include/asm-i386/processor.h

   1 /*
   2  * include/asm-i386/processor.h
   3  *
   4  * Copyright (C) 1994 Linus Torvalds
   5  */
   6
   7 #ifndef __ASM_I386_PROCESSOR_H
   8 #define __ASM_I386_PROCESSOR_H
   9
  10 #include <asm/vm86.h>
  11 #include <asm/math_emu.h>
  12 #include <asm/segment.h>
  13 #include <asm/page.h>
  14 #include <asm/types.h>
  15 #include <asm/sigcontext.h>
  16 #include <asm/cpufeature.h>
  17 #include <asm/msr.h>
  18 #include <asm/system.h>
  19 #include <linux/cache.h>
  20 #include <linux/threads.h>
  21 #include <asm/percpu.h>
  22 #include <linux/cpumask.h>
  23 #include <linux/init.h>
  24
  25 /* flag for disabling the tsc */
  26 extern int tsc_disable;
  27
  28 struct desc_struct {
  29         unsigned long a,b;
  30 };
  31
  32 #define desc_empty(desc) \
  33                 (!((desc)->a | (desc)->b))
  34
  35 #define desc_equal(desc1, desc2) \
  36                 (((desc1)->a == (desc2)->a) && ((desc1)->b == (desc2)->b))
  37 /*
  38  * Default implementation of macro that returns current
  39  * instruction pointer ("program counter").
  40  */
  41 #define current_text_addr() ({ void *pc; __asm__("movl $1f,%0\n1:":"=g" (pc)); pc; })
  42
  43 /*
  44  *  CPU type and hardware bug flags. Kept separately for each CPU.
  45  *  Members of this structure are referenced in head.S, so think twice
  46  *  before touching them. [mj]
  47  */
  48
  49 struct cpuinfo_x86 {
  50         __u8    x86;            /* CPU family */
  51         __u8    x86_vendor;     /* CPU vendor */
  52         __u8    x86_model;
  53         __u8    x86_mask;
  54         char    wp_works_ok;    /* It doesn't on 386's */
  55         char    hlt_works_ok;   /* Problems on some 486Dx4's and old 386's */
  56         char    hard_math;
  57         char    rfu;
  58         int     cpuid_level;    /* Maximum supported CPUID level, -1=no CPUID */
  59         unsigned long   x86_capability[NCAPINTS];
  60         char    x86_vendor_id[16];
  61         char    x86_model_id[64];
  62         int     x86_cache_size;  /* in KB - valid for CPUS which support this
  63                                     call  */
  64         int     x86_cache_alignment;    /* In bytes */
  65         char    fdiv_bug;
  66         char    f00f_bug;
  67         char    coma_bug;
  68         char    pad0;
  69         int     x86_power;
  70         unsigned long loops_per_jiffy;
  71 #ifdef CONFIG_SMP
  72         cpumask_t llc_shared_map;       /* cpus sharing the last level cache */
  73 #endif
  74         unsigned char x86_max_cores;    /* cpuid returned max cores value */
  75         unsigned char apicid;
  76         unsigned short x86_clflush_size;
  77 #ifdef CONFIG_SMP
  78         unsigned char booted_cores;     /* number of cores as seen by OS */
  79         __u8 phys_proc_id;              /* Physical processor id. */
  80         __u8 cpu_core_id;               /* Core id */
  81 #endif
  82 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
  83
  84 #define X86_VENDOR_INTEL 0
  85 #define X86_VENDOR_CYRIX 1
  86 #define X86_VENDOR_AMD 2
  87 #define X86_VENDOR_UMC 3
  88 #define X86_VENDOR_NEXGEN 4
  89 #define X86_VENDOR_CENTAUR 5
  90 #define X86_VENDOR_RISE 6
  91 #define X86_VENDOR_TRANSMETA 7
  92 #define X86_VENDOR_NSC 8
  93 #define X86_VENDOR_NUM 9
  94 #define X86_VENDOR_UNKNOWN 0xff
  95
  96 /*
  97  * capabilities of CPUs
  98  */
  99
 100 extern struct cpuinfo_x86 boot_cpu_data;
 101 extern struct cpuinfo_x86 new_cpu_data;
 102 extern struct tss_struct doublefault_tss;
