diff --git a/sys/kern/kern_exec.c b/sys/kern/kern_exec.c
--- a/sys/kern/kern_exec.c
+++ b/sys/kern/kern_exec.c
@@ -307,8 +307,17 @@
 	p = td->td_proc;
 	if ((p->p_flag & P_HADTHREADS) != 0) {
 		PROC_LOCK(p);
+		while (p->p_singlethr > 0) {
+			error = msleep(&p->p_singlethr, &p->p_mtx,
+			    PWAIT | PCATCH, "exec1t", 0);
+			if (error != 0) {
+				error = ERESTART;
+				goto unlock;
+			}
+		}
 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
 			error = ERESTART;
+unlock:
 		PROC_UNLOCK(p);
 	}
 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c
--- a/sys/kern/kern_exit.c
+++ b/sys/kern/kern_exit.c
@@ -213,6 +213,15 @@
 	__unreachable();
 }
 
+void
+proc_set_p2_wexit(struct proc *p)
+{
+	PROC_LOCK_ASSERT(p, MA_OWNED);
+	p->p_flag2 |= P2_WEXIT;
+	while (p->p_singlethr > 0)
+		msleep(&p->p_singlethr, &p->p_mtx, PWAIT | PCATCH, "exit1t", 0);
+}
+
 /*
  * Exit: deallocate address space and other resources, change proc state to
  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
@@ -251,6 +260,8 @@
 	 * MUST abort all other threads before proceeding past here.
 	 */
 	PROC_LOCK(p);
+	proc_set_p2_wexit(p);
+
 	/*
 	 * First check if some other thread or external request got
 	 * here before us.  If so, act appropriately: exit or suspend.
diff --git a/sys/kern/kern_proc.c b/sys/kern/kern_proc.c
--- a/sys/kern/kern_proc.c
+++ b/sys/kern/kern_proc.c
@@ -3399,10 +3399,10 @@
 static struct sx stop_all_proc_blocker;
 SX_SYSINIT(stop_all_proc_blocker, &stop_all_proc_blocker, "sapblk");
 
-void
+bool
 stop_all_proc_block(void)
 {
-	sx_xlock(&stop_all_proc_blocker);
+	return (sx_xlock_sig(&stop_all_proc_blocker) == 0);
 }
 
 void
@@ -3426,7 +3426,8 @@
 	int r, gen;
 	bool restart, seen_stopped, seen_exiting, stopped_some;
 
-	stop_all_proc_block();
+	if (!stop_all_proc_block())
+		return;
 
 	cp = curproc;
 allproc_loop:
@@ -3446,7 +3447,7 @@
 			PROC_UNLOCK(p);
 			continue;
 		}
-		if ((p->p_flag & P_WEXIT) != 0) {
+		if ((p->p_flag2 & P2_WEXIT) != 0) {
 			seen_exiting = true;
 			PROC_UNLOCK(p);
 			continue;
diff --git a/sys/kern/kern_procctl.c b/sys/kern/kern_procctl.c
--- a/sys/kern/kern_procctl.c
+++ b/sys/kern/kern_procctl.c
@@ -254,7 +254,7 @@
 	PROC_LOCK(p);
 }
 
-static bool
+static void
 reap_kill_proc_locked(struct thread *td, struct proc *p2,
     ksiginfo_t *ksi, struct procctl_reaper_kill *rk, int *error)
 {
@@ -270,7 +270,7 @@
 			rk->rk_fpid = p2->p_pid;
 			*error = error1;
 		}
-		return (true);
+		return;
 	}
 
 	/*
@@ -291,47 +291,38 @@
 	 * race.
 	 */
 	need_stop = p2 != td->td_proc &&
-	    (p2->p_flag & (P_KPROC | P_SYSTEM)) == 0 &&
+	    (td->td_proc->p_flag2 & P2_WEXIT) == 0 &&
+	    (p2->p_flag & (P_KPROC | P_SYSTEM | P_STOPPED)) == 0 &&
 	    (rk->rk_flags & REAPER_KILL_CHILDREN) == 0;
 
 	if (need_stop) {
-		if (P_SHOULDSTOP(p2) == P_STOPPED_SINGLE)
-			return (false);	/* retry later */
 		xlocked = sx_xlocked(&proctree_lock);
 		sx_unlock(&proctree_lock);
 		r = thread_single(p2, SINGLE_ALLPROC);
-		if (r != 0) {
-			reap_kill_proc_relock(p2, xlocked);
-			return (false);
-		}
+		reap_kill_proc_relock(p2, xlocked);
+		if (r != 0)
+			need_stop = false;
 	}
 
 	pksignal(p2, rk->rk_sig, ksi);
 	rk->rk_killed++;
 	*error = error1;
 
-	if (need_stop) {
-		reap_kill_proc_relock(p2, xlocked);
+	if (need_stop)
 		thread_single_end(p2, SINGLE_ALLPROC);
-	}
-	return (true);
 }
 
-static bool
+static void
 reap_kill_proc(struct thread *td, struct proc *p2, ksiginfo_t *ksi,
     struct procctl_reaper_kill *rk, int *error)
 {
-	bool res;
-
-	res = true;
 	PROC_LOCK(p2);
-	if ((p2->p_flag & P_WEXIT) == 0) {
+	if ((p2->p_flag2 & P2_WEXIT) == 0) {
 		_PHOLD_LITE(p2);
-		res = reap_kill_proc_locked(td, p2, ksi, rk, error);
+		reap_kill_proc_locked(td, p2, ksi, rk, error);
 		_PRELE(p2);
 	}
 	PROC_UNLOCK(p2);
-	return (res);
 }
 
 struct reap_kill_tracker {
@@ -346,11 +337,25 @@
 {
 	struct reap_kill_tracker *t;
 
+	PROC_LOCK(p2);
+	if ((p2->p_flag2 & P2_WEXIT) != 0) {
+		PROC_UNLOCK(p2);
+		return;
+	}
+	_PHOLD_LITE(p2);
+	PROC_UNLOCK(p2);
 	t = malloc(sizeof(struct reap_kill_tracker), M_TEMP, M_WAITOK);
 	t->parent = p2;
 	TAILQ_INSERT_TAIL(tracker, t, link);
 }
 
+static void
+reap_kill_sched_free(struct reap_kill_tracker *t)
+{
+	PRELE(t->parent);
+	free(t, M_TEMP);
+}
+
 static void
 reap_kill_children(struct thread *td, struct proc *reaper,
     struct procctl_reaper_kill *rk, ksiginfo_t *ksi, int *error)
@@ -380,8 +385,20 @@
 	TAILQ_INIT(&tracker);
 	reap_kill_sched(&tracker, reaper);
 	while ((t = TAILQ_FIRST(&tracker)) != NULL) {
-		MPASS((t->parent->p_treeflag & P_TREE_REAPER) != 0);
 		TAILQ_REMOVE(&tracker, t, link);
+
+		/*
+		 * Since reap_kill_proc() drops proctree_lock sx, it
+		 * is possible that the tracker reapper is no longer.
+		 * In this case the subtree is reparented to the new
+		 * reaper, which should handle it.
+		 */
+		if ((t->parent->p_treeflag & P_TREE_REAPER) == 0) {
+			reap_kill_sched_free(t);
+			res = true;
+			continue;
+		}
+
 		LIST_FOREACH(p2, &t->parent->p_reaplist, p_reapsibling) {
 			if (t->parent == reaper &&
 			    (rk->rk_flags & REAPER_KILL_SUBTREE) != 0 &&
@@ -391,11 +408,10 @@
 				reap_kill_sched(&tracker, p2);
 			if (alloc_unr_specific(pids, p2->p_pid) != p2->p_pid)
 				continue;
-			if (!reap_kill_proc(td, p2, ksi, rk, error))
-				free_unr(pids, p2->p_pid);
+			reap_kill_proc(td, p2, ksi, rk, error);
 			res = true;
 		}
-		free(t, M_TEMP);
+		reap_kill_sched_free(t);
 	}
 	return (res);
 }
@@ -412,8 +428,21 @@
 	 * repeated.
 	 */
 	init_unrhdr(&pids, 1, PID_MAX, UNR_NO_MTX);
+	PROC_LOCK(td->td_proc);
+	if ((td->td_proc->p_flag2 & P2_WEXIT) != 0) {
+		PROC_UNLOCK(td->td_proc);
+		goto out;
+	}
+	td->td_proc->p_singlethr++;
+	PROC_UNLOCK(td->td_proc);
 	while (reap_kill_subtree_once(td, p, reaper, rk, ksi, error, &pids))
 	       ;
+	PROC_LOCK(td->td_proc);
+	td->td_proc->p_singlethr--;
+	if (td->td_proc->p_singlethr == 0)
+		wakeup(&p->p_singlethr);
+	PROC_UNLOCK(td->td_proc);
+out:
 	clean_unrhdr(&pids);
 	clear_unrhdr(&pids);
 }
@@ -1050,8 +1079,8 @@
 	sapblk = false;
 	if (cmd_info->sapblk != NULL) {
 		sapblk = cmd_info->sapblk(td, data);
-		if (sapblk)
-			stop_all_proc_block();
+		if (sapblk && !stop_all_proc_block())
+			return (ERESTART);
 	}
 
 	switch (cmd_info->lock_tree) {
diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c
--- a/sys/kern/kern_sig.c
+++ b/sys/kern/kern_sig.c
@@ -3406,6 +3406,8 @@
 	struct proc *p = td->td_proc;
 
