Merge branches 'release' and 'autoload' into release
[linux-2.6] / arch / x86 / kernel / i387.c
1 /*
2  *  Copyright (C) 1994 Linus Torvalds
3  *
4  *  Pentium III FXSR, SSE support
5  *  General FPU state handling cleanups
6  *      Gareth Hughes <gareth@valinux.com>, May 2000
7  */
8
9 #include <linux/sched.h>
10 #include <linux/module.h>
11 #include <linux/regset.h>
12 #include <asm/processor.h>
13 #include <asm/i387.h>
14 #include <asm/math_emu.h>
15 #include <asm/sigcontext.h>
16 #include <asm/user.h>
17 #include <asm/ptrace.h>
18 #include <asm/uaccess.h>
19
20 #ifdef CONFIG_X86_64
21
22 #include <asm/sigcontext32.h>
23 #include <asm/user32.h>
24
25 #else
26
27 #define save_i387_ia32          save_i387
28 #define restore_i387_ia32       restore_i387
29
30 #define _fpstate_ia32           _fpstate
31 #define user_i387_ia32_struct   user_i387_struct
32 #define user32_fxsr_struct      user_fxsr_struct
33
34 #endif
35
36 #ifdef CONFIG_MATH_EMULATION
37 #define HAVE_HWFP (boot_cpu_data.hard_math)
38 #else
39 #define HAVE_HWFP 1
40 #endif
41
42 unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
43
44 void mxcsr_feature_mask_init(void)
45 {
46         unsigned long mask = 0;
47         clts();
48         if (cpu_has_fxsr) {
49                 memset(&current->thread.i387.fxsave, 0,
50                        sizeof(struct i387_fxsave_struct));
51                 asm volatile("fxsave %0" : : "m" (current->thread.i387.fxsave));
52                 mask = current->thread.i387.fxsave.mxcsr_mask;
53                 if (mask == 0)
54                         mask = 0x0000ffbf;
55         }
56         mxcsr_feature_mask &= mask;
57         stts();
58 }
59
60 #ifdef CONFIG_X86_64
61 /*
62  * Called at bootup to set up the initial FPU state that is later cloned
63  * into all processes.
64  */
65 void __cpuinit fpu_init(void)
66 {
67         unsigned long oldcr0 = read_cr0();
68         extern void __bad_fxsave_alignment(void);
69
70         if (offsetof(struct task_struct, thread.i387.fxsave) & 15)
71                 __bad_fxsave_alignment();
72         set_in_cr4(X86_CR4_OSFXSR);
73         set_in_cr4(X86_CR4_OSXMMEXCPT);
74
75         write_cr0(oldcr0 & ~((1UL<<3)|(1UL<<2))); /* clear TS and EM */
76
77         mxcsr_feature_mask_init();
78         /* clean state in init */
79         current_thread_info()->status = 0;
80         clear_used_math();
81 }
82 #endif  /* CONFIG_X86_64 */
83
84 /*
85  * The _current_ task is using the FPU for the first time
86  * so initialize it and set the mxcsr to its default
87  * value at reset if we support XMM instructions and then
88  * remeber the current task has used the FPU.
89  */
90 void init_fpu(struct task_struct *tsk)
91 {
92         if (tsk_used_math(tsk)) {
93                 if (tsk == current)
94                         unlazy_fpu(tsk);
95                 return;
96         }
97
98         if (cpu_has_fxsr) {
99                 memset(&tsk->thread.i387.fxsave, 0,
100                        sizeof(struct i387_fxsave_struct));
101                 tsk->thread.i387.fxsave.cwd = 0x37f;
102                 if (cpu_has_xmm)
103                         tsk->thread.i387.fxsave.mxcsr = MXCSR_DEFAULT;
104         } else {
105                 memset(&tsk->thread.i387.fsave, 0,
106                        sizeof(struct i387_fsave_struct));
107                 tsk->thread.i387.fsave.cwd = 0xffff037fu;
108                 tsk->thread.i387.fsave.swd = 0xffff0000u;
109                 tsk->thread.i387.fsave.twd = 0xffffffffu;
110                 tsk->thread.i387.fsave.fos = 0xffff0000u;
111         }
112         /*
113          * Only the device not available exception or ptrace can call init_fpu.
114          */
115         set_stopped_child_used_math(tsk);
116 }
117
118 int fpregs_active(struct task_struct *target, const struct user_regset *regset)
119 {
120         return tsk_used_math(target) ? regset->n : 0;
121 }
122
123 int xfpregs_active(struct task_struct *target, const struct user_regset *regset)
124 {
125         return (cpu_has_fxsr && tsk_used_math(target)) ? regset->n : 0;
126 }
127
128 int xfpregs_get(struct task_struct *target, const struct user_regset *regset,
129                 unsigned int pos, unsigned int count,
130                 void *kbuf, void __user *ubuf)
131 {
132         if (!cpu_has_fxsr)
133                 return -ENODEV;
134
135         unlazy_fpu(target);
136
137         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
138                                    &target->thread.i387.fxsave, 0, -1);
139 }
140
141 int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
142                 unsigned int pos, unsigned int count,
143                 const void *kbuf, const void __user *ubuf)
144 {
145         int ret;
146
147         if (!cpu_has_fxsr)
148                 return -ENODEV;
149
150         unlazy_fpu(target);
151         set_stopped_child_used_math(target);
152
153         ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
154                                  &target->thread.i387.fxsave, 0, -1);
155
156         /*
157          * mxcsr reserved bits must be masked to zero for security reasons.
158          */
159         target->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
160
161         return ret;
162 }
163
164 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
165
166 /*
167  * FPU tag word conversions.
168  */
169
170 static inline unsigned short twd_i387_to_fxsr(unsigned short twd)
171 {
172         unsigned int tmp; /* to avoid 16 bit prefixes in the code */
173
174         /* Transform each pair of bits into 01 (valid) or 00 (empty) */
175         tmp = ~twd;
176         tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */
177         /* and move the valid bits to the lower byte. */
178         tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */
179         tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */
180         tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */
181         return tmp;
182 }
183
184 #define FPREG_ADDR(f, n)        ((void *)&(f)->st_space + (n) * 16);
185 #define FP_EXP_TAG_VALID        0
186 #define FP_EXP_TAG_ZERO         1
187 #define FP_EXP_TAG_SPECIAL      2
188 #define FP_EXP_TAG_EMPTY        3
189
190 static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave)
191 {
192         struct _fpxreg *st;
193         u32 tos = (fxsave->swd >> 11) & 7;
194         u32 twd = (unsigned long) fxsave->twd;
195         u32 tag;
196         u32 ret = 0xffff0000u;
197         int i;
198
199         for (i = 0; i < 8; i++, twd >>= 1) {
200                 if (twd & 0x1) {
201                         st = FPREG_ADDR(fxsave, (i - tos) & 7);
202
203                         switch (st->exponent & 0x7fff) {
204                         case 0x7fff:
205                                 tag = FP_EXP_TAG_SPECIAL;
206                                 break;
207                         case 0x0000:
208                                 if (!st->significand[0] &&
209                                     !st->significand[1] &&
210                                     !st->significand[2] &&
211                                     !st->significand[3])
212                                         tag = FP_EXP_TAG_ZERO;
213                                 else
214                                         tag = FP_EXP_TAG_SPECIAL;
215                                 break;
216                         default:
217                                 if (st->significand[3] & 0x8000)
218                                         tag = FP_EXP_TAG_VALID;
219                                 else
220                                         tag = FP_EXP_TAG_SPECIAL;
221                                 break;
222                         }
223                 } else {
224                         tag = FP_EXP_TAG_EMPTY;
225                 }
226                 ret |= tag << (2 * i);
227         }
228         return ret;
229 }
230
231 /*
232  * FXSR floating point environment conversions.
