common.c 35.7 KB
Newer Older
1
#include <linux/bootmem.h>
2
#include <linux/linkage.h>
3
#include <linux/bitops.h>
4
#include <linux/kernel.h>
5
#include <linux/module.h>
6 7
#include <linux/percpu.h>
#include <linux/string.h>
8
#include <linux/ctype.h>
L
Linus Torvalds 已提交
9
#include <linux/delay.h>
10 11
#include <linux/sched.h>
#include <linux/init.h>
12
#include <linux/kprobes.h>
13
#include <linux/kgdb.h>
L
Linus Torvalds 已提交
14
#include <linux/smp.h>
15
#include <linux/io.h>
16
#include <linux/syscore_ops.h>
17 18

#include <asm/stackprotector.h>
19
#include <asm/perf_event.h>
L
Linus Torvalds 已提交
20
#include <asm/mmu_context.h>
21
#include <asm/archrandom.h>
22 23
#include <asm/hypervisor.h>
#include <asm/processor.h>
24
#include <asm/tlbflush.h>
25
#include <asm/debugreg.h>
26
#include <asm/sections.h>
27
#include <asm/vsyscall.h>
A
Alan Cox 已提交
28 29
#include <linux/topology.h>
#include <linux/cpumask.h>
30
#include <asm/pgtable.h>
A
Arun Sharma 已提交
31
#include <linux/atomic.h>
32 33 34 35
#include <asm/proto.h>
#include <asm/setup.h>
#include <asm/apic.h>
#include <asm/desc.h>
36
#include <asm/fpu/internal.h>
37
#include <asm/mtrr.h>
A
Alan Cox 已提交
38
#include <linux/numa.h>
39 40
#include <asm/asm.h>
#include <asm/cpu.h>
41
#include <asm/mce.h>
42
#include <asm/msr.h>
43
#include <asm/pat.h>
44 45
#include <asm/microcode.h>
#include <asm/microcode_intel.h>
46 47

#ifdef CONFIG_X86_LOCAL_APIC
T
Tejun Heo 已提交
48
#include <asm/uv/uv.h>
L
Linus Torvalds 已提交
49 50 51 52
#endif

#include "cpu.h"

53 54
/* all of these masks are initialized in setup_cpu_local_masks() */
cpumask_var_t cpu_initialized_mask;
55 56
cpumask_var_t cpu_callout_mask;
cpumask_var_t cpu_callin_mask;
57 58 59 60

/* representing cpus for which sibling maps can be computed */
cpumask_var_t cpu_sibling_setup_mask;

B
Brian Gerst 已提交
61
/* correctly size the local cpu masks */
62
void __init setup_cpu_local_masks(void)
B
Brian Gerst 已提交
63 64 65 66 67 68 69
{
	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
	alloc_bootmem_cpumask_var(&cpu_callin_mask);
	alloc_bootmem_cpumask_var(&cpu_callout_mask);
	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
}

70
static void default_init(struct cpuinfo_x86 *c)
71 72
{
#ifdef CONFIG_X86_64
73
	cpu_detect_cache_sizes(c);
74 75 76 77 78 79 80 81 82 83 84 85 86
#else
	/* Not much we can do here... */
	/* Check if at least it has cpuid */
	if (c->cpuid_level == -1) {
		/* No cpuid. It must be an ancient CPU */
		if (c->x86 == 4)
			strcpy(c->x86_model_id, "486");
		else if (c->x86 == 3)
			strcpy(c->x86_model_id, "386");
	}
#endif
}

87
static const struct cpu_dev default_cpu = {
88 89 90 91 92
	.c_init		= default_init,
	.c_vendor	= "Unknown",
	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
};

93
static const struct cpu_dev *this_cpu = &default_cpu;
94

B
Brian Gerst 已提交
95
DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
Y
Yinghai Lu 已提交
96
#ifdef CONFIG_X86_64
B
Brian Gerst 已提交
97 98 99 100 101
	/*
	 * We need valid kernel segments for data and code in long mode too
	 * IRET will check the segment types  kkeil 2000/10/28
	 * Also sysret mandates a special GDT layout
	 *
102
	 * TLS descriptors are currently at a different place compared to i386.
B
Brian Gerst 已提交
103 104
	 * Hopefully nobody expects them at a fixed place (Wine?)
	 */
A
Akinobu Mita 已提交
105 106 107 108 109 110
	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
Y
Yinghai Lu 已提交
111
#else
A
Akinobu Mita 已提交
112 113 114 115
	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
116 117 118 119 120
	/*
	 * Segments used for calling PnP BIOS have byte granularity.
	 * They code segments and data segments have fixed 64k limits,
	 * the transfer segment sizes are set at run time.
	 */
121
	/* 32-bit code */
A
Akinobu Mita 已提交
122
	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
123
	/* 16-bit code */
A
Akinobu Mita 已提交
124
	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
125
	/* 16-bit data */
A
Akinobu Mita 已提交
126
	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
127
	/* 16-bit data */
A
Akinobu Mita 已提交
128
	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
129
	/* 16-bit data */
A
Akinobu Mita 已提交
130
	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
131 132 133 134
	/*
	 * The APM segments have byte granularity and their bases
	 * are set at run time.  All have 64k limits.
	 */
135
	/* 32-bit code */
A
Akinobu Mita 已提交
136
	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
137
	/* 16-bit code */
A
Akinobu Mita 已提交
138
	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
139
	/* data */
140
	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
141

A
Akinobu Mita 已提交
142 143
	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
144
	GDT_STACK_CANARY_INIT
Y
Yinghai Lu 已提交
145
#endif
B
Brian Gerst 已提交
146
} };
147
EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
148

149
static int __init x86_mpx_setup(char *s)
150
{
151
	/* require an exact match without trailing characters */
152 153
	if (strlen(s))
		return 0;
154

155 156 157
	/* do not emit a message if the feature is not present */
	if (!boot_cpu_has(X86_FEATURE_MPX))
		return 1;
158

159 160
	setup_clear_cpu_cap(X86_FEATURE_MPX);
	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
161 162
	return 1;
}
163
__setup("nompx", x86_mpx_setup);
164

165
#ifdef CONFIG_X86_32
166 167
static int cachesize_override = -1;
static int disable_x86_serial_nr = 1;
L
Linus Torvalds 已提交
168

169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187
static int __init cachesize_setup(char *str)
{
	get_option(&str, &cachesize_override);
	return 1;
}
__setup("cachesize=", cachesize_setup);

static int __init x86_sep_setup(char *s)
{
	setup_clear_cpu_cap(X86_FEATURE_SEP);
	return 1;
}
__setup("nosep", x86_sep_setup);

/* Standard macro to see if a specific flag is changeable */
static inline int flag_is_changeable_p(u32 flag)
{
	u32 f1, f2;

188 189 190 191 192 193 194
	/*
	 * Cyrix and IDT cpus allow disabling of CPUID
	 * so the code below may return different results
	 * when it is executed before and after enabling
	 * the CPUID. Add "volatile" to not allow gcc to
	 * optimize the subsequent calls to this function.
	 */
I
Ingo Molnar 已提交
195 196 197 198 199 200 201 202 203 204 205
	asm volatile ("pushfl		\n\t"
		      "pushfl		\n\t"
		      "popl %0		\n\t"
		      "movl %0, %1	\n\t"
		      "xorl %2, %0	\n\t"
		      "pushl %0		\n\t"
		      "popfl		\n\t"
		      "pushfl		\n\t"
		      "popl %0		\n\t"
		      "popfl		\n\t"

206 207
		      : "=&r" (f1), "=&r" (f2)
		      : "ir" (flag));
208 209 210 211 212

	return ((f1^f2) & flag) != 0;
}

/* Probe for the CPUID instruction */
213
int have_cpuid_p(void)
214 215 216 217
{
	return flag_is_changeable_p(X86_EFLAGS_ID);
}

218
static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
219
{
I
Ingo Molnar 已提交
220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235
	unsigned long lo, hi;

	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
		return;

	/* Disable processor serial number: */

	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
	lo |= 0x200000;
	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);

	printk(KERN_NOTICE "CPU serial number disabled.\n");
	clear_cpu_cap(c, X86_FEATURE_PN);

	/* Disabling the serial number may affect the cpuid level */
	c->cpuid_level = cpuid_eax(0);
236 237 238 239 240 241 242 243
}

static int __init x86_serial_nr_setup(char *s)
{
	disable_x86_serial_nr = 0;
	return 1;
}
__setup("serialnumber", x86_serial_nr_setup);
244
#else
245 246 247 248 249 250 251
static inline int flag_is_changeable_p(u32 flag)
{
	return 1;
}
static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
{
}
252
#endif
253

254 255
static __init int setup_disable_smep(char *arg)
{
256
	setup_clear_cpu_cap(X86_FEATURE_SMEP);
257 258 259 260
	return 1;
}
__setup("nosmep", setup_disable_smep);

261
static __always_inline void setup_smep(struct cpuinfo_x86 *c)
262
{
263
	if (cpu_has(c, X86_FEATURE_SMEP))
A
Andy Lutomirski 已提交
264
		cr4_set_bits(X86_CR4_SMEP);
265 266
}

267 268
static __init int setup_disable_smap(char *arg)
{
269
	setup_clear_cpu_cap(X86_FEATURE_SMAP);
270 271 272 273
	return 1;
}
__setup("nosmap", setup_disable_smap);

274 275
static __always_inline void setup_smap(struct cpuinfo_x86 *c)
{
276
	unsigned long eflags = native_save_fl();
277 278 279 280

	/* This should have been cleared long ago */
	BUG_ON(eflags & X86_EFLAGS_AC);

281 282
	if (cpu_has(c, X86_FEATURE_SMAP)) {
#ifdef CONFIG_X86_SMAP
A
Andy Lutomirski 已提交
283
		cr4_set_bits(X86_CR4_SMAP);
284
#else
A
Andy Lutomirski 已提交
285
		cr4_clear_bits(X86_CR4_SMAP);
286 287
#endif
	}
288 289
}

290 291 292 293 294 295 296 297 298
/*
 * Some CPU features depend on higher CPUID levels, which may not always
 * be available due to CPUID level capping or broken virtualization
 * software.  Add those features to this table to auto-disable them.
 */
struct cpuid_dependent_feature {
	u32 feature;
	u32 level;
};
I
Ingo Molnar 已提交
299

300
static const struct cpuid_dependent_feature
301 302 303 304 305 306 307
cpuid_dependent_features[] = {
	{ X86_FEATURE_MWAIT,		0x00000005 },
	{ X86_FEATURE_DCA,		0x00000009 },
	{ X86_FEATURE_XSAVE,		0x0000000d },
	{ 0, 0 }
};

308
static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
309 310
{
	const struct cpuid_dependent_feature *df;
311

312
	for (df = cpuid_dependent_features; df->feature; df++) {
I
Ingo Molnar 已提交
313 314 315

		if (!cpu_has(c, df->feature))
			continue;
316 317 318 319 320 321 322
		/*
		 * Note: cpuid_level is set to -1 if unavailable, but
		 * extended_extended_level is set to 0 if unavailable
		 * and the legitimate extended levels are all negative
		 * when signed; hence the weird messing around with
		 * signs here...
		 */
I
Ingo Molnar 已提交
323
		if (!((s32)df->level < 0 ?
324
		     (u32)df->level > (u32)c->extended_cpuid_level :
I
Ingo Molnar 已提交
325 326 327 328 329 330 331 332
		     (s32)df->level > (s32)c->cpuid_level))
			continue;

		clear_cpu_cap(c, df->feature);
		if (!warn)
			continue;

		printk(KERN_WARNING
333 334
		       "CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
				x86_cap_flag(df->feature), df->level);
335
	}
336
}
337

338 339 340
/*
 * Naming convention should be: <Name> [(<Codename>)]
 * This table only is used unless init_<vendor>() below doesn't set it;
I
Ingo Molnar 已提交
341 342
 * in particular, if CPUID levels 0x80000002..4 are supported, this
 * isn't used
343 344 345
 */

/* Look up CPU names by table lookup. */
346
static const char *table_lookup_model(struct cpuinfo_x86 *c)
347
{
348 349
#ifdef CONFIG_X86_32
	const struct legacy_cpu_model_info *info;
350 351 352 353 354 355 356

	if (c->x86_model >= 16)
		return NULL;	/* Range check */

	if (!this_cpu)
		return NULL;

357
	info = this_cpu->legacy_models;
358

359
	while (info->family) {
360 361 362 363
		if (info->family == c->x86)
			return info->model_names[c->x86_model];
		info++;
	}
364
#endif
365 366 367
	return NULL;		/* Not found */
}

368 369
__u32 cpu_caps_cleared[NCAPINTS];
__u32 cpu_caps_set[NCAPINTS];
370

371 372 373 374 375 376 377 378
void load_percpu_segment(int cpu)
{
#ifdef CONFIG_X86_32
	loadsegment(fs, __KERNEL_PERCPU);
#else
	loadsegment(gs, 0);
	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
#endif
379
	load_stack_canary_segment();
380 381
}

I
Ingo Molnar 已提交
382 383 384 385
/*
 * Current gdt points %fs at the "master" per-cpu area: after this,
 * it's on the real one.
 */
386
void switch_to_new_gdt(int cpu)
387 388 389
{
	struct desc_ptr gdt_descr;

390
	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
391 392
	gdt_descr.size = GDT_SIZE - 1;
	load_gdt(&gdt_descr);
393
	/* Reload the per-cpu base */
394 395

	load_percpu_segment(cpu);
396 397
}

398
static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
L
Linus Torvalds 已提交
399

400
static void get_model_name(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
401 402
{
	unsigned int *v;
403
	char *p, *q, *s;
L
Linus Torvalds 已提交
404

405
	if (c->extended_cpuid_level < 0x80000004)
406
		return;
L
Linus Torvalds 已提交
407

I
Ingo Molnar 已提交
408
	v = (unsigned int *)c->x86_model_id;
L
Linus Torvalds 已提交
409 410 411 412 413
	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
	c->x86_model_id[48] = 0;

414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
	/* Trim whitespace */
	p = q = s = &c->x86_model_id[0];

	while (*p == ' ')
		p++;

	while (*p) {
		/* Note the last non-whitespace index */
		if (!isspace(*p))
			s = q;

		*q++ = *p++;
	}

	*(s + 1) = '\0';
L
Linus Torvalds 已提交
429 430
}

431
void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
432
{
433
	unsigned int n, dummy, ebx, ecx, edx, l2size;
L
Linus Torvalds 已提交
434

435
	n = c->extended_cpuid_level;
L
Linus Torvalds 已提交
436 437

	if (n >= 0x80000005) {
438 439
		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
		c->x86_cache_size = (ecx>>24) + (edx>>24);
440 441 442 443
#ifdef CONFIG_X86_64
		/* On K8 L1 TLB is inclusive, so don't count it */
		c->x86_tlbsize = 0;
#endif
L
Linus Torvalds 已提交
444 445 446 447 448
	}

	if (n < 0x80000006)	/* Some chips just has a large L1. */
		return;

449
	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
L
Linus Torvalds 已提交
450
	l2size = ecx >> 16;
451

452 453 454
#ifdef CONFIG_X86_64
	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
#else
L
Linus Torvalds 已提交
455
	/* do processor-specific cache resizing */
456 457
	if (this_cpu->legacy_cache_size)
		l2size = this_cpu->legacy_cache_size(c, l2size);
L
Linus Torvalds 已提交
458 459 460 461 462

	/* Allow user to override all this if necessary. */
	if (cachesize_override != -1)
		l2size = cachesize_override;

463
	if (l2size == 0)
L
Linus Torvalds 已提交
464
		return;		/* Again, no L2 cache is possible */
465
#endif
L
Linus Torvalds 已提交
466 467 468 469

	c->x86_cache_size = l2size;
}

470 471 472 473 474 475
u16 __read_mostly tlb_lli_4k[NR_INFO];
u16 __read_mostly tlb_lli_2m[NR_INFO];
u16 __read_mostly tlb_lli_4m[NR_INFO];
u16 __read_mostly tlb_lld_4k[NR_INFO];
u16 __read_mostly tlb_lld_2m[NR_INFO];
u16 __read_mostly tlb_lld_4m[NR_INFO];
476
u16 __read_mostly tlb_lld_1g[NR_INFO];
477

478
static void cpu_detect_tlb(struct cpuinfo_x86 *c)
479 480 481 482
{
	if (this_cpu->c_detect_tlb)
		this_cpu->c_detect_tlb(c);

483
	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
484
		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
485 486 487 488 489
		tlb_lli_4m[ENTRIES]);

	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
490 491
}

492
void detect_ht(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
493
{
B
Borislav Petkov 已提交
494
#ifdef CONFIG_SMP
495 496
	u32 eax, ebx, ecx, edx;
	int index_msb, core_bits;
497
	static bool printed;
L
Linus Torvalds 已提交
498

499
	if (!cpu_has(c, X86_FEATURE_HT))
500
		return;
L
Linus Torvalds 已提交
501

502 503
	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
		goto out;
L
Linus Torvalds 已提交
504

505 506
	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
		return;
L
Linus Torvalds 已提交
507

508
	cpuid(1, &eax, &ebx, &ecx, &edx);
L
Linus Torvalds 已提交
509

510 511 512
	smp_num_siblings = (ebx & 0xff0000) >> 16;

	if (smp_num_siblings == 1) {
513
		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
I
Ingo Molnar 已提交
514 515
		goto out;
	}
516

I
Ingo Molnar 已提交
517 518
	if (smp_num_siblings <= 1)
		goto out;
519

I
Ingo Molnar 已提交
520 521
	index_msb = get_count_order(smp_num_siblings);
	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
522

I
Ingo Molnar 已提交
523
	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
524

I
Ingo Molnar 已提交
525
	index_msb = get_count_order(smp_num_siblings);
526

I
Ingo Molnar 已提交
527
	core_bits = get_count_order(c->x86_max_cores);
528

I
Ingo Molnar 已提交
529 530
	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
				       ((1 << core_bits) - 1);
L
Linus Torvalds 已提交
531

532
out:
533
	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
534 535 536 537
		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
		       c->phys_proc_id);
		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
		       c->cpu_core_id);
538
		printed = 1;
539 540
	}
#endif
541
}
L
Linus Torvalds 已提交
542

543
static void get_cpu_vendor(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
544 545
{
	char *v = c->x86_vendor_id;
I
Ingo Molnar 已提交
546
	int i;
L
Linus Torvalds 已提交
547 548

	for (i = 0; i < X86_VENDOR_NUM; i++) {
Y
Yinghai Lu 已提交
549 550 551 552 553 554
		if (!cpu_devs[i])
			break;

		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
		    (cpu_devs[i]->c_ident[1] &&
		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
I
Ingo Molnar 已提交
555

Y
Yinghai Lu 已提交
556 557 558
			this_cpu = cpu_devs[i];
			c->x86_vendor = this_cpu->c_x86_vendor;
			return;
L
Linus Torvalds 已提交
559 560
		}
	}
Y
Yinghai Lu 已提交
561

562 563 564
	printk_once(KERN_ERR
			"CPU: vendor_id '%s' unknown, using generic init.\n" \
			"CPU: Your system may be unstable.\n", v);
Y
Yinghai Lu 已提交
565

566 567
	c->x86_vendor = X86_VENDOR_UNKNOWN;
	this_cpu = &default_cpu;
L
Linus Torvalds 已提交
568 569
}