 103 DECLARE_PER_CPU(struct tss_struct, init_tss);
 104
 105 #ifdef CONFIG_SMP
 106 extern struct cpuinfo_x86 cpu_data[];
 107 #define current_cpu_data cpu_data[smp_processor_id()]
 108 #else
 109 #define cpu_data (&boot_cpu_data)
 110 #define current_cpu_data boot_cpu_data
 111 #endif
 112
 113 extern  int cpu_llc_id[NR_CPUS];
 114 extern char ignore_fpu_irq;
 115
 116 void __init cpu_detect(struct cpuinfo_x86 *c);
 117
 118 extern void identify_cpu(struct cpuinfo_x86 *);
 119 extern void print_cpu_info(struct cpuinfo_x86 *);
 120 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
 121 extern unsigned short num_cache_leaves;
 122
 123 #ifdef CONFIG_X86_HT
 124 extern void detect_ht(struct cpuinfo_x86 *c);
 125 #else
 126 static inline void detect_ht(struct cpuinfo_x86 *c) {}
 127 #endif
 128
 129 /*
 130  * EFLAGS bits
 131  */
 132 #define X86_EFLAGS_CF   0x00000001 /* Carry Flag */
 133 #define X86_EFLAGS_PF   0x00000004 /* Parity Flag */
 134 #define X86_EFLAGS_AF   0x00000010 /* Auxillary carry Flag */
 135 #define X86_EFLAGS_ZF   0x00000040 /* Zero Flag */
 136 #define X86_EFLAGS_SF   0x00000080 /* Sign Flag */
 137 #define X86_EFLAGS_TF   0x00000100 /* Trap Flag */
 138 #define X86_EFLAGS_IF   0x00000200 /* Interrupt Flag */
 139 #define X86_EFLAGS_DF   0x00000400 /* Direction Flag */
 140 #define X86_EFLAGS_OF   0x00000800 /* Overflow Flag */
 141 #define X86_EFLAGS_IOPL 0x00003000 /* IOPL mask */
 142 #define X86_EFLAGS_NT   0x00004000 /* Nested Task */
 143 #define X86_EFLAGS_RF   0x00010000 /* Resume Flag */
 144 #define X86_EFLAGS_VM   0x00020000 /* Virtual Mode */
 145 #define X86_EFLAGS_AC   0x00040000 /* Alignment Check */
 146 #define X86_EFLAGS_VIF  0x00080000 /* Virtual Interrupt Flag */
 147 #define X86_EFLAGS_VIP  0x00100000 /* Virtual Interrupt Pending */
 148 #define X86_EFLAGS_ID   0x00200000 /* CPUID detection flag */
 149
 150 static inline fastcall void native_cpuid(unsigned int *eax, unsigned int *ebx,
 151                                          unsigned int *ecx, unsigned int *edx)
 152 {
 153         /* ecx is often an input as well as an output. */
 154         __asm__("cpuid"
 155                 : "=a" (*eax),
 156                   "=b" (*ebx),
 157                   "=c" (*ecx),
 158                   "=d" (*edx)
 159                 : "0" (*eax), "2" (*ecx));
 160 }
 161
 162 #define load_cr3(pgdir) write_cr3(__pa(pgdir))
 163
 164 /*
 165  * Intel CPU features in CR4
 166  */
 167 #define X86_CR4_VME             0x0001  /* enable vm86 extensions */
 168 #define X86_CR4_PVI             0x0002  /* virtual interrupts flag enable */
 169 #define X86_CR4_TSD             0x0004  /* disable time stamp at ipl 3 */
 170 #define X86_CR4_DE              0x0008  /* enable debugging extensions */
 171 #define X86_CR4_PSE             0x0010  /* enable page size extensions */
 172 #define X86_CR4_PAE             0x0020  /* enable physical address extensions */
 173 #define X86_CR4_MCE             0x0040  /* Machine check enable */
 174 #define X86_CR4_PGE             0x0080  /* enable global pages */
 175 #define X86_CR4_PCE             0x0100  /* enable performance counters at ipl 3 */
 176 #define X86_CR4_OSFXSR          0x0200  /* enable fast FPU save and restore */
 177 #define X86_CR4_OSXMMEXCPT      0x0400  /* enable unmasked SSE exceptions */
 178
 179 /*
 180  * Save the cr4 feature set we're using (ie
 181  * Pentium 4MB enable and PPro Global page
 182  * enable), so that any CPU's that boot up
 183  * after us can get the correct flags.
