arch/i386/kernel/i387.c

   1 /*
   2  *  linux/arch/i386/kernel/i387.c
   3  *
   4  *  Copyright (C) 1994 Linus Torvalds
   5  *
   6  *  Pentium III FXSR, SSE support
   7  *  General FPU state handling cleanups
   8  *      Gareth Hughes <gareth@valinux.com>, May 2000
   9  */
  10
  11 #include <linux/config.h>
  12 #include <linux/sched.h>
  13 #include <asm/processor.h>
  14 #include <asm/i387.h>
  15 #include <asm/math_emu.h>
  16 #include <asm/sigcontext.h>
  17 #include <asm/user.h>
  18 #include <asm/ptrace.h>
  19 #include <asm/uaccess.h>
  20
  21 #ifdef CONFIG_MATH_EMULATION
  22 #define HAVE_HWFP (boot_cpu_data.hard_math)
  23 #else
  24 #define HAVE_HWFP 1
  25 #endif
  26
  27 unsigned long mxcsr_feature_mask = 0xffffffff;
  28
  29 void mxcsr_feature_mask_init(void)
  30 {
  31         unsigned long mask = 0;
  32         clts();
  33         if (cpu_has_fxsr) {
  34                 memset(&current->thread.i387.fxsave, 0, sizeof(struct i387_fxsave_struct));
  35                 asm volatile("fxsave %0" : : "m" (current->thread.i387.fxsave));
  36                 mask = current->thread.i387.fxsave.mxcsr_mask;
  37                 if (mask == 0) mask = 0x0000ffbf;
  38         }
  39         mxcsr_feature_mask &= mask;
  40         stts();
  41 }
  42
  43 /*
  44  * The _current_ task is using the FPU for the first time
  45  * so initialize it and set the mxcsr to its default
  46  * value at reset if we support XMM instructions and then
  47  * remeber the current task has used the FPU.
  48  */
  49 void init_fpu(struct task_struct *tsk)
  50 {
  51         if (cpu_has_fxsr) {
  52                 memset(&tsk->thread.i387.fxsave, 0, sizeof(struct i387_fxsave_struct));
  53                 tsk->thread.i387.fxsave.cwd = 0x37f;
  54                 if (cpu_has_xmm)
  55                         tsk->thread.i387.fxsave.mxcsr = 0x1f80;
  56         } else {
  57                 memset(&tsk->thread.i387.fsave, 0, sizeof(struct i387_fsave_struct));
  58                 tsk->thread.i387.fsave.cwd = 0xffff037fu;
  59                 tsk->thread.i387.fsave.swd = 0xffff0000u;
  60                 tsk->thread.i387.fsave.twd = 0xffffffffu;
  61                 tsk->thread.i387.fsave.fos = 0xffff0000u;
  62         }
  63         tsk->used_math = 1;
  64 }
  65
  66 /*
  67  * FPU lazy state save handling.
  68  */
  69
  70 void kernel_fpu_begin(void)
  71 {
  72         struct thread_info *thread = current_thread_info();
  73
  74         preempt_disable();
  75         if (thread->status & TS_USEDFPU) {
  76                 __save_init_fpu(thread->task);
  77                 return;
  78         }
  79         clts();
  80 }
  81
  82 void restore_fpu( struct task_struct *tsk )
  83 {
  84         if ( cpu_has_fxsr ) {
  85                 asm volatile( "fxrstor %0"
  86                               : : "m" (tsk->thread.i387.fxsave) );
  87         } else {
  88                 asm volatile( "frstor %0"
  89                               : : "m" (tsk->thread.i387.fsave) );
  90         }
  91 }
  92
  93 /*
  94  * FPU tag word conversions.
