]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - arch/x86/kernel/cpu/common.c
x86: Add support for Intel Cache QoS Monitoring (CQM) detection
[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>
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;
422 char *p, *q;
423
3da99c97 424 if (c->extended_cpuid_level < 0x80000004)
1b05d60d 425 return;
1da177e4 426
0f3fa48a 427 v = (unsigned int *)c->x86_model_id;
1da177e4
LT
428 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
429 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
430 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
431 c->x86_model_id[48] = 0;
432
0f3fa48a
IM
433 /*
434 * Intel chips right-justify this string for some dumb reason;
435 * undo that brain damage:
436 */
1da177e4 437 p = q = &c->x86_model_id[0];
34048c9e 438 while (*p == ' ')
9766cdbc 439 p++;
34048c9e 440 if (p != q) {
9766cdbc
JSR
441 while (*p)
442 *q++ = *p++;
443 while (q <= &c->x86_model_id[48])
444 *q++ = '\0'; /* Zero-pad the rest */
1da177e4 445 }
1da177e4
LT
446}
447
148f9bb8 448void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
1da177e4 449{
9d31d35b 450 unsigned int n, dummy, ebx, ecx, edx, l2size;
1da177e4 451
3da99c97 452 n = c->extended_cpuid_level;
1da177e4
LT
453
454 if (n >= 0x80000005) {
9d31d35b 455 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
9d31d35b 456 c->x86_cache_size = (ecx>>24) + (edx>>24);
140fc727
YL
457#ifdef CONFIG_X86_64
458 /* On K8 L1 TLB is inclusive, so don't count it */
459 c->x86_tlbsize = 0;
460#endif
1da177e4
LT
461 }
462
463 if (n < 0x80000006) /* Some chips just has a large L1. */
464 return;
465
0a488a53 466 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
1da177e4 467 l2size = ecx >> 16;
34048c9e 468
140fc727
YL
469#ifdef CONFIG_X86_64
470 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
471#else
1da177e4 472 /* do processor-specific cache resizing */
09dc68d9
JB
473 if (this_cpu->legacy_cache_size)
474 l2size = this_cpu->legacy_cache_size(c, l2size);
1da177e4
LT
475
476 /* Allow user to override all this if necessary. */
477 if (cachesize_override != -1)
478 l2size = cachesize_override;
479
34048c9e 480 if (l2size == 0)
1da177e4 481 return; /* Again, no L2 cache is possible */
140fc727 482#endif
1da177e4
LT
483
484 c->x86_cache_size = l2size;
1da177e4
LT
485}
486
e0ba94f1
AS
487u16 __read_mostly tlb_lli_4k[NR_INFO];
488u16 __read_mostly tlb_lli_2m[NR_INFO];
489u16 __read_mostly tlb_lli_4m[NR_INFO];
490u16 __read_mostly tlb_lld_4k[NR_INFO];
491u16 __read_mostly tlb_lld_2m[NR_INFO];
492u16 __read_mostly tlb_lld_4m[NR_INFO];
dd360393 493u16 __read_mostly tlb_lld_1g[NR_INFO];
e0ba94f1 494
148f9bb8 495void cpu_detect_tlb(struct cpuinfo_x86 *c)
e0ba94f1
AS
496{
497 if (this_cpu->c_detect_tlb)
498 this_cpu->c_detect_tlb(c);
499
dd360393 500 printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n"
e9f4e0a9 501 "Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
e0ba94f1
AS
502 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
503 tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES],
c4211f42 504 tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES],
e9f4e0a9 505 tlb_lld_1g[ENTRIES]);
e0ba94f1
AS
506}
507
148f9bb8 508void detect_ht(struct cpuinfo_x86 *c)
1da177e4 509{
97e4db7c 510#ifdef CONFIG_X86_HT
0a488a53
YL
511 u32 eax, ebx, ecx, edx;
512 int index_msb, core_bits;
2eaad1fd 513 static bool printed;
1da177e4 514
0a488a53 515 if (!cpu_has(c, X86_FEATURE_HT))
9d31d35b 516 return;
1da177e4 517
0a488a53
YL
518 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
519 goto out;
1da177e4 520
1cd78776
YL
521 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
522 return;
1da177e4 523
0a488a53 524 cpuid(1, &eax, &ebx, &ecx, &edx);
1da177e4 525
9d31d35b
YL
526 smp_num_siblings = (ebx & 0xff0000) >> 16;
527
528 if (smp_num_siblings == 1) {
2eaad1fd 529 printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
0f3fa48a
IM
530 goto out;
531 }
9d31d35b 532
0f3fa48a
IM
533 if (smp_num_siblings <= 1)
534 goto out;
9d31d35b 535
0f3fa48a
IM
536 index_msb = get_count_order(smp_num_siblings);
537 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
9d31d35b 538
0f3fa48a 539 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
9d31d35b 540
0f3fa48a 541 index_msb = get_count_order(smp_num_siblings);
9d31d35b 542
0f3fa48a 543 core_bits = get_count_order(c->x86_max_cores);
9d31d35b 544
0f3fa48a
IM
545 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
546 ((1 << core_bits) - 1);
1da177e4 547
0a488a53 548out:
2eaad1fd 549 if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
0a488a53
YL
550 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
551 c->phys_proc_id);
552 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
553 c->cpu_core_id);
2eaad1fd 554 printed = 1;
9d31d35b 555 }
9d31d35b 556#endif
97e4db7c 557}
1da177e4 558
148f9bb8 559static void get_cpu_vendor(struct cpuinfo_x86 *c)
1da177e4
LT
560{
561 char *v = c->x86_vendor_id;
0f3fa48a 562 int i;
1da177e4
LT
563
564 for (i = 0; i < X86_VENDOR_NUM; i++) {
10a434fc
YL
565 if (!cpu_devs[i])
566 break;
567
568 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
569 (cpu_devs[i]->c_ident[1] &&
570 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
0f3fa48a 571
10a434fc
YL
572 this_cpu = cpu_devs[i];
573 c->x86_vendor = this_cpu->c_x86_vendor;
574 return;
1da177e4
LT
575 }
576 }
10a434fc 577
a9c56953
MK
578 printk_once(KERN_ERR
579 "CPU: vendor_id '%s' unknown, using generic init.\n" \
580 "CPU: Your system may be unstable.\n", v);
10a434fc 581
fe38d855
CE
582 c->x86_vendor = X86_VENDOR_UNKNOWN;
583 this_cpu = &default_cpu;
1da177e4
LT
584}
585
148f9bb8 586void cpu_detect(struct cpuinfo_x86 *c)
1da177e4 587{
1da177e4 588 /* Get vendor name */
4a148513
HH
589 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
590 (unsigned int *)&c->x86_vendor_id[0],
591 (unsigned int *)&c->x86_vendor_id[8],
592 (unsigned int *)&c->x86_vendor_id[4]);
1da177e4 593
1da177e4 594 c->x86 = 4;
9d31d35b 595 /* Intel-defined flags: level 0x00000001 */
1da177e4
LT
596 if (c->cpuid_level >= 0x00000001) {
597 u32 junk, tfms, cap0, misc;
0f3fa48a 598
1da177e4 599 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
9d31d35b
YL
600 c->x86 = (tfms >> 8) & 0xf;
601 c->x86_model = (tfms >> 4) & 0xf;
602 c->x86_mask = tfms & 0xf;
0f3fa48a 603
f5f786d0 604 if (c->x86 == 0xf)
1da177e4 605 c->x86 += (tfms >> 20) & 0xff;
f5f786d0 606 if (c->x86 >= 0x6)
9d31d35b 607 c->x86_model += ((tfms >> 16) & 0xf) << 4;
0f3fa48a 608
d4387bd3 609 if (cap0 & (1<<19)) {
d4387bd3 610 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
9d31d35b 611 c->x86_cache_alignment = c->x86_clflush_size;
d4387bd3 612 }
1da177e4 613 }
1da177e4 614}
3da99c97 615
148f9bb8 616void get_cpu_cap(struct cpuinfo_x86 *c)
093af8d7
YL
617{
618 u32 tfms, xlvl;
3da99c97 619 u32 ebx;
093af8d7 620
3da99c97
YL
621 /* Intel-defined flags: level 0x00000001 */
622 if (c->cpuid_level >= 0x00000001) {
623 u32 capability, excap;
0f3fa48a 624
3da99c97
YL
625 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
626 c->x86_capability[0] = capability;
627 c->x86_capability[4] = excap;
628 }
093af8d7 629
bdc802dc
PA
630 /* Additional Intel-defined flags: level 0x00000007 */
631 if (c->cpuid_level >= 0x00000007) {
632 u32 eax, ebx, ecx, edx;
633
634 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
635
2494b030 636 c->x86_capability[9] = ebx;
bdc802dc
PA
637 }
638
6229ad27
FY
639 /* Extended state features: level 0x0000000d */
640 if (c->cpuid_level >= 0x0000000d) {
641 u32 eax, ebx, ecx, edx;
642
643 cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
644
645 c->x86_capability[10] = eax;
646 }
647
cbc82b17
PWJ
648 /* Additional Intel-defined flags: level 0x0000000F */
649 if (c->cpuid_level >= 0x0000000F) {
650 u32 eax, ebx, ecx, edx;
651
652 /* QoS sub-leaf, EAX=0Fh, ECX=0 */
653 cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
654 c->x86_capability[11] = edx;
655 if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
656 /* will be overridden if occupancy monitoring exists */
657 c->x86_cache_max_rmid = ebx;
658
659 /* QoS sub-leaf, EAX=0Fh, ECX=1 */
660 cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
661 c->x86_capability[12] = edx;
662 if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) {
663 c->x86_cache_max_rmid = ecx;
664 c->x86_cache_occ_scale = ebx;
665 }
666 } else {
667 c->x86_cache_max_rmid = -1;
668 c->x86_cache_occ_scale = -1;
669 }
670 }
671
3da99c97
YL
672 /* AMD-defined flags: level 0x80000001 */
673 xlvl = cpuid_eax(0x80000000);
674 c->extended_cpuid_level = xlvl;
0f3fa48a 675
3da99c97
YL
676 if ((xlvl & 0xffff0000) == 0x80000000) {
677 if (xlvl >= 0x80000001) {
678 c->x86_capability[1] = cpuid_edx(0x80000001);
679 c->x86_capability[6] = cpuid_ecx(0x80000001);
093af8d7 680 }
093af8d7 681 }
093af8d7 682
5122c890
YL
683 if (c->extended_cpuid_level >= 0x80000008) {
684 u32 eax = cpuid_eax(0x80000008);
685
686 c->x86_virt_bits = (eax >> 8) & 0xff;
687 c->x86_phys_bits = eax & 0xff;
093af8d7 688 }
13c6c532
JB
689#ifdef CONFIG_X86_32
690 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
691 c->x86_phys_bits = 36;
5122c890 692#endif
e3224234
YL
693
694 if (c->extended_cpuid_level >= 0x80000007)
695 c->x86_power = cpuid_edx(0x80000007);
093af8d7 696
1dedefd1 697 init_scattered_cpuid_features(c);
093af8d7 698}
1da177e4 699
148f9bb8 700static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
aef93c8b
YL
701{
702#ifdef CONFIG_X86_32
703 int i;
704
705 /*
706 * First of all, decide if this is a 486 or higher
707 * It's a 486 if we can modify the AC flag
708 */
709 if (flag_is_changeable_p(X86_EFLAGS_AC))
710 c->x86 = 4;
711 else
712 c->x86 = 3;
713
714 for (i = 0; i < X86_VENDOR_NUM; i++)
715 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
716 c->x86_vendor_id[0] = 0;
717 cpu_devs[i]->c_identify(c);
718 if (c->x86_vendor_id[0]) {
719 get_cpu_vendor(c);
720 break;
721 }
722 }
723#endif
724}
725
34048c9e
PC
726/*
727 * Do minimum CPU detection early.
