEATURE_PAT);
71617bf1 978 break;
a860b63c
YL
979 }
980
981}
982
983/*
984 * This does the hard work of actually picking apart the CPU stuff...
985 */
986void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
987{
988 int i;
989
990 early_identify_cpu(c);
991
1d67953f
VP
992 init_scattered_cpuid_features(c);
993
1e9f28fa
SS
994 c->apicid = phys_pkg_id(0);
995
1da177e4
LT
996 /*
997 * Vendor-specific initialization. In this section we
998 * canonicalize the feature flags, meaning if there are
999 * features a certain CPU supports which CPUID doesn't
1000 * tell us, CPUID claiming incorrect flags, or other bugs,
1001 * we handle them here.
1002 *
1003 * At the end of this section, c->x86_capability better
1004 * indicate the features this CPU genuinely supports!
1005 */
1006 switch (c->x86_vendor) {
1007 case X86_VENDOR_AMD:
1008 init_amd(c);
1009 break;
1010
1011 case X86_VENDOR_INTEL:
1012 init_intel(c);
1013 break;
1014
1015 case X86_VENDOR_UNKNOWN:
1016 default:
1017 display_cacheinfo(c);
1018 break;
1019 }
1020
04e1ba85 1021 detect_ht(c);
1da177e4
LT
1022
1023 /*
1024 * On SMP, boot_cpu_data holds the common feature set between
1025 * all CPUs; so make sure that we indicate which features are
1026 * common between the CPUs. The first time this routine gets
1027 * executed, c == &boot_cpu_data.
1028 */
1029 if (c != &boot_cpu_data) {
1030 /* AND the already accumulated flags with these */
04e1ba85 1031 for (i = 0; i < NCAPINTS; i++)
1da177e4
LT
1032 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1033 }
1034
7d851c8d
AK
1035 /* Clear all flags overriden by options */
1036 for (i = 0; i < NCAPINTS; i++)
12c247a6 1037 c->x86_capability[i] &= ~cleared_cpu_caps[i];
7d851c8d 1038
1da177e4
LT
1039#ifdef CONFIG_X86_MCE
1040 mcheck_init(c);
1041#endif
74ff305b
HS
1042 select_idle_routine(c);
1043
1da177e4 1044#ifdef CONFIG_NUMA
3019e8eb 1045 numa_add_cpu(smp_processor_id());
1da177e4 1046#endif
2b16a235 1047
1da177e4 1048}
1da177e4 1049
7a636af6
GOC
1050void __cpuinit identify_boot_cpu(void)
1051{
1052 identify_cpu(&boot_cpu_data);
1053}
1054
1055void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
1056{
1057 BUG_ON(c == &boot_cpu_data);
1058 identify_cpu(c);
1059 mtrr_ap_init();
1060}
1061
191679fd
AK
1062static __init int setup_noclflush(char *arg)
1063{
1064 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1065 return 1;
1066}
1067__setup("noclflush", setup_noclflush);
1068
e6982c67 1069void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1da177e4
LT
1070{
1071 if (c->x86_model_id[0])
d8ff0bbf 1072 printk(KERN_CONT "%s", c->x86_model_id);
1da177e4 1073
04e1ba85
TG
1074 if (c->x86_mask || c->cpuid_level >= 0)
1075 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1da177e4 1076 else
04e1ba85 1077 printk(KERN_CONT "\n");
1da177e4
LT
1078}
1079
ac72e788
AK
1080static __init int setup_disablecpuid(char *arg)
1081{
1082 int bit;
1083 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1084 setup_clear_cpu_cap(bit);
1085 else
1086 return 0;
1087 return 1;
1088}
1089__setup("clearcpuid=", setup_disablecpuid);