include/asm-x86/processor.h

   1 #ifndef __ASM_X86_PROCESSOR_H
   2 #define __ASM_X86_PROCESSOR_H
   3
   4 #include <asm/processor-flags.h>
   5
   6 /* migration helpers, for KVM - will be removed in 2.6.25: */
   7 #include <asm/vm86.h>
   8 #define Xgt_desc_struct desc_ptr
   9
  10 /* Forward declaration, a strange C thing */
  11 struct task_struct;
  12 struct mm_struct;
  13
  14 #include <asm/vm86.h>
  15 #include <asm/math_emu.h>
  16 #include <asm/segment.h>
  17 #include <asm/types.h>
  18 #include <asm/sigcontext.h>
  19 #include <asm/current.h>
  20 #include <asm/cpufeature.h>
  21 #include <asm/system.h>
  22 #include <asm/page.h>
  23 #include <asm/percpu.h>
  24 #include <asm/msr.h>
  25 #include <asm/desc_defs.h>
  26 #include <asm/nops.h>
  27 #include <linux/personality.h>
  28 #include <linux/cpumask.h>
  29 #include <linux/cache.h>
  30 #include <linux/threads.h>
  31 #include <linux/init.h>
  32
  33 /*
  34  * Default implementation of macro that returns current
  35  * instruction pointer ("program counter").
  36  */
  37 static inline void *current_text_addr(void)
  38 {
  39         void *pc;
  40         asm volatile("mov $1f,%0\n1:":"=r" (pc));
  41         return pc;
  42 }
  43
  44 #ifdef CONFIG_X86_VSMP
  45 #define ARCH_MIN_TASKALIGN      (1 << INTERNODE_CACHE_SHIFT)
  46 #define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
  47 #else
  48 #define ARCH_MIN_TASKALIGN      16
  49 #define ARCH_MIN_MMSTRUCT_ALIGN 0
  50 #endif
  51
  52 /*
  53  *  CPU type and hardware bug flags. Kept separately for each CPU.
  54  *  Members of this structure are referenced in head.S, so think twice
  55  *  before touching them. [mj]
  56  */
  57
  58 struct cpuinfo_x86 {
  59         __u8    x86;            /* CPU family */
  60         __u8    x86_vendor;     /* CPU vendor */
  61         __u8    x86_model;
  62         __u8    x86_mask;
  63 #ifdef CONFIG_X86_32
  64         char    wp_works_ok;    /* It doesn't on 386's */
  65         char    hlt_works_ok;   /* Problems on some 486Dx4's and old 386's */
  66         char    hard_math;
  67         char    rfu;
  68         char    fdiv_bug;
  69         char    f00f_bug;
  70         char    coma_bug;
  71         char    pad0;
  72 #else
  73         /* number of 4K pages in DTLB/ITLB combined(in pages)*/
  74         int     x86_tlbsize;
  75         __u8    x86_virt_bits, x86_phys_bits;
  76         /* cpuid returned core id bits */
  77         __u8    x86_coreid_bits;
  78         /* Max extended CPUID function supported */
  79         __u32   extended_cpuid_level;
  80 #endif
  81         int     cpuid_level;    /* Maximum supported CPUID level, -1=no CPUID */
  82         __u32   x86_capability[NCAPINTS];
  83         char    x86_vendor_id[16];
  84         char    x86_model_id[64];
  85         int     x86_cache_size;  /* in KB - valid for CPUS which support this
  86                                     call  */
  87         int     x86_cache_alignment;    /* In bytes */
  88         int     x86_power;
  89         unsigned long loops_per_jiffy;
  90 #ifdef CONFIG_SMP
  91         cpumask_t llc_shared_map;       /* cpus sharing the last level cache */
  92 #endif
  93         u16 x86_max_cores;              /* cpuid returned max cores value */
  94         u16 apicid;
  95         u16 x86_clflush_size;
  96 #ifdef CONFIG_SMP
  97         u16 booted_cores;               /* number of cores as seen by OS */
  98         u16 phys_proc_id;               /* Physical processor id. */
  99         u16 cpu_core_id;                /* Core id */
 100         u16 cpu_index;                  /* index into per_cpu list */
 101 #endif
 102 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
 103
 104 #define X86_VENDOR_INTEL 0
 105 #define X86_VENDOR_CYRIX 1
 106 #define X86_VENDOR_AMD 2
 107 #define X86_VENDOR_UMC 3
 108 #define X86_VENDOR_NEXGEN 4
 109 #define X86_VENDOR_CENTAUR 5
 110 #define X86_VENDOR_TRANSMETA 7
 111 #define X86_VENDOR_NSC 8
 112 #define X86_VENDOR_NUM 9
 113 #define X86_VENDOR_UNKNOWN 0xff
 114
 115 /*
 116  * capabilities of CPUs
 117  */
 118 extern struct cpuinfo_x86 boot_cpu_data;
 119 extern struct cpuinfo_x86 new_cpu_data;
 120 extern struct tss_struct doublefault_tss;
 121 extern __u32 cleared_cpu_caps[NCAPINTS];
 122
 123 #ifdef CONFIG_SMP
 124 DECLARE_PER_CPU(struct cpuinfo_x86, cpu_info);
 125 #define cpu_data(cpu)           per_cpu(cpu_info, cpu)
 126 #define current_cpu_data        cpu_data(smp_processor_id())
 127 #else
 128 #define cpu_data(cpu)           boot_cpu_data
 129 #define current_cpu_data        boot_cpu_data
 130 #endif
 131
 132 void cpu_detect(struct cpuinfo_x86 *c);
 133
 134 extern void identify_cpu(struct cpuinfo_x86 *);
 135 extern void identify_boot_cpu(void);
 136 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
 137 extern void print_cpu_info(struct cpuinfo_x86 *);
 138 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
 139 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
 140 extern unsigned short num_cache_leaves;
 141
 142 #if defined(CONFIG_X86_HT) || defined(CONFIG_X86_64)
 143 extern void detect_ht(struct cpuinfo_x86 *c);
 144 #else
 145 static inline void detect_ht(struct cpuinfo_x86 *c) {}
 146 #endif
 147
 148 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
 149                                          unsigned int *ecx, unsigned int *edx)
 150 {
 151         /* ecx is often an input as well as an output. */
 152         __asm__("cpuid"
 153                 : "=a" (*eax),
 154                   "=b" (*ebx),
 155                   "=c" (*ecx),
 156                   "=d" (*edx)
 157                 : "0" (*eax), "2" (*ecx));
 158 }
 159
 160 static inline void load_cr3(pgd_t *pgdir)
 161 {
 162         write_cr3(__pa(pgdir));
 163 }
 164
 165 #ifdef CONFIG_X86_32
 166 /* This is the TSS defined by the hardware. */
 167 struct x86_hw_tss {
 168         unsigned short  back_link, __blh;
 169         unsigned long   sp0;
 170         unsigned short  ss0, __ss0h;
 171         unsigned long   sp1;
 172         unsigned short  ss1, __ss1h;    /* ss1 caches MSR_IA32_SYSENTER_CS */
 173         unsigned long   sp2;
 174         unsigned short  ss2, __ss2h;
 175         unsigned long   __cr3;
 176         unsigned long   ip;
 177         unsigned long   flags;
 178         unsigned long   ax, cx, dx, bx;
 179         unsigned long   sp, bp, si, di;
 180         unsigned short  es, __esh;
 181         unsigned short  cs, __csh;
 182         unsigned short  ss, __ssh;
 183         unsigned short  ds, __dsh;
 184         unsigned short  fs, __fsh;
 185         unsigned short  gs, __gsh;
 186         unsigned short  ldt, __ldth;
 187         unsigned short  trace, io_bitmap_base;
 188 } __attribute__((packed));
 189 #else
 190 struct x86_hw_tss {
 191         u32 reserved1;
 192         u64 sp0;
 193         u64 sp1;
 194         u64 sp2;
 195         u64 reserved2;
 196         u64 ist[7];
 197         u32 reserved3;
 198         u32 reserved4;
 199         u16 reserved5;
 200         u16 io_bitmap_base;
 201 } __attribute__((packed)) ____cacheline_aligned;
 202 #endif
 203
 204 /*
 205  * Size of io_bitmap.
 206  */
 207 #define IO_BITMAP_BITS  65536
 208 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
 209 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
 210 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
 211 #define INVALID_IO_BITMAP_OFFSET 0x8000
 212 #define INVALID_IO_BITMAP_OFFSET_LAZY 0x9000
 213
 214 struct tss_struct {
 215         struct x86_hw_tss x86_tss;
 216
 217         /*
 218          * The extra 1 is there because the CPU will access an
 219          * additional byte beyond the end of the IO permission
 220          * bitmap. The extra byte must be all 1 bits, and must
 221          * be within the limit.
 222          */
 223         unsigned long   io_bitmap[IO_BITMAP_LONGS + 1];
 224         /*
 225          * Cache the current maximum and the last task that used the bitmap:
 226          */
 227         unsigned long io_bitmap_max;
 228         struct thread_struct *io_bitmap_owner;
 229         /*
 230          * pads the TSS to be cacheline-aligned (size is 0x100)
 231          */
 232         unsigned long __cacheline_filler[35];
 233         /*
 234          * .. and then another 0x100 bytes for emergency kernel stack
 235          */
 236         unsigned long stack[64];
 237 } __attribute__((packed));
 238
 239 DECLARE_PER_CPU(struct tss_struct, init_tss);
 240
 241 /* Save the original ist values for checking stack pointers during debugging */
 242 struct orig_ist {
 243         unsigned long ist[7];
 244 };
 245
 246 #define MXCSR_DEFAULT           0x1f80
 247
 248 struct i387_fsave_struct {
 249         u32     cwd;
 250         u32     swd;
 251         u32     twd;
 252         u32     fip;
 253         u32     fcs;
 254         u32     foo;
 255         u32     fos;
 256         u32     st_space[20];   /* 8*10 bytes for each FP-reg = 80 bytes */
 257         u32     status;         /* software status information */
 258 };
 259
 260 struct i387_fxsave_struct {
 261         u16     cwd;
 262         u16     swd;
 263         u16     twd;
 264         u16     fop;
 265         union {
 266                 struct {
 267                         u64     rip;
 268                         u64     rdp;
 269                 };
 270                 struct {
 271                         u32     fip;
 272                         u32     fcs;
 273                         u32     foo;
 274                         u32     fos;
 275                 };
 276         };
 277         u32     mxcsr;
 278         u32     mxcsr_mask;
 279         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
 280         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
 281         u32     padding[24];
 282 } __attribute__((aligned(16)));
 283
 284 struct i387_soft_struct {
 285         u32     cwd;
 286         u32     swd;
 287         u32     twd;
 288         u32     fip;
 289         u32     fcs;
 290         u32     foo;
 291         u32     fos;
 292         u32     st_space[20];   /* 8*10 bytes for each FP-reg = 80 bytes */
 293         u8      ftop, changed, lookahead, no_update, rm, alimit;
 294         struct info     *info;
 295         u32     entry_eip;
 296 };
 297
 298 union i387_union {
 299         struct i387_fsave_struct        fsave;
 300         struct i387_fxsave_struct       fxsave;
 301         struct i387_soft_struct         soft;
 302 };
 303
 304 #ifdef CONFIG_X86_32
 305 /*
 306  * the following now lives in the per cpu area:
 307  * extern       int cpu_llc_id[NR_CPUS];
 308  */
 309 DECLARE_PER_CPU(u8, cpu_llc_id);
 310 #else
 311 DECLARE_PER_CPU(struct orig_ist, orig_ist);
 312 #endif
 313
 314 extern void print_cpu_info(struct cpuinfo_x86 *);
 315 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
 316 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
 317 extern unsigned short num_cache_leaves;
 318
 319 struct thread_struct {
 320 /* cached TLS descriptors. */
 321         struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
 322         unsigned long   sp0;
 323         unsigned long   sp;
 324 #ifdef CONFIG_X86_32
 325         unsigned long   sysenter_cs;
 326 #else
 327         unsigned long   usersp; /* Copy from PDA */
 328         unsigned short  es, ds, fsindex, gsindex;
 329 #endif
 330         unsigned long   ip;
 331         unsigned long   fs;
 332         unsigned long   gs;
 333 /* Hardware debugging registers */
 334         unsigned long   debugreg0;
 335         unsigned long   debugreg1;
 336         unsigned long   debugreg2;
 337         unsigned long   debugreg3;
 338         unsigned long   debugreg6;
 339         unsigned long   debugreg7;
 340 /* fault info */
 341         unsigned long   cr2, trap_no, error_code;
 342 /* floating point info */
 343         union i387_union        i387 __attribute__((aligned(16)));;
 344 #ifdef CONFIG_X86_32
 345 /* virtual 86 mode info */
 346         struct vm86_struct __user *vm86_info;
 347         unsigned long           screen_bitmap;
 348         unsigned long           v86flags, v86mask, saved_sp0;
 349         unsigned int            saved_fs, saved_gs;
 350 #endif
 351 /* IO permissions */
