2 * Copyright (C) 1994 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * General FPU state handling cleanups
6 * Gareth Hughes <gareth@valinux.com>, May 2000
8 #include <linux/module.h>
9 #include <linux/regset.h>
10 #include <linux/sched.h>
12 #include <asm/sigcontext.h>
13 #include <asm/processor.h>
14 #include <asm/math_emu.h>
15 #include <asm/uaccess.h>
16 #include <asm/ptrace.h>
21 # include <asm/sigcontext32.h>
22 # include <asm/user32.h>
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
31 #ifdef CONFIG_MATH_EMULATION
32 # define HAVE_HWFP (boot_cpu_data.hard_math)
37 static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
38 unsigned int xstate_size;
39 static struct i387_fxsave_struct fx_scratch __cpuinitdata;
41 void __cpuinit mxcsr_feature_mask_init(void)
43 unsigned long mask = 0;
47 memset(&fx_scratch, 0, sizeof(struct i387_fxsave_struct));
48 asm volatile("fxsave %0" : : "m" (fx_scratch));
49 mask = fx_scratch.mxcsr_mask;
53 mxcsr_feature_mask &= mask;
57 void __init init_thread_xstate(void)
60 xstate_size = sizeof(struct i387_soft_struct);
70 xstate_size = sizeof(struct i387_fxsave_struct);
73 xstate_size = sizeof(struct i387_fsave_struct);
79 * Called at bootup to set up the initial FPU state that is later cloned
82 void __cpuinit fpu_init(void)
84 unsigned long oldcr0 = read_cr0();
86 set_in_cr4(X86_CR4_OSFXSR);
87 set_in_cr4(X86_CR4_OSXMMEXCPT);
89 write_cr0(oldcr0 & ~(X86_CR0_TS|X86_CR0_EM)); /* clear TS and EM */
92 * Boot processor to setup the FP and extended state context info.
94 if (!smp_processor_id())
98 mxcsr_feature_mask_init();
99 /* clean state in init */
100 current_thread_info()->status = 0;
103 #endif /* CONFIG_X86_64 */
106 * The _current_ task is using the FPU for the first time
107 * so initialize it and set the mxcsr to its default
108 * value at reset if we support XMM instructions and then
109 * remeber the current task has used the FPU.
111 int init_fpu(struct task_struct *tsk)
113 if (tsk_used_math(tsk)) {
114 if (HAVE_HWFP && tsk == current)
120 * Memory allocation at the first usage of the FPU and other state.
122 if (!tsk->thread.xstate) {
123 tsk->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
125 if (!tsk->thread.xstate)
131 memset(tsk->thread.xstate, 0, xstate_size);
133 set_stopped_child_used_math(tsk);
139 struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave;
141 memset(fx, 0, xstate_size);
144 fx->mxcsr = MXCSR_DEFAULT;
146 struct i387_fsave_struct *fp = &tsk->thread.xstate->fsave;
147 memset(fp, 0, xstate_size);
148 fp->cwd = 0xffff037fu;
149 fp->swd = 0xffff0000u;
150 fp->twd = 0xffffffffu;
151 fp->fos = 0xffff0000u;
154 * Only the device not available exception or ptrace can call init_fpu.
156 set_stopped_child_used_math(tsk);
160 int fpregs_active(struct task_struct *target, const struct user_regset *regset)
162 return tsk_used_math(target) ? regset->n : 0;
165 int xfpregs_active(struct task_struct *target, const struct user_regset *regset)
167 return (cpu_has_fxsr && tsk_used_math(target)) ? regset->n : 0;
170 int xfpregs_get(struct task_struct *target, const struct user_regset *regset,
171 unsigned int pos, unsigned int count,
172 void *kbuf, void __user *ubuf)
179 ret = init_fpu(target);
183 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
184 &target->thread.xstate->fxsave, 0, -1);
187 int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
188 unsigned int pos, unsigned int count,
189 const void *kbuf, const void __user *ubuf)
196 ret = init_fpu(target);
200 set_stopped_child_used_math(target);
202 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
203 &target->thread.xstate->fxsave, 0, -1);
206 * mxcsr reserved bits must be masked to zero for security reasons.
208 target->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask;
213 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
216 * FPU tag word conversions.
219 static inline unsigned short twd_i387_to_fxsr(unsigned short twd)
221 unsigned int tmp; /* to avoid 16 bit prefixes in the code */
223 /* Transform each pair of bits into 01 (valid) or 00 (empty) */
225 tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */
226 /* and move the valid bits to the lower byte. */
227 tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */
228 tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */
229 tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */
234 #define FPREG_ADDR(f, n) ((void *)&(f)->st_space + (n) * 16);
235 #define FP_EXP_TAG_VALID 0
236 #define FP_EXP_TAG_ZERO 1
237 #define FP_EXP_TAG_SPECIAL 2
238 #define FP_EXP_TAG_EMPTY 3
240 static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave)
243 u32 tos = (fxsave->swd >> 11) & 7;
244 u32 twd = (unsigned long) fxsave->twd;
246 u32 ret = 0xffff0000u;
249 for (i = 0; i < 8; i++, twd >>= 1) {
251 st = FPREG_ADDR(fxsave, (i - tos) & 7);
253 switch (st->exponent & 0x7fff) {
255 tag = FP_EXP_TAG_SPECIAL;
258 if (!st->significand[0] &&
259 !st->significand[1] &&
260 !st->significand[2] &&
262 tag = FP_EXP_TAG_ZERO;
264 tag = FP_EXP_TAG_SPECIAL;
267 if (st->significand[3] & 0x8000)
268 tag = FP_EXP_TAG_VALID;
270 tag = FP_EXP_TAG_SPECIAL;
274 tag = FP_EXP_TAG_EMPTY;
276 ret |= tag << (2 * i);
282 * FXSR floating point environment conversions.
286 convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk)
288 struct i387_fxsave_struct *fxsave = &tsk->thread.xstate->fxsave;
289 struct _fpreg *to = (struct _fpreg *) &env->st_space[0];
290 struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0];
293 env->cwd = fxsave->cwd | 0xffff0000u;
294 env->swd = fxsave->swd | 0xffff0000u;
295 env->twd = twd_fxsr_to_i387(fxsave);
298 env->fip = fxsave->rip;
299 env->foo = fxsave->rdp;
300 if (tsk == current) {
302 * should be actually ds/cs at fpu exception time, but
303 * that information is not available in 64bit mode.
305 asm("mov %%ds, %[fos]" : [fos] "=r" (env->fos));
306 asm("mov %%cs, %[fcs]" : [fcs] "=r" (env->fcs));
308 struct pt_regs *regs = task_pt_regs(tsk);
310 env->fos = 0xffff0000 | tsk->thread.ds;
314 env->fip = fxsave->fip;
315 env->fcs = (u16) fxsave->fcs | ((u32) fxsave->fop << 16);
316 env->foo = fxsave->foo;
317 env->fos = fxsave->fos;
320 for (i = 0; i < 8; ++i)
321 memcpy(&to[i], &from[i], sizeof(to[0]));
324 static void convert_to_fxsr(struct task_struct *tsk,
325 const struct user_i387_ia32_struct *env)
328 struct i387_fxsave_struct *fxsave = &tsk->thread.xstate->fxsave;
329 struct _fpreg *from = (struct _fpreg *) &env->st_space[0];
330 struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0];
333 fxsave->cwd = env->cwd;
334 fxsave->swd = env->swd;
335 fxsave->twd = twd_i387_to_fxsr(env->twd);
336 fxsave->fop = (u16) ((u32) env->fcs >> 16);
338 fxsave->rip = env->fip;
339 fxsave->rdp = env->foo;
340 /* cs and ds ignored */
342 fxsave->fip = env->fip;
343 fxsave->fcs = (env->fcs & 0xffff);
344 fxsave->foo = env->foo;
345 fxsave->fos = env->fos;
348 for (i = 0; i < 8; ++i)
349 memcpy(&to[i], &from[i], sizeof(from[0]));
352 int fpregs_get(struct task_struct *target, const struct user_regset *regset,
353 unsigned int pos, unsigned int count,
354 void *kbuf, void __user *ubuf)
356 struct user_i387_ia32_struct env;
359 ret = init_fpu(target);
364 return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf);
367 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
368 &target->thread.xstate->fsave, 0,
372 if (kbuf && pos == 0 && count == sizeof(env)) {
373 convert_from_fxsr(kbuf, target);
377 convert_from_fxsr(&env, target);
379 return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
382 int fpregs_set(struct task_struct *target, const struct user_regset *regset,
383 unsigned int pos, unsigned int count,
384 const void *kbuf, const void __user *ubuf)
386 struct user_i387_ia32_struct env;
389 ret = init_fpu(target);
393 set_stopped_child_used_math(target);
396 return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf);
399 return user_regset_copyin(&pos, &count, &kbuf, &ubuf,
400 &target->thread.xstate->fsave, 0, -1);
403 if (pos > 0 || count < sizeof(env))
404 convert_from_fxsr(&env, target);
406 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
408 convert_to_fxsr(target, &env);
414 * Signal frame handlers.
417 static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf)
419 struct task_struct *tsk = current;
420 struct i387_fsave_struct *fp = &tsk->thread.xstate->fsave;
423 fp->status = fp->swd;
424 if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct)))
429 static int save_i387_fxsave(struct _fpstate_ia32 __user *buf)
431 struct task_struct *tsk = current;
432 struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave;
433 struct user_i387_ia32_struct env;
438 convert_from_fxsr(&env, tsk);
439 if (__copy_to_user(buf, &env, sizeof(env)))
442 err |= __put_user(fx->swd, &buf->status);
443 err |= __put_user(X86_FXSR_MAGIC, &buf->magic);
447 if (__copy_to_user(&buf->_fxsr_env[0], fx,
448 sizeof(struct i387_fxsave_struct)))
453 int save_i387_ia32(struct _fpstate_ia32 __user *buf)
458 * This will cause a "finit" to be triggered by the next
459 * attempted FPU operation by the 'current' process.
464 return fpregs_soft_get(current, NULL,
465 0, sizeof(struct user_i387_ia32_struct),
470 return save_i387_fxsave(buf);
472 return save_i387_fsave(buf);
475 static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf)
477 struct task_struct *tsk = current;
479 return __copy_from_user(&tsk->thread.xstate->fsave, buf,
480 sizeof(struct i387_fsave_struct));
483 static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf)
485 struct task_struct *tsk = current;
486 struct user_i387_ia32_struct env;
489 err = __copy_from_user(&tsk->thread.xstate->fxsave, &buf->_fxsr_env[0],
490 sizeof(struct i387_fxsave_struct));
491 /* mxcsr reserved bits must be masked to zero for security reasons */
492 tsk->thread.xstate->fxsave.mxcsr &= mxcsr_feature_mask;
493 if (err || __copy_from_user(&env, buf, sizeof(env)))
495 convert_to_fxsr(tsk, &env);
500 int restore_i387_ia32(struct _fpstate_ia32 __user *buf)
503 struct task_struct *tsk = current;
516 err = restore_i387_fxsave(buf);
518 err = restore_i387_fsave(buf);
520 err = fpregs_soft_set(current, NULL,
521 0, sizeof(struct user_i387_ia32_struct),
530 * FPU state for core dumps.
531 * This is only used for a.out dumps now.
532 * It is declared generically using elf_fpregset_t (which is
533 * struct user_i387_struct) but is in fact only used for 32-bit
534 * dumps, so on 64-bit it is really struct user_i387_ia32_struct.
536 int dump_fpu(struct pt_regs *regs, struct user_i387_struct *fpu)
538 struct task_struct *tsk = current;
541 fpvalid = !!used_math();
543 fpvalid = !fpregs_get(tsk, NULL,
544 0, sizeof(struct user_i387_ia32_struct),
549 EXPORT_SYMBOL(dump_fpu);
551 #endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */