]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - arch/x86/kernel/cpu/common.c
x86/xen: Probe target addresses in set_aliased_prot() before the hypercall
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kernel / cpu / common.c
CommitLineData
f0fc4aff 1#include <linux/bootmem.h>
9766cdbc 2#include <linux/linkage.h>
f0fc4aff 3#include <linux/bitops.h>
9766cdbc 4#include <linux/kernel.h>
f0fc4aff 5#include <linux/module.h>
9766cdbc
JSR
6#include <linux/percpu.h>
7#include <linux/string.h>
ee098e1a 8#include <linux/ctype.h>
1da177e4 9#include <linux/delay.h>
9766cdbc
JSR
10#include <linux/sched.h>
11#include <linux/init.h>
0f46efeb 12#include <linux/kprobes.h>
9766cdbc 13#include <linux/kgdb.h>
1da177e4 14#include <linux/smp.h>
9766cdbc
JSR
15#include <linux/io.h>
16
17#include <asm/stackprotector.h>
cdd6c482 18#include <asm/perf_event.h>
1da177e4 19#include <asm/mmu_context.h>
49d859d7 20#include <asm/archrandom.h>
9766cdbc
JSR
21#include <asm/hypervisor.h>
22#include <asm/processor.h>
1e02ce4c 23#include <asm/tlbflush.h>
f649e938 24#include <asm/debugreg.h>
9766cdbc 25#include <asm/sections.h>
f40c3300 26#include <asm/vsyscall.h>
8bdbd962
AC
27#include <linux/topology.h>
28#include <linux/cpumask.h>
9766cdbc 29#include <asm/pgtable.h>
60063497 30#include <linux/atomic.h>
9766cdbc
JSR
31#include <asm/proto.h>
32#include <asm/setup.h>
33#include <asm/apic.h>
34#include <asm/desc.h>
78f7f1e5 35#include <asm/fpu/internal.h>
27b07da7 36#include <asm/mtrr.h>
8bdbd962 37#include <linux/numa.h>
9766cdbc
JSR
38#include <asm/asm.h>
39#include <asm/cpu.h>
a03a3e28 40#include <asm/mce.h>
9766cdbc 41#include <asm/msr.h>
8d4a4300 42#include <asm/pat.h>
d288e1cf
FY
43#include <asm/microcode.h>
44#include <asm/microcode_intel.h>
e641f5f5
IM
45
46#ifdef CONFIG_X86_LOCAL_APIC
bdbcdd48 47#include <asm/uv/uv.h>
1da177e4
LT
48#endif
49
50#include "cpu.h"
51
c2d1cec1 52/* all of these masks are initialized in setup_cpu_local_masks() */
c2d1cec1 53cpumask_var_t cpu_initialized_mask;
9766cdbc
JSR
54cpumask_var_t cpu_callout_mask;
55cpumask_var_t cpu_callin_mask;
c2d1cec1
MT
56
57/* representing cpus for which sibling maps can be computed */
58cpumask_var_t cpu_sibling_setup_mask;
59
2f2f52ba 60/* correctly size the local cpu masks */
4369f1fb 61void __init setup_cpu_local_masks(void)
2f2f52ba
BG
62{
63 alloc_bootmem_cpumask_var(&cpu_initialized_mask);
64 alloc_bootmem_cpumask_var(&cpu_callin_mask);
65 alloc_bootmem_cpumask_var(&cpu_callout_mask);
66 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
67}
68
148f9bb8 69static void default_init(struct cpuinfo_x86 *c)
e8055139
OZ
70{
71#ifdef CONFIG_X86_64
27c13ece 72 cpu_detect_cache_sizes(c);
e8055139
OZ
73#else
74 /* Not much we can do here... */
75 /* Check if at least it has cpuid */
76 if (c->cpuid_level == -1) {
77 /* No cpuid. It must be an ancient CPU */
78 if (c->x86 == 4)
79 strcpy(c->x86_model_id, "486");
80 else if (c->x86 == 3)
81 strcpy(c->x86_model_id, "386");
82 }
83#endif
84}
85
148f9bb8 86static const struct cpu_dev default_cpu = {
e8055139
OZ
87 .c_init = default_init,
88 .c_vendor = "Unknown",
89 .c_x86_vendor = X86_VENDOR_UNKNOWN,
90};
91
148f9bb8 92static const struct cpu_dev *this_cpu = &default_cpu;
0a488a53 93
06deef89 94DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
950ad7ff 95#ifdef CONFIG_X86_64
06deef89
BG
96 /*
97 * We need valid kernel segments for data and code in long mode too
98 * IRET will check the segment types kkeil 2000/10/28
99 * Also sysret mandates a special GDT layout
100 *
9766cdbc 101 * TLS descriptors are currently at a different place compared to i386.
06deef89
BG
102 * Hopefully nobody expects them at a fixed place (Wine?)
103 */
1e5de182
AM
104 [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
105 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
106 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
107 [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
108 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
109 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
950ad7ff 110#else
1e5de182
AM
111 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
112 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
113 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
114 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
bf504672
RR
115 /*
116 * Segments used for calling PnP BIOS have byte granularity.
117 * They code segments and data segments have fixed 64k limits,
118 * the transfer segment sizes are set at run time.
119 */
6842ef0e 120 /* 32-bit code */
1e5de182 121 [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
6842ef0e 122 /* 16-bit code */
1e5de182 123 [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
6842ef0e 124 /* 16-bit data */
1e5de182 125 [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff),
6842ef0e 126 /* 16-bit data */
1e5de182 127 [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0),
6842ef0e 128 /* 16-bit data */
1e5de182 129 [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0),
bf504672
RR
130 /*
131 * The APM segments have byte granularity and their bases
132 * are set at run time. All have 64k limits.
133 */
6842ef0e 134 /* 32-bit code */
1e5de182 135 [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
bf504672 136 /* 16-bit code */
1e5de182 137 [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
6842ef0e 138 /* data */
72c4d853 139 [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff),
bf504672 140
1e5de182
AM
141 [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
142 [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
60a5317f 143 GDT_STACK_CANARY_INIT
950ad7ff 144#endif
06deef89 145} };
7a61d35d 146EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
ae1ee11b 147
8c3641e9 148static int __init x86_mpx_setup(char *s)
0c752a93 149{
8c3641e9 150 /* require an exact match without trailing characters */
2cd3949f
DH
151 if (strlen(s))
152 return 0;
0c752a93 153
8c3641e9
DH
154 /* do not emit a message if the feature is not present */
155 if (!boot_cpu_has(X86_FEATURE_MPX))
156 return 1;
6bad06b7 157
8c3641e9
DH
158 setup_clear_cpu_cap(X86_FEATURE_MPX);
159 pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
b6f42a4a
FY
160 return 1;
161}
8c3641e9 162__setup("nompx", x86_mpx_setup);
b6f42a4a 163
ba51dced 164#ifdef CONFIG_X86_32
148f9bb8
PG
165static int cachesize_override = -1;
166static int disable_x86_serial_nr = 1;
1da177e4 167
0a488a53
YL
168static int __init cachesize_setup(char *str)
169{
170 get_option(&str, &cachesize_override);
171 return 1;
172}
173__setup("cachesize=", cachesize_setup);
174
0a488a53
YL
175static int __init x86_sep_setup(char *s)
176{
177 setup_clear_cpu_cap(X86_FEATURE_SEP);
178 return 1;
179}
180__setup("nosep", x86_sep_setup);
181
182/* Standard macro to see if a specific flag is changeable */
183static inline int flag_is_changeable_p(u32 flag)
184{
185 u32 f1, f2;
186
94f6bac1
KH
187 /*
188 * Cyrix and IDT cpus allow disabling of CPUID
189 * so the code below may return different results
190 * when it is executed before and after enabling
191 * the CPUID. Add "volatile" to not allow gcc to
192 * optimize the subsequent calls to this function.
