Index: sys/amd64/amd64/machdep.c
===================================================================
--- sys/amd64/amd64/machdep.c
+++ sys/amd64/amd64/machdep.c
@@ -604,7 +604,6 @@
 	regs->tf_fs = _ufssel;
 	regs->tf_gs = _ugssel;
 	regs->tf_flags = TF_HASSEGS;
-	td->td_retval[1] = 0;
 
 	/*
 	 * Reset the hardware debug registers if they were in use.
Index: sys/amd64/ia32/ia32_signal.c
===================================================================
--- sys/amd64/ia32/ia32_signal.c
+++ sys/amd64/ia32/ia32_signal.c
@@ -967,5 +967,4 @@
 
 	/* Return via doreti so that we can change to a different %cs */
 	set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET);
-	td->td_retval[1] = 0;
 }
Index: sys/arm/cloudabi32/cloudabi32_sysvec.c
===================================================================
--- sys/arm/cloudabi32/cloudabi32_sysvec.c
+++ sys/arm/cloudabi32/cloudabi32_sysvec.c
@@ -61,7 +61,7 @@
 	 * tpidrurw to the TCB.
 	 */
 	regs = td->td_frame;
-	regs->tf_r0 = td->td_retval[0] =
+	regs->tf_r0 =
 	    stack + roundup(sizeof(cloudabi32_tcb_t), sizeof(register_t));
 	(void)cpu_set_user_tls(td, (void *)stack);
 }
Index: sys/arm64/arm64/machdep.c
===================================================================
--- sys/arm64/arm64/machdep.c
+++ sys/arm64/arm64/machdep.c
@@ -310,12 +310,7 @@
 
 	memset(tf, 0, sizeof(struct trapframe));
 
-	/*
-	 * We need to set x0 for init as it doesn't call
-	 * cpu_set_syscall_retval to copy the value. We also
-	 * need to set td_retval for the cases where we do.
-	 */
-	tf->tf_x[0] = td->td_retval[0] = stack;
+	tf->tf_x[0] = stack;
 	tf->tf_sp = STACKALIGN(stack);
 	tf->tf_lr = imgp->entry_addr;
 	tf->tf_elr = imgp->entry_addr;
Index: sys/arm64/cloudabi64/cloudabi64_sysvec.c
===================================================================
--- sys/arm64/cloudabi64/cloudabi64_sysvec.c
+++ sys/arm64/cloudabi64/cloudabi64_sysvec.c
@@ -61,7 +61,7 @@
 	 * tpidr_el0 to the TCB.
 	 */
 	regs = td->td_frame;
-	regs->tf_x[0] = td->td_retval[0] =
+	regs->tf_x[0] =
 	    stack + roundup(sizeof(cloudabi64_tcb_t), sizeof(register_t));
 	(void)cpu_set_user_tls(td, (void *)stack);
 }
Index: sys/compat/linux/linux_emul.c
===================================================================
--- sys/compat/linux/linux_emul.c
+++ sys/compat/linux/linux_emul.c
@@ -186,7 +186,7 @@
 
 	error = kern_execve(td, eargs, NULL);
 	post_execve(td, error, oldvmspace);
-	if (error != 0)
+	if (error != EJUSTRETURN)
 		return (error);
 
 	/*
@@ -213,7 +213,7 @@
 		free(em, M_TEMP);
 		free(pem, M_LINUX);
 	}
-	return (0);
+	return (EJUSTRETURN);
 }
 
 void 
Index: sys/i386/i386/machdep.c
===================================================================
--- sys/i386/i386/machdep.c
+++ sys/i386/i386/machdep.c
@@ -1126,6 +1126,7 @@
 	set_fsbase(td, 0);
 	set_gsbase(td, 0);
 
+	/* Make sure edx is 0x0 on entry. Linux binaries depend on it. */
 	bzero((char *)regs, sizeof(struct trapframe));
 	regs->tf_eip = imgp->entry_addr;
 	regs->tf_esp = stack;
@@ -1168,13 +1169,6 @@
 	 * clean FP state if it uses the FPU again.
 	 */
 	fpstate_drop(td);
-
-	/*
-	 * XXX - Linux emulator
-	 * Make sure sure edx is 0x0 on entry. Linux binaries depend
-	 * on it.
-	 */
-	td->td_retval[1] = 0;
 }
 
 void
Index: sys/kern/init_main.c
===================================================================
--- sys/kern/init_main.c
+++ sys/kern/init_main.c
@@ -797,7 +797,7 @@
 		 * Otherwise, return via fork_trampoline() all the way
 		 * to user mode as init!
 		 */
-		if ((error = sys_execve(td, &args)) == 0) {
+		if ((error = sys_execve(td, &args)) == EJUSTRETURN) {
 			mtx_unlock(&Giant);
 			return;
 		}
Index: sys/kern/kern_exec.c
===================================================================
--- sys/kern/kern_exec.c
+++ sys/kern/kern_exec.c
@@ -318,7 +318,7 @@
 		 * If success, we upgrade to SINGLE_EXIT state to
 		 * force other threads to suicide.
 		 */
-		if (error == 0)
+		if (error == EJUSTRETURN)
 			thread_single(p, SINGLE_EXIT);
 		else
 			thread_single_end(p, SINGLE_BOUNDARY);
@@ -962,7 +962,13 @@
 		ktrprocctor(p);
 #endif
 
-	return (error);
+	/*
+	 * We don't want cpu_set_syscall_retval() to overwrite any of
+	 * the register values put in place by exec_setregs().
+	 * Implementations of cpu_set_syscall_retval() will leave
+	 * registers unmodified when returning EJUSTRETURN.
+	 */
+	return (error == 0 ? EJUSTRETURN : error);
 }
 
 int
Index: sys/powerpc/powerpc/exec_machdep.c
===================================================================
--- sys/powerpc/powerpc/exec_machdep.c
+++ sys/powerpc/powerpc/exec_machdep.c
@@ -520,22 +520,11 @@
 	 *	- ps_strings is a NetBSD extention, and will be
 	 * 	  ignored by executables which are strictly
 	 *	  compliant with the SVR4 ABI.
-	 *
-	 * XXX We have to set both regs and retval here due to different
-	 * XXX calling convention in trap.c and init_main.c.
 	 */
 
 	/* Collect argc from the user stack */
 	argc = fuword((void *)stack);
 
-        /*
-         * XXX PG: these get overwritten in the syscall return code.
-         * execve() should return EJUSTRETURN, like it does on NetBSD.
-         * Emulate by setting the syscall return value cells. The
-         * registers still have to be set for init's fork trampoline.
-         */
-        td->td_retval[0] = argc;
-        td->td_retval[1] = stack + sizeof(register_t);
 	tf->fixreg[3] = argc;
 	tf->fixreg[4] = stack + sizeof(register_t);
 	tf->fixreg[5] = stack + (2 + argc)*sizeof(register_t);
@@ -572,8 +561,6 @@
 
 	argc = fuword32((void *)stack);
 
-        td->td_retval[0] = argc;
-        td->td_retval[1] = stack + sizeof(uint32_t);
 	tf->fixreg[3] = argc;
 	tf->fixreg[4] = stack + sizeof(uint32_t);
 	tf->fixreg[5] = stack + (2 + argc)*sizeof(uint32_t);
Index: sys/riscv/riscv/machdep.c
===================================================================
--- sys/riscv/riscv/machdep.c
+++ sys/riscv/riscv/machdep.c
@@ -279,12 +279,7 @@
 
 	memset(tf, 0, sizeof(struct trapframe));
 
-	/*
-	 * We need to set a0 for init as it doesn't call
-	 * cpu_set_syscall_retval to copy the value. We also
-	 * need to set td_retval for the cases where we do.
-	 */
-	tf->tf_a[0] = td->td_retval[0] = stack;
+	tf->tf_a[0] = stack;
 	tf->tf_sp = STACKALIGN(stack);
 	tf->tf_ra = imgp->entry_addr;
 	tf->tf_sepc = imgp->entry_addr;
Index: sys/sparc64/sparc64/machdep.c
===================================================================
--- sys/sparc64/sparc64/machdep.c
+++ sys/sparc64/sparc64/machdep.c
@@ -1009,9 +1009,6 @@
 	 * header, it turns out that just always using TSO performs best.
 	 */
 	tf->tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_MM_TSO;
-
-	td->td_retval[0] = tf->tf_out[0];
-	td->td_retval[1] = tf->tf_out[1];
 }
 
 int
Index: sys/sparc64/sparc64/vm_machdep.c
===================================================================
--- sys/sparc64/sparc64/vm_machdep.c
+++ sys/sparc64/sparc64/vm_machdep.c
@@ -213,9 +213,6 @@
 	tf->tf_out[6] = sp - SPOFF - sizeof(struct frame);
 	tf->tf_tpc = (uint64_t)entry;
 	tf->tf_tnpc = tf->tf_tpc + 4;
-
-	td->td_retval[0] = tf->tf_out[0];
-	td->td_retval[1] = tf->tf_out[1];
 }
 
 int