 184  */
 185 extern unsigned long mmu_cr4_features;
 186
 187 static inline void set_in_cr4 (unsigned long mask)
 188 {
 189         unsigned cr4;
 190         mmu_cr4_features |= mask;
 191         cr4 = read_cr4();
 192         cr4 |= mask;
 193         write_cr4(cr4);
 194 }
 195
 196 static inline void clear_in_cr4 (unsigned long mask)
 197 {
 198         unsigned cr4;
 199         mmu_cr4_features &= ~mask;
 200         cr4 = read_cr4();
 201         cr4 &= ~mask;
 202         write_cr4(cr4);
 203 }
 204
 205 /*
 206  *      NSC/Cyrix CPU configuration register indexes
 207  */
 208
 209 #define CX86_PCR0 0x20
 210 #define CX86_GCR  0xb8
 211 #define CX86_CCR0 0xc0
 212 #define CX86_CCR1 0xc1
 213 #define CX86_CCR2 0xc2
 214 #define CX86_CCR3 0xc3
 215 #define CX86_CCR4 0xe8
 216 #define CX86_CCR5 0xe9
 217 #define CX86_CCR6 0xea
 218 #define CX86_CCR7 0xeb
 219 #define CX86_PCR1 0xf0
 220 #define CX86_DIR0 0xfe
 221 #define CX86_DIR1 0xff
 222 #define CX86_ARR_BASE 0xc4
 223 #define CX86_RCR_BASE 0xdc
 224
 225 /*
 226  *      NSC/Cyrix CPU indexed register access macros
 227  */
 228
 229 #define getCx86(reg) ({ outb((reg), 0x22); inb(0x23); })
 230
 231 #define setCx86(reg, data) do { \
 232         outb((reg), 0x22); \
 233         outb((data), 0x23); \
 234 } while (0)
 235
 236 /* Stop speculative execution */
 237 static inline void sync_core(void)
 238 {
 239         int tmp;
 240         asm volatile("cpuid" : "=a" (tmp) : "0" (1) : "ebx","ecx","edx","memory");
 241 }
 242
 243 static inline void __monitor(const void *eax, unsigned long ecx,
 244                 unsigned long edx)
 245 {
 246         /* "monitor %eax,%ecx,%edx;" */
 247         asm volatile(
 248                 ".byte 0x0f,0x01,0xc8;"
 249                 : :"a" (eax), "c" (ecx), "d"(edx));
 250 }
 251
 252 static inline void __mwait(unsigned long eax, unsigned long ecx)
 253 {
 254         /* "mwait %eax,%ecx;" */
 255         asm volatile(
 256                 ".byte 0x0f,0x01,0xc9;"
 257                 : :"a" (eax), "c" (ecx));
 258 }
 259
 260 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
 261
 262 /* from system description table in BIOS.  Mostly for MCA use, but
 263 others may find it useful. */
 264 extern unsigned int machine_id;
 265 extern unsigned int machine_submodel_id;
 266 extern unsigned int BIOS_revision;
 267 extern unsigned int mca_pentium_flag;
 268
 269 /* Boot loader type from the setup header */
 270 extern int bootloader_type;
 271
 272 /*
 273  * User space process size: 3GB (default).