233  */
234
235 static void convert_from_fxsr(struct user_i387_ia32_struct *env,
236                               struct task_struct *tsk)
237 {
238         struct i387_fxsave_struct *fxsave = &tsk->thread.i387.fxsave;
239         struct _fpreg *to = (struct _fpreg *) &env->st_space[0];
240         struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0];
241         int i;
242
243         env->cwd = fxsave->cwd | 0xffff0000u;
244         env->swd = fxsave->swd | 0xffff0000u;
245         env->twd = twd_fxsr_to_i387(fxsave);
246
247 #ifdef CONFIG_X86_64
248         env->fip = fxsave->rip;
249         env->foo = fxsave->rdp;
250         if (tsk == current) {
251                 /*
252                  * should be actually ds/cs at fpu exception time, but
253                  * that information is not available in 64bit mode.
254                  */
255                 asm("mov %%ds,%0" : "=r" (env->fos));
256                 asm("mov %%cs,%0" : "=r" (env->fcs));
257         } else {
258                 struct pt_regs *regs = task_pt_regs(tsk);
259                 env->fos = 0xffff0000 | tsk->thread.ds;
260                 env->fcs = regs->cs;
261         }
262 #else
263         env->fip = fxsave->fip;
264         env->fcs = fxsave->fcs;
265         env->foo = fxsave->foo;
266         env->fos = fxsave->fos;
267 #endif
268
269         for (i = 0; i < 8; ++i)
270                 memcpy(&to[i], &from[i], sizeof(to[0]));
271 }
272
273 static void convert_to_fxsr(struct task_struct *tsk,
274                             const struct user_i387_ia32_struct *env)
275
276 {
277         struct i387_fxsave_struct *fxsave = &tsk->thread.i387.fxsave;
278         struct _fpreg *from = (struct _fpreg *) &env->st_space[0];
279         struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0];
280         int i;
281
282         fxsave->cwd = env->cwd;
283         fxsave->swd = env->swd;
284         fxsave->twd = twd_i387_to_fxsr(env->twd);
285         fxsave->fop = (u16) ((u32) env->fcs >> 16);
286 #ifdef CONFIG_X86_64
287         fxsave->rip = env->fip;
288         fxsave->rdp = env->foo;
289         /* cs and ds ignored */
290 #else
291         fxsave->fip = env->fip;
292         fxsave->fcs = (env->fcs & 0xffff);
293         fxsave->foo = env->foo;
294         fxsave->fos = env->fos;
295 #endif
296
297         for (i = 0; i < 8; ++i)
298                 memcpy(&to[i], &from[i], sizeof(from[0]));
299 }
300
301 int fpregs_get(struct task_struct *target, const struct user_regset *regset,
302                unsigned int pos, unsigned int count,
303                void *kbuf, void __user *ubuf)
304 {
305         struct user_i387_ia32_struct env;
306
307         if (!HAVE_HWFP)
308                 return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf);
309
310         unlazy_fpu(target);
311
312         if (!cpu_has_fxsr)
313                 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
314                                            &target->thread.i387.fsave, 0, -1);
315
316         if (kbuf && pos == 0 && count == sizeof(env)) {
317                 convert_from_fxsr(kbuf, target);
318                 return 0;
319         }
320
321         convert_from_fxsr(&env, target);
322         return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
323 }
324
325 int fpregs_set(struct task_struct *target, const struct user_regset *regset,
326                unsigned int pos, unsigned int count,
327                const void *kbuf, const void __user *ubuf)
328 {
329         struct user_i387_ia32_struct env;
330         int ret;
331
332         if (!HAVE_HWFP)
333                 return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf);
334
335         unlazy_fpu(target);
336         set_stopped_child_used_math(target);
337
338         if (!cpu_has_fxsr)
339                 return user_regset_copyin(&pos, &count, &kbuf, &ubuf,
340                                           &target->thread.i387.fsave, 0, -1);
341
342         if (pos > 0 || count < sizeof(env))
343                 convert_from_fxsr(&env, target);
344
345         ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
346         if (!ret)
347                 convert_to_fxsr(target, &env);
348
349         return ret;
350 }
351
352 /*
353  * Signal frame handlers.
354  */
355
356 static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf)
357 {
358         struct task_struct *tsk = current;
359
360         unlazy_fpu(tsk);
361         tsk->thread.i387.fsave.status = tsk->thread.i387.fsave.swd;
362         if (__copy_to_user(buf, &tsk->thread.i387.fsave,
363                            sizeof(struct i387_fsave_struct)))
364                 return -1;
365         return 1;
366 }
367
368 static int save_i387_fxsave(struct _fpstate_ia32 __user *buf)
369 {
370         struct task_struct *tsk = current;
371         struct user_i387_ia32_struct env;
372         int err = 0;
373
374         unlazy_fpu(tsk);
375
376         convert_from_fxsr(&env, tsk);
377         if (__copy_to_user(buf, &env, sizeof(env)))
378                 return -1;
379
380         err |= __put_user(tsk->thread.i387.fxsave.swd, &buf->status);
381         err |= __put_user(X86_FXSR_MAGIC, &buf->magic);
382         if (err)
383                 return -1;
384
385         if (__copy_to_user(&buf->_fxsr_env[0], &tsk->thread.i387.fxsave,
386                            sizeof(struct i387_fxsave_struct)))
387                 return -1;
388         return 1;
389 }
390
391 int save_i387_ia32(struct _fpstate_ia32 __user *buf)
392 {
393         if (!used_math())
394                 return 0;
395
396         /* This will cause a "finit" to be triggered by the next
397          * attempted FPU operation by the 'current' process.
398          */
399         clear_used_math();
400
401         if (HAVE_HWFP) {
402                 if (cpu_has_fxsr) {
403                         return save_i387_fxsave(buf);
404                 } else {
405                         return save_i387_fsave(buf);
406                 }
407         } else {
408                 return fpregs_soft_get(current, NULL,
409                                        0, sizeof(struct user_i387_ia32_struct),
410                                        NULL, buf) ? -1 : 1;
411         }
412 }
413
414 static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf)
415 {
416         struct task_struct *tsk = current;
417         clear_fpu(tsk);
418         return __copy_from_user(&tsk->thread.i387.fsave, buf,
419                                 sizeof(struct i387_fsave_struct));
420 }
421
422 static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf)
423 {
424         int err;
425         struct task_struct *tsk = current;
426         struct user_i387_ia32_struct env;
427         clear_fpu(tsk);
428         err = __copy_from_user(&tsk->thread.i387.fxsave, &buf->_fxsr_env[0],
429                                sizeof(struct i387_fxsave_struct));
430         /* mxcsr reserved bits must be masked to zero for security reasons */
431         tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask;
432         if (err || __copy_from_user(&env, buf, sizeof(env)))
433                 return 1;
434         convert_to_fxsr(tsk, &env);
435         return 0;
436 }
437
438 int restore_i387_ia32(struct _fpstate_ia32 __user *buf)
439 {
440         int err;
441
442         if (HAVE_HWFP) {
443                 if (cpu_has_fxsr) {
444                         err = restore_i387_fxsave(buf);
445                 } else {
446                         err = restore_i387_fsave(buf);
447                 }
448         } else {
449                 err = fpregs_soft_set(current, NULL,
450                                       0, sizeof(struct user_i387_ia32_struct),
451                                       NULL, buf) != 0;
452         }
453         set_used_math();
454         return err;
455 }
456
457 /*
458  * FPU state for core dumps.
459  * This is only used for a.out dumps now.
460  * It is declared generically using elf_fpregset_t (which is
461  * struct user_i387_struct) but is in fact only used for 32-bit
462  * dumps, so on 64-bit it is really struct user_i387_ia32_struct.
463  */
464 int dump_fpu(struct pt_regs *regs, struct user_i387_struct *fpu)
465 {
466         int fpvalid;
467         struct task_struct *tsk = current;
468
469         fpvalid = !!used_math();
470         if (fpvalid)
471                 fpvalid = !fpregs_get(tsk, NULL,
472                                       0, sizeof(struct user_i387_ia32_struct),
473                                       fpu, NULL);
474
475         return fpvalid;
476 }
477 EXPORT_SYMBOL(dump_fpu);
478
479 #endif  /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */