arch/x86/include/asm/processor.h

   1 #ifndef _ASM_X86_PROCESSOR_H
   2 #define _ASM_X86_PROCESSOR_H
   3
   4 #include <asm/processor-flags.h>
   5
   6 /* Forward declaration, a strange C thing */
   7 struct task_struct;
   8 struct mm_struct;
   9
  10 #include <asm/vm86.h>
  11 #include <asm/math_emu.h>
  12 #include <asm/segment.h>
  13 #include <asm/types.h>
  14 #include <asm/sigcontext.h>
  15 #include <asm/current.h>
  16 #include <asm/cpufeature.h>
  17 #include <asm/system.h>
  18 #include <asm/page.h>
  19 #include <asm/pgtable_types.h>
  20 #include <asm/percpu.h>
  21 #include <asm/msr.h>
  22 #include <asm/desc_defs.h>
  23 #include <asm/nops.h>
  24 #include <asm/ds.h>
  25
  26 #include <linux/personality.h>
  27 #include <linux/cpumask.h>
  28 #include <linux/cache.h>
  29 #include <linux/threads.h>
  30 #include <linux/init.h>
  31
  32 /*
  33  * Default implementation of macro that returns current
  34  * instruction pointer ("program counter").
  35  */
  36 static inline void *current_text_addr(void)
  37 {
  38         void *pc;
  39
  40         asm volatile("mov $1f, %0; 1:":"=r" (pc));
  41
  42         return pc;
  43 }
  44
  45 #ifdef CONFIG_X86_VSMP
  46 # define ARCH_MIN_TASKALIGN             (1 << INTERNODE_CACHE_SHIFT)
  47 # define ARCH_MIN_MMSTRUCT_ALIGN        (1 << INTERNODE_CACHE_SHIFT)
  48 #else
  49 # define ARCH_MIN_TASKALIGN             16
  50 # define ARCH_MIN_MMSTRUCT_ALIGN        0
  51 #endif
  52
  53 /*
  54  *  CPU type and hardware bug flags. Kept separately for each CPU.
  55  *  Members of this structure are referenced in head.S, so think twice
  56  *  before touching them. [mj]
  57  */
  58
  59 struct cpuinfo_x86 {
  60         __u8                    x86;            /* CPU family */
  61         __u8                    x86_vendor;     /* CPU vendor */
  62         __u8                    x86_model;
  63         __u8                    x86_mask;
  64 #ifdef CONFIG_X86_32
  65         char                    wp_works_ok;    /* It doesn't on 386's */
  66
  67         /* Problems on some 486Dx4's and old 386's: */
  68         char                    hlt_works_ok;
  69         char                    hard_math;
  70         char                    rfu;
  71         char                    fdiv_bug;
  72         char                    f00f_bug;
  73         char                    coma_bug;
  74         char                    pad0;
  75 #else
  76         /* Number of 4K pages in DTLB/ITLB combined(in pages): */
  77         int                     x86_tlbsize;
  78         __u8                    x86_virt_bits;
  79         __u8                    x86_phys_bits;
  80 #endif
  81         /* CPUID returned core id bits: */
  82         __u8                    x86_coreid_bits;
  83         /* Max extended CPUID function supported: */
  84         __u32                   extended_cpuid_level;
  85         /* Maximum supported CPUID level, -1=no CPUID: */
  86         int                     cpuid_level;
  87         __u32                   x86_capability[NCAPINTS];
  88         char                    x86_vendor_id[16];
  89         char                    x86_model_id[64];
  90         /* in KB - valid for CPUS which support this call: */
  91         int                     x86_cache_size;
  92         int                     x86_cache_alignment;    /* In bytes */
  93         int                     x86_power;
  94         unsigned long           loops_per_jiffy;
  95 #ifdef CONFIG_SMP
  96         /* cpus sharing the last level cache: */
  97         cpumask_t               llc_shared_map;
  98 #endif
  99         /* cpuid returned max cores value: */
 100         u16                      x86_max_cores;
 101         u16                     apicid;
 102         u16                     initial_apicid;
 103         u16                     x86_clflush_size;
 104 #ifdef CONFIG_SMP
 105         /* number of cores as seen by the OS: */
 106         u16                     booted_cores;
 107         /* Physical processor id: */
 108         u16                     phys_proc_id;
 109         /* Core id: */
 110         u16                     cpu_core_id;
 111         /* Index into per_cpu list: */
 112         u16                     cpu_index;
 113 #endif
 114         unsigned int            x86_hyper_vendor;
 115 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
 116
 117 #define X86_VENDOR_INTEL        0
 118 #define X86_VENDOR_CYRIX        1
 119 #define X86_VENDOR_AMD          2
 120 #define X86_VENDOR_UMC          3
 121 #define X86_VENDOR_CENTAUR      5
 122 #define X86_VENDOR_TRANSMETA    7
 123 #define X86_VENDOR_NSC          8
 124 #define X86_VENDOR_NUM          9
 125
 126 #define X86_VENDOR_UNKNOWN      0xff
 127
 128 #define X86_HYPER_VENDOR_NONE  0
 129 #define X86_HYPER_VENDOR_VMWARE 1
 130
 131 /*
 132  * capabilities of CPUs
 133  */
 134 extern struct cpuinfo_x86       boot_cpu_data;
 135 extern struct cpuinfo_x86       new_cpu_data;
 136
 137 extern struct tss_struct        doublefault_tss;
 138 extern __u32                    cleared_cpu_caps[NCAPINTS];
 139
 140 #ifdef CONFIG_SMP
 141 DECLARE_PER_CPU(struct cpuinfo_x86, cpu_info);
 142 #define cpu_data(cpu)           per_cpu(cpu_info, cpu)
 143 #define current_cpu_data        __get_cpu_var(cpu_info)
 144 #else
 145 #define cpu_data(cpu)           boot_cpu_data
 146 #define current_cpu_data        boot_cpu_data
 147 #endif
 148
 149 extern const struct seq_operations cpuinfo_op;
 150
 151 static inline int hlt_works(int cpu)
 152 {
 153 #ifdef CONFIG_X86_32
 154         return cpu_data(cpu).hlt_works_ok;
 155 #else
 156         return 1;
 157 #endif
 158 }
 159
 160 #define cache_line_size()       (boot_cpu_data.x86_cache_alignment)
 161
 162 extern void cpu_detect(struct cpuinfo_x86 *c);
 163
 164 extern struct pt_regs *idle_regs(struct pt_regs *);
 165
 166 extern void early_cpu_init(void);
 167 extern void identify_boot_cpu(void);
 168 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
 169 extern void print_cpu_info(struct cpuinfo_x86 *);
 170 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
 171 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
 172 extern unsigned short num_cache_leaves;
 173
 174 extern void detect_extended_topology(struct cpuinfo_x86 *c);
 175 extern void detect_ht(struct cpuinfo_x86 *c);
 176
 177 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
 178                                 unsigned int *ecx, unsigned int *edx)
 179 {
 180         /* ecx is often an input as well as an output. */
 181         asm("cpuid"
 182             : "=a" (*eax),
 183               "=b" (*ebx),
 184               "=c" (*ecx),
 185               "=d" (*edx)
 186             : "0" (*eax), "2" (*ecx));
 187 }
 188
 189 static inline void load_cr3(pgd_t *pgdir)
 190 {
 191         write_cr3(__pa(pgdir));
 192 }
 193
 194 #ifdef CONFIG_X86_32
 195 /* This is the TSS defined by the hardware. */
 196 struct x86_hw_tss {
 197         unsigned short          back_link, __blh;
 198         unsigned long           sp0;
 199         unsigned short          ss0, __ss0h;
 200         unsigned long           sp1;
 201         /* ss1 caches MSR_IA32_SYSENTER_CS: */
 202         unsigned short          ss1, __ss1h;
 203         unsigned long           sp2;
 204         unsigned short          ss2, __ss2h;
 205         unsigned long           __cr3;
 206         unsigned long           ip;
 207         unsigned long           flags;
 208         unsigned long           ax;
 209         unsigned long           cx;
 210         unsigned long           dx;
 211         unsigned long           bx;
 212         unsigned long           sp;
 213         unsigned long           bp;
 214         unsigned long           si;
 215         unsigned long           di;
 216         unsigned short          es, __esh;
 217         unsigned short          cs, __csh;
 218         unsigned short          ss, __ssh;
 219         unsigned short          ds, __dsh;
 220         unsigned short          fs, __fsh;
 221         unsigned short          gs, __gsh;
 222         unsigned short          ldt, __ldth;
 223         unsigned short          trace;
 224         unsigned short          io_bitmap_base;
 225
 226 } __attribute__((packed));
 227 #else
 228 struct x86_hw_tss {
 229         u32                     reserved1;
 230         u64                     sp0;
 231         u64                     sp1;
 232         u64                     sp2;
 233         u64                     reserved2;
 234         u64                     ist[7];
 235         u32                     reserved3;
 236         u32                     reserved4;
 237         u16                     reserved5;
 238         u16                     io_bitmap_base;
 239
 240 } __attribute__((packed)) ____cacheline_aligned;
 241 #endif
 242
 243 /*
 244  * IO-bitmap sizes:
 245  */
 246 #define IO_BITMAP_BITS                  65536
 247 #define IO_BITMAP_BYTES                 (IO_BITMAP_BITS/8)
 248 #define IO_BITMAP_LONGS                 (IO_BITMAP_BYTES/sizeof(long))
 249 #define IO_BITMAP_OFFSET                offsetof(struct tss_struct, io_bitmap)
 250 #define INVALID_IO_BITMAP_OFFSET        0x8000
 251
 252 struct tss_struct {
 253         /*
 254          * The hardware state:
 255          */
 256         struct x86_hw_tss       x86_tss;
 257
 258         /*
 259          * The extra 1 is there because the CPU will access an
 260          * additional byte beyond the end of the IO permission
 261          * bitmap. The extra byte must be all 1 bits, and must
 262          * be within the limit.
 263          */
 264         unsigned long           io_bitmap[IO_BITMAP_LONGS + 1];
 265
 266         /*
 267          * .. and then another 0x100 bytes for the emergency kernel stack:
 268          */
 269         unsigned long           stack[64];
 270
 271 } ____cacheline_aligned;
 272
 273 DECLARE_PER_CPU(struct tss_struct, init_tss);
 274
 275 /*
 276  * Save the original ist values for checking stack pointers during debugging
 277  */
 278 struct orig_ist {
 279         unsigned long           ist[7];
 280 };
 281
 282 #define MXCSR_DEFAULT           0x1f80
 283
 284 struct i387_fsave_struct {
 285         u32                     cwd;    /* FPU Control Word             */
 286         u32                     swd;    /* FPU Status Word              */
 287         u32                     twd;    /* FPU Tag Word                 */
 288         u32                     fip;    /* FPU IP Offset                */
 289         u32                     fcs;    /* FPU IP Selector              */
 290         u32                     foo;    /* FPU Operand Pointer Offset   */
 291         u32                     fos;    /* FPU Operand Pointer Selector */
 292
 293         /* 8*10 bytes for each FP-reg = 80 bytes:                       */
 294         u32                     st_space[20];
 295
 296         /* Software status information [not touched by FSAVE ]:         */
 297         u32                     status;
 298 };
 299
 300 struct i387_fxsave_struct {
 301         u16                     cwd; /* Control Word                    */
 302         u16                     swd; /* Status Word                     */
 303         u16                     twd; /* Tag Word                        */
 304         u16                     fop; /* Last Instruction Opcode         */
 305         union {
 306                 struct {
 307                         u64     rip; /* Instruction Pointer             */
 308                         u64     rdp; /* Data Pointer                    */
 309                 };
 310                 struct {
 311                         u32     fip; /* FPU IP Offset                   */
 312                         u32     fcs; /* FPU IP Selector                 */
 313                         u32     foo; /* FPU Operand Offset              */
 314                         u32     fos; /* FPU Operand Selector            */
 315                 };
 316         };
 317         u32                     mxcsr;          /* MXCSR Register State */
 318         u32                     mxcsr_mask;     /* MXCSR Mask           */
 319
 320         /* 8*16 bytes for each FP-reg = 128 bytes:                      */
 321         u32                     st_space[32];
 322
 323         /* 16*16 bytes for each XMM-reg = 256 bytes:                    */
 324         u32                     xmm_space[64];
 325
 326         u32                     padding[12];
 327
 328         union {
 329                 u32             padding1[12];
 330                 u32             sw_reserved[12];
 331         };
 332
 333 } __attribute__((aligned(16)));
 334
 335 struct i387_soft_struct {
 336         u32                     cwd;
 337         u32                     swd;
 338         u32                     twd;
 339         u32                     fip;
 340         u32                     fcs;
 341         u32                     foo;
 342         u32                     fos;
 343         /* 8*10 bytes for each FP-reg = 80 bytes: */
 344         u32                     st_space[20];
 345         u8                      ftop;
 346         u8                      changed;
 347         u8                      lookahead;
 348         u8                      no_update;
 349         u8                      rm;
 350         u8                      alimit;
 351         struct math_emu_info    *info;
 352         u32                     entry_eip;
 353 };
 354
 355 struct xsave_hdr_struct {
 356         u64 xstate_bv;
 357         u64 reserved1[2];
 358         u64 reserved2[5];
 359 } __attribute__((packed));
 360
 361 struct xsave_struct {
 362         struct i387_fxsave_struct i387;
 363         struct xsave_hdr_struct xsave_hdr;
 364         /* new processor state extensions will go here */
 365 } __attribute__ ((packed, aligned (64)));
 366
 367 union thread_xstate {
 368         struct i387_fsave_struct        fsave;
 369         struct i387_fxsave_struct       fxsave;
 370         struct i387_soft_struct         soft;
 371         struct xsave_struct             xsave;
 372 };
 373
 374 #ifdef CONFIG_X86_64
 375 DECLARE_PER_CPU(struct orig_ist, orig_ist);
 376
 377 union irq_stack_union {
 378         char irq_stack[IRQ_STACK_SIZE];
 379         /*
 380          * GCC hardcodes the stack canary as %gs:40.  Since the
 381          * irq_stack is the object at %gs:0, we reserve the bottom
 382          * 48 bytes of the irq stack for the canary.
 383          */
 384         struct {
 385                 char gs_base[40];
 386                 unsigned long stack_canary;
 387         };
 388 };
 389
 390 DECLARE_PER_CPU(union irq_stack_union, irq_stack_union);
 391 DECLARE_INIT_PER_CPU(irq_stack_union);
 392
 393 DECLARE_PER_CPU(char *, irq_stack_ptr);
 394 #else   /* X86_64 */
 395 #ifdef CONFIG_CC_STACKPROTECTOR
 396 DECLARE_PER_CPU(unsigned long, stack_canary);
 397 #endif
 398 #endif  /* X86_64 */
 399
 400 extern unsigned int xstate_size;
 401 extern void free_thread_xstate(struct task_struct *);
 402 extern struct kmem_cache *task_xstate_cachep;
 403 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
 404 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
 405 extern unsigned short num_cache_leaves;
 406
 407 struct thread_struct {
 408         /* Cached TLS descriptors: */
 409         struct desc_struct      tls_array[GDT_ENTRY_TLS_ENTRIES];
 410         unsigned long           sp0;
 411         unsigned long           sp;
 412 #ifdef CONFIG_X86_32
 413         unsigned long           sysenter_cs;
 414 #else
 415         unsigned long           usersp; /* Copy from PDA */
 416         unsigned short          es;
 417         unsigned short          ds;
 418         unsigned short          fsindex;
 419         unsigned short          gsindex;
 420 #endif
 421         unsigned long           ip;
 422         unsigned long           fs;
 423         unsigned long           gs;
 424         /* Hardware debugging registers: */
 425         unsigned long           debugreg0;
 426         unsigned long           debugreg1;
 427         unsigned long           debugreg2;
 428         unsigned long           debugreg3;
 429         unsigned long           debugreg6;
 430         unsigned long           debugreg7;
 431         /* Fault info: */
 432         unsigned long           cr2;
 433         unsigned long           trap_no;
 434         unsigned long           error_code;
 435         /* floating point and extended processor state */
 436         union thread_xstate     *xstate;
 437 #ifdef CONFIG_X86_32
 438         /* Virtual 86 mode info */
 439         struct vm86_struct __user *vm86_info;
 440         unsigned long           screen_bitmap;
 441         unsigned long           v86flags;
 442         unsigned long           v86mask;
 443         unsigned long           saved_sp0;
 444         unsigned int            saved_fs;
 445         unsigned int            saved_gs;
 446 #endif
 447         /* IO permissions: */
 448         unsigned long           *io_bitmap_ptr;
 449         unsigned long           iopl;
 450         /* Max allowed port in the bitmap, in bytes: */
 451         unsigned                io_bitmap_max;