  95  */
  96
  97 static inline unsigned short twd_i387_to_fxsr( unsigned short twd )
  98 {
  99         unsigned int tmp; /* to avoid 16 bit prefixes in the code */
 100
 101         /* Transform each pair of bits into 01 (valid) or 00 (empty) */
 102         tmp = ~twd;
 103         tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */
 104         /* and move the valid bits to the lower byte. */
 105         tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */
 106         tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */
 107         tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */
 108         return tmp;
 109 }
 110
 111 static inline unsigned long twd_fxsr_to_i387( struct i387_fxsave_struct *fxsave )
 112 {
 113         struct _fpxreg *st = NULL;
 114         unsigned long twd = (unsigned long) fxsave->twd;
 115         unsigned long tag;
 116         unsigned long ret = 0xffff0000u;
 117         int i;
 118
 119 #define FPREG_ADDR(f, n)        ((char *)&(f)->st_space + (n) * 16);
 120
 121         for ( i = 0 ; i < 8 ; i++ ) {
 122                 if ( twd & 0x1 ) {
 123                         st = (struct _fpxreg *) FPREG_ADDR( fxsave, i );
 124
 125                         switch ( st->exponent & 0x7fff ) {
 126                         case 0x7fff:
 127                                 tag = 2;                /* Special */
 128                                 break;
 129                         case 0x0000:
 130                                 if ( !st->significand[0] &&
 131                                      !st->significand[1] &&
 132                                      !st->significand[2] &&
 133                                      !st->significand[3] ) {
 134                                         tag = 1;        /* Zero */
 135                                 } else {
 136                                         tag = 2;        /* Special */
 137                                 }
 138                                 break;
 139                         default:
 140                                 if ( st->significand[3] & 0x8000 ) {
 141                                         tag = 0;        /* Valid */
 142                                 } else {
 143                                         tag = 2;        /* Special */
 144                                 }
 145                                 break;
 146                         }
 147                 } else {
 148                         tag = 3;                        /* Empty */
 149                 }
 150                 ret |= (tag << (2 * i));
 151                 twd = twd >> 1;
 152         }
 153         return ret;
 154 }
 155
 156 /*
 157  * FPU state interaction.
 158  */
 159
 160 unsigned short get_fpu_cwd( struct task_struct *tsk )
 161 {
 162         if ( cpu_has_fxsr ) {
 163                 return tsk->thread.i387.fxsave.cwd;
 164         } else {
 165                 return (unsigned short)tsk->thread.i387.fsave.cwd;
 166         }
 167 }
 168
 169 unsigned short get_fpu_swd( struct task_struct *tsk )
 170 {
 171         if ( cpu_has_fxsr ) {
 172                 return tsk->thread.i387.fxsave.swd;
 173         } else {
 174                 return (unsigned short)tsk->thread.i387.fsave.swd;
 175         }
 176 }
 177
 178 unsigned short get_fpu_twd( struct task_struct *tsk )
 179 {
 180         if ( cpu_has_fxsr ) {
 181                 return tsk->thread.i387.fxsave.twd;
 182         } else {
 183                 return (unsigned short)tsk->thread.i387.fsave.twd;
 184         }
 185 }
 186
 187 unsigned short get_fpu_mxcsr( struct task_struct *tsk )
 188 {
 189         if ( cpu_has_xmm ) {
 190                 return tsk->thread.i387.fxsave.mxcsr;
 191         } else {
 192                 return 0x1f80;
 193         }
 194 }
 195
 196 void set_fpu_cwd( struct task_struct *tsk, unsigned short cwd )
 197 {
 198         if ( cpu_has_fxsr ) {
 199                 tsk->thread.i387.fxsave.cwd = cwd;
 200         } else {
 201                 tsk->thread.i387.fsave.cwd = ((long)cwd | 0xffff0000u);
 202         }
 203 }
 204
 205 void set_fpu_swd( struct task_struct *tsk, unsigned short swd )
 206 {
 207         if ( cpu_has_fxsr ) {
 208                 tsk->thread.i387.fxsave.swd = swd;
 209         } else {
 210                 tsk->thread.i387.fsave.swd = ((long)swd | 0xffff0000u);
 211         }
 212 }
 213
 214 void set_fpu_twd( struct task_struct *tsk, unsigned short twd )
 215 {
 216         if ( cpu_has_fxsr ) {
 217                 tsk->thread.i387.fxsave.twd = twd_i387_to_fxsr(twd);
 218         } else {
 219                 tsk->thread.i387.fsave.twd = ((long)twd | 0xffff0000u);
 220         }
 221 }
 222
 223 /*
 224  * FXSR floating point environment conversions.
