Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
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 | */ | |
129f6946 | 8 | #include <linux/module.h> |
44210111 | 9 | #include <linux/regset.h> |
f668964e IM |
10 | #include <linux/sched.h> |
11 | ||
12 | #include <asm/sigcontext.h> | |
1da177e4 | 13 | #include <asm/processor.h> |
1da177e4 | 14 | #include <asm/math_emu.h> |
1da177e4 | 15 | #include <asm/uaccess.h> |
f668964e IM |
16 | #include <asm/ptrace.h> |
17 | #include <asm/i387.h> | |
18 | #include <asm/user.h> | |
1da177e4 | 19 | |
44210111 | 20 | #ifdef CONFIG_X86_64 |
f668964e IM |
21 | # include <asm/sigcontext32.h> |
22 | # include <asm/user32.h> | |
44210111 | 23 | #else |
f668964e IM |
24 | # define save_i387_ia32 save_i387 |
25 | # define restore_i387_ia32 restore_i387 | |
26 | # define _fpstate_ia32 _fpstate | |
27 | # define user_i387_ia32_struct user_i387_struct | |
28 | # define user32_fxsr_struct user_fxsr_struct | |
44210111 RM |
29 | #endif |
30 | ||
1da177e4 | 31 | #ifdef CONFIG_MATH_EMULATION |
f668964e | 32 | # define HAVE_HWFP (boot_cpu_data.hard_math) |
1da177e4 | 33 | #else |
f668964e | 34 | # define HAVE_HWFP 1 |
1da177e4 LT |
35 | #endif |
36 | ||
f668964e | 37 | static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu; |
1da177e4 LT |
38 | |
39 | void mxcsr_feature_mask_init(void) | |
40 | { | |
41 | unsigned long mask = 0; | |
f668964e | 42 | |
1da177e4 LT |
43 | clts(); |
44 | if (cpu_has_fxsr) { | |
3b095a04 CG |
45 | memset(¤t->thread.i387.fxsave, 0, |
46 | sizeof(struct i387_fxsave_struct)); | |
47 | asm volatile("fxsave %0" : : "m" (current->thread.i387.fxsave)); | |
1da177e4 | 48 | mask = current->thread.i387.fxsave.mxcsr_mask; |
3b095a04 CG |
49 | if (mask == 0) |
50 | mask = 0x0000ffbf; | |
51 | } | |
1da177e4 LT |
52 | mxcsr_feature_mask &= mask; |
53 | stts(); | |
54 | } | |
55 | ||
44210111 RM |
56 | #ifdef CONFIG_X86_64 |
57 | /* | |
58 | * Called at bootup to set up the initial FPU state that is later cloned | |
59 | * into all processes. | |
60 | */ | |
61 | void __cpuinit fpu_init(void) | |
62 | { | |
63 | unsigned long oldcr0 = read_cr0(); | |
64 | extern void __bad_fxsave_alignment(void); | |
65 | ||
66 | if (offsetof(struct task_struct, thread.i387.fxsave) & 15) | |
67 | __bad_fxsave_alignment(); | |
f668964e | 68 | |
44210111 RM |
69 | set_in_cr4(X86_CR4_OSFXSR); |
70 | set_in_cr4(X86_CR4_OSXMMEXCPT); | |
71 | ||
f668964e | 72 | write_cr0(oldcr0 & ~(X86_CR0_TS|X86_CR0_EM)); /* clear TS and EM */ |
44210111 RM |
73 | |
74 | mxcsr_feature_mask_init(); | |
75 | /* clean state in init */ | |
76 | current_thread_info()->status = 0; | |
77 | clear_used_math(); | |
78 | } | |
79 | #endif /* CONFIG_X86_64 */ | |
80 | ||
1da177e4 LT |
81 | /* |
82 | * The _current_ task is using the FPU for the first time | |
83 | * so initialize it and set the mxcsr to its default | |
84 | * value at reset if we support XMM instructions and then | |
85 | * remeber the current task has used the FPU. | |
