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 | 10 | #include <linux/sched.h> |
5a0e3ad6 | 11 | #include <linux/slab.h> |
f668964e IM |
12 | |
13 | #include <asm/sigcontext.h> | |
1da177e4 | 14 | #include <asm/processor.h> |
1da177e4 | 15 | #include <asm/math_emu.h> |
1da177e4 | 16 | #include <asm/uaccess.h> |
f668964e IM |
17 | #include <asm/ptrace.h> |
18 | #include <asm/i387.h> | |
1361b83a | 19 | #include <asm/fpu-internal.h> |
f668964e | 20 | #include <asm/user.h> |
1da177e4 | 21 | |
44210111 | 22 | #ifdef CONFIG_X86_64 |
f668964e IM |
23 | # include <asm/sigcontext32.h> |
24 | # include <asm/user32.h> | |
44210111 | 25 | #else |
ab513701 SS |
26 | # define save_i387_xstate_ia32 save_i387_xstate |
27 | # define restore_i387_xstate_ia32 restore_i387_xstate | |
f668964e | 28 | # define _fpstate_ia32 _fpstate |
ab513701 | 29 | # define _xstate_ia32 _xstate |
3c1c7f10 | 30 | # define sig_xstate_ia32_size sig_xstate_size |
c37b5efe | 31 | # define fx_sw_reserved_ia32 fx_sw_reserved |
f668964e IM |
32 | # define user_i387_ia32_struct user_i387_struct |
33 | # define user32_fxsr_struct user_fxsr_struct | |
44210111 RM |
34 | #endif |
35 | ||
8546c008 LT |
36 | /* |
37 | * Were we in an interrupt that interrupted kernel mode? | |
38 | * | |
39 | * We can do a kernel_fpu_begin/end() pair *ONLY* if that | |
40 | * pair does nothing at all: the thread must not have fpu (so | |
41 | * that we don't try to save the FPU state), and TS must | |
42 | * be set (so that the clts/stts pair does nothing that is | |
43 | * visible in the interrupted kernel thread). | |
44 | */ | |
45 | static inline bool interrupted_kernel_fpu_idle(void) | |
46 | { | |
47 | return !__thread_has_fpu(current) && | |
48 | (read_cr0() & X86_CR0_TS); | |
49 | } | |
50 | ||
51 | /* | |
52 | * Were we in user mode (or vm86 mode) when we were | |
53 | * interrupted? | |
54 | * | |
55 | * Doing kernel_fpu_begin/end() is ok if we are running | |
56 | * in an interrupt context from user mode - we'll just | |
57 | * save the FPU state as required. | |
58 | */ | |
59 | static inline bool interrupted_user_mode(void) | |
60 | { | |
61 | struct pt_regs *regs = get_irq_regs(); | |
62 | return regs && user_mode_vm(regs); | |
63 | } | |
64 | ||
65 | /* | |
66 | * Can we use the FPU in kernel mode with the | |
67 | * whole "kernel_fpu_begin/end()" sequence? | |
68 | * | |
69 | * It's always ok in process context (ie "not interrupt") | |
70 | * but it is sometimes ok even from an irq. | |
71 | */ | |
72 | bool irq_fpu_usable(void) | |
73 | { | |
74 | return !in_interrupt() || | |
75 | interrupted_user_mode() || | |
76 | interrupted_kernel_fpu_idle(); | |
77 | } | |
78 | EXPORT_SYMBOL(irq_fpu_usable); | |
79 | ||
80 | void kernel_fpu_begin(void) | |
81 | { | |
82 | struct task_struct *me = current; | |
83 | ||
84 | WARN_ON_ONCE(!irq_fpu_usable()); | |
85 | preempt_disable(); | |
86 | if (__thread_has_fpu(me)) { | |
87 | __save_init_fpu(me); | |
88 | __thread_clear_has_fpu(me); | |
89 | /* We do 'stts()' in kernel_fpu_end() */ | |
90 | } else { | |
91 | percpu_write(fpu_owner_task, NULL); | |
92 | clts(); | |
93 | } | |
94 | } | |
95 | EXPORT_SYMBOL(kernel_fpu_begin); | |
96 | ||
97 | void kernel_fpu_end(void) | |
98 | { | |
99 | stts(); | |
100 | preempt_enable(); | |
101 | } | |
102 | EXPORT_SYMBOL(kernel_fpu_end); | |
103 | ||
104 | void unlazy_fpu(struct task_struct *tsk) | |
105 | { | |
106 | preempt_disable(); | |
107 | if (__thread_has_fpu(tsk)) { | |
108 | __save_init_fpu(tsk); | |
109 | __thread_fpu_end(tsk); | |
110 | } else | |
111 | tsk->fpu_counter = 0; | |
112 | preempt_enable(); | |
113 | } | |
114 | EXPORT_SYMBOL(unlazy_fpu); | |
115 | ||
1da177e4 | 116 | #ifdef CONFIG_MATH_EMULATION |
f668964e | 117 | # define HAVE_HWFP (boot_cpu_data.hard_math) |
1da177e4 | 118 | #else |
f668964e | 119 | # define HAVE_HWFP 1 |
1da177e4 LT |
120 | #endif |
121 | ||
f668964e | 122 | static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu; |
61c4628b | 123 | unsigned int xstate_size; |
f45755b8 | 124 | EXPORT_SYMBOL_GPL(xstate_size); |
3c1c7f10 | 125 | unsigned int sig_xstate_ia32_size = sizeof(struct _fpstate_ia32); |
61c4628b | 126 | static struct i387_fxsave_struct fx_scratch __cpuinitdata; |
1da177e4 | 127 | |
1361b83a | 128 | static void __cpuinit mxcsr_feature_mask_init(void) |
1da177e4 LT |
129 | { |
130 | unsigned long mask = 0; | |
f668964e | 131 | |
1da177e4 LT |
132 | clts(); |
133 | if (cpu_has_fxsr) { | |
61c4628b SS |
134 | memset(&fx_scratch, 0, sizeof(struct i387_fxsave_struct)); |
135 | asm volatile("fxsave %0" : : "m" (fx_scratch)); | |
136 | mask = fx_scratch.mxcsr_mask; | |
3b095a04 CG |
137 | if (mask == 0) |
138 | mask = 0x0000ffbf; | |
139 | } | |
1da177e4 LT |
140 | mxcsr_feature_mask &= mask; |
141 | stts(); | |
142 | } | |
143 | ||
0e49bf66 | 144 | static void __cpuinit init_thread_xstate(void) |
61c4628b | 145 | { |
0e49bf66 RR |
146 | /* |
147 | * Note that xstate_size might be overwriten later during | |
148 | * xsave_init(). | |
149 | */ | |
150 | ||
e8a496ac | 151 | if (!HAVE_HWFP) { |
1f999ab5 RR |
152 | /* |
153 | * Disable xsave as we do not support it if i387 | |
154 | * emulation is enabled. | |
155 | */ | |
156 | setup_clear_cpu_cap(X86_FEATURE_XSAVE); | |
157 | setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); | |
e8a496ac SS |
158 | xstate_size = sizeof(struct i387_soft_struct); |
159 | return; | |
160 | } | |
161 | ||
61c4628b SS |
162 | if (cpu_has_fxsr) |
163 | xstate_size = sizeof(struct i387_fxsave_struct); | |
61c4628b SS |
164 | else |
165 | xstate_size = sizeof(struct i387_fsave_struct); | |
61c4628b SS |
166 | } |
167 | ||
44210111 RM |
168 | /* |
169 | * Called at bootup to set up the initial FPU state that is later cloned | |
170 | * into all processes. | |
171 | */ | |
0e49bf66 | 172 | |
44210111 RM |
173 | void __cpuinit fpu_init(void) |
174 | { | |
6ac8bac2 BG |
175 | unsigned long cr0; |
176 | unsigned long cr4_mask = 0; | |
44210111 | 177 | |
6ac8bac2 BG |
178 | if (cpu_has_fxsr) |
179 | cr4_mask |= X86_CR4_OSFXSR; | |
180 | if (cpu_has_xmm) | |
181 | cr4_mask |= X86_CR4_OSXMMEXCPT; | |
182 | if (cr4_mask) | |
183 | set_in_cr4(cr4_mask); | |
184 | ||
185 | cr0 = read_cr0(); | |
186 | cr0 &= ~(X86_CR0_TS|X86_CR0_EM); /* clear TS and EM */ | |
187 | if (!HAVE_HWFP) | |
188 | cr0 |= X86_CR0_EM; | |
189 | write_cr0(cr0); | |
44210111 | 190 | |
dc1e35c6 SS |
191 | if (!smp_processor_id()) |
192 | init_thread_xstate(); | |
dc1e35c6 | 193 | |
44210111 RM |
194 | mxcsr_feature_mask_init(); |
195 | /* clean state in init */ | |
c9ad4882 | 196 | current_thread_info()->status = 0; |
44210111 RM |
197 | clear_used_math(); |
198 | } | |
0e49bf66 | 199 | |
5ee481da | 200 | void fpu_finit(struct fpu *fpu) |
1da177e4 | 201 | { |
e8a496ac | 202 | if (!HAVE_HWFP) { |
86603283 AK |
203 | finit_soft_fpu(&fpu->state->soft); |
204 | return; | |
e8a496ac | 205 | } |
e8a496ac | 206 | |
1da177e4 | 207 | if (cpu_has_fxsr) { |
86603283 | 208 | struct i387_fxsave_struct *fx = &fpu->state->fxsave; |
61c4628b SS |
209 | |
210 | memset(fx, 0, xstate_size); | |
211 | fx->cwd = 0x37f; | |
1da177e4 | 212 | if (cpu_has_xmm) |
61c4628b | 213 | fx->mxcsr = MXCSR_DEFAULT; |
1da177e4 | 214 | } else { |
86603283 | 215 | struct i387_fsave_struct *fp = &fpu->state->fsave; |
61c4628b SS |
216 | memset(fp, 0, xstate_size); |
217 | fp->cwd = 0xffff037fu; | |
218 | fp->swd = 0xffff0000u; | |
219 | fp->twd = 0xffffffffu; | |
220 | fp->fos = 0xffff0000u; | |
1da177e4 | 221 | } |
86603283 | 222 | } |
5ee481da | 223 | EXPORT_SYMBOL_GPL(fpu_finit); |
86603283 AK |
224 | |
225 | /* | |
226 | * The _current_ task is using the FPU for the first time | |
227 | * so initialize it and set the mxcsr to its default | |
228 | * value at reset if we support XMM instructions and then | |
0d2eb44f | 229 | * remember the current task has used the FPU. |
86603283 AK |
230 | */ |
231 | int init_fpu(struct task_struct *tsk) | |
232 | { | |
233 | int ret; | |
234 | ||
235 | if (tsk_used_math(tsk)) { | |
236 | if (HAVE_HWFP && tsk == current) | |
237 | unlazy_fpu(tsk); | |
089f9fba | 238 | tsk->thread.fpu.last_cpu = ~0; |
86603283 AK |
239 | return 0; |
240 | } | |
241 | ||
44210111 | 242 | /* |
86603283 | 243 | * Memory allocation at the first usage of the FPU and other state. |
44210111 | 244 | */ |
86603283 AK |
245 | ret = fpu_alloc(&tsk->thread.fpu); |
246 | if (ret) | |
247 | return ret; | |
248 | ||
249 | fpu_finit(&tsk->thread.fpu); | |
250 | ||
1da177e4 | 251 | set_stopped_child_used_math(tsk); |
aa283f49 | 252 | return 0; |
1da177e4 | 253 | } |
e5c30142 | 254 | EXPORT_SYMBOL_GPL(init_fpu); |
1da177e4 | 255 | |
5b3efd50 SS |
256 | /* |
257 | * The xstateregs_active() routine is the same as the fpregs_active() routine, | |
258 | * as the "regset->n" for the xstate regset will be updated based on the feature | |
259 | * capabilites supported by the xsave. | |
260 | */ | |
44210111 RM |
261 | int fpregs_active(struct task_struct *target, const struct user_regset *regset) |
262 | { | |
263 | return tsk_used_math(target) ? regset->n : 0; | |
264 | } | |
1da177e4 | 265 | |
44210111 | 266 | int xfpregs_active(struct task_struct *target, const struct user_regset *regset) |
1da177e4 | 267 | { |
44210111 RM |
268 | return (cpu_has_fxsr && tsk_used_math(target)) ? regset->n : 0; |
269 | } | |
1da177e4 | 270 | |
44210111 RM |
271 | int xfpregs_get(struct task_struct *target, const struct user_regset *regset, |
272 | unsigned int pos, unsigned int count, | |
273 | void *kbuf, void __user *ubuf) | |
274 | { | |
aa283f49 SS |
275 | int ret; |
276 | ||
44210111 RM |
277 | if (!cpu_has_fxsr) |
278 | return -ENODEV; | |
279 | ||
aa283f49 SS |
280 | ret = init_fpu(target); |
281 | if (ret) | |
282 | return ret; | |
44210111 | 283 | |
29104e10 SS |
284 | sanitize_i387_state(target); |
285 | ||
44210111 | 286 | return user_regset_copyout(&pos, &count, &kbuf, &ubuf, |
86603283 | 287 | &target->thread.fpu.state->fxsave, 0, -1); |
1da177e4 | 288 | } |
44210111 RM |
289 | |
290 | int xfpregs_set(struct task_struct *target, const struct user_regset *regset, | |
291 | unsigned int pos, unsigned int count, | |
292 | const void *kbuf, const void __user *ubuf) | |
293 | { | |
294 | int ret; | |
295 | ||
296 | if (!cpu_has_fxsr) | |
297 | return -ENODEV; | |
298 | ||
aa283f49 SS |
299 | ret = init_fpu(target); |
300 | if (ret) | |
301 | return ret; | |
302 | ||
29104e10 SS |
303 | sanitize_i387_state(target); |
304 | ||
44210111 | 305 | ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, |
86603283 | 306 | &target->thread.fpu.state->fxsave, 0, -1); |
44210111 RM |
307 | |
308 | /* | |
309 | * mxcsr reserved bits must be masked to zero for security reasons. | |
310 | */ | |
86603283 | 311 | target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask; |
44210111 | 312 | |
42deec6f SS |
313 | /* |
314 | * update the header bits in the xsave header, indicating the | |
315 | * presence of FP and SSE state. | |
316 | */ | |
317 | if (cpu_has_xsave) | |
86603283 | 318 | target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE; |
42deec6f | 319 | |
44210111 RM |
320 | return ret; |
321 | } | |
322 | ||
5b3efd50 SS |
323 | int xstateregs_get(struct task_struct *target, const struct user_regset *regset, |
324 | unsigned int pos, unsigned int count, | |
325 | void *kbuf, void __user *ubuf) | |
326 | { | |
327 | int ret; | |
328 | ||
329 | if (!cpu_has_xsave) | |
330 | return -ENODEV; | |
331 | ||
332 | ret = init_fpu(target); | |
333 | if (ret) | |
334 | return ret; | |
335 | ||
336 | /* | |
ff7fbc72 SS |
337 | * Copy the 48bytes defined by the software first into the xstate |
338 | * memory layout in the thread struct, so that we can copy the entire | |
339 | * xstateregs to the user using one user_regset_copyout(). | |
5b3efd50 | 340 | */ |
86603283 | 341 | memcpy(&target->thread.fpu.state->fxsave.sw_reserved, |
ff7fbc72 | 342 | xstate_fx_sw_bytes, sizeof(xstate_fx_sw_bytes)); |
5b3efd50 SS |
343 | |
344 | /* | |
ff7fbc72 | 345 | * Copy the xstate memory layout. |
5b3efd50 SS |
346 | */ |
347 | ret = user_regset_copyout(&pos, &count, &kbuf, &ubuf, | |
86603283 | 348 | &target->thread.fpu.state->xsave, 0, -1); |
5b3efd50 SS |
349 | return ret; |
350 | } | |
351 | ||
352 | int xstateregs_set(struct task_struct *target, const struct user_regset *regset, | |
353 | unsigned int pos, unsigned int count, | |
354 | const void *kbuf, const void __user *ubuf) | |
355 | { | |
356 | int ret; | |
357 | struct xsave_hdr_struct *xsave_hdr; | |
358 | ||
359 | if (!cpu_has_xsave) | |
360 | return -ENODEV; | |
361 | ||
362 | ret = init_fpu(target); | |
363 | if (ret) | |
364 | return ret; | |
365 | ||
366 | ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, | |
86603283 | 367 | &target->thread.fpu.state->xsave, 0, -1); |
5b3efd50 SS |
368 | |
369 | /* | |
370 | * mxcsr reserved bits must be masked to zero for security reasons. | |
371 | */ | |
86603283 | 372 | target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask; |
5b3efd50 | 373 | |
86603283 | 374 | xsave_hdr = &target->thread.fpu.state->xsave.xsave_hdr; |
5b3efd50 SS |
375 | |
376 | xsave_hdr->xstate_bv &= pcntxt_mask; | |
377 | /* | |
378 | * These bits must be zero. | |
379 | */ | |
380 | xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0; | |
381 | ||
382 | return ret; | |
383 | } | |
384 | ||
44210111 | 385 | #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION |
1da177e4 | 386 | |
1da177e4 LT |
387 | /* |
388 | * FPU tag word conversions. | |
389 | */ | |
390 | ||
3b095a04 | 391 | static inline unsigned short twd_i387_to_fxsr(unsigned short twd) |
1da177e4 LT |
392 | { |
393 | unsigned int tmp; /* to avoid 16 bit prefixes in the code */ | |
3b095a04 | 394 | |
1da177e4 | 395 | /* Transform each pair of bits into 01 (valid) or 00 (empty) */ |
3b095a04 | 396 | tmp = ~twd; |
44210111 | 397 | tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */ |
3b095a04 CG |
398 | /* and move the valid bits to the lower byte. */ |
399 | tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */ | |
400 | tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */ | |
401 | tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */ | |
f668964e | 402 | |
3b095a04 | 403 | return tmp; |
1da177e4 LT |
404 | } |
405 | ||
497888cf | 406 | #define FPREG_ADDR(f, n) ((void *)&(f)->st_space + (n) * 16) |
44210111 RM |
407 | #define FP_EXP_TAG_VALID 0 |
408 | #define FP_EXP_TAG_ZERO 1 | |
409 | #define FP_EXP_TAG_SPECIAL 2 | |
410 | #define FP_EXP_TAG_EMPTY 3 | |
411 | ||
412 | static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave) | |
413 | { | |
414 | struct _fpxreg *st; | |
415 | u32 tos = (fxsave->swd >> 11) & 7; | |
416 | u32 twd = (unsigned long) fxsave->twd; | |
417 | u32 tag; | |
418 | u32 ret = 0xffff0000u; | |
419 | int i; | |
1da177e4 | 420 | |
44210111 | 421 | for (i = 0; i < 8; i++, twd >>= 1) { |
3b095a04 CG |
422 | if (twd & 0x1) { |
423 | st = FPREG_ADDR(fxsave, (i - tos) & 7); | |
1da177e4 | 424 | |
3b095a04 | 425 | switch (st->exponent & 0x7fff) { |
1da177e4 | 426 | case 0x7fff: |
44210111 | 427 | tag = FP_EXP_TAG_SPECIAL; |
1da177e4 LT |
428 | break; |
429 | case 0x0000: | |
3b095a04 CG |
430 | if (!st->significand[0] && |
431 | !st->significand[1] && | |
432 | !st->significand[2] && | |
44210111 RM |
433 | !st->significand[3]) |
434 | tag = FP_EXP_TAG_ZERO; | |
435 | else | |
436 | tag = FP_EXP_TAG_SPECIAL; | |
1da177e4 LT |
437 | break; |
438 | default: | |
44210111 RM |
439 | if (st->significand[3] & 0x8000) |
440 | tag = FP_EXP_TAG_VALID; | |
441 | else | |
442 | tag = FP_EXP_TAG_SPECIAL; | |
1da177e4 LT |
443 | break; |
444 | } | |
445 | } else { | |
44210111 | 446 | tag = FP_EXP_TAG_EMPTY; |
1da177e4 | 447 | } |
44210111 | 448 | ret |= tag << (2 * i); |
1da177e4 LT |
449 | } |
450 | return ret; | |
451 | } | |
452 | ||
453 | /* | |
44210111 | 454 | * FXSR floating point environment conversions. |
1da177e4 LT |
455 | */ |
456 | ||
f668964e IM |
457 | static void |
458 | convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk) | |
1da177e4 | 459 | { |
86603283 | 460 | struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave; |
44210111 RM |
461 | struct _fpreg *to = (struct _fpreg *) &env->st_space[0]; |
462 | struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0]; | |
463 | int i; | |
1da177e4 | 464 | |
44210111 RM |
465 | env->cwd = fxsave->cwd | 0xffff0000u; |
466 | env->swd = fxsave->swd | 0xffff0000u; | |
467 | env->twd = twd_fxsr_to_i387(fxsave); | |
468 | ||
469 | #ifdef CONFIG_X86_64 | |
470 | env->fip = fxsave->rip; | |
471 | env->foo = fxsave->rdp; | |
10c11f30 BG |
472 | /* |
473 | * should be actually ds/cs at fpu exception time, but | |
474 | * that information is not available in 64bit mode. | |
475 | */ | |
476 | env->fcs = task_pt_regs(tsk)->cs; | |
44210111 | 477 | if (tsk == current) { |
10c11f30 | 478 | savesegment(ds, env->fos); |
1da177e4 | 479 | } else { |
10c11f30 | 480 | env->fos = tsk->thread.ds; |
1da177e4 | 481 | } |
10c11f30 | 482 | env->fos |= 0xffff0000; |
44210111 RM |
483 | #else |
484 | env->fip = fxsave->fip; | |
609b5297 | 485 | env->fcs = (u16) fxsave->fcs | ((u32) fxsave->fop << 16); |
44210111 RM |
486 | env->foo = fxsave->foo; |
487 | env->fos = fxsave->fos; | |
488 | #endif | |
1da177e4 | 489 | |
44210111 RM |
490 | for (i = 0; i < 8; ++i) |
491 | memcpy(&to[i], &from[i], sizeof(to[0])); | |
1da177e4 LT |
492 | } |
493 | ||
44210111 RM |
494 | static void convert_to_fxsr(struct task_struct *tsk, |
495 | const struct user_i387_ia32_struct *env) | |
1da177e4 | 496 | |
1da177e4 | 497 | { |
86603283 | 498 | struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave; |
44210111 RM |
499 | struct _fpreg *from = (struct _fpreg *) &env->st_space[0]; |
500 | struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0]; | |
501 | int i; | |
1da177e4 | 502 | |
44210111 RM |
503 | fxsave->cwd = env->cwd; |
504 | fxsave->swd = env->swd; | |
505 | fxsave->twd = twd_i387_to_fxsr(env->twd); | |
506 | fxsave->fop = (u16) ((u32) env->fcs >> 16); | |
507 | #ifdef CONFIG_X86_64 | |
508 | fxsave->rip = env->fip; | |
509 | fxsave->rdp = env->foo; | |
510 | /* cs and ds ignored */ | |
511 | #else | |
512 | fxsave->fip = env->fip; | |
513 | fxsave->fcs = (env->fcs & 0xffff); | |
514 | fxsave->foo = env->foo; | |
515 | fxsave->fos = env->fos; | |
516 | #endif | |
1da177e4 | 517 | |
44210111 RM |
518 | for (i = 0; i < 8; ++i) |
519 | memcpy(&to[i], &from[i], sizeof(from[0])); | |
1da177e4 LT |
520 | } |
521 | ||
44210111 RM |
522 | int fpregs_get(struct task_struct *target, const struct user_regset *regset, |
523 | unsigned int pos, unsigned int count, | |
524 | void *kbuf, void __user *ubuf) | |
1da177e4 | 525 | { |
44210111 | 526 | struct user_i387_ia32_struct env; |