570
void cpu_detect(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
571 572
{
	/* Get vendor name */
573 574 575 576
	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
	      (unsigned int *)&c->x86_vendor_id[0],
	      (unsigned int *)&c->x86_vendor_id[8],
	      (unsigned int *)&c->x86_vendor_id[4]);
L
Linus Torvalds 已提交
577 578

	c->x86 = 4;
579
	/* Intel-defined flags: level 0x00000001 */
L
Linus Torvalds 已提交
580 581
	if (c->cpuid_level >= 0x00000001) {
		u32 junk, tfms, cap0, misc;
I
Ingo Molnar 已提交
582

L
Linus Torvalds 已提交
583
		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
584 585 586
		c->x86 = (tfms >> 8) & 0xf;
		c->x86_model = (tfms >> 4) & 0xf;
		c->x86_mask = tfms & 0xf;
I
Ingo Molnar 已提交
587

588
		if (c->x86 == 0xf)
L
Linus Torvalds 已提交
589
			c->x86 += (tfms >> 20) & 0xff;
590
		if (c->x86 >= 0x6)
591
			c->x86_model += ((tfms >> 16) & 0xf) << 4;
I
Ingo Molnar 已提交
592

H
Huang, Ying 已提交
593 594
		if (cap0 & (1<<19)) {
			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
595
			c->x86_cache_alignment = c->x86_clflush_size;
H
Huang, Ying 已提交
596
		}
L
Linus Torvalds 已提交
597 598
	}
}
599

600
void get_cpu_cap(struct cpuinfo_x86 *c)
601 602
{
	u32 tfms, xlvl;
603
	u32 ebx;
604

605 606 607
	/* Intel-defined flags: level 0x00000001 */
	if (c->cpuid_level >= 0x00000001) {
		u32 capability, excap;
I
Ingo Molnar 已提交
608

609 610 611 612
		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
		c->x86_capability[0] = capability;
		c->x86_capability[4] = excap;
	}
613

614 615 616 617 618 619
	/* Additional Intel-defined flags: level 0x00000007 */
	if (c->cpuid_level >= 0x00000007) {
		u32 eax, ebx, ecx, edx;

		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);

620
		c->x86_capability[9] = ebx;
621 622
	}

623 624 625 626 627 628 629 630 631
	/* Extended state features: level 0x0000000d */
	if (c->cpuid_level >= 0x0000000d) {
		u32 eax, ebx, ecx, edx;

		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);

		c->x86_capability[10] = eax;
	}

632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655
	/* Additional Intel-defined flags: level 0x0000000F */
	if (c->cpuid_level >= 0x0000000F) {
		u32 eax, ebx, ecx, edx;

		/* QoS sub-leaf, EAX=0Fh, ECX=0 */
		cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
		c->x86_capability[11] = edx;
		if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
			/* will be overridden if occupancy monitoring exists */
			c->x86_cache_max_rmid = ebx;

			/* QoS sub-leaf, EAX=0Fh, ECX=1 */
			cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
			c->x86_capability[12] = edx;
			if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) {
				c->x86_cache_max_rmid = ecx;
				c->x86_cache_occ_scale = ebx;
			}
		} else {
			c->x86_cache_max_rmid = -1;
			c->x86_cache_occ_scale = -1;
		}
	}

656 657 658
	/* AMD-defined flags: level 0x80000001 */
	xlvl = cpuid_eax(0x80000000);
	c->extended_cpuid_level = xlvl;
I
Ingo Molnar 已提交
659

660 661 662 663
	if ((xlvl & 0xffff0000) == 0x80000000) {
		if (xlvl >= 0x80000001) {
			c->x86_capability[1] = cpuid_edx(0x80000001);
			c->x86_capability[6] = cpuid_ecx(0x80000001);
664 665 666
		}
	}

667 668 669 670 671
	if (c->extended_cpuid_level >= 0x80000008) {
		u32 eax = cpuid_eax(0x80000008);

		c->x86_virt_bits = (eax >> 8) & 0xff;
		c->x86_phys_bits = eax & 0xff;
W
Wan Zongshun 已提交
672
		c->x86_capability[13] = cpuid_ebx(0x80000008);
673
	}
674 675 676
#ifdef CONFIG_X86_32
	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
		c->x86_phys_bits = 36;
677
#endif
678 679 680

	if (c->extended_cpuid_level >= 0x80000007)
		c->x86_power = cpuid_edx(0x80000007);
681

682
	init_scattered_cpuid_features(c);
683
}
L
Linus Torvalds 已提交
684

685
static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
Y
Yinghai Lu 已提交
686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710
{
#ifdef CONFIG_X86_32
	int i;

	/*
	 * First of all, decide if this is a 486 or higher
	 * It's a 486 if we can modify the AC flag
	 */
	if (flag_is_changeable_p(X86_EFLAGS_AC))
		c->x86 = 4;
	else
		c->x86 = 3;

	for (i = 0; i < X86_VENDOR_NUM; i++)
		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
			c->x86_vendor_id[0] = 0;
			cpu_devs[i]->c_identify(c);
			if (c->x86_vendor_id[0]) {
				get_cpu_vendor(c);
				break;
			}
		}
#endif
}

711 712 713 714 715 716 717 718 719
/*
 * Do minimum CPU detection early.
 * Fields really needed: vendor, cpuid_level, family, model, mask,
 * cache alignment.
 * The others are not touched to avoid unwanted side effects.
 *
 * WARNING: this function is only called on the BP.  Don't add code here
 * that is supposed to run on all CPUs.
 */
720
static void __init early_identify_cpu(struct cpuinfo_x86 *c)
721
{
722 723
#ifdef CONFIG_X86_64
	c->x86_clflush_size = 64;
724 725
	c->x86_phys_bits = 36;
	c->x86_virt_bits = 48;
726
#else
H
Huang, Ying 已提交
727
	c->x86_clflush_size = 32;
728 729
	c->x86_phys_bits = 32;
	c->x86_virt_bits = 32;
730
#endif
731
	c->x86_cache_alignment = c->x86_clflush_size;
732

733
	memset(&c->x86_capability, 0, sizeof c->x86_capability);
734
	c->extended_cpuid_level = 0;
735

Y
Yinghai Lu 已提交
736 737 738 739
	if (!have_cpuid_p())
		identify_cpu_without_cpuid(c);

	/* cyrix could have cpuid enabled via c_identify()*/
740 741 742 743
	if (!have_cpuid_p())
		return;

	cpu_detect(c);
744 745
	get_cpu_vendor(c);
	get_cpu_cap(c);
746

Y
Yinghai Lu 已提交
747 748
	if (this_cpu->c_early_init)
		this_cpu->c_early_init(c);
749

750
	c->cpu_index = 0;
751
	filter_cpuid_features(c, false);
752

B
Borislav Petkov 已提交
753 754
	if (this_cpu->c_bsp_init)
		this_cpu->c_bsp_init(c);
755 756

	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
757
	fpu__init_system(c);
758 759
}

760 761
void __init early_cpu_init(void)
{
762
	const struct cpu_dev *const *cdev;
Y
Yinghai Lu 已提交
763 764
	int count = 0;

765
#ifdef CONFIG_PROCESSOR_SELECT
766
	printk(KERN_INFO "KERNEL supported cpus:\n");
767 768
#endif

Y
Yinghai Lu 已提交
769
	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
770
		const struct cpu_dev *cpudev = *cdev;
771

Y
Yinghai Lu 已提交
772 773 774 775 776
		if (count >= X86_VENDOR_NUM)
			break;
		cpu_devs[count] = cpudev;
		count++;

777
#ifdef CONFIG_PROCESSOR_SELECT
778 779 780 781 782 783 784 785 786
		{
			unsigned int j;

			for (j = 0; j < 2; j++) {
				if (!cpudev->c_ident[j])
					continue;
				printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
					cpudev->c_ident[j]);
			}
Y
Yinghai Lu 已提交
787
		}
788
#endif
Y
Yinghai Lu 已提交
789
	}
790
	early_identify_cpu(&boot_cpu_data);
791
}
792

793
/*
B
Borislav Petkov 已提交
794 795 796 797 798
 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
 * unfortunately, that's not true in practice because of early VIA
 * chips and (more importantly) broken virtualizers that are not easy
 * to detect. In the latter case it doesn't even *fail* reliably, so
 * probing for it doesn't even work. Disable it completely on 32-bit
799
 * unless we can find a reliable way to detect all the broken cases.
B
Borislav Petkov 已提交
800
 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
801
 */
802
static void detect_nopl(struct cpuinfo_x86 *c)
803
{
B
Borislav Petkov 已提交
804
#ifdef CONFIG_X86_32
805
	clear_cpu_cap(c, X86_FEATURE_NOPL);
B
Borislav Petkov 已提交
806 807 808
#else
	set_cpu_cap(c, X86_FEATURE_NOPL);
#endif
809 810
}

811
static void generic_identify(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
812
{
Y
Yinghai Lu 已提交
813
	c->extended_cpuid_level = 0;
L
Linus Torvalds 已提交
814

815
	if (!have_cpuid_p())
Y
Yinghai Lu 已提交
816
		identify_cpu_without_cpuid(c);
817

Y
Yinghai Lu 已提交
818
	/* cyrix could have cpuid enabled via c_identify()*/
I
Ingo Molnar 已提交
819
	if (!have_cpuid_p())
Y
Yinghai Lu 已提交
820
		return;
L
Linus Torvalds 已提交
821

822
	cpu_detect(c);
L
Linus Torvalds 已提交
823

824
	get_cpu_vendor(c);
L
Linus Torvalds 已提交
825

826
	get_cpu_cap(c);
L
Linus Torvalds 已提交
827

828 829
	if (c->cpuid_level >= 0x00000001) {
		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
830
#ifdef CONFIG_X86_32
B
Borislav Petkov 已提交
831
# ifdef CONFIG_SMP
832
		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
833
# else
834
		c->apicid = c->initial_apicid;
835 836 837
# endif
#endif
		c->phys_proc_id = c->initial_apicid;
838
	}
L
Linus Torvalds 已提交
839

840
	get_model_name(c); /* Default name */
L
Linus Torvalds 已提交
841

842
	detect_nopl(c);
L
Linus Torvalds 已提交
843 844
}

845 846 847 848 849 850 851 852 853 854 855 856 857 858
static void x86_init_cache_qos(struct cpuinfo_x86 *c)
{
	/*
	 * The heavy lifting of max_rmid and cache_occ_scale are handled
	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
	 * in case CQM bits really aren't there in this CPU.
	 */
	if (c != &boot_cpu_data) {
		boot_cpu_data.x86_cache_max_rmid =
			min(boot_cpu_data.x86_cache_max_rmid,
			    c->x86_cache_max_rmid);
	}
}

L
Linus Torvalds 已提交
859 860 861
/*
 * This does the hard work of actually picking apart the CPU stuff...
 */
862
static void identify_cpu(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
863 864 865 866 867 868 869 870 871
{
	int i;

	c->loops_per_jiffy = loops_per_jiffy;
	c->x86_cache_size = -1;
	c->x86_vendor = X86_VENDOR_UNKNOWN;
	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
	c->x86_vendor_id[0] = '\0'; /* Unset */
	c->x86_model_id[0] = '\0';  /* Unset */
872
	c->x86_max_cores = 1;
873
	c->x86_coreid_bits = 0;
874
#ifdef CONFIG_X86_64
875
	c->x86_clflush_size = 64;
876 877
	c->x86_phys_bits = 36;
	c->x86_virt_bits = 48;
878 879
#else
	c->cpuid_level = -1;	/* CPUID not detected */
880
	c->x86_clflush_size = 32;
881 882
	c->x86_phys_bits = 32;
	c->x86_virt_bits = 32;
883 884
#endif
	c->x86_cache_alignment = c->x86_clflush_size;
L
Linus Torvalds 已提交
885 886 887 888
	memset(&c->x86_capability, 0, sizeof c->x86_capability);

	generic_identify(c);

889
	if (this_cpu->c_identify)
L
Linus Torvalds 已提交
890 891
		this_cpu->c_identify(c);

892 893 894 895 896 897
	/* Clear/Set all flags overriden by options, after probe */
	for (i = 0; i < NCAPINTS; i++) {
		c->x86_capability[i] &= ~cpu_caps_cleared[i];
		c->x86_capability[i] |= cpu_caps_set[i];
	}

898
#ifdef CONFIG_X86_64
899
	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
900 901
#endif

L
Linus Torvalds 已提交
902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917
	/*
	 * Vendor-specific initialization.  In this section we
	 * canonicalize the feature flags, meaning if there are
	 * features a certain CPU supports which CPUID doesn't
	 * tell us, CPUID claiming incorrect flags, or other bugs,
	 * we handle them here.
	 *
	 * At the end of this section, c->x86_capability better
	 * indicate the features this CPU genuinely supports!
	 */
	if (this_cpu->c_init)
		this_cpu->c_init(c);

	/* Disable the PN if appropriate */
	squash_the_stupid_serial_number(c);

918 919 920 921
	/* Set up SMEP/SMAP */
	setup_smep(c);
	setup_smap(c);

L
Linus Torvalds 已提交
922
	/*
I
Ingo Molnar 已提交
923 924
	 * The vendor-specific functions might have changed features.
	 * Now we do "generic changes."
L
Linus Torvalds 已提交
925 926
	 */

927 928 929
	/* Filter out anything that depends on CPUID levels we don't have */
	filter_cpuid_features(c, true);

L
Linus Torvalds 已提交
930
	/* If the model name is still unset, do table lookup. */
931
	if (!c->x86_model_id[0]) {
932
		const char *p;
L
Linus Torvalds 已提交
933
		p = table_lookup_model(c);
934
		if (p)
L
Linus Torvalds 已提交
935 936 937 938
			strcpy(c->x86_model_id, p);
		else
			/* Last resort... */
			sprintf(c->x86_model_id, "%02x/%02x",
939
				c->x86, c->x86_model);
L
Linus Torvalds 已提交
940 941
	}

942 943 944 945
#ifdef CONFIG_X86_64
	detect_ht(c);
#endif

946
	init_hypervisor(c);
947
	x86_init_rdrand(c);
948
	x86_init_cache_qos(c);
949 950 951 952 953 954 955 956 957 958

	/*
	 * Clear/Set all flags overriden by options, need do it
	 * before following smp all cpus cap AND.
	 */
	for (i = 0; i < NCAPINTS; i++) {
		c->x86_capability[i] &= ~cpu_caps_cleared[i];
		c->x86_capability[i] |= cpu_caps_set[i];
	}

L
Linus Torvalds 已提交
959 960 961 962 963 964
	/*
	 * On SMP, boot_cpu_data holds the common feature set between
	 * all CPUs; so make sure that we indicate which features are
	 * common between the CPUs.  The first time this routine gets
	 * executed, c == &boot_cpu_data.
	 */
965
	if (c != &boot_cpu_data) {
L
Linus Torvalds 已提交
966
		/* AND the already accumulated flags with these */
967
		for (i = 0; i < NCAPINTS; i++)
L
Linus Torvalds 已提交
968
			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
969 970 971 972

		/* OR, i.e. replicate the bug flags */
		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
L
Linus Torvalds 已提交
973 974 975
	}

	/* Init Machine Check Exception if available. */
976
	mcheck_cpu_init(c);
977 978

	select_idle_routine(c);
979

980
#ifdef CONFIG_NUMA
981 982
	numa_add_cpu(smp_processor_id());
#endif
983
}
S
Shaohua Li 已提交
984

985 986 987 988
/*
 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
 * on 32-bit kernels:
 */
989 990 991
#ifdef CONFIG_X86_32
void enable_sep_cpu(void)
{
992 993
	struct tss_struct *tss;
	int cpu;
994

995 996 997 998 999 1000 1001
	cpu = get_cpu();
	tss = &per_cpu(cpu_tss, cpu);

	if (!boot_cpu_has(X86_FEATURE_SEP))
		goto out;

	/*
1002 1003
	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
	 * see the big comment in struct x86_hw_tss's definition.
1004
	 */
1005 1006

	tss->x86_tss.ss1 = __KERNEL_CS;
1007 1008
	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);

1009 1010 1011
	wrmsr(MSR_IA32_SYSENTER_ESP,
	      (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
	      0);
1012

1013
	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1014 1015

out:
1016 1017
	put_cpu();
}
1018 1019
#endif

1020 1021 1022
void __init identify_boot_cpu(void)
{
	identify_cpu(&boot_cpu_data);
1023
	init_amd_e400_c1e_mask();
1024
#ifdef CONFIG_X86_32
1025
	sysenter_setup();
L
Li Shaohua 已提交
1026
	enable_sep_cpu();
1027
#endif
1028
	cpu_detect_tlb(&boot_cpu_data);
1029
}
S
Shaohua Li 已提交
1030

1031
void identify_secondary_cpu(struct cpuinfo_x86 *c)
1032 1033 1034
{
	BUG_ON(c == &boot_cpu_data);
	identify_cpu(c);
1035
#ifdef CONFIG_X86_32
1036
	enable_sep_cpu();
1037
#endif
1038
	mtrr_ap_init();
L
Linus Torvalds 已提交
1039 1040
}

1041
struct msr_range {
I
Ingo Molnar 已提交
1042 1043
	unsigned	min;
	unsigned	max;
1044
};
L
Linus Torvalds 已提交
1045

1046
static const struct msr_range msr_range_array[] = {
1047 1048 1049 1050 1051
	{ 0x00000000, 0x00000418},
	{ 0xc0000000, 0xc000040b},
	{ 0xc0010000, 0xc0010142},
	{ 0xc0011000, 0xc001103b},
};
L
Linus Torvalds 已提交
1052

1053
static void __print_cpu_msr(void)
1054
{
I
Ingo Molnar 已提交
1055
	unsigned index_min, index_max;
1056 1057 1058 1059 1060 1061 1062
	unsigned index;
	u64 val;
	int i;

	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
		index_min = msr_range_array[i].min;
		index_max = msr_range_array[i].max;
I
Ingo Molnar 已提交
1063

1064
		for (index = index_min; index < index_max; index++) {
1065
			if (rdmsrl_safe(index, &val))
1066 1067
				continue;
			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
L
Linus Torvalds 已提交
1068
		}
1069 1070
	}
}
1071

1072
static int show_msr;
I
Ingo Molnar 已提交
1073

1074 1075 1076
static __init int setup_show_msr(char *arg)
{
	int num;
1077

1078
	get_option(&arg, &num);
1079

1080 1081 1082
	if (num > 0)
		show_msr = num;
	return 1;
L
Linus Torvalds 已提交
1083
}
1084
__setup("show_msr=", setup_show_msr);
L
Linus Torvalds 已提交
1085

A
Andi Kleen 已提交
1086 1087
static __init int setup_noclflush(char *arg)
{
1088
	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1089
	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
A
Andi Kleen 已提交
1090 1091 1092 1093
	return 1;
}
__setup("noclflush", setup_noclflush);

1094
void print_cpu_info(struct cpuinfo_x86 *c)
L
Linus Torvalds 已提交
1095
{
1096
	const char *vendor = NULL;
L
Linus Torvalds 已提交
1097

I
Ingo Molnar 已提交
1098
	if (c->x86_vendor < X86_VENDOR_NUM) {
L
Linus Torvalds 已提交
1099
		vendor = this_cpu->c_vendor;
I
Ingo Molnar 已提交
1100 1101 1102 1103
	} else {
		if (c->cpuid_level >= 0)
			vendor = c->x86_vendor_id;
	}
L
Linus Torvalds 已提交
1104

1105
	if (vendor && !strstr(c->x86_model_id, vendor))
1106
		printk(KERN_CONT "%s ", vendor);
L
Linus Torvalds 已提交
1107

1108
	if (c->x86_model_id[0])
1109
		printk(KERN_CONT "%s", c->x86_model_id);
L
Linus Torvalds 已提交
1110
	else
1111
		printk(KERN_CONT "%d86", c->x86);
L
Linus Torvalds 已提交
1112

1113
	printk(KERN_CONT " (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1114

1115
	if (c->x86_mask || c->cpuid_level >= 0)
1116
		printk(KERN_CONT ", stepping: 0x%x)\n", c->x86_mask);
L
Linus Torvalds 已提交
1117
	else
1118
		printk(KERN_CONT ")\n");
1119

1120
	print_cpu_msr(c);
1121 1122
}

1123
void print_cpu_msr(struct cpuinfo_x86 *c)
1124
{
1125
	if (c->cpu_index < show_msr)
1126
		__print_cpu_msr();
L
Linus Torvalds 已提交
1127 1128
}

1129 1130 1131
static __init int setup_disablecpuid(char *arg)
{
	int bit;
I
Ingo Molnar 已提交
1132

1133 1134 1135 1136
	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
		setup_clear_cpu_cap(bit);
	else
		return 0;
I
Ingo Molnar 已提交
1137

1138 1139 1140 1141
	return 1;
}
__setup("clearcpuid=", setup_disablecpuid);

1142
#ifdef CONFIG_X86_64
1143
struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1144 1145
struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
				    (unsigned long) debug_idt_table };
1146

1147
DEFINE_PER_CPU_FIRST(union irq_stack_union,
1148
		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
I
Ingo Molnar 已提交
1149

1150
/*
1151 1152
 * The following percpu variables are hot.  Align current_task to
 * cacheline size such that they fall in the same cacheline.
1153 1154 1155 1156
 */
DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
	&init_task;
EXPORT_PER_CPU_SYMBOL(current_task);
1157

1158 1159 1160
DEFINE_PER_CPU(char *, irq_stack_ptr) =
	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;

1161
DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1162

1163 1164 1165
DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
EXPORT_PER_CPU_SYMBOL(__preempt_count);

I
Ingo Molnar 已提交
1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176
/*
 * Special IST stacks which the CPU switches to when it calls
 * an IST-marked descriptor entry. Up to 7 stacks (hardware
 * limit), all of them are 4K, except the debug stack which
 * is 8K.
 */
static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
};

1177
static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1178
	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1179 1180 1181

/* May not be marked __init: used by software suspend */
void syscall_init(void)
L
Linus Torvalds 已提交
1182
{
1183 1184 1185 1186 1187 1188
	/*
	 * LSTAR and STAR live in a bit strange symbiosis.
	 * They both write to the same internal register. STAR allows to
	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
	 */
	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1189
	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1190 1191

#ifdef CONFIG_IA32_EMULATION
1192
	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1193
	/*
1194 1195 1196 1197
	 * This only works on Intel CPUs.
	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
	 * This does not cause SYSENTER to jump to the wrong location, because
	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1198 1199 1200
	 */
	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1201
	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1202
#else
1203
	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1204
	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1205 1206
	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1207
#endif
1208

1209 1210
	/* Flags to clear on syscall */
	wrmsrl(MSR_SYSCALL_MASK,
1211
	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1212
	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
L
Linus Torvalds 已提交
1213
}
1214

1215 1216 1217 1218 1219 1220
/*
 * Copies of the original ist values from the tss are only accessed during
 * debugging, no special alignment required.
 */
DEFINE_PER_CPU(struct orig_ist, orig_ist);

1221
static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1222
DEFINE_PER_CPU(int, debug_stack_usage);
1223 1224 1225

int is_debug_stack(unsigned long addr)
{
1226 1227 1228
	return __this_cpu_read(debug_stack_usage) ||
		(addr <= __this_cpu_read(debug_stack_addr) &&
		 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
1229
}
1230
NOKPROBE_SYMBOL(is_debug_stack);
1231

1232
DEFINE_PER_CPU(u32, debug_idt_ctr);
1233

1234 1235
void debug_stack_set_zero(void)
{
1236 1237
	this_cpu_inc(debug_idt_ctr);
	load_current_idt();
1238
}
1239
NOKPROBE_SYMBOL(debug_stack_set_zero);
1240 1241 1242

void debug_stack_reset(void)
{
1243
	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1244
		return;
1245 1246
	if (this_cpu_dec_return(debug_idt_ctr) == 0)
		load_current_idt();
1247
}
1248
NOKPROBE_SYMBOL(debug_stack_reset);
1249

I
Ingo Molnar 已提交
1250
#else	/* CONFIG_X86_64 */
1251

1252 1253
DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
EXPORT_PER_CPU_SYMBOL(current_task);
1254 1255
DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
EXPORT_PER_CPU_SYMBOL(__preempt_count);
1256

1257 1258 1259 1260 1261 1262 1263 1264 1265
/*
 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
 * the top of the kernel stack.  Use an extra percpu variable to track the
 * top of the kernel stack directly.
 */
DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
	(unsigned long)&init_thread_union + THREAD_SIZE;
EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);

1266
#ifdef CONFIG_CC_STACKPROTECTOR
1267
DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1268
#endif
1269

I
Ingo Molnar 已提交
1270
#endif	/* CONFIG_X86_64 */
1271

1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286
/*
 * Clear all 6 debug registers:
 */
static void clear_all_debug_regs(void)
{
	int i;

	for (i = 0; i < 8; i++) {
		/* Ignore db4, db5 */
		if ((i == 4) || (i == 5))
			continue;

		set_debugreg(0, i);
	}
}
1287

1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301
#ifdef CONFIG_KGDB
/*
 * Restore debug regs if using kgdbwait and you have a kernel debugger
 * connection established.
 */
static void dbg_restore_debug_regs(void)
{
	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
		arch_kgdb_ops.correct_hw_break();
}
#else /* ! CONFIG_KGDB */
#define dbg_restore_debug_regs()
#endif /* ! CONFIG_KGDB */

1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314
static void wait_for_master_cpu(int cpu)
{
#ifdef CONFIG_SMP
	/*
	 * wait for ACK from master CPU before continuing
	 * with AP initialization
	 */
	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
		cpu_relax();
#endif
}

1315 1316 1317 1318 1319
/*
 * cpu_init() initializes state that is per-CPU. Some data is already
 * initialized (naturally) in the bootstrap process, such as the GDT
 * and IDT. We reload them nevertheless, this function acts as a
 * 'CPU state barrier', nothing should get across.
1320
 * A lot of state is already set up in PDA init for 64 bit
1321
 */
1322
#ifdef CONFIG_X86_64
I
Ingo Molnar 已提交
1323

1324
void cpu_init(void)
1325
{
1326
	struct orig_ist *oist;
1327
	struct task_struct *me;
I
Ingo Molnar 已提交
1328 1329
	struct tss_struct *t;
	unsigned long v;
1330
	int cpu = stack_smp_processor_id();
1331 1332
	int i;

1333 1334
	wait_for_master_cpu(cpu);

1335 1336 1337 1338 1339 1340
	/*
	 * Initialize the CR4 shadow before doing anything that could
	 * try to read it.
	 */
	cr4_init_shadow();

1341 1342 1343 1344 1345 1346
	/*
	 * Load microcode on this cpu if a valid microcode is available.
	 * This is early microcode loading procedure.
	 */
	load_ucode_ap();

1347
	t = &per_cpu(cpu_tss, cpu);
1348
	oist = &per_cpu(orig_ist, cpu);
I
Ingo Molnar 已提交
1349

1350
#ifdef CONFIG_NUMA
1351
	if (this_cpu_read(numa_node) == 0 &&
1352 1353
	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
		set_numa_node(early_cpu_to_node(cpu));
1354
#endif
1355 1356 1357

	me = current;

1358
	pr_debug("Initializing CPU#%d\n", cpu);
1359

A
Andy Lutomirski 已提交
1360
	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1361 1362 1363 1364 1365 1366

	/*
	 * Initialize the per-CPU GDT with the boot GDT,
	 * and set up the GDT descriptor:
	 */

1367
	switch_to_new_gdt(cpu);
1368 1369
	loadsegment(fs, 0);

1370
	load_current_idt();
1371 1372 1373 1374 1375 1376 1377 1378

	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
	syscall_init();

	wrmsrl(MSR_FS_BASE, 0);
	wrmsrl(MSR_KERNEL_GS_BASE, 0);
	barrier();

1379
	x86_configure_nx();
1380
	x2apic_setup();
1381 1382 1383 1384

	/*
	 * set up and load the per-CPU TSS
	 */
1385
	if (!oist->ist[0]) {
1386
		char *estacks = per_cpu(exception_stacks, cpu);
I
Ingo Molnar 已提交
1387

1388
		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
I
Ingo Molnar 已提交
1389
			estacks += exception_stack_sizes[v];
1390
			oist->ist[v] = t->x86_tss.ist[v] =
1391
					(unsigned long)estacks;
1392 1393
			if (v == DEBUG_STACK-1)
				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1394 1395 1396 1397
		}
	}

	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
I
Ingo Molnar 已提交
1398

1399 1400 1401 1402 1403 1404 1405 1406 1407
	/*
	 * <= is required because the CPU will access up to
	 * 8 bits beyond the end of the IO permission bitmap.
	 */
	for (i = 0; i <= IO_BITMAP_LONGS; i++)
		t->io_bitmap[i] = ~0UL;

	atomic_inc(&init_mm.mm_count);
	me->active_mm = &init_mm;
S
Stoyan Gaydarov 已提交
1408
	BUG_ON(me->mm);
1409 1410 1411 1412 1413
	enter_lazy_tlb(&init_mm, me);

	load_sp0(t, &current->thread);
	set_tss_desc(cpu, t);
	load_TR_desc();
1414
	load_mm_ldt(&init_mm);
1415

1416 1417
	clear_all_debug_regs();
	dbg_restore_debug_regs();
1418

I
Ingo Molnar 已提交
1419
	fpu__init_cpu();
1420 1421 1422 1423 1424 1425 1426

	if (is_uv_system())
		uv_cpu_init();
}

#else

1427
void cpu_init(void)
1428
{
1429 1430
	int cpu = smp_processor_id();
	struct task_struct *curr = current;
1431
	struct tss_struct *t = &per_cpu(cpu_tss, cpu);
1432
	struct thread_struct *thread = &curr->thread;
1433

1434
	wait_for_master_cpu(cpu);
1435

1436 1437 1438 1439 1440 1441
	/*
	 * Initialize the CR4 shadow before doing anything that could
	 * try to read it.
	 */
	cr4_init_shadow();

1442
	show_ucode_info_early();
1443 1444 1445

	printk(KERN_INFO "Initializing CPU#%d\n", cpu);

D
Dave Hansen 已提交
1446
	if (cpu_feature_enabled(X86_FEATURE_VME) || cpu_has_tsc || cpu_has_de)
A
Andy Lutomirski 已提交
1447
		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1448

1449
	load_current_idt();
1450
	switch_to_new_gdt(cpu);
L
Linus Torvalds 已提交
1451 1452 1453 1454 1455

	/*
	 * Set up and load the per-CPU TSS and LDT
	 */
	atomic_inc(&init_mm.mm_count);
1456
	curr->active_mm = &init_mm;
S
Stoyan Gaydarov 已提交
1457
	BUG_ON(curr->mm);
1458
	enter_lazy_tlb(&init_mm, curr);
L
Linus Torvalds 已提交
1459

1460
	load_sp0(t, thread);
1461
	set_tss_desc(cpu, t);
L
Linus Torvalds 已提交
1462
	load_TR_desc();
1463
	load_mm_ldt(&init_mm);
L
Linus Torvalds 已提交
1464

1465 1466
	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);

1467
#ifdef CONFIG_DOUBLEFAULT
L
Linus Torvalds 已提交
1468 1469
	/* Set up doublefault TSS pointer in the GDT */
	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1470
#endif
L
Linus Torvalds 已提交
1471

1472
	clear_all_debug_regs();
1473
	dbg_restore_debug_regs();
L
Linus Torvalds 已提交
1474

I
Ingo Molnar 已提交
1475
	fpu__init_cpu();
L
Linus Torvalds 已提交
1476
}
1477
#endif
1478 1479 1480 1481 1482 1483 1484 1485

#ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
void warn_pre_alternatives(void)
{
	WARN(1, "You're using static_cpu_has before alternatives have run!\n");
}
EXPORT_SYMBOL_GPL(warn_pre_alternatives);
#endif
1486 1487 1488 1489 1490 1491

inline bool __static_cpu_has_safe(u16 bit)
{
	return boot_cpu_has(bit);
}
EXPORT_SYMBOL_GPL(__static_cpu_has_safe);
1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508

static void bsp_resume(void)
{
	if (this_cpu->c_bsp_resume)
		this_cpu->c_bsp_resume(&boot_cpu_data);
}

static struct syscore_ops cpu_syscore_ops = {
	.resume		= bsp_resume,
};

static int __init init_cpu_syscore(void)
{
	register_syscore_ops(&cpu_syscore_ops);
	return 0;
}
core_initcall(init_cpu_syscore);