]>
Commit | Line | Data |
---|---|---|
1eeaed76 RM |
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 | * x86-64 work by Andi Kleen 2002 | |
8 | */ | |
9 | ||
77ef50a5 VN |
10 | #ifndef ASM_X86__I387_H |
11 | #define ASM_X86__I387_H | |
1eeaed76 RM |
12 | |
13 | #include <linux/sched.h> | |
14 | #include <linux/kernel_stat.h> | |
15 | #include <linux/regset.h> | |
92c37fa3 | 16 | #include <asm/asm.h> |
1eeaed76 RM |
17 | #include <asm/processor.h> |
18 | #include <asm/sigcontext.h> | |
19 | #include <asm/user.h> | |
20 | #include <asm/uaccess.h> | |
dc1e35c6 | 21 | #include <asm/xsave.h> |
1eeaed76 RM |
22 | |
23 | extern void fpu_init(void); | |
1eeaed76 | 24 | extern void mxcsr_feature_mask_init(void); |
aa283f49 | 25 | extern int init_fpu(struct task_struct *child); |
1eeaed76 | 26 | extern asmlinkage void math_state_restore(void); |
61c4628b | 27 | extern void init_thread_xstate(void); |
1eeaed76 RM |
28 | |
29 | extern user_regset_active_fn fpregs_active, xfpregs_active; | |
30 | extern user_regset_get_fn fpregs_get, xfpregs_get, fpregs_soft_get; | |
31 | extern user_regset_set_fn fpregs_set, xfpregs_set, fpregs_soft_set; | |
32 | ||
33 | #ifdef CONFIG_IA32_EMULATION | |
34 | struct _fpstate_ia32; | |
35 | extern int save_i387_ia32(struct _fpstate_ia32 __user *buf); | |
36 | extern int restore_i387_ia32(struct _fpstate_ia32 __user *buf); | |
37 | #endif | |
38 | ||
b359e8a4 SS |
39 | #define X87_FSW_ES (1 << 7) /* Exception Summary */ |
40 | ||
1eeaed76 RM |
41 | #ifdef CONFIG_X86_64 |
42 | ||
43 | /* Ignore delayed exceptions from user space */ | |
44 | static inline void tolerant_fwait(void) | |
45 | { | |
46 | asm volatile("1: fwait\n" | |
47 | "2:\n" | |
affe6637 | 48 | _ASM_EXTABLE(1b, 2b)); |
1eeaed76 RM |
49 | } |
50 | ||
b359e8a4 | 51 | static inline int fxrstor_checking(struct i387_fxsave_struct *fx) |
1eeaed76 RM |
52 | { |
53 | int err; | |
54 | ||
55 | asm volatile("1: rex64/fxrstor (%[fx])\n\t" | |
56 | "2:\n" | |
57 | ".section .fixup,\"ax\"\n" | |
58 | "3: movl $-1,%[err]\n" | |
59 | " jmp 2b\n" | |
60 | ".previous\n" | |
affe6637 | 61 | _ASM_EXTABLE(1b, 3b) |
1eeaed76 RM |
62 | : [err] "=r" (err) |
63 | #if 0 /* See comment in __save_init_fpu() below. */ | |
64 | : [fx] "r" (fx), "m" (*fx), "0" (0)); | |
65 | #else | |
66 | : [fx] "cdaSDb" (fx), "m" (*fx), "0" (0)); | |
67 | #endif | |
1eeaed76 RM |
68 | return err; |
69 | } | |
70 | ||
b359e8a4 SS |
71 | static inline int restore_fpu_checking(struct task_struct *tsk) |
72 | { | |
73 | if (task_thread_info(tsk)->status & TS_XSAVE) | |
74 | return xrstor_checking(&tsk->thread.xstate->xsave); | |
75 | else | |
76 | return fxrstor_checking(&tsk->thread.xstate->fxsave); | |
77 | } | |
1eeaed76 RM |
78 | |
79 | /* AMD CPUs don't save/restore FDP/FIP/FOP unless an exception | |
80 | is pending. Clear the x87 state here by setting it to fixed | |
81 | values. The kernel data segment can be sometimes 0 and sometimes | |
82 | new user value. Both should be ok. | |
83 | Use the PDA as safe address because it should be already in L1. */ | |
b359e8a4 | 84 | static inline void clear_fpu_state(struct task_struct *tsk) |
1eeaed76 | 85 | { |
b359e8a4 SS |
86 | struct xsave_struct *xstate = &tsk->thread.xstate->xsave; |
87 | struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave; | |
88 | ||
89 | /* | |
90 | * xsave header may indicate the init state of the FP. | |
91 | */ | |
92 | if ((task_thread_info(tsk)->status & TS_XSAVE) && | |
93 | !(xstate->xsave_hdr.xstate_bv & XSTATE_FP)) | |
94 | return; | |
95 | ||
1eeaed76 | 96 | if (unlikely(fx->swd & X87_FSW_ES)) |
affe6637 | 97 | asm volatile("fnclex"); |
1eeaed76 | 98 | alternative_input(ASM_NOP8 ASM_NOP2, |
affe6637 JP |
99 | " emms\n" /* clear stack tags */ |
100 | " fildl %%gs:0", /* load to clear state */ | |
101 | X86_FEATURE_FXSAVE_LEAK); | |
1eeaed76 RM |
102 | } |
103 | ||
104 | static inline int save_i387_checking(struct i387_fxsave_struct __user *fx) | |
105 | { | |
106 | int err; | |
107 | ||
108 | asm volatile("1: rex64/fxsave (%[fx])\n\t" | |
109 | "2:\n" | |
110 | ".section .fixup,\"ax\"\n" | |
111 | "3: movl $-1,%[err]\n" | |
112 | " jmp 2b\n" | |
113 | ".previous\n" | |
affe6637 | 114 | _ASM_EXTABLE(1b, 3b) |
1eeaed76 RM |
115 | : [err] "=r" (err), "=m" (*fx) |
116 | #if 0 /* See comment in __fxsave_clear() below. */ | |
117 | : [fx] "r" (fx), "0" (0)); | |
118 | #else | |
119 | : [fx] "cdaSDb" (fx), "0" (0)); | |
120 | #endif | |
affe6637 JP |
121 | if (unlikely(err) && |
122 | __clear_user(fx, sizeof(struct i387_fxsave_struct))) | |
1eeaed76 RM |
123 | err = -EFAULT; |
124 | /* No need to clear here because the caller clears USED_MATH */ | |
125 | return err; | |
126 | } | |
127 | ||
b359e8a4 | 128 | static inline void fxsave(struct task_struct *tsk) |
1eeaed76 RM |
129 | { |
130 | /* Using "rex64; fxsave %0" is broken because, if the memory operand | |
131 | uses any extended registers for addressing, a second REX prefix | |
132 | will be generated (to the assembler, rex64 followed by semicolon | |
133 | is a separate instruction), and hence the 64-bitness is lost. */ | |
134 | #if 0 | |
135 | /* Using "fxsaveq %0" would be the ideal choice, but is only supported | |
136 | starting with gas 2.16. */ | |
137 | __asm__ __volatile__("fxsaveq %0" | |
61c4628b | 138 | : "=m" (tsk->thread.xstate->fxsave)); |
1eeaed76 RM |
139 | #elif 0 |
140 | /* Using, as a workaround, the properly prefixed form below isn't | |
141 | accepted by any binutils version so far released, complaining that | |
142 | the same type of prefix is used twice if an extended register is | |
143 | needed for addressing (fix submitted to mainline 2005-11-21). */ | |
144 | __asm__ __volatile__("rex64/fxsave %0" | |
61c4628b | 145 | : "=m" (tsk->thread.xstate->fxsave)); |
1eeaed76 RM |
146 | #else |
147 | /* This, however, we can work around by forcing the compiler to select | |
148 | an addressing mode that doesn't require extended registers. */ | |
61c4628b SS |
149 | __asm__ __volatile__("rex64/fxsave (%1)" |
150 | : "=m" (tsk->thread.xstate->fxsave) | |
151 | : "cdaSDb" (&tsk->thread.xstate->fxsave)); | |
1eeaed76 | 152 | #endif |
b359e8a4 SS |
153 | } |
154 | ||
155 | static inline void __save_init_fpu(struct task_struct *tsk) | |
156 | { | |
157 | if (task_thread_info(tsk)->status & TS_XSAVE) | |
158 | xsave(tsk); | |
159 | else | |
160 | fxsave(tsk); | |
161 | ||
162 | clear_fpu_state(tsk); | |
1eeaed76 RM |
163 | task_thread_info(tsk)->status &= ~TS_USEDFPU; |
164 | } | |
165 | ||
1eeaed76 RM |
166 | #else /* CONFIG_X86_32 */ |
167 | ||
e8a496ac SS |
168 | extern void finit(void); |
169 | ||
1eeaed76 RM |
170 | static inline void tolerant_fwait(void) |
171 | { | |
172 | asm volatile("fnclex ; fwait"); | |
173 | } | |
174 | ||
175 | static inline void restore_fpu(struct task_struct *tsk) | |
176 | { | |
b359e8a4 SS |
177 | if (task_thread_info(tsk)->status & TS_XSAVE) { |
178 | xrstor_checking(&tsk->thread.xstate->xsave); | |
179 | return; | |
180 | } | |
1eeaed76 RM |
181 | /* |
182 | * The "nop" is needed to make the instructions the same | |
183 | * length. | |
184 | */ | |
185 | alternative_input( | |
186 | "nop ; frstor %1", | |
187 | "fxrstor %1", | |
188 | X86_FEATURE_FXSR, | |
61c4628b | 189 | "m" (tsk->thread.xstate->fxsave)); |
1eeaed76 RM |
190 | } |
191 | ||
192 | /* We need a safe address that is cheap to find and that is already | |
193 | in L1 during context switch. The best choices are unfortunately | |
194 | different for UP and SMP */ | |
195 | #ifdef CONFIG_SMP | |
196 | #define safe_address (__per_cpu_offset[0]) | |
197 | #else | |
198 | #define safe_address (kstat_cpu(0).cpustat.user) | |
199 | #endif | |
200 | ||
201 | /* | |
202 | * These must be called with preempt disabled | |
203 | */ | |
204 | static inline void __save_init_fpu(struct task_struct *tsk) | |
205 | { | |
b359e8a4 SS |
206 | if (task_thread_info(tsk)->status & TS_XSAVE) { |
207 | struct xsave_struct *xstate = &tsk->thread.xstate->xsave; | |
208 | struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave; | |
209 | ||
210 | xsave(tsk); | |
211 | ||
212 | /* | |
213 | * xsave header may indicate the init state of the FP. | |
214 | */ | |
215 | if (!(xstate->xsave_hdr.xstate_bv & XSTATE_FP)) | |
216 | goto end; | |
217 | ||
218 | if (unlikely(fx->swd & X87_FSW_ES)) | |
219 | asm volatile("fnclex"); | |
220 | ||
221 | /* | |
222 | * we can do a simple return here or be paranoid :) | |
223 | */ | |
224 | goto clear_state; | |
225 | } | |
226 | ||
1eeaed76 RM |
227 | /* Use more nops than strictly needed in case the compiler |
228 | varies code */ | |
229 | alternative_input( | |
230 | "fnsave %[fx] ;fwait;" GENERIC_NOP8 GENERIC_NOP4, | |
231 | "fxsave %[fx]\n" | |
232 | "bt $7,%[fsw] ; jnc 1f ; fnclex\n1:", | |
233 | X86_FEATURE_FXSR, | |
61c4628b SS |
234 | [fx] "m" (tsk->thread.xstate->fxsave), |
235 | [fsw] "m" (tsk->thread.xstate->fxsave.swd) : "memory"); | |
b359e8a4 | 236 | clear_state: |
1eeaed76 RM |
237 | /* AMD K7/K8 CPUs don't save/restore FDP/FIP/FOP unless an exception |
238 | is pending. Clear the x87 state here by setting it to fixed | |
239 | values. safe_address is a random variable that should be in L1 */ | |
240 | alternative_input( | |
241 | GENERIC_NOP8 GENERIC_NOP2, | |
242 | "emms\n\t" /* clear stack tags */ | |
243 | "fildl %[addr]", /* set F?P to defined value */ | |
244 | X86_FEATURE_FXSAVE_LEAK, | |
245 | [addr] "m" (safe_address)); | |
b359e8a4 | 246 | end: |
1eeaed76 RM |
247 | task_thread_info(tsk)->status &= ~TS_USEDFPU; |
248 | } | |
249 | ||
250 | /* | |
251 | * Signal frame handlers... | |
252 | */ | |
253 | extern int save_i387(struct _fpstate __user *buf); | |
254 | extern int restore_i387(struct _fpstate __user *buf); | |
255 | ||
256 | #endif /* CONFIG_X86_64 */ | |
257 | ||
258 | static inline void __unlazy_fpu(struct task_struct *tsk) | |
259 | { | |
260 | if (task_thread_info(tsk)->status & TS_USEDFPU) { | |
261 | __save_init_fpu(tsk); | |
262 | stts(); | |
263 | } else | |
264 | tsk->fpu_counter = 0; | |
265 | } | |
266 | ||
267 | static inline void __clear_fpu(struct task_struct *tsk) | |
268 | { | |
269 | if (task_thread_info(tsk)->status & TS_USEDFPU) { | |
270 | tolerant_fwait(); | |
271 | task_thread_info(tsk)->status &= ~TS_USEDFPU; | |
272 | stts(); | |
273 | } | |
274 | } | |
275 | ||
276 | static inline void kernel_fpu_begin(void) | |
277 | { | |
278 | struct thread_info *me = current_thread_info(); | |
279 | preempt_disable(); | |
280 | if (me->status & TS_USEDFPU) | |
281 | __save_init_fpu(me->task); | |
282 | else | |
283 | clts(); | |
284 | } | |
285 | ||
286 | static inline void kernel_fpu_end(void) | |
287 | { | |
288 | stts(); | |
289 | preempt_enable(); | |
290 | } | |
291 | ||
292 | #ifdef CONFIG_X86_64 | |
293 | ||
294 | static inline void save_init_fpu(struct task_struct *tsk) | |
295 | { | |
296 | __save_init_fpu(tsk); | |
297 | stts(); | |
298 | } | |
299 | ||
300 | #define unlazy_fpu __unlazy_fpu | |
301 | #define clear_fpu __clear_fpu | |
302 | ||
303 | #else /* CONFIG_X86_32 */ | |
304 | ||
305 | /* | |
306 | * These disable preemption on their own and are safe | |
307 | */ | |
308 | static inline void save_init_fpu(struct task_struct *tsk) | |
309 | { | |
310 | preempt_disable(); | |
311 | __save_init_fpu(tsk); | |
312 | stts(); | |
313 | preempt_enable(); | |
314 | } | |
315 | ||
316 | static inline void unlazy_fpu(struct task_struct *tsk) | |
317 | { | |
318 | preempt_disable(); | |
319 | __unlazy_fpu(tsk); | |
320 | preempt_enable(); | |
321 | } | |
322 | ||
323 | static inline void clear_fpu(struct task_struct *tsk) | |
324 | { | |
325 | preempt_disable(); | |
326 | __clear_fpu(tsk); | |
327 | preempt_enable(); | |
328 | } | |
329 | ||
330 | #endif /* CONFIG_X86_64 */ | |
331 | ||
1eeaed76 RM |
332 | /* |
333 | * i387 state interaction | |
334 | */ | |
335 | static inline unsigned short get_fpu_cwd(struct task_struct *tsk) | |
336 | { | |
337 | if (cpu_has_fxsr) { | |
61c4628b | 338 | return tsk->thread.xstate->fxsave.cwd; |
1eeaed76 | 339 | } else { |
1679f271 | 340 | return (unsigned short)tsk->thread.xstate->fsave.cwd; |
1eeaed76 RM |
341 | } |
342 | } | |
343 | ||
344 | static inline unsigned short get_fpu_swd(struct task_struct *tsk) | |
345 | { | |
346 | if (cpu_has_fxsr) { | |
61c4628b | 347 | return tsk->thread.xstate->fxsave.swd; |
1eeaed76 | 348 | } else { |
1679f271 | 349 | return (unsigned short)tsk->thread.xstate->fsave.swd; |
1eeaed76 RM |
350 | } |
351 | } | |
352 | ||
353 | static inline unsigned short get_fpu_mxcsr(struct task_struct *tsk) | |
354 | { | |
355 | if (cpu_has_xmm) { | |
61c4628b | 356 | return tsk->thread.xstate->fxsave.mxcsr; |
1eeaed76 RM |
357 | } else { |
358 | return MXCSR_DEFAULT; | |
359 | } | |
360 | } | |
361 | ||
77ef50a5 | 362 | #endif /* ASM_X86__I387_H */ |