Linux Audio

Check our new training course

Loading...
v3.1
   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#include <linux/percpu.h>
   7#include <linux/string.h>
 
   8#include <linux/delay.h>
   9#include <linux/sched.h>
  10#include <linux/init.h>
 
  11#include <linux/kgdb.h>
  12#include <linux/smp.h>
  13#include <linux/io.h>
 
  14
  15#include <asm/stackprotector.h>
  16#include <asm/perf_event.h>
  17#include <asm/mmu_context.h>
 
  18#include <asm/hypervisor.h>
  19#include <asm/processor.h>
 
 
  20#include <asm/sections.h>
 
  21#include <linux/topology.h>
  22#include <linux/cpumask.h>
  23#include <asm/pgtable.h>
  24#include <linux/atomic.h>
  25#include <asm/proto.h>
  26#include <asm/setup.h>
  27#include <asm/apic.h>
  28#include <asm/desc.h>
  29#include <asm/i387.h>
  30#include <asm/mtrr.h>
  31#include <linux/numa.h>
  32#include <asm/asm.h>
 
  33#include <asm/cpu.h>
  34#include <asm/mce.h>
  35#include <asm/msr.h>
  36#include <asm/pat.h>
 
 
  37
  38#ifdef CONFIG_X86_LOCAL_APIC
  39#include <asm/uv/uv.h>
  40#endif
  41
  42#include "cpu.h"
  43
  44/* all of these masks are initialized in setup_cpu_local_masks() */
  45cpumask_var_t cpu_initialized_mask;
  46cpumask_var_t cpu_callout_mask;
  47cpumask_var_t cpu_callin_mask;
  48
  49/* representing cpus for which sibling maps can be computed */
  50cpumask_var_t cpu_sibling_setup_mask;
  51
  52/* correctly size the local cpu masks */
  53void __init setup_cpu_local_masks(void)
  54{
  55	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
  56	alloc_bootmem_cpumask_var(&cpu_callin_mask);
  57	alloc_bootmem_cpumask_var(&cpu_callout_mask);
  58	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
  59}
  60
  61static void __cpuinit default_init(struct cpuinfo_x86 *c)
  62{
  63#ifdef CONFIG_X86_64
  64	cpu_detect_cache_sizes(c);
  65#else
  66	/* Not much we can do here... */
  67	/* Check if at least it has cpuid */
  68	if (c->cpuid_level == -1) {
  69		/* No cpuid. It must be an ancient CPU */
  70		if (c->x86 == 4)
  71			strcpy(c->x86_model_id, "486");
  72		else if (c->x86 == 3)
  73			strcpy(c->x86_model_id, "386");
  74	}
  75#endif
  76}
  77
  78static const struct cpu_dev __cpuinitconst default_cpu = {
  79	.c_init		= default_init,
  80	.c_vendor	= "Unknown",
  81	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
  82};
  83
  84static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
  85
  86DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
  87#ifdef CONFIG_X86_64
  88	/*
  89	 * We need valid kernel segments for data and code in long mode too
  90	 * IRET will check the segment types  kkeil 2000/10/28
  91	 * Also sysret mandates a special GDT layout
  92	 *
  93	 * TLS descriptors are currently at a different place compared to i386.
  94	 * Hopefully nobody expects them at a fixed place (Wine?)
  95	 */
  96	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
  97	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
  98	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
  99	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
 100	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
 101	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
 102#else
 103	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
 104	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
 105	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
 106	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
 107	/*
 108	 * Segments used for calling PnP BIOS have byte granularity.
 109	 * They code segments and data segments have fixed 64k limits,
 110	 * the transfer segment sizes are set at run time.
 111	 */
 112	/* 32-bit code */
 113	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
 114	/* 16-bit code */
 115	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
 116	/* 16-bit data */
 117	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
 118	/* 16-bit data */
 119	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
 120	/* 16-bit data */
 121	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
 122	/*
 123	 * The APM segments have byte granularity and their bases
 124	 * are set at run time.  All have 64k limits.
 125	 */
 126	/* 32-bit code */
 127	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
 128	/* 16-bit code */
 129	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
 130	/* data */
 131	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
 132
 133	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
 134	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
 135	GDT_STACK_CANARY_INIT
 136#endif
 137} };
 138EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
 139
 140static int __init x86_xsave_setup(char *s)
 141{
 142	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
 143	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
 
 
 
 
 
 
 
 
 144	return 1;
 145}
 146__setup("noxsave", x86_xsave_setup);
 147
 148static int __init x86_xsaveopt_setup(char *s)
 149{
 150	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
 151	return 1;
 
 
 
 
 
 
 
 
 
 152}
 153__setup("noxsaveopt", x86_xsaveopt_setup);
 154
 155#ifdef CONFIG_X86_32
 156static int cachesize_override __cpuinitdata = -1;
 157static int disable_x86_serial_nr __cpuinitdata = 1;
 158
 159static int __init cachesize_setup(char *str)
 160{
 161	get_option(&str, &cachesize_override);
 162	return 1;
 163}
 164__setup("cachesize=", cachesize_setup);
 165
 166static int __init x86_fxsr_setup(char *s)
 167{
 168	setup_clear_cpu_cap(X86_FEATURE_FXSR);
 169	setup_clear_cpu_cap(X86_FEATURE_XMM);
 170	return 1;
 171}
 172__setup("nofxsr", x86_fxsr_setup);
 173
 174static int __init x86_sep_setup(char *s)
 175{
 176	setup_clear_cpu_cap(X86_FEATURE_SEP);
 177	return 1;
 178}
 179__setup("nosep", x86_sep_setup);
 180
 181/* Standard macro to see if a specific flag is changeable */
 182static inline int flag_is_changeable_p(u32 flag)
 183{
 184	u32 f1, f2;
 185
 186	/*
 187	 * Cyrix and IDT cpus allow disabling of CPUID
 188	 * so the code below may return different results
 189	 * when it is executed before and after enabling
 190	 * the CPUID. Add "volatile" to not allow gcc to
 191	 * optimize the subsequent calls to this function.
 192	 */
 193	asm volatile ("pushfl		\n\t"
 194		      "pushfl		\n\t"
 195		      "popl %0		\n\t"
 196		      "movl %0, %1	\n\t"
 197		      "xorl %2, %0	\n\t"
 198		      "pushl %0		\n\t"
 199		      "popfl		\n\t"
 200		      "pushfl		\n\t"
 201		      "popl %0		\n\t"
 202		      "popfl		\n\t"
 203
 204		      : "=&r" (f1), "=&r" (f2)
 205		      : "ir" (flag));
 206
 207	return ((f1^f2) & flag) != 0;
 208}
 209
 210/* Probe for the CPUID instruction */
 211static int __cpuinit have_cpuid_p(void)
 212{
 213	return flag_is_changeable_p(X86_EFLAGS_ID);
 214}
 215
 216static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
 217{
 218	unsigned long lo, hi;
 219
 220	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
 221		return;
 222
 223	/* Disable processor serial number: */
 224
 225	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
 226	lo |= 0x200000;
 227	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
 228
 229	printk(KERN_NOTICE "CPU serial number disabled.\n");
 230	clear_cpu_cap(c, X86_FEATURE_PN);
 231
 232	/* Disabling the serial number may affect the cpuid level */
 233	c->cpuid_level = cpuid_eax(0);
 234}
 235
 236static int __init x86_serial_nr_setup(char *s)
 237{
 238	disable_x86_serial_nr = 0;
 239	return 1;
 240}
 241__setup("serialnumber", x86_serial_nr_setup);
 242#else
 243static inline int flag_is_changeable_p(u32 flag)
 244{
 245	return 1;
 246}
 247/* Probe for the CPUID instruction */
 248static inline int have_cpuid_p(void)
 249{
 250	return 1;
 251}
 252static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
 253{
 254}
 255#endif
 256
 257static int disable_smep __cpuinitdata;
 258static __init int setup_disable_smep(char *arg)
 259{
 260	disable_smep = 1;
 
 
 261	return 1;
 262}
 263__setup("nosmep", setup_disable_smep);
 264
 265static __cpuinit void setup_smep(struct cpuinfo_x86 *c)
 266{
 267	if (cpu_has(c, X86_FEATURE_SMEP)) {
 268		if (unlikely(disable_smep)) {
 269			setup_clear_cpu_cap(X86_FEATURE_SMEP);
 270			clear_in_cr4(X86_CR4_SMEP);
 271		} else
 272			set_in_cr4(X86_CR4_SMEP);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 273	}
 274}
 275
 276/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 277 * Some CPU features depend on higher CPUID levels, which may not always
 278 * be available due to CPUID level capping or broken virtualization
 279 * software.  Add those features to this table to auto-disable them.
 280 */
 281struct cpuid_dependent_feature {
 282	u32 feature;
 283	u32 level;
 284};
 285
 286static const struct cpuid_dependent_feature __cpuinitconst
 287cpuid_dependent_features[] = {
 288	{ X86_FEATURE_MWAIT,		0x00000005 },
 289	{ X86_FEATURE_DCA,		0x00000009 },
 290	{ X86_FEATURE_XSAVE,		0x0000000d },
 291	{ 0, 0 }
 292};
 293
 294static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
 295{
 296	const struct cpuid_dependent_feature *df;
 297
 298	for (df = cpuid_dependent_features; df->feature; df++) {
 299
 300		if (!cpu_has(c, df->feature))
 301			continue;
 302		/*
 303		 * Note: cpuid_level is set to -1 if unavailable, but
 304		 * extended_extended_level is set to 0 if unavailable
 305		 * and the legitimate extended levels are all negative
 306		 * when signed; hence the weird messing around with
 307		 * signs here...
 308		 */
 309		if (!((s32)df->level < 0 ?
 310		     (u32)df->level > (u32)c->extended_cpuid_level :
 311		     (s32)df->level > (s32)c->cpuid_level))
 312			continue;
 313
 314		clear_cpu_cap(c, df->feature);
 315		if (!warn)
 316			continue;
 317
 318		printk(KERN_WARNING
 319		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
 320				x86_cap_flags[df->feature], df->level);
 321	}
 322}
 323
 324/*
 325 * Naming convention should be: <Name> [(<Codename>)]
 326 * This table only is used unless init_<vendor>() below doesn't set it;
 327 * in particular, if CPUID levels 0x80000002..4 are supported, this
 328 * isn't used
 329 */
 330
 331/* Look up CPU names by table lookup. */
 332static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
 333{
 334	const struct cpu_model_info *info;
 
 335
 336	if (c->x86_model >= 16)
 337		return NULL;	/* Range check */
 338
 339	if (!this_cpu)
 340		return NULL;
 341
 342	info = this_cpu->c_models;
 343
 344	while (info && info->family) {
 345		if (info->family == c->x86)
 346			return info->model_names[c->x86_model];
 347		info++;
 348	}
 
 349	return NULL;		/* Not found */
 350}
 351
 352__u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
 353__u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
 354
 355void load_percpu_segment(int cpu)
 356{
 357#ifdef CONFIG_X86_32
 358	loadsegment(fs, __KERNEL_PERCPU);
 359#else
 360	loadsegment(gs, 0);
 361	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
 362#endif
 363	load_stack_canary_segment();
 364}
 365
 366/*
 367 * Current gdt points %fs at the "master" per-cpu area: after this,
 368 * it's on the real one.
 369 */
 370void switch_to_new_gdt(int cpu)
 371{
 372	struct desc_ptr gdt_descr;
 373
 374	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
 375	gdt_descr.size = GDT_SIZE - 1;
 376	load_gdt(&gdt_descr);
 377	/* Reload the per-cpu base */
 378
 379	load_percpu_segment(cpu);
 380}
 381
 382static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
 383
 384static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
 385{
 386	unsigned int *v;
 387	char *p, *q;
 388
 389	if (c->extended_cpuid_level < 0x80000004)
 390		return;
 391
 392	v = (unsigned int *)c->x86_model_id;
 393	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
 394	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
 395	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
 396	c->x86_model_id[48] = 0;
 397
 398	/*
 399	 * Intel chips right-justify this string for some dumb reason;
 400	 * undo that brain damage:
 401	 */
 402	p = q = &c->x86_model_id[0];
 403	while (*p == ' ')
 404		p++;
 405	if (p != q) {
 406		while (*p)
 407			*q++ = *p++;
 408		while (q <= &c->x86_model_id[48])
 409			*q++ = '\0';	/* Zero-pad the rest */
 
 
 410	}
 
 
 411}
 412
 413void __cpuinit cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
 414{
 415	unsigned int n, dummy, ebx, ecx, edx, l2size;
 416
 417	n = c->extended_cpuid_level;
 418
 419	if (n >= 0x80000005) {
 420		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
 421		c->x86_cache_size = (ecx>>24) + (edx>>24);
 422#ifdef CONFIG_X86_64
 423		/* On K8 L1 TLB is inclusive, so don't count it */
 424		c->x86_tlbsize = 0;
 425#endif
 426	}
 427
 428	if (n < 0x80000006)	/* Some chips just has a large L1. */
 429		return;
 430
 431	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
 432	l2size = ecx >> 16;
 433
 434#ifdef CONFIG_X86_64
 435	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
 436#else
 437	/* do processor-specific cache resizing */
 438	if (this_cpu->c_size_cache)
 439		l2size = this_cpu->c_size_cache(c, l2size);
 440
 441	/* Allow user to override all this if necessary. */
 442	if (cachesize_override != -1)
 443		l2size = cachesize_override;
 444
 445	if (l2size == 0)
 446		return;		/* Again, no L2 cache is possible */
 447#endif
 448
 449	c->x86_cache_size = l2size;
 450}
 451
 452void __cpuinit detect_ht(struct cpuinfo_x86 *c)
 
 
 
 
 
 
 
 
 453{
 454#ifdef CONFIG_X86_HT
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 455	u32 eax, ebx, ecx, edx;
 456	int index_msb, core_bits;
 457	static bool printed;
 458
 459	if (!cpu_has(c, X86_FEATURE_HT))
 460		return;
 461
 462	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
 463		goto out;
 464
 465	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
 466		return;
 467
 468	cpuid(1, &eax, &ebx, &ecx, &edx);
 469
 470	smp_num_siblings = (ebx & 0xff0000) >> 16;
 471
 472	if (smp_num_siblings == 1) {
 473		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
 474		goto out;
 475	}
 476
 477	if (smp_num_siblings <= 1)
 478		goto out;
 479
 480	index_msb = get_count_order(smp_num_siblings);
 481	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
 482
 483	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
 484
 485	index_msb = get_count_order(smp_num_siblings);
 486
 487	core_bits = get_count_order(c->x86_max_cores);
 488
 489	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
 490				       ((1 << core_bits) - 1);
 491
 492out:
 493	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
 494		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
 495		       c->phys_proc_id);
 496		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
 497		       c->cpu_core_id);
 498		printed = 1;
 499	}
 500#endif
 501}
 502
 503static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
 504{
 505	char *v = c->x86_vendor_id;
 506	int i;
 507
 508	for (i = 0; i < X86_VENDOR_NUM; i++) {
 509		if (!cpu_devs[i])
 510			break;
 511
 512		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
 513		    (cpu_devs[i]->c_ident[1] &&
 514		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
 515
 516			this_cpu = cpu_devs[i];
 517			c->x86_vendor = this_cpu->c_x86_vendor;
 518			return;
 519		}
 520	}
 521
 522	printk_once(KERN_ERR
 523			"CPU: vendor_id '%s' unknown, using generic init.\n" \
 524			"CPU: Your system may be unstable.\n", v);
 525
 526	c->x86_vendor = X86_VENDOR_UNKNOWN;
 527	this_cpu = &default_cpu;
 528}
 529
 530void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
 531{
 532	/* Get vendor name */
 533	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
 534	      (unsigned int *)&c->x86_vendor_id[0],
 535	      (unsigned int *)&c->x86_vendor_id[8],
 536	      (unsigned int *)&c->x86_vendor_id[4]);
 537
 538	c->x86 = 4;
 539	/* Intel-defined flags: level 0x00000001 */
 540	if (c->cpuid_level >= 0x00000001) {
 541		u32 junk, tfms, cap0, misc;
 542
 543		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
 544		c->x86 = (tfms >> 8) & 0xf;
 545		c->x86_model = (tfms >> 4) & 0xf;
 546		c->x86_mask = tfms & 0xf;
 547
 548		if (c->x86 == 0xf)
 549			c->x86 += (tfms >> 20) & 0xff;
 550		if (c->x86 >= 0x6)
 551			c->x86_model += ((tfms >> 16) & 0xf) << 4;
 552
 553		if (cap0 & (1<<19)) {
 554			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
 555			c->x86_cache_alignment = c->x86_clflush_size;
 556		}
 557	}
 558}
 559
 560void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
 561{
 562	u32 tfms, xlvl;
 563	u32 ebx;
 564
 565	/* Intel-defined flags: level 0x00000001 */
 566	if (c->cpuid_level >= 0x00000001) {
 567		u32 capability, excap;
 568
 569		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
 570		c->x86_capability[0] = capability;
 571		c->x86_capability[4] = excap;
 572	}
 573
 
 
 
 
 574	/* Additional Intel-defined flags: level 0x00000007 */
 575	if (c->cpuid_level >= 0x00000007) {
 576		u32 eax, ebx, ecx, edx;
 577
 578		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
 
 
 
 
 
 
 
 579
 580		c->x86_capability[9] = ebx;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 581	}
 582
 583	/* AMD-defined flags: level 0x80000001 */
 584	xlvl = cpuid_eax(0x80000000);
 585	c->extended_cpuid_level = xlvl;
 
 
 
 
 586
 587	if ((xlvl & 0xffff0000) == 0x80000000) {
 588		if (xlvl >= 0x80000001) {
 589			c->x86_capability[1] = cpuid_edx(0x80000001);
 590			c->x86_capability[6] = cpuid_ecx(0x80000001);
 591		}
 592	}
 593
 
 
 
 
 
 
 
 594	if (c->extended_cpuid_level >= 0x80000008) {
 595		u32 eax = cpuid_eax(0x80000008);
 596
 597		c->x86_virt_bits = (eax >> 8) & 0xff;
 598		c->x86_phys_bits = eax & 0xff;
 
 599	}
 600#ifdef CONFIG_X86_32
 601	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
 602		c->x86_phys_bits = 36;
 603#endif
 604
 605	if (c->extended_cpuid_level >= 0x80000007)
 606		c->x86_power = cpuid_edx(0x80000007);
 607
 608	init_scattered_cpuid_features(c);
 609}
 610
 611static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
 612{
 613#ifdef CONFIG_X86_32
 614	int i;
 615
 616	/*
 617	 * First of all, decide if this is a 486 or higher
 618	 * It's a 486 if we can modify the AC flag
 619	 */
 620	if (flag_is_changeable_p(X86_EFLAGS_AC))
 621		c->x86 = 4;
 622	else
 623		c->x86 = 3;
 624
 625	for (i = 0; i < X86_VENDOR_NUM; i++)
 626		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
 627			c->x86_vendor_id[0] = 0;
 628			cpu_devs[i]->c_identify(c);
 629			if (c->x86_vendor_id[0]) {
 630				get_cpu_vendor(c);
 631				break;
 632			}
 633		}
 634#endif
 635}
 636
 637/*
 638 * Do minimum CPU detection early.
 639 * Fields really needed: vendor, cpuid_level, family, model, mask,
 640 * cache alignment.
 641 * The others are not touched to avoid unwanted side effects.
 642 *
 643 * WARNING: this function is only called on the BP.  Don't add code here
 644 * that is supposed to run on all CPUs.
 645 */
 646static void __init early_identify_cpu(struct cpuinfo_x86 *c)
 647{
 648#ifdef CONFIG_X86_64
 649	c->x86_clflush_size = 64;
 650	c->x86_phys_bits = 36;
 651	c->x86_virt_bits = 48;
 652#else
 653	c->x86_clflush_size = 32;
 654	c->x86_phys_bits = 32;
 655	c->x86_virt_bits = 32;
 656#endif
 657	c->x86_cache_alignment = c->x86_clflush_size;
 658
 659	memset(&c->x86_capability, 0, sizeof c->x86_capability);
 660	c->extended_cpuid_level = 0;
 661
 662	if (!have_cpuid_p())
 663		identify_cpu_without_cpuid(c);
 664
 665	/* cyrix could have cpuid enabled via c_identify()*/
 666	if (!have_cpuid_p())
 667		return;
 
 
 668
 669	cpu_detect(c);
 670
 671	get_cpu_vendor(c);
 672
 673	get_cpu_cap(c);
 
 674
 675	if (this_cpu->c_early_init)
 676		this_cpu->c_early_init(c);
 677
 678#ifdef CONFIG_SMP
 679	c->cpu_index = 0;
 680#endif
 681	filter_cpuid_features(c, false);
 682
 683	setup_smep(c);
 
 684}
 685
 686void __init early_cpu_init(void)
 687{
 688	const struct cpu_dev *const *cdev;
 689	int count = 0;
 690
 691#ifdef CONFIG_PROCESSOR_SELECT
 692	printk(KERN_INFO "KERNEL supported cpus:\n");
 693#endif
 694
 695	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
 696		const struct cpu_dev *cpudev = *cdev;
 697
 698		if (count >= X86_VENDOR_NUM)
 699			break;
 700		cpu_devs[count] = cpudev;
 701		count++;
 702
 703#ifdef CONFIG_PROCESSOR_SELECT
 704		{
 705			unsigned int j;
 706
 707			for (j = 0; j < 2; j++) {
 708				if (!cpudev->c_ident[j])
 709					continue;
 710				printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
 711					cpudev->c_ident[j]);
 712			}
 713		}
 714#endif
 715	}
 716	early_identify_cpu(&boot_cpu_data);
 717}
 718
 719/*
 720 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
 721 * unfortunately, that's not true in practice because of early VIA
 722 * chips and (more importantly) broken virtualizers that are not easy
 723 * to detect. In the latter case it doesn't even *fail* reliably, so
 724 * probing for it doesn't even work. Disable it completely on 32-bit
 725 * unless we can find a reliable way to detect all the broken cases.
 726 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
 727 */
 728static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
 729{
 730#ifdef CONFIG_X86_32
 731	clear_cpu_cap(c, X86_FEATURE_NOPL);
 732#else
 733	set_cpu_cap(c, X86_FEATURE_NOPL);
 734#endif
 735}
 736
 737static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 738{
 739	c->extended_cpuid_level = 0;
 740
 741	if (!have_cpuid_p())
 742		identify_cpu_without_cpuid(c);
 743
 744	/* cyrix could have cpuid enabled via c_identify()*/
 745	if (!have_cpuid_p())
 746		return;
 747
 748	cpu_detect(c);
 749
 750	get_cpu_vendor(c);
 751
 752	get_cpu_cap(c);
 753
 754	if (c->cpuid_level >= 0x00000001) {
 755		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
 756#ifdef CONFIG_X86_32
 757# ifdef CONFIG_X86_HT
 758		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
 759# else
 760		c->apicid = c->initial_apicid;
 761# endif
 762#endif
 763
 764#ifdef CONFIG_X86_HT
 765		c->phys_proc_id = c->initial_apicid;
 766#endif
 767	}
 768
 769	setup_smep(c);
 770
 771	get_model_name(c); /* Default name */
 772
 773	detect_nopl(c);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 774}
 775
 776/*
 777 * This does the hard work of actually picking apart the CPU stuff...
 778 */
 779static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
 780{
 781	int i;
 782
 783	c->loops_per_jiffy = loops_per_jiffy;
 784	c->x86_cache_size = -1;
 785	c->x86_vendor = X86_VENDOR_UNKNOWN;
 786	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
 787	c->x86_vendor_id[0] = '\0'; /* Unset */
 788	c->x86_model_id[0] = '\0';  /* Unset */
 789	c->x86_max_cores = 1;
 790	c->x86_coreid_bits = 0;
 
 791#ifdef CONFIG_X86_64
 792	c->x86_clflush_size = 64;
 793	c->x86_phys_bits = 36;
 794	c->x86_virt_bits = 48;
 795#else
 796	c->cpuid_level = -1;	/* CPUID not detected */
 797	c->x86_clflush_size = 32;
 798	c->x86_phys_bits = 32;
 799	c->x86_virt_bits = 32;
 800#endif
 801	c->x86_cache_alignment = c->x86_clflush_size;
 802	memset(&c->x86_capability, 0, sizeof c->x86_capability);
 803
 804	generic_identify(c);
 805
 806	if (this_cpu->c_identify)
 807		this_cpu->c_identify(c);
 808
 809	/* Clear/Set all flags overriden by options, after probe */
 810	for (i = 0; i < NCAPINTS; i++) {
 811		c->x86_capability[i] &= ~cpu_caps_cleared[i];
 812		c->x86_capability[i] |= cpu_caps_set[i];
 813	}
 814
 815#ifdef CONFIG_X86_64
 816	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
 817#endif
 818
 819	/*
 820	 * Vendor-specific initialization.  In this section we
 821	 * canonicalize the feature flags, meaning if there are
 822	 * features a certain CPU supports which CPUID doesn't
 823	 * tell us, CPUID claiming incorrect flags, or other bugs,
 824	 * we handle them here.
 825	 *
 826	 * At the end of this section, c->x86_capability better
 827	 * indicate the features this CPU genuinely supports!
 828	 */
 829	if (this_cpu->c_init)
 830		this_cpu->c_init(c);
 831
 832	/* Disable the PN if appropriate */
 833	squash_the_stupid_serial_number(c);
 834
 
 
 
 
 835	/*
 836	 * The vendor-specific functions might have changed features.
 837	 * Now we do "generic changes."
 838	 */
 839
 840	/* Filter out anything that depends on CPUID levels we don't have */
 841	filter_cpuid_features(c, true);
 842
 843	/* If the model name is still unset, do table lookup. */
 844	if (!c->x86_model_id[0]) {
 845		const char *p;
 846		p = table_lookup_model(c);
 847		if (p)
 848			strcpy(c->x86_model_id, p);
 849		else
 850			/* Last resort... */
 851			sprintf(c->x86_model_id, "%02x/%02x",
 852				c->x86, c->x86_model);
 853	}
 854
 855#ifdef CONFIG_X86_64
 856	detect_ht(c);
 857#endif
 858
 859	init_hypervisor(c);
 
 
 
 860
 861	/*
 862	 * Clear/Set all flags overriden by options, need do it
 863	 * before following smp all cpus cap AND.
 864	 */
 865	for (i = 0; i < NCAPINTS; i++) {
 866		c->x86_capability[i] &= ~cpu_caps_cleared[i];
 867		c->x86_capability[i] |= cpu_caps_set[i];
 868	}
 869
 870	/*
 871	 * On SMP, boot_cpu_data holds the common feature set between
 872	 * all CPUs; so make sure that we indicate which features are
 873	 * common between the CPUs.  The first time this routine gets
 874	 * executed, c == &boot_cpu_data.
 875	 */
 876	if (c != &boot_cpu_data) {
 877		/* AND the already accumulated flags with these */
 878		for (i = 0; i < NCAPINTS; i++)
 879			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
 
 
 
 
 880	}
 881
 882	/* Init Machine Check Exception if available. */
 883	mcheck_cpu_init(c);
 884
 885	select_idle_routine(c);
 886
 887#ifdef CONFIG_NUMA
 888	numa_add_cpu(smp_processor_id());
 889#endif
 890}
 891
 892#ifdef CONFIG_X86_64
 893static void vgetcpu_set_mode(void)
 
 
 
 
 894{
 895	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
 896		vgetcpu_mode = VGETCPU_RDTSCP;
 897	else
 898		vgetcpu_mode = VGETCPU_LSL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 899}
 900#endif
 901
 902void __init identify_boot_cpu(void)
 903{
 904	identify_cpu(&boot_cpu_data);
 905	init_amd_e400_c1e_mask();
 906#ifdef CONFIG_X86_32
 907	sysenter_setup();
 908	enable_sep_cpu();
 909#else
 910	vgetcpu_set_mode();
 911#endif
 
 912}
 913
 914void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
 915{
 916	BUG_ON(c == &boot_cpu_data);
 917	identify_cpu(c);
 918#ifdef CONFIG_X86_32
 919	enable_sep_cpu();
 920#endif
 921	mtrr_ap_init();
 
 922}
 923
 924struct msr_range {
 925	unsigned	min;
 926	unsigned	max;
 927};
 928
 929static const struct msr_range msr_range_array[] __cpuinitconst = {
 930	{ 0x00000000, 0x00000418},
 931	{ 0xc0000000, 0xc000040b},
 932	{ 0xc0010000, 0xc0010142},
 933	{ 0xc0011000, 0xc001103b},
 934};
 935
 936static void __cpuinit print_cpu_msr(void)
 937{
 938	unsigned index_min, index_max;
 939	unsigned index;
 940	u64 val;
 941	int i;
 942
 943	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
 944		index_min = msr_range_array[i].min;
 945		index_max = msr_range_array[i].max;
 946
 947		for (index = index_min; index < index_max; index++) {
 948			if (rdmsrl_amd_safe(index, &val))
 949				continue;
 950			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
 951		}
 952	}
 953}
 954
 955static int show_msr __cpuinitdata;
 956
 957static __init int setup_show_msr(char *arg)
 958{
 959	int num;
 960
 961	get_option(&arg, &num);
 962
 963	if (num > 0)
 964		show_msr = num;
 965	return 1;
 966}
 967__setup("show_msr=", setup_show_msr);
 968
 969static __init int setup_noclflush(char *arg)
 970{
 971	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
 
 972	return 1;
 973}
 974__setup("noclflush", setup_noclflush);
 975
 976void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
 977{
 978	const char *vendor = NULL;
 979
 980	if (c->x86_vendor < X86_VENDOR_NUM) {
 981		vendor = this_cpu->c_vendor;
 982	} else {
 983		if (c->cpuid_level >= 0)
 984			vendor = c->x86_vendor_id;
 985	}
 986
 987	if (vendor && !strstr(c->x86_model_id, vendor))
 988		printk(KERN_CONT "%s ", vendor);
 989
 990	if (c->x86_model_id[0])
 991		printk(KERN_CONT "%s", c->x86_model_id);
 992	else
 993		printk(KERN_CONT "%d86", c->x86);
 
 
 994
 995	if (c->x86_mask || c->cpuid_level >= 0)
 996		printk(KERN_CONT " stepping %02x\n", c->x86_mask);
 997	else
 998		printk(KERN_CONT "\n");
 999
1000#ifdef CONFIG_SMP
1001	if (c->cpu_index < show_msr)
1002		print_cpu_msr();
1003#else
1004	if (show_msr)
1005		print_cpu_msr();
1006#endif
1007}
1008
1009static __init int setup_disablecpuid(char *arg)
1010{
1011	int bit;
1012
1013	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1014		setup_clear_cpu_cap(bit);
1015	else
1016		return 0;
1017
1018	return 1;
1019}
1020__setup("clearcpuid=", setup_disablecpuid);
1021
1022#ifdef CONFIG_X86_64
1023struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
 
 
 
 
 
 
 
1024
1025DEFINE_PER_CPU_FIRST(union irq_stack_union,
1026		     irq_stack_union) __aligned(PAGE_SIZE);
1027
1028/*
1029 * The following four percpu variables are hot.  Align current_task to
1030 * cacheline size such that all four fall in the same cacheline.
1031 */
1032DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1033	&init_task;
1034EXPORT_PER_CPU_SYMBOL(current_task);
1035
1036DEFINE_PER_CPU(unsigned long, kernel_stack) =
1037	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1038EXPORT_PER_CPU_SYMBOL(kernel_stack);
1039
1040DEFINE_PER_CPU(char *, irq_stack_ptr) =
1041	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
 
 
1042
1043DEFINE_PER_CPU(unsigned int, irq_count) = -1;
 
1044
1045/*
1046 * Special IST stacks which the CPU switches to when it calls
1047 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1048 * limit), all of them are 4K, except the debug stack which
1049 * is 8K.
1050 */
1051static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1052	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1053	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1054};
1055
1056static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1057	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1058
1059/* May not be marked __init: used by software suspend */
1060void syscall_init(void)
1061{
1062	/*
1063	 * LSTAR and STAR live in a bit strange symbiosis.
1064	 * They both write to the same internal register. STAR allows to
1065	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1066	 */
1067	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1068	wrmsrl(MSR_LSTAR, system_call);
1069	wrmsrl(MSR_CSTAR, ignore_sysret);
1070
1071#ifdef CONFIG_IA32_EMULATION
1072	syscall32_cpu_init();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1073#endif
1074
1075	/* Flags to clear on syscall */
1076	wrmsrl(MSR_SYSCALL_MASK,
1077	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
 
1078}
1079
1080unsigned long kernel_eflags;
1081
1082/*
1083 * Copies of the original ist values from the tss are only accessed during
1084 * debugging, no special alignment required.
1085 */
1086DEFINE_PER_CPU(struct orig_ist, orig_ist);
1087
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1088#else	/* CONFIG_X86_64 */
1089
1090DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1091EXPORT_PER_CPU_SYMBOL(current_task);
 
 
 
 
 
 
 
 
 
 
 
1092
1093#ifdef CONFIG_CC_STACKPROTECTOR
1094DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1095#endif
1096
1097/* Make sure %fs and %gs are initialized properly in idle threads */
1098struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1099{
1100	memset(regs, 0, sizeof(struct pt_regs));
1101	regs->fs = __KERNEL_PERCPU;
1102	regs->gs = __KERNEL_STACK_CANARY;
1103
1104	return regs;
1105}
1106#endif	/* CONFIG_X86_64 */
1107
1108/*
1109 * Clear all 6 debug registers:
1110 */
1111static void clear_all_debug_regs(void)
1112{
1113	int i;
1114
1115	for (i = 0; i < 8; i++) {
1116		/* Ignore db4, db5 */
1117		if ((i == 4) || (i == 5))
1118			continue;
1119
1120		set_debugreg(0, i);
1121	}
1122}
1123
1124#ifdef CONFIG_KGDB
1125/*
1126 * Restore debug regs if using kgdbwait and you have a kernel debugger
1127 * connection established.
1128 */
1129static void dbg_restore_debug_regs(void)
1130{
1131	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1132		arch_kgdb_ops.correct_hw_break();
1133}
1134#else /* ! CONFIG_KGDB */
1135#define dbg_restore_debug_regs()
1136#endif /* ! CONFIG_KGDB */
1137
 
 
 
 
 
 
 
 
 
 
 
 
 
1138/*
1139 * cpu_init() initializes state that is per-CPU. Some data is already
1140 * initialized (naturally) in the bootstrap process, such as the GDT
1141 * and IDT. We reload them nevertheless, this function acts as a
1142 * 'CPU state barrier', nothing should get across.
1143 * A lot of state is already set up in PDA init for 64 bit
1144 */
1145#ifdef CONFIG_X86_64
1146
1147void __cpuinit cpu_init(void)
1148{
1149	struct orig_ist *oist;
1150	struct task_struct *me;
1151	struct tss_struct *t;
1152	unsigned long v;
1153	int cpu;
1154	int i;
1155
1156	cpu = stack_smp_processor_id();
1157	t = &per_cpu(init_tss, cpu);
 
 
 
 
 
 
 
 
 
 
1158	oist = &per_cpu(orig_ist, cpu);
1159
1160#ifdef CONFIG_NUMA
1161	if (cpu != 0 && percpu_read(numa_node) == 0 &&
1162	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1163		set_numa_node(early_cpu_to_node(cpu));
1164#endif
1165
1166	me = current;
1167
1168	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1169		panic("CPU#%d already initialized!\n", cpu);
1170
1171	pr_debug("Initializing CPU#%d\n", cpu);
1172
1173	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1174
1175	/*
1176	 * Initialize the per-CPU GDT with the boot GDT,
1177	 * and set up the GDT descriptor:
1178	 */
1179
1180	switch_to_new_gdt(cpu);
1181	loadsegment(fs, 0);
1182
1183	load_idt((const struct desc_ptr *)&idt_descr);
1184
1185	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1186	syscall_init();
1187
1188	wrmsrl(MSR_FS_BASE, 0);
1189	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1190	barrier();
1191
1192	x86_configure_nx();
1193	if (cpu != 0)
1194		enable_x2apic();
1195
1196	/*
1197	 * set up and load the per-CPU TSS
1198	 */
1199	if (!oist->ist[0]) {
1200		char *estacks = per_cpu(exception_stacks, cpu);
1201
1202		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1203			estacks += exception_stack_sizes[v];
1204			oist->ist[v] = t->x86_tss.ist[v] =
1205					(unsigned long)estacks;
 
 
1206		}
1207	}
1208
1209	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1210
1211	/*
1212	 * <= is required because the CPU will access up to
1213	 * 8 bits beyond the end of the IO permission bitmap.
1214	 */
1215	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1216		t->io_bitmap[i] = ~0UL;
1217
1218	atomic_inc(&init_mm.mm_count);
1219	me->active_mm = &init_mm;
1220	BUG_ON(me->mm);
1221	enter_lazy_tlb(&init_mm, me);
1222
1223	load_sp0(t, &current->thread);
1224	set_tss_desc(cpu, t);
1225	load_TR_desc();
1226	load_LDT(&init_mm.context);
1227
1228	clear_all_debug_regs();
1229	dbg_restore_debug_regs();
1230
1231	fpu_init();
1232	xsave_init();
1233
1234	raw_local_save_flags(kernel_eflags);
1235
1236	if (is_uv_system())
1237		uv_cpu_init();
1238}
1239
1240#else
1241
1242void __cpuinit cpu_init(void)
1243{
1244	int cpu = smp_processor_id();
1245	struct task_struct *curr = current;
1246	struct tss_struct *t = &per_cpu(init_tss, cpu);
1247	struct thread_struct *thread = &curr->thread;
1248
1249	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1250		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1251		for (;;)
1252			local_irq_enable();
1253	}
1254
1255	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
 
 
 
 
1256
1257	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1258		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1259
1260	load_idt(&idt_descr);
 
 
 
 
 
 
 
1261	switch_to_new_gdt(cpu);
1262
1263	/*
1264	 * Set up and load the per-CPU TSS and LDT
1265	 */
1266	atomic_inc(&init_mm.mm_count);
1267	curr->active_mm = &init_mm;
1268	BUG_ON(curr->mm);
1269	enter_lazy_tlb(&init_mm, curr);
1270
1271	load_sp0(t, thread);
1272	set_tss_desc(cpu, t);
1273	load_TR_desc();
1274	load_LDT(&init_mm.context);
1275
1276	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1277
1278#ifdef CONFIG_DOUBLEFAULT
1279	/* Set up doublefault TSS pointer in the GDT */
1280	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1281#endif
1282
1283	clear_all_debug_regs();
1284	dbg_restore_debug_regs();
1285
1286	fpu_init();
1287	xsave_init();
1288}
1289#endif
v4.10.11
   1#include <linux/bootmem.h>
   2#include <linux/linkage.h>
   3#include <linux/bitops.h>
   4#include <linux/kernel.h>
   5#include <linux/export.h>
   6#include <linux/percpu.h>
   7#include <linux/string.h>
   8#include <linux/ctype.h>
   9#include <linux/delay.h>
  10#include <linux/sched.h>
  11#include <linux/init.h>
  12#include <linux/kprobes.h>
  13#include <linux/kgdb.h>
  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>
  20#include <asm/mmu_context.h>
  21#include <asm/archrandom.h>
  22#include <asm/hypervisor.h>
  23#include <asm/processor.h>
  24#include <asm/tlbflush.h>
  25#include <asm/debugreg.h>
  26#include <asm/sections.h>
  27#include <asm/vsyscall.h>
  28#include <linux/topology.h>
  29#include <linux/cpumask.h>
  30#include <asm/pgtable.h>
  31#include <linux/atomic.h>
  32#include <asm/proto.h>
  33#include <asm/setup.h>
  34#include <asm/apic.h>
  35#include <asm/desc.h>
  36#include <asm/fpu/internal.h>
  37#include <asm/mtrr.h>
  38#include <linux/numa.h>
  39#include <asm/asm.h>
  40#include <asm/bugs.h>
  41#include <asm/cpu.h>
  42#include <asm/mce.h>
  43#include <asm/msr.h>
  44#include <asm/pat.h>
  45#include <asm/microcode.h>
  46#include <asm/microcode_intel.h>
  47
  48#ifdef CONFIG_X86_LOCAL_APIC
  49#include <asm/uv/uv.h>
  50#endif
  51
  52#include "cpu.h"
  53
  54/* all of these masks are initialized in setup_cpu_local_masks() */
  55cpumask_var_t cpu_initialized_mask;
  56cpumask_var_t cpu_callout_mask;
  57cpumask_var_t cpu_callin_mask;
  58
  59/* representing cpus for which sibling maps can be computed */
  60cpumask_var_t cpu_sibling_setup_mask;
  61
  62/* correctly size the local cpu masks */
  63void __init setup_cpu_local_masks(void)
  64{
  65	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
  66	alloc_bootmem_cpumask_var(&cpu_callin_mask);
  67	alloc_bootmem_cpumask_var(&cpu_callout_mask);
  68	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
  69}
  70
  71static void default_init(struct cpuinfo_x86 *c)
  72{
  73#ifdef CONFIG_X86_64
  74	cpu_detect_cache_sizes(c);
  75#else
  76	/* Not much we can do here... */
  77	/* Check if at least it has cpuid */
  78	if (c->cpuid_level == -1) {
  79		/* No cpuid. It must be an ancient CPU */
  80		if (c->x86 == 4)
  81			strcpy(c->x86_model_id, "486");
  82		else if (c->x86 == 3)
  83			strcpy(c->x86_model_id, "386");
  84	}
  85#endif
  86}
  87
  88static const struct cpu_dev default_cpu = {
  89	.c_init		= default_init,
  90	.c_vendor	= "Unknown",
  91	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
  92};
  93
  94static const struct cpu_dev *this_cpu = &default_cpu;
  95
  96DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
  97#ifdef CONFIG_X86_64
  98	/*
  99	 * We need valid kernel segments for data and code in long mode too
 100	 * IRET will check the segment types  kkeil 2000/10/28
 101	 * Also sysret mandates a special GDT layout
 102	 *
 103	 * TLS descriptors are currently at a different place compared to i386.
 104	 * Hopefully nobody expects them at a fixed place (Wine?)
 105	 */
 106	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
 107	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
 108	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
 109	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
 110	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
 111	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
 112#else
 113	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
 114	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
 115	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
 116	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
 117	/*
 118	 * Segments used for calling PnP BIOS have byte granularity.
 119	 * They code segments and data segments have fixed 64k limits,
 120	 * the transfer segment sizes are set at run time.
 121	 */
 122	/* 32-bit code */
 123	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
 124	/* 16-bit code */
 125	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
 126	/* 16-bit data */
 127	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
 128	/* 16-bit data */
 129	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
 130	/* 16-bit data */
 131	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
 132	/*
 133	 * The APM segments have byte granularity and their bases
 134	 * are set at run time.  All have 64k limits.
 135	 */
 136	/* 32-bit code */
 137	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
 138	/* 16-bit code */
 139	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
 140	/* data */
 141	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
 142
 143	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
 144	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
 145	GDT_STACK_CANARY_INIT
 146#endif
 147} };
 148EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
 149
 150static int __init x86_mpx_setup(char *s)
 151{
 152	/* require an exact match without trailing characters */
 153	if (strlen(s))
 154		return 0;
 155
 156	/* do not emit a message if the feature is not present */
 157	if (!boot_cpu_has(X86_FEATURE_MPX))
 158		return 1;
 159
 160	setup_clear_cpu_cap(X86_FEATURE_MPX);
 161	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
 162	return 1;
 163}
 164__setup("nompx", x86_mpx_setup);
 165
 166static int __init x86_noinvpcid_setup(char *s)
 167{
 168	/* noinvpcid doesn't accept parameters */
 169	if (s)
 170		return -EINVAL;
 171
 172	/* do not emit a message if the feature is not present */
 173	if (!boot_cpu_has(X86_FEATURE_INVPCID))
 174		return 0;
 175
 176	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
 177	pr_info("noinvpcid: INVPCID feature disabled\n");
 178	return 0;
 179}
 180early_param("noinvpcid", x86_noinvpcid_setup);
 181
 182#ifdef CONFIG_X86_32
 183static int cachesize_override = -1;
 184static int disable_x86_serial_nr = 1;
 185
 186static int __init cachesize_setup(char *str)
 187{
 188	get_option(&str, &cachesize_override);
 189	return 1;
 190}
 191__setup("cachesize=", cachesize_setup);
 192
 
 
 
 
 
 
 
 
 193static int __init x86_sep_setup(char *s)
 194{
 195	setup_clear_cpu_cap(X86_FEATURE_SEP);
 196	return 1;
 197}
 198__setup("nosep", x86_sep_setup);
 199
 200/* Standard macro to see if a specific flag is changeable */
 201static inline int flag_is_changeable_p(u32 flag)
 202{
 203	u32 f1, f2;
 204
 205	/*
 206	 * Cyrix and IDT cpus allow disabling of CPUID
 207	 * so the code below may return different results
 208	 * when it is executed before and after enabling
 209	 * the CPUID. Add "volatile" to not allow gcc to
 210	 * optimize the subsequent calls to this function.
 211	 */
 212	asm volatile ("pushfl		\n\t"
 213		      "pushfl		\n\t"
 214		      "popl %0		\n\t"
 215		      "movl %0, %1	\n\t"
 216		      "xorl %2, %0	\n\t"
 217		      "pushl %0		\n\t"
 218		      "popfl		\n\t"
 219		      "pushfl		\n\t"
 220		      "popl %0		\n\t"
 221		      "popfl		\n\t"
 222
 223		      : "=&r" (f1), "=&r" (f2)
 224		      : "ir" (flag));
 225
 226	return ((f1^f2) & flag) != 0;
 227}
 228
 229/* Probe for the CPUID instruction */
 230int have_cpuid_p(void)
 231{
 232	return flag_is_changeable_p(X86_EFLAGS_ID);
 233}
 234
 235static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
 236{
 237	unsigned long lo, hi;
 238
 239	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
 240		return;
 241
 242	/* Disable processor serial number: */
 243
 244	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
 245	lo |= 0x200000;
 246	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
 247
 248	pr_notice("CPU serial number disabled.\n");
 249	clear_cpu_cap(c, X86_FEATURE_PN);
 250
 251	/* Disabling the serial number may affect the cpuid level */
 252	c->cpuid_level = cpuid_eax(0);
 253}
 254
 255static int __init x86_serial_nr_setup(char *s)
 256{
 257	disable_x86_serial_nr = 0;
 258	return 1;
 259}
 260__setup("serialnumber", x86_serial_nr_setup);
 261#else
 262static inline int flag_is_changeable_p(u32 flag)
 263{
 264	return 1;
 265}
 
 
 
 
 
 266static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
 267{
 268}
 269#endif
 270
 
 271static __init int setup_disable_smep(char *arg)
 272{
 273	setup_clear_cpu_cap(X86_FEATURE_SMEP);
 274	/* Check for things that depend on SMEP being enabled: */
 275	check_mpx_erratum(&boot_cpu_data);
 276	return 1;
 277}
 278__setup("nosmep", setup_disable_smep);
 279
 280static __always_inline void setup_smep(struct cpuinfo_x86 *c)
 281{
 282	if (cpu_has(c, X86_FEATURE_SMEP))
 283		cr4_set_bits(X86_CR4_SMEP);
 284}
 285
 286static __init int setup_disable_smap(char *arg)
 287{
 288	setup_clear_cpu_cap(X86_FEATURE_SMAP);
 289	return 1;
 290}
 291__setup("nosmap", setup_disable_smap);
 292
 293static __always_inline void setup_smap(struct cpuinfo_x86 *c)
 294{
 295	unsigned long eflags = native_save_fl();
 296
 297	/* This should have been cleared long ago */
 298	BUG_ON(eflags & X86_EFLAGS_AC);
 299
 300	if (cpu_has(c, X86_FEATURE_SMAP)) {
 301#ifdef CONFIG_X86_SMAP
 302		cr4_set_bits(X86_CR4_SMAP);
 303#else
 304		cr4_clear_bits(X86_CR4_SMAP);
 305#endif
 306	}
 307}
 308
 309/*
 310 * Protection Keys are not available in 32-bit mode.
 311 */
 312static bool pku_disabled;
 313
 314static __always_inline void setup_pku(struct cpuinfo_x86 *c)
 315{
 316	/* check the boot processor, plus compile options for PKU: */
 317	if (!cpu_feature_enabled(X86_FEATURE_PKU))
 318		return;
 319	/* checks the actual processor's cpuid bits: */
 320	if (!cpu_has(c, X86_FEATURE_PKU))
 321		return;
 322	if (pku_disabled)
 323		return;
 324
 325	cr4_set_bits(X86_CR4_PKE);
 326	/*
 327	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
 328	 * cpuid bit to be set.  We need to ensure that we
 329	 * update that bit in this CPU's "cpu_info".
 330	 */
 331	get_cpu_cap(c);
 332}
 333
 334#ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
 335static __init int setup_disable_pku(char *arg)
 336{
 337	/*
 338	 * Do not clear the X86_FEATURE_PKU bit.  All of the
 339	 * runtime checks are against OSPKE so clearing the
 340	 * bit does nothing.
 341	 *
 342	 * This way, we will see "pku" in cpuinfo, but not
 343	 * "ospke", which is exactly what we want.  It shows
 344	 * that the CPU has PKU, but the OS has not enabled it.
 345	 * This happens to be exactly how a system would look
 346	 * if we disabled the config option.
 347	 */
 348	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
 349	pku_disabled = true;
 350	return 1;
 351}
 352__setup("nopku", setup_disable_pku);
 353#endif /* CONFIG_X86_64 */
 354
 355/*
 356 * Some CPU features depend on higher CPUID levels, which may not always
 357 * be available due to CPUID level capping or broken virtualization
 358 * software.  Add those features to this table to auto-disable them.
 359 */
 360struct cpuid_dependent_feature {
 361	u32 feature;
 362	u32 level;
 363};
 364
 365static const struct cpuid_dependent_feature
 366cpuid_dependent_features[] = {
 367	{ X86_FEATURE_MWAIT,		0x00000005 },
 368	{ X86_FEATURE_DCA,		0x00000009 },
 369	{ X86_FEATURE_XSAVE,		0x0000000d },
 370	{ 0, 0 }
 371};
 372
 373static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
 374{
 375	const struct cpuid_dependent_feature *df;
 376
 377	for (df = cpuid_dependent_features; df->feature; df++) {
 378
 379		if (!cpu_has(c, df->feature))
 380			continue;
 381		/*
 382		 * Note: cpuid_level is set to -1 if unavailable, but
 383		 * extended_extended_level is set to 0 if unavailable
 384		 * and the legitimate extended levels are all negative
 385		 * when signed; hence the weird messing around with
 386		 * signs here...
 387		 */
 388		if (!((s32)df->level < 0 ?
 389		     (u32)df->level > (u32)c->extended_cpuid_level :
 390		     (s32)df->level > (s32)c->cpuid_level))
 391			continue;
 392
 393		clear_cpu_cap(c, df->feature);
 394		if (!warn)
 395			continue;
 396
 397		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
 398			x86_cap_flag(df->feature), df->level);
 
 399	}
 400}
 401
 402/*
 403 * Naming convention should be: <Name> [(<Codename>)]
 404 * This table only is used unless init_<vendor>() below doesn't set it;
 405 * in particular, if CPUID levels 0x80000002..4 are supported, this
 406 * isn't used
 407 */
 408
 409/* Look up CPU names by table lookup. */
 410static const char *table_lookup_model(struct cpuinfo_x86 *c)
 411{
 412#ifdef CONFIG_X86_32
 413	const struct legacy_cpu_model_info *info;
 414
 415	if (c->x86_model >= 16)
 416		return NULL;	/* Range check */
 417
 418	if (!this_cpu)
 419		return NULL;
 420
 421	info = this_cpu->legacy_models;
 422
 423	while (info->family) {
 424		if (info->family == c->x86)
 425			return info->model_names[c->x86_model];
 426		info++;
 427	}
 428#endif
 429	return NULL;		/* Not found */
 430}
 431
 432__u32 cpu_caps_cleared[NCAPINTS];
 433__u32 cpu_caps_set[NCAPINTS];
 434
 435void load_percpu_segment(int cpu)
 436{
 437#ifdef CONFIG_X86_32
 438	loadsegment(fs, __KERNEL_PERCPU);
 439#else
 440	__loadsegment_simple(gs, 0);
 441	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
 442#endif
 443	load_stack_canary_segment();
 444}
 445
 446/*
 447 * Current gdt points %fs at the "master" per-cpu area: after this,
 448 * it's on the real one.
 449 */
 450void switch_to_new_gdt(int cpu)
 451{
 452	struct desc_ptr gdt_descr;
 453
 454	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
 455	gdt_descr.size = GDT_SIZE - 1;
 456	load_gdt(&gdt_descr);
 457	/* Reload the per-cpu base */
 458
 459	load_percpu_segment(cpu);
 460}
 461
 462static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
 463
 464static void get_model_name(struct cpuinfo_x86 *c)
 465{
 466	unsigned int *v;
 467	char *p, *q, *s;
 468
 469	if (c->extended_cpuid_level < 0x80000004)
 470		return;
 471
 472	v = (unsigned int *)c->x86_model_id;
 473	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
 474	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
 475	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
 476	c->x86_model_id[48] = 0;
 477
 478	/* Trim whitespace */
 479	p = q = s = &c->x86_model_id[0];
 480
 
 
 481	while (*p == ' ')
 482		p++;
 483
 484	while (*p) {
 485		/* Note the last non-whitespace index */
 486		if (!isspace(*p))
 487			s = q;
 488
 489		*q++ = *p++;
 490	}
 491
 492	*(s + 1) = '\0';
 493}
 494
 495void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
 496{
 497	unsigned int n, dummy, ebx, ecx, edx, l2size;
 498
 499	n = c->extended_cpuid_level;
 500
 501	if (n >= 0x80000005) {
 502		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
 503		c->x86_cache_size = (ecx>>24) + (edx>>24);
 504#ifdef CONFIG_X86_64
 505		/* On K8 L1 TLB is inclusive, so don't count it */
 506		c->x86_tlbsize = 0;
 507#endif
 508	}
 509
 510	if (n < 0x80000006)	/* Some chips just has a large L1. */
 511		return;
 512
 513	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
 514	l2size = ecx >> 16;
 515
 516#ifdef CONFIG_X86_64
 517	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
 518#else
 519	/* do processor-specific cache resizing */
 520	if (this_cpu->legacy_cache_size)
 521		l2size = this_cpu->legacy_cache_size(c, l2size);
 522
 523	/* Allow user to override all this if necessary. */
 524	if (cachesize_override != -1)
 525		l2size = cachesize_override;
 526
 527	if (l2size == 0)
 528		return;		/* Again, no L2 cache is possible */
 529#endif
 530
 531	c->x86_cache_size = l2size;
 532}
 533
 534u16 __read_mostly tlb_lli_4k[NR_INFO];
 535u16 __read_mostly tlb_lli_2m[NR_INFO];
 536u16 __read_mostly tlb_lli_4m[NR_INFO];
 537u16 __read_mostly tlb_lld_4k[NR_INFO];
 538u16 __read_mostly tlb_lld_2m[NR_INFO];
 539u16 __read_mostly tlb_lld_4m[NR_INFO];
 540u16 __read_mostly tlb_lld_1g[NR_INFO];
 541
 542static void cpu_detect_tlb(struct cpuinfo_x86 *c)
 543{
 544	if (this_cpu->c_detect_tlb)
 545		this_cpu->c_detect_tlb(c);
 546
 547	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
 548		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
 549		tlb_lli_4m[ENTRIES]);
 550
 551	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
 552		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
 553		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
 554}
 555
 556void detect_ht(struct cpuinfo_x86 *c)
 557{
 558#ifdef CONFIG_SMP
 559	u32 eax, ebx, ecx, edx;
 560	int index_msb, core_bits;
 561	static bool printed;
 562
 563	if (!cpu_has(c, X86_FEATURE_HT))
 564		return;
 565
 566	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
 567		goto out;
 568
 569	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
 570		return;
 571
 572	cpuid(1, &eax, &ebx, &ecx, &edx);
 573
 574	smp_num_siblings = (ebx & 0xff0000) >> 16;
 575
 576	if (smp_num_siblings == 1) {
 577		pr_info_once("CPU0: Hyper-Threading is disabled\n");
 578		goto out;
 579	}
 580
 581	if (smp_num_siblings <= 1)
 582		goto out;
 583
 584	index_msb = get_count_order(smp_num_siblings);
 585	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
 586
 587	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
 588
 589	index_msb = get_count_order(smp_num_siblings);
 590
 591	core_bits = get_count_order(c->x86_max_cores);
 592
 593	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
 594				       ((1 << core_bits) - 1);
 595
 596out:
 597	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
 598		pr_info("CPU: Physical Processor ID: %d\n",
 599			c->phys_proc_id);
 600		pr_info("CPU: Processor Core ID: %d\n",
 601			c->cpu_core_id);
 602		printed = 1;
 603	}
 604#endif
 605}
 606
 607static void get_cpu_vendor(struct cpuinfo_x86 *c)
 608{
 609	char *v = c->x86_vendor_id;
 610	int i;
 611
 612	for (i = 0; i < X86_VENDOR_NUM; i++) {
 613		if (!cpu_devs[i])
 614			break;
 615
 616		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
 617		    (cpu_devs[i]->c_ident[1] &&
 618		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
 619
 620			this_cpu = cpu_devs[i];
 621			c->x86_vendor = this_cpu->c_x86_vendor;
 622			return;
 623		}
 624	}
 625
 626	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
 627		    "CPU: Your system may be unstable.\n", v);
 
 628
 629	c->x86_vendor = X86_VENDOR_UNKNOWN;
 630	this_cpu = &default_cpu;
 631}
 632
 633void cpu_detect(struct cpuinfo_x86 *c)
 634{
 635	/* Get vendor name */
 636	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
 637	      (unsigned int *)&c->x86_vendor_id[0],
 638	      (unsigned int *)&c->x86_vendor_id[8],
 639	      (unsigned int *)&c->x86_vendor_id[4]);
 640
 641	c->x86 = 4;
 642	/* Intel-defined flags: level 0x00000001 */
 643	if (c->cpuid_level >= 0x00000001) {
 644		u32 junk, tfms, cap0, misc;
 645
 646		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
 647		c->x86		= x86_family(tfms);
 648		c->x86_model	= x86_model(tfms);
 649		c->x86_mask	= x86_stepping(tfms);
 
 
 
 
 
 650
 651		if (cap0 & (1<<19)) {
 652			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
 653			c->x86_cache_alignment = c->x86_clflush_size;
 654		}
 655	}
 656}
 657
 658void get_cpu_cap(struct cpuinfo_x86 *c)
 659{
 660	u32 eax, ebx, ecx, edx;
 
 661
 662	/* Intel-defined flags: level 0x00000001 */
 663	if (c->cpuid_level >= 0x00000001) {
 664		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
 665
 666		c->x86_capability[CPUID_1_ECX] = ecx;
 667		c->x86_capability[CPUID_1_EDX] = edx;
 
 668	}
 669
 670	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
 671	if (c->cpuid_level >= 0x00000006)
 672		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
 673
 674	/* Additional Intel-defined flags: level 0x00000007 */
 675	if (c->cpuid_level >= 0x00000007) {
 
 
 676		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
 677		c->x86_capability[CPUID_7_0_EBX] = ebx;
 678		c->x86_capability[CPUID_7_ECX] = ecx;
 679	}
 680
 681	/* Extended state features: level 0x0000000d */
 682	if (c->cpuid_level >= 0x0000000d) {
 683		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
 684
 685		c->x86_capability[CPUID_D_1_EAX] = eax;
 686	}
 687
 688	/* Additional Intel-defined flags: level 0x0000000F */
 689	if (c->cpuid_level >= 0x0000000F) {
 690
 691		/* QoS sub-leaf, EAX=0Fh, ECX=0 */
 692		cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
 693		c->x86_capability[CPUID_F_0_EDX] = edx;
 694
 695		if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
 696			/* will be overridden if occupancy monitoring exists */
 697			c->x86_cache_max_rmid = ebx;
 698
 699			/* QoS sub-leaf, EAX=0Fh, ECX=1 */
 700			cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
 701			c->x86_capability[CPUID_F_1_EDX] = edx;
 702
 703			if ((cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) ||
 704			      ((cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL)) ||
 705			       (cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)))) {
 706				c->x86_cache_max_rmid = ecx;
 707				c->x86_cache_occ_scale = ebx;
 708			}
 709		} else {
 710			c->x86_cache_max_rmid = -1;
 711			c->x86_cache_occ_scale = -1;
 712		}
 713	}
 714
 715	/* AMD-defined flags: level 0x80000001 */
 716	eax = cpuid_eax(0x80000000);
 717	c->extended_cpuid_level = eax;
 718
 719	if ((eax & 0xffff0000) == 0x80000000) {
 720		if (eax >= 0x80000001) {
 721			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
 722
 723			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
 724			c->x86_capability[CPUID_8000_0001_EDX] = edx;
 
 
 725		}
 726	}
 727
 728	if (c->extended_cpuid_level >= 0x80000007) {
 729		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
 730
 731		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
 732		c->x86_power = edx;
 733	}
 734
 735	if (c->extended_cpuid_level >= 0x80000008) {
 736		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
 737
 738		c->x86_virt_bits = (eax >> 8) & 0xff;
 739		c->x86_phys_bits = eax & 0xff;
 740		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
 741	}
 742#ifdef CONFIG_X86_32
 743	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
 744		c->x86_phys_bits = 36;
 745#endif
 746
 747	if (c->extended_cpuid_level >= 0x8000000a)
 748		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
 749
 750	init_scattered_cpuid_features(c);
 751}
 752
 753static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
 754{
 755#ifdef CONFIG_X86_32
 756	int i;
 757
 758	/*
 759	 * First of all, decide if this is a 486 or higher
 760	 * It's a 486 if we can modify the AC flag
 761	 */
 762	if (flag_is_changeable_p(X86_EFLAGS_AC))
 763		c->x86 = 4;
 764	else
 765		c->x86 = 3;
 766
 767	for (i = 0; i < X86_VENDOR_NUM; i++)
 768		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
 769			c->x86_vendor_id[0] = 0;
 770			cpu_devs[i]->c_identify(c);
 771			if (c->x86_vendor_id[0]) {
 772				get_cpu_vendor(c);
 773				break;
 774			}
 775		}
 776#endif
 777}
 778
 779/*
 780 * Do minimum CPU detection early.
 781 * Fields really needed: vendor, cpuid_level, family, model, mask,
 782 * cache alignment.
 783 * The others are not touched to avoid unwanted side effects.
 784 *
 785 * WARNING: this function is only called on the BP.  Don't add code here
 786 * that is supposed to run on all CPUs.
 787 */
 788static void __init early_identify_cpu(struct cpuinfo_x86 *c)
 789{
 790#ifdef CONFIG_X86_64
 791	c->x86_clflush_size = 64;
 792	c->x86_phys_bits = 36;
 793	c->x86_virt_bits = 48;
 794#else
 795	c->x86_clflush_size = 32;
 796	c->x86_phys_bits = 32;
 797	c->x86_virt_bits = 32;
 798#endif
 799	c->x86_cache_alignment = c->x86_clflush_size;
 800
 801	memset(&c->x86_capability, 0, sizeof c->x86_capability);
 802	c->extended_cpuid_level = 0;
 803
 804	if (!have_cpuid_p())
 805		identify_cpu_without_cpuid(c);
 806
 807	/* cyrix could have cpuid enabled via c_identify()*/
 808	if (have_cpuid_p()) {
 809		cpu_detect(c);
 810		get_cpu_vendor(c);
 811		get_cpu_cap(c);
 812
 813		if (this_cpu->c_early_init)
 814			this_cpu->c_early_init(c);
 
 815
 816		c->cpu_index = 0;
 817		filter_cpuid_features(c, false);
 818
 819		if (this_cpu->c_bsp_init)
 820			this_cpu->c_bsp_init(c);
 821	}
 
 
 
 
 822
 823	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
 824	fpu__init_system(c);
 825}
 826
 827void __init early_cpu_init(void)
 828{
 829	const struct cpu_dev *const *cdev;
 830	int count = 0;
 831
 832#ifdef CONFIG_PROCESSOR_SELECT
 833	pr_info("KERNEL supported cpus:\n");
 834#endif
 835
 836	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
 837		const struct cpu_dev *cpudev = *cdev;
 838
 839		if (count >= X86_VENDOR_NUM)
 840			break;
 841		cpu_devs[count] = cpudev;
 842		count++;
 843
 844#ifdef CONFIG_PROCESSOR_SELECT
 845		{
 846			unsigned int j;
 847
 848			for (j = 0; j < 2; j++) {
 849				if (!cpudev->c_ident[j])
 850					continue;
 851				pr_info("  %s %s\n", cpudev->c_vendor,
 852					cpudev->c_ident[j]);
 853			}
 854		}
 855#endif
 856	}
 857	early_identify_cpu(&boot_cpu_data);
 858}
 859
 860/*
 861 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
 862 * unfortunately, that's not true in practice because of early VIA
 863 * chips and (more importantly) broken virtualizers that are not easy
 864 * to detect. In the latter case it doesn't even *fail* reliably, so
 865 * probing for it doesn't even work. Disable it completely on 32-bit
 866 * unless we can find a reliable way to detect all the broken cases.
 867 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
 868 */
 869static void detect_nopl(struct cpuinfo_x86 *c)
 870{
 871#ifdef CONFIG_X86_32
 872	clear_cpu_cap(c, X86_FEATURE_NOPL);
 873#else
 874	set_cpu_cap(c, X86_FEATURE_NOPL);
 875#endif
 876}
 877
 878static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
 879{
 880#ifdef CONFIG_X86_64
 881	/*
 882	 * Empirically, writing zero to a segment selector on AMD does
 883	 * not clear the base, whereas writing zero to a segment
 884	 * selector on Intel does clear the base.  Intel's behavior
 885	 * allows slightly faster context switches in the common case
 886	 * where GS is unused by the prev and next threads.
 887	 *
 888	 * Since neither vendor documents this anywhere that I can see,
 889	 * detect it directly instead of hardcoding the choice by
 890	 * vendor.
 891	 *
 892	 * I've designated AMD's behavior as the "bug" because it's
 893	 * counterintuitive and less friendly.
 894	 */
 895
 896	unsigned long old_base, tmp;
 897	rdmsrl(MSR_FS_BASE, old_base);
 898	wrmsrl(MSR_FS_BASE, 1);
 899	loadsegment(fs, 0);
 900	rdmsrl(MSR_FS_BASE, tmp);
 901	if (tmp != 0)
 902		set_cpu_bug(c, X86_BUG_NULL_SEG);
 903	wrmsrl(MSR_FS_BASE, old_base);
 904#endif
 905}
 906
 907static void generic_identify(struct cpuinfo_x86 *c)
 908{
 909	c->extended_cpuid_level = 0;
 910
 911	if (!have_cpuid_p())
 912		identify_cpu_without_cpuid(c);
 913
 914	/* cyrix could have cpuid enabled via c_identify()*/
 915	if (!have_cpuid_p())
 916		return;
 917
 918	cpu_detect(c);
 919
 920	get_cpu_vendor(c);
 921
 922	get_cpu_cap(c);
 923
 924	if (c->cpuid_level >= 0x00000001) {
 925		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
 926#ifdef CONFIG_X86_32
 927# ifdef CONFIG_SMP
 928		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
 929# else
 930		c->apicid = c->initial_apicid;
 931# endif
 932#endif
 
 
 933		c->phys_proc_id = c->initial_apicid;
 
 934	}
 935
 
 
 936	get_model_name(c); /* Default name */
 937
 938	detect_nopl(c);
 939
 940	detect_null_seg_behavior(c);
 941
 942	/*
 943	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
 944	 * systems that run Linux at CPL > 0 may or may not have the
 945	 * issue, but, even if they have the issue, there's absolutely
 946	 * nothing we can do about it because we can't use the real IRET
 947	 * instruction.
 948	 *
 949	 * NB: For the time being, only 32-bit kernels support
 950	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
 951	 * whether to apply espfix using paravirt hooks.  If any
 952	 * non-paravirt system ever shows up that does *not* have the
 953	 * ESPFIX issue, we can change this.
 954	 */
 955#ifdef CONFIG_X86_32
 956# ifdef CONFIG_PARAVIRT
 957	do {
 958		extern void native_iret(void);
 959		if (pv_cpu_ops.iret == native_iret)
 960			set_cpu_bug(c, X86_BUG_ESPFIX);
 961	} while (0);
 962# else
 963	set_cpu_bug(c, X86_BUG_ESPFIX);
 964# endif
 965#endif
 966}
 967
 968static void x86_init_cache_qos(struct cpuinfo_x86 *c)
 969{
 970	/*
 971	 * The heavy lifting of max_rmid and cache_occ_scale are handled
 972	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
 973	 * in case CQM bits really aren't there in this CPU.
 974	 */
 975	if (c != &boot_cpu_data) {
 976		boot_cpu_data.x86_cache_max_rmid =
 977			min(boot_cpu_data.x86_cache_max_rmid,
 978			    c->x86_cache_max_rmid);
 979	}
 980}
 981
 982/*
 983 * Validate that ACPI/mptables have the same information about the
 984 * effective APIC id and update the package map.
 985 */
 986static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
 987{
 988#ifdef CONFIG_SMP
 989	unsigned int apicid, cpu = smp_processor_id();
 990
 991	apicid = apic->cpu_present_to_apicid(cpu);
 992
 993	if (apicid != c->apicid) {
 994		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
 995		       cpu, apicid, c->initial_apicid);
 996	}
 997	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
 998#else
 999	c->logical_proc_id = 0;
1000#endif
1001}
1002
1003/*
1004 * This does the hard work of actually picking apart the CPU stuff...
1005 */
1006static void identify_cpu(struct cpuinfo_x86 *c)
1007{
1008	int i;
1009
1010	c->loops_per_jiffy = loops_per_jiffy;
1011	c->x86_cache_size = -1;
1012	c->x86_vendor = X86_VENDOR_UNKNOWN;
1013	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
1014	c->x86_vendor_id[0] = '\0'; /* Unset */
1015	c->x86_model_id[0] = '\0';  /* Unset */
1016	c->x86_max_cores = 1;
1017	c->x86_coreid_bits = 0;
1018	c->cu_id = 0xff;
1019#ifdef CONFIG_X86_64
1020	c->x86_clflush_size = 64;
1021	c->x86_phys_bits = 36;
1022	c->x86_virt_bits = 48;
1023#else
1024	c->cpuid_level = -1;	/* CPUID not detected */
1025	c->x86_clflush_size = 32;
1026	c->x86_phys_bits = 32;
1027	c->x86_virt_bits = 32;
1028#endif
1029	c->x86_cache_alignment = c->x86_clflush_size;
1030	memset(&c->x86_capability, 0, sizeof c->x86_capability);
1031
1032	generic_identify(c);
1033
1034	if (this_cpu->c_identify)
1035		this_cpu->c_identify(c);
1036
1037	/* Clear/Set all flags overridden by options, after probe */
1038	for (i = 0; i < NCAPINTS; i++) {
1039		c->x86_capability[i] &= ~cpu_caps_cleared[i];
1040		c->x86_capability[i] |= cpu_caps_set[i];
1041	}
1042
1043#ifdef CONFIG_X86_64
1044	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1045#endif
1046
1047	/*
1048	 * Vendor-specific initialization.  In this section we
1049	 * canonicalize the feature flags, meaning if there are
1050	 * features a certain CPU supports which CPUID doesn't
1051	 * tell us, CPUID claiming incorrect flags, or other bugs,
1052	 * we handle them here.
1053	 *
1054	 * At the end of this section, c->x86_capability better
1055	 * indicate the features this CPU genuinely supports!
1056	 */
1057	if (this_cpu->c_init)
1058		this_cpu->c_init(c);
1059
1060	/* Disable the PN if appropriate */
1061	squash_the_stupid_serial_number(c);
1062
1063	/* Set up SMEP/SMAP */
1064	setup_smep(c);
1065	setup_smap(c);
1066
1067	/*
1068	 * The vendor-specific functions might have changed features.
1069	 * Now we do "generic changes."
1070	 */
1071
1072	/* Filter out anything that depends on CPUID levels we don't have */
1073	filter_cpuid_features(c, true);
1074
1075	/* If the model name is still unset, do table lookup. */
1076	if (!c->x86_model_id[0]) {
1077		const char *p;
1078		p = table_lookup_model(c);
1079		if (p)
1080			strcpy(c->x86_model_id, p);
1081		else
1082			/* Last resort... */
1083			sprintf(c->x86_model_id, "%02x/%02x",
1084				c->x86, c->x86_model);
1085	}
1086
1087#ifdef CONFIG_X86_64
1088	detect_ht(c);
1089#endif
1090
1091	init_hypervisor(c);
1092	x86_init_rdrand(c);
1093	x86_init_cache_qos(c);
1094	setup_pku(c);
1095
1096	/*
1097	 * Clear/Set all flags overridden by options, need do it
1098	 * before following smp all cpus cap AND.
1099	 */
1100	for (i = 0; i < NCAPINTS; i++) {
1101		c->x86_capability[i] &= ~cpu_caps_cleared[i];
1102		c->x86_capability[i] |= cpu_caps_set[i];
1103	}
1104
1105	/*
1106	 * On SMP, boot_cpu_data holds the common feature set between
1107	 * all CPUs; so make sure that we indicate which features are
1108	 * common between the CPUs.  The first time this routine gets
1109	 * executed, c == &boot_cpu_data.
1110	 */
1111	if (c != &boot_cpu_data) {
1112		/* AND the already accumulated flags with these */
1113		for (i = 0; i < NCAPINTS; i++)
1114			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1115
1116		/* OR, i.e. replicate the bug flags */
1117		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1118			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1119	}
1120
1121	/* Init Machine Check Exception if available. */
1122	mcheck_cpu_init(c);
1123
1124	select_idle_routine(c);
1125
1126#ifdef CONFIG_NUMA
1127	numa_add_cpu(smp_processor_id());
1128#endif
1129}
1130
1131/*
1132 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1133 * on 32-bit kernels:
1134 */
1135#ifdef CONFIG_X86_32
1136void enable_sep_cpu(void)
1137{
1138	struct tss_struct *tss;
1139	int cpu;
1140
1141	if (!boot_cpu_has(X86_FEATURE_SEP))
1142		return;
1143
1144	cpu = get_cpu();
1145	tss = &per_cpu(cpu_tss, cpu);
1146
1147	/*
1148	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1149	 * see the big comment in struct x86_hw_tss's definition.
1150	 */
1151
1152	tss->x86_tss.ss1 = __KERNEL_CS;
1153	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1154
1155	wrmsr(MSR_IA32_SYSENTER_ESP,
1156	      (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
1157	      0);
1158
1159	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1160
1161	put_cpu();
1162}
1163#endif
1164
1165void __init identify_boot_cpu(void)
1166{
1167	identify_cpu(&boot_cpu_data);
 
1168#ifdef CONFIG_X86_32
1169	sysenter_setup();
1170	enable_sep_cpu();
 
 
1171#endif
1172	cpu_detect_tlb(&boot_cpu_data);
1173}
1174
1175void identify_secondary_cpu(struct cpuinfo_x86 *c)
1176{
1177	BUG_ON(c == &boot_cpu_data);
1178	identify_cpu(c);
1179#ifdef CONFIG_X86_32
1180	enable_sep_cpu();
1181#endif
1182	mtrr_ap_init();
1183	validate_apic_and_package_id(c);
1184}
1185
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1186static __init int setup_noclflush(char *arg)
1187{
1188	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1189	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1190	return 1;
1191}
1192__setup("noclflush", setup_noclflush);
1193
1194void print_cpu_info(struct cpuinfo_x86 *c)
1195{
1196	const char *vendor = NULL;
1197
1198	if (c->x86_vendor < X86_VENDOR_NUM) {
1199		vendor = this_cpu->c_vendor;
1200	} else {
1201		if (c->cpuid_level >= 0)
1202			vendor = c->x86_vendor_id;
1203	}
1204
1205	if (vendor && !strstr(c->x86_model_id, vendor))
1206		pr_cont("%s ", vendor);
1207
1208	if (c->x86_model_id[0])
1209		pr_cont("%s", c->x86_model_id);
1210	else
1211		pr_cont("%d86", c->x86);
1212
1213	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1214
1215	if (c->x86_mask || c->cpuid_level >= 0)
1216		pr_cont(", stepping: 0x%x)\n", c->x86_mask);
1217	else
1218		pr_cont(")\n");
 
 
 
 
 
 
 
 
1219}
1220
1221static __init int setup_disablecpuid(char *arg)
1222{
1223	int bit;
1224
1225	if (get_option(&arg, &bit) && bit >= 0 && bit < NCAPINTS * 32)
1226		setup_clear_cpu_cap(bit);
1227	else
1228		return 0;
1229
1230	return 1;
1231}
1232__setup("clearcpuid=", setup_disablecpuid);
1233
1234#ifdef CONFIG_X86_64
1235struct desc_ptr idt_descr __ro_after_init = {
1236	.size = NR_VECTORS * 16 - 1,
1237	.address = (unsigned long) idt_table,
1238};
1239const struct desc_ptr debug_idt_descr = {
1240	.size = NR_VECTORS * 16 - 1,
1241	.address = (unsigned long) debug_idt_table,
1242};
1243
1244DEFINE_PER_CPU_FIRST(union irq_stack_union,
1245		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
1246
1247/*
1248 * The following percpu variables are hot.  Align current_task to
1249 * cacheline size such that they fall in the same cacheline.
1250 */
1251DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1252	&init_task;
1253EXPORT_PER_CPU_SYMBOL(current_task);
1254
 
 
 
 
1255DEFINE_PER_CPU(char *, irq_stack_ptr) =
1256	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE;
1257
1258DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1259
1260DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1261EXPORT_PER_CPU_SYMBOL(__preempt_count);
1262
1263/*
1264 * Special IST stacks which the CPU switches to when it calls
1265 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1266 * limit), all of them are 4K, except the debug stack which
1267 * is 8K.
1268 */
1269static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1270	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1271	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1272};
1273
1274static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1275	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1276
1277/* May not be marked __init: used by software suspend */
1278void syscall_init(void)
1279{
1280	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1281	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
 
 
 
 
 
 
1282
1283#ifdef CONFIG_IA32_EMULATION
1284	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1285	/*
1286	 * This only works on Intel CPUs.
1287	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1288	 * This does not cause SYSENTER to jump to the wrong location, because
1289	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1290	 */
1291	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1292	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1293	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1294#else
1295	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1296	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1297	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1298	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1299#endif
1300
1301	/* Flags to clear on syscall */
1302	wrmsrl(MSR_SYSCALL_MASK,
1303	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1304	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1305}
1306
 
 
1307/*
1308 * Copies of the original ist values from the tss are only accessed during
1309 * debugging, no special alignment required.
1310 */
1311DEFINE_PER_CPU(struct orig_ist, orig_ist);
1312
1313static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1314DEFINE_PER_CPU(int, debug_stack_usage);
1315
1316int is_debug_stack(unsigned long addr)
1317{
1318	return __this_cpu_read(debug_stack_usage) ||
1319		(addr <= __this_cpu_read(debug_stack_addr) &&
1320		 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
1321}
1322NOKPROBE_SYMBOL(is_debug_stack);
1323
1324DEFINE_PER_CPU(u32, debug_idt_ctr);
1325
1326void debug_stack_set_zero(void)
1327{
1328	this_cpu_inc(debug_idt_ctr);
1329	load_current_idt();
1330}
1331NOKPROBE_SYMBOL(debug_stack_set_zero);
1332
1333void debug_stack_reset(void)
1334{
1335	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1336		return;
1337	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1338		load_current_idt();
1339}
1340NOKPROBE_SYMBOL(debug_stack_reset);
1341
1342#else	/* CONFIG_X86_64 */
1343
1344DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1345EXPORT_PER_CPU_SYMBOL(current_task);
1346DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1347EXPORT_PER_CPU_SYMBOL(__preempt_count);
1348
1349/*
1350 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1351 * the top of the kernel stack.  Use an extra percpu variable to track the
1352 * top of the kernel stack directly.
1353 */
1354DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1355	(unsigned long)&init_thread_union + THREAD_SIZE;
1356EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1357
1358#ifdef CONFIG_CC_STACKPROTECTOR
1359DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1360#endif
1361
 
 
 
 
 
 
 
 
 
1362#endif	/* CONFIG_X86_64 */
1363
1364/*
1365 * Clear all 6 debug registers:
1366 */
1367static void clear_all_debug_regs(void)
1368{
1369	int i;
1370
1371	for (i = 0; i < 8; i++) {
1372		/* Ignore db4, db5 */
1373		if ((i == 4) || (i == 5))
1374			continue;
1375
1376		set_debugreg(0, i);
1377	}
1378}
1379
1380#ifdef CONFIG_KGDB
1381/*
1382 * Restore debug regs if using kgdbwait and you have a kernel debugger
1383 * connection established.
1384 */
1385static void dbg_restore_debug_regs(void)
1386{
1387	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1388		arch_kgdb_ops.correct_hw_break();
1389}
1390#else /* ! CONFIG_KGDB */
1391#define dbg_restore_debug_regs()
1392#endif /* ! CONFIG_KGDB */
1393
1394static void wait_for_master_cpu(int cpu)
1395{
1396#ifdef CONFIG_SMP
1397	/*
1398	 * wait for ACK from master CPU before continuing
1399	 * with AP initialization
1400	 */
1401	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1402	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1403		cpu_relax();
1404#endif
1405}
1406
1407/*
1408 * cpu_init() initializes state that is per-CPU. Some data is already
1409 * initialized (naturally) in the bootstrap process, such as the GDT
1410 * and IDT. We reload them nevertheless, this function acts as a
1411 * 'CPU state barrier', nothing should get across.
1412 * A lot of state is already set up in PDA init for 64 bit
1413 */
1414#ifdef CONFIG_X86_64
1415
1416void cpu_init(void)
1417{
1418	struct orig_ist *oist;
1419	struct task_struct *me;
1420	struct tss_struct *t;
1421	unsigned long v;
1422	int cpu = raw_smp_processor_id();
1423	int i;
1424
1425	wait_for_master_cpu(cpu);
1426
1427	/*
1428	 * Initialize the CR4 shadow before doing anything that could
1429	 * try to read it.
1430	 */
1431	cr4_init_shadow();
1432
1433	if (cpu)
1434		load_ucode_ap();
1435
1436	t = &per_cpu(cpu_tss, cpu);
1437	oist = &per_cpu(orig_ist, cpu);
1438
1439#ifdef CONFIG_NUMA
1440	if (this_cpu_read(numa_node) == 0 &&
1441	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1442		set_numa_node(early_cpu_to_node(cpu));
1443#endif
1444
1445	me = current;
1446
 
 
 
1447	pr_debug("Initializing CPU#%d\n", cpu);
1448
1449	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1450
1451	/*
1452	 * Initialize the per-CPU GDT with the boot GDT,
1453	 * and set up the GDT descriptor:
1454	 */
1455
1456	switch_to_new_gdt(cpu);
1457	loadsegment(fs, 0);
1458
1459	load_current_idt();
1460
1461	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1462	syscall_init();
1463
1464	wrmsrl(MSR_FS_BASE, 0);
1465	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1466	barrier();
1467
1468	x86_configure_nx();
1469	x2apic_setup();
 
1470
1471	/*
1472	 * set up and load the per-CPU TSS
1473	 */
1474	if (!oist->ist[0]) {
1475		char *estacks = per_cpu(exception_stacks, cpu);
1476
1477		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1478			estacks += exception_stack_sizes[v];
1479			oist->ist[v] = t->x86_tss.ist[v] =
1480					(unsigned long)estacks;
1481			if (v == DEBUG_STACK-1)
1482				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1483		}
1484	}
1485
1486	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1487
1488	/*
1489	 * <= is required because the CPU will access up to
1490	 * 8 bits beyond the end of the IO permission bitmap.
1491	 */
1492	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1493		t->io_bitmap[i] = ~0UL;
1494
1495	atomic_inc(&init_mm.mm_count);
1496	me->active_mm = &init_mm;
1497	BUG_ON(me->mm);
1498	enter_lazy_tlb(&init_mm, me);
1499
1500	load_sp0(t, &current->thread);
1501	set_tss_desc(cpu, t);
1502	load_TR_desc();
1503	load_mm_ldt(&init_mm);
1504
1505	clear_all_debug_regs();
1506	dbg_restore_debug_regs();
1507
1508	fpu__init_cpu();
 
 
 
1509
1510	if (is_uv_system())
1511		uv_cpu_init();
1512}
1513
1514#else
1515
1516void cpu_init(void)
1517{
1518	int cpu = smp_processor_id();
1519	struct task_struct *curr = current;
1520	struct tss_struct *t = &per_cpu(cpu_tss, cpu);
1521	struct thread_struct *thread = &curr->thread;
1522
1523	wait_for_master_cpu(cpu);
 
 
 
 
1524
1525	/*
1526	 * Initialize the CR4 shadow before doing anything that could
1527	 * try to read it.
1528	 */
1529	cr4_init_shadow();
1530
1531	show_ucode_info_early();
 
1532
1533	pr_info("Initializing CPU#%d\n", cpu);
1534
1535	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1536	    boot_cpu_has(X86_FEATURE_TSC) ||
1537	    boot_cpu_has(X86_FEATURE_DE))
1538		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1539
1540	load_current_idt();
1541	switch_to_new_gdt(cpu);
1542
1543	/*
1544	 * Set up and load the per-CPU TSS and LDT
1545	 */
1546	atomic_inc(&init_mm.mm_count);
1547	curr->active_mm = &init_mm;
1548	BUG_ON(curr->mm);
1549	enter_lazy_tlb(&init_mm, curr);
1550
1551	load_sp0(t, thread);
1552	set_tss_desc(cpu, t);
1553	load_TR_desc();
1554	load_mm_ldt(&init_mm);
1555
1556	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1557
1558#ifdef CONFIG_DOUBLEFAULT
1559	/* Set up doublefault TSS pointer in the GDT */
1560	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1561#endif
1562
1563	clear_all_debug_regs();
1564	dbg_restore_debug_regs();
1565
1566	fpu__init_cpu();
 
1567}
1568#endif
1569
1570static void bsp_resume(void)
1571{
1572	if (this_cpu->c_bsp_resume)
1573		this_cpu->c_bsp_resume(&boot_cpu_data);
1574}
1575
1576static struct syscore_ops cpu_syscore_ops = {
1577	.resume		= bsp_resume,
1578};
1579
1580static int __init init_cpu_syscore(void)
1581{
1582	register_syscore_ops(&cpu_syscore_ops);
1583	return 0;
1584}
1585core_initcall(init_cpu_syscore);