[deliverable/linux.git] / include / asm-x86_64 / i387.h

/*
 * include/asm-x86_64/i387.h
 *
 * Copyright (C) 1994 Linus Torvalds
 *
 * Pentium III FXSR, SSE support
 * General FPU state handling cleanups
 *	Gareth Hughes <gareth@valinux.com>, May 2000
 * x86-64 work by Andi Kleen 2002
 */

#ifndef __ASM_X86_64_I387_H
#define __ASM_X86_64_I387_H

#include <linux/sched.h>
#include <asm/processor.h>
#include <asm/sigcontext.h>
#include <asm/user.h>
#include <asm/thread_info.h>
#include <asm/uaccess.h>

extern void fpu_init(void);
extern unsigned int mxcsr_feature_mask;
extern void mxcsr_feature_mask_init(void);
extern void init_fpu(struct task_struct *child);
extern int save_i387(struct _fpstate __user *buf);

/*
 * FPU lazy state save handling...
 */

#define unlazy_fpu(tsk) do { \
	if ((tsk)->thread_info->status & TS_USEDFPU) \
		save_init_fpu(tsk); \
} while (0)

/* Ignore delayed exceptions from user space */
static inline void tolerant_fwait(void)
{
	asm volatile("1: fwait\n"
		     "2:\n"
		     "   .section __ex_table,\"a\"\n"
		     "	.align 8\n"
		     "	.quad 1b,2b\n"
		     "	.previous\n");
}

#define clear_fpu(tsk) do { \
	if ((tsk)->thread_info->status & TS_USEDFPU) {		\
		tolerant_fwait();				\
		(tsk)->thread_info->status &= ~TS_USEDFPU;	\
		stts();						\
	}							\
} while (0)

/*
 * ptrace request handers...
 */
extern int get_fpregs(struct user_i387_struct __user *buf,
		      struct task_struct *tsk);
extern int set_fpregs(struct task_struct *tsk,
		      struct user_i387_struct __user *buf);

/*
 * i387 state interaction
 */
#define get_fpu_mxcsr(t) ((t)->thread.i387.fxsave.mxcsr)
#define get_fpu_cwd(t) ((t)->thread.i387.fxsave.cwd)
#define get_fpu_fxsr_twd(t) ((t)->thread.i387.fxsave.twd)
#define get_fpu_swd(t) ((t)->thread.i387.fxsave.swd)
#define set_fpu_cwd(t,val) ((t)->thread.i387.fxsave.cwd = (val))
#define set_fpu_swd(t,val) ((t)->thread.i387.fxsave.swd = (val))
#define set_fpu_fxsr_twd(t,val) ((t)->thread.i387.fxsave.twd = (val))

static inline int restore_fpu_checking(struct i387_fxsave_struct *fx) 
{ 
	int err;

	asm volatile("1:  rex64/fxrstor (%[fx])\n\t"
		     "2:\n"
		     ".section .fixup,\"ax\"\n"
		     "3:  movl $-1,%[err]\n"
		     "    jmp  2b\n"
		     ".previous\n"
		     ".section __ex_table,\"a\"\n"
		     "   .align 8\n"
		     "   .quad  1b,3b\n"
		     ".previous"
		     : [err] "=r" (err)
#if 0 /* See comment in __fxsave_clear() below. */
		     : [fx] "r" (fx), "m" (*fx), "0" (0));
#else
		     : [fx] "cdaSDb" (fx), "m" (*fx), "0" (0));
#endif
	if (unlikely(err))
		init_fpu(current);
	return err;
} 

static inline int save_i387_checking(struct i387_fxsave_struct __user *fx) 
{ 
	int err;

	asm volatile("1:  rex64/fxsave (%[fx])\n\t"
		     "2:\n"
		     ".section .fixup,\"ax\"\n"
		     "3:  movl $-1,%[err]\n"
		     "    jmp  2b\n"
		     ".previous\n"
		     ".section __ex_table,\"a\"\n"
		     "   .align 8\n"
		     "   .quad  1b,3b\n"
		     ".previous"
		     : [err] "=r" (err), "=m" (*fx)
#if 0 /* See comment in __fxsave_clear() below. */
		     : [fx] "r" (fx), "0" (0));
#else
		     : [fx] "cdaSDb" (fx), "0" (0));
#endif
	if (unlikely(err))
		__clear_user(fx, sizeof(struct i387_fxsave_struct));
	return err;
} 

static inline void __fxsave_clear(struct task_struct *tsk)
{
	/* Using "rex64; fxsave %0" is broken because, if the memory operand
	   uses any extended registers for addressing, a second REX prefix
	   will be generated (to the assembler, rex64 followed by semicolon
	   is a separate instruction), and hence the 64-bitness is lost. */
#if 0
	/* Using "fxsaveq %0" would be the ideal choice, but is only supported
	   starting with gas 2.16. */
	__asm__ __volatile__("fxsaveq %0"
			     : "=m" (tsk->thread.i387.fxsave));
#elif 0
	/* Using, as a workaround, the properly prefixed form below isn't
	   accepted by any binutils version so far released, complaining that
	   the same type of prefix is used twice if an extended register is
	   needed for addressing (fix submitted to mainline 2005-11-21). */
	__asm__ __volatile__("rex64/fxsave %0"
			     : "=m" (tsk->thread.i387.fxsave));
#else
	/* This, however, we can work around by forcing the compiler to select
	   an addressing mode that doesn't require extended registers. */
	__asm__ __volatile__("rex64/fxsave %P2(%1)"
			     : "=m" (tsk->thread.i387.fxsave)
			     : "cdaSDb" (tsk),
				"i" (offsetof(__typeof__(*tsk),
					      thread.i387.fxsave)));
#endif
	__asm__ __volatile__("fnclex");
}

static inline void kernel_fpu_begin(void)
{
	struct thread_info *me = current_thread_info();
	preempt_disable();
	if (me->status & TS_USEDFPU) {
		__fxsave_clear(me->task);
		me->status &= ~TS_USEDFPU;
		return;
	}
	clts();
}

static inline void kernel_fpu_end(void)
{
	stts();
	preempt_enable();
}

static inline void save_init_fpu( struct task_struct *tsk )
{
 	__fxsave_clear(tsk);
	tsk->thread_info->status &= ~TS_USEDFPU;
	stts();
}

/* 
 * This restores directly out of user space. Exceptions are handled.
 */
static inline int restore_i387(struct _fpstate __user *buf)
{
	return restore_fpu_checking((__force struct i387_fxsave_struct *)buf);
}

#endif /* __ASM_X86_64_I387_H */
Commit	Line	Data
1da177e4 LT	1	/*
	2	* include/asm-x86_64/i387.h
	3	*
	4	* Copyright (C) 1994 Linus Torvalds
	5	*
	6	* Pentium III FXSR, SSE support
	7	* General FPU state handling cleanups
	8	* Gareth Hughes <gareth@valinux.com>, May 2000
	9	* x86-64 work by Andi Kleen 2002
	10	*/
	11
	12	#ifndef __ASM_X86_64_I387_H
	13	#define __ASM_X86_64_I387_H
	14
	15	#include <linux/sched.h>
	16	#include <asm/processor.h>
	17	#include <asm/sigcontext.h>
	18	#include <asm/user.h>
	19	#include <asm/thread_info.h>
	20	#include <asm/uaccess.h>
	21
	22	extern void fpu_init(void);
	23	extern unsigned int mxcsr_feature_mask;
	24	extern void mxcsr_feature_mask_init(void);
	25	extern void init_fpu(struct task_struct *child);
	26	extern int save_i387(struct _fpstate __user *buf);
	27
	28	/*
	29	* FPU lazy state save handling...
	30	*/
	31
	32	#define unlazy_fpu(tsk) do { \
	33	if ((tsk)->thread_info->status & TS_USEDFPU) \
	34	save_init_fpu(tsk); \
	35	} while (0)
	36
	37	/* Ignore delayed exceptions from user space */
	38	static inline void tolerant_fwait(void)
	39	{
	40	asm volatile("1: fwait\n"
	41	"2:\n"
	42	" .section __ex_table,\"a\"\n"
	43	" .align 8\n"
	44	" .quad 1b,2b\n"
	45	" .previous\n");
	46	}
	47
	48	#define clear_fpu(tsk) do { \
	49	if ((tsk)->thread_info->status & TS_USEDFPU) { \
	50	tolerant_fwait(); \
	51	(tsk)->thread_info->status &= ~TS_USEDFPU; \
	52	stts(); \
	53	} \
	54	} while (0)
	55
	56	/*
	57	* ptrace request handers...
	58	*/
	59	extern int get_fpregs(struct user_i387_struct __user *buf,
	60	struct task_struct *tsk);
	61	extern int set_fpregs(struct task_struct *tsk,
	62	struct user_i387_struct __user *buf);
	63
	64	/*
65	* i387 state interaction
66	*/
67	#define get_fpu_mxcsr(t) ((t)->thread.i387.fxsave.mxcsr)
68	#define get_fpu_cwd(t) ((t)->thread.i387.fxsave.cwd)
69	#define get_fpu_fxsr_twd(t) ((t)->thread.i387.fxsave.twd)
70	#define get_fpu_swd(t) ((t)->thread.i387.fxsave.swd)
71	#define set_fpu_cwd(t,val) ((t)->thread.i387.fxsave.cwd = (val))
72	#define set_fpu_swd(t,val) ((t)->thread.i387.fxsave.swd = (val))
73	#define set_fpu_fxsr_twd(t,val) ((t)->thread.i387.fxsave.twd = (val))
74
75	static inline int restore_fpu_checking(struct i387_fxsave_struct *fx)
76	{
77	int err;
7180d4fb JB	78
7180d4fb JB	79	asm volatile("1: rex64/fxrstor (%[fx])\n\t"
1da177e4 LT	80	"2:\n"
	81	".section .fixup,\"ax\"\n"
	82	"3: movl $-1,%[err]\n"
	83	" jmp 2b\n"
	84	".previous\n"
	85	".section __ex_table,\"a\"\n"
	86	" .align 8\n"
	87	" .quad 1b,3b\n"
	88	".previous"
	89	: [err] "=r" (err)
7180d4fb JB	90	#if 0 /* See comment in __fxsave_clear() below. */
	91	: [fx] "r" (fx), "m" (*fx), "0" (0));
	92	#else
	93	: [fx] "cdaSDb" (fx), "m" (*fx), "0" (0));
	94	#endif
1da177e4 LT	95	if (unlikely(err))
	96	init_fpu(current);
	97	return err;
	98	}
	99
	100	static inline int save_i387_checking(struct i387_fxsave_struct __user *fx)
	101	{
	102	int err;
7180d4fb JB	103
7180d4fb JB	104	asm volatile("1: rex64/fxsave (%[fx])\n\t"
1da177e4 LT	105	"2:\n"
	106	".section .fixup,\"ax\"\n"
	107	"3: movl $-1,%[err]\n"
	108	" jmp 2b\n"
	109	".previous\n"
	110	".section __ex_table,\"a\"\n"
	111	" .align 8\n"
	112	" .quad 1b,3b\n"
	113	".previous"
7180d4fb JB	114	: [err] "=r" (err), "=m" (*fx)
	115	#if 0 /* See comment in __fxsave_clear() below. */
	116	: [fx] "r" (fx), "0" (0));
	117	#else
	118	: [fx] "cdaSDb" (fx), "0" (0));
	119	#endif
1da177e4 LT	120	if (unlikely(err))
	121	__clear_user(fx, sizeof(struct i387_fxsave_struct));
	122	return err;
	123	}
	124
7180d4fb JB	125	static inline void __fxsave_clear(struct task_struct *tsk)
	126	{
	127	/* Using "rex64; fxsave %0" is broken because, if the memory operand
	128	uses any extended registers for addressing, a second REX prefix
	129	will be generated (to the assembler, rex64 followed by semicolon
	130	is a separate instruction), and hence the 64-bitness is lost. */
	131	#if 0
	132	/* Using "fxsaveq %0" would be the ideal choice, but is only supported
	133	starting with gas 2.16. */
	134	__asm__ __volatile__("fxsaveq %0"
	135	: "=m" (tsk->thread.i387.fxsave));
	136	#elif 0
	137	/* Using, as a workaround, the properly prefixed form below isn't
	138	accepted by any binutils version so far released, complaining that
	139	the same type of prefix is used twice if an extended register is
	140	needed for addressing (fix submitted to mainline 2005-11-21). */
	141	__asm__ __volatile__("rex64/fxsave %0"
	142	: "=m" (tsk->thread.i387.fxsave));
	143	#else
	144	/* This, however, we can work around by forcing the compiler to select
	145	an addressing mode that doesn't require extended registers. */
	146	__asm__ __volatile__("rex64/fxsave %P2(%1)"
	147	: "=m" (tsk->thread.i387.fxsave)
	148	: "cdaSDb" (tsk),
	149	"i" (offsetof(__typeof__(*tsk),
	150	thread.i387.fxsave)));
	151	#endif
	152	__asm__ __volatile__("fnclex");
	153	}
	154
1da177e4 LT	155	static inline void kernel_fpu_begin(void)
	156	{
	157	struct thread_info *me = current_thread_info();
	158	preempt_disable();
7180d4fb JB	159	if (me->status & TS_USEDFPU) {
7180d4fb JB	160	__fxsave_clear(me->task);
1da177e4 LT	161	me->status &= ~TS_USEDFPU;
	162	return;
	163	}
	164	clts();
	165	}
	166
	167	static inline void kernel_fpu_end(void)
	168	{
	169	stts();
	170	preempt_enable();
	171	}
	172
	173	static inline void save_init_fpu( struct task_struct *tsk )
	174	{
7180d4fb	175	__fxsave_clear(tsk);
1da177e4 LT	176	tsk->thread_info->status &= ~TS_USEDFPU;
	177	stts();
	178	}
	179
	180	/*
	181	* This restores directly out of user space. Exceptions are handled.
	182	*/
	183	static inline int restore_i387(struct _fpstate __user *buf)
	184	{
	185	return restore_fpu_checking((__force struct i387_fxsave_struct *)buf);
	186	}
	187
	188	#endif /* __ASM_X86_64_I387_H */