728 * Fields really needed: vendor, cpuid_level, family, model, mask,
729 * cache alignment.
730 * The others are not touched to avoid unwanted side effects.
731 *
732 * WARNING: this function is only called on the BP. Don't add code here
733 * that is supposed to run on all CPUs.
734 */
3da99c97 735static void __init early_identify_cpu(struct cpuinfo_x86 *c)
d7cd5611 736{
6627d242
YL
737#ifdef CONFIG_X86_64
738 c->x86_clflush_size = 64;
13c6c532
JB
739 c->x86_phys_bits = 36;
740 c->x86_virt_bits = 48;
6627d242 741#else
d4387bd3 742 c->x86_clflush_size = 32;
13c6c532
JB
743 c->x86_phys_bits = 32;
744 c->x86_virt_bits = 32;
6627d242 745#endif
0a488a53 746 c->x86_cache_alignment = c->x86_clflush_size;
d7cd5611 747
3da99c97 748 memset(&c->x86_capability, 0, sizeof c->x86_capability);
0a488a53 749 c->extended_cpuid_level = 0;
d7cd5611 750
aef93c8b
YL
751 if (!have_cpuid_p())
752 identify_cpu_without_cpuid(c);
753
754 /* cyrix could have cpuid enabled via c_identify()*/
d7cd5611
RR
755 if (!have_cpuid_p())
756 return;
757
758 cpu_detect(c);
3da99c97 759 get_cpu_vendor(c);
3da99c97 760 get_cpu_cap(c);
60e019eb 761 fpu_detect(c);
12cf105c 762
10a434fc
YL
763 if (this_cpu->c_early_init)
764 this_cpu->c_early_init(c);
093af8d7 765
f6e9456c 766 c->cpu_index = 0;
b38b0665 767 filter_cpuid_features(c, false);
de5397ad 768
a110b5ec
BP
769 if (this_cpu->c_bsp_init)
770 this_cpu->c_bsp_init(c);
c3b83598
BP
771
772 setup_force_cpu_cap(X86_FEATURE_ALWAYS);
d7cd5611
RR
773}
774
9d31d35b
YL
775void __init early_cpu_init(void)
776{
02dde8b4 777 const struct cpu_dev *const *cdev;
10a434fc
YL
778 int count = 0;
779
ac23f253 780#ifdef CONFIG_PROCESSOR_SELECT
9766cdbc 781 printk(KERN_INFO "KERNEL supported cpus:\n");
31c997ca
IM
782#endif
783
10a434fc 784 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
02dde8b4 785 const struct cpu_dev *cpudev = *cdev;
9d31d35b 786
10a434fc
YL
787 if (count >= X86_VENDOR_NUM)
788 break;
789 cpu_devs[count] = cpudev;
790 count++;
791
ac23f253 792#ifdef CONFIG_PROCESSOR_SELECT
31c997ca
IM
793 {
794 unsigned int j;
795
796 for (j = 0; j < 2; j++) {
797 if (!cpudev->c_ident[j])
798 continue;
799 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
800 cpudev->c_ident[j]);
801 }
10a434fc 802 }
0388423d 803#endif
10a434fc 804 }
9d31d35b 805 early_identify_cpu(&boot_cpu_data);
d7cd5611 806}
093af8d7 807
b6734c35 808/*
366d4a43
BP
809 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
810 * unfortunately, that's not true in practice because of early VIA
811 * chips and (more importantly) broken virtualizers that are not easy
812 * to detect. In the latter case it doesn't even *fail* reliably, so
813 * probing for it doesn't even work. Disable it completely on 32-bit
ba0593bf 814 * unless we can find a reliable way to detect all the broken cases.
366d4a43 815 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
b6734c35 816 */
148f9bb8 817static void detect_nopl(struct cpuinfo_x86 *c)
b6734c35 818{
366d4a43 819#ifdef CONFIG_X86_32
b6734c35 820 clear_cpu_cap(c, X86_FEATURE_NOPL);
366d4a43
BP
821#else
822 set_cpu_cap(c, X86_FEATURE_NOPL);
823#endif
d7cd5611
RR
824}
825
148f9bb8 826static void generic_identify(struct cpuinfo_x86 *c)
1da177e4 827{
aef93c8b 828 c->extended_cpuid_level = 0;
1da177e4 829
3da99c97 830 if (!have_cpuid_p())
aef93c8b 831 identify_cpu_without_cpuid(c);
1d67953f 832
aef93c8b 833 /* cyrix could have cpuid enabled via c_identify()*/
a9853dd6 834 if (!have_cpuid_p())
aef93c8b 835 return;
1da177e4 836
3da99c97 837 cpu_detect(c);
1da177e4 838
3da99c97 839 get_cpu_vendor(c);
1da177e4 840
3da99c97 841 get_cpu_cap(c);
1da177e4 842
3da99c97
YL
843 if (c->cpuid_level >= 0x00000001) {
844 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
b89d3b3e
YL
845#ifdef CONFIG_X86_32
846# ifdef CONFIG_X86_HT
cb8cc442 847 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
b89d3b3e 848# else
3da99c97 849 c->apicid = c->initial_apicid;
b89d3b3e
YL
850# endif
851#endif
b89d3b3e 852 c->phys_proc_id = c->initial_apicid;
3da99c97 853 }
1da177e4 854
1b05d60d 855 get_model_name(c); /* Default name */
1da177e4 856
3da99c97 857 detect_nopl(c);
1da177e4 858}
1da177e4 859
cbc82b17
PWJ
860static void x86_init_cache_qos(struct cpuinfo_x86 *c)
861{
862 /*
863 * The heavy lifting of max_rmid and cache_occ_scale are handled
864 * in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu
865 * in case CQM bits really aren't there in this CPU.
866 */
867 if (c != &boot_cpu_data) {
868 boot_cpu_data.x86_cache_max_rmid =
869 min(boot_cpu_data.x86_cache_max_rmid,
870 c->x86_cache_max_rmid);
871 }
872}
873
1da177e4
LT
874/*
875 * This does the hard work of actually picking apart the CPU stuff...
876 */
148f9bb8 877static void identify_cpu(struct cpuinfo_x86 *c)
1da177e4
LT
878{
879 int i;
880
881 c->loops_per_jiffy = loops_per_jiffy;
882 c->x86_cache_size = -1;
883 c->x86_vendor = X86_VENDOR_UNKNOWN;
1da177e4
LT
884 c->x86_model = c->x86_mask = 0; /* So far unknown... */
885 c->x86_vendor_id[0] = '\0'; /* Unset */
886 c->x86_model_id[0] = '\0'; /* Unset */
94605eff 887 c->x86_max_cores = 1;
102bbe3a 888 c->x86_coreid_bits = 0;
11fdd252 889#ifdef CONFIG_X86_64
102bbe3a 890 c->x86_clflush_size = 64;
13c6c532
JB
891 c->x86_phys_bits = 36;
892 c->x86_virt_bits = 48;
102bbe3a
YL
893#else
894 c->cpuid_level = -1; /* CPUID not detected */
770d132f 895 c->x86_clflush_size = 32;
13c6c532
JB
896 c->x86_phys_bits = 32;
897 c->x86_virt_bits = 32;
102bbe3a
YL
898#endif
899 c->x86_cache_alignment = c->x86_clflush_size;
1da177e4
LT
900 memset(&c->x86_capability, 0, sizeof c->x86_capability);
901
1da177e4
LT
902 generic_identify(c);
903
3898534d 904 if (this_cpu->c_identify)
1da177e4
LT
905 this_cpu->c_identify(c);
906
2759c328
YL
907 /* Clear/Set all flags overriden by options, after probe */
908 for (i = 0; i < NCAPINTS; i++) {
909 c->x86_capability[i] &= ~cpu_caps_cleared[i];
910 c->x86_capability[i] |= cpu_caps_set[i];
911 }
912
102bbe3a 913#ifdef CONFIG_X86_64
cb8cc442 914 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
102bbe3a
YL
915#endif
916
1da177e4
LT
917 /*
918 * Vendor-specific initialization. In this section we
919 * canonicalize the feature flags, meaning if there are
920 * features a certain CPU supports which CPUID doesn't
921 * tell us, CPUID claiming incorrect flags, or other bugs,
922 * we handle them here.
923 *
924 * At the end of this section, c->x86_capability better
925 * indicate the features this CPU genuinely supports!
926 */
927 if (this_cpu->c_init)
928 this_cpu->c_init(c);
929
930 /* Disable the PN if appropriate */
931 squash_the_stupid_serial_number(c);
932
b2cc2a07
PA
933 /* Set up SMEP/SMAP */
934 setup_smep(c);
935 setup_smap(c);
936
1da177e4 937 /*
0f3fa48a
IM
938 * The vendor-specific functions might have changed features.
939 * Now we do "generic changes."