193 */
0f3fa48a
IM
194 asm volatile ("pushfl \n\t"
195 "pushfl \n\t"
196 "popl %0 \n\t"
197 "movl %0, %1 \n\t"
198 "xorl %2, %0 \n\t"
199 "pushl %0 \n\t"
200 "popfl \n\t"
201 "pushfl \n\t"
202 "popl %0 \n\t"
203 "popfl \n\t"
204
94f6bac1
KH
205 : "=&r" (f1), "=&r" (f2)
206 : "ir" (flag));
0a488a53
YL
207
208 return ((f1^f2) & flag) != 0;
209}
210
211/* Probe for the CPUID instruction */
148f9bb8 212int have_cpuid_p(void)
0a488a53
YL
213{
214 return flag_is_changeable_p(X86_EFLAGS_ID);
215}
216
148f9bb8 217static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
0a488a53 218{
0f3fa48a
IM
219 unsigned long lo, hi;
220
221 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
222 return;
223
224 /* Disable processor serial number: */
225
226 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
227 lo |= 0x200000;
228 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
229
230 printk(KERN_NOTICE "CPU serial number disabled.\n");
231 clear_cpu_cap(c, X86_FEATURE_PN);
232
233 /* Disabling the serial number may affect the cpuid level */
234 c->cpuid_level = cpuid_eax(0);
0a488a53
YL
235}
236
237static int __init x86_serial_nr_setup(char *s)
238{
239 disable_x86_serial_nr = 0;
240 return 1;
241}
242__setup("serialnumber", x86_serial_nr_setup);
ba51dced 243#else
102bbe3a
YL
244static inline int flag_is_changeable_p(u32 flag)
245{
246 return 1;
247}
102bbe3a
YL
248static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
249{
250}
ba51dced 251#endif
0a488a53 252
de5397ad
FY
253static __init int setup_disable_smep(char *arg)
254{
b2cc2a07 255 setup_clear_cpu_cap(X86_FEATURE_SMEP);
de5397ad
FY
256 return 1;
257}
258__setup("nosmep", setup_disable_smep);
259
b2cc2a07 260static __always_inline void setup_smep(struct cpuinfo_x86 *c)
de5397ad 261{
b2cc2a07 262 if (cpu_has(c, X86_FEATURE_SMEP))
375074cc 263 cr4_set_bits(X86_CR4_SMEP);
de5397ad
FY
264}
265
52b6179a
PA
266static __init int setup_disable_smap(char *arg)
267{
b2cc2a07 268 setup_clear_cpu_cap(X86_FEATURE_SMAP);
52b6179a
PA
269 return 1;
270}
271__setup("nosmap", setup_disable_smap);
272
b2cc2a07
PA
273static __always_inline void setup_smap(struct cpuinfo_x86 *c)
274{
275 unsigned long eflags;
276
277 /* This should have been cleared long ago */
278 raw_local_save_flags(eflags);
279 BUG_ON(eflags & X86_EFLAGS_AC);
280
03bbd596
PA
281 if (cpu_has(c, X86_FEATURE_SMAP)) {
282#ifdef CONFIG_X86_SMAP
375074cc 283 cr4_set_bits(X86_CR4_SMAP);
03bbd596 284#else
375074cc 285 cr4_clear_bits(X86_CR4_SMAP);
03bbd596
PA
286#endif
287 }
de5397ad
FY
288}
289
b38b0665
PA
290/*
291 * Some CPU features depend on higher CPUID levels, which may not always
292 * be available due to CPUID level capping or broken virtualization
293 * software. Add those features to this table to auto-disable them.
294 */
295struct cpuid_dependent_feature {
296 u32 feature;
297 u32 level;
298};
0f3fa48a 299
148f9bb8 300static const struct cpuid_dependent_feature
b38b0665
PA
301cpuid_dependent_features[] = {
302 { X86_FEATURE_MWAIT, 0x00000005 },
303 { X86_FEATURE_DCA, 0x00000009 },
304 { X86_FEATURE_XSAVE, 0x0000000d },
305 { 0, 0 }
306};
307
148f9bb8 308static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
b38b0665
PA
309{
310 const struct cpuid_dependent_feature *df;
9766cdbc 311
b38b0665 312 for (df = cpuid_dependent_features; df->feature; df++) {
0f3fa48a
IM
313
314 if (!cpu_has(c, df->feature))
315 continue;
b38b0665
PA
316 /*
317 * Note: cpuid_level is set to -1 if unavailable, but
318 * extended_extended_level is set to 0 if unavailable
319 * and the legitimate extended levels are all negative
320 * when signed; hence the weird messing around with
321 * signs here...
322 */
0f3fa48a 323 if (!((s32)df->level < 0 ?
f6db44df 324 (u32)df->level > (u32)c->extended_cpuid_level :
0f3fa48a
IM
325 (s32)df->level > (s32)c->cpuid_level))
326 continue;
327
328 clear_cpu_cap(c, df->feature);
329 if (!warn)
330 continue;
331
332 printk(KERN_WARNING
9def39be
JT
333 "CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
334 x86_cap_flag(df->feature), df->level);
b38b0665 335 }
f6db44df 336}
b38b0665 337
102bbe3a
YL
338/*
339 * Naming convention should be: <Name> [(<Codename>)]
340 * This table only is used unless init_<vendor>() below doesn't set it;
0f3fa48a
IM
341 * in particular, if CPUID levels 0x80000002..4 are supported, this
342 * isn't used
102bbe3a
YL
343 */
344
345/* Look up CPU names by table lookup. */
148f9bb8 346static const char *table_lookup_model(struct cpuinfo_x86 *c)
102bbe3a 347{
09dc68d9
JB
348#ifdef CONFIG_X86_32
349 const struct legacy_cpu_model_info *info;
102bbe3a
YL
350
351 if (c->x86_model >= 16)
352 return NULL; /* Range check */
353
354 if (!this_cpu)
355 return NULL;
356
09dc68d9 357 info = this_cpu->legacy_models;
102bbe3a 358
09dc68d9 359 while (info->family) {
102bbe3a
YL
360 if (info->family == c->x86)
361 return info->model_names[c->x86_model];
362 info++;
363 }
09dc68d9 364#endif
102bbe3a
YL
365 return NULL; /* Not found */
366}
367
148f9bb8
PG
368__u32 cpu_caps_cleared[NCAPINTS];
369__u32 cpu_caps_set[NCAPINTS];
7d851c8d 370
11e3a840
JF
371void load_percpu_segment(int cpu)
372{
373#ifdef CONFIG_X86_32
374 loadsegment(fs, __KERNEL_PERCPU);
375#else
376 loadsegment(gs, 0);
377 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
378#endif
60a5317f 379 load_stack_canary_segment();
11e3a840
JF
380}
381
0f3fa48a
IM
382/*
383 * Current gdt points %fs at the "master" per-cpu area: after this,
384 * it's on the real one.
