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 tos = (fxsave->swd >> 11) & 7;
 115         unsigned long twd = (unsigned long) fxsave->twd;
 116         unsigned long tag;
 117         unsigned long ret = 0xffff0000u;
 118         int i;
 119
 120 #define FPREG_ADDR(f, n)        ((void *)&(f)->st_space + (n) * 16);
 121
 122         for ( i = 0 ; i < 8 ; i++ ) {
 123                 if ( twd & 0x1 ) {
 124                         st = FPREG_ADDR( fxsave, (i - tos) & 7 );
 125
 126                         switch ( st->exponent & 0x7fff ) {
 127                         case 0x7fff:
 128                                 tag = 2;                /* Special */
 129                                 break;
 130                         case 0x0000:
 131                                 if ( !st->significand[0] &&
 132                                      !st->significand[1] &&
 133                                      !st->significand[2] &&
 134                                      !st->significand[3] ) {
 135                                         tag = 1;        /* Zero */
 136                                 } else {
 137                                         tag = 2;        /* Special */
 138                                 }
 139                                 break;
 140                         default:
 141                                 if ( st->significand[3] & 0x8000 ) {
 142                                         tag = 0;        /* Valid */
 143                                 } else {
 144                                         tag = 2;        /* Special */
 145                                 }
 146                                 break;
 147                         }
 148                 } else {
 149                         tag = 3;                        /* Empty */
 150                 }
 151                 ret |= (tag << (2 * i));
 152                 twd = twd >> 1;
 153         }
 154         return ret;
 155 }
 156
 157 /*
 158  * FPU state interaction.
 159  */
 160
 161 unsigned short get_fpu_cwd( struct task_struct *tsk )
 162 {
 163         if ( cpu_has_fxsr ) {
 164                 return tsk->thread.i387.fxsave.cwd;
 165         } else {
 166                 return (unsigned short)tsk->thread.i387.fsave.cwd;
 167         }
 168 }
 169
 170 unsigned short get_fpu_swd( struct task_struct *tsk )
 171 {
 172         if ( cpu_has_fxsr ) {
 173                 return tsk->thread.i387.fxsave.swd;
 174         } else {
 175                 return (unsigned short)tsk->thread.i387.fsave.swd;
 176         }
 177 }
 178
 179 unsigned short get_fpu_twd( struct task_struct *tsk )
 180 {
 181         if ( cpu_has_fxsr ) {
 182                 return tsk->thread.i387.fxsave.twd;
 183         } else {
 184                 return (unsigned short)tsk->thread.i387.fsave.twd;
 185         }
 186 }
 187
 188 unsigned short get_fpu_mxcsr( struct task_struct *tsk )
 189 {
 190         if ( cpu_has_xmm ) {
 191                 return tsk->thread.i387.fxsave.mxcsr;
 192         } else {
 193                 return 0x1f80;
 194         }
 195 }
 196
 197 void set_fpu_cwd( struct task_struct *tsk, unsigned short cwd )
 198 {
 199         if ( cpu_has_fxsr ) {
 200                 tsk->thread.i387.fxsave.cwd = cwd;
 201         } else {
 202                 tsk->thread.i387.fsave.cwd = ((long)cwd | 0xffff0000u);
 203         }
 204 }
 205
 206 void set_fpu_swd( struct task_struct *tsk, unsigned short swd )
 207 {
 208         if ( cpu_has_fxsr ) {
 209                 tsk->thread.i387.fxsave.swd = swd;
 210         } else {
 211                 tsk->thread.i387.fsave.swd = ((long)swd | 0xffff0000u);
 212         }
 213 }
 214
 215 void set_fpu_twd( struct task_struct *tsk, unsigned short twd )
 216 {
 217         if ( cpu_has_fxsr ) {
 218                 tsk->thread.i387.fxsave.twd = twd_i387_to_fxsr(twd);
 219         } else {
 220                 tsk->thread.i387.fsave.twd = ((long)twd | 0xffff0000u);
 221         }
 222 }
 223
 224 /*
 225  * FXSR floating point environment conversions.
 226  */
 227
 228 static int convert_fxsr_to_user( struct _fpstate __user *buf,
 229                                         struct i387_fxsave_struct *fxsave )
 230 {
 231         unsigned long env[7];
 232         struct _fpreg __user *to;
 233         struct _fpxreg *from;
 234         int i;
 235
 236         env[0] = (unsigned long)fxsave->cwd | 0xffff0000ul;
 237         env[1] = (unsigned long)fxsave->swd | 0xffff0000ul;
 238         env[2] = twd_fxsr_to_i387(fxsave);
 239         env[3] = fxsave->fip;
 240         env[4] = fxsave->fcs | ((unsigned long)fxsave->fop << 16);
 241         env[5] = fxsave->foo;
 242         env[6] = fxsave->fos;
 243
 244         if ( __copy_to_user( buf, env, 7 * sizeof(unsigned long) ) )
 245                 return 1;
 246
 247         to = &buf->_st[0];
 248         from = (struct _fpxreg *) &fxsave->st_space[0];
 249         for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
 250                 unsigned long __user *t = (unsigned long __user *)to;
 251                 unsigned long *f = (unsigned long *)from;
 252
 253                 if (__put_user(*f, t) ||
 254                                 __put_user(*(f + 1), t + 1) ||
 255                                 __put_user(from->exponent, &to->exponent))
 256                         return 1;
 257         }
 258         return 0;
 259 }
 260
 261 static int convert_fxsr_from_user( struct i387_fxsave_struct *fxsave,
 262                                           struct _fpstate __user *buf )
 263 {
 264         unsigned long env[7];
 265         struct _fpxreg *to;
 266         struct _fpreg __user *from;
 267         int i;
 268
 269         if ( __copy_from_user( env, buf, 7 * sizeof(long) ) )
 270                 return 1;
 271
 272         fxsave->cwd = (unsigned short)(env[0] & 0xffff);
 273         fxsave->swd = (unsigned short)(env[1] & 0xffff);
 274         fxsave->twd = twd_i387_to_fxsr((unsigned short)(env[2] & 0xffff));
 275         fxsave->fip = env[3];
 276         fxsave->fop = (unsigned short)((env[4] & 0xffff0000ul) >> 16);
 277         fxsave->fcs = (env[4] & 0xffff);
 278         fxsave->foo = env[5];
 279         fxsave->fos = env[6];
 280
 281         to = (struct _fpxreg *) &fxsave->st_space[0];
 282         from = &buf->_st[0];
 283         for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
 284                 unsigned long *t = (unsigned long *)to;
 285                 unsigned long __user *f = (unsigned long __user *)from;
 286
 287                 if (__get_user(*t, f) ||
 288                                 __get_user(*(t + 1), f + 1) ||
 289                                 __get_user(to->exponent, &from->exponent))
 290                         return 1;
 291         }
 292         return 0;
 293 }
 294
 295 /*
 296  * Signal frame handlers.
 297  */
 298
 299 static inline int save_i387_fsave( struct _fpstate __user *buf )
 300 {
 301         struct task_struct *tsk = current;
 302
 303         unlazy_fpu( tsk );
 304         tsk->thread.i387.fsave.status = tsk->thread.i387.fsave.swd;
 305         if ( __copy_to_user( buf, &tsk->thread.i387.fsave,
 306                              sizeof(struct i387_fsave_struct) ) )
 307                 return -1;
 308         return 1;
 309 }
 310
 311 static int save_i387_fxsave( struct _fpstate __user *buf )
 312 {
 313         struct task_struct *tsk = current;
 314         int err = 0;
 315
 316         unlazy_fpu( tsk );
 317
 318         if ( convert_fxsr_to_user( buf, &tsk->thread.i387.fxsave ) )
 319                 return -1;
 320
 321         err |= __put_user( tsk->thread.i387.fxsave.swd, &buf->status );
 322         err |= __put_user( X86_FXSR_MAGIC, &buf->magic );
 323         if ( err )
 324                 return -1;
 325
 326         if ( __copy_to_user( &buf->_fxsr_env[0], &tsk->thread.i387.fxsave,
 327                              sizeof(struct i387_fxsave_struct) ) )
 328                 return -1;
 329         return 1;
 330 }
 331
 332 int save_i387( struct _fpstate __user *buf )
 333 {
 334         if ( !current->used_math )
 335                 return 0;
 336
 337         /* This will cause a "finit" to be triggered by the next
 338          * attempted FPU operation by the 'current' process.
 339          */
 340         current->used_math = 0;
 341
 342         if ( HAVE_HWFP ) {
 343                 if ( cpu_has_fxsr ) {
 344                         return save_i387_fxsave( buf );
 345                 } else {
 346                         return save_i387_fsave( buf );
 347                 }
 348         } else {
 349                 return save_i387_soft( &current->thread.i387.soft, buf );
 350         }
 351 }
 352
 353 static inline int restore_i387_fsave( struct _fpstate __user *buf )
 354 {
 355         struct task_struct *tsk = current;
 356         clear_fpu( tsk );
 357         return __copy_from_user( &tsk->thread.i387.fsave, buf,
 358                                  sizeof(struct i387_fsave_struct) );
 359 }
 360
 361 static int restore_i387_fxsave( struct _fpstate __user *buf )
 362 {
 363         int err;
 364         struct task_struct *tsk = current;
 365         clear_fpu( tsk );
 366         err = __copy_from_user( &tsk->thread.i387.fxsave, &buf->_fxsr_env[0],
 367                                 sizeof(struct i387_fxsave_struct) );
 368         /* mxcsr reserved bits must be masked to zero for security reasons */
 369         tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
 370         return err ? 1 : convert_fxsr_from_user( &tsk->thread.i387.fxsave, buf );
 371 }
 372
 373 int restore_i387( struct _fpstate __user *buf )
 374 {
 375         int err;
 376
 377         if ( HAVE_HWFP ) {
 378                 if ( cpu_has_fxsr ) {
 379                         err = restore_i387_fxsave( buf );
 380                 } else {
 381                         err = restore_i387_fsave( buf );
 382                 }
 383         } else {
 384                 err = restore_i387_soft( &current->thread.i387.soft, buf );
 385         }
 386         current->used_math = 1;
 387         return err;
 388 }
 389
 390 /*
 391  * ptrace request handlers.
 392  */
 393
 394 static inline int get_fpregs_fsave( struct user_i387_struct __user *buf,
 395                                     struct task_struct *tsk )
 396 {
 397         return __copy_to_user( buf, &tsk->thread.i387.fsave,
 398                                sizeof(struct user_i387_struct) );
 399 }
 400
 401 static inline int get_fpregs_fxsave( struct user_i387_struct __user *buf,
 402                                      struct task_struct *tsk )
 403 {
 404         return convert_fxsr_to_user( (struct _fpstate __user *)buf,
 405                                      &tsk->thread.i387.fxsave );
 406 }
 407
 408 int get_fpregs( struct user_i387_struct __user *buf, struct task_struct *tsk )
 409 {
 410         if ( HAVE_HWFP ) {
 411                 if ( cpu_has_fxsr ) {
 412                         return get_fpregs_fxsave( buf, tsk );
 413                 } else {
 414                         return get_fpregs_fsave( buf, tsk );
 415                 }
 416         } else {
 417                 return save_i387_soft( &tsk->thread.i387.soft,
 418                                        (struct _fpstate __user *)buf );
 419         }
 420 }
 421
 422 static inline int set_fpregs_fsave( struct task_struct *tsk,
 423                                     struct user_i387_struct __user *buf )
 424 {
 425         return __copy_from_user( &tsk->thread.i387.fsave, buf,
 426                                  sizeof(struct user_i387_struct) );
 427 }
 428
 429 static inline int set_fpregs_fxsave( struct task_struct *tsk,
 430                                      struct user_i387_struct __user *buf )
 431 {
 432         return convert_fxsr_from_user( &tsk->thread.i387.fxsave,
 433                                        (struct _fpstate __user *)buf );
 434 }
 435
 436 int set_fpregs( struct task_struct *tsk, struct user_i387_struct __user *buf )
 437 {
 438         if ( HAVE_HWFP ) {
 439                 if ( cpu_has_fxsr ) {
 440                         return set_fpregs_fxsave( tsk, buf );
 441                 } else {
 442                         return set_fpregs_fsave( tsk, buf );
 443                 }
 444         } else {
 445                 return restore_i387_soft( &tsk->thread.i387.soft,
 446                                           (struct _fpstate __user *)buf );
 447         }
 448 }
 449
 450 int get_fpxregs( struct user_fxsr_struct __user *buf, struct task_struct *tsk )
 451 {
 452         if ( cpu_has_fxsr ) {
 453                 if (__copy_to_user( buf, &tsk->thread.i387.fxsave,
 454                                     sizeof(struct user_fxsr_struct) ))
 455                         return -EFAULT;
 456                 return 0;
 457         } else {
 458                 return -EIO;
 459         }
 460 }
 461
 462 int set_fpxregs( struct task_struct *tsk, struct user_fxsr_struct __user *buf )
 463 {
 464         int ret = 0;
 465
 466         if ( cpu_has_fxsr ) {
 467                 if (__copy_from_user( &tsk->thread.i387.fxsave, buf,
 468                                   sizeof(struct user_fxsr_struct) ))
 469                         ret = -EFAULT;
 470                 /* mxcsr reserved bits must be masked to zero for security reasons */
 471                 tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
 472         } else {
 473                 ret = -EIO;
 474         }
 475         return ret;
 476 }
 477
 478 /*
 479  * FPU state for core dumps.
 480  */
 481
 482 static inline void copy_fpu_fsave( struct task_struct *tsk,
 483                                    struct user_i387_struct *fpu )
 484 {
 485         memcpy( fpu, &tsk->thread.i387.fsave,
 486                 sizeof(struct user_i387_struct) );
 487 }
 488
 489 static inline void copy_fpu_fxsave( struct task_struct *tsk,
 490                                    struct user_i387_struct *fpu )
 491 {
 492         unsigned short *to;
 493         unsigned short *from;
 494         int i;
 495
 496         memcpy( fpu, &tsk->thread.i387.fxsave, 7 * sizeof(long) );
 497
 498         to = (unsigned short *)&fpu->st_space[0];
 499         from = (unsigned short *)&tsk->thread.i387.fxsave.st_space[0];
 500         for ( i = 0 ; i < 8 ; i++, to += 5, from += 8 ) {
 501                 memcpy( to, from, 5 * sizeof(unsigned short) );
 502         }
 503 }
 504
 505 int dump_fpu( struct pt_regs *regs, struct user_i387_struct *fpu )
 506 {
 507         int fpvalid;
 508         struct task_struct *tsk = current;
 509
 510         fpvalid = tsk->used_math;
 511         if ( fpvalid ) {
 512                 unlazy_fpu( tsk );
 513                 if ( cpu_has_fxsr ) {
 514                         copy_fpu_fxsave( tsk, fpu );
 515                 } else {
 516                         copy_fpu_fsave( tsk, fpu );
 517                 }
 518         }
 519
 520         return fpvalid;
 521 }
 522
 523 int dump_task_fpu(struct task_struct *tsk, struct user_i387_struct *fpu)
 524 {
 525         int fpvalid = tsk->used_math;
 526
 527         if (fpvalid) {
 528                 if (tsk == current)
 529                         unlazy_fpu(tsk);
 530                 if (cpu_has_fxsr)
 531                         copy_fpu_fxsave(tsk, fpu);
 532                 else
 533                         copy_fpu_fsave(tsk, fpu);
 534         }
 535         return fpvalid;
 536 }
 537
 538 int dump_task_extended_fpu(struct task_struct *tsk, struct user_fxsr_struct *fpu)
 539 {
 540         int fpvalid = tsk->used_math && cpu_has_fxsr;
 541
 542         if (fpvalid) {
 543                 if (tsk == current)
 544                        unlazy_fpu(tsk);
 545                 memcpy(fpu, &tsk->thread.i387.fxsave, sizeof(*fpu));
 546         }
 547         return fpvalid;
 548 }