2 * linux/arch/i386/kernel/i387.c
4 * Copyright (C) 1994 Linus Torvalds
6 * Pentium III FXSR, SSE support
7 * General FPU state handling cleanups
8 * Gareth Hughes <gareth@valinux.com>, May 2000
11 #include <linux/sched.h>
12 #include <linux/module.h>
13 #include <asm/processor.h>
15 #include <asm/math_emu.h>
16 #include <asm/sigcontext.h>
18 #include <asm/ptrace.h>
19 #include <asm/uaccess.h>
21 #ifdef CONFIG_MATH_EMULATION
22 #define HAVE_HWFP (boot_cpu_data.hard_math)
27 static unsigned long mxcsr_feature_mask __read_mostly = 0xffffffff;
29 void mxcsr_feature_mask_init(void)
31 unsigned long mask = 0;
34 memset(¤t->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;
39 mxcsr_feature_mask &= mask;
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.
49 void init_fpu(struct task_struct *tsk)
52 memset(&tsk->thread.i387.fxsave, 0, sizeof(struct i387_fxsave_struct));
53 tsk->thread.i387.fxsave.cwd = 0x37f;
55 tsk->thread.i387.fxsave.mxcsr = 0x1f80;
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;
63 /* only the device not available exception or ptrace can call init_fpu */
64 set_stopped_child_used_math(tsk);
68 * FPU lazy state save handling.
71 void kernel_fpu_begin(void)
73 struct thread_info *thread = current_thread_info();
76 if (thread->status & TS_USEDFPU) {
77 __save_init_fpu(thread->task);
82 EXPORT_SYMBOL_GPL(kernel_fpu_begin);
85 * FPU tag word conversions.
88 static inline unsigned short twd_i387_to_fxsr( unsigned short twd )
90 unsigned int tmp; /* to avoid 16 bit prefixes in the code */
92 /* Transform each pair of bits into 01 (valid) or 00 (empty) */
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 */
102 static inline unsigned long twd_fxsr_to_i387( struct i387_fxsave_struct *fxsave )
104 struct _fpxreg *st = NULL;
105 unsigned long tos = (fxsave->swd >> 11) & 7;
106 unsigned long twd = (unsigned long) fxsave->twd;
108 unsigned long ret = 0xffff0000u;
111 #define FPREG_ADDR(f, n) ((void *)&(f)->st_space + (n) * 16);
113 for ( i = 0 ; i < 8 ; i++ ) {
115 st = FPREG_ADDR( fxsave, (i - tos) & 7 );
117 switch ( st->exponent & 0x7fff ) {
119 tag = 2; /* Special */
122 if ( !st->significand[0] &&
123 !st->significand[1] &&
124 !st->significand[2] &&
125 !st->significand[3] ) {
128 tag = 2; /* Special */
132 if ( st->significand[3] & 0x8000 ) {
135 tag = 2; /* Special */
142 ret |= (tag << (2 * i));
149 * FPU state interaction.
152 unsigned short get_fpu_cwd( struct task_struct *tsk )
154 if ( cpu_has_fxsr ) {
155 return tsk->thread.i387.fxsave.cwd;
157 return (unsigned short)tsk->thread.i387.fsave.cwd;
161 unsigned short get_fpu_swd( struct task_struct *tsk )
163 if ( cpu_has_fxsr ) {
164 return tsk->thread.i387.fxsave.swd;
166 return (unsigned short)tsk->thread.i387.fsave.swd;
171 unsigned short get_fpu_twd( struct task_struct *tsk )
173 if ( cpu_has_fxsr ) {
174 return tsk->thread.i387.fxsave.twd;
176 return (unsigned short)tsk->thread.i387.fsave.twd;
181 unsigned short get_fpu_mxcsr( struct task_struct *tsk )
184 return tsk->thread.i387.fxsave.mxcsr;
192 void set_fpu_cwd( struct task_struct *tsk, unsigned short cwd )
194 if ( cpu_has_fxsr ) {
195 tsk->thread.i387.fxsave.cwd = cwd;
197 tsk->thread.i387.fsave.cwd = ((long)cwd | 0xffff0000u);
201 void set_fpu_swd( struct task_struct *tsk, unsigned short swd )
203 if ( cpu_has_fxsr ) {
204 tsk->thread.i387.fxsave.swd = swd;
206 tsk->thread.i387.fsave.swd = ((long)swd | 0xffff0000u);
210 void set_fpu_twd( struct task_struct *tsk, unsigned short twd )
212 if ( cpu_has_fxsr ) {
213 tsk->thread.i387.fxsave.twd = twd_i387_to_fxsr(twd);
215 tsk->thread.i387.fsave.twd = ((long)twd | 0xffff0000u);
222 * FXSR floating point environment conversions.
226 convert_fxsr_env_to_i387(unsigned long env[7],
227 struct i387_fxsave_struct *fxsave)
229 env[0] = (unsigned long)fxsave->cwd | 0xffff0000ul;
230 env[1] = (unsigned long)fxsave->swd | 0xffff0000ul;
231 env[2] = twd_fxsr_to_i387(fxsave);
232 env[3] = fxsave->fip;
233 env[4] = fxsave->fcs | ((unsigned long)fxsave->fop << 16);
234 env[5] = fxsave->foo;
235 env[6] = fxsave->fos;
238 static int convert_fxsr_to_user(struct _fpstate __user *buf,
239 struct i387_fxsave_struct *fxsave)
241 unsigned long env[7];
242 struct _fpreg __user *to;
243 struct _fpxreg *from;
246 convert_fxsr_env_to_i387(env, fxsave);
247 if ( __copy_to_user( buf, env, 7 * sizeof(unsigned long) ) )
251 from = (struct _fpxreg *) &fxsave->st_space[0];
252 for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
253 unsigned long __user *t = (unsigned long __user *)to;
254 unsigned long *f = (unsigned long *)from;
256 if (__put_user(*f, t) ||
257 __put_user(*(f + 1), t + 1) ||
258 __put_user(from->exponent, &to->exponent))
265 convert_fxsr_env_from_i387(struct i387_fxsave_struct *fxsave,
266 const unsigned long env[7])
268 fxsave->cwd = (unsigned short)(env[0] & 0xffff);
269 fxsave->swd = (unsigned short)(env[1] & 0xffff);
270 fxsave->twd = twd_i387_to_fxsr((unsigned short)(env[2] & 0xffff));
271 fxsave->fip = env[3];
272 fxsave->fop = (unsigned short)((env[4] & 0xffff0000ul) >> 16);
273 fxsave->fcs = (env[4] & 0xffff);
274 fxsave->foo = env[5];
275 fxsave->fos = env[6];
278 static int convert_fxsr_from_user( struct i387_fxsave_struct *fxsave,
279 struct _fpstate __user *buf )
281 unsigned long env[7];
283 struct _fpreg __user *from;
286 if ( __copy_from_user( env, buf, 7 * sizeof(long) ) )
289 convert_fxsr_env_from_i387(fxsave, env);
291 to = (struct _fpxreg *) &fxsave->st_space[0];
293 for ( i = 0 ; i < 8 ; i++, to++, from++ ) {
294 unsigned long *t = (unsigned long *)to;
295 unsigned long __user *f = (unsigned long __user *)from;
297 if (__get_user(*t, f) ||
298 __get_user(*(t + 1), f + 1) ||
299 __get_user(to->exponent, &from->exponent))
306 * Signal frame handlers.
309 static inline int save_i387_fsave( struct _fpstate __user *buf )
311 struct task_struct *tsk = current;
314 tsk->thread.i387.fsave.status = tsk->thread.i387.fsave.swd;
315 if ( __copy_to_user( buf, &tsk->thread.i387.fsave,
316 sizeof(struct i387_fsave_struct) ) )
321 static int save_i387_fxsave( struct _fpstate __user *buf )
323 struct task_struct *tsk = current;
328 if ( convert_fxsr_to_user( buf, &tsk->thread.i387.fxsave ) )
331 err |= __put_user( tsk->thread.i387.fxsave.swd, &buf->status );
332 err |= __put_user( X86_FXSR_MAGIC, &buf->magic );
336 if ( __copy_to_user( &buf->_fxsr_env[0], &tsk->thread.i387.fxsave,
337 sizeof(struct i387_fxsave_struct) ) )
342 int save_i387( struct _fpstate __user *buf )
347 /* This will cause a "finit" to be triggered by the next
348 * attempted FPU operation by the 'current' process.
353 if ( cpu_has_fxsr ) {
354 return save_i387_fxsave( buf );
356 return save_i387_fsave( buf );
359 return save_i387_soft( ¤t->thread.i387.soft, buf );
363 static inline int restore_i387_fsave( struct _fpstate __user *buf )
365 struct task_struct *tsk = current;
367 return __copy_from_user( &tsk->thread.i387.fsave, buf,
368 sizeof(struct i387_fsave_struct) );
371 static int restore_i387_fxsave( struct _fpstate __user *buf )
374 struct task_struct *tsk = current;
376 err = __copy_from_user( &tsk->thread.i387.fxsave, &buf->_fxsr_env[0],
377 sizeof(struct i387_fxsave_struct) );
378 /* mxcsr reserved bits must be masked to zero for security reasons */
379 tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
380 return err ? 1 : convert_fxsr_from_user( &tsk->thread.i387.fxsave, buf );
383 int restore_i387( struct _fpstate __user *buf )
388 if ( cpu_has_fxsr ) {
389 err = restore_i387_fxsave( buf );
391 err = restore_i387_fsave( buf );
394 err = restore_i387_soft( ¤t->thread.i387.soft, buf );
401 * ptrace request handlers.
404 static inline void get_fpregs_fsave(struct user_i387_struct *buf,
405 struct task_struct *tsk)
407 memcpy(buf, &tsk->thread.i387.fsave, sizeof(struct user_i387_struct));
410 static inline void get_fpregs_fxsave(struct user_i387_struct *buf,
411 struct task_struct *tsk)
414 const struct _fpxreg *from;
417 convert_fxsr_env_to_i387((unsigned long *) buf,
418 &tsk->thread.i387.fxsave);
420 to = (struct _fpreg *) buf->st_space;
421 from = (const struct _fpxreg *) &tsk->thread.i387.fxsave.st_space[0];
422 for (i = 0; i < 8; i++, to++, from++)
423 *to = *(const struct _fpreg *) from;
426 int get_fpregs(struct user_i387_struct *buf, struct task_struct *tsk)
430 get_fpregs_fxsave(buf, tsk);
432 get_fpregs_fsave(buf, tsk);
435 return save_i387_soft( &tsk->thread.i387.soft,
436 (struct _fpstate __user *)buf );
440 static inline void set_fpregs_fsave(struct task_struct *tsk,
441 const struct user_i387_struct *buf)
443 memcpy(&tsk->thread.i387.fsave, buf, sizeof(struct user_i387_struct));
446 static inline void set_fpregs_fxsave(struct task_struct *tsk,
447 const struct user_i387_struct *buf)
450 const struct _fpreg *from;
453 convert_fxsr_env_from_i387(&tsk->thread.i387.fxsave,
454 (unsigned long *) buf);
456 to = (struct _fpxreg *) &tsk->thread.i387.fxsave.st_space[0];
457 from = (const struct _fpreg *) buf->st_space;
458 for (i = 0; i < 8; i++, to++, from++)
459 *(struct _fpreg *) to = *from;
462 int set_fpregs(struct task_struct *tsk, const struct user_i387_struct *buf)
466 set_fpregs_fxsave(tsk, buf);
468 set_fpregs_fsave(tsk, buf);
471 return restore_i387_soft( &tsk->thread.i387.soft,
472 (struct _fpstate __user *)buf );
476 void updated_fpxregs(struct task_struct *tsk)
478 /* mxcsr reserved bits must be masked to zero for security reasons */
479 tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
483 * FPU state for core dumps.
486 static inline void copy_fpu_fsave( struct task_struct *tsk,
487 struct user_i387_struct *fpu )
489 memcpy( fpu, &tsk->thread.i387.fsave,
490 sizeof(struct user_i387_struct) );
493 static inline void copy_fpu_fxsave( struct task_struct *tsk,
494 struct user_i387_struct *fpu )
497 unsigned short *from;
500 memcpy( fpu, &tsk->thread.i387.fxsave, 7 * sizeof(long) );
502 to = (unsigned short *)&fpu->st_space[0];
503 from = (unsigned short *)&tsk->thread.i387.fxsave.st_space[0];
504 for ( i = 0 ; i < 8 ; i++, to += 5, from += 8 ) {
505 memcpy( to, from, 5 * sizeof(unsigned short) );
509 int dump_fpu( struct pt_regs *regs, struct user_i387_struct *fpu )
512 struct task_struct *tsk = current;
514 fpvalid = !!used_math();
517 if ( cpu_has_fxsr ) {
518 copy_fpu_fxsave( tsk, fpu );
520 copy_fpu_fsave( tsk, fpu );
526 EXPORT_SYMBOL(dump_fpu);
528 int dump_task_fpu(struct task_struct *tsk, struct user_i387_struct *fpu)
530 int fpvalid = !!tsk_used_math(tsk);
536 copy_fpu_fxsave(tsk, fpu);
538 copy_fpu_fsave(tsk, fpu);
543 int dump_task_extended_fpu(struct task_struct *tsk, struct user_fxsr_struct *fpu)
545 int fpvalid = tsk_used_math(tsk) && cpu_has_fxsr;
550 memcpy(fpu, &tsk->thread.i387.fxsave, sizeof(*fpu));