385 */
552be871 386void switch_to_new_gdt(int cpu)
9d31d35b
YL
387{
388 struct desc_ptr gdt_descr;
389
2697fbd5 390 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
9d31d35b
YL
391 gdt_descr.size = GDT_SIZE - 1;
392 load_gdt(&gdt_descr);
2697fbd5 393 /* Reload the per-cpu base */
11e3a840
JF
394
395 load_percpu_segment(cpu);
9d31d35b
YL
396}
397
148f9bb8 398static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
1da177e4 399
148f9bb8 400static void get_model_name(struct cpuinfo_x86 *c)
1da177e4
LT
401{
402 unsigned int *v;
ee098e1a 403 char *p, *q, *s;
1da177e4 404
3da99c97 405 if (c->extended_cpuid_level < 0x80000004)
1b05d60d 406 return;
1da177e4 407
0f3fa48a 408 v = (unsigned int *)c->x86_model_id;
1da177e4
LT
409 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
410 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
411 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
412 c->x86_model_id[48] = 0;
413
ee098e1a
BP
414 /* Trim whitespace */
415 p = q = s = &c->x86_model_id[0];
416
417 while (*p == ' ')
418 p++;
419
420 while (*p) {
421 /* Note the last non-whitespace index */
422 if (!isspace(*p))
423 s = q;
424
425 *q++ = *p++;
426 }
427
428 *(s + 1) = '\0';
1da177e4
LT
429}
430
148f9bb8 431void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
1da177e4 432{
9d31d35b 433 unsigned int n, dummy, ebx, ecx, edx, l2size;
1da177e4 434
3da99c97 435 n = c->extended_cpuid_level;
1da177e4
LT
436
437 if (n >= 0x80000005) {
9d31d35b 438 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
9d31d35b 439 c->x86_cache_size = (ecx>>24) + (edx>>24);
140fc727
YL
440#ifdef CONFIG_X86_64
441 /* On K8 L1 TLB is inclusive, so don't count it */
442 c->x86_tlbsize = 0;
443#endif
1da177e4
LT
444 }
445
446 if (n < 0x80000006) /* Some chips just has a large L1. */
447 return;
448
0a488a53 449 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
1da177e4 450 l2size = ecx >> 16;
34048c9e 451
140fc727
YL
452#ifdef CONFIG_X86_64
453 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
454#else
1da177e4 455 /* do processor-specific cache resizing */
09dc68d9
JB
456 if (this_cpu->legacy_cache_size)
457 l2size = this_cpu->legacy_cache_size(c, l2size);
1da177e4
LT
458
459 /* Allow user to override all this if necessary. */
460 if (cachesize_override != -1)
461 l2size = cachesize_override;
462
34048c9e 463 if (l2size == 0)
1da177e4 464 return; /* Again, no L2 cache is possible */
140fc727 465#endif
1da177e4
LT
466
467 c->x86_cache_size = l2size;
1da177e4
LT
468}
469
e0ba94f1
AS
470u16 __read_mostly tlb_lli_4k[NR_INFO];
471u16 __read_mostly tlb_lli_2m[NR_INFO];
472u16 __read_mostly tlb_lli_4m[NR_INFO];
473u16 __read_mostly tlb_lld_4k[NR_INFO];
474u16 __read_mostly tlb_lld_2m[NR_INFO];
475u16 __read_mostly tlb_lld_4m[NR_INFO];
dd360393 476u16 __read_mostly tlb_lld_1g[NR_INFO];
e0ba94f1 477
f94fe119 478static void cpu_detect_tlb(struct cpuinfo_x86 *c)
e0ba94f1
AS
479{
480 if (this_cpu->c_detect_tlb)
481 this_cpu->c_detect_tlb(c);
482
f94fe119 483 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
e0ba94f1 484 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
f94fe119
SH
485 tlb_lli_4m[ENTRIES]);
486
487 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
488 tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
489 tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
e0ba94f1
AS
490}
491
148f9bb8 492void detect_ht(struct cpuinfo_x86 *c)
1da177e4 493{
c8e56d20 494#ifdef CONFIG_SMP
0a488a53
YL
495 u32 eax, ebx, ecx, edx;
496 int index_msb, core_bits;
2eaad1fd 497 static bool printed;
1da177e4 498
0a488a53 499 if (!cpu_has(c, X86_FEATURE_HT))
9d31d35b 500 return;
1da177e4 501
0a488a53
YL
502 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
503 goto out;
1da177e4 504
1cd78776
YL
505 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
506 return;
1da177e4 507
0a488a53 508 cpuid(1, &eax, &ebx, &ecx, &edx);
1da177e4 509
9d31d35b
YL
510 smp_num_siblings = (ebx & 0xff0000) >> 16;
511
512 if (smp_num_siblings == 1) {
2eaad1fd 513 printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
0f3fa48a
IM
514 goto out;
515 }
9d31d35b 516
0f3fa48a
IM
517 if (smp_num_siblings <= 1)
518 goto out;
9d31d35b 519
0f3fa48a
IM
520 index_msb = get_count_order(smp_num_siblings);
521 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
9d31d35b 522
0f3fa48a 523 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
9d31d35b 524
0f3fa48a 525 index_msb = get_count_order(smp_num_siblings);
9d31d35b 526
0f3fa48a 527 core_bits = get_count_order(c->x86_max_cores);
9d31d35b 528
0f3fa48a
IM
529 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
530 ((1 << core_bits) - 1);
1da177e4 531
0a488a53 532out:
2eaad1fd 533 if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
0a488a53
YL
534 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
535 c->phys_proc_id);
536 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
537 c->cpu_core_id);
2eaad1fd 538 printed = 1;
9d31d35b 539 }
9d31d35b 540#endif
97e4db7c 541}
1da177e4 542
148f9bb8 543static void get_cpu_vendor(struct cpuinfo_x86 *c)
1da177e4
LT
544{
545 char *v = c->x86_vendor_id;
0f3fa48a 546 int i;
1da177e4
LT
547
548 for (i = 0; i < X86_VENDOR_NUM; i++) {
10a434fc
YL
549 if (!cpu_devs[i])