86 | */ | |
87 | void init_fpu(struct task_struct *tsk) | |
88 | { | |
44210111 RM |
89 | if (tsk_used_math(tsk)) { |
90 | if (tsk == current) | |
91 | unlazy_fpu(tsk); | |
92 | return; | |
93 | } | |
94 | ||
1da177e4 | 95 | if (cpu_has_fxsr) { |
3b095a04 CG |
96 | memset(&tsk->thread.i387.fxsave, 0, |
97 | sizeof(struct i387_fxsave_struct)); | |
1da177e4 LT |
98 | tsk->thread.i387.fxsave.cwd = 0x37f; |
99 | if (cpu_has_xmm) | |
44210111 | 100 | tsk->thread.i387.fxsave.mxcsr = MXCSR_DEFAULT; |
1da177e4 | 101 | } else { |
3b095a04 CG |
102 | memset(&tsk->thread.i387.fsave, 0, |
103 | sizeof(struct i387_fsave_struct)); | |
1da177e4 LT |
104 | tsk->thread.i387.fsave.cwd = 0xffff037fu; |
105 | tsk->thread.i387.fsave.swd = 0xffff0000u; | |
106 | tsk->thread.i387.fsave.twd = 0xffffffffu; | |
107 | tsk->thread.i387.fsave.fos = 0xffff0000u; | |
108 | } | |
44210111 RM |
109 | /* |
110 | * Only the device not available exception or ptrace can call init_fpu. | |
111 | */ | |
1da177e4 LT |
112 | set_stopped_child_used_math(tsk); |
113 | } | |
114 | ||
44210111 RM |
115 | int fpregs_active(struct task_struct *target, const struct user_regset *regset) |
116 | { | |
117 | return tsk_used_math(target) ? regset->n : 0; | |
118 | } | |
1da177e4 | 119 | |
44210111 | 120 | int xfpregs_active(struct task_struct *target, const struct user_regset *regset) |
1da177e4 | 121 | { |
44210111 RM |
122 | return (cpu_has_fxsr && tsk_used_math(target)) ? regset->n : 0; |
123 | } | |
1da177e4 | 124 | |
44210111 RM |
125 | int xfpregs_get(struct task_struct *target, const struct user_regset *regset, |
126 | unsigned int pos, unsigned int count, | |
127 | void *kbuf, void __user *ubuf) | |
128 | { | |
129 | if (!cpu_has_fxsr) | |
130 | return -ENODEV; | |
131 | ||
18a86221 | 132 | init_fpu(target); |
44210111 RM |
133 | |
134 | return user_regset_copyout(&pos, &count, &kbuf, &ubuf, | |
135 | &target->thread.i387.fxsave, 0, -1); | |
1da177e4 | 136 | } |
44210111 RM |
137 | |
138 | int xfpregs_set(struct task_struct *target, const struct user_regset *regset, | |
139 | unsigned int pos, unsigned int count, | |
140 | const void *kbuf, const void __user *ubuf) | |
141 | { | |
142 | int ret; | |
143 | ||
144 | if (!cpu_has_fxsr) | |
145 | return -ENODEV; | |
146 | ||
18a86221 | 147 | init_fpu(target); |
44210111 RM |
148 | set_stopped_child_used_math(target); |
149 | ||
150 | ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, | |
151 | &target->thread.i387.fxsave, 0, -1); | |
152 | ||
153 | /* | |
154 | * mxcsr reserved bits must be masked to zero for security reasons. | |
155 | */ | |
156 | target->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask; | |
157 | ||
158 | return ret; | |
159 | } | |
160 | ||
161 | #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION | |
1da177e4 | 162 | |
1da177e4 LT |
163 | /* |
164 | * FPU tag word conversions. | |
165 | */ | |
166 | ||
3b095a04 | 167 | static inline unsigned short twd_i387_to_fxsr(unsigned short twd) |
1da177e4 LT |
168 | { |
169 | unsigned int tmp; /* to avoid 16 bit prefixes in the code */ | |
3b095a04 | 170 | |
1da177e4 | 171 | /* Transform each pair of bits into 01 (valid) or 00 (empty) */ |
3b095a04 | 172 | tmp = ~twd; |
44210111 | 173 | tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */ |
3b095a04 CG |
174 | /* and move the valid bits to the lower byte. */ |
175 | tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */ | |
176 | tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */ | |
177 | tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */ | |
f668964e | 178 | |
3b095a04 | 179 | return tmp; |
1da177e4 LT |
180 | } |
181 | ||
1da177e4 | 182 | #define FPREG_ADDR(f, n) ((void *)&(f)->st_space + (n) * 16); |
44210111 RM |
183 | #define FP_EXP_TAG_VALID 0 |
184 | #define FP_EXP_TAG_ZERO 1 | |
185 | #define FP_EXP_TAG_SPECIAL 2 | |
186 | #define FP_EXP_TAG_EMPTY 3 | |
187 | ||
188 | static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave) | |
189 | { | |
190 | struct _fpxreg *st; | |
191 | u32 tos = (fxsave->swd >> 11) & 7; | |
192 | u32 twd = (unsigned long) fxsave->twd; | |
193 | u32 tag; | |
194 | u32 ret = 0xffff0000u; | |
195 | int i; | |
1da177e4 | 196 | |
44210111 | 197 | for (i = 0; i < 8; i++, twd >>= 1) { |
3b095a04 CG |
198 | if (twd & 0x1) { |
199 | st = FPREG_ADDR(fxsave, (i - tos) & 7); | |
1da177e4 | 200 | |
3b095a04 | 201 | switch (st->exponent & 0x7fff) { |
1da177e4 | 202 | case 0x7fff: |
44210111 | 203 | tag = FP_EXP_TAG_SPECIAL; |
1da177e4 LT |
204 | break; |
205 | case 0x0000: | |
3b095a04 CG |
206 | if (!st->significand[0] && |
207 | !st->significand[1] && | |
208 | !st->significand[2] && | |
44210111 RM |
209 | !st->significand[3]) |
210 | tag = FP_EXP_TAG_ZERO; | |
211 | else | |
212 | tag = FP_EXP_TAG_SPECIAL; | |
1da177e4 LT |
213 | break; |
214 | default: | |
44210111 RM |
215 | if (st->significand[3] & 0x8000) |
216 | tag = FP_EXP_TAG_VALID; | |
217 | else | |
218 | tag = FP_EXP_TAG_SPECIAL; | |
1da177e4 LT |
219 | break; |
220 | } | |
221 | } else { | |
44210111 | 222 | tag = FP_EXP_TAG_EMPTY; |
1da177e4 | 223 | } |
44210111 | 224 | ret |= tag << (2 * i); |
1da177e4 LT |
225 | } |
226 | return ret; | |
227 | } | |
228 | ||
229 | /* | |
44210111 | 230 | * FXSR floating point environment conversions. |
1da177e4 LT |
231 | */ |
232 | ||
f668964e IM |
233 | static void |
234 | convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk) | |
1da177e4 | 235 | { |
44210111 RM |
236 | struct i387_fxsave_struct *fxsave = &tsk->thread.i387.fxsave; |
237 | struct _fpreg *to = (struct _fpreg *) &env->st_space[0]; | |
238 | struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0]; | |
239 | int i; | |
1da177e4 | 240 | |
44210111 RM |
241 | env->cwd = fxsave->cwd | 0xffff0000u; |
242 | env->swd = fxsave->swd | 0xffff0000u; | |
243 | env->twd = twd_fxsr_to_i387(fxsave); | |
244 | ||
245 | #ifdef CONFIG_X86_64 | |
246 | env->fip = fxsave->rip; | |
247 | env->foo = fxsave->rdp; | |
248 | if (tsk == current) { | |
249 | /* | |
250 | * should be actually ds/cs at fpu exception time, but | |
251 | * that information is not available in 64bit mode. | |
252 | */ | |
f668964e IM |
253 | asm("mov %%ds, %[fos]" : [fos] "=r" (env->fos)); |
254 | asm("mov %%cs, %[fcs]" : [fcs] "=r" (env->fcs)); | |
1da177e4 | 255 | } else { |
44210111 | 256 | struct pt_regs *regs = task_pt_regs(tsk); |
f668964e | 257 | |
44210111 RM |
258 | env->fos = 0xffff0000 | tsk->thread.ds; |
259 | env->fcs = regs->cs; | |
1da177e4 | 260 | } |
44210111 RM |
261 | #else |
262 | env->fip = fxsave->fip; | |
609b5297 | 263 | env->fcs = (u16) fxsave->fcs | ((u32) fxsave->fop << 16); |
44210111 RM |
264 | env->foo = fxsave->foo; |
265 | env->fos = fxsave->fos; | |
266 | #endif | |
1da177e4 | 267 | |
44210111 RM |
268 | for (i = 0; i < 8; ++i) |
269 | memcpy(&to[i], &from[i], sizeof(to[0])); | |
1da177e4 LT |
270 | } |
271 | ||
44210111 RM |
272 | static void convert_to_fxsr(struct task_struct *tsk, |
273 | const struct user_i387_ia32_struct *env) | |
1da177e4 | 274 | |
1da177e4 | 275 | { |
44210111 RM |
276 | struct i387_fxsave_struct *fxsave = &tsk->thread.i387.fxsave; |
277 | struct _fpreg *from = (struct _fpreg *) &env->st_space[0]; | |
278 | struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0]; | |
279 | int i; | |
1da177e4 | 280 | |
44210111 RM |
281 | fxsave->cwd = env->cwd; |
282 | fxsave->swd = env->swd; | |
283 | fxsave->twd = twd_i387_to_fxsr(env->twd); | |
284 | fxsave->fop = (u16) ((u32) env->fcs >> 16); | |
285 | #ifdef CONFIG_X86_64 | |
286 | fxsave->rip = env->fip; | |
287 | fxsave->rdp = env->foo; | |
288 | /* cs and ds ignored */ | |
289 | #else | |
290 | fxsave->fip = env->fip; | |
291 | fxsave->fcs = (env->fcs & 0xffff); | |
292 | fxsave->foo = env->foo; | |
293 | fxsave->fos = env->fos; | |
294 | #endif | |
1da177e4 | 295 | |
44210111 RM |
296 | for (i = 0; i < 8; ++i) |
297 | memcpy(&to[i], &from[i], sizeof(from[0])); | |
1da177e4 LT |
298 | } |
299 | ||
44210111 RM |
300 | int fpregs_get(struct task_struct *target, const struct user_regset *regset, |
301 | unsigned int pos, unsigned int count, | |
302 | void *kbuf, void __user *ubuf) | |
1da177e4 | 303 | { |
44210111 | 304 | struct user_i387_ia32_struct env; |
1da177e4 | 305 | |
44210111 RM |
306 | if (!HAVE_HWFP) |
307 | return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf); | |
1da177e4 | 308 | |
18a86221 | 309 | init_fpu(target); |
1da177e4 | 310 | |
f668964e | 311 | if (!cpu_has_fxsr) { |
44210111 RM |
312 | return user_regset_copyout(&pos, &count, &kbuf, &ubuf, |
313 | &target->thread.i387.fsave, 0, -1); | |
f668964e | 314 | } |
1da177e4 | 315 | |
44210111 RM |
316 | if (kbuf && pos == 0 && count == sizeof(env)) { |
317 | convert_from_fxsr(kbuf, target); | |
318 | return 0; | |
1da177e4 | 319 | } |
44210111 RM |
320 | |
321 | convert_from_fxsr(&env, target); | |
f668964e | 322 | |
44210111 | 323 | return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &env, 0, -1); |
