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         struct _fpreg tmp[8]; /* 80 bytes scratch area */
 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 = tmp;
 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                 *t = *f;
 254                 *(t + 1) = *(f+1);
 255                 to->exponent = from->exponent;
 256         }
 257         if (copy_to_user(buf->_st, tmp, sizeof(struct _fpreg [8])))
 258                 return 1;
 259         return 0;
 260 }
 261
 262 static int convert_fxsr_from_user( struct i387_fxsave_struct *fxsave,
 263                                           struct _fpstate __user *buf )
 264 {
 265         struct _fpreg tmp[8]; /* 80 bytes scratch area */
 266         unsigned long env[7];
 267         struct _fpxreg *to;
 268         struct _fpreg __user *from;
 269         int i;
 270
 271         if ( __copy_from_user( env, buf, 7 * sizeof(long) ) )
 272                 return 1;
 273         if (copy_from_user(tmp, buf->_st, sizeof(struct _fpreg [8])))
 274                 return 1;
 275
 276         fxsave->cwd = (unsigned short)(env[0] & 0xffff);
 277         fxsave->swd = (unsigned short)(env[1] & 0xffff);
 278         fxsave->twd = twd_i387_to_fxsr((unsigned short)(env[2] & 0xffff));
 279         fxsave->fip = env[3];
 280         fxsave->fop = (unsigned short)((env[4] & 0xffff0000ul) >> 16);
 281         fxsave->fcs = (env[4] & 0xffff);
 282         fxsave->foo = env[5];
 283         fxsave->fos = env[6];
 284
 285         to = (struct _fpxreg *) &fxsave->st_space[0];
 286         from = tmp;
 287         for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
 288                 unsigned long *t = (unsigned long *)to;
 289                 unsigned long __user *f = (unsigned long __user *)from;
 290
 291                 *t = *f;
 292                 *(t + 1) = *(f + 1);
 293                 to->exponent = from->exponent;
 294         }
 295         return 0;
 296 }
 297
 298 /*
 299  * Signal frame handlers.
 300  */
 301
 302 static inline int save_i387_fsave( struct _fpstate __user *buf )
 303 {
 304         struct task_struct *tsk = current;
 305
 306         unlazy_fpu( tsk );
 307         tsk->thread.i387.fsave.status = tsk->thread.i387.fsave.swd;
 308         if ( __copy_to_user( buf, &tsk->thread.i387.fsave,
 309                              sizeof(struct i387_fsave_struct) ) )
 310                 return -1;
 311         return 1;
 312 }
 313
 314 static int save_i387_fxsave( struct _fpstate __user *buf )
 315 {
 316         struct task_struct *tsk = current;
 317         int err = 0;
 318
 319         unlazy_fpu( tsk );
 320
 321         if ( convert_fxsr_to_user( buf, &tsk->thread.i387.fxsave ) )
 322                 return -1;
 323
 324         err |= __put_user( tsk->thread.i387.fxsave.swd, &buf->status );
 325         err |= __put_user( X86_FXSR_MAGIC, &buf->magic );
 326         if ( err )
 327                 return -1;
 328
 329         if ( __copy_to_user( &buf->_fxsr_env[0], &tsk->thread.i387.fxsave,
 330                              sizeof(struct i387_fxsave_struct) ) )
 331                 return -1;
 332         return 1;
 333 }
 334
 335 int save_i387( struct _fpstate __user *buf )
 336 {
 337         if ( !current->used_math )
 338                 return 0;
 339
 340         /* This will cause a "finit" to be triggered by the next
 341          * attempted FPU operation by the 'current' process.
 342          */
 343         current->used_math = 0;
 344
 345         if ( HAVE_HWFP ) {
 346                 if ( cpu_has_fxsr ) {
 347                         return save_i387_fxsave( buf );
 348                 } else {
 349                         return save_i387_fsave( buf );
 350                 }
 351         } else {
 352                 return save_i387_soft( &current->thread.i387.soft, buf );
 353         }
 354 }
 355
 356 static inline int restore_i387_fsave( struct _fpstate __user *buf )
 357 {
 358         struct task_struct *tsk = current;
 359         clear_fpu( tsk );
 360         return __copy_from_user( &tsk->thread.i387.fsave, buf,
 361                                  sizeof(struct i387_fsave_struct) );
 362 }
 363
 364 static int restore_i387_fxsave( struct _fpstate __user *buf )
 365 {
 366         int err;
 367         struct task_struct *tsk = current;
 368         clear_fpu( tsk );
 369         err = __copy_from_user( &tsk->thread.i387.fxsave, &buf->_fxsr_env[0],
 370                                 sizeof(struct i387_fxsave_struct) );
 371         /* mxcsr reserved bits must be masked to zero for security reasons */
 372         tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
 373         return err ? 1 : convert_fxsr_from_user( &tsk->thread.i387.fxsave, buf );
 374 }
 375
 376 int restore_i387( struct _fpstate __user *buf )
 377 {
 378         int err;
 379
 380         if ( HAVE_HWFP ) {
 381                 if ( cpu_has_fxsr ) {
 382                         err = restore_i387_fxsave( buf );
 383                 } else {
 384                         err = restore_i387_fsave( buf );
 385                 }
 386         } else {
 387                 err = restore_i387_soft( &current->thread.i387.soft, buf );
 388         }
 389         current->used_math = 1;
 390         return err;
 391 }
 392
 393 /*
 394  * ptrace request handlers.
 395  */
 396
 397 static inline int get_fpregs_fsave( struct user_i387_struct __user *buf,
 398                                     struct task_struct *tsk )
 399 {
 400         return __copy_to_user( buf, &tsk->thread.i387.fsave,
 401                                sizeof(struct user_i387_struct) );
 402 }
 403
 404 static inline int get_fpregs_fxsave( struct user_i387_struct __user *buf,
 405                                      struct task_struct *tsk )
 406 {
 407         return convert_fxsr_to_user( (struct _fpstate __user *)buf,
 408                                      &tsk->thread.i387.fxsave );
 409 }
 410
 411 int get_fpregs( struct user_i387_struct __user *buf, struct task_struct *tsk )
 412 {
 413         if ( HAVE_HWFP ) {
 414                 if ( cpu_has_fxsr ) {
 415                         return get_fpregs_fxsave( buf, tsk );
 416                 } else {
 417                         return get_fpregs_fsave( buf, tsk );
 418                 }
 419         } else {
 420                 return save_i387_soft( &tsk->thread.i387.soft,
 421                                        (struct _fpstate __user *)buf );
 422         }
 423 }
 424
 425 static inline int set_fpregs_fsave( struct task_struct *tsk,
 426                                     struct user_i387_struct __user *buf )
 427 {
 428         return __copy_from_user( &tsk->thread.i387.fsave, buf,
 429                                  sizeof(struct user_i387_struct) );
 430 }
 431
 432 static inline int set_fpregs_fxsave( struct task_struct *tsk,
 433                                      struct user_i387_struct __user *buf )
 434 {
 435         return convert_fxsr_from_user( &tsk->thread.i387.fxsave,
 436                                        (struct _fpstate __user *)buf );
 437 }
 438
 439 int set_fpregs( struct task_struct *tsk, struct user_i387_struct __user *buf )
 440 {
 441         if ( HAVE_HWFP ) {
 442                 if ( cpu_has_fxsr ) {
 443                         return set_fpregs_fxsave( tsk, buf );
 444                 } else {
 445                         return set_fpregs_fsave( tsk, buf );
 446                 }
 447         } else {
 448                 return restore_i387_soft( &tsk->thread.i387.soft,
 449                                           (struct _fpstate __user *)buf );
 450         }
 451 }
 452
 453 int get_fpxregs( struct user_fxsr_struct __user *buf, struct task_struct *tsk )
 454 {
 455         if ( cpu_has_fxsr ) {
 456                 if (__copy_to_user( buf, &tsk->thread.i387.fxsave,
 457                                     sizeof(struct user_fxsr_struct) ))
 458                         return -EFAULT;
 459                 return 0;
 460         } else {
 461                 return -EIO;
 462         }
 463 }
 464
 465 int set_fpxregs( struct task_struct *tsk, struct user_fxsr_struct __user *buf )
 466 {
 467         int ret = 0;
 468
 469         if ( cpu_has_fxsr ) {
 470                 if (__copy_from_user( &tsk->thread.i387.fxsave, buf,
 471                                   sizeof(struct user_fxsr_struct) ))
 472                         ret = -EFAULT;
 473                 /* mxcsr reserved bits must be masked to zero for security reasons */
 474                 tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
 475         } else {
 476                 ret = -EIO;
 477         }
 478         return ret;
 479 }
 480
 481 /*
 482  * FPU state for core dumps.
 483  */
 484
 485 static inline void copy_fpu_fsave( struct task_struct *tsk,
 486                                    struct user_i387_struct *fpu )
 487 {
 488         memcpy( fpu, &tsk->thread.i387.fsave,
 489                 sizeof(struct user_i387_struct) );
 490 }
 491
 492 static inline void copy_fpu_fxsave( struct task_struct *tsk,
 493                                    struct user_i387_struct *fpu )
 494 {
 495         unsigned short *to;
 496         unsigned short *from;
 497         int i;
 498
 499         memcpy( fpu, &tsk->thread.i387.fxsave, 7 * sizeof(long) );
 500
 501         to = (unsigned short *)&fpu->st_space[0];
 502         from = (unsigned short *)&tsk->thread.i387.fxsave.st_space[0];
 503         for ( i = 0 ; i < 8 ; i++, to += 5, from += 8 ) {
 504                 memcpy( to, from, 5 * sizeof(unsigned short) );
 505         }
 506 }
 507
 508 int dump_fpu( struct pt_regs *regs, struct user_i387_struct *fpu )
 509 {
 510         int fpvalid;
 511         struct task_struct *tsk = current;
 512
 513         fpvalid = tsk->used_math;
 514         if ( fpvalid ) {
 515                 unlazy_fpu( tsk );
 516                 if ( cpu_has_fxsr ) {
 517                         copy_fpu_fxsave( tsk, fpu );
 518                 } else {
 519                         copy_fpu_fsave( tsk, fpu );
 520                 }
 521         }
 522
 523         return fpvalid;
 524 }
 525
 526 int dump_extended_fpu( struct pt_regs *regs, struct user_fxsr_struct *fpu )
 527 {
 528         int fpvalid;
 529         struct task_struct *tsk = current;
 530
 531         fpvalid = tsk->used_math && cpu_has_fxsr;
 532         if ( fpvalid ) {
 533                 unlazy_fpu( tsk );
 534                 memcpy( fpu, &tsk->thread.i387.fxsave,
 535                         sizeof(struct user_fxsr_struct) );
 536         }
 537
 538         return fpvalid;
 539 }
 540
 541 int dump_task_fpu(struct task_struct *tsk, struct user_i387_struct *fpu)
 542 {
 543         int fpvalid = tsk->used_math;
 544
 545         if (fpvalid) {
 546                 if (tsk == current)
 547                         unlazy_fpu(tsk);
 548                 if (cpu_has_fxsr)
 549                         copy_fpu_fxsave(tsk, fpu);
 550                 else
 551                         copy_fpu_fsave(tsk, fpu);
 552         }
 553         return fpvalid;
 554 }
 555
 556 int dump_task_extended_fpu(struct task_struct *tsk, struct user_fxsr_struct *fpu)
 557 {
 558         int fpvalid = tsk->used_math && cpu_has_fxsr;
 559
 560         if (fpvalid) {
 561                 if (tsk == current)
 562                        unlazy_fpu(tsk);
 563                 memcpy(fpu, &tsk->thread.i387.fxsave, sizeof(*fpu));
 564         }
 565         return fpvalid;
 566 }