 452 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
 453         unsigned long   debugctlmsr;
 454 #ifdef CONFIG_X86_DS
 455 /* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
 456         struct ds_context       *ds_ctx;
 457 #endif /* CONFIG_X86_DS */
 458 #ifdef CONFIG_X86_PTRACE_BTS
 459 /* the signal to send on a bts buffer overflow */
 460         unsigned int    bts_ovfl_signal;
 461 #endif /* CONFIG_X86_PTRACE_BTS */
 462 };
 463
 464 static inline unsigned long native_get_debugreg(int regno)
 465 {
 466         unsigned long val = 0;  /* Damn you, gcc! */
 467
 468         switch (regno) {
 469         case 0:
 470                 asm("mov %%db0, %0" :"=r" (val));
 471                 break;
 472         case 1:
 473                 asm("mov %%db1, %0" :"=r" (val));
 474                 break;
 475         case 2:
 476                 asm("mov %%db2, %0" :"=r" (val));
 477                 break;
 478         case 3:
 479                 asm("mov %%db3, %0" :"=r" (val));
 480                 break;
 481         case 6:
 482                 asm("mov %%db6, %0" :"=r" (val));
 483                 break;
 484         case 7:
 485                 asm("mov %%db7, %0" :"=r" (val));
 486                 break;
 487         default:
 488                 BUG();
 489         }
 490         return val;
 491 }
 492
 493 static inline void native_set_debugreg(int regno, unsigned long value)
 494 {
 495         switch (regno) {
 496         case 0:
 497                 asm("mov %0, %%db0"     ::"r" (value));
 498                 break;
 499         case 1:
 500                 asm("mov %0, %%db1"     ::"r" (value));
 501                 break;
 502         case 2:
 503                 asm("mov %0, %%db2"     ::"r" (value));
 504                 break;
 505         case 3:
 506                 asm("mov %0, %%db3"     ::"r" (value));
 507                 break;
 508         case 6:
 509                 asm("mov %0, %%db6"     ::"r" (value));
 510                 break;
 511         case 7:
 512                 asm("mov %0, %%db7"     ::"r" (value));
 513                 break;
 514         default:
 515                 BUG();
 516         }
 517 }
 518
 519 /*
 520  * Set IOPL bits in EFLAGS from given mask
 521  */
 522 static inline void native_set_iopl_mask(unsigned mask)
 523 {
 524 #ifdef CONFIG_X86_32
 525         unsigned int reg;
 526
 527         asm volatile ("pushfl;"
 528                       "popl %0;"
 529                       "andl %1, %0;"
 530                       "orl %2, %0;"
 531                       "pushl %0;"
 532                       "popfl"
 533                       : "=&r" (reg)
 534                       : "i" (~X86_EFLAGS_IOPL), "r" (mask));
 535 #endif
 536 }
 537
 538 static inline void
 539 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
 540 {
 541         tss->x86_tss.sp0 = thread->sp0;
 542 #ifdef CONFIG_X86_32
 543         /* Only happens when SEP is enabled, no need to test "SEP"arately: */
 544         if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
 545                 tss->x86_tss.ss1 = thread->sysenter_cs;
 546                 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
 547         }
 548 #endif
 549 }
 550
 551 static inline void native_swapgs(void)
 552 {
 553 #ifdef CONFIG_X86_64
 554         asm volatile("swapgs" ::: "memory");
 555 #endif
 556 }
 557
 558 #ifdef CONFIG_PARAVIRT
 559 #include <asm/paravirt.h>
 560 #else
 561 #define __cpuid                 native_cpuid
 562 #define paravirt_enabled()      0
 563
 564 /*
 565  * These special macros can be used to get or set a debugging register
 566  */
 567 #define get_debugreg(var, register)                             \
 568         (var) = native_get_debugreg(register)
 569 #define set_debugreg(value, register)                           \
 570         native_set_debugreg(register, value)
 571
 572 static inline void load_sp0(struct tss_struct *tss,
 573                             struct thread_struct *thread)
 574 {
 575         native_load_sp0(tss, thread);
 576 }
 577
 578 #define set_iopl_mask native_set_iopl_mask
 579 #endif /* CONFIG_PARAVIRT */
 580
 581 /*
 582  * Save the cr4 feature set we're using (ie
 583  * Pentium 4MB enable and PPro Global page
 584  * enable), so that any CPU's that boot up
 585  * after us can get the correct flags.
 586  */
 587 extern unsigned long            mmu_cr4_features;
 588
 589 static inline void set_in_cr4(unsigned long mask)
 590 {
 591         unsigned cr4;
 592
 593         mmu_cr4_features |= mask;
 594         cr4 = read_cr4();
 595         cr4 |= mask;
 596         write_cr4(cr4);
 597 }
 598
 599 static inline void clear_in_cr4(unsigned long mask)
 600 {
 601         unsigned cr4;
 602
 603         mmu_cr4_features &= ~mask;
 604         cr4 = read_cr4();
 605         cr4 &= ~mask;
 606         write_cr4(cr4);
 607 }
 608
 609 typedef struct {
 610         unsigned long           seg;
 611 } mm_segment_t;
 612
 613
 614 /*
 615  * create a kernel thread without removing it from tasklists
 616  */
 617 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
 618
 619 /* Free all resources held by a thread. */
 620 extern void release_thread(struct task_struct *);
 621
 622 /* Prepare to copy thread state - unlazy all lazy state */
 623 extern void prepare_to_copy(struct task_struct *tsk);
 624
 625 unsigned long get_wchan(struct task_struct *p);
 626
 627 /*
 628  * Generic CPUID function
 629  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
 630  * resulting in stale register contents being returned.
 631  */
 632 static inline void cpuid(unsigned int op,
 633                          unsigned int *eax, unsigned int *ebx,
 634                          unsigned int *ecx, unsigned int *edx)
 635 {
 636         *eax = op;
 637         *ecx = 0;
 638         __cpuid(eax, ebx, ecx, edx);
 639 }
 640
 641 /* Some CPUID calls want 'count' to be placed in ecx */
 642 static inline void cpuid_count(unsigned int op, int count,
 643                                unsigned int *eax, unsigned int *ebx,
 644                                unsigned int *ecx, unsigned int *edx)
 645 {
 646         *eax = op;
 647         *ecx = count;
 648         __cpuid(eax, ebx, ecx, edx);
 649 }
 650
 651 /*
 652  * CPUID functions returning a single datum
 653  */
 654 static inline unsigned int cpuid_eax(unsigned int op)
 655 {
 656         unsigned int eax, ebx, ecx, edx;
 657
 658         cpuid(op, &eax, &ebx, &ecx, &edx);
 659
 660         return eax;
 661 }
 662
 663 static inline unsigned int cpuid_ebx(unsigned int op)
 664 {
 665         unsigned int eax, ebx, ecx, edx;
 666
 667         cpuid(op, &eax, &ebx, &ecx, &edx);
 668
 669         return ebx;
 670 }
 671
 672 static inline unsigned int cpuid_ecx(unsigned int op)
 673 {
 674         unsigned int eax, ebx, ecx, edx;
 675
 676         cpuid(op, &eax, &ebx, &ecx, &edx);
 677
 678         return ecx;
 679 }
 680
 681 static inline unsigned int cpuid_edx(unsigned int op)
 682 {
 683         unsigned int eax, ebx, ecx, edx;
 684
 685         cpuid(op, &eax, &ebx, &ecx, &edx);
 686
 687         return edx;
 688 }
 689
 690 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
 691 static inline void rep_nop(void)
 692 {
 693         asm volatile("rep; nop" ::: "memory");
 694 }
 695
 696 static inline void cpu_relax(void)
 697 {
 698         rep_nop();
 699 }
 700
 701 /* Stop speculative execution: */
 702 static inline void sync_core(void)
 703 {
 704         int tmp;
 705
 706         asm volatile("cpuid" : "=a" (tmp) : "0" (1)
 707                      : "ebx", "ecx", "edx", "memory");
 708 }
 709
 710 static inline void __monitor(const void *eax, unsigned long ecx,
 711                              unsigned long edx)
 712 {
 713         /* "monitor %eax, %ecx, %edx;" */
 714         asm volatile(".byte 0x0f, 0x01, 0xc8;"
 715                      :: "a" (eax), "c" (ecx), "d"(edx));
 716 }
 717
 718 static inline void __mwait(unsigned long eax, unsigned long ecx)
 719 {
 720         /* "mwait %eax, %ecx;" */
 721         asm volatile(".byte 0x0f, 0x01, 0xc9;"
 722                      :: "a" (eax), "c" (ecx));
 723 }
 724
 725 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
 726 {
 727         trace_hardirqs_on();
 728         /* "mwait %eax, %ecx;" */
 729         asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
 730                      :: "a" (eax), "c" (ecx));
 731 }
 732
 733 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
 734
 735 extern void select_idle_routine(const struct cpuinfo_x86 *c);
 736
 737 extern unsigned long            boot_option_idle_override;
 738 extern unsigned long            idle_halt;
 739 extern unsigned long            idle_nomwait;
 740
 741 /*
 742  * on systems with caches, caches must be flashed as the absolute
 743  * last instruction before going into a suspended halt.  Otherwise,
 744  * dirty data can linger in the cache and become stale on resume,
 745  * leading to strange errors.
 746  *
 747  * perform a variety of operations to guarantee that the compiler
 748  * will not reorder instructions.  wbinvd itself is serializing
 749  * so the processor will not reorder.
 750  *
 751  * Systems without cache can just go into halt.
 752  */
 753 static inline void wbinvd_halt(void)
 754 {
 755         mb();
 756         /* check for clflush to determine if wbinvd is legal */
 757         if (cpu_has_clflush)
 758                 asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
 759         else
 760                 while (1)
 761                         halt();
 762 }
 763
 764 extern void enable_sep_cpu(void);
 765 extern int sysenter_setup(void);
 766
 767 /* Defined in head.S */
 768 extern struct desc_ptr          early_gdt_descr;
 769
 770 extern void cpu_set_gdt(int);
 771 extern void switch_to_new_gdt(int);
 772 extern void load_percpu_segment(int);
 773 extern void cpu_init(void);
 774
 775 static inline unsigned long get_debugctlmsr(void)
 776 {
 777     unsigned long debugctlmsr = 0;
 778
 779 #ifndef CONFIG_X86_DEBUGCTLMSR
 780         if (boot_cpu_data.x86 < 6)
 781                 return 0;
 782 #endif
 783         rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
 784
 785     return debugctlmsr;
 786 }
 787
 788 static inline void update_debugctlmsr(unsigned long debugctlmsr)
 789 {
 790 #ifndef CONFIG_X86_DEBUGCTLMSR
 791         if (boot_cpu_data.x86 < 6)
 792                 return;
 793 #endif
 794         wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
 795 }
 796
 797 /*
 798  * from system description table in BIOS. Mostly for MCA use, but
 799  * others may find it useful:
 800  */
 801 extern unsigned int             machine_id;
 802 extern unsigned int             machine_submodel_id;
 803 extern unsigned int             BIOS_revision;
 804
 805 /* Boot loader type from the setup header: */
 806 extern int                      bootloader_type;
 807
 808 extern char                     ignore_fpu_irq;
 809
 810 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
 811 #define ARCH_HAS_PREFETCHW
 812 #define ARCH_HAS_SPINLOCK_PREFETCH
 813
 814 #ifdef CONFIG_X86_32
 815 # define BASE_PREFETCH          ASM_NOP4
 816 # define ARCH_HAS_PREFETCH
 817 #else
 818 # define BASE_PREFETCH          "prefetcht0 (%1)"
 819 #endif
 820
 821 /*
 822  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
 823  *
 824  * It's not worth to care about 3dnow prefetches for the K6
 825  * because they are microcoded there and very slow.
 826  */
 827 static inline void prefetch(const void *x)
 828 {
 829         alternative_input(BASE_PREFETCH,
 830                           "prefetchnta (%1)",
 831                           X86_FEATURE_XMM,
 832                           "r" (x));
 833 }
 834
 835 /*
 836  * 3dnow prefetch to get an exclusive cache line.
 837  * Useful for spinlocks to avoid one state transition in the
 838  * cache coherency protocol:
 839  */
 840 static inline void prefetchw(const void *x)
 841 {
 842         alternative_input(BASE_PREFETCH,
 843                           "prefetchw (%1)",
 844                           X86_FEATURE_3DNOW,
 845                           "r" (x));
 846 }
 847
 848 static inline void spin_lock_prefetch(const void *x)
 849 {
 850         prefetchw(x);
 851 }
 852
 853 #ifdef CONFIG_X86_32
 854 /*
 855  * User space process size: 3GB (default).
 856  */
 857 #define TASK_SIZE               PAGE_OFFSET
 858 #define TASK_SIZE_MAX           TASK_SIZE
 859 #define STACK_TOP               TASK_SIZE
 860 #define STACK_TOP_MAX           STACK_TOP
 861
 862 #define INIT_THREAD  {                                                    \
 863         .sp0                    = sizeof(init_stack) + (long)&init_stack, \
 864         .vm86_info              = NULL,                                   \
 865         .sysenter_cs            = __KERNEL_CS,                            \
 866         .io_bitmap_ptr          = NULL,                                   \
 867         .fs                     = __KERNEL_PERCPU,                        \
 868 }
 869
 870 /*
 871  * Note that the .io_bitmap member must be extra-big. This is because
 872  * the CPU will access an additional byte beyond the end of the IO
 873  * permission bitmap. The extra byte must be all 1 bits, and must
 874  * be within the limit.
 875  */
 876 #define INIT_TSS  {                                                       \
 877         .x86_tss = {                                                      \
 878                 .sp0            = sizeof(init_stack) + (long)&init_stack, \
 879                 .ss0            = __KERNEL_DS,                            \
 880                 .ss1            = __KERNEL_CS,                            \
 881                 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,               \
 882          },                                                               \
 883         .io_bitmap              = { [0 ... IO_BITMAP_LONGS] = ~0 },       \
 884 }
 885
 886 extern unsigned long thread_saved_pc(struct task_struct *tsk);
 887
 888 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
 889 #define KSTK_TOP(info)                                                 \
 890 ({                                                                     \
 891        unsigned long *__ptr = (unsigned long *)(info);                 \
 892        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
 893 })
 894
 895 /*
 896  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
 897  * This is necessary to guarantee that the entire "struct pt_regs"
 898  * is accessable even if the CPU haven't stored the SS/ESP registers
 899  * on the stack (interrupt gate does not save these registers
 900  * when switching to the same priv ring).
 901  * Therefore beware: accessing the ss/esp fields of the
 902  * "struct pt_regs" is possible, but they may contain the
 903  * completely wrong values.
 904  */
 905 #define task_pt_regs(task)                                             \
 906 ({                                                                     \
 907        struct pt_regs *__regs__;                                       \
 908        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
 909        __regs__ - 1;                                                   \
 910 })
 911
 912 #define KSTK_ESP(task)          (task_pt_regs(task)->sp)
 913
 914 #else
 915 /*
 916  * User space process size. 47bits minus one guard page.
 917  */
 918 #define TASK_SIZE_MAX   ((1UL << 47) - PAGE_SIZE)
 919
 920 /* This decides where the kernel will search for a free chunk of vm
 921  * space during mmap's.
 922  */
 923 #define IA32_PAGE_OFFSET        ((current->personality & ADDR_LIMIT_3GB) ? \
 924                                         0xc0000000 : 0xFFFFe000)
 925
 926 #define TASK_SIZE               (test_thread_flag(TIF_IA32) ? \
 927                                         IA32_PAGE_OFFSET : TASK_SIZE_MAX)
 928 #define TASK_SIZE_OF(child)     ((test_tsk_thread_flag(child, TIF_IA32)) ? \
 929                                         IA32_PAGE_OFFSET : TASK_SIZE_MAX)
 930
 931 #define STACK_TOP               TASK_SIZE
 932 #define STACK_TOP_MAX           TASK_SIZE_MAX
 933
 934 #define INIT_THREAD  { \
 935         .sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
 936 }
 937
 938 #define INIT_TSS  { \
 939         .x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
 940 }
 941
 942 /*
 943  * Return saved PC of a blocked thread.
 944  * What is this good for? it will be always the scheduler or ret_from_fork.
 945  */
 946 #define thread_saved_pc(t)      (*(unsigned long *)((t)->thread.sp - 8))
 947
 948 #define task_pt_regs(tsk)       ((struct pt_regs *)(tsk)->thread.sp0 - 1)
 949 #define KSTK_ESP(tsk)           -1 /* sorry. doesn't work for syscall. */
 950 #endif /* CONFIG_X86_64 */
 951
 952 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
 953                                                unsigned long new_sp);
 954
 955 /*
 956  * This decides where the kernel will search for a free chunk of vm
 957  * space during mmap's.
 958  */
 959 #define TASK_UNMAPPED_BASE      (PAGE_ALIGN(TASK_SIZE / 3))
 960
 961 #define KSTK_EIP(task)          (task_pt_regs(task)->ip)
 962
 963 /* Get/set a process' ability to use the timestamp counter instruction */
 964 #define GET_TSC_CTL(adr)        get_tsc_mode((adr))
 965 #define SET_TSC_CTL(val)        set_tsc_mode((val))
 966
 967 extern int get_tsc_mode(unsigned long adr);
 968 extern int set_tsc_mode(unsigned int val);
 969
 970 #endif /* _ASM_X86_PROCESSOR_H */