 225  */
 226
 227 static int convert_fxsr_to_user( struct _fpstate __user *buf,
 228                                         struct i387_fxsave_struct *fxsave )
 229 {
 230         unsigned long env[7];
 231         struct _fpreg __user *to;
 232         struct _fpxreg *from;
 233         int i;
 234
 235         env[0] = (unsigned long)fxsave->cwd | 0xffff0000ul;
 236         env[1] = (unsigned long)fxsave->swd | 0xffff0000ul;
 237         env[2] = twd_fxsr_to_i387(fxsave);
 238         env[3] = fxsave->fip;
 239         env[4] = fxsave->fcs | ((unsigned long)fxsave->fop << 16);
 240         env[5] = fxsave->foo;
 241         env[6] = fxsave->fos;
 242
 243         if ( __copy_to_user( buf, env, 7 * sizeof(unsigned long) ) )
 244                 return 1;
 245
 246         to = &buf->_st[0];
 247         from = (struct _fpxreg *) &fxsave->st_space[0];
 248         for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
 249                 unsigned long __user *t = (unsigned long __user *)to;
 250                 unsigned long *f = (unsigned long *)from;
 251
 252                 if (__put_user(*f, t) ||
 253                                 __put_user(*(f + 1), t + 1) ||
 254                                 __put_user(from->exponent, &to->exponent))
 255                         return 1;
 256         }
 257         return 0;
 258 }
 259
 260 static int convert_fxsr_from_user( struct i387_fxsave_struct *fxsave,
 261                                           struct _fpstate __user *buf )
 262 {
 263         unsigned long env[7];
 264         struct _fpxreg *to;
 265         struct _fpreg __user *from;
 266         int i;
 267
 268         if ( __copy_from_user( env, buf, 7 * sizeof(long) ) )
 269                 return 1;
 270
 271         fxsave->cwd = (unsigned short)(env[0] & 0xffff);
 272         fxsave->swd = (unsigned short)(env[1] & 0xffff);
 273         fxsave->twd = twd_i387_to_fxsr((unsigned short)(env[2] & 0xffff));
 274         fxsave->fip = env[3];
 275         fxsave->fop = (unsigned short)((env[4] & 0xffff0000ul) >> 16);
 276         fxsave->fcs = (env[4] & 0xffff);
 277         fxsave->foo = env[5];
 278         fxsave->fos = env[6];
 279
 280         to = (struct _fpxreg *) &fxsave->st_space[0];
 281         from = &buf->_st[0];
 282         for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
 283                 unsigned long *t = (unsigned long *)to;
 284                 unsigned long __user *f = (unsigned long __user *)from;
 285
 286                 if (__get_user(*t, f) ||
 287                                 __get_user(*(t + 1), f + 1) ||
 288                                 __get_user(to->exponent, &from->exponent))
 289                         return 1;
 290         }
 291         return 0;
 292 }
 293
 294 /*
 295  * Signal frame handlers.
 296  */
 297
 298 static inline int save_i387_fsave( struct _fpstate __user *buf )
 299 {
 300         struct task_struct *tsk = current;
 301
 302         unlazy_fpu( tsk );
 303         tsk->thread.i387.fsave.status = tsk->thread.i387.fsave.swd;
 304         if ( __copy_to_user( buf, &tsk->thread.i387.fsave,
 305                              sizeof(struct i387_fsave_struct) ) )
 306                 return -1;
 307         return 1;
 308 }
 309
 310 static int save_i387_fxsave( struct _fpstate __user *buf )
 311 {
 312         struct task_struct *tsk = current;
 313         int err = 0;
 314
 315         unlazy_fpu( tsk );
 316
 317         if ( convert_fxsr_to_user( buf, &tsk->thread.i387.fxsave ) )
 318                 return -1;
 319
 320         err |= __put_user( tsk->thread.i387.fxsave.swd, &buf->status );
 321         err |= __put_user( X86_FXSR_MAGIC, &buf->magic );
 322         if ( err )
 323                 return -1;
 324
 325         if ( __copy_to_user( &buf->_fxsr_env[0], &tsk->thread.i387.fxsave,
 326                              sizeof(struct i387_fxsave_struct) ) )
 327                 return -1;
 328         return 1;
 329 }
 330
 331 int save_i387( struct _fpstate __user *buf )
 332 {
 333         if ( !current->used_math )
 334                 return 0;
 335
 336         /* This will cause a "finit" to be triggered by the next
 337          * attempted FPU operation by the 'current' process.
 338          */
 339         current->used_math = 0;
 340
 341         if ( HAVE_HWFP ) {
 342                 if ( cpu_has_fxsr ) {
 343                         return save_i387_fxsave( buf );
 344                 } else {
 345                         return save_i387_fsave( buf );
 346                 }
 347         } else {
 348                 return save_i387_soft( &current->thread.i387.soft, buf );
 349         }
 350 }
 351
 352 static inline int restore_i387_fsave( struct _fpstate __user *buf )
 353 {
 354         struct task_struct *tsk = current;
 355         clear_fpu( tsk );
 356         return __copy_from_user( &tsk->thread.i387.fsave, buf,
 357                                  sizeof(struct i387_fsave_struct) );
 358 }
 359
 360 static int restore_i387_fxsave( struct _fpstate __user *buf )
 361 {
 362         int err;
 363         struct task_struct *tsk = current;
 364         clear_fpu( tsk );
 365         err = __copy_from_user( &tsk->thread.i387.fxsave, &buf->_fxsr_env[0],
 366                                 sizeof(struct i387_fxsave_struct) );
 367         /* mxcsr reserved bits must be masked to zero for security reasons */
 368         tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
 369         return err ? 1 : convert_fxsr_from_user( &tsk->thread.i387.fxsave, buf );
 370 }
 371
 372 int restore_i387( struct _fpstate __user *buf )
 373 {
 374         int err;
 375
 376         if ( HAVE_HWFP ) {
 377                 if ( cpu_has_fxsr ) {
 378                         err = restore_i387_fxsave( buf );
 379                 } else {
 380                         err = restore_i387_fsave( buf );
 381                 }
 382         } else {
 383                 err = restore_i387_soft( &current->thread.i387.soft, buf );
 384         }
 385         current->used_math = 1;
 386         return err;
 387 }
 388
 389 /*
 390  * ptrace request handlers.
 391  */
 392
 393 static inline int get_fpregs_fsave( struct user_i387_struct __user *buf,
 394                                     struct task_struct *tsk )
 395 {
 396         return __copy_to_user( buf, &tsk->thread.i387.fsave,
 397                                sizeof(struct user_i387_struct) );
 398 }
 399
 400 static inline int get_fpregs_fxsave( struct user_i387_struct __user *buf,
 401                                      struct task_struct *tsk )
 402 {
 403         return convert_fxsr_to_user( (struct _fpstate __user *)buf,
 404                                      &tsk->thread.i387.fxsave );
 405 }
 406
 407 int get_fpregs( struct user_i387_struct __user *buf, struct task_struct *tsk )
 408 {
 409         if ( HAVE_HWFP ) {
 410                 if ( cpu_has_fxsr ) {
 411                         return get_fpregs_fxsave( buf, tsk );
 412                 } else {
 413                         return get_fpregs_fsave( buf, tsk );
 414                 }
 415         } else {
 416                 return save_i387_soft( &tsk->thread.i387.soft,
 417                                        (struct _fpstate __user *)buf );
 418         }
 419 }
 420
 421 static inline int set_fpregs_fsave( struct task_struct *tsk,
 422                                     struct user_i387_struct __user *buf )
 423 {
 424         return __copy_from_user( &tsk->thread.i387.fsave, buf,
 425                                  sizeof(struct user_i387_struct) );
 426 }
 427
 428 static inline int set_fpregs_fxsave( struct task_struct *tsk,
 429                                      struct user_i387_struct __user *buf )
 430 {
 431         return convert_fxsr_from_user( &tsk->thread.i387.fxsave,
 432                                        (struct _fpstate __user *)buf );
 433 }
 434
 435 int set_fpregs( struct task_struct *tsk, struct user_i387_struct __user *buf )
 436 {
 437         if ( HAVE_HWFP ) {
 438                 if ( cpu_has_fxsr ) {
 439                         return set_fpregs_fxsave( tsk, buf );
 440                 } else {
 441                         return set_fpregs_fsave( tsk, buf );
 442                 }
 443         } else {
 444                 return restore_i387_soft( &tsk->thread.i387.soft,
 445                                           (struct _fpstate __user *)buf );
 446         }
 447 }
 448
 449 int get_fpxregs( struct user_fxsr_struct __user *buf, struct task_struct *tsk )
 450 {
 451         if ( cpu_has_fxsr ) {
 452                 if (__copy_to_user( buf, &tsk->thread.i387.fxsave,
 453                                     sizeof(struct user_fxsr_struct) ))
 454                         return -EFAULT;
 455                 return 0;
 456         } else {
 457                 return -EIO;
 458         }
 459 }
 460
 461 int set_fpxregs( struct task_struct *tsk, struct user_fxsr_struct __user *buf )
 462 {
 463         int ret = 0;
 464
 465         if ( cpu_has_fxsr ) {
 466                 if (__copy_from_user( &tsk->thread.i387.fxsave, buf,
 467                                   sizeof(struct user_fxsr_struct) ))
 468                         ret = -EFAULT;
 469                 /* mxcsr reserved bits must be masked to zero for security reasons */
 470                 tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
 471         } else {
 472                 ret = -EIO;
 473         }
 474         return ret;
 475 }
 476
 477 /*
 478  * FPU state for core dumps.
 479  */
 480
 481 static inline void copy_fpu_fsave( struct task_struct *tsk,
 482                                    struct user_i387_struct *fpu )
 483 {
 484         memcpy( fpu, &tsk->thread.i387.fsave,
 485                 sizeof(struct user_i387_struct) );
 486 }
 487
 488 static inline void copy_fpu_fxsave( struct task_struct *tsk,
 489                                    struct user_i387_struct *fpu )
 490 {
 491         unsigned short *to;
 492         unsigned short *from;
 493         int i;
 494
 495         memcpy( fpu, &tsk->thread.i387.fxsave, 7 * sizeof(long) );
 496
 497         to = (unsigned short *)&fpu->st_space[0];
 498         from = (unsigned short *)&tsk->thread.i387.fxsave.st_space[0];
 499         for ( i = 0 ; i < 8 ; i++, to += 5, from += 8 ) {
 500                 memcpy( to, from, 5 * sizeof(unsigned short) );
 501         }
 502 }
 503
 504 int dump_fpu( struct pt_regs *regs, struct user_i387_struct *fpu )
 505 {
 506         int fpvalid;
 507         struct task_struct *tsk = current;
 508
 509         fpvalid = tsk->used_math;
 510         if ( fpvalid ) {
 511                 unlazy_fpu( tsk );
 512                 if ( cpu_has_fxsr ) {
 513                         copy_fpu_fxsave( tsk, fpu );
 514                 } else {
 515                         copy_fpu_fsave( tsk, fpu );
 516                 }
 517         }
 518
 519         return fpvalid;
 520 }
 521
 522 int dump_extended_fpu( struct pt_regs *regs, struct user_fxsr_struct *fpu )
 523 {
 524         int fpvalid;
 525         struct task_struct *tsk = current;
 526
 527         fpvalid = tsk->used_math && cpu_has_fxsr;
 528         if ( fpvalid ) {
 529                 unlazy_fpu( tsk );
 530                 memcpy( fpu, &tsk->thread.i387.fxsave,
 531                         sizeof(struct user_fxsr_struct) );
 532         }
 533
 534         return fpvalid;
 535 }
 536
 537 int dump_task_fpu(struct task_struct *tsk, struct user_i387_struct *fpu)
 538 {
 539         int fpvalid = tsk->used_math;
 540
 541         if (fpvalid) {
 542                 if (tsk == current)
 543                         unlazy_fpu(tsk);
 544                 if (cpu_has_fxsr)
 545                         copy_fpu_fxsave(tsk, fpu);
 546                 else
 547                         copy_fpu_fsave(tsk, fpu);
 548         }
 549         return fpvalid;
 550 }
 551
 552 int dump_task_extended_fpu(struct task_struct *tsk, struct user_fxsr_struct *fpu)
 553 {
 554         int fpvalid = tsk->used_math && cpu_has_fxsr;
 555
 556         if (fpvalid) {
 557                 if (tsk == current)
 558                        unlazy_fpu(tsk);
 559                 memcpy(fpu, &tsk->thread.i387.fxsave, sizeof(*fpu));
 560         }
 561         return fpvalid;
 562 }