aa283f49 | 527 | int ret; |
1da177e4 | 528 | |
aa283f49 SS |
529 | ret = init_fpu(target); |
530 | if (ret) | |
531 | return ret; | |
1da177e4 | 532 | |
e8a496ac SS |
533 | if (!HAVE_HWFP) |
534 | return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf); | |
535 | ||
f668964e | 536 | if (!cpu_has_fxsr) { |
44210111 | 537 | return user_regset_copyout(&pos, &count, &kbuf, &ubuf, |
86603283 | 538 | &target->thread.fpu.state->fsave, 0, |
61c4628b | 539 | -1); |
f668964e | 540 | } |
1da177e4 | 541 | |
29104e10 SS |
542 | sanitize_i387_state(target); |
543 | ||
44210111 RM |
544 | if (kbuf && pos == 0 && count == sizeof(env)) { |
545 | convert_from_fxsr(kbuf, target); | |
546 | return 0; | |
1da177e4 | 547 | } |
44210111 RM |
548 | |
549 | convert_from_fxsr(&env, target); | |
f668964e | 550 | |
44210111 | 551 | return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &env, 0, -1); |
1da177e4 LT |
552 | } |
553 | ||
44210111 RM |
554 | int fpregs_set(struct task_struct *target, const struct user_regset *regset, |
555 | unsigned int pos, unsigned int count, | |
556 | const void *kbuf, const void __user *ubuf) | |
1da177e4 | 557 | { |
44210111 RM |
558 | struct user_i387_ia32_struct env; |
559 | int ret; | |
1da177e4 | 560 | |
aa283f49 SS |
561 | ret = init_fpu(target); |
562 | if (ret) | |
563 | return ret; | |
564 | ||
29104e10 SS |
565 | sanitize_i387_state(target); |
566 | ||
e8a496ac SS |
567 | if (!HAVE_HWFP) |
568 | return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf); | |
569 | ||
f668964e | 570 | if (!cpu_has_fxsr) { |
44210111 | 571 | return user_regset_copyin(&pos, &count, &kbuf, &ubuf, |
86603283 | 572 | &target->thread.fpu.state->fsave, 0, -1); |
f668964e | 573 | } |
44210111 RM |
574 | |
575 | if (pos > 0 || count < sizeof(env)) | |
576 | convert_from_fxsr(&env, target); | |
577 | ||
578 | ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &env, 0, -1); | |
579 | if (!ret) | |
580 | convert_to_fxsr(target, &env); | |
581 | ||
42deec6f SS |
582 | /* |
583 | * update the header bit in the xsave header, indicating the | |
584 | * presence of FP. | |
585 | */ | |
586 | if (cpu_has_xsave) | |
86603283 | 587 | target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FP; |
44210111 | 588 | return ret; |
1da177e4 LT |
589 | } |
590 | ||
591 | /* | |
592 | * Signal frame handlers. | |
593 | */ | |
594 | ||
44210111 | 595 | static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf) |
1da177e4 LT |
596 | { |
597 | struct task_struct *tsk = current; | |
86603283 | 598 | struct i387_fsave_struct *fp = &tsk->thread.fpu.state->fsave; |
1da177e4 | 599 | |
61c4628b SS |
600 | fp->status = fp->swd; |
601 | if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct))) | |
1da177e4 LT |
602 | return -1; |
603 | return 1; | |
604 | } | |
605 | ||
44210111 | 606 | static int save_i387_fxsave(struct _fpstate_ia32 __user *buf) |
1da177e4 LT |
607 | { |
608 | struct task_struct *tsk = current; | |
86603283 | 609 | struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave; |
44210111 | 610 | struct user_i387_ia32_struct env; |
1da177e4 LT |
611 | int err = 0; |
612 | ||
44210111 RM |
613 | convert_from_fxsr(&env, tsk); |
614 | if (__copy_to_user(buf, &env, sizeof(env))) | |
1da177e4 LT |
615 | return -1; |
616 | ||
61c4628b | 617 | err |= __put_user(fx->swd, &buf->status); |
3b095a04 CG |
618 | err |= __put_user(X86_FXSR_MAGIC, &buf->magic); |
619 | if (err) | |
1da177e4 LT |
620 | return -1; |
621 | ||
c37b5efe | 622 | if (__copy_to_user(&buf->_fxsr_env[0], fx, xstate_size)) |
1da177e4 LT |
623 | return -1; |
624 | return 1; | |
625 | } | |
626 | ||
c37b5efe SS |
627 | static int save_i387_xsave(void __user *buf) |
628 | { | |
04944b79 | 629 | struct task_struct *tsk = current; |
c37b5efe SS |
630 | struct _fpstate_ia32 __user *fx = buf; |
631 | int err = 0; | |
632 | ||
29104e10 SS |
633 | |
634 | sanitize_i387_state(tsk); | |
635 | ||
04944b79 SS |
636 | /* |
637 | * For legacy compatible, we always set FP/SSE bits in the bit | |
638 | * vector while saving the state to the user context. | |
639 | * This will enable us capturing any changes(during sigreturn) to | |
640 | * the FP/SSE bits by the legacy applications which don't touch | |
641 | * xstate_bv in the xsave header. | |
642 | * | |
643 | * xsave aware applications can change the xstate_bv in the xsave | |
644 | * header as well as change any contents in the memory layout. | |
645 | * xrestore as part of sigreturn will capture all the changes. | |
646 | */ | |
86603283 | 647 | tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE; |
04944b79 | 648 | |
c37b5efe SS |
649 | if (save_i387_fxsave(fx) < 0) |
650 | return -1; | |
651 | ||
652 | err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved_ia32, | |
653 | sizeof(struct _fpx_sw_bytes)); | |
654 | err |= __put_user(FP_XSTATE_MAGIC2, | |
655 | (__u32 __user *) (buf + sig_xstate_ia32_size | |
656 | - FP_XSTATE_MAGIC2_SIZE)); | |
657 | if (err) | |
658 | return -1; | |
659 | ||
660 | return 1; | |
661 | } | |
662 | ||
ab513701 | 663 | int save_i387_xstate_ia32(void __user *buf) |
1da177e4 | 664 | { |
ab513701 SS |
665 | struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf; |
666 | struct task_struct *tsk = current; | |
667 | ||
3b095a04 | 668 | if (!used_math()) |
1da177e4 | 669 | return 0; |
ab513701 SS |
670 | |
671 | if (!access_ok(VERIFY_WRITE, buf, sig_xstate_ia32_size)) | |
672 | return -EACCES; | |
f668964e IM |
673 | /* |
674 | * This will cause a "finit" to be triggered by the next | |
1da177e4 LT |
675 | * attempted FPU operation by the 'current' process. |
676 | */ | |
677 | clear_used_math(); | |
678 | ||
f668964e | 679 | if (!HAVE_HWFP) { |
44210111 RM |
680 | return fpregs_soft_get(current, NULL, |
681 | 0, sizeof(struct user_i387_ia32_struct), | |
ab513701 | 682 | NULL, fp) ? -1 : 1; |
1da177e4 | 683 | } |
f668964e | 684 | |
ab513701 SS |
685 | unlazy_fpu(tsk); |
686 | ||
c37b5efe SS |
687 | if (cpu_has_xsave) |
688 | return save_i387_xsave(fp); | |
f668964e | 689 | if (cpu_has_fxsr) |
ab513701 | 690 | return save_i387_fxsave(fp); |
f668964e | 691 | else |
ab513701 | 692 | return save_i387_fsave(fp); |
1da177e4 LT |
693 | } |
694 | ||
44210111 | 695 | static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf) |
1da177e4 LT |
696 | { |
697 | struct task_struct *tsk = current; | |
f668964e | 698 | |
86603283 | 699 | return __copy_from_user(&tsk->thread.fpu.state->fsave, buf, |
3b095a04 | 700 | sizeof(struct i387_fsave_struct)); |
1da177e4 LT |
701 | } |
702 | ||
c37b5efe SS |
703 | static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf, |
704 | unsigned int size) | |
1da177e4 | 705 | { |
1da177e4 | 706 | struct task_struct *tsk = current; |
44210111 | 707 | struct user_i387_ia32_struct env; |
f668964e IM |
708 | int err; |
709 | ||
86603283 | 710 | err = __copy_from_user(&tsk->thread.fpu.state->fxsave, &buf->_fxsr_env[0], |
c37b5efe | 711 | size); |
1da177e4 | 712 | /* mxcsr reserved bits must be masked to zero for security reasons */ |
86603283 | 713 | tsk->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask; |
44210111 RM |
714 | if (err || __copy_from_user(&env, buf, sizeof(env))) |
715 | return 1; | |
716 | convert_to_fxsr(tsk, &env); | |
f668964e | 717 | |
44210111 | 718 | return 0; |
1da177e4 LT |
719 | } |
720 | ||
c37b5efe SS |
721 | static int restore_i387_xsave(void __user *buf) |
722 | { | |
723 | struct _fpx_sw_bytes fx_sw_user; | |
724 | struct _fpstate_ia32 __user *fx_user = | |
725 | ((struct _fpstate_ia32 __user *) buf); | |
726 | struct i387_fxsave_struct __user *fx = | |
727 | (struct i387_fxsave_struct __user *) &fx_user->_fxsr_env[0]; | |
728 | struct xsave_hdr_struct *xsave_hdr = | |
86603283 | 729 | ¤t->thread.fpu.state->xsave.xsave_hdr; |
6152e4b1 | 730 | u64 mask; |
c37b5efe SS |
731 | int err; |
732 | ||
733 | if (check_for_xstate(fx, buf, &fx_sw_user)) | |
734 | goto fx_only; | |
735 | ||
6152e4b1 | 736 | mask = fx_sw_user.xstate_bv; |
c37b5efe SS |
737 | |
738 | err = restore_i387_fxsave(buf, fx_sw_user.xstate_size); | |
739 | ||
6152e4b1 | 740 | xsave_hdr->xstate_bv &= pcntxt_mask; |
c37b5efe SS |
741 | /* |
742 | * These bits must be zero. | |
743 | */ | |
744 | xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0; | |
745 | ||
746 | /* | |
747 | * Init the state that is not present in the memory layout | |
748 | * and enabled by the OS. | |
749 | */ | |
6152e4b1 PA |
750 | mask = ~(pcntxt_mask & ~mask); |
751 | xsave_hdr->xstate_bv &= mask; | |
c37b5efe SS |
752 | |
753 | return err; | |
754 | fx_only: | |
755 | /* | |
756 | * Couldn't find the extended state information in the memory | |
757 | * layout. Restore the FP/SSE and init the other extended state | |
758 | * enabled by the OS. | |
759 | */ | |
760 | xsave_hdr->xstate_bv = XSTATE_FPSSE; | |
761 | return restore_i387_fxsave(buf, sizeof(struct i387_fxsave_struct)); | |
762 | } | |
763 | ||
ab513701 | 764 | int restore_i387_xstate_ia32(void __user *buf) |
1da177e4 LT |
765 | { |
766 | int err; | |
e8a496ac | 767 | struct task_struct *tsk = current; |
ab513701 | 768 | struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf; |
1da177e4 | 769 | |
e8a496ac | 770 | if (HAVE_HWFP) |
fd3c3ed5 SS |
771 | clear_fpu(tsk); |
772 | ||
ab513701 SS |
773 | if (!buf) { |
774 | if (used_math()) { | |
775 | clear_fpu(tsk); | |
776 | clear_used_math(); | |
777 | } | |
778 | ||
779 | return 0; | |
780 | } else | |
781 | if (!access_ok(VERIFY_READ, buf, sig_xstate_ia32_size)) | |
782 | return -EACCES; | |
783 | ||
e8a496ac SS |
784 | if (!used_math()) { |
785 | err = init_fpu(tsk); | |
786 | if (err) | |
787 | return err; | |
788 | } | |
fd3c3ed5 | 789 | |
e8a496ac | 790 | if (HAVE_HWFP) { |
c37b5efe SS |
791 | if (cpu_has_xsave) |
792 | err = restore_i387_xsave(buf); | |
793 | else if (cpu_has_fxsr) | |
794 | err = restore_i387_fxsave(fp, sizeof(struct | |
795 | i387_fxsave_struct)); | |
f668964e | 796 | else |
ab513701 | 797 | err = restore_i387_fsave(fp); |
1da177e4 | 798 | } else { |
44210111 RM |
799 | err = fpregs_soft_set(current, NULL, |
800 | 0, sizeof(struct user_i387_ia32_struct), | |
ab513701 | 801 | NULL, fp) != 0; |
1da177e4 LT |
802 | } |
803 | set_used_math(); | |
f668964e | 804 | |
1da177e4 LT |
805 | return err; |
806 | } | |
807 | ||
1da177e4 LT |
808 | /* |
809 | * FPU state for core dumps. | |
60b3b9af RM |
810 | * This is only used for a.out dumps now. |
811 | * It is declared generically using elf_fpregset_t (which is | |
812 | * struct user_i387_struct) but is in fact only used for 32-bit | |
813 | * dumps, so on 64-bit it is really struct user_i387_ia32_struct. | |
1da177e4 | 814 | */ |
3b095a04 | 815 | int dump_fpu(struct pt_regs *regs, struct user_i387_struct *fpu) |
1da177e4 | 816 | { |
1da177e4 | 817 | struct task_struct *tsk = current; |
f668964e | 818 | int fpvalid; |
1da177e4 LT |
819 | |
820 | fpvalid = !!used_math(); | |
60b3b9af RM |
821 | if (fpvalid) |
822 | fpvalid = !fpregs_get(tsk, NULL, | |
823 | 0, sizeof(struct user_i387_ia32_struct), | |
824 | fpu, NULL); | |
1da177e4 LT |
825 | |
826 | return fpvalid; | |
827 | } | |
129f6946 | 828 | EXPORT_SYMBOL(dump_fpu); |
1da177e4 | 829 | |
60b3b9af | 830 | #endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */ |