Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
[linux-2.6] / 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/sched.h>
12 #include <linux/module.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 static unsigned long mxcsr_feature_mask __read_mostly = 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         /* only the device not available exception or ptrace can call init_fpu */
64         set_stopped_child_used_math(tsk);
65 }
66
67 /*
68  * FPU lazy state save handling.
69  */
70
71 void kernel_fpu_begin(void)
72 {
73         struct thread_info *thread = current_thread_info();
74
75         preempt_disable();
76         if (thread->status & TS_USEDFPU) {
77                 __save_init_fpu(thread->task);
78                 return;
79         }
80         clts();
81 }
82 EXPORT_SYMBOL_GPL(kernel_fpu_begin);
83
84 /*
85  * FPU tag word conversions.
86  */
87
88 static inline unsigned short twd_i387_to_fxsr( unsigned short twd )
89 {
90         unsigned int tmp; /* to avoid 16 bit prefixes in the code */
91  
92         /* Transform each pair of bits into 01 (valid) or 00 (empty) */
93         tmp = ~twd;
94         tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */
95         /* and move the valid bits to the lower byte. */
96         tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */
97         tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */
98         tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */
99         return tmp;
100 }
101
102 static inline unsigned long twd_fxsr_to_i387( struct i387_fxsave_struct *fxsave )
103 {
104         struct _fpxreg *st = NULL;
105         unsigned long tos = (fxsave->swd >> 11) & 7;
106         unsigned long twd = (unsigned long) fxsave->twd;
107         unsigned long tag;
108         unsigned long ret = 0xffff0000u;
109         int i;
110
111 #define FPREG_ADDR(f, n)        ((void *)&(f)->st_space + (n) * 16);
112
113         for ( i = 0 ; i < 8 ; i++ ) {
114                 if ( twd & 0x1 ) {
115                         st = FPREG_ADDR( fxsave, (i - tos) & 7 );
116
117                         switch ( st->exponent & 0x7fff ) {
118                         case 0x7fff:
119                                 tag = 2;                /* Special */
120                                 break;
121                         case 0x0000:
122                                 if ( !st->significand[0] &&
123                                      !st->significand[1] &&
124                                      !st->significand[2] &&
125                                      !st->significand[3] ) {
126                                         tag = 1;        /* Zero */
127                                 } else {
128                                         tag = 2;        /* Special */
129                                 }
130                                 break;
131                         default:
132                                 if ( st->significand[3] & 0x8000 ) {
133                                         tag = 0;        /* Valid */
134                                 } else {
135                                         tag = 2;        /* Special */
136                                 }
137                                 break;
138                         }
139                 } else {
140                         tag = 3;                        /* Empty */
141                 }
142                 ret |= (tag << (2 * i));
143                 twd = twd >> 1;
144         }
145         return ret;
146 }
147
148 /*
149  * FPU state interaction.
150  */
151
152 unsigned short get_fpu_cwd( struct task_struct *tsk )
153 {
154         if ( cpu_has_fxsr ) {
155                 return tsk->thread.i387.fxsave.cwd;
156         } else {
157                 return (unsigned short)tsk->thread.i387.fsave.cwd;
158         }
159 }
160
161 unsigned short get_fpu_swd( struct task_struct *tsk )
162 {
163         if ( cpu_has_fxsr ) {
164                 return tsk->thread.i387.fxsave.swd;
165         } else {
166                 return (unsigned short)tsk->thread.i387.fsave.swd;
167         }
168 }
169
170 #if 0
171 unsigned short get_fpu_twd( struct task_struct *tsk )
172 {
173         if ( cpu_has_fxsr ) {
174                 return tsk->thread.i387.fxsave.twd;
175         } else {
176                 return (unsigned short)tsk->thread.i387.fsave.twd;
177         }
178 }
179 #endif  /*  0  */
180
181 unsigned short get_fpu_mxcsr( struct task_struct *tsk )
182 {
183         if ( cpu_has_xmm ) {
184                 return tsk->thread.i387.fxsave.mxcsr;
185         } else {
186                 return 0x1f80;
187         }
188 }
189
190 #if 0
191
192 void set_fpu_cwd( struct task_struct *tsk, unsigned short cwd )
193 {
194         if ( cpu_has_fxsr ) {
195                 tsk->thread.i387.fxsave.cwd = cwd;
196         } else {
197                 tsk->thread.i387.fsave.cwd = ((long)cwd | 0xffff0000u);
198         }
199 }
200
201 void set_fpu_swd( struct task_struct *tsk, unsigned short swd )
202 {
203         if ( cpu_has_fxsr ) {
204                 tsk->thread.i387.fxsave.swd = swd;
205         } else {
206                 tsk->thread.i387.fsave.swd = ((long)swd | 0xffff0000u);
207         }
208 }
209
210 void set_fpu_twd( struct task_struct *tsk, unsigned short twd )
211 {
212         if ( cpu_has_fxsr ) {
213                 tsk->thread.i387.fxsave.twd = twd_i387_to_fxsr(twd);
214         } else {
215                 tsk->thread.i387.fsave.twd = ((long)twd | 0xffff0000u);
216         }
217 }
218
219 #endif  /*  0  */
220
221 /*
222  * FXSR floating point environment conversions.
223  */
224
225 static int convert_fxsr_to_user( struct _fpstate __user *buf,
226                                         struct i387_fxsave_struct *fxsave )
227 {
228         unsigned long env[7];
229         struct _fpreg __user *to;
230         struct _fpxreg *from;
231         int i;
232
233         env[0] = (unsigned long)fxsave->cwd | 0xffff0000ul;
234         env[1] = (unsigned long)fxsave->swd | 0xffff0000ul;
235         env[2] = twd_fxsr_to_i387(fxsave);
236         env[3] = fxsave->fip;
237         env[4] = fxsave->fcs | ((unsigned long)fxsave->fop << 16);
238         env[5] = fxsave->foo;
239         env[6] = fxsave->fos;
240
241         if ( __copy_to_user( buf, env, 7 * sizeof(unsigned long) ) )
242                 return 1;
243
244         to = &buf->_st[0];
245         from = (struct _fpxreg *) &fxsave->st_space[0];
246         for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
247                 unsigned long __user *t = (unsigned long __user *)to;
248                 unsigned long *f = (unsigned long *)from;
249
250                 if (__put_user(*f, t) ||
251                                 __put_user(*(f + 1), t + 1) ||
252                                 __put_user(from->exponent, &to->exponent))
253                         return 1;
254         }
255         return 0;
256 }
257
258 static int convert_fxsr_from_user( struct i387_fxsave_struct *fxsave,
259                                           struct _fpstate __user *buf )
260 {
261         unsigned long env[7];
262         struct _fpxreg *to;
263         struct _fpreg __user *from;
264         int i;
265
266         if ( __copy_from_user( env, buf, 7 * sizeof(long) ) )
267                 return 1;
268
269         fxsave->cwd = (unsigned short)(env[0] & 0xffff);
270         fxsave->swd = (unsigned short)(env[1] & 0xffff);
271         fxsave->twd = twd_i387_to_fxsr((unsigned short)(env[2] & 0xffff));
272         fxsave->fip = env[3];
273         fxsave->fop = (unsigned short)((env[4] & 0xffff0000ul) >> 16);
274         fxsave->fcs = (env[4] & 0xffff);
275         fxsave->foo = env[5];
276         fxsave->fos = env[6];
277
278         to = (struct _fpxreg *) &fxsave->st_space[0];
279         from = &buf->_st[0];
280         for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
281                 unsigned long *t = (unsigned long *)to;
282                 unsigned long __user *f = (unsigned long __user *)from;
283
284                 if (__get_user(*t, f) ||
285                                 __get_user(*(t + 1), f + 1) ||
286                                 __get_user(to->exponent, &from->exponent))
287                         return 1;
288         }
289         return 0;
290 }
291
292 /*
293  * Signal frame handlers.
294  */
295
296 static inline int save_i387_fsave( struct _fpstate __user *buf )
297 {
298         struct task_struct *tsk = current;
299
300         unlazy_fpu( tsk );
301         tsk->thread.i387.fsave.status = tsk->thread.i387.fsave.swd;
302         if ( __copy_to_user( buf, &tsk->thread.i387.fsave,
303                              sizeof(struct i387_fsave_struct) ) )
304                 return -1;
305         return 1;
306 }
307
308 static int save_i387_fxsave( struct _fpstate __user *buf )
309 {
310         struct task_struct *tsk = current;
311         int err = 0;
312
313         unlazy_fpu( tsk );
314
315         if ( convert_fxsr_to_user( buf, &tsk->thread.i387.fxsave ) )
316                 return -1;
317
318         err |= __put_user( tsk->thread.i387.fxsave.swd, &buf->status );
319         err |= __put_user( X86_FXSR_MAGIC, &buf->magic );
320         if ( err )
321                 return -1;
322
323         if ( __copy_to_user( &buf->_fxsr_env[0], &tsk->thread.i387.fxsave,
324                              sizeof(struct i387_fxsave_struct) ) )
325                 return -1;
326         return 1;
327 }
328
329 int save_i387( struct _fpstate __user *buf )
330 {
331         if ( !used_math() )
332                 return 0;
333
334         /* This will cause a "finit" to be triggered by the next
335          * attempted FPU operation by the 'current' process.
336          */
337         clear_used_math();
338
339         if ( HAVE_HWFP ) {
340                 if ( cpu_has_fxsr ) {
341                         return save_i387_fxsave( buf );
342                 } else {
343                         return save_i387_fsave( buf );
344                 }
345         } else {
346                 return save_i387_soft( &current->thread.i387.soft, buf );
347         }
348 }
349
350 static inline int restore_i387_fsave( struct _fpstate __user *buf )
351 {
352         struct task_struct *tsk = current;
353         clear_fpu( tsk );
354         return __copy_from_user( &tsk->thread.i387.fsave, buf,
355                                  sizeof(struct i387_fsave_struct) );
356 }
357
358 static int restore_i387_fxsave( struct _fpstate __user *buf )
359 {
360         int err;
361         struct task_struct *tsk = current;
362         clear_fpu( tsk );
363         err = __copy_from_user( &tsk->thread.i387.fxsave, &buf->_fxsr_env[0],
364                                 sizeof(struct i387_fxsave_struct) );
365         /* mxcsr reserved bits must be masked to zero for security reasons */
366         tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
367         return err ? 1 : convert_fxsr_from_user( &tsk->thread.i387.fxsave, buf );
368 }
369
370 int restore_i387( struct _fpstate __user *buf )
371 {
372         int err;
373
374         if ( HAVE_HWFP ) {
375                 if ( cpu_has_fxsr ) {
376                         err = restore_i387_fxsave( buf );
377                 } else {
378                         err = restore_i387_fsave( buf );
379                 }
380         } else {
381                 err = restore_i387_soft( &current->thread.i387.soft, buf );
382         }
383         set_used_math();
384         return err;
385 }
386
387 /*
388  * ptrace request handlers.
389  */
390
391 static inline int get_fpregs_fsave( struct user_i387_struct __user *buf,
392                                     struct task_struct *tsk )
393 {
394         return __copy_to_user( buf, &tsk->thread.i387.fsave,
395                                sizeof(struct user_i387_struct) );
396 }
397
398 static inline int get_fpregs_fxsave( struct user_i387_struct __user *buf,
399                                      struct task_struct *tsk )
400 {
401         return convert_fxsr_to_user( (struct _fpstate __user *)buf,
402                                      &tsk->thread.i387.fxsave );
403 }
404
405 int get_fpregs( struct user_i387_struct __user *buf, struct task_struct *tsk )
406 {
407         if ( HAVE_HWFP ) {
408                 if ( cpu_has_fxsr ) {
409                         return get_fpregs_fxsave( buf, tsk );
410                 } else {
411                         return get_fpregs_fsave( buf, tsk );
412                 }
413         } else {
414                 return save_i387_soft( &tsk->thread.i387.soft,
415                                        (struct _fpstate __user *)buf );
416         }
417 }
418
419 static inline int set_fpregs_fsave( struct task_struct *tsk,
420                                     struct user_i387_struct __user *buf )
421 {
422         return __copy_from_user( &tsk->thread.i387.fsave, buf,
423                                  sizeof(struct user_i387_struct) );
424 }
425
426 static inline int set_fpregs_fxsave( struct task_struct *tsk,
427                                      struct user_i387_struct __user *buf )
428 {
429         return convert_fxsr_from_user( &tsk->thread.i387.fxsave,
430                                        (struct _fpstate __user *)buf );
431 }
432
433 int set_fpregs( struct task_struct *tsk, struct user_i387_struct __user *buf )
434 {
435         if ( HAVE_HWFP ) {
436                 if ( cpu_has_fxsr ) {
437                         return set_fpregs_fxsave( tsk, buf );
438                 } else {
439                         return set_fpregs_fsave( tsk, buf );
440                 }
441         } else {
442                 return restore_i387_soft( &tsk->thread.i387.soft,
443                                           (struct _fpstate __user *)buf );
444         }
445 }
446
447 int get_fpxregs( struct user_fxsr_struct __user *buf, struct task_struct *tsk )
448 {
449         if ( cpu_has_fxsr ) {
450                 if (__copy_to_user( buf, &tsk->thread.i387.fxsave,
451                                     sizeof(struct user_fxsr_struct) ))
452                         return -EFAULT;
453                 return 0;
454         } else {
455                 return -EIO;
456         }
457 }
458
459 int set_fpxregs( struct task_struct *tsk, struct user_fxsr_struct __user *buf )
460 {
461         int ret = 0;
462
463         if ( cpu_has_fxsr ) {
464                 if (__copy_from_user( &tsk->thread.i387.fxsave, buf,
465                                   sizeof(struct user_fxsr_struct) ))
466                         ret = -EFAULT;
467                 /* mxcsr reserved bits must be masked to zero for security reasons */
468                 tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
469         } else {
470                 ret = -EIO;
471         }
472         return ret;
473 }
474
475 /*
476  * FPU state for core dumps.
477  */
478
479 static inline void copy_fpu_fsave( struct task_struct *tsk,
480                                    struct user_i387_struct *fpu )
481 {
482         memcpy( fpu, &tsk->thread.i387.fsave,
483                 sizeof(struct user_i387_struct) );
484 }
485
486 static inline void copy_fpu_fxsave( struct task_struct *tsk,
487                                    struct user_i387_struct *fpu )
488 {
489         unsigned short *to;
490         unsigned short *from;
491         int i;
492
493         memcpy( fpu, &tsk->thread.i387.fxsave, 7 * sizeof(long) );
494
495         to = (unsigned short *)&fpu->st_space[0];
496         from = (unsigned short *)&tsk->thread.i387.fxsave.st_space[0];
497         for ( i = 0 ; i < 8 ; i++, to += 5, from += 8 ) {
498                 memcpy( to, from, 5 * sizeof(unsigned short) );
499         }
500 }
501
502 int dump_fpu( struct pt_regs *regs, struct user_i387_struct *fpu )
503 {
504         int fpvalid;
505         struct task_struct *tsk = current;
506
507         fpvalid = !!used_math();
508         if ( fpvalid ) {
509                 unlazy_fpu( tsk );
510                 if ( cpu_has_fxsr ) {
511                         copy_fpu_fxsave( tsk, fpu );
512                 } else {
513                         copy_fpu_fsave( tsk, fpu );
514                 }
515         }
516
517         return fpvalid;
518 }
519 EXPORT_SYMBOL(dump_fpu);
520
521 int dump_task_fpu(struct task_struct *tsk, struct user_i387_struct *fpu)
522 {
523         int fpvalid = !!tsk_used_math(tsk);
524
525         if (fpvalid) {
526                 if (tsk == current)
527                         unlazy_fpu(tsk);
528                 if (cpu_has_fxsr)
529                         copy_fpu_fxsave(tsk, fpu);
530                 else
531                         copy_fpu_fsave(tsk, fpu);
532         }
533         return fpvalid;
534 }
535
536 int dump_task_extended_fpu(struct task_struct *tsk, struct user_fxsr_struct *fpu)
537 {
538         int fpvalid = tsk_used_math(tsk) && cpu_has_fxsr;
539
540         if (fpvalid) {
541                 if (tsk == current)
542                        unlazy_fpu(tsk);
543                 memcpy(fpu, &tsk->thread.i387.fxsave, sizeof(*fpu));
544         }
545         return fpvalid;
546 }