]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - arch/x86/kernel/cpu/common.c
x86/cpu: Strip any /proc/cpuinfo model name field whitespace
[mirror_ubuntu-jammy-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>
1da177e4 8#include <linux/delay.h>
9766cdbc
JSR
9#include <linux/sched.h>
10#include <linux/init.h>
0f46efeb 11#include <linux/kprobes.h>
9766cdbc 12#include <linux/kgdb.h>
1da177e4 13#include <linux/smp.h>
9766cdbc
JSR
14#include <linux/io.h>
15
16#include <asm/stackprotector.h>
cdd6c482 17#include <asm/perf_event.h>
1da177e4 18#include <asm/mmu_context.h>
49d859d7 19#include <asm/archrandom.h>
9766cdbc
JSR
20#include <asm/hypervisor.h>
21#include <asm/processor.h>
1e02ce4c 22#include <asm/tlbflush.h>
f649e938 23#include <asm/debugreg.h>
9766cdbc 24#include <asm/sections.h>
f40c3300 25#include <asm/vsyscall.h>
8bdbd962
AC
26#include <linux/topology.h>
27#include <linux/cpumask.h>
9766cdbc 28#include <asm/pgtable.h>
60063497 29#include <linux/atomic.h>
9766cdbc
JSR
30#include <asm/proto.h>
31#include <asm/setup.h>
32#include <asm/apic.h>
33#include <asm/desc.h>
34#include <asm/i387.h>
1361b83a 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
0c752a93
SS
148static int __init x86_xsave_setup(char *s)
149{
2cd3949f
DH
150 if (strlen(s))
151 return 0;
0c752a93 152 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
6bad06b7 153 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
b6f42a4a 154 setup_clear_cpu_cap(X86_FEATURE_XSAVES);
c6fd893d
SS
155 setup_clear_cpu_cap(X86_FEATURE_AVX);
156 setup_clear_cpu_cap(X86_FEATURE_AVX2);
0c752a93
SS
157 return 1;
158}
159__setup("noxsave", x86_xsave_setup);
160
6bad06b7
SS
161static int __init x86_xsaveopt_setup(char *s)
162{
163 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
164 return 1;
165}
166__setup("noxsaveopt", x86_xsaveopt_setup);
167
b6f42a4a
FY
168static int __init x86_xsaves_setup(char *s)
169{
170 setup_clear_cpu_cap(X86_FEATURE_XSAVES);
171 return 1;
172}
173__setup("noxsaves", x86_xsaves_setup);
174
ba51dced 175#ifdef CONFIG_X86_32
148f9bb8
PG
176static int cachesize_override = -1;
177static int disable_x86_serial_nr = 1;
1da177e4 178
0a488a53
YL
179static int __init cachesize_setup(char *str)
180{
181 get_option(&str, &cachesize_override);
182 return 1;
183}
184__setup("cachesize=", cachesize_setup);
185
0a488a53
YL
186static int __init x86_fxsr_setup(char *s)
187{
188 setup_clear_cpu_cap(X86_FEATURE_FXSR);
189 setup_clear_cpu_cap(X86_FEATURE_XMM);
190 return 1;
191}
192__setup("nofxsr", x86_fxsr_setup);
193
194static int __init x86_sep_setup(char *s)
195{
196 setup_clear_cpu_cap(X86_FEATURE_SEP);
197 return 1;
198}
199__setup("nosep", x86_sep_setup);
200
201/* Standard macro to see if a specific flag is changeable */
202static inline int flag_is_changeable_p(u32 flag)
203{
204 u32 f1, f2;
205
94f6bac1
KH
206 /*
207 * Cyrix and IDT cpus allow disabling of CPUID
208 * so the code below may return different results
209 * when it is executed before and after enabling
210 * the CPUID. Add "volatile" to not allow gcc to
211 * optimize the subsequent calls to this function.
212 */
0f3fa48a
IM
213 asm volatile ("pushfl \n\t"
214 "pushfl \n\t"
215 "popl %0 \n\t"
216 "movl %0, %1 \n\t"
217 "xorl %2, %0 \n\t"
218 "pushl %0 \n\t"
219 "popfl \n\t"
220 "pushfl \n\t"
221 "popl %0 \n\t"
222 "popfl \n\t"
223
94f6bac1
KH
224 : "=&r" (f1), "=&r" (f2)
225 : "ir" (flag));
0a488a53
YL
226
227 return ((f1^f2) & flag) != 0;
228}
229
230/* Probe for the CPUID instruction */
148f9bb8 231int have_cpuid_p(void)
0a488a53
YL
232{
233 return flag_is_changeable_p(X86_EFLAGS_ID);
234}
235
148f9bb8 236static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
0a488a53 237{
0f3fa48a
IM
238 unsigned long lo, hi;
239
240 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
241 return;
242
243 /* Disable processor serial number: */
244
245 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
246 lo |= 0x200000;
247 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
248
249 printk(KERN_NOTICE "CPU serial number disabled.\n");
250 clear_cpu_cap(c, X86_FEATURE_PN);
251
252 /* Disabling the serial number may affect the cpuid level */
253 c->cpuid_level = cpuid_eax(0);
0a488a53
YL
254}
255
256static int __init x86_serial_nr_setup(char *s)
257{
258 disable_x86_serial_nr = 0;
259 return 1;
260}
261__setup("serialnumber", x86_serial_nr_setup);
ba51dced 262#else
102bbe3a
YL
263static inline int flag_is_changeable_p(u32 flag)
264{
265 return 1;
266}
102bbe3a
YL
267static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
268{
269}
ba51dced 270#endif
0a488a53 271
de5397ad
FY
272static __init int setup_disable_smep(char *arg)
273{
b2cc2a07 274 setup_clear_cpu_cap(X86_FEATURE_SMEP);
de5397ad
FY
275 return 1;
276}
277__setup("nosmep", setup_disable_smep);
278
b2cc2a07 279static __always_inline void setup_smep(struct cpuinfo_x86 *c)
de5397ad 280{
b2cc2a07 281 if (cpu_has(c, X86_FEATURE_SMEP))
375074cc 282 cr4_set_bits(X86_CR4_SMEP);
de5397ad
FY
283}
284
52b6179a
PA
285static __init int setup_disable_smap(char *arg)
286{
b2cc2a07 287 setup_clear_cpu_cap(X86_FEATURE_SMAP);
52b6179a
PA
288 return 1;
289}
290__setup("nosmap", setup_disable_smap);
291
b2cc2a07
PA
292static __always_inline void setup_smap(struct cpuinfo_x86 *c)
293{
294 unsigned long eflags;
295
296 /* This should have been cleared long ago */
297 raw_local_save_flags(eflags);
298 BUG_ON(eflags & X86_EFLAGS_AC);
299
03bbd596
PA
300 if (cpu_has(c, X86_FEATURE_SMAP)) {
301#ifdef CONFIG_X86_SMAP
375074cc 302 cr4_set_bits(X86_CR4_SMAP);
03bbd596 303#else
375074cc 304 cr4_clear_bits(X86_CR4_SMAP);
03bbd596
PA
305#endif
306 }
de5397ad
FY
307}
308
b38b0665
PA
309/*
310 * Some CPU features depend on higher CPUID levels, which may not always
311 * be available due to CPUID level capping or broken virtualization
312 * software. Add those features to this table to auto-disable them.
313 */
314struct cpuid_dependent_feature {
315 u32 feature;
316 u32 level;
317};
0f3fa48a 318
148f9bb8 319static const struct cpuid_dependent_feature
b38b0665
PA
320cpuid_dependent_features[] = {
321 { X86_FEATURE_MWAIT, 0x00000005 },
322 { X86_FEATURE_DCA, 0x00000009 },
323 { X86_FEATURE_XSAVE, 0x0000000d },
324 { 0, 0 }
325};
326
148f9bb8 327static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
b38b0665
PA
328{
329 const struct cpuid_dependent_feature *df;
9766cdbc 330
b38b0665 331 for (df = cpuid_dependent_features; df->feature; df++) {
0f3fa48a
IM
332
333 if (!cpu_has(c, df->feature))
334 continue;
b38b0665
PA
335 /*
336 * Note: cpuid_level is set to -1 if unavailable, but
337 * extended_extended_level is set to 0 if unavailable
338 * and the legitimate extended levels are all negative
339 * when signed; hence the weird messing around with
340 * signs here...
341 */
0f3fa48a 342 if (!((s32)df->level < 0 ?
f6db44df 343 (u32)df->level > (u32)c->extended_cpuid_level :
0f3fa48a
IM
344 (s32)df->level > (s32)c->cpuid_level))
345 continue;
346
347 clear_cpu_cap(c, df->feature);
348 if (!warn)
349 continue;
350
351 printk(KERN_WARNING
9def39be
JT
352 "CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
353 x86_cap_flag(df->feature), df->level);
b38b0665 354 }
f6db44df 355}
b38b0665 356
102bbe3a
YL
357/*
358 * Naming convention should be: <Name> [(<Codename>)]
359 * This table only is used unless init_<vendor>() below doesn't set it;
0f3fa48a
IM
360 * in particular, if CPUID levels 0x80000002..4 are supported, this
361 * isn't used
102bbe3a
YL
362 */
363
364/* Look up CPU names by table lookup. */
148f9bb8 365static const char *table_lookup_model(struct cpuinfo_x86 *c)
102bbe3a 366{
09dc68d9
JB
367#ifdef CONFIG_X86_32
368 const struct legacy_cpu_model_info *info;
102bbe3a
YL
369
370 if (c->x86_model >= 16)
371 return NULL; /* Range check */
372
373 if (!this_cpu)
374 return NULL;
375
09dc68d9 376 info = this_cpu->legacy_models;
102bbe3a 377
09dc68d9 378 while (info->family) {
102bbe3a
YL
379 if (info->family == c->x86)
380 return info->model_names[c->x86_model];
381 info++;
382 }
09dc68d9 383#endif
102bbe3a
YL
384 return NULL; /* Not found */
385}
386
148f9bb8
PG
387__u32 cpu_caps_cleared[NCAPINTS];
388__u32 cpu_caps_set[NCAPINTS];
7d851c8d 389
11e3a840
JF
390void load_percpu_segment(int cpu)
391{
392#ifdef CONFIG_X86_32
393 loadsegment(fs, __KERNEL_PERCPU);
394#else
395 loadsegment(gs, 0);
396 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
397#endif
60a5317f 398 load_stack_canary_segment();
11e3a840
JF
399}
400
0f3fa48a
IM
401/*
402 * Current gdt points %fs at the "master" per-cpu area: after this,
403 * it's on the real one.
404 */
552be871 405void switch_to_new_gdt(int cpu)
9d31d35b
YL
406{
407 struct desc_ptr gdt_descr;
408
2697fbd5 409 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
9d31d35b
YL
410 gdt_descr.size = GDT_SIZE - 1;
411 load_gdt(&gdt_descr);
2697fbd5 412 /* Reload the per-cpu base */
11e3a840
JF
413
414 load_percpu_segment(cpu);
9d31d35b
YL
415}
416
148f9bb8 417static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
1da177e4 418
148f9bb8 419static void get_model_name(struct cpuinfo_x86 *c)
1da177e4
LT
420{
421 unsigned int *v;
1da177e4 422
3da99c97 423 if (c->extended_cpuid_level < 0x80000004)
1b05d60d 424 return;
1da177e4 425
0f3fa48a 426 v = (unsigned int *)c->x86_model_id;
1da177e4
LT
427 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
428 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
429 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
430 c->x86_model_id[48] = 0;
431
0f3fa48a 432 /*
adafb98d
PB
433 * Remove leading whitespace on Intel processors and trailing
434 * whitespace on AMD processors.
0f3fa48a 435 */
adafb98d 436 memmove(c->x86_model_id, strim(c->x86_model_id), 48);
1da177e4
LT
437}
438
148f9bb8 439void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
1da177e4 440{
9d31d35b 441 unsigned int n, dummy, ebx, ecx, edx, l2size;
1da177e4 442
3da99c97 443 n = c->extended_cpuid_level;
1da177e4
LT
444
445 if (n >= 0x80000005) {
9d31d35b 446 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
9d31d35b 447 c->x86_cache_size = (ecx>>24) + (edx>>24);
140fc727
YL
448#ifdef CONFIG_X86_64
449 /* On K8 L1 TLB is inclusive, so don't count it */
450 c->x86_tlbsize = 0;
451#endif
1da177e4
LT
452 }
453
454 if (n < 0x80000006) /* Some chips just has a large L1. */
455 return;
456
0a488a53 457 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
1da177e4 458 l2size = ecx >> 16;
34048c9e 459
140fc727
YL
460#ifdef CONFIG_X86_64
461 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
462#else
1da177e4 463 /* do processor-specific cache resizing */
09dc68d9
JB
464 if (this_cpu->legacy_cache_size)
465 l2size = this_cpu->legacy_cache_size(c, l2size);
1da177e4
LT
466
467 /* Allow user to override all this if necessary. */
468 if (cachesize_override != -1)
469 l2size = cachesize_override;
470
34048c9e 471 if (l2size == 0)
1da177e4 472 return; /* Again, no L2 cache is possible */
140fc727 473#endif
1da177e4
LT
474
475 c->x86_cache_size = l2size;
1da177e4
LT
476}
477
e0ba94f1
AS
478u16 __read_mostly tlb_lli_4k[NR_INFO];
479u16 __read_mostly tlb_lli_2m[NR_INFO];
480u16 __read_mostly tlb_lli_4m[NR_INFO];
481u16 __read_mostly tlb_lld_4k[NR_INFO];
482u16 __read_mostly tlb_lld_2m[NR_INFO];
483u16 __read_mostly tlb_lld_4m[NR_INFO];
dd360393 484u16 __read_mostly tlb_lld_1g[NR_INFO];
e0ba94f1 485
f94fe119 486static void cpu_detect_tlb(struct cpuinfo_x86 *c)
e0ba94f1
AS
487{
488 if (this_cpu->c_detect_tlb)
489 this_cpu->c_detect_tlb(c);
490
f94fe119 491 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
e0ba94f1 492 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
f94fe119
SH
493 tlb_lli_4m[ENTRIES]);
494
495 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
496 tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
497 tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
e0ba94f1
AS
498}
499
148f9bb8 500void detect_ht(struct cpuinfo_x86 *c)
1da177e4 501{
97e4db7c 502#ifdef CONFIG_X86_HT
0a488a53
YL
503 u32 eax, ebx, ecx, edx;
504 int index_msb, core_bits;
2eaad1fd 505 static bool printed;
1da177e4 506
0a488a53 507 if (!cpu_has(c, X86_FEATURE_HT))
9d31d35b 508 return;
1da177e4 509
0a488a53
YL
510 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
511 goto out;
1da177e4 512
1cd78776
YL
513 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
514 return;
1da177e4 515
0a488a53 516 cpuid(1, &eax, &ebx, &ecx, &edx);
1da177e4 517
9d31d35b
YL
518 smp_num_siblings = (ebx & 0xff0000) >> 16;
519
520 if (smp_num_siblings == 1) {
2eaad1fd 521 printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
0f3fa48a
IM
522 goto out;
523 }
9d31d35b 524
0f3fa48a
IM
525 if (smp_num_siblings <= 1)
526 goto out;
9d31d35b 527
0f3fa48a
IM
528 index_msb = get_count_order(smp_num_siblings);
529 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
9d31d35b 530
0f3fa48a 531 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
9d31d35b 532
0f3fa48a 533 index_msb = get_count_order(smp_num_siblings);
9d31d35b 534
0f3fa48a 535 core_bits = get_count_order(c->x86_max_cores);
9d31d35b 536
0f3fa48a
IM
537 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
538 ((1 << core_bits) - 1);
1da177e4 539
0a488a53 540out:
2eaad1fd 541 if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
0a488a53
YL
542 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
543 c->phys_proc_id);
544 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
545 c->cpu_core_id);
2eaad1fd 546 printed = 1;
9d31d35b 547 }
9d31d35b 548#endif
97e4db7c 549}
1da177e4 550
148f9bb8 551static void get_cpu_vendor(struct cpuinfo_x86 *c)
1da177e4
LT
552{
553 char *v = c->x86_vendor_id;
0f3fa48a 554 int i;
1da177e4
LT
555
556 for (i = 0; i < X86_VENDOR_NUM; i++) {
10a434fc
YL
557 if (!cpu_devs[i])
558 break;
559
560 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
561 (cpu_devs[i]->c_ident[1] &&
562 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
0f3fa48a 563
10a434fc
YL
564 this_cpu = cpu_devs[i];
565 c->x86_vendor = this_cpu->c_x86_vendor;
566 return;
1da177e4
LT
567 }
568 }
10a434fc 569
a9c56953
MK
570 printk_once(KERN_ERR
571 "CPU: vendor_id '%s' unknown, using generic init.\n" \
572 "CPU: Your system may be unstable.\n", v);
10a434fc 573
fe38d855
CE
574 c->x86_vendor = X86_VENDOR_UNKNOWN;
575 this_cpu = &default_cpu;
1da177e4
LT
576}
577
148f9bb8 578void cpu_detect(struct cpuinfo_x86 *c)
1da177e4 579{
1da177e4 580 /* Get vendor name */
4a148513
HH
581 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
582 (unsigned int *)&c->x86_vendor_id[0],
583 (unsigned int *)&c->x86_vendor_id[8],
584 (unsigned int *)&c->x86_vendor_id[4]);
1da177e4 585
1da177e4 586 c->x86 = 4;
9d31d35b 587 /* Intel-defined flags: level 0x00000001 */
1da177e4
LT
588 if (c->cpuid_level >= 0x00000001) {
589 u32 junk, tfms, cap0, misc;
0f3fa48a 590
1da177e4 591 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
9d31d35b
YL
592 c->x86 = (tfms >> 8) & 0xf;
593 c->x86_model = (tfms >> 4) & 0xf;
594 c->x86_mask = tfms & 0xf;
0f3fa48a 595
f5f786d0 596 if (c->x86 == 0xf)
1da177e4 597 c->x86 += (tfms >> 20) & 0xff;
f5f786d0 598 if (c->x86 >= 0x6)
9d31d35b 599 c->x86_model += ((tfms >> 16) & 0xf) << 4;
0f3fa48a 600
d4387bd3 601 if (cap0 & (1<<19)) {
d4387bd3 602 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
9d31d35b 603 c->x86_cache_alignment = c->x86_clflush_size;
d4387bd3 604 }
1da177e4 605 }
1da177e4 606}
3da99c97 607
148f9bb8 608void get_cpu_cap(struct cpuinfo_x86 *c)
093af8d7
YL
609{
610 u32 tfms, xlvl;
3da99c97 611 u32 ebx;
093af8d7 612
3da99c97
YL
613 /* Intel-defined flags: level 0x00000001 */