550 break;
551
552 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
553 (cpu_devs[i]->c_ident[1] &&
554 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
0f3fa48a 555
10a434fc
YL
556 this_cpu = cpu_devs[i];
557 c->x86_vendor = this_cpu->c_x86_vendor;
558 return;
1da177e4
LT
559 }
560 }
10a434fc 561
a9c56953
MK
562 printk_once(KERN_ERR
563 "CPU: vendor_id '%s' unknown, using generic init.\n" \
564 "CPU: Your system may be unstable.\n", v);
10a434fc 565
fe38d855
CE
566 c->x86_vendor = X86_VENDOR_UNKNOWN;
567 this_cpu = &default_cpu;
1da177e4
LT
568}
569
148f9bb8 570void cpu_detect(struct cpuinfo_x86 *c)
1da177e4 571{
1da177e4 572 /* Get vendor name */
4a148513
HH
573 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
574 (unsigned int *)&c->x86_vendor_id[0],
575 (unsigned int *)&c->x86_vendor_id[8],
576 (unsigned int *)&c->x86_vendor_id[4]);
1da177e4 577
1da177e4 578 c->x86 = 4;
9d31d35b 579 /* Intel-defined flags: level 0x00000001 */
1da177e4
LT
580 if (c->cpuid_level >= 0x00000001) {
581 u32 junk, tfms, cap0, misc;
0f3fa48a 582
1da177e4 583 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
9d31d35b
YL
584 c->x86 = (tfms >> 8) & 0xf;
585 c->x86_model = (tfms >> 4) & 0xf;
586 c->x86_mask = tfms & 0xf;
0f3fa48a 587
f5f786d0 588 if (c->x86 == 0xf)
1da177e4 589 c->x86 += (tfms >> 20) & 0xff;
f5f786d0 590 if (c->x86 >= 0x6)
9d31d35b 591 c->x86_model += ((tfms >> 16) & 0xf) << 4;
0f3fa48a 592
d4387bd3 593 if (cap0 & (1<<19)) {
d4387bd3 594 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
9d31d35b 595 c->x86_cache_alignment = c->x86_clflush_size;
d4387bd3 596 }
1da177e4 597 }
1da177e4 598}
3da99c97 599
148f9bb8 600void get_cpu_cap(struct cpuinfo_x86 *c)
093af8d7
YL
601{
602 u32 tfms, xlvl;
3da99c97 603 u32 ebx;
093af8d7 604
3da99c97
YL
605 /* Intel-defined flags: level 0x00000001 */
606 if (c->cpuid_level >= 0x00000001) {
607 u32 capability, excap;
0f3fa48a 608
3da99c97
YL
609 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
610 c->x86_capability[0] = capability;
611 c->x86_capability[4] = excap;
612 }
093af8d7 613
bdc802dc
PA
614 /* Additional Intel-defined flags: level 0x00000007 */
615 if (c->cpuid_level >= 0x00000007) {
616 u32 eax, ebx, ecx, edx;
617
618 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
619
2494b030 620 c->x86_capability[9] = ebx;
bdc802dc
PA
621 }
622
6229ad27
FY
623 /* Extended state features: level 0x0000000d */
624 if (c->cpuid_level >= 0x0000000d) {
625 u32 eax, ebx, ecx, edx;
626
627 cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
628
629 c->x86_capability[10] = eax;
630 }
631
cbc82b17
PWJ
632 /* Additional Intel-defined flags: level 0x0000000F */
633 if (c->cpuid_level >= 0x0000000F) {
634 u32 eax, ebx, ecx, edx;
635
636 /* QoS sub-leaf, EAX=0Fh, ECX=0 */
637 cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
638 c->x86_capability[11] = edx;
639 if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
640 /* will be overridden if occupancy monitoring exists */
641 c->x86_cache_max_rmid = ebx;
642
643 /* QoS sub-leaf, EAX=0Fh, ECX=1 */
644 cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
645 c->x86_capability[12] = edx;
646 if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) {
647 c->x86_cache_max_rmid = ecx;
648 c->x86_cache_occ_scale = ebx;
649 }
650 } else {
651 c->x86_cache_max_rmid = -1;
652 c->x86_cache_occ_scale = -1;
653 }
654 }
655
3da99c97
YL
656 /* AMD-defined flags: level 0x80000001 */
657 xlvl = cpuid_eax(0x80000000);
658 c->extended_cpuid_level = xlvl;
0f3fa48a 659
3da99c97
YL
660 if ((xlvl & 0xffff0000) == 0x80000000) {
661 if (xlvl >= 0x80000001) {
662 c->x86_capability[1] = cpuid_edx(0x80000001);
663 c->x86_capability[6] = cpuid_ecx(0x80000001);
093af8d7 664 }
093af8d7 665 }
093af8d7 666
5122c890
YL
667 if (c->extended_cpuid_level >= 0x80000008) {
668 u32 eax = cpuid_eax(0x80000008);
669
670 c->x86_virt_bits = (eax >> 8) & 0xff;
671 c->x86_phys_bits = eax & 0xff;
093af8d7 672 }
13c6c532
JB
673#ifdef CONFIG_X86_32
674 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
675 c->x86_phys_bits = 36;
5122c890 676#endif
e3224234
YL
677
678 if (c->extended_cpuid_level >= 0x80000007)
679 c->x86_power = cpuid_edx(0x80000007);
093af8d7 680
1dedefd1 681 init_scattered_cpuid_features(c);
093af8d7 682}
1da177e4 683
148f9bb8 684static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
aef93c8b
YL
685{
686#ifdef CONFIG_X86_32
687 int i;
688
689 /*
690 * First of all, decide if this is a 486 or higher
691 * It's a 486 if we can modify the AC flag
692 */
693 if (flag_is_changeable_p(X86_EFLAGS_AC))
694 c->x86 = 4;
695 else
696 c->x86 = 3;
697
698 for (i = 0; i < X86_VENDOR_NUM; i++)
699 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
700 c->x86_vendor_id[0] = 0;
701 cpu_devs[i]->c_identify(c);
702 if (c->x86_vendor_id[0]) {
703 get_cpu_vendor(c);
704 break;
705 }
706 }
707#endif
708}
709
34048c9e
PC
710/*
711 * Do minimum CPU detection early.
712 * Fields really needed: vendor, cpuid_level, family, model, mask,
713 * cache alignment.
714 * The others are not touched to avoid unwanted side effects.
715 *
716 * WARNING: this function is only called on the BP. Don't add code here
717 * that is supposed to run on all CPUs.
