Split the function of the PCB_FPU flags into two: PCB_FPU now indicates that

the actual FPU is enabled, while PCB_FPREGS indicates that the FPU state
structure in the PCB is valid. This separation reflects the situation on
FPU-less systems in which the FP state is used by the emulator but we don't
actually want to try to turn on the non-existant FPU.

Use this flag to save and restore FP regs properly on both AIM and Book-E.
As a side effect, this sets up hard-FP and Altivec on Book-E CPUs with such
abilities except for a trap handler to call enable_fpu()/enable_altivec().

6 files changed, 25 insertions, 34 deletions

diff --git a/sys/conf/files.powerpc b/sys/conf/files.powerpc
index 763c113..b23015b 100644
--- a/sys/conf/files.powerpc
+++ b/sys/conf/files.powerpc
@@ -165,7 +165,7 @@ powerpc/powermac/uninorth.c	optional	powermac
 powerpc/powermac/uninorthpci.c	optional	powermac pci
 powerpc/powermac/vcoregpio.c	optional	powermac 
 powerpc/powermac/windtunnel.c	optional	powermac windtunnel
-powerpc/powerpc/altivec.c	optional	aim
+powerpc/powerpc/altivec.c	standard
 powerpc/powerpc/autoconf.c	standard
 powerpc/powerpc/bcopy.c		standard
 powerpc/powerpc/bus_machdep.c	standard
@@ -182,7 +182,7 @@ powerpc/powerpc/dump_machdep.c	standard
 powerpc/powerpc/elf32_machdep.c	optional	powerpc | compat_freebsd32
 powerpc/powerpc/elf64_machdep.c	optional	powerpc64
 powerpc/powerpc/exec_machdep.c	standard
-powerpc/powerpc/fpu.c		optional	aim
+powerpc/powerpc/fpu.c		standard
 powerpc/powerpc/fuswintr.c	standard
 powerpc/powerpc/gdb_machdep.c	optional	gdb
 powerpc/powerpc/in_cksum.c	optional	inet | inet6
diff --git a/sys/powerpc/booke/trap.c b/sys/powerpc/booke/trap.c
index 959d8b1..72c4f47 100644
--- a/sys/powerpc/booke/trap.c
+++ b/sys/powerpc/booke/trap.c
@@ -194,6 +194,11 @@ trap(struct trapframe *frame)
 
 		case EXC_PGM:	/* Program exception */
 #ifdef FPU_EMU
+			if (!(td->td_pcb->pcb_flags & PCB_FPREGS)) {
+				bzero(&td->td_pcb->pcb_fpu,
+				    sizeof(td->td_pcb->pcb_fpu));
+				td->td_pcb->pcb_flags |= PCB_FPREGS;
+			}
 			sig = fpu_emulate(frame,
 			    (struct fpreg *)&td->td_pcb->pcb_fpu);
 #else
diff --git a/sys/powerpc/include/pcb.h b/sys/powerpc/include/pcb.h
index 2076f42..a1a8132 100644
--- a/sys/powerpc/include/pcb.h
+++ b/sys/powerpc/include/pcb.h
@@ -47,8 +47,9 @@ struct pcb {
 	faultbuf	*pcb_onfault;		/* For use during
 						    copyin/copyout */
 	int		pcb_flags;
-#define	PCB_FPU		1	/* Process had FPU initialized */
-#define	PCB_VEC		2	/* Process had Altivec initialized */
+#define	PCB_FPU		1	/* Process uses FPU */
+#define	PCB_FPREGS	2	/* Process had FPU registers initialized */
+#define	PCB_VEC		4	/* Process had Altivec initialized */
 	struct fpu {
 		double	fpr[32];
 		double	fpscr;	/* FPSCR stored as double for easier access */
diff --git a/sys/powerpc/powerpc/exec_machdep.c b/sys/powerpc/powerpc/exec_machdep.c
index 038be9b..4b4be6c 100644
--- a/sys/powerpc/powerpc/exec_machdep.c
+++ b/sys/powerpc/powerpc/exec_machdep.c
@@ -381,19 +381,20 @@ grab_mcontext(struct thread *td, mcontext_t *mcp, int flags)
 		mcp->mc_gpr[4] = 0;
 	}
 
-#ifdef AIM
 	/*
 	 * This assumes that floating-point context is *not* lazy,
 	 * so if the thread has used FP there would have been a
 	 * FP-unavailable exception that would have set things up
 	 * correctly.
 	 */
-	if (pcb->pcb_flags & PCB_FPU) {
-		KASSERT(td == curthread,
-			("get_mcontext: fp save not curthread"));
-		critical_enter();
-		save_fpu(td);
-		critical_exit();
+	if (pcb->pcb_flags & PCB_FPREGS) {
+		if (pcb->pcb_flags & PCB_FPU) {
+			KASSERT(td == curthread,
+				("get_mcontext: fp save not curthread"));
+			critical_enter();
+			save_fpu(td);
+			critical_exit();
+		}
 		mcp->mc_flags |= _MC_FP_VALID;
 		memcpy(&mcp->mc_fpscr, &pcb->pcb_fpu.fpscr, sizeof(double));
 		memcpy(mcp->mc_fpreg, pcb->pcb_fpu.fpr, 32*sizeof(double));
@@ -414,7 +415,6 @@ grab_mcontext(struct thread *td, mcontext_t *mcp, int flags)
 		mcp->mc_vrsave =  pcb->pcb_vec.vrsave;
 		memcpy(mcp->mc_avec, pcb->pcb_vec.vr, sizeof(mcp->mc_avec));
 	}
-#endif
 
 	mcp->mc_len = sizeof(*mcp);
 
@@ -467,13 +467,9 @@ set_mcontext(struct thread *td, const mcontext_t *mcp)
 	else
 		tf->fixreg[2] = tls;
 
-#ifdef AIM
 	if (mcp->mc_flags & _MC_FP_VALID) {
-		if ((pcb->pcb_flags & PCB_FPU) != PCB_FPU) {
-			critical_enter();
-			enable_fpu(td);
-			critical_exit();
-		}
+		/* enable_fpu() will happen lazily on a fault */
+		pcb->pcb_flags |= PCB_FPREGS;
 		memcpy(&pcb->pcb_fpu.fpscr, &mcp->mc_fpscr, sizeof(double));
 		memcpy(pcb->pcb_fpu.fpr, mcp->mc_fpreg, 32*sizeof(double));
 	}
@@ -488,7 +484,6 @@ set_mcontext(struct thread *td, const mcontext_t *mcp)
 		pcb->pcb_vec.vrsave = mcp->mc_vrsave;
 		memcpy(pcb->pcb_vec.vr, mcp->mc_avec, sizeof(mcp->mc_avec));
 	}
-#endif
 
 	return (0);
 }
@@ -625,7 +620,7 @@ fill_fpregs(struct thread *td, struct fpreg *fpregs)
 
 	pcb = td->td_pcb;
 
-	if ((pcb->pcb_flags & PCB_FPU) == 0)
+	if ((pcb->pcb_flags & PCB_FPREGS) == 0)
 		memset(fpregs, 0, sizeof(struct fpreg));
 	else
 		memcpy(fpregs, &pcb->pcb_fpu, sizeof(struct fpreg));
@@ -654,14 +649,11 @@ set_dbregs(struct thread *td, struct dbreg *dbregs)
 int
 set_fpregs(struct thread *td, struct fpreg *fpregs)
 {
-#ifdef AIM
 	struct pcb *pcb;
 
 	pcb = td->td_pcb;
-	if ((pcb->pcb_flags & PCB_FPU) == 0)
-		enable_fpu(td);
+	pcb->pcb_flags |= PCB_FPREGS;
 	memcpy(&pcb->pcb_fpu, fpregs, sizeof(struct fpreg));
-#endif
 
 	return (0);
 }
@@ -1021,14 +1013,10 @@ cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
 	tf->fixreg[3] = (register_t)arg;
 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
 		tf->srr0 = (register_t)entry;
-	    #ifdef AIM
 		tf->srr1 = PSL_USERSET | PSL_FE_DFLT;
 		#ifdef __powerpc64__
 		tf->srr1 &= ~PSL_SF;
 		#endif
-	    #else
-		tf->srr1 = PSL_USERSET;
-	    #endif
 	} else {
 	    #ifdef __powerpc64__
 		register_t entry_desc[3];
diff --git a/sys/powerpc/powerpc/fpu.c b/sys/powerpc/powerpc/fpu.c
index c4cc326..b719218 100644
--- a/sys/powerpc/powerpc/fpu.c
+++ b/sys/powerpc/powerpc/fpu.c
@@ -66,10 +66,11 @@ enable_fpu(struct thread *td)
 	 * initialise the FPU registers and FPSCR to 0, and set the flag
 	 * to indicate that the FPU is in use.
 	 */
+	pcb->pcb_flags |= PCB_FPU;
 	tf->srr1 |= PSL_FP;
-	if (!(pcb->pcb_flags & PCB_FPU)) {
+	if (!(pcb->pcb_flags & PCB_FPREGS)) {
 		memset(&pcb->pcb_fpu, 0, sizeof pcb->pcb_fpu);
-		pcb->pcb_flags |= PCB_FPU;
+		pcb->pcb_flags |= PCB_FPREGS;
 	}
 
 	/*
diff --git a/sys/powerpc/powerpc/swtch32.S b/sys/powerpc/powerpc/swtch32.S
index 5d88c2f..84f9ead 100644
--- a/sys/powerpc/powerpc/swtch32.S
+++ b/sys/powerpc/powerpc/swtch32.S
@@ -100,7 +100,6 @@ ENTRY(cpu_switch)
 	mr	%r16,%r5		/* and the new lock */
 	mr	%r17,%r6		/* and the PCB */
 	
-#ifdef AIM
 	lwz	%r7,PCB_FLAGS(%r17)
 	/* Save FPU context if needed */
 	andi.	%r7, %r7, PCB_FPU
@@ -116,7 +115,6 @@ ENTRY(cpu_switch)
 	bl	save_vec
 	
 .L2:
-#endif
 	mr	%r3,%r14		/* restore old thread ptr */
 	bl	pmap_deactivate		/* Deactivate the current pmap */
 
@@ -143,7 +141,6 @@ blocked_loop:
 	mr	%r3,%r2			/* Get new thread ptr */
 	bl	pmap_activate		/* Activate the new address space */
 
-#ifdef AIM
 	lwz	%r6, PCB_FLAGS(%r17)
 	/* Restore FPU context if needed */
 	andi.	%r6, %r6, PCB_FPU
@@ -160,7 +157,6 @@ blocked_loop:
 	bl	enable_vec
 
 .L4:
-#endif
 	/* thread to restore is in r3 */
 	mr	%r3,%r17		/* Recover PCB ptr */
 	lmw	%r12,PCB_CONTEXT(%r3)	/* Load the non-volatile GP regs */