Index: sys/amd64/amd64/machdep.c
===================================================================
--- sys/amd64/amd64/machdep.c
+++ sys/amd64/amd64/machdep.c
@@ -1960,15 +1960,26 @@
 }
 
 int
-ptrace_single_step(struct thread *td)
+ptrace_single_step(struct thread *td, bool onetime)
 {
-	td->td_frame->tf_rflags |= PSL_T;
+	bool t_set;
+
+	t_set = (td->td_frame->tf_rflags & PSL_T) != 0;
+	if (!t_set) {
+		td->td_frame->tf_rflags |= PSL_T;
+		if (onetime) {
+			PROC_LOCK(td->td_proc);
+			td->td_dbgflags |= TDB_STEP;
+			PROC_UNLOCK(td->td_proc);
+		}
+	}
 	return (0);
 }
 
 int
 ptrace_clear_single_step(struct thread *td)
 {
+
 	td->td_frame->tf_rflags &= ~PSL_T;
 	return (0);
 }
Index: sys/amd64/amd64/trap.c
===================================================================
--- sys/amd64/amd64/trap.c
+++ sys/amd64/amd64/trap.c
@@ -266,19 +266,27 @@
 			ucode = ILL_PRVOPC;
 			break;
 
-		case T_BPTFLT:		/* bpt instruction fault */
 		case T_TRCTRAP:		/* trace trap */
 			enable_intr();
-#ifdef KDTRACE_HOOKS
-			if (type == T_BPTFLT) {
-				if (dtrace_pid_probe_ptr != NULL &&
-				    dtrace_pid_probe_ptr(frame) == 0)
-					return;
+			PROC_LOCK(td->td_proc);
+			if ((td->td_dbgflags & TDB_STEP) != 0) {
+				td->td_frame->tf_rflags &= ~PSL_T;
+				td->td_dbgflags &= ~TDB_STEP;
 			}
+			PROC_UNLOCK(td->td_proc);
+			ucode = TRAP_TRACE;
+			signo = SIGTRAP;
+			break;
+
+		case T_BPTFLT:		/* bpt instruction fault */
+			enable_intr();
+#ifdef KDTRACE_HOOKS
+			if (dtrace_pid_probe_ptr != NULL &&
+			    dtrace_pid_probe_ptr(frame) == 0)
+				return;
 #endif
-			frame->tf_rflags &= ~PSL_T;
+			ucode = TRAP_BRKPT;
 			signo = SIGTRAP;
-			ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 			break;
 
 		case T_ARITHTRAP:	/* arithmetic trap */
Index: sys/i386/i386/machdep.c
===================================================================
--- sys/i386/i386/machdep.c
+++ sys/i386/i386/machdep.c
@@ -2769,15 +2769,26 @@
 }
 
 int
-ptrace_single_step(struct thread *td)
+ptrace_single_step(struct thread *td, bool onetime)
 {
-	td->td_frame->tf_eflags |= PSL_T;
+	bool t_set;
+
+	t_set = (td->td_frame->tf_eflags & PSL_T) != 0;
+	if (!t_set) {
+		td->td_frame->tf_eflags |= PSL_T;
+		if (onetime) {
+			PROC_LOCK(td->td_proc);
+			td->td_dbgflags |= TDB_STEP;
+			PROC_UNLOCK(td->td_proc);
+		}
+	}
 	return (0);
 }
 
 int
 ptrace_clear_single_step(struct thread *td)
 {
+
 	td->td_frame->tf_eflags &= ~PSL_T;
 	return (0);
 }
Index: sys/i386/i386/trap.c
===================================================================
--- sys/i386/i386/trap.c
+++ sys/i386/i386/trap.c
@@ -320,20 +320,28 @@
 			ucode = ILL_PRVOPC;
 			break;
 
-		case T_BPTFLT:		/* bpt instruction fault */
 		case T_TRCTRAP:		/* trace trap */
 			enable_intr();
-#ifdef KDTRACE_HOOKS
-			if (type == T_BPTFLT) {
-				if (dtrace_pid_probe_ptr != NULL &&
-				    dtrace_pid_probe_ptr(frame) == 0)
-					return;
+			PROC_LOCK(td->td_proc);
+			if ((td->td_dbgflags & TDB_STEP) != 0) {
+				td->td_frame->tf_eflags &= ~PSL_T;
+				td->td_dbgflags &= ~TDB_STEP;
 			}
-#endif
+			PROC_UNLOCK(td->td_proc);
 user_trctrap_out:
-			frame->tf_eflags &= ~PSL_T;
 			signo = SIGTRAP;
-			ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
+			ucode = TRAP_TRACE;
+			break;
+
+		case T_BPTFLT:		/* bpt instruction fault */
+			enable_intr();
+#ifdef KDTRACE_HOOKS
+			if (dtrace_pid_probe_ptr != NULL &&
+			    dtrace_pid_probe_ptr(frame) == 0)
+				return;
+#endif
+			signo = SIGTRAP;
+			ucode = TRAP_BRKPT;
 			break;
 
 		case T_ARITHTRAP:	/* arithmetic trap */
@@ -352,7 +360,6 @@
 			if (frame->tf_eflags & PSL_VM) {
 				signo = vm86_emulate((struct vm86frame *)frame);
 				if (signo == SIGTRAP) {
-					type = T_TRCTRAP;
 					load_dr6(rdr6() | 0x4000);
 					goto user_trctrap_out;
 				}
Index: sys/kern/sys_process.c
===================================================================
--- sys/kern/sys_process.c
+++ sys/kern/sys_process.c
@@ -221,7 +221,7 @@
 proc_sstep(struct thread *td)
 {
 
-	PROC_ACTION(ptrace_single_step(td));
+	PROC_ACTION(ptrace_single_step(td, true));
 }
 
 int
@@ -951,7 +951,7 @@
 	case PT_SETSTEP:
 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
 		    p->p_pid);
-		error = ptrace_single_step(td2);
+		error = ptrace_single_step(td2, false);
 		break;
 
 	case PT_SUSPEND:
@@ -1053,7 +1053,7 @@
 		case PT_STEP:
 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
 			    td2->td_tid, p->p_pid, data);
-			error = ptrace_single_step(td2);
+			error = ptrace_single_step(td2, true);
 			if (error)
 				goto out;
 			break;
Index: sys/sys/proc.h
===================================================================
--- sys/sys/proc.h
+++ sys/sys/proc.h
@@ -445,6 +445,7 @@
 #define	TDB_EXIT	0x00000400 /* Exiting LWP indicator for ptrace() */
 #define	TDB_VFORK	0x00000800 /* vfork indicator for ptrace() */
 #define	TDB_FSTP	0x00001000 /* The thread is PT_ATTACH leader */
+#define	TDB_STEP	0x00002000 /* (x86) PSL_T set for PT_STEP */
 
 /*
  * "Private" flags kept in td_pflags:
Index: sys/sys/ptrace.h
===================================================================
--- sys/sys/ptrace.h
+++ sys/sys/ptrace.h
@@ -172,7 +172,7 @@
 #ifdef _KERNEL
 
 int	ptrace_set_pc(struct thread *_td, unsigned long _addr);
-int	ptrace_single_step(struct thread *_td);
+int	ptrace_single_step(struct thread *_td, bool onetime);
 int	ptrace_clear_single_step(struct thread *_td);
 
 #ifdef __HAVE_PTRACE_MACHDEP