 274  */
 275 #define TASK_SIZE       (PAGE_OFFSET)
 276
 277 /* This decides where the kernel will search for a free chunk of vm
 278  * space during mmap's.
 279  */
 280 #define TASK_UNMAPPED_BASE      (PAGE_ALIGN(TASK_SIZE / 3))
 281
 282 #define HAVE_ARCH_PICK_MMAP_LAYOUT
 283
 284 /*
 285  * Size of io_bitmap.
 286  */
 287 #define IO_BITMAP_BITS  65536
 288 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
 289 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
 290 #define IO_BITMAP_OFFSET offsetof(struct tss_struct,io_bitmap)
 291 #define INVALID_IO_BITMAP_OFFSET 0x8000
 292 #define INVALID_IO_BITMAP_OFFSET_LAZY 0x9000
 293
 294 struct i387_fsave_struct {
 295         long    cwd;
 296         long    swd;
 297         long    twd;
 298         long    fip;
 299         long    fcs;
 300         long    foo;
 301         long    fos;
 302         long    st_space[20];   /* 8*10 bytes for each FP-reg = 80 bytes */
 303         long    status;         /* software status information */
 304 };
 305
 306 struct i387_fxsave_struct {
 307         unsigned short  cwd;
 308         unsigned short  swd;
 309         unsigned short  twd;
 310         unsigned short  fop;
 311         long    fip;
 312         long    fcs;
 313         long    foo;
 314         long    fos;
 315         long    mxcsr;
 316         long    mxcsr_mask;
 317         long    st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
 318         long    xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
 319         long    padding[56];
 320 } __attribute__ ((aligned (16)));
 321
 322 struct i387_soft_struct {
 323         long    cwd;
 324         long    swd;
 325         long    twd;
 326         long    fip;
 327         long    fcs;
 328         long    foo;
 329         long    fos;
 330         long    st_space[20];   /* 8*10 bytes for each FP-reg = 80 bytes */
 331         unsigned char   ftop, changed, lookahead, no_update, rm, alimit;
 332         struct info     *info;
 333         unsigned long   entry_eip;
 334 };
 335
 336 union i387_union {
 337         struct i387_fsave_struct        fsave;
 338         struct i387_fxsave_struct       fxsave;
 339         struct i387_soft_struct soft;
 340 };
 341
 342 typedef struct {
 343         unsigned long seg;
 344 } mm_segment_t;
 345
 346 struct thread_struct;
 347
 348 struct tss_struct {
 349         unsigned short  back_link,__blh;
 350         unsigned long   esp0;
 351         unsigned short  ss0,__ss0h;
 352         unsigned long   esp1;
 353         unsigned short  ss1,__ss1h;     /* ss1 is used to cache MSR_IA32_SYSENTER_CS */
 354         unsigned long   esp2;
 355         unsigned short  ss2,__ss2h;
 356         unsigned long   __cr3;
 357         unsigned long   eip;
 358         unsigned long   eflags;
 359         unsigned long   eax,ecx,edx,ebx;
 360         unsigned long   esp;
 361         unsigned long   ebp;
 362         unsigned long   esi;
 363         unsigned long   edi;
 364         unsigned short  es, __esh;
 365         unsigned short  cs, __csh;
 366         unsigned short  ss, __ssh;
 367         unsigned short  ds, __dsh;
 368         unsigned short  fs, __fsh;
 369         unsigned short  gs, __gsh;
 370         unsigned short  ldt, __ldth;
 371         unsigned short  trace, io_bitmap_base;
 372         /*
 373          * The extra 1 is there because the CPU will access an
 374          * additional byte beyond the end of the IO permission
 375          * bitmap. The extra byte must be all 1 bits, and must
 376          * be within the limit.
 377          */
 378         unsigned long   io_bitmap[IO_BITMAP_LONGS + 1];
 379         /*
 380          * Cache the current maximum and the last task that used the bitmap:
 381          */
 382         unsigned long io_bitmap_max;
 383         struct thread_struct *io_bitmap_owner;
 384         /*
 385          * pads the TSS to be cacheline-aligned (size is 0x100)
 386          */
 387         unsigned long __cacheline_filler[35];
 388         /*
 389          * .. and then another 0x100 bytes for emergency kernel stack
 390          */
 391         unsigned long stack[64];
 392 } __attribute__((packed));
 393
 394 #define ARCH_MIN_TASKALIGN      16
 395
 396 struct thread_struct {
 397 /* cached TLS descriptors. */
 398         struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
 399         unsigned long   esp0;
 400         unsigned long   sysenter_cs;
 401         unsigned long   eip;
 402         unsigned long   esp;
 403         unsigned long   fs;
 404         unsigned long   gs;
 405 /* Hardware debugging registers */
 406         unsigned long   debugreg[8];  /* %%db0-7 debug registers */
 407 /* fault info */
 408         unsigned long   cr2, trap_no, error_code;
 409 /* floating point info */
 410         union i387_union        i387;
 411 /* virtual 86 mode info */
 412         struct vm86_struct __user * vm86_info;
 413         unsigned long           screen_bitmap;
 414         unsigned long           v86flags, v86mask, saved_esp0;
 415         unsigned int            saved_fs, saved_gs;
 416 /* IO permissions */
 417         unsigned long   *io_bitmap_ptr;
 418         unsigned long   iopl;
 419 /* max allowed port in the bitmap, in bytes: */
 420         unsigned long   io_bitmap_max;
 421 };
 422
 423 #define INIT_THREAD  {                                                  \
 424         .esp0 = sizeof(init_stack) + (long)&init_stack,                 \
 425         .vm86_info = NULL,                                              \
 426         .sysenter_cs = __KERNEL_CS,                                     \
 427         .io_bitmap_ptr = NULL,                                          \
 428         .fs = __KERNEL_PDA,                                             \
 429 }
 430
 431 /*
 432  * Note that the .io_bitmap member must be extra-big. This is because
 433  * the CPU will access an additional byte beyond the end of the IO
 434  * permission bitmap. The extra byte must be all 1 bits, and must
 435  * be within the limit.
 436  */
 437 #define INIT_TSS  {                                                     \
 438         .esp0           = sizeof(init_stack) + (long)&init_stack,       \
 439         .ss0            = __KERNEL_DS,                                  \
 440         .ss1            = __KERNEL_CS,                                  \
 441         .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,                     \
 442         .io_bitmap      = { [ 0 ... IO_BITMAP_LONGS] = ~0 },            \
 443 }
 444
 445 #define start_thread(regs, new_eip, new_esp) do {               \
 446         __asm__("movl %0,%%gs": :"r" (0));                      \
 447         regs->xfs = 0;                                          \
 448         set_fs(USER_DS);                                        \
 449         regs->xds = __USER_DS;                                  \
 450         regs->xes = __USER_DS;                                  \
 451         regs->xss = __USER_DS;                                  \
 452         regs->xcs = __USER_CS;                                  \
 453         regs->eip = new_eip;                                    \
 454         regs->esp = new_esp;                                    \
 455 } while (0)
 456
 457 /* Forward declaration, a strange C thing */
 458 struct task_struct;
 459 struct mm_struct;
 460
 461 /* Free all resources held by a thread. */
 462 extern void release_thread(struct task_struct *);
 463
 464 /* Prepare to copy thread state - unlazy all lazy status */
 465 extern void prepare_to_copy(struct task_struct *tsk);
 466
 467 /*
 468  * create a kernel thread without removing it from tasklists
 469  */
 470 extern int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags);
 471
 472 extern unsigned long thread_saved_pc(struct task_struct *tsk);
 473 void show_trace(struct task_struct *task, struct pt_regs *regs, unsigned long *stack);
 474
 475 unsigned long get_wchan(struct task_struct *p);
 476
 477 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
 478 #define KSTK_TOP(info)                                                 \
 479 ({                                                                     \
 480        unsigned long *__ptr = (unsigned long *)(info);                 \
 481        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
 482 })
 483
 484 /*
 485  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
 486  * This is necessary to guarantee that the entire "struct pt_regs"
 487  * is accessable even if the CPU haven't stored the SS/ESP registers
 488  * on the stack (interrupt gate does not save these registers
 489  * when switching to the same priv ring).
 490  * Therefore beware: accessing the xss/esp fields of the
 491  * "struct pt_regs" is possible, but they may contain the
 492  * completely wrong values.
 493  */
 494 #define task_pt_regs(task)                                             \
 495 ({                                                                     \
 496        struct pt_regs *__regs__;                                       \
 497        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
 498        __regs__ - 1;                                                   \
 499 })
 500
 501 #define KSTK_EIP(task) (task_pt_regs(task)->eip)
 502 #define KSTK_ESP(task) (task_pt_regs(task)->esp)
 503
 504
 505 struct microcode_header {
 506         unsigned int hdrver;
 507         unsigned int rev;
 508         unsigned int date;
 509         unsigned int sig;
 510         unsigned int cksum;
 511         unsigned int ldrver;
 512         unsigned int pf;
 513         unsigned int datasize;
 514         unsigned int totalsize;
 515         unsigned int reserved[3];
 516 };
 517
 518 struct microcode {
 519         struct microcode_header hdr;
 520         unsigned int bits[0];
 521 };
 522
 523 typedef struct microcode microcode_t;
 524 typedef struct microcode_header microcode_header_t;
 525
 526 /* microcode format is extended from prescott processors */
 527 struct extended_signature {
 528         unsigned int sig;
 529         unsigned int pf;
 530         unsigned int cksum;
 531 };
 532
 533 struct extended_sigtable {
 534         unsigned int count;
 535         unsigned int cksum;
 536         unsigned int reserved[3];
 537         struct extended_signature sigs[0];
 538 };
 539
 540 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
 541 static inline void rep_nop(void)
 542 {
 543         __asm__ __volatile__("rep;nop": : :"memory");
 544 }
 545
 546 #define cpu_relax()     rep_nop()
 547
 548 #ifdef CONFIG_PARAVIRT
 549 #include <asm/paravirt.h>
 550 #else
 551 #define paravirt_enabled() 0
 552 #define __cpuid native_cpuid
 553
 554 static inline void load_esp0(struct tss_struct *tss, struct thread_struct *thread)
 555 {
 556         tss->esp0 = thread->esp0;
 557         /* This can only happen when SEP is enabled, no need to test "SEP"arately */
 558         if (unlikely(tss->ss1 != thread->sysenter_cs)) {
 559                 tss->ss1 = thread->sysenter_cs;
 560                 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
 561         }
 562 }
 563
 564 /*
 565  * These special macros can be used to get or set a debugging register
 566  */
 567 #define get_debugreg(var, register)                             \
 568                 __asm__("movl %%db" #register ", %0"            \
 569                         :"=r" (var))
 570 #define set_debugreg(value, register)                   \
 571                 __asm__("movl %0,%%db" #register                \
 572                         : /* no output */                       \
 573                         :"r" (value))
 574
 575 #define set_iopl_mask native_set_iopl_mask
 576 #endif /* CONFIG_PARAVIRT */
 577
 578 /*
 579  * Set IOPL bits in EFLAGS from given mask
 580  */
 581 static fastcall inline void native_set_iopl_mask(unsigned mask)
 582 {
 583         unsigned int reg;
 584         __asm__ __volatile__ ("pushfl;"
 585                               "popl %0;"
 586                               "andl %1, %0;"
 587                               "orl %2, %0;"
 588                               "pushl %0;"
 589                               "popfl"
 590                                 : "=&r" (reg)
 591                                 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
 592 }
 593
 594 /*
 595  * Generic CPUID function
 596  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
 597  * resulting in stale register contents being returned.
 598  */
 599 static inline void cpuid(unsigned int op, unsigned int *eax, unsigned int *ebx, unsigned int *ecx, unsigned int *edx)
 600 {
 601         *eax = op;
 602         *ecx = 0;
 603         __cpuid(eax, ebx, ecx, edx);
 604 }
 605
 606 /* Some CPUID calls want 'count' to be placed in ecx */
 607 static inline void cpuid_count(int op, int count, int *eax, int *ebx, int *ecx,
 608                                int *edx)
 609 {
 610         *eax = op;
 611         *ecx = count;
 612         __cpuid(eax, ebx, ecx, edx);
 613 }
 614
 615 /*
 616  * CPUID functions returning a single datum
 617  */
 618 static inline unsigned int cpuid_eax(unsigned int op)
 619 {
 620         unsigned int eax, ebx, ecx, edx;
 621
 622         cpuid(op, &eax, &ebx, &ecx, &edx);
 623         return eax;
 624 }
 625 static inline unsigned int cpuid_ebx(unsigned int op)
 626 {
 627         unsigned int eax, ebx, ecx, edx;
 628
 629         cpuid(op, &eax, &ebx, &ecx, &edx);
 630         return ebx;
 631 }
 632 static inline unsigned int cpuid_ecx(unsigned int op)
 633 {
 634         unsigned int eax, ebx, ecx, edx;
 635
 636         cpuid(op, &eax, &ebx, &ecx, &edx);
 637         return ecx;
 638 }
 639 static inline unsigned int cpuid_edx(unsigned int op)
 640 {
 641         unsigned int eax, ebx, ecx, edx;
 642
 643         cpuid(op, &eax, &ebx, &ecx, &edx);
 644         return edx;
 645 }
 646
 647 /* generic versions from gas */
 648 #define GENERIC_NOP1    ".byte 0x90\n"
 649 #define GENERIC_NOP2            ".byte 0x89,0xf6\n"
 650 #define GENERIC_NOP3        ".byte 0x8d,0x76,0x00\n"
 651 #define GENERIC_NOP4        ".byte 0x8d,0x74,0x26,0x00\n"
 652 #define GENERIC_NOP5        GENERIC_NOP1 GENERIC_NOP4
 653 #define GENERIC_NOP6    ".byte 0x8d,0xb6,0x00,0x00,0x00,0x00\n"
 654 #define GENERIC_NOP7    ".byte 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00\n"
 655 #define GENERIC_NOP8    GENERIC_NOP1 GENERIC_NOP7
 656
 657 /* Opteron nops */
 658 #define K8_NOP1 GENERIC_NOP1
 659 #define K8_NOP2 ".byte 0x66,0x90\n"
 660 #define K8_NOP3 ".byte 0x66,0x66,0x90\n"
 661 #define K8_NOP4 ".byte 0x66,0x66,0x66,0x90\n"
 662 #define K8_NOP5 K8_NOP3 K8_NOP2
 663 #define K8_NOP6 K8_NOP3 K8_NOP3
 664 #define K8_NOP7 K8_NOP4 K8_NOP3
 665 #define K8_NOP8 K8_NOP4 K8_NOP4
 666
 667 /* K7 nops */
 668 /* uses eax dependencies (arbitary choice) */
 669 #define K7_NOP1  GENERIC_NOP1
 670 #define K7_NOP2 ".byte 0x8b,0xc0\n"
 671 #define K7_NOP3 ".byte 0x8d,0x04,0x20\n"
 672 #define K7_NOP4 ".byte 0x8d,0x44,0x20,0x00\n"
 673 #define K7_NOP5 K7_NOP4 ASM_NOP1
 674 #define K7_NOP6 ".byte 0x8d,0x80,0,0,0,0\n"
 675 #define K7_NOP7        ".byte 0x8D,0x04,0x05,0,0,0,0\n"
 676 #define K7_NOP8        K7_NOP7 ASM_NOP1
 677
 678 #ifdef CONFIG_MK8
 679 #define ASM_NOP1 K8_NOP1
 680 #define ASM_NOP2 K8_NOP2
 681 #define ASM_NOP3 K8_NOP3
 682 #define ASM_NOP4 K8_NOP4
 683 #define ASM_NOP5 K8_NOP5
 684 #define ASM_NOP6 K8_NOP6
 685 #define ASM_NOP7 K8_NOP7
 686 #define ASM_NOP8 K8_NOP8
 687 #elif defined(CONFIG_MK7)
 688 #define ASM_NOP1 K7_NOP1
 689 #define ASM_NOP2 K7_NOP2
 690 #define ASM_NOP3 K7_NOP3
 691 #define ASM_NOP4 K7_NOP4
 692 #define ASM_NOP5 K7_NOP5
 693 #define ASM_NOP6 K7_NOP6
 694 #define ASM_NOP7 K7_NOP7
 695 #define ASM_NOP8 K7_NOP8
 696 #else
 697 #define ASM_NOP1 GENERIC_NOP1
 698 #define ASM_NOP2 GENERIC_NOP2
 699 #define ASM_NOP3 GENERIC_NOP3
 700 #define ASM_NOP4 GENERIC_NOP4
 701 #define ASM_NOP5 GENERIC_NOP5
 702 #define ASM_NOP6 GENERIC_NOP6
 703 #define ASM_NOP7 GENERIC_NOP7
 704 #define ASM_NOP8 GENERIC_NOP8
 705 #endif
 706
 707 #define ASM_NOP_MAX 8
 708
 709 /* Prefetch instructions for Pentium III and AMD Athlon */
 710 /* It's not worth to care about 3dnow! prefetches for the K6
 711    because they are microcoded there and very slow.
 712    However we don't do prefetches for pre XP Athlons currently
 713    That should be fixed. */
 714 #define ARCH_HAS_PREFETCH
 715 static inline void prefetch(const void *x)
 716 {
 717         alternative_input(ASM_NOP4,
 718                           "prefetchnta (%1)",
 719                           X86_FEATURE_XMM,
 720                           "r" (x));
 721 }
 722
 723 #define ARCH_HAS_PREFETCH
 724 #define ARCH_HAS_PREFETCHW
 725 #define ARCH_HAS_SPINLOCK_PREFETCH
 726
 727 /* 3dnow! prefetch to get an exclusive cache line. Useful for
 728    spinlocks to avoid one state transition in the cache coherency protocol. */
 729 static inline void prefetchw(const void *x)
 730 {
 731         alternative_input(ASM_NOP4,
 732                           "prefetchw (%1)",
 733                           X86_FEATURE_3DNOW,
 734                           "r" (x));
 735 }
 736 #define spin_lock_prefetch(x)   prefetchw(x)
 737
 738 extern void select_idle_routine(const struct cpuinfo_x86 *c);
 739
 740 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
 741
 742 extern unsigned long boot_option_idle_override;
 743 extern void enable_sep_cpu(void);
 744 extern int sysenter_setup(void);
 745
 746 extern int init_gdt(int cpu, struct task_struct *idle);
 747 extern void cpu_set_gdt(int);
 748 extern void secondary_cpu_init(void);
 749
 750 #endif /* __ASM_I386_PROCESSOR_H */