1da177e4
LT
940 */
941
b38b0665
PA
942 /* Filter out anything that depends on CPUID levels we don't have */
943 filter_cpuid_features(c, true);
944
1da177e4 945 /* If the model name is still unset, do table lookup. */
34048c9e 946 if (!c->x86_model_id[0]) {
02dde8b4 947 const char *p;
1da177e4 948 p = table_lookup_model(c);
34048c9e 949 if (p)
1da177e4
LT
950 strcpy(c->x86_model_id, p);
951 else
952 /* Last resort... */
953 sprintf(c->x86_model_id, "%02x/%02x",
54a20f8c 954 c->x86, c->x86_model);
1da177e4
LT
955 }
956
102bbe3a
YL
957#ifdef CONFIG_X86_64
958 detect_ht(c);
959#endif
960
88b094fb 961 init_hypervisor(c);
49d859d7 962 x86_init_rdrand(c);
cbc82b17 963 x86_init_cache_qos(c);
3e0c3737
YL
964
965 /*
966 * Clear/Set all flags overriden by options, need do it
967 * before following smp all cpus cap AND.
968 */
969 for (i = 0; i < NCAPINTS; i++) {
970 c->x86_capability[i] &= ~cpu_caps_cleared[i];
971 c->x86_capability[i] |= cpu_caps_set[i];
972 }
973
1da177e4
LT
974 /*
975 * On SMP, boot_cpu_data holds the common feature set between
976 * all CPUs; so make sure that we indicate which features are
977 * common between the CPUs. The first time this routine gets
978 * executed, c == &boot_cpu_data.
979 */
34048c9e 980 if (c != &boot_cpu_data) {
1da177e4 981 /* AND the already accumulated flags with these */
9d31d35b 982 for (i = 0; i < NCAPINTS; i++)
1da177e4 983 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
65fc985b
BP
984
985 /* OR, i.e. replicate the bug flags */
986 for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
987 c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1da177e4
LT
988 }
989
990 /* Init Machine Check Exception if available. */
5e09954a 991 mcheck_cpu_init(c);
30d432df
AK
992
993 select_idle_routine(c);
102bbe3a 994
de2d9445 995#ifdef CONFIG_NUMA
102bbe3a
YL
996 numa_add_cpu(smp_processor_id());
997#endif
a6c4e076 998}
31ab269a 999
e04d645f 1000#ifdef CONFIG_X86_64
e48510f4 1001#ifdef CONFIG_IA32_EMULATION
cfda7bb9
AL
1002/* May not be __init: called during resume */
1003static void syscall32_cpu_init(void)
1004{
1005 /* Load these always in case some future AMD CPU supports
1006 SYSENTER from compat mode too. */
1007 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1008 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1009 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target);
1010
1011 wrmsrl(MSR_CSTAR, ia32_cstar_target);
1012}
e48510f4
AM
1013#endif /* CONFIG_IA32_EMULATION */
1014#endif /* CONFIG_X86_64 */
cfda7bb9
AL
1015
1016#ifdef CONFIG_X86_32
1017void enable_sep_cpu(void)
1018{
1019 int cpu = get_cpu();
1020 struct tss_struct *tss = &per_cpu(init_tss, cpu);
1021
1022 if (!boot_cpu_has(X86_FEATURE_SEP)) {
1023 put_cpu();
1024 return;
1025 }
1026
1027 tss->x86_tss.ss1 = __KERNEL_CS;
1028 tss->x86_tss.sp1 = sizeof(struct tss_struct) + (unsigned long) tss;
1029 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
1030 wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.sp1, 0);
1031 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) ia32_sysenter_target, 0);
1032 put_cpu();
1033}
e04d645f
GC
1034#endif
1035
a6c4e076
JF
1036void __init identify_boot_cpu(void)
1037{
1038 identify_cpu(&boot_cpu_data);
02c68a02 1039 init_amd_e400_c1e_mask();
102bbe3a 1040#ifdef CONFIG_X86_32
a6c4e076 1041 sysenter_setup();
6fe940d6 1042 enable_sep_cpu();
102bbe3a 1043#endif
5b556332 1044 cpu_detect_tlb(&boot_cpu_data);
a6c4e076 1045}
3b520b23 1046
148f9bb8 1047void identify_secondary_cpu(struct cpuinfo_x86 *c)
a6c4e076
JF
1048{
1049 BUG_ON(c == &boot_cpu_data);
1050 identify_cpu(c);
102bbe3a 1051#ifdef CONFIG_X86_32
a6c4e076 1052 enable_sep_cpu();
102bbe3a 1053#endif
a6c4e076 1054 mtrr_ap_init();
1da177e4
LT
1055}
1056
a0854a46 1057struct msr_range {
0f3fa48a
IM
1058 unsigned min;
1059 unsigned max;
a0854a46 1060};
1da177e4 1061
148f9bb8 1062static const struct msr_range msr_range_array[] = {
a0854a46
YL
1063 { 0x00000000, 0x00000418},
1064 { 0xc0000000, 0xc000040b},
1065 { 0xc0010000, 0xc0010142},
1066 { 0xc0011000, 0xc001103b},
1067};
1da177e4 1068
148f9bb8 1069static void __print_cpu_msr(void)
a0854a46 1070{
0f3fa48a 1071 unsigned index_min, index_max;
a0854a46
YL
1072 unsigned index;
1073 u64 val;
1074 int i;
a0854a46
YL
1075
1076 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1077 index_min = msr_range_array[i].min;
1078 index_max = msr_range_array[i].max;
0f3fa48a 1079
a0854a46 1080 for (index = index_min; index < index_max; index++) {
ecd431d9 1081 if (rdmsrl_safe(index, &val))
a0854a46
YL
1082 continue;
1083 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1da177e4 1084 }
a0854a46
YL
1085 }
1086}
94605eff 1087
148f9bb8 1088static int show_msr;
0f3fa48a 1089
a0854a46
YL
1090static __init int setup_show_msr(char *arg)
1091{
1092 int num;
3dd9d514 1093
a0854a46 1094 get_option(&arg, &num);
3dd9d514 1095
a0854a46
YL
1096 if (num > 0)
1097 show_msr = num;
1098 return 1;
1da177e4 1099}
a0854a46 1100__setup("show_msr=", setup_show_msr);
1da177e4 1101
191679fd
AK
1102static __init int setup_noclflush(char *arg)
1103{
840d2830 1104 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
da4aaa7d 1105 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
191679fd
AK
1106 return 1;
1107}
1108__setup("noclflush", setup_noclflush);
1109
148f9bb8 1110void print_cpu_info(struct cpuinfo_x86 *c)
1da177e4 1111{
02dde8b4 1112 const char *vendor = NULL;
1da177e4 1113
0f3fa48a 1114 if (c->x86_vendor < X86_VENDOR_NUM) {
1da177e4 1115 vendor = this_cpu->c_vendor;
0f3fa48a
IM
1116 } else {
1117 if (c->cpuid_level >= 0)
1118 vendor = c->x86_vendor_id;
1119 }
1da177e4 1120
bd32a8cf 1121 if (vendor && !strstr(c->x86_model_id, vendor))
9d31d35b 1122 printk(KERN_CONT "%s ", vendor);
1da177e4 1123
9d31d35b 1124 if (c->x86_model_id[0])
924e101a 1125 printk(KERN_CONT "%s", strim(c->x86_model_id));
1da177e4 1126 else
9d31d35b 1127 printk(KERN_CONT "%d86", c->x86);
1da177e4 1128
924e101a
BP
1129 printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1130
34048c9e 1131 if (c->x86_mask || c->cpuid_level >= 0)
924e101a 1132 printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1da177e4 1133 else
924e101a 1134 printk(KERN_CONT ")\n");
a0854a46 1135
0b8b8078 1136 print_cpu_msr(c);
21c3fcf3
YL
1137}
1138
148f9bb8 1139void print_cpu_msr(struct cpuinfo_x86 *c)
21c3fcf3 1140{
a0854a46 1141 if (c->cpu_index < show_msr)
21c3fcf3 1142 __print_cpu_msr();
1da177e4
LT
1143}
1144
ac72e788
AK
1145static __init int setup_disablecpuid(char *arg)
1146{
1147 int bit;
0f3fa48a 1148
ac72e788
AK
1149 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1150 setup_clear_cpu_cap(bit);
1151 else
1152 return 0;
0f3fa48a 1153
ac72e788
AK
1154 return 1;
1155}
1156__setup("clearcpuid=", setup_disablecpuid);
1157
198d208d
SR
1158DEFINE_PER_CPU(unsigned long, kernel_stack) =
1159 (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1160EXPORT_PER_CPU_SYMBOL(kernel_stack);
1161
d5494d4f 1162#ifdef CONFIG_X86_64
9ff80942 1163struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
629f4f9d
SA
1164struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1165 (unsigned long) debug_idt_table };
d5494d4f 1166
947e76cd 1167DEFINE_PER_CPU_FIRST(union irq_stack_union,
277d5b40 1168 irq_stack_union) __aligned(PAGE_SIZE) __visible;
0f3fa48a 1169
bdf977b3
TH
1170/*
1171 * The following four percpu variables are hot. Align current_task to
1172 * cacheline size such that all four fall in the same cacheline.
1173 */
1174DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1175 &init_task;
1176EXPORT_PER_CPU_SYMBOL(current_task);
d5494d4f 1177
bdf977b3
TH
1178DEFINE_PER_CPU(char *, irq_stack_ptr) =
1179 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1180
277d5b40 1181DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
d5494d4f 1182
c2daa3be
PZ
1183DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1184EXPORT_PER_CPU_SYMBOL(__preempt_count);
1185
7e16838d
LT
1186DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1187
0f3fa48a
IM
1188/*
1189 * Special IST stacks which the CPU switches to when it calls
1190 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1191 * limit), all of them are 4K, except the debug stack which
1192 * is 8K.
1193 */
1194static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1195 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1196 [DEBUG_STACK - 1] = DEBUG_STKSZ
1197};
1198
92d65b23 1199static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
3e352aa8 1200 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
d5494d4f 1201
d5494d4f
YL
1202/* May not be marked __init: used by software suspend */
1203void syscall_init(void)
1da177e4 1204{
d5494d4f
YL
1205 /*
1206 * LSTAR and STAR live in a bit strange symbiosis.
1207 * They both write to the same internal register. STAR allows to
1208 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1209 */
1210 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
1211 wrmsrl(MSR_LSTAR, system_call);
1212 wrmsrl(MSR_CSTAR, ignore_sysret);
03ae5768 1213
d5494d4f
YL
1214#ifdef CONFIG_IA32_EMULATION
1215 syscall32_cpu_init();
1216#endif
03ae5768 1217
d5494d4f
YL
1218 /* Flags to clear on syscall */
1219 wrmsrl(MSR_SYSCALL_MASK,
63bcff2a 1220 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
8c7aa698 1221 X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1da177e4 1222}
62111195 1223
d5494d4f
YL
1224/*
1225 * Copies of the original ist values from the tss are only accessed during
1226 * debugging, no special alignment required.
1227 */
1228DEFINE_PER_CPU(struct orig_ist, orig_ist);
1229
228bdaa9 1230static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
42181186 1231DEFINE_PER_CPU(int, debug_stack_usage);
228bdaa9
SR
1232
1233int is_debug_stack(unsigned long addr)
1234{
89cbc767
CL
1235 return __this_cpu_read(debug_stack_usage) ||
1236 (addr <= __this_cpu_read(debug_stack_addr) &&
1237 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
228bdaa9 1238}
0f46efeb 1239NOKPROBE_SYMBOL(is_debug_stack);
228bdaa9 1240
629f4f9d 1241DEFINE_PER_CPU(u32, debug_idt_ctr);
f8988175 1242
228bdaa9
SR
1243void debug_stack_set_zero(void)
1244{
629f4f9d
SA
1245 this_cpu_inc(debug_idt_ctr);
1246 load_current_idt();
228bdaa9 1247}
0f46efeb 1248NOKPROBE_SYMBOL(debug_stack_set_zero);
228bdaa9
SR
1249
1250void debug_stack_reset(void)
1251{
629f4f9d 1252 if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
f8988175 1253 return;
629f4f9d
SA
1254 if (this_cpu_dec_return(debug_idt_ctr) == 0)
1255 load_current_idt();
228bdaa9 1256}
0f46efeb 1257NOKPROBE_SYMBOL(debug_stack_reset);
228bdaa9 1258
0f3fa48a 1259#else /* CONFIG_X86_64 */
d5494d4f 1260
bdf977b3
TH
1261DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1262EXPORT_PER_CPU_SYMBOL(current_task);
c2daa3be
PZ
1263DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1264EXPORT_PER_CPU_SYMBOL(__preempt_count);
27e74da9 1265DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
bdf977b3 1266
60a5317f 1267#ifdef CONFIG_CC_STACKPROTECTOR
53f82452 1268DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
60a5317f 1269#endif
d5494d4f 1270
0f3fa48a 1271#endif /* CONFIG_X86_64 */
c5413fbe 1272
9766cdbc
JSR
1273/*
1274 * Clear all 6 debug registers:
1275 */
1276static void clear_all_debug_regs(void)
1277{
1278 int i;
1279
1280 for (i = 0; i < 8; i++) {
1281 /* Ignore db4, db5 */
1282 if ((i == 4) || (i == 5))
1283 continue;
1284
1285 set_debugreg(0, i);
1286 }
1287}
c5413fbe 1288
0bb9fef9
JW
1289#ifdef CONFIG_KGDB
1290/*
1291 * Restore debug regs if using kgdbwait and you have a kernel debugger
1292 * connection established.
1293 */
1294static void dbg_restore_debug_regs(void)
1295{
1296 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1297 arch_kgdb_ops.correct_hw_break();
1298}
1299#else /* ! CONFIG_KGDB */
1300#define dbg_restore_debug_regs()
1301#endif /* ! CONFIG_KGDB */
1302
ce4b1b16
IM
1303static void wait_for_master_cpu(int cpu)
1304{
1305#ifdef CONFIG_SMP
1306 /*
1307 * wait for ACK from master CPU before continuing
1308 * with AP initialization
1309 */
1310 WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1311 while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1312 cpu_relax();
1313#endif
1314}
1315
d2cbcc49
RR
1316/*
1317 * cpu_init() initializes state that is per-CPU. Some data is already
1318 * initialized (naturally) in the bootstrap process, such as the GDT
1319 * and IDT. We reload them nevertheless, this function acts as a
1320 * 'CPU state barrier', nothing should get across.
1ba76586 1321 * A lot of state is already set up in PDA init for 64 bit
d2cbcc49 1322 */
1ba76586 1323#ifdef CONFIG_X86_64
0f3fa48a 1324
148f9bb8 1325void cpu_init(void)
1ba76586 1326{
0fe1e009 1327 struct orig_ist *oist;
1ba76586 1328 struct task_struct *me;
0f3fa48a
IM
1329 struct tss_struct *t;
1330 unsigned long v;
ce4b1b16 1331 int cpu = stack_smp_processor_id();
1ba76586
YL
1332 int i;
1333
ce4b1b16
IM
1334 wait_for_master_cpu(cpu);
1335
1e02ce4c
AL
1336 /*
1337 * Initialize the CR4 shadow before doing anything that could
1338 * try to read it.
1339 */
1340 cr4_init_shadow();
1341
e6ebf5de
FY
1342 /*
1343 * Load microcode on this cpu if a valid microcode is available.
1344 * This is early microcode loading procedure.
1345 */
1346 load_ucode_ap();
1347
0f3fa48a 1348 t = &per_cpu(init_tss, cpu);
0fe1e009 1349 oist = &per_cpu(orig_ist, cpu);
0f3fa48a 1350
e7a22c1e 1351#ifdef CONFIG_NUMA
27fd185f 1352 if (this_cpu_read(numa_node) == 0 &&
e534c7c5
LS
1353 early_cpu_to_node(cpu) != NUMA_NO_NODE)
1354 set_numa_node(early_cpu_to_node(cpu));
e7a22c1e 1355#endif
1ba76586
YL
1356
1357 me = current;
1358
2eaad1fd 1359 pr_debug("Initializing CPU#%d\n", cpu);
1ba76586 1360
375074cc 1361 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1ba76586
YL
1362
1363 /*
1364 * Initialize the per-CPU GDT with the boot GDT,
1365 * and set up the GDT descriptor:
1366 */
1367
552be871 1368 switch_to_new_gdt(cpu);
2697fbd5
BG
1369 loadsegment(fs, 0);
1370
cf910e83 1371 load_current_idt();
1ba76586
YL
1372
1373 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1374 syscall_init();
1375
1376 wrmsrl(MSR_FS_BASE, 0);
1377 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1378 barrier();
1379
4763ed4d 1380 x86_configure_nx();
27fd185f 1381 enable_x2apic();
1ba76586
YL
1382
1383 /*
1384 * set up and load the per-CPU TSS
1385 */
0fe1e009 1386 if (!oist->ist[0]) {
92d65b23 1387 char *estacks = per_cpu(exception_stacks, cpu);
0f3fa48a 1388
1ba76586 1389 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
0f3fa48a 1390 estacks += exception_stack_sizes[v];
0fe1e009 1391 oist->ist[v] = t->x86_tss.ist[v] =
1ba76586 1392 (unsigned long)estacks;
228bdaa9
SR
1393 if (v == DEBUG_STACK-1)
1394 per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1ba76586
YL
1395 }
1396 }
1397
1398 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
0f3fa48a 1399
1ba76586
YL
1400 /*
1401 * <= is required because the CPU will access up to
1402 * 8 bits beyond the end of the IO permission bitmap.
1403 */
1404 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1405 t->io_bitmap[i] = ~0UL;
1406
1407 atomic_inc(&init_mm.mm_count);
1408 me->active_mm = &init_mm;
8c5dfd25 1409 BUG_ON(me->mm);
1ba76586
YL
1410 enter_lazy_tlb(&init_mm, me);
1411
1412 load_sp0(t, &current->thread);
1413 set_tss_desc(cpu, t);
1414 load_TR_desc();
1415 load_LDT(&init_mm.context);
1416
0bb9fef9
JW
1417 clear_all_debug_regs();
1418 dbg_restore_debug_regs();
1ba76586
YL
1419
1420 fpu_init();
1421
1ba76586
YL
1422 if (is_uv_system())
1423 uv_cpu_init();
1424}
1425
1426#else
1427
148f9bb8 1428void cpu_init(void)
9ee79a3d 1429{
d2cbcc49
RR
1430 int cpu = smp_processor_id();
1431 struct task_struct *curr = current;
34048c9e 1432 struct tss_struct *t = &per_cpu(init_tss, cpu);
9ee79a3d 1433 struct thread_struct *thread = &curr->thread;
62111195 1434
ce4b1b16 1435 wait_for_master_cpu(cpu);
e6ebf5de 1436
ce4b1b16 1437 show_ucode_info_early();
62111195
JF
1438
1439 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1440
9298b815 1441 if (cpu_feature_enabled(X86_FEATURE_VME) || cpu_has_tsc || cpu_has_de)
375074cc 1442 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
62111195 1443
cf910e83 1444 load_current_idt();
552be871 1445 switch_to_new_gdt(cpu);
1da177e4 1446
1da177e4
LT
1447 /*
1448 * Set up and load the per-CPU TSS and LDT
1449 */
1450 atomic_inc(&init_mm.mm_count);
62111195 1451 curr->active_mm = &init_mm;
8c5dfd25 1452 BUG_ON(curr->mm);
62111195 1453 enter_lazy_tlb(&init_mm, curr);
1da177e4 1454
faca6227 1455 load_sp0(t, thread);
34048c9e 1456 set_tss_desc(cpu, t);
1da177e4
LT
1457 load_TR_desc();
1458 load_LDT(&init_mm.context);
1459
f9a196b8
TG
1460 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1461
22c4e308 1462#ifdef CONFIG_DOUBLEFAULT
1da177e4
LT
1463 /* Set up doublefault TSS pointer in the GDT */
1464 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
22c4e308 1465#endif
1da177e4 1466
9766cdbc 1467 clear_all_debug_regs();
0bb9fef9 1468 dbg_restore_debug_regs();
1da177e4 1469
0e49bf66 1470 fpu_init();
1da177e4 1471}
1ba76586 1472#endif
5700f743
BP
1473
1474#ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1475void warn_pre_alternatives(void)
1476{
1477 WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1478}
1479EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1480#endif
4a90a99c
BP
1481
1482inline bool __static_cpu_has_safe(u16 bit)
1483{
1484 return boot_cpu_has(bit);
1485}
1486EXPORT_SYMBOL_GPL(__static_cpu_has_safe);