718 */
3da99c97 719static void __init early_identify_cpu(struct cpuinfo_x86 *c)
d7cd5611 720{
6627d242
YL
721#ifdef CONFIG_X86_64
722 c->x86_clflush_size = 64;
13c6c532
JB
723 c->x86_phys_bits = 36;
724 c->x86_virt_bits = 48;
6627d242 725#else
d4387bd3 726 c->x86_clflush_size = 32;
13c6c532
JB
727 c->x86_phys_bits = 32;
728 c->x86_virt_bits = 32;
6627d242 729#endif
0a488a53 730 c->x86_cache_alignment = c->x86_clflush_size;
d7cd5611 731
3da99c97 732 memset(&c->x86_capability, 0, sizeof c->x86_capability);
0a488a53 733 c->extended_cpuid_level = 0;
d7cd5611 734
aef93c8b
YL
735 if (!have_cpuid_p())
736 identify_cpu_without_cpuid(c);
737
738 /* cyrix could have cpuid enabled via c_identify()*/
d7cd5611
RR
739 if (!have_cpuid_p())
740 return;
741
742 cpu_detect(c);
3da99c97 743 get_cpu_vendor(c);
3da99c97 744 get_cpu_cap(c);
12cf105c 745
10a434fc
YL
746 if (this_cpu->c_early_init)
747 this_cpu->c_early_init(c);
093af8d7 748
f6e9456c 749 c->cpu_index = 0;
b38b0665 750 filter_cpuid_features(c, false);
de5397ad 751
a110b5ec
BP
752 if (this_cpu->c_bsp_init)
753 this_cpu->c_bsp_init(c);
c3b83598
BP
754
755 setup_force_cpu_cap(X86_FEATURE_ALWAYS);
db52ef74 756 fpu__init_system(c);
d7cd5611
RR
757}
758
9d31d35b
YL
759void __init early_cpu_init(void)
760{
02dde8b4 761 const struct cpu_dev *const *cdev;
10a434fc
YL
762 int count = 0;
763
ac23f253 764#ifdef CONFIG_PROCESSOR_SELECT
9766cdbc 765 printk(KERN_INFO "KERNEL supported cpus:\n");
31c997ca
IM
766#endif
767
10a434fc 768 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
02dde8b4 769 const struct cpu_dev *cpudev = *cdev;
9d31d35b 770
10a434fc
YL
771 if (count >= X86_VENDOR_NUM)
772 break;
773 cpu_devs[count] = cpudev;
774 count++;
775
ac23f253 776#ifdef CONFIG_PROCESSOR_SELECT
31c997ca
IM
777 {
778 unsigned int j;
779
780 for (j = 0; j < 2; j++) {
781 if (!cpudev->c_ident[j])
782 continue;
783 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
784 cpudev->c_ident[j]);
785 }
10a434fc 786 }
0388423d 787#endif
10a434fc 788 }
9d31d35b 789 early_identify_cpu(&boot_cpu_data);
d7cd5611 790}
093af8d7 791
b6734c35 792/*
366d4a43
BP
793 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
794 * unfortunately, that's not true in practice because of early VIA
795 * chips and (more importantly) broken virtualizers that are not easy
796 * to detect. In the latter case it doesn't even *fail* reliably, so
797 * probing for it doesn't even work. Disable it completely on 32-bit
ba0593bf 798 * unless we can find a reliable way to detect all the broken cases.
366d4a43 799 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
b6734c35 800 */
148f9bb8 801static void detect_nopl(struct cpuinfo_x86 *c)
b6734c35 802{
366d4a43 803#ifdef CONFIG_X86_32
b6734c35 804 clear_cpu_cap(c, X86_FEATURE_NOPL);
366d4a43
BP
805#else
806 set_cpu_cap(c, X86_FEATURE_NOPL);
807#endif
d7cd5611
RR
808}
809
148f9bb8 810static void generic_identify(struct cpuinfo_x86 *c)
1da177e4 811{
aef93c8b 812 c->extended_cpuid_level = 0;
1da177e4 813
3da99c97 814 if (!have_cpuid_p())
aef93c8b 815 identify_cpu_without_cpuid(c);
1d67953f 816
aef93c8b 817 /* cyrix could have cpuid enabled via c_identify()*/
a9853dd6 818 if (!have_cpuid_p())
aef93c8b 819 return;
1da177e4 820
3da99c97 821 cpu_detect(c);
1da177e4 822
3da99c97 823 get_cpu_vendor(c);
1da177e4 824
3da99c97 825 get_cpu_cap(c);
1da177e4 826
3da99c97
YL
827 if (c->cpuid_level >= 0x00000001) {
828 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
b89d3b3e 829#ifdef CONFIG_X86_32
c8e56d20 830# ifdef CONFIG_SMP
cb8cc442 831 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
b89d3b3e 832# else
3da99c97 833 c->apicid = c->initial_apicid;
b89d3b3e
YL
834# endif
835#endif
b89d3b3e 836 c->phys_proc_id = c->initial_apicid;
3da99c97 837 }
1da177e4 838
1b05d60d 839 get_model_name(c); /* Default name */
1da177e4 840
3da99c97 841 detect_nopl(c);
1da177e4 842}
1da177e4 843
cbc82b17
PWJ
844static void x86_init_cache_qos(struct cpuinfo_x86 *c)
845{
846 /*
847 * The heavy lifting of max_rmid and cache_occ_scale are handled
848 * in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu
849 * in case CQM bits really aren't there in this CPU.
850 */
851 if (c != &boot_cpu_data) {
852 boot_cpu_data.x86_cache_max_rmid =
853 min(boot_cpu_data.x86_cache_max_rmid,
854 c->x86_cache_max_rmid);
855 }
856}
857
1da177e4
LT
858/*
859 * This does the hard work of actually picking apart the CPU stuff...
860 */
148f9bb8 861static void identify_cpu(struct cpuinfo_x86 *c)
1da177e4
LT
862{
863 int i;
864
865 c->loops_per_jiffy = loops_per_jiffy;
866 c->x86_cache_size = -1;
867 c->x86_vendor = X86_VENDOR_UNKNOWN;
1da177e4
LT
868 c->x86_model = c->x86_mask = 0; /* So far unknown... */
869 c->x86_vendor_id[0] = '\0'; /* Unset */
870 c->x86_model_id[0] = '\0'; /* Unset */
94605eff 871 c->x86_max_cores = 1;
102bbe3a 872 c->x86_coreid_bits = 0;
11fdd252 873#ifdef CONFIG_X86_64
102bbe3a 874 c->x86_clflush_size = 64;
13c6c532
JB
875 c->x86_phys_bits = 36;
876 c->x86_virt_bits = 48;
102bbe3a
YL
877#else
878 c->cpuid_level = -1; /* CPUID not detected */
770d132f 879 c->x86_clflush_size = 32;
13c6c532
JB
880 c->x86_phys_bits = 32;
881 c->x86_virt_bits = 32;
102bbe3a
YL
882#endif
883 c->x86_cache_alignment = c->x86_clflush_size;
1da177e4
LT
884 memset(&c->x86_capability, 0, sizeof c->x86_capability);
885
1da177e4
LT
886 generic_identify(c);
887
3898534d 888 if (this_cpu->c_identify)
1da177e4
LT
889 this_cpu->c_identify(c);
890
2759c328
YL
891 /* Clear/Set all flags overriden by options, after probe */
892 for (i = 0; i < NCAPINTS; i++) {
893 c->x86_capability[i] &= ~cpu_caps_cleared[i];
894 c->x86_capability[i] |= cpu_caps_set[i];
895 }
896
102bbe3a 897#ifdef CONFIG_X86_64
cb8cc442 898 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
102bbe3a
YL
899#endif
900
1da177e4
LT
901 /*
902 * Vendor-specific initialization. In this section we
903 * canonicalize the feature flags, meaning if there are
904 * features a certain CPU supports which CPUID doesn't
905 * tell us, CPUID claiming incorrect flags, or other bugs,
906 * we handle them here.
907 *
908 * At the end of this section, c->x86_capability better
909 * indicate the features this CPU genuinely supports!
910 */
911 if (this_cpu->c_init)
912 this_cpu->c_init(c);
913
914 /* Disable the PN if appropriate */
915 squash_the_stupid_serial_number(c);
916
b2cc2a07
PA
917 /* Set up SMEP/SMAP */
918 setup_smep(c);
919 setup_smap(c);
920
1da177e4 921 /*
0f3fa48a
IM
922 * The vendor-specific functions might have changed features.
923 * Now we do "generic changes."
1da177e4
LT
924 */
925
b38b0665
PA
926 /* Filter out anything that depends on CPUID levels we don't have */
927 filter_cpuid_features(c, true);
928
1da177e4 929 /* If the model name is still unset, do table lookup. */
34048c9e 930 if (!c->x86_model_id[0]) {
02dde8b4 931 const char *p;
1da177e4 932 p = table_lookup_model(c);
34048c9e 933 if (p)
1da177e4
LT
934 strcpy(c->x86_model_id, p);
935 else
936 /* Last resort... */
937 sprintf(c->x86_model_id, "%02x/%02x",
54a20f8c 938 c->x86, c->x86_model);
1da177e4
LT
939 }
940
102bbe3a
YL
941#ifdef CONFIG_X86_64
942 detect_ht(c);
943#endif
944
88b094fb 945 init_hypervisor(c);
49d859d7 946 x86_init_rdrand(c);
cbc82b17 947 x86_init_cache_qos(c);
3e0c3737
YL
948
949 /*
950 * Clear/Set all flags overriden by options, need do it
951 * before following smp all cpus cap AND.
952 */
953 for (i = 0; i < NCAPINTS; i++) {
954 c->x86_capability[i] &= ~cpu_caps_cleared[i];
955 c->x86_capability[i] |= cpu_caps_set[i];
956 }
957
1da177e4
LT
958 /*
959 * On SMP, boot_cpu_data holds the common feature set between
960 * all CPUs; so make sure that we indicate which features are
961 * common between the CPUs. The first time this routine gets
962 * executed, c == &boot_cpu_data.
963 */
34048c9e 964 if (c != &boot_cpu_data) {
1da177e4 965 /* AND the already accumulated flags with these */
9d31d35b 966 for (i = 0; i < NCAPINTS; i++)
1da177e4 967 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
65fc985b
BP
968
969 /* OR, i.e. replicate the bug flags */
970 for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
971 c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1da177e4
LT
972 }
973
974 /* Init Machine Check Exception if available. */
5e09954a 975 mcheck_cpu_init(c);
30d432df
AK
976
977 select_idle_routine(c);
102bbe3a 978
de2d9445 979#ifdef CONFIG_NUMA
102bbe3a
YL
980 numa_add_cpu(smp_processor_id());
981#endif
a6c4e076 982}
31ab269a 983
8b6c0ab1
IM
984/*
985 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
986 * on 32-bit kernels:
987 */
cfda7bb9
AL
988#ifdef CONFIG_X86_32
989void enable_sep_cpu(void)
990{
8b6c0ab1
IM
991 struct tss_struct *tss;
992 int cpu;
cfda7bb9 993
8b6c0ab1
IM
994 cpu = get_cpu();
995 tss = &per_cpu(cpu_tss, cpu);
996
997 if (!boot_cpu_has(X86_FEATURE_SEP))
998 goto out;
999
1000 /*
cf9328cc
AL
1001 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1002 * see the big comment in struct x86_hw_tss's definition.
8b6c0ab1 1003 */
cfda7bb9
AL
1004
1005 tss->x86_tss.ss1 = __KERNEL_CS;
8b6c0ab1
IM
1006 wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1007
cf9328cc
AL
1008 wrmsr(MSR_IA32_SYSENTER_ESP,
1009 (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
1010 0);
8b6c0ab1 1011
4c8cd0c5 1012 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
8b6c0ab1
IM
1013
1014out:
cfda7bb9
AL
1015 put_cpu();
1016}
e04d645f
GC
1017#endif
1018
a6c4e076
JF
1019void __init identify_boot_cpu(void)
1020{
1021 identify_cpu(&boot_cpu_data);
02c68a02 1022 init_amd_e400_c1e_mask();
102bbe3a 1023#ifdef CONFIG_X86_32
a6c4e076 1024 sysenter_setup();
6fe940d6 1025 enable_sep_cpu();
102bbe3a 1026#endif
5b556332 1027 cpu_detect_tlb(&boot_cpu_data);
a6c4e076 1028}
3b520b23 1029
148f9bb8 1030void identify_secondary_cpu(struct cpuinfo_x86 *c)
a6c4e076
JF
1031{
1032 BUG_ON(c == &boot_cpu_data);
1033 identify_cpu(c);
102bbe3a 1034#ifdef CONFIG_X86_32
a6c4e076 1035 enable_sep_cpu();
102bbe3a 1036#endif
a6c4e076 1037 mtrr_ap_init();
1da177e4
LT
1038}
1039
a0854a46 1040struct msr_range {
0f3fa48a
IM
1041 unsigned min;
1042 unsigned max;
a0854a46 1043};
1da177e4 1044
148f9bb8 1045static const struct msr_range msr_range_array[] = {
a0854a46
YL
1046 { 0x00000000, 0x00000418},
1047 { 0xc0000000, 0xc000040b},
1048 { 0xc0010000, 0xc0010142},
1049 { 0xc0011000, 0xc001103b},
1050};
1da177e4 1051
148f9bb8 1052static void __print_cpu_msr(void)
a0854a46 1053{
0f3fa48a 1054 unsigned index_min, index_max;
a0854a46
YL
1055 unsigned index;
1056 u64 val;
1057 int i;
a0854a46
YL
1058
1059 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1060 index_min = msr_range_array[i].min;
1061 index_max = msr_range_array[i].max;
0f3fa48a 1062
a0854a46 1063 for (index = index_min; index < index_max; index++) {
ecd431d9 1064 if (rdmsrl_safe(index, &val))
a0854a46
YL
1065 continue;
1066 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1da177e4 1067 }
a0854a46
YL
1068 }
1069}
94605eff 1070
148f9bb8 1071static int show_msr;
0f3fa48a 1072
a0854a46
YL
1073static __init int setup_show_msr(char *arg)
1074{
1075 int num;
3dd9d514 1076
a0854a46 1077 get_option(&arg, &num);
3dd9d514 1078
a0854a46
YL
1079 if (num > 0)
1080 show_msr = num;
1081 return 1;
1da177e4 1082}
a0854a46 1083__setup("show_msr=", setup_show_msr);
1da177e4 1084
191679fd
AK
1085static __init int setup_noclflush(char *arg)
1086{
840d2830 1087 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
da4aaa7d 1088 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
191679fd
AK
1089 return 1;
1090}
1091__setup("noclflush", setup_noclflush);
1092
148f9bb8 1093void print_cpu_info(struct cpuinfo_x86 *c)
1da177e4 1094{
02dde8b4 1095 const char *vendor = NULL;
1da177e4 1096
0f3fa48a 1097 if (c->x86_vendor < X86_VENDOR_NUM) {
1da177e4 1098 vendor = this_cpu->c_vendor;
0f3fa48a
IM
1099 } else {
1100 if (c->cpuid_level >= 0)
1101 vendor = c->x86_vendor_id;
1102 }
1da177e4 1103
bd32a8cf 1104 if (vendor && !strstr(c->x86_model_id, vendor))
9d31d35b 1105 printk(KERN_CONT "%s ", vendor);
1da177e4 1106
9d31d35b 1107 if (c->x86_model_id[0])
adafb98d 1108 printk(KERN_CONT "%s", c->x86_model_id);
1da177e4 1109 else
9d31d35b 1110 printk(KERN_CONT "%d86", c->x86);
1da177e4 1111
924e101a
BP
1112 printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1113
34048c9e 1114 if (c->x86_mask || c->cpuid_level >= 0)
924e101a 1115 printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1da177e4 1116 else
924e101a 1117 printk(KERN_CONT ")\n");
a0854a46 1118
0b8b8078 1119 print_cpu_msr(c);
21c3fcf3
YL
1120}
1121
148f9bb8 1122void print_cpu_msr(struct cpuinfo_x86 *c)
21c3fcf3 1123{
a0854a46 1124 if (c->cpu_index < show_msr)
21c3fcf3 1125 __print_cpu_msr();
1da177e4
LT
1126}
1127
ac72e788
AK
1128static __init int setup_disablecpuid(char *arg)
1129{
1130 int bit;
0f3fa48a 1131
ac72e788
AK
1132 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1133 setup_clear_cpu_cap(bit);
1134 else
1135 return 0;
0f3fa48a 1136
ac72e788
AK
1137 return 1;
1138}
1139__setup("clearcpuid=", setup_disablecpuid);
1140
d5494d4f 1141#ifdef CONFIG_X86_64
9ff80942 1142struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
629f4f9d
SA
1143struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1144 (unsigned long) debug_idt_table };
d5494d4f 1145
947e76cd 1146DEFINE_PER_CPU_FIRST(union irq_stack_union,
277d5b40 1147 irq_stack_union) __aligned(PAGE_SIZE) __visible;
0f3fa48a 1148
bdf977b3 1149/*
a7fcf28d
AL
1150 * The following percpu variables are hot. Align current_task to
1151 * cacheline size such that they fall in the same cacheline.
bdf977b3
TH
1152 */
1153DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1154 &init_task;
1155EXPORT_PER_CPU_SYMBOL(current_task);
d5494d4f 1156
bdf977b3
TH
1157DEFINE_PER_CPU(char *, irq_stack_ptr) =
1158 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1159
277d5b40 1160DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
d5494d4f 1161
c2daa3be
PZ
1162DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1163EXPORT_PER_CPU_SYMBOL(__preempt_count);
1164
0f3fa48a
IM
1165/*
1166 * Special IST stacks which the CPU switches to when it calls
1167 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1168 * limit), all of them are 4K, except the debug stack which
1169 * is 8K.
1170 */
1171static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1172 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1173 [DEBUG_STACK - 1] = DEBUG_STKSZ
1174};
1175
92d65b23 1176static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
3e352aa8 1177 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
d5494d4f 1178
d5494d4f
YL
1179/* May not be marked __init: used by software suspend */
1180void syscall_init(void)
1da177e4 1181{
d5494d4f
YL
1182 /*
1183 * LSTAR and STAR live in a bit strange symbiosis.
1184 * They both write to the same internal register. STAR allows to
1185 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1186 */
1187 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
b2502b41 1188 wrmsrl(MSR_LSTAR, entry_SYSCALL_64);
d56fe4bf
IM
1189
1190#ifdef CONFIG_IA32_EMULATION
2cd23553 1191 wrmsrl(MSR_CSTAR, entry_SYSCALL_compat);
a76c7f46 1192 /*
487d1edb
DV
1193 * This only works on Intel CPUs.
1194 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1195 * This does not cause SYSENTER to jump to the wrong location, because
1196 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
a76c7f46
DV
1197 */
1198 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1199 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
4c8cd0c5 1200 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
d56fe4bf
IM
1201#else
1202 wrmsrl(MSR_CSTAR, ignore_sysret);
6b51311c 1203 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
d56fe4bf
IM
1204 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1205 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
d5494d4f 1206#endif
03ae5768 1207
d5494d4f
YL
1208 /* Flags to clear on syscall */
1209 wrmsrl(MSR_SYSCALL_MASK,
63bcff2a 1210 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
8c7aa698 1211 X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1da177e4 1212}
62111195 1213
d5494d4f
YL
1214/*
1215 * Copies of the original ist values from the tss are only accessed during
1216 * debugging, no special alignment required.
1217 */
1218DEFINE_PER_CPU(struct orig_ist, orig_ist);
1219
228bdaa9 1220static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
42181186 1221DEFINE_PER_CPU(int, debug_stack_usage);
228bdaa9
SR
1222
1223int is_debug_stack(unsigned long addr)
1224{
89cbc767
CL
1225 return __this_cpu_read(debug_stack_usage) ||
1226 (addr <= __this_cpu_read(debug_stack_addr) &&
1227 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
228bdaa9 1228}
0f46efeb 1229NOKPROBE_SYMBOL(is_debug_stack);
228bdaa9 1230
629f4f9d 1231DEFINE_PER_CPU(u32, debug_idt_ctr);
f8988175 1232
228bdaa9
SR
1233void debug_stack_set_zero(void)
1234{
629f4f9d
SA
1235 this_cpu_inc(debug_idt_ctr);
1236 load_current_idt();
228bdaa9 1237}
0f46efeb 1238NOKPROBE_SYMBOL(debug_stack_set_zero);
228bdaa9
SR
1239
1240void debug_stack_reset(void)
1241{
629f4f9d 1242 if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
f8988175 1243 return;
629f4f9d
SA
1244 if (this_cpu_dec_return(debug_idt_ctr) == 0)
1245 load_current_idt();
228bdaa9 1246}
0f46efeb 1247NOKPROBE_SYMBOL(debug_stack_reset);
228bdaa9 1248
0f3fa48a 1249#else /* CONFIG_X86_64 */
d5494d4f 1250
bdf977b3
TH
1251DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1252EXPORT_PER_CPU_SYMBOL(current_task);
c2daa3be
PZ
1253DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1254EXPORT_PER_CPU_SYMBOL(__preempt_count);
bdf977b3 1255
a7fcf28d
AL
1256/*
1257 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1258 * the top of the kernel stack. Use an extra percpu variable to track the
1259 * top of the kernel stack directly.
1260 */
1261DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1262 (unsigned long)&init_thread_union + THREAD_SIZE;
1263EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1264
60a5317f 1265#ifdef CONFIG_CC_STACKPROTECTOR
53f82452 1266DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
60a5317f 1267#endif
d5494d4f 1268
0f3fa48a 1269#endif /* CONFIG_X86_64 */
c5413fbe 1270
9766cdbc
JSR
1271/*
1272 * Clear all 6 debug registers:
1273 */
1274static void clear_all_debug_regs(void)
1275{
1276 int i;
1277
1278 for (i = 0; i < 8; i++) {
1279 /* Ignore db4, db5 */
1280 if ((i == 4) || (i == 5))
1281 continue;
1282
1283 set_debugreg(0, i);
1284 }
1285}
c5413fbe 1286
0bb9fef9
JW
1287#ifdef CONFIG_KGDB
1288/*
1289 * Restore debug regs if using kgdbwait and you have a kernel debugger
1290 * connection established.
1291 */
1292static void dbg_restore_debug_regs(void)
1293{
1294 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1295 arch_kgdb_ops.correct_hw_break();
1296}
1297#else /* ! CONFIG_KGDB */
1298#define dbg_restore_debug_regs()
1299#endif /* ! CONFIG_KGDB */
1300
ce4b1b16
IM
1301static void wait_for_master_cpu(int cpu)
1302{
1303#ifdef CONFIG_SMP
1304 /*
1305 * wait for ACK from master CPU before continuing
1306 * with AP initialization
1307 */
1308 WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1309 while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1310 cpu_relax();
1311#endif
1312}
1313
d2cbcc49
RR
1314/*
1315 * cpu_init() initializes state that is per-CPU. Some data is already
1316 * initialized (naturally) in the bootstrap process, such as the GDT
1317 * and IDT. We reload them nevertheless, this function acts as a
1318 * 'CPU state barrier', nothing should get across.
1ba76586 1319 * A lot of state is already set up in PDA init for 64 bit
d2cbcc49 1320 */
1ba76586 1321#ifdef CONFIG_X86_64
0f3fa48a 1322
148f9bb8 1323void cpu_init(void)
1ba76586 1324{
0fe1e009 1325 struct orig_ist *oist;
1ba76586 1326 struct task_struct *me;
0f3fa48a
IM
1327 struct tss_struct *t;
1328 unsigned long v;
ce4b1b16 1329 int cpu = stack_smp_processor_id();
1ba76586
YL
1330 int i;
1331
ce4b1b16
IM
1332 wait_for_master_cpu(cpu);
1333
1e02ce4c
AL
1334 /*
1335 * Initialize the CR4 shadow before doing anything that could
1336 * try to read it.
1337 */
1338 cr4_init_shadow();
1339
e6ebf5de
FY
1340 /*
1341 * Load microcode on this cpu if a valid microcode is available.
1342 * This is early microcode loading procedure.
1343 */
1344 load_ucode_ap();
1345
24933b82 1346 t = &per_cpu(cpu_tss, cpu);
0fe1e009 1347 oist = &per_cpu(orig_ist, cpu);
0f3fa48a 1348
e7a22c1e 1349#ifdef CONFIG_NUMA
27fd185f 1350 if (this_cpu_read(numa_node) == 0 &&
e534c7c5
LS
1351 early_cpu_to_node(cpu) != NUMA_NO_NODE)
1352 set_numa_node(early_cpu_to_node(cpu));
e7a22c1e 1353#endif
1ba76586
YL
1354
1355 me = current;
1356
2eaad1fd 1357 pr_debug("Initializing CPU#%d\n", cpu);
1ba76586 1358
375074cc 1359 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1ba76586
YL
1360
1361 /*
1362 * Initialize the per-CPU GDT with the boot GDT,
1363 * and set up the GDT descriptor:
1364 */
1365
552be871 1366 switch_to_new_gdt(cpu);
2697fbd5
BG
1367 loadsegment(fs, 0);
1368
cf910e83 1369 load_current_idt();
1ba76586
YL
1370
1371 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1372 syscall_init();
1373
1374 wrmsrl(MSR_FS_BASE, 0);
1375 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1376 barrier();
1377
4763ed4d 1378 x86_configure_nx();
659006bf 1379 x2apic_setup();
1ba76586
YL
1380
1381 /*
1382 * set up and load the per-CPU TSS
1383 */
0fe1e009 1384 if (!oist->ist[0]) {
92d65b23 1385 char *estacks = per_cpu(exception_stacks, cpu);
0f3fa48a 1386
1ba76586 1387 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
0f3fa48a 1388 estacks += exception_stack_sizes[v];
0fe1e009 1389 oist->ist[v] = t->x86_tss.ist[v] =
1ba76586 1390 (unsigned long)estacks;
228bdaa9
SR
1391 if (v == DEBUG_STACK-1)
1392 per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1ba76586
YL
1393 }
1394 }
1395
1396 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
0f3fa48a 1397
1ba76586
YL
1398 /*
1399 * <= is required because the CPU will access up to
1400 * 8 bits beyond the end of the IO permission bitmap.
1401 */
1402 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1403 t->io_bitmap[i] = ~0UL;
1404
1405 atomic_inc(&init_mm.mm_count);
1406 me->active_mm = &init_mm;
8c5dfd25 1407 BUG_ON(me->mm);
1ba76586
YL
1408 enter_lazy_tlb(&init_mm, me);
1409
1410 load_sp0(t, &current->thread);
1411 set_tss_desc(cpu, t);
1412 load_TR_desc();
1413 load_LDT(&init_mm.context);
1414
0bb9fef9
JW
1415 clear_all_debug_regs();
1416 dbg_restore_debug_regs();
1ba76586 1417
21c4cd10 1418 fpu__init_cpu();
1ba76586 1419
1ba76586
YL
1420 if (is_uv_system())
1421 uv_cpu_init();
1422}
1423
1424#else
1425
148f9bb8 1426void cpu_init(void)
9ee79a3d 1427{
d2cbcc49
RR
1428 int cpu = smp_processor_id();
1429 struct task_struct *curr = current;
24933b82 1430 struct tss_struct *t = &per_cpu(cpu_tss, cpu);
9ee79a3d 1431 struct thread_struct *thread = &curr->thread;
62111195 1432
ce4b1b16 1433 wait_for_master_cpu(cpu);
e6ebf5de 1434
5b2bdbc8
SR
1435 /*
1436 * Initialize the CR4 shadow before doing anything that could
1437 * try to read it.
1438 */
1439 cr4_init_shadow();
1440
ce4b1b16 1441 show_ucode_info_early();
62111195
JF
1442
1443 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1444
9298b815 1445 if (cpu_feature_enabled(X86_FEATURE_VME) || cpu_has_tsc || cpu_has_de)
375074cc 1446 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
62111195 1447
cf910e83 1448 load_current_idt();
552be871 1449 switch_to_new_gdt(cpu);
1da177e4 1450
1da177e4
LT
1451 /*
1452 * Set up and load the per-CPU TSS and LDT
1453 */
1454 atomic_inc(&init_mm.mm_count);
62111195 1455 curr->active_mm = &init_mm;
8c5dfd25 1456 BUG_ON(curr->mm);
62111195 1457 enter_lazy_tlb(&init_mm, curr);
1da177e4 1458
faca6227 1459 load_sp0(t, thread);
34048c9e 1460 set_tss_desc(cpu, t);
1da177e4
LT
1461 load_TR_desc();
1462 load_LDT(&init_mm.context);
1463
f9a196b8
TG
1464 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1465
22c4e308 1466#ifdef CONFIG_DOUBLEFAULT
1da177e4
LT
1467 /* Set up doublefault TSS pointer in the GDT */
1468 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
22c4e308 1469#endif
1da177e4 1470
9766cdbc 1471 clear_all_debug_regs();
0bb9fef9 1472 dbg_restore_debug_regs();
1da177e4 1473
21c4cd10 1474 fpu__init_cpu();
1da177e4 1475}
1ba76586 1476#endif
5700f743
BP
1477
1478#ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1479void warn_pre_alternatives(void)
1480{
1481 WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1482}
1483EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1484#endif
4a90a99c
BP
1485
1486inline bool __static_cpu_has_safe(u16 bit)
1487{
1488 return boot_cpu_has(bit);
1489}
1490EXPORT_SYMBOL_GPL(__static_cpu_has_safe);