1da177e4 LT |
324 | } |
325 | ||
44210111 RM |
326 | int fpregs_set(struct task_struct *target, const struct user_regset *regset, |
327 | unsigned int pos, unsigned int count, | |
328 | const void *kbuf, const void __user *ubuf) | |
1da177e4 | 329 | { |
44210111 RM |
330 | struct user_i387_ia32_struct env; |
331 | int ret; | |
1da177e4 | 332 | |
44210111 RM |
333 | if (!HAVE_HWFP) |
334 | return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf); | |
1da177e4 | 335 | |
18a86221 | 336 | init_fpu(target); |
44210111 RM |
337 | set_stopped_child_used_math(target); |
338 | ||
f668964e | 339 | if (!cpu_has_fxsr) { |
44210111 RM |
340 | return user_regset_copyin(&pos, &count, &kbuf, &ubuf, |
341 | &target->thread.i387.fsave, 0, -1); | |
f668964e | 342 | } |
44210111 RM |
343 | |
344 | if (pos > 0 || count < sizeof(env)) | |
345 | convert_from_fxsr(&env, target); | |
346 | ||
347 | ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &env, 0, -1); | |
348 | if (!ret) | |
349 | convert_to_fxsr(target, &env); | |
350 | ||
351 | return ret; | |
1da177e4 LT |
352 | } |
353 | ||
354 | /* | |
355 | * Signal frame handlers. | |
356 | */ | |
357 | ||
44210111 | 358 | static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf) |
1da177e4 LT |
359 | { |
360 | struct task_struct *tsk = current; | |
361 | ||
3b095a04 | 362 | unlazy_fpu(tsk); |
1da177e4 | 363 | tsk->thread.i387.fsave.status = tsk->thread.i387.fsave.swd; |
3b095a04 CG |
364 | if (__copy_to_user(buf, &tsk->thread.i387.fsave, |
365 | sizeof(struct i387_fsave_struct))) | |
1da177e4 LT |
366 | return -1; |
367 | return 1; | |
368 | } | |
369 | ||
44210111 | 370 | static int save_i387_fxsave(struct _fpstate_ia32 __user *buf) |
1da177e4 LT |
371 | { |
372 | struct task_struct *tsk = current; | |
44210111 | 373 | struct user_i387_ia32_struct env; |
1da177e4 LT |
374 | int err = 0; |
375 | ||
3b095a04 | 376 | unlazy_fpu(tsk); |
1da177e4 | 377 | |
44210111 RM |
378 | convert_from_fxsr(&env, tsk); |
379 | if (__copy_to_user(buf, &env, sizeof(env))) | |
1da177e4 LT |
380 | return -1; |
381 | ||
3b095a04 CG |
382 | err |= __put_user(tsk->thread.i387.fxsave.swd, &buf->status); |
383 | err |= __put_user(X86_FXSR_MAGIC, &buf->magic); | |
384 | if (err) | |
1da177e4 LT |
385 | return -1; |
386 | ||
3b095a04 CG |
387 | if (__copy_to_user(&buf->_fxsr_env[0], &tsk->thread.i387.fxsave, |
388 | sizeof(struct i387_fxsave_struct))) | |
1da177e4 LT |
389 | return -1; |
390 | return 1; | |
391 | } | |
392 | ||
44210111 | 393 | int save_i387_ia32(struct _fpstate_ia32 __user *buf) |
1da177e4 | 394 | { |
3b095a04 | 395 | if (!used_math()) |
1da177e4 | 396 | return 0; |
f668964e IM |
397 | /* |
398 | * This will cause a "finit" to be triggered by the next | |
1da177e4 LT |
399 | * attempted FPU operation by the 'current' process. |
400 | */ | |
401 | clear_used_math(); | |
402 | ||
f668964e | 403 | if (!HAVE_HWFP) { |
44210111 RM |
404 | return fpregs_soft_get(current, NULL, |
405 | 0, sizeof(struct user_i387_ia32_struct), | |
406 | NULL, buf) ? -1 : 1; | |
1da177e4 | 407 | } |
f668964e IM |
408 | |
409 | if (cpu_has_fxsr) | |
410 | return save_i387_fxsave(buf); | |
411 | else | |
412 | return save_i387_fsave(buf); | |
1da177e4 LT |
413 | } |
414 | ||
44210111 | 415 | static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf) |
1da177e4 LT |
416 | { |
417 | struct task_struct *tsk = current; | |
f668964e | 418 | |
3b095a04 CG |
419 | clear_fpu(tsk); |
420 | return __copy_from_user(&tsk->thread.i387.fsave, buf, | |
421 | sizeof(struct i387_fsave_struct)); | |
1da177e4 LT |
422 | } |
423 | ||
44210111 | 424 | static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf) |
1da177e4 | 425 | { |
1da177e4 | 426 | struct task_struct *tsk = current; |
44210111 | 427 | struct user_i387_ia32_struct env; |
f668964e IM |
428 | int err; |
429 | ||
3b095a04 CG |
430 | clear_fpu(tsk); |
431 | err = __copy_from_user(&tsk->thread.i387.fxsave, &buf->_fxsr_env[0], | |
432 | sizeof(struct i387_fxsave_struct)); | |
1da177e4 LT |
433 | /* mxcsr reserved bits must be masked to zero for security reasons */ |
434 | tsk->thread.i387.fxsave.mxcsr &= mxcsr_feature_mask; | |
44210111 RM |
435 | if (err || __copy_from_user(&env, buf, sizeof(env))) |
436 | return 1; | |
437 | convert_to_fxsr(tsk, &env); | |
f668964e | 438 | |
44210111 | 439 | return 0; |
1da177e4 LT |
440 | } |
441 | ||
44210111 | 442 | int restore_i387_ia32(struct _fpstate_ia32 __user *buf) |
1da177e4 LT |
443 | { |
444 | int err; | |
445 | ||
3b095a04 | 446 | if (HAVE_HWFP) { |
f668964e | 447 | if (cpu_has_fxsr) |
3b095a04 | 448 | err = restore_i387_fxsave(buf); |
f668964e | 449 | else |
3b095a04 | 450 | err = restore_i387_fsave(buf); |
1da177e4 | 451 | } else { |
44210111 RM |
452 | err = fpregs_soft_set(current, NULL, |
453 | 0, sizeof(struct user_i387_ia32_struct), | |
454 | NULL, buf) != 0; | |
1da177e4 LT |
455 | } |
456 | set_used_math(); | |
f668964e | 457 | |
1da177e4 LT |
458 | return err; |
459 | } | |
460 | ||
1da177e4 LT |
461 | /* |
462 | * FPU state for core dumps. | |
60b3b9af RM |
463 | * This is only used for a.out dumps now. |
464 | * It is declared generically using elf_fpregset_t (which is | |
465 | * struct user_i387_struct) but is in fact only used for 32-bit | |
466 | * dumps, so on 64-bit it is really struct user_i387_ia32_struct. | |
1da177e4 | 467 | */ |
3b095a04 | 468 | int dump_fpu(struct pt_regs *regs, struct user_i387_struct *fpu) |
1da177e4 | 469 | { |
1da177e4 | 470 | struct task_struct *tsk = current; |
f668964e | 471 | int fpvalid; |
1da177e4 LT |
472 | |
473 | fpvalid = !!used_math(); | |
60b3b9af RM |
474 | if (fpvalid) |
475 | fpvalid = !fpregs_get(tsk, NULL, | |
476 | 0, sizeof(struct user_i387_ia32_struct), | |
477 | fpu, NULL); | |
1da177e4 LT |
478 | |
479 | return fpvalid; | |
480 | } | |
129f6946 | 481 | EXPORT_SYMBOL(dump_fpu); |
1da177e4 | 482 | |
60b3b9af | 483 | #endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */ |