A few cleanups to the FPU code on amd64:

- fpudna() always returned 1 since amd64 CPUs always have FPUs.  Change
  the function to return void and adjust the calling code in trap() to
  assume the return 1 case is the only case.
- Remove fpu_cleanstate_ready as it is always true when it is tested.
  Also, only initialize fpu_cleanstate when fpuinit() is called on the BSP.

Reviewed by:	bde

3 files changed, 22 insertions, 33 deletions

diff --git a/sys/amd64/amd64/fpu.c b/sys/amd64/amd64/fpu.c
index 85e5ec9..111d6d8 100644
--- a/sys/amd64/amd64/fpu.c
+++ b/sys/amd64/amd64/fpu.c
@@ -102,10 +102,11 @@ SYSCTL_INT(_hw, HW_FLOATINGPT, floatingpoint, CTLFLAG_RD,
     NULL, 1, "Floating point instructions executed in hardware");
 
 static	struct savefpu		fpu_cleanstate;
-static	bool_t			fpu_cleanstate_ready;
 
 /*
- * Initialize floating point unit.
+ * Initialize the floating point unit.  On the boot CPU we generate a
+ * clean state that is used to initialize the floating point unit when
+ * it is first used by a process.
  */
 void
 fpuinit(void)
@@ -115,22 +116,22 @@ fpuinit(void)
 	u_short control;
 
 	savecrit = intr_disable();
-	PCPU_SET(fpcurthread, 0);
 	stop_emulating();
 	fninit();
 	control = __INITIAL_FPUCW__;
 	fldcw(&control);
 	mxcsr = __INITIAL_MXCSR__;
 	ldmxcsr(mxcsr);
-	fxsave(&fpu_cleanstate);
-	if (fpu_cleanstate.sv_env.en_mxcsr_mask)
-		cpu_mxcsr_mask = fpu_cleanstate.sv_env.en_mxcsr_mask;
-	else
-		cpu_mxcsr_mask = 0xFFBF;
+	if (PCPU_GET(cpuid) == 0) {
+		fxsave(&fpu_cleanstate);
+		if (fpu_cleanstate.sv_env.en_mxcsr_mask)
+			cpu_mxcsr_mask = fpu_cleanstate.sv_env.en_mxcsr_mask;
+		else
+			cpu_mxcsr_mask = 0xFFBF;
+		bzero(fpu_cleanstate.sv_fp, sizeof(fpu_cleanstate.sv_fp));
+		bzero(fpu_cleanstate.sv_xmm, sizeof(fpu_cleanstate.sv_xmm));
+	}
 	start_emulating();
-	bzero(fpu_cleanstate.sv_fp, sizeof(fpu_cleanstate.sv_fp));
-	bzero(fpu_cleanstate.sv_xmm, sizeof(fpu_cleanstate.sv_xmm));
-	fpu_cleanstate_ready = 1;
 	intr_restore(savecrit);
 }
 
@@ -384,8 +385,8 @@ fputrap()
 
 static int err_count = 0;
 
-int
-fpudna()
+void
+fpudna(void)
 {
 	struct pcb *pcb;
 	register_t s;
@@ -395,7 +396,7 @@ fpudna()
 		printf("fpudna: fpcurthread == curthread %d times\n",
 		    ++err_count);
 		stop_emulating();
-		return (1);
+		return;
 	}
 	if (PCPU_GET(fpcurthread) != NULL) {
 		printf("fpudna: fpcurthread = %p (%d), curthread = %p (%d)\n",
@@ -428,8 +429,6 @@ fpudna()
 	} else
 		fxrstor(&pcb->pcb_save);
 	intr_restore(s);
-
-	return (1);
 }
 
 /*
@@ -457,10 +456,7 @@ fpugetregs(struct thread *td, struct savefpu *addr)
 	register_t s;
 
 	if ((td->td_pcb->pcb_flags & PCB_FPUINITDONE) == 0) {
-		if (fpu_cleanstate_ready)
-			bcopy(&fpu_cleanstate, addr, sizeof(fpu_cleanstate));
-		else
-			bzero(addr, sizeof(*addr));
+		bcopy(&fpu_cleanstate, addr, sizeof(fpu_cleanstate));
 		return (_MC_FPOWNED_NONE);
 	}
 	s = intr_disable();
diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c
index 8d710cd..ccd6c39 100644
--- a/sys/amd64/amd64/trap.c
+++ b/sys/amd64/amd64/trap.c
@@ -416,13 +416,8 @@ trap(struct trapframe *frame)
 
 		case T_DNA:
 			/* transparent fault (due to context switch "late") */
-			if (fpudna())
-				goto userout;
-			printf("pid %d killed due to lack of floating point\n",
-				p->p_pid);
-			i = SIGKILL;
-			ucode = 0;
-			break;
+			fpudna();
+			goto userout;
 
 		case T_FPOPFLT:		/* FPU operand fetch fault */
 			ucode = ILL_COPROC;
@@ -450,11 +445,9 @@ trap(struct trapframe *frame)
 			 * XXX this should be fatal unless the kernel has
 			 * registered such use.
 			 */
-			if (fpudna()) {
-				printf("fpudna in kernel mode!\n");
-				goto out;
-			}
-			break;
+			fpudna();
+			printf("fpudna in kernel mode!\n");
+			goto out;
 
 		case T_STKFLT:		/* stack fault */
 			break;
diff --git a/sys/amd64/include/fpu.h b/sys/amd64/include/fpu.h
index 9b99fd8..272f94a 100644
--- a/sys/amd64/include/fpu.h
+++ b/sys/amd64/include/fpu.h
@@ -97,7 +97,7 @@ struct  savefpu {
 #define	__INITIAL_MXCSR_MASK__	0xFFBF
 
 #ifdef _KERNEL
-int	fpudna(void);
+void	fpudna(void);
 void	fpudrop(void);
 void	fpuexit(struct thread *td);
 int	fpuformat(void);