 352         unsigned long   *io_bitmap_ptr;
 353         unsigned long   iopl;
 354 /* max allowed port in the bitmap, in bytes: */
 355         unsigned io_bitmap_max;
 356 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
 357         unsigned long   debugctlmsr;
 358 /* Debug Store - if not 0 points to a DS Save Area configuration;
 359  *               goes into MSR_IA32_DS_AREA */
 360         unsigned long   ds_area_msr;
 361 };
 362
 363 static inline unsigned long native_get_debugreg(int regno)
 364 {
 365         unsigned long val = 0;  /* Damn you, gcc! */
 366
 367         switch (regno) {
 368         case 0:
 369                 asm("mov %%db0, %0" :"=r" (val)); break;
 370         case 1:
 371                 asm("mov %%db1, %0" :"=r" (val)); break;
 372         case 2:
 373                 asm("mov %%db2, %0" :"=r" (val)); break;
 374         case 3:
 375                 asm("mov %%db3, %0" :"=r" (val)); break;
 376         case 6:
 377                 asm("mov %%db6, %0" :"=r" (val)); break;
 378         case 7:
 379                 asm("mov %%db7, %0" :"=r" (val)); break;
 380         default:
 381                 BUG();
 382         }
 383         return val;
 384 }
 385
 386 static inline void native_set_debugreg(int regno, unsigned long value)
 387 {
 388         switch (regno) {
 389         case 0:
 390                 asm("mov %0,%%db0"      : /* no output */ :"r" (value));
 391                 break;
 392         case 1:
 393                 asm("mov %0,%%db1"      : /* no output */ :"r" (value));
 394                 break;
 395         case 2:
 396                 asm("mov %0,%%db2"      : /* no output */ :"r" (value));
 397                 break;
 398         case 3:
 399                 asm("mov %0,%%db3"      : /* no output */ :"r" (value));
 400                 break;
 401         case 6:
 402                 asm("mov %0,%%db6"      : /* no output */ :"r" (value));
 403                 break;
 404         case 7:
 405                 asm("mov %0,%%db7"      : /* no output */ :"r" (value));
 406                 break;
 407         default:
 408                 BUG();
 409         }
 410 }
 411
 412 /*
 413  * Set IOPL bits in EFLAGS from given mask
 414  */
 415 static inline void native_set_iopl_mask(unsigned mask)
 416 {
 417 #ifdef CONFIG_X86_32
 418         unsigned int reg;
 419         __asm__ __volatile__ ("pushfl;"
 420                               "popl %0;"
 421                               "andl %1, %0;"
 422                               "orl %2, %0;"
 423                               "pushl %0;"
 424                               "popfl"
 425                                 : "=&r" (reg)
 426                                 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
 427 #endif
 428 }
 429
 430 static inline void native_load_sp0(struct tss_struct *tss,
 431                                    struct thread_struct *thread)
 432 {
 433         tss->x86_tss.sp0 = thread->sp0;
 434 #ifdef CONFIG_X86_32
 435         /* Only happens when SEP is enabled, no need to test "SEP"arately */
 436         if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
 437                 tss->x86_tss.ss1 = thread->sysenter_cs;
 438                 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
 439         }
 440 #endif
 441 }
 442
 443 static inline void native_swapgs(void)
 444 {
 445 #ifdef CONFIG_X86_64
 446         asm volatile("swapgs" ::: "memory");
 447 #endif
 448 }
 449
 450 #ifdef CONFIG_PARAVIRT
 451 #include <asm/paravirt.h>
 452 #else
 453 #define __cpuid native_cpuid
 454 #define paravirt_enabled() 0
 455
 456 /*
 457  * These special macros can be used to get or set a debugging register
 458  */
 459 #define get_debugreg(var, register)                             \
 460         (var) = native_get_debugreg(register)
 461 #define set_debugreg(value, register)                           \
 462         native_set_debugreg(register, value)
 463
 464 static inline void load_sp0(struct tss_struct *tss,
 465                             struct thread_struct *thread)
 466 {
 467         native_load_sp0(tss, thread);
 468 }
 469
 470 #define set_iopl_mask native_set_iopl_mask
 471 #define SWAPGS  swapgs
 472 #endif /* CONFIG_PARAVIRT */
 473
 474 /*
 475  * Save the cr4 feature set we're using (ie
 476  * Pentium 4MB enable and PPro Global page
 477  * enable), so that any CPU's that boot up
 478  * after us can get the correct flags.
 479  */
 480 extern unsigned long mmu_cr4_features;
 481
 482 static inline void set_in_cr4(unsigned long mask)
 483 {
 484         unsigned cr4;
 485         mmu_cr4_features |= mask;
 486         cr4 = read_cr4();
 487         cr4 |= mask;
 488         write_cr4(cr4);
 489 }
 490
 491 static inline void clear_in_cr4(unsigned long mask)
 492 {
 493         unsigned cr4;
 494         mmu_cr4_features &= ~mask;
 495         cr4 = read_cr4();
 496         cr4 &= ~mask;
 497         write_cr4(cr4);
 498 }
 499
 500 struct microcode_header {
 501         unsigned int hdrver;
 502         unsigned int rev;
 503         unsigned int date;
 504         unsigned int sig;
 505         unsigned int cksum;
 506         unsigned int ldrver;
 507         unsigned int pf;
 508         unsigned int datasize;
 509         unsigned int totalsize;
 510         unsigned int reserved[3];
 511 };
 512
 513 struct microcode {
 514         struct microcode_header hdr;
 515         unsigned int bits[0];
 516 };
 517
 518 typedef struct microcode microcode_t;
 519 typedef struct microcode_header microcode_header_t;
 520
 521 /* microcode format is extended from prescott processors */
 522 struct extended_signature {
 523         unsigned int sig;
 524         unsigned int pf;
 525         unsigned int cksum;
 526 };
 527
 528 struct extended_sigtable {
 529         unsigned int count;
 530         unsigned int cksum;
 531         unsigned int reserved[3];
 532         struct extended_signature sigs[0];
 533 };
 534
 535 typedef struct {
 536         unsigned long seg;
 537 } mm_segment_t;
 538
 539
 540 /*
 541  * create a kernel thread without removing it from tasklists
 542  */
 543 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
 544
 545 /* Free all resources held by a thread. */
 546 extern void release_thread(struct task_struct *);
 547
 548 /* Prepare to copy thread state - unlazy all lazy status */
 549 extern void prepare_to_copy(struct task_struct *tsk);
 550
 551 unsigned long get_wchan(struct task_struct *p);
 552
 553 /*
 554  * Generic CPUID function
 555  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
 556  * resulting in stale register contents being returned.
 557  */
 558 static inline void cpuid(unsigned int op,
 559                          unsigned int *eax, unsigned int *ebx,
 560                          unsigned int *ecx, unsigned int *edx)
 561 {
 562         *eax = op;
 563         *ecx = 0;
 564         __cpuid(eax, ebx, ecx, edx);
 565 }
 566
 567 /* Some CPUID calls want 'count' to be placed in ecx */
 568 static inline void cpuid_count(unsigned int op, int count,
 569                                unsigned int *eax, unsigned int *ebx,
 570                                unsigned int *ecx, unsigned int *edx)
 571 {
 572         *eax = op;
 573         *ecx = count;
 574         __cpuid(eax, ebx, ecx, edx);
 575 }
 576
 577 /*
 578  * CPUID functions returning a single datum
 579  */
 580 static inline unsigned int cpuid_eax(unsigned int op)
 581 {
 582         unsigned int eax, ebx, ecx, edx;
 583
 584         cpuid(op, &eax, &ebx, &ecx, &edx);
 585         return eax;
 586 }
 587 static inline unsigned int cpuid_ebx(unsigned int op)
 588 {
 589         unsigned int eax, ebx, ecx, edx;
 590
 591         cpuid(op, &eax, &ebx, &ecx, &edx);
 592         return ebx;
 593 }
 594 static inline unsigned int cpuid_ecx(unsigned int op)
 595 {
 596         unsigned int eax, ebx, ecx, edx;
 597
 598         cpuid(op, &eax, &ebx, &ecx, &edx);
 599         return ecx;
 600 }
 601 static inline unsigned int cpuid_edx(unsigned int op)
 602 {
 603         unsigned int eax, ebx, ecx, edx;
 604
 605         cpuid(op, &eax, &ebx, &ecx, &edx);
 606         return edx;
 607 }
 608
 609 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
 610 static inline void rep_nop(void)
 611 {
 612         __asm__ __volatile__("rep;nop": : :"memory");
 613 }
 614
 615 /* Stop speculative execution */
 616 static inline void sync_core(void)
 617 {
 618         int tmp;
 619         asm volatile("cpuid" : "=a" (tmp) : "0" (1)
 620                                           : "ebx", "ecx", "edx", "memory");
 621 }
 622
 623 #define cpu_relax()   rep_nop()
 624
 625 static inline void __monitor(const void *eax, unsigned long ecx,
 626                 unsigned long edx)
 627 {
 628         /* "monitor %eax,%ecx,%edx;" */
 629         asm volatile(
 630                 ".byte 0x0f,0x01,0xc8;"
 631                 : :"a" (eax), "c" (ecx), "d"(edx));
 632 }
 633
 634 static inline void __mwait(unsigned long eax, unsigned long ecx)
 635 {
 636         /* "mwait %eax,%ecx;" */
 637         asm volatile(
 638                 ".byte 0x0f,0x01,0xc9;"
 639                 : :"a" (eax), "c" (ecx));
 640 }
 641
 642 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
 643 {
 644         /* "mwait %eax,%ecx;" */
 645         asm volatile(
 646                 "sti; .byte 0x0f,0x01,0xc9;"
 647                 : :"a" (eax), "c" (ecx));
 648 }
 649
 650 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
 651
 652 extern int force_mwait;
 653
 654 extern void select_idle_routine(const struct cpuinfo_x86 *c);
 655
 656 extern unsigned long boot_option_idle_override;
 657
 658 extern void enable_sep_cpu(void);
 659 extern int sysenter_setup(void);
 660
 661 /* Defined in head.S */
 662 extern struct desc_ptr early_gdt_descr;
 663
 664 extern void cpu_set_gdt(int);
 665 extern void switch_to_new_gdt(void);
 666 extern void cpu_init(void);
 667 extern void init_gdt(int cpu);
 668
 669 /* from system description table in BIOS.  Mostly for MCA use, but
 670  * others may find it useful. */
 671 extern unsigned int machine_id;
 672 extern unsigned int machine_submodel_id;
 673 extern unsigned int BIOS_revision;
 674 extern unsigned int mca_pentium_flag;
 675
 676 /* Boot loader type from the setup header */
 677 extern int bootloader_type;
 678
 679 extern char ignore_fpu_irq;
 680 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
 681
 682 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
 683 #define ARCH_HAS_PREFETCHW
 684 #define ARCH_HAS_SPINLOCK_PREFETCH
 685
 686 #ifdef CONFIG_X86_32
 687 #define BASE_PREFETCH   ASM_NOP4
 688 #define ARCH_HAS_PREFETCH
 689 #else
 690 #define BASE_PREFETCH   "prefetcht0 (%1)"
 691 #endif
 692
 693 /* Prefetch instructions for Pentium III and AMD Athlon */
 694 /* It's not worth to care about 3dnow! prefetches for the K6
 695    because they are microcoded there and very slow.
 696    However we don't do prefetches for pre XP Athlons currently
 697    That should be fixed. */
 698 static inline void prefetch(const void *x)
 699 {
 700         alternative_input(BASE_PREFETCH,
 701                           "prefetchnta (%1)",
 702                           X86_FEATURE_XMM,
 703                           "r" (x));
 704 }
 705
 706 /* 3dnow! prefetch to get an exclusive cache line. Useful for
 707    spinlocks to avoid one state transition in the cache coherency protocol. */
 708 static inline void prefetchw(const void *x)
 709 {
 710         alternative_input(BASE_PREFETCH,
 711                           "prefetchw (%1)",
 712                           X86_FEATURE_3DNOW,
 713                           "r" (x));
 714 }
 715
 716 #define spin_lock_prefetch(x)   prefetchw(x)
 717 #ifdef CONFIG_X86_32
 718 /*
 719  * User space process size: 3GB (default).
 720  */
 721 #define TASK_SIZE       (PAGE_OFFSET)
 722
 723 #define INIT_THREAD  {                                                  \
 724         .sp0 = sizeof(init_stack) + (long)&init_stack,                  \
 725         .vm86_info = NULL,                                              \
 726         .sysenter_cs = __KERNEL_CS,                                     \
 727         .io_bitmap_ptr = NULL,                                          \
 728         .fs = __KERNEL_PERCPU,                                          \
 729 }
 730
 731 /*
 732  * Note that the .io_bitmap member must be extra-big. This is because
 733  * the CPU will access an additional byte beyond the end of the IO
 734  * permission bitmap. The extra byte must be all 1 bits, and must
 735  * be within the limit.
