x86, xsave: context switch support using xsave/xrstor

Uses xsave/xrstor (instead of traditional fxsave/fxrstor) in context switch
when available.

Introduces TS_XSAVE flag, which determine the need to use xsave/xrstor
instructions during context switch instead of the legacy fxsave/fxrstor
instructions. Thread-synchronous status word is already in L1 cache during
this code patch and thus minimizes the performance penality compared to
(cpu_has_xsave) checks.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

arch/x86/kernel/cpu/common.c

@@ -709,7 +709,10 @@ void __cpuinit cpu_init(void)
 	/*
 	 * Force FPU initialization:
 	 */
-	current_thread_info()->status = 0;
+	if (cpu_has_xsave)
+		current_thread_info()->status = TS_XSAVE;
+	else
+		current_thread_info()->status = 0;
 	clear_used_math();
 	mxcsr_feature_mask_init();
 

arch/x86/kernel/i387.c

@@ -97,7 +97,10 @@ void __cpuinit fpu_init(void)
 
 	mxcsr_feature_mask_init();
 	/* clean state in init */
-	current_thread_info()->status = 0;
+	if (cpu_has_xsave)
+		current_thread_info()->status = TS_XSAVE;
+	else
+		current_thread_info()->status = 0;
 	clear_used_math();
 }
 #endif	/* CONFIG_X86_64 */

arch/x86/kernel/traps_64.c

@@ -1134,7 +1134,7 @@ asmlinkage void math_state_restore(void)
 	/*
 	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
 	 */
-	if (unlikely(restore_fpu_checking(&me->thread.xstate->fxsave))) {
+	if (unlikely(restore_fpu_checking(me))) {
 		stts();
 		force_sig(SIGSEGV, me);
 		return;

include/asm-x86/i387.h

@@ -36,6 +36,8 @@ extern int save_i387_ia32(struct _fpstate_ia32 __user *buf);
 extern int restore_i387_ia32(struct _fpstate_ia32 __user *buf);
 #endif
 
+#define X87_FSW_ES (1 << 7)	/* Exception Summary */
+
 #ifdef CONFIG_X86_64
 
 /* Ignore delayed exceptions from user space */
@@ -46,7 +48,7 @@ static inline void tolerant_fwait(void)
 		     _ASM_EXTABLE(1b, 2b));
 }
 
-static inline int restore_fpu_checking(struct i387_fxsave_struct *fx)
+static inline int fxrstor_checking(struct i387_fxsave_struct *fx)
 {
 	int err;
 
@@ -66,15 +68,31 @@ static inline int restore_fpu_checking(struct i387_fxsave_struct *fx)
 	return err;
 }
 
-#define X87_FSW_ES (1 << 7)	/* Exception Summary */
+static inline int restore_fpu_checking(struct task_struct *tsk)
+{
+	if (task_thread_info(tsk)->status & TS_XSAVE)
+		return xrstor_checking(&tsk->thread.xstate->xsave);
+	else
+		return fxrstor_checking(&tsk->thread.xstate->fxsave);
+}
 
 /* AMD CPUs don't save/restore FDP/FIP/FOP unless an exception
    is pending. Clear the x87 state here by setting it to fixed
    values. The kernel data segment can be sometimes 0 and sometimes
    new user value. Both should be ok.
    Use the PDA as safe address because it should be already in L1. */
-static inline void clear_fpu_state(struct i387_fxsave_struct *fx)
+static inline void clear_fpu_state(struct task_struct *tsk)
 {
+	struct xsave_struct *xstate = &tsk->thread.xstate->xsave;
+	struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave;
+
+	/*
+	 * xsave header may indicate the init state of the FP.
+	 */
+	if ((task_thread_info(tsk)->status & TS_XSAVE) &&
+	    !(xstate->xsave_hdr.xstate_bv & XSTATE_FP))
+		return;
+
 	if (unlikely(fx->swd & X87_FSW_ES))
 		asm volatile("fnclex");
 	alternative_input(ASM_NOP8 ASM_NOP2,
@@ -107,7 +125,7 @@ static inline int save_i387_checking(struct i387_fxsave_struct __user *fx)
 	return err;
 }
 
-static inline void __save_init_fpu(struct task_struct *tsk)
+static inline void fxsave(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
@@ -132,7 +150,16 @@ static inline void __save_init_fpu(struct task_struct *tsk)
 			     : "=m" (tsk->thread.xstate->fxsave)
 			     : "cdaSDb" (&tsk->thread.xstate->fxsave));
 #endif
-	clear_fpu_state(&tsk->thread.xstate->fxsave);
+}
+
+static inline void __save_init_fpu(struct task_struct *tsk)
+{
+	if (task_thread_info(tsk)->status & TS_XSAVE)
+		xsave(tsk);
+	else
+		fxsave(tsk);
+
+	clear_fpu_state(tsk);
 	task_thread_info(tsk)->status &= ~TS_USEDFPU;
 }
 
@@ -147,6 +174,10 @@ static inline void tolerant_fwait(void)
 
 static inline void restore_fpu(struct task_struct *tsk)
 {
+	if (task_thread_info(tsk)->status & TS_XSAVE) {
+		xrstor_checking(&tsk->thread.xstate->xsave);
+		return;
+	}
 	/*
 	 * The "nop" is needed to make the instructions the same
 	 * length.
@@ -172,6 +203,27 @@ static inline void restore_fpu(struct task_struct *tsk)
  */
 static inline void __save_init_fpu(struct task_struct *tsk)
 {
+	if (task_thread_info(tsk)->status & TS_XSAVE) {
+		struct xsave_struct *xstate = &tsk->thread.xstate->xsave;
+		struct i387_fxsave_struct *fx = &tsk->thread.xstate->fxsave;
+
+		xsave(tsk);
+
+		/*
+		 * xsave header may indicate the init state of the FP.
+		 */
+		if (!(xstate->xsave_hdr.xstate_bv & XSTATE_FP))
+			goto end;
+
+		if (unlikely(fx->swd & X87_FSW_ES))
+			asm volatile("fnclex");
+
+		/*
+		 * we can do a simple return here or be paranoid :)
+		 */
+		goto clear_state;
+	}
+
 	/* Use more nops than strictly needed in case the compiler
 	   varies code */
 	alternative_input(
@@ -181,6 +233,7 @@ static inline void __save_init_fpu(struct task_struct *tsk)
 		X86_FEATURE_FXSR,
 		[fx] "m" (tsk->thread.xstate->fxsave),
 		[fsw] "m" (tsk->thread.xstate->fxsave.swd) : "memory");
+clear_state:
 	/* AMD K7/K8 CPUs don't save/restore FDP/FIP/FOP unless an exception
 	   is pending.  Clear the x87 state here by setting it to fixed
 	   values. safe_address is a random variable that should be in L1 */
@@ -190,6 +243,7 @@ static inline void __save_init_fpu(struct task_struct *tsk)
 		"fildl %[addr]", 	/* set F?P to defined value */
 		X86_FEATURE_FXSAVE_LEAK,
 		[addr] "m" (safe_address));
+end:
 	task_thread_info(tsk)->status &= ~TS_USEDFPU;
 }
 

include/asm-x86/processor.h

@@ -362,6 +362,7 @@ union thread_xstate {
 	struct i387_fsave_struct	fsave;
 	struct i387_fxsave_struct	fxsave;
 	struct i387_soft_struct		soft;
+	struct xsave_struct		xsave;
 };
 
 #ifdef CONFIG_X86_64

include/asm-x86/thread_info.h

@@ -239,6 +239,7 @@ static inline struct thread_info *stack_thread_info(void)
 #define TS_POLLING		0x0004	/* true if in idle loop
 					   and not sleeping */
 #define TS_RESTORE_SIGMASK	0x0008	/* restore signal mask in do_signal() */
+#define TS_XSAVE		0x0010	/* Use xsave/xrstor */
 
 #define tsk_is_polling(t) (task_thread_info(t)->status & TS_POLLING)
 

include/asm-x86/xsave.h

@@ -17,10 +17,43 @@
 #define XCNTXT_LMASK	(XSTATE_FP | XSTATE_SSE)
 #define XCNTXT_HMASK	0x0
 
+#ifdef CONFIG_X86_64
+#define REX_PREFIX	"0x48, "
+#else
+#define REX_PREFIX
+#endif
+
 extern unsigned int xstate_size, pcntxt_hmask, pcntxt_lmask;
 extern struct xsave_struct *init_xstate_buf;
 
 extern void xsave_cntxt_init(void);
 extern void xsave_init(void);
+extern int init_fpu(struct task_struct *child);
 
+static inline int xrstor_checking(struct xsave_struct *fx)
+{
+	int err;
+
+	asm volatile("1: .byte " REX_PREFIX "0x0f,0xae,0x2f\n\t"
+		     "2:\n"
+		     ".section .fixup,\"ax\"\n"
+		     "3:  movl $-1,%[err]\n"
+		     "    jmp  2b\n"
+		     ".previous\n"
+		     _ASM_EXTABLE(1b, 3b)
+		     : [err] "=r" (err)
+		     : "D" (fx), "m" (*fx), "a" (-1), "d" (-1), "0" (0)
+		     : "memory");
+
+	return err;
+}
+
+static inline void xsave(struct task_struct *tsk)
+{
+	/* This, however, we can work around by forcing the compiler to select
+	   an addressing mode that doesn't require extended registers. */
+	__asm__ __volatile__(".byte " REX_PREFIX "0x0f,0xae,0x27"
+			     : : "D" (&(tsk->thread.xstate->xsave)),
+				 "a" (-1), "d"(-1) : "memory");
+}
 #endif