614 if (c->cpuid_level >= 0x00000001) {
615 u32 capability, excap;
0f3fa48a 616
3da99c97
YL
617 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
618 c->x86_capability[0] = capability;
619 c->x86_capability[4] = excap;
620 }
093af8d7 621
bdc802dc
PA
622 /* Additional Intel-defined flags: level 0x00000007 */
623 if (c->cpuid_level >= 0x00000007) {
624 u32 eax, ebx, ecx, edx;
625
626 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
627
2494b030 628 c->x86_capability[9] = ebx;
bdc802dc
PA
629 }
630
6229ad27
FY
631 /* Extended state features: level 0x0000000d */
632 if (c->cpuid_level >= 0x0000000d) {
633 u32 eax, ebx, ecx, edx;
634
635 cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
636
637 c->x86_capability[10] = eax;
638 }
639
cbc82b17
PWJ
640 /* Additional Intel-defined flags: level 0x0000000F */
641 if (c->cpuid_level >= 0x0000000F) {
642 u32 eax, ebx, ecx, edx;
643
644 /* QoS sub-leaf, EAX=0Fh, ECX=0 */
645 cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
646 c->x86_capability[11] = edx;
647 if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
648 /* will be overridden if occupancy monitoring exists */
649 c->x86_cache_max_rmid = ebx;
650
651 /* QoS sub-leaf, EAX=0Fh, ECX=1 */
652 cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
653 c->x86_capability[12] = edx;
654 if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) {
655 c->x86_cache_max_rmid = ecx;
656 c->x86_cache_occ_scale = ebx;
657 }
658 } else {
659 c->x86_cache_max_rmid = -1;
660 c->x86_cache_occ_scale = -1;
661 }
662 }
663
3da99c97
YL
664 /* AMD-defined flags: level 0x80000001 */
665 xlvl = cpuid_eax(0x80000000);
666 c->extended_cpuid_level = xlvl;
0f3fa48a 667
3da99c97
YL
668 if ((xlvl & 0xffff0000) == 0x80000000) {
669 if (xlvl >= 0x80000001) {
670 c->x86_capability[1] = cpuid_edx(0x80000001);
671 c->x86_capability[6] = cpuid_ecx(0x80000001);
093af8d7 672 }
093af8d7 673 }
093af8d7 674
5122c890
YL
675 if (c->extended_cpuid_level >= 0x80000008) {
676 u32 eax = cpuid_eax(0x80000008);
677
678 c->x86_virt_bits = (eax >> 8) & 0xff;
679 c->x86_phys_bits = eax & 0xff;
093af8d7 680 }
13c6c532
JB
681#ifdef CONFIG_X86_32
682 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
683 c->x86_phys_bits = 36;
5122c890 684#endif
e3224234
YL
685
686 if (c->extended_cpuid_level >= 0x80000007)
687 c->x86_power = cpuid_edx(0x80000007);
093af8d7 688
1dedefd1 689 init_scattered_cpuid_features(c);
093af8d7 690}
1da177e4 691
148f9bb8 692static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
aef93c8b
YL
693{
694#ifdef CONFIG_X86_32
695 int i;
696
697 /*
698 * First of all, decide if this is a 486 or higher
699 * It's a 486 if we can modify the AC flag
700 */
701 if (flag_is_changeable_p(X86_EFLAGS_AC))
702 c->x86 = 4;
703 else
704 c->x86 = 3;
705
706 for (i = 0; i < X86_VENDOR_NUM; i++)
707 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
708 c->x86_vendor_id[0] = 0;
709 cpu_devs[i]->c_identify(c);
710 if (c->x86_vendor_id[0]) {
711 get_cpu_vendor(c);
712 break;
713 }
714 }
715#endif
716}
717
34048c9e
PC
718/*
719 * Do minimum CPU detection early.
720 * Fields really needed: vendor, cpuid_level, family, model, mask,
721 * cache alignment.
722 * The others are not touched to avoid unwanted side effects.
723 *
724 * WARNING: this function is only called on the BP. Don't add code here
725 * that is supposed to run on all CPUs.
726 */
3da99c97 727static void __init early_identify_cpu(struct cpuinfo_x86 *c)
d7cd5611 728{
6627d242
YL
729#ifdef CONFIG_X86_64
730 c->x86_clflush_size = 64;
13c6c532
JB
731 c->x86_phys_bits = 36;
732 c->x86_virt_bits = 48;
6627d242 733#else
d4387bd3 734 c->x86_clflush_size = 32;
13c6c532
JB
735 c->x86_phys_bits = 32;
736 c->x86_virt_bits = 32;
6627d242 737#endif
0a488a53 738 c->x86_cache_alignment = c->x86_clflush_size;
d7cd5611 739
3da99c97 740 memset(&c->x86_capability, 0, sizeof c->x86_capability);
0a488a53 741 c->extended_cpuid_level = 0;
d7cd5611 742
aef93c8b
YL
743 if (!have_cpuid_p())
744 identify_cpu_without_cpuid(c);
745
746 /* cyrix could have cpuid enabled via c_identify()*/
d7cd5611
RR
747 if (!have_cpuid_p())
748 return;
749
750 cpu_detect(c);
3da99c97 751 get_cpu_vendor(c);
3da99c97 752 get_cpu_cap(c);
60e019eb 753 fpu_detect(c);
12cf105c 754
10a434fc
YL
755 if (this_cpu->c_early_init)
756 this_cpu->c_early_init(c);
093af8d7 757
f6e9456c 758 c->cpu_index = 0;
b38b0665 759 filter_cpuid_features(c, false);
de5397ad 760
a110b5ec
BP
761 if (this_cpu->c_bsp_init)
762 this_cpu->c_bsp_init(c);
c3b83598
BP
763
764 setup_force_cpu_cap(X86_FEATURE_ALWAYS);
d7cd5611
RR
765}
766
9d31d35b
YL
767void __init early_cpu_init(void)
768{
02dde8b4 769 const struct cpu_dev *const *cdev;
10a434fc
YL
770 int count = 0;
771
ac23f253 772#ifdef CONFIG_PROCESSOR_SELECT
9766cdbc 773 printk(KERN_INFO "KERNEL supported cpus:\n");
31c997ca
IM
774#endif
775
10a434fc 776 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
02dde8b4 777 const struct cpu_dev *cpudev = *cdev;
9d31d35b 778
10a434fc
YL
779 if (count >= X86_VENDOR_NUM)
780 break;
781 cpu_devs[count] = cpudev;
782 count++;
783
ac23f253 784#ifdef CONFIG_PROCESSOR_SELECT
31c997ca
IM
785 {
786 unsigned int j;
787
788 for (j = 0; j < 2; j++) {
789 if (!cpudev->c_ident[j])
790 continue;
791 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
792 cpudev->c_ident[j]);
793 }
10a434fc 794 }
0388423d 795#endif
10a434fc 796 }
9d31d35b 797 early_identify_cpu(&boot_cpu_data);
d7cd5611 798}
093af8d7 799
b6734c35 800/*
366d4a43
BP
801 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
802 * unfortunately, that's not true in practice because of early VIA
803 * chips and (more importantly) broken virtualizers that are not easy
804 * to detect. In the latter case it doesn't even *fail* reliably, so
805 * probing for it doesn't even work. Disable it completely on 32-bit
ba0593bf 806 * unless we can find a reliable way to detect all the broken cases.
366d4a43 807 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
b6734c35 808 */
148f9bb8 809static void detect_nopl(struct cpuinfo_x86 *c)
b6734c35 810{
366d4a43 811#ifdef CONFIG_X86_32
b6734c35 812 clear_cpu_cap(c, X86_FEATURE_NOPL);
366d4a43
BP
813#else
814 set_cpu_cap(c, X86_FEATURE_NOPL);
815#endif
d7cd5611
RR
816}
817
148f9bb8 818static void generic_identify(struct cpuinfo_x86 *c)
1da177e4 819{
aef93c8b 820 c->extended_cpuid_level = 0;
1da177e4 821
3da99c97 822 if (!have_cpuid_p())
aef93c8b 823 identify_cpu_without_cpuid(c);
1d67953f 824
aef93c8b 825 /* cyrix could have cpuid enabled via c_identify()*/
a9853dd6 826 if (!have_cpuid_p())
aef93c8b 827 return;
1da177e4 828
3da99c97 829 cpu_detect(c);
1da177e4 830
3da99c97 831 get_cpu_vendor(c);
1da177e4 832
3da99c97 833 get_cpu_cap(c);
1da177e4 834
3da99c97
YL
835 if (c->cpuid_level >= 0x00000001) {
836 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
b89d3b3e
YL
837#ifdef CONFIG_X86_32
838# ifdef CONFIG_X86_HT
cb8cc442 839 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
b89d3b3e 840# else
3da99c97 841 c->apicid = c->initial_apicid;
b89d3b3e
YL
842# endif
843#endif
b89d3b3e 844 c->phys_proc_id = c->initial_apicid;
3da99c97 845 }
1da177e4 846
1b05d60d 847 get_model_name(c); /* Default name */
1da177e4 848
3da99c97 849 detect_nopl(c);
1da177e4 850}
1da177e4 851
cbc82b17
PWJ
852static void x86_init_cache_qos(struct cpuinfo_x86 *c)
853{
854 /*
855 * The heavy lifting of max_rmid and cache_occ_scale are handled
856 * in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu
857 * in case CQM bits really aren't there in this CPU.
858 */
859 if (c != &boot_cpu_data) {
860 boot_cpu_data.x86_cache_max_rmid =
861 min(boot_cpu_data.x86_cache_max_rmid,
862 c->x86_cache_max_rmid);
863 }
864}
865
1da177e4
LT
866/*
867 * This does the hard work of actually picking apart the CPU stuff...
868 */
148f9bb8 869static void identify_cpu(struct cpuinfo_x86 *c)
1da177e4
LT
870{
871 int i;
872
873 c->loops_per_jiffy = loops_per_jiffy;
874 c->x86_cache_size = -1;
875 c->x86_vendor = X86_VENDOR_UNKNOWN;
1da177e4
LT
876 c->x86_model = c->x86_mask = 0; /* So far unknown... */
877 c->x86_vendor_id[0] = '\0'; /* Unset */
878 c->x86_model_id[0] = '\0'; /* Unset */
94605eff 879 c->x86_max_cores = 1;
102bbe3a 880 c->x86_coreid_bits = 0;
11fdd252 881#ifdef CONFIG_X86_64
102bbe3a 882 c->x86_clflush_size = 64;
13c6c532
JB
883 c->x86_phys_bits = 36;
884 c->x86_virt_bits = 48;
102bbe3a
YL
885#else
886 c->cpuid_level = -1; /* CPUID not detected */
770d132f 887 c->x86_clflush_size = 32;
13c6c532
JB
888 c->x86_phys_bits = 32;
889 c->x86_virt_bits = 32;
102bbe3a
YL
890#endif
891 c->x86_cache_alignment = c->x86_clflush_size;
1da177e4
LT
892 memset(&c->x86_capability, 0, sizeof c->x86_capability);
893
1da177e4
LT
894 generic_identify(c);
895
3898534d 896 if (this_cpu->c_identify)
1da177e4
LT
897 this_cpu->c_identify(c);
898
2759c328
YL
899 /* Clear/Set all flags overriden by options, after probe */
900 for (i = 0; i < NCAPINTS; i++) {
901 c->x86_capability[i] &= ~cpu_caps_cleared[i];
902 c->x86_capability[i] |= cpu_caps_set[i];
903 }
904
102bbe3a 905#ifdef CONFIG_X86_64
cb8cc442 906 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
102bbe3a
YL
907#endif
908
1da177e4
LT
909 /*
910 * Vendor-specific initialization. In this section we
911 * canonicalize the feature flags, meaning if there are
912 * features a certain CPU supports which CPUID doesn't
913 * tell us, CPUID claiming incorrect flags, or other bugs,
914 * we handle them here.
915 *
916 * At the end of this section, c->x86_capability better
917 * indicate the features this CPU genuinely supports!
918 */
919 if (this_cpu->c_init)
920 this_cpu->c_init(c);
921
922 /* Disable the PN if appropriate */
923 squash_the_stupid_serial_number(c);
924
b2cc2a07
PA
925 /* Set up SMEP/SMAP */
926 setup_smep(c);
927 setup_smap(c);
928
1da177e4 929 /*
0f3fa48a
IM
930 * The vendor-specific functions might have changed features.
931 * Now we do "generic changes."
1da177e4
LT
932 */
933
b38b0665
PA
934 /* Filter out anything that depends on CPUID levels we don't have */
935 filter_cpuid_features(c, true);
936
1da177e4 937 /* If the model name is still unset, do table lookup. */
34048c9e 938 if (!c->x86_model_id[0]) {
02dde8b4 939 const char *p;
1da177e4 940 p = table_lookup_model(c);
34048c9e 941 if (p)
1da177e4
LT
942 strcpy(c->x86_model_id, p);
943 else
944 /* Last resort... */
945 sprintf(c->x86_model_id, "%02x/%02x",
54a20f8c 946 c->x86, c->x86_model);
1da177e4
LT
947 }
948
102bbe3a
YL
949#ifdef CONFIG_X86_64
950 detect_ht(c);
951#endif
952
88b094fb 953 init_hypervisor(c);
49d859d7 954 x86_init_rdrand(c);
cbc82b17 955 x86_init_cache_qos(c);
3e0c3737
YL
956
957 /*
958 * Clear/Set all flags overriden by options, need do it
959 * before following smp all cpus cap AND.
960 */
961 for (i = 0; i < NCAPINTS; i++) {
962 c->x86_capability[i] &= ~cpu_caps_cleared[i];
963 c->x86_capability[i] |= cpu_caps_set[i];
964 }
965
1da177e4
LT
966 /*
967 * On SMP, boot_cpu_data holds the common feature set between
968 * all CPUs; so make sure that we indicate which features are
969 * common between the CPUs. The first time this routine gets
970 * executed, c == &boot_cpu_data.
971 */
34048c9e 972 if (c != &boot_cpu_data) {
1da177e4 973 /* AND the already accumulated flags with these */
9d31d35b 974 for (i = 0; i < NCAPINTS; i++)
1da177e4 975 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
65fc985b
BP
976
977 /* OR, i.e. replicate the bug flags */
978 for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
979 c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1da177e4
LT
980 }
981
982 /* Init Machine Check Exception if available. */
5e09954a 983 mcheck_cpu_init(c);
30d432df
AK
984
985 select_idle_routine(c);
102bbe3a 986
de2d9445 987#ifdef CONFIG_NUMA
102bbe3a
YL
988 numa_add_cpu(smp_processor_id());
989#endif
a6c4e076 990}
31ab269a 991
8b6c0ab1
IM
992/*
993 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
994 * on 32-bit kernels:
995 */
cfda7bb9
AL
996#ifdef CONFIG_X86_32
997void enable_sep_cpu(void)
998{
8b6c0ab1
IM
999 struct tss_struct *tss;
1000 int cpu;
cfda7bb9 1001
8b6c0ab1
IM
1002 cpu = get_cpu();
1003 tss = &per_cpu(cpu_tss, cpu);
1004
1005 if (!boot_cpu_has(X86_FEATURE_SEP))
1006 goto out;
1007
1008 /*
cf9328cc
AL
1009 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1010 * see the big comment in struct x86_hw_tss's definition.
8b6c0ab1 1011 */
cfda7bb9
AL
1012
1013 tss->x86_tss.ss1 = __KERNEL_CS;
8b6c0ab1
IM
1014 wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1015
cf9328cc
AL
1016 wrmsr(MSR_IA32_SYSENTER_ESP,
1017 (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
1018 0);
8b6c0ab1
IM
1019
1020 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)ia32_sysenter_target, 0);
1021
1022out:
cfda7bb9
AL
1023 put_cpu();
1024}
e04d645f
GC
1025#endif
1026
a6c4e076
JF
1027void __init identify_boot_cpu(void)
1028{
1029 identify_cpu(&boot_cpu_data);
02c68a02 1030 init_amd_e400_c1e_mask();
102bbe3a 1031#ifdef CONFIG_X86_32
a6c4e076 1032 sysenter_setup();
6fe940d6 1033 enable_sep_cpu();
102bbe3a 1034#endif
5b556332 1035 cpu_detect_tlb(&boot_cpu_data);
a6c4e076 1036}
3b520b23 1037
148f9bb8 1038void identify_secondary_cpu(struct cpuinfo_x86 *c)
a6c4e076
JF
1039{
1040 BUG_ON(c == &boot_cpu_data);
1041 identify_cpu(c);
102bbe3a 1042#ifdef CONFIG_X86_32
a6c4e076 1043 enable_sep_cpu();
102bbe3a 1044#endif
a6c4e076 1045 mtrr_ap_init();
1da177e4
LT
1046}
1047
a0854a46 1048struct msr_range {
0f3fa48a
IM
1049 unsigned min;
1050 unsigned max;
a0854a46 1051};
1da177e4 1052
148f9bb8 1053static const struct msr_range msr_range_array[] = {
a0854a46
YL
1054 { 0x00000000, 0x00000418},
1055 { 0xc0000000, 0xc000040b},
1056 { 0xc0010000, 0xc0010142},
1057 { 0xc0011000, 0xc001103b},
1058};
1da177e4 1059
148f9bb8 1060static void __print_cpu_msr(void)
a0854a46 1061{
0f3fa48a 1062 unsigned index_min, index_max;
a0854a46
YL
1063 unsigned index;
1064 u64 val;
1065 int i;
a0854a46
YL
1066
1067 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1068 index_min = msr_range_array[i].min;
1069 index_max = msr_range_array[i].max;
0f3fa48a 1070
a0854a46 1071 for (index = index_min; index < index_max; index++) {
ecd431d9 1072 if (rdmsrl_safe(index, &val))
a0854a46
YL
1073 continue;
1074 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1da177e4 1075 }
a0854a46
YL
1076 }
1077}
94605eff 1078
148f9bb8 1079static int show_msr;
0f3fa48a 1080
a0854a46
YL
1081static __init int setup_show_msr(char *arg)
1082{
1083 int num;
3dd9d514 1084
a0854a46 1085 get_option(&arg, &num);
3dd9d514 1086
a0854a46
YL
1087 if (num > 0)
1088 show_msr = num;
1089 return 1;
1da177e4 1090}
a0854a46 1091__setup("show_msr=", setup_show_msr);
1da177e4 1092
191679fd
AK
1093static __init int setup_noclflush(char *arg)
1094{
840d2830 1095 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
da4aaa7d 1096 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
191679fd
AK
1097 return 1;
1098}
1099__setup("noclflush", setup_noclflush);
1100
148f9bb8 1101void print_cpu_info(struct cpuinfo_x86 *c)
1da177e4 1102{
02dde8b4 1103 const char *vendor = NULL;
1da177e4 1104
0f3fa48a 1105 if (c->x86_vendor < X86_VENDOR_NUM) {
1da177e4 1106 vendor = this_cpu->c_vendor;
0f3fa48a
IM
1107 } else {
1108 if (c->cpuid_level >= 0)
1109 vendor = c->x86_vendor_id;
1110 }
1da177e4 1111
bd32a8cf 1112 if (vendor && !strstr(c->x86_model_id, vendor))
9d31d35b 1113 printk(KERN_CONT "%s ", vendor);
1da177e4 1114
9d31d35b 1115 if (c->x86_model_id[0])
adafb98d 1116 printk(KERN_CONT "%s", c->x86_model_id);
1da177e4 1117 else
9d31d35b 1118 printk(KERN_CONT "%d86", c->x86);
1da177e4 1119
924e101a
BP
1120 printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1121
34048c9e 1122 if (c->x86_mask || c->cpuid_level >= 0)
924e101a 1123 printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1da177e4 1124 else
924e101a 1125 printk(KERN_CONT ")\n");
a0854a46 1126
0b8b8078 1127 print_cpu_msr(c);
21c3fcf3
YL
1128}
1129
148f9bb8 1130void print_cpu_msr(struct cpuinfo_x86 *c)
21c3fcf3 1131{
a0854a46 1132 if (c->cpu_index < show_msr)
21c3fcf3 1133 __print_cpu_msr();
1da177e4
LT
1134}
1135
ac72e788
AK
1136static __init int setup_disablecpuid(char *arg)
1137{
1138 int bit;
0f3fa48a 1139
ac72e788
AK
1140 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1141 setup_clear_cpu_cap(bit);
1142 else
1143 return 0;
0f3fa48a 1144
ac72e788
AK
1145 return 1;
1146}
1147__setup("clearcpuid=", setup_disablecpuid);
1148
198d208d 1149DEFINE_PER_CPU(unsigned long, kernel_stack) =
ef593260 1150 (unsigned long)&init_thread_union + THREAD_SIZE;
198d208d
SR
1151EXPORT_PER_CPU_SYMBOL(kernel_stack);
1152
d5494d4f 1153#ifdef CONFIG_X86_64
9ff80942 1154struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
629f4f9d
SA
1155struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1156 (unsigned long) debug_idt_table };
d5494d4f 1157
947e76cd 1158DEFINE_PER_CPU_FIRST(union irq_stack_union,
277d5b40 1159 irq_stack_union) __aligned(PAGE_SIZE) __visible;
0f3fa48a 1160
bdf977b3 1161/*
a7fcf28d
AL
1162 * The following percpu variables are hot. Align current_task to
1163 * cacheline size such that they fall in the same cacheline.
bdf977b3
TH
1164 */
1165DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1166 &init_task;
1167EXPORT_PER_CPU_SYMBOL(current_task);
d5494d4f 1168
bdf977b3
TH
1169DEFINE_PER_CPU(char *, irq_stack_ptr) =
1170 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1171
277d5b40 1172DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
d5494d4f 1173
c2daa3be
PZ
1174DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1175EXPORT_PER_CPU_SYMBOL(__preempt_count);
1176
7e16838d
LT
1177DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1178
0f3fa48a
IM
1179/*
1180 * Special IST stacks which the CPU switches to when it calls
1181 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1182 * limit), all of them are 4K, except the debug stack which
1183 * is 8K.
1184 */
1185static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1186 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1187 [DEBUG_STACK - 1] = DEBUG_STKSZ
1188};
1189
92d65b23 1190static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
3e352aa8 1191 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
d5494d4f 1192
d5494d4f
YL
1193/* May not be marked __init: used by software suspend */
1194void syscall_init(void)
1da177e4 1195{
d5494d4f
YL
1196 /*
1197 * LSTAR and STAR live in a bit strange symbiosis.
1198 * They both write to the same internal register. STAR allows to
1199 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1200 */
1201 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
1202 wrmsrl(MSR_LSTAR, system_call);
d56fe4bf
IM
1203
1204#ifdef CONFIG_IA32_EMULATION
a76c7f46
DV
1205 wrmsrl(MSR_CSTAR, ia32_cstar_target);
1206 /*
487d1edb
DV
1207 * This only works on Intel CPUs.
1208 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1209 * This does not cause SYSENTER to jump to the wrong location, because
1210 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
a76c7f46
DV
1211 */
1212 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1213 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1214 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target);
d56fe4bf
IM
1215#else
1216 wrmsrl(MSR_CSTAR, ignore_sysret);
6b51311c 1217 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
d56fe4bf
IM
1218 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1219 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
d5494d4f 1220#endif
03ae5768 1221
d5494d4f
YL
1222 /* Flags to clear on syscall */
1223 wrmsrl(MSR_SYSCALL_MASK,
63bcff2a 1224 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
8c7aa698 1225 X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1da177e4 1226}
62111195 1227
d5494d4f
YL
1228/*
1229 * Copies of the original ist values from the tss are only accessed during
1230 * debugging, no special alignment required.
1231 */
1232DEFINE_PER_CPU(struct orig_ist, orig_ist);
1233
228bdaa9 1234static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
42181186 1235DEFINE_PER_CPU(int, debug_stack_usage);
228bdaa9
SR
1236
1237int is_debug_stack(unsigned long addr)
1238{
89cbc767
CL
1239 return __this_cpu_read(debug_stack_usage) ||
1240 (addr <= __this_cpu_read(debug_stack_addr) &&
1241 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
228bdaa9 1242}
0f46efeb 1243NOKPROBE_SYMBOL(is_debug_stack);
228bdaa9 1244
629f4f9d 1245DEFINE_PER_CPU(u32, debug_idt_ctr);
f8988175 1246
228bdaa9
SR
1247void debug_stack_set_zero(void)
1248{
629f4f9d
SA
1249 this_cpu_inc(debug_idt_ctr);
1250 load_current_idt();
228bdaa9 1251}
0f46efeb 1252NOKPROBE_SYMBOL(debug_stack_set_zero);
228bdaa9
SR
1253
1254void debug_stack_reset(void)
1255{
629f4f9d 1256 if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
f8988175 1257 return;
629f4f9d
SA
1258 if (this_cpu_dec_return(debug_idt_ctr) == 0)
1259 load_current_idt();
228bdaa9 1260}
0f46efeb 1261NOKPROBE_SYMBOL(debug_stack_reset);
228bdaa9 1262
0f3fa48a 1263#else /* CONFIG_X86_64 */
d5494d4f 1264
bdf977b3
TH
1265DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1266EXPORT_PER_CPU_SYMBOL(current_task);
c2daa3be
PZ
1267DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1268EXPORT_PER_CPU_SYMBOL(__preempt_count);
27e74da9 1269DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
bdf977b3 1270
a7fcf28d
AL
1271/*
1272 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1273 * the top of the kernel stack. Use an extra percpu variable to track the
1274 * top of the kernel stack directly.
1275 */
1276DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1277 (unsigned long)&init_thread_union + THREAD_SIZE;
1278EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1279
60a5317f 1280#ifdef CONFIG_CC_STACKPROTECTOR
53f82452 1281DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
60a5317f 1282#endif
d5494d4f 1283
0f3fa48a 1284#endif /* CONFIG_X86_64 */
c5413fbe 1285
9766cdbc
JSR
1286/*
1287 * Clear all 6 debug registers:
1288 */
1289static void clear_all_debug_regs(void)
1290{
1291 int i;
1292
1293 for (i = 0; i < 8; i++) {
1294 /* Ignore db4, db5 */
1295 if ((i == 4) || (i == 5))
1296 continue;
1297
1298 set_debugreg(0, i);
1299 }
1300}
c5413fbe 1301
0bb9fef9
JW
1302#ifdef CONFIG_KGDB
1303/*
1304 * Restore debug regs if using kgdbwait and you have a kernel debugger
1305 * connection established.
1306 */
1307static void dbg_restore_debug_regs(void)
1308{
1309 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1310 arch_kgdb_ops.correct_hw_break();
1311}
1312#else /* ! CONFIG_KGDB */
1313#define dbg_restore_debug_regs()
1314#endif /* ! CONFIG_KGDB */
1315
ce4b1b16
IM
1316static void wait_for_master_cpu(int cpu)
1317{
1318#ifdef CONFIG_SMP
1319 /*
1320 * wait for ACK from master CPU before continuing
1321 * with AP initialization
1322 */
1323 WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1324 while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1325 cpu_relax();
1326#endif
1327}
1328
d2cbcc49
RR
1329/*
1330 * cpu_init() initializes state that is per-CPU. Some data is already
1331 * initialized (naturally) in the bootstrap process, such as the GDT
1332 * and IDT. We reload them nevertheless, this function acts as a
1333 * 'CPU state barrier', nothing should get across.
1ba76586 1334 * A lot of state is already set up in PDA init for 64 bit
d2cbcc49 1335 */
1ba76586 1336#ifdef CONFIG_X86_64
0f3fa48a 1337
148f9bb8 1338void cpu_init(void)
1ba76586 1339{
0fe1e009 1340 struct orig_ist *oist;
1ba76586 1341 struct task_struct *me;
0f3fa48a
IM
1342 struct tss_struct *t;
1343 unsigned long v;
ce4b1b16 1344 int cpu = stack_smp_processor_id();
1ba76586
YL
1345 int i;
1346
ce4b1b16
IM
1347 wait_for_master_cpu(cpu);
1348
1e02ce4c
AL
1349 /*
1350 * Initialize the CR4 shadow before doing anything that could
1351 * try to read it.
1352 */
1353 cr4_init_shadow();
1354
e6ebf5de
FY
1355 /*
1356 * Load microcode on this cpu if a valid microcode is available.
1357 * This is early microcode loading procedure.
1358 */
1359 load_ucode_ap();
1360
24933b82 1361 t = &per_cpu(cpu_tss, cpu);
0fe1e009 1362 oist = &per_cpu(orig_ist, cpu);
0f3fa48a 1363
e7a22c1e 1364#ifdef CONFIG_NUMA
27fd185f 1365 if (this_cpu_read(numa_node) == 0 &&
e534c7c5
LS
1366 early_cpu_to_node(cpu) != NUMA_NO_NODE)
1367 set_numa_node(early_cpu_to_node(cpu));
e7a22c1e 1368#endif
1ba76586
YL
1369
1370 me = current;
1371
2eaad1fd 1372 pr_debug("Initializing CPU#%d\n", cpu);
1ba76586 1373
375074cc 1374 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1ba76586
YL
1375
1376 /*
1377 * Initialize the per-CPU GDT with the boot GDT,
1378 * and set up the GDT descriptor:
1379 */
1380
552be871 1381 switch_to_new_gdt(cpu);
2697fbd5
BG
1382 loadsegment(fs, 0);
1383
cf910e83 1384 load_current_idt();
1ba76586
YL
1385
1386 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1387 syscall_init();
1388
1389 wrmsrl(MSR_FS_BASE, 0);
1390 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1391 barrier();
1392
4763ed4d 1393 x86_configure_nx();
659006bf 1394 x2apic_setup();
1ba76586
YL
1395
1396 /*
1397 * set up and load the per-CPU TSS
1398 */
0fe1e009 1399 if (!oist->ist[0]) {
92d65b23 1400 char *estacks = per_cpu(exception_stacks, cpu);
0f3fa48a 1401
1ba76586 1402 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
0f3fa48a 1403 estacks += exception_stack_sizes[v];
0fe1e009 1404 oist->ist[v] = t->x86_tss.ist[v] =
1ba76586 1405 (unsigned long)estacks;
228bdaa9
SR
1406 if (v == DEBUG_STACK-1)
1407 per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1ba76586
YL
1408 }
1409 }
1410
1411 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
0f3fa48a 1412
1ba76586
YL
1413 /*
1414 * <= is required because the CPU will access up to
1415 * 8 bits beyond the end of the IO permission bitmap.
1416 */
1417 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1418 t->io_bitmap[i] = ~0UL;
1419
1420 atomic_inc(&init_mm.mm_count);
1421 me->active_mm = &init_mm;
8c5dfd25 1422 BUG_ON(me->mm);
1ba76586
YL
1423 enter_lazy_tlb(&init_mm, me);
1424
1425 load_sp0(t, &current->thread);
1426 set_tss_desc(cpu, t);
1427 load_TR_desc();
1428 load_LDT(&init_mm.context);
1429
0bb9fef9
JW
1430 clear_all_debug_regs();
1431 dbg_restore_debug_regs();
1ba76586
YL
1432
1433 fpu_init();
1434
1ba76586
YL
1435 if (is_uv_system())
1436 uv_cpu_init();
1437}
1438
1439#else
1440
148f9bb8 1441void cpu_init(void)
9ee79a3d 1442{
d2cbcc49
RR
1443 int cpu = smp_processor_id();
1444 struct task_struct *curr = current;
24933b82 1445 struct tss_struct *t = &per_cpu(cpu_tss, cpu);
9ee79a3d 1446 struct thread_struct *thread = &curr->thread;
62111195 1447
ce4b1b16 1448 wait_for_master_cpu(cpu);
e6ebf5de 1449
5b2bdbc8
SR
1450 /*
1451 * Initialize the CR4 shadow before doing anything that could
1452 * try to read it.
1453 */
1454 cr4_init_shadow();
1455
ce4b1b16 1456 show_ucode_info_early();
62111195
JF
1457
1458 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1459
9298b815 1460 if (cpu_feature_enabled(X86_FEATURE_VME) || cpu_has_tsc || cpu_has_de)
375074cc 1461 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
62111195 1462
cf910e83 1463 load_current_idt();
552be871 1464 switch_to_new_gdt(cpu);
1da177e4 1465
1da177e4
LT
1466 /*
1467 * Set up and load the per-CPU TSS and LDT
1468 */
1469 atomic_inc(&init_mm.mm_count);
62111195 1470 curr->active_mm = &init_mm;
8c5dfd25 1471 BUG_ON(curr->mm);
62111195 1472 enter_lazy_tlb(&init_mm, curr);
1da177e4 1473
faca6227 1474 load_sp0(t, thread);
34048c9e 1475 set_tss_desc(cpu, t);
1da177e4
LT
1476 load_TR_desc();
1477 load_LDT(&init_mm.context);
1478
f9a196b8
TG
1479 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1480
22c4e308 1481#ifdef CONFIG_DOUBLEFAULT
1da177e4
LT
1482 /* Set up doublefault TSS pointer in the GDT */
1483 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
22c4e308 1484#endif
1da177e4 1485
9766cdbc 1486 clear_all_debug_regs();
0bb9fef9 1487 dbg_restore_debug_regs();
1da177e4 1488
0e49bf66 1489 fpu_init();
1da177e4 1490}
1ba76586 1491#endif
5700f743
BP
1492
1493#ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1494void warn_pre_alternatives(void)
1495{
1496 WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1497}
1498EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1499#endif
4a90a99c
BP
1500
1501inline bool __static_cpu_has_safe(u16 bit)
1502{
1503 return boot_cpu_has(bit);
1504}
1505EXPORT_SYMBOL_GPL(__static_cpu_has_safe);