 736  */
 737 #define INIT_TSS  {                                                     \
 738         .x86_tss = {                                                    \
 739                 .sp0            = sizeof(init_stack) + (long)&init_stack, \
 740                 .ss0            = __KERNEL_DS,                          \
 741                 .ss1            = __KERNEL_CS,                          \
 742                 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,             \
 743          },                                                             \
 744         .io_bitmap      = { [0 ... IO_BITMAP_LONGS] = ~0 },             \
 745 }
 746
 747 #define start_thread(regs, new_eip, new_esp) do {               \
 748         __asm__("movl %0,%%gs": :"r" (0));                      \
 749         regs->fs = 0;                                           \
 750         set_fs(USER_DS);                                        \
 751         regs->ds = __USER_DS;                                   \
 752         regs->es = __USER_DS;                                   \
 753         regs->ss = __USER_DS;                                   \
 754         regs->cs = __USER_CS;                                   \
 755         regs->ip = new_eip;                                     \
 756         regs->sp = new_esp;                                     \
 757 } while (0)
 758
 759
 760 extern unsigned long thread_saved_pc(struct task_struct *tsk);
 761
 762 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
 763 #define KSTK_TOP(info)                                                 \
 764 ({                                                                     \
 765        unsigned long *__ptr = (unsigned long *)(info);                 \
 766        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
 767 })
 768
 769 /*
 770  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
 771  * This is necessary to guarantee that the entire "struct pt_regs"
 772  * is accessable even if the CPU haven't stored the SS/ESP registers
 773  * on the stack (interrupt gate does not save these registers
 774  * when switching to the same priv ring).
 775  * Therefore beware: accessing the ss/esp fields of the
 776  * "struct pt_regs" is possible, but they may contain the
 777  * completely wrong values.
 778  */
 779 #define task_pt_regs(task)                                             \
 780 ({                                                                     \
 781        struct pt_regs *__regs__;                                       \
 782        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
 783        __regs__ - 1;                                                   \
 784 })
 785
 786 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
 787
 788 #else
 789 /*
 790  * User space process size. 47bits minus one guard page.
 791  */
 792 #define TASK_SIZE64     (0x800000000000UL - 4096)
 793
 794 /* This decides where the kernel will search for a free chunk of vm
 795  * space during mmap's.
 796  */
 797 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
 798                            0xc0000000 : 0xFFFFe000)
 799
 800 #define TASK_SIZE               (test_thread_flag(TIF_IA32) ? \
 801                                  IA32_PAGE_OFFSET : TASK_SIZE64)
 802 #define TASK_SIZE_OF(child)     ((test_tsk_thread_flag(child, TIF_IA32)) ? \
 803                                   IA32_PAGE_OFFSET : TASK_SIZE64)
 804
 805 #define INIT_THREAD  { \
 806         .sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
 807 }
 808
 809 #define INIT_TSS  { \
 810         .x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
 811 }
 812
 813 #define start_thread(regs, new_rip, new_rsp) do {                            \
 814         asm volatile("movl %0,%%fs; movl %0,%%es; movl %0,%%ds": :"r" (0));  \
 815         load_gs_index(0);                                                    \
 816         (regs)->ip = (new_rip);                                              \
 817         (regs)->sp = (new_rsp);                                              \
 818         write_pda(oldrsp, (new_rsp));                                        \
 819         (regs)->cs = __USER_CS;                                              \
 820         (regs)->ss = __USER_DS;                                              \
 821         (regs)->flags = 0x200;                                               \
 822         set_fs(USER_DS);                                                     \
 823 } while (0)
 824
 825 /*
 826  * Return saved PC of a blocked thread.
 827  * What is this good for? it will be always the scheduler or ret_from_fork.
 828  */
 829 #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8))
 830
 831 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
 832 #define KSTK_ESP(tsk) -1 /* sorry. doesn't work for syscall. */
 833 #endif /* CONFIG_X86_64 */
 834
 835 /* This decides where the kernel will search for a free chunk of vm
 836  * space during mmap's.
 837  */
 838 #define TASK_UNMAPPED_BASE      (PAGE_ALIGN(TASK_SIZE / 3))
 839
 840 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
 841
 842 #endif