 	PROC_LOCK_ASSERT(p, MA_OWNED);
+	proc_set_p2_wexit(p);
+
 	p->p_acflag |= AXSIG;
 	/*
 	 * We must be single-threading to generate a core dump.  This
diff --git a/sys/kern/kern_thread.c b/sys/kern/kern_thread.c
--- a/sys/kern/kern_thread.c
+++ b/sys/kern/kern_thread.c
@@ -99,7 +99,7 @@
     "struct proc KBI p_pid");
 _Static_assert(offsetof(struct proc, p_filemon) == 0x3c8,
     "struct proc KBI p_filemon");
-_Static_assert(offsetof(struct proc, p_comm) == 0x3e0,
+_Static_assert(offsetof(struct proc, p_comm) == 0x3e4,
     "struct proc KBI p_comm");
 _Static_assert(offsetof(struct proc, p_emuldata) == 0x4c8,
     "struct proc KBI p_emuldata");
@@ -1148,20 +1148,16 @@
 		 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP
 		 * is used to avoid immediate un-suspend.
 		 */
-		if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY |
-		    TDF_ALLPROCSUSP)) == 0) {
+		if (TD_IS_SUSPENDED(td2) && (td2->td_flags &
+		    TDF_ALLPROCSUSP) == 0) {
 			wakeup_swapper |= thread_unsuspend_one(td2, p, false);
 			thread_lock(td2);
 			goto restart;
 		}
 		if (TD_CAN_ABORT(td2)) {
-			if ((td2->td_flags & TDF_SBDRY) == 0) {
-				thread_suspend_one(td2);
-				td2->td_flags |= TDF_ALLPROCSUSP;
-			} else {
-				wakeup_swapper |= sleepq_abort(td2, ERESTART);
-				return (wakeup_swapper);
-			}
+			td2->td_flags |= TDF_ALLPROCSUSP;
+			wakeup_swapper |= sleepq_abort(td2, ERESTART);
+			return (wakeup_swapper);
 		}
 		break;
 	default:
@@ -1207,10 +1203,18 @@
 	mtx_assert(&Giant, MA_NOTOWNED);
 	PROC_LOCK_ASSERT(p, MA_OWNED);
 
-	if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
+	/*
+	 * Is someone already single threading?
+	 * Or may be singlethreading is not needed at all.
+	 */
+	if (mode == SINGLE_ALLPROC) {
+		while ((p->p_flag & P_STOPPED_SINGLE) != 0) {
+			if ((p->p_flag2 & P2_WEXIT) != 0)
+				return (1);
+			msleep(&p->p_flag, &p->p_mtx, PCATCH, "thrsgl", 0);
+		}
+	} else if ((p->p_flag & P_HADTHREADS) == 0)
 		return (0);
-
-	/* Is someone already single threading? */
 	if (p->p_singlethread != NULL && p->p_singlethread != td)
 		return (1);
 
@@ -1242,7 +1246,7 @@
 			if (TD_IS_INHIBITED(td2)) {
 				wakeup_swapper |= weed_inhib(mode, td2, p);
 #ifdef SMP
-			} else if (TD_IS_RUNNING(td2) && td != td2) {
+			} else if (TD_IS_RUNNING(td2)) {
 				forward_signal(td2);
 				thread_unlock(td2);
 #endif
@@ -1669,6 +1673,7 @@
 	PROC_SUNLOCK(p);
 	if (wakeup_swapper)
 		kick_proc0();
+	wakeup(&p->p_flag);
 }
 
 /*
diff --git a/sys/sys/proc.h b/sys/sys/proc.h
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@@ -691,6 +691,8 @@
 	int		p_pendingexits; /* (c) Count of pending thread exits. */
 	struct filemon	*p_filemon;	/* (c) filemon-specific data. */
 	int		p_pdeathsig;	/* (c) Signal from parent on exit. */
+	int		p_singlethr;	/* (?) Count of threads doing
+					   external thread_single() */
 /* End area that is zeroed on creation. */
 #define	p_endzero	p_magic
 
@@ -847,6 +849,8 @@
 #define	P2_NO_NEW_PRIVS		0x00008000	/* Ignore setuid */
 #define	P2_WXORX_DISABLE	0x00010000	/* WX mappings enabled */
 #define	P2_WXORX_ENABLE_EXEC	0x00020000	/* WXORX enabled after exec */
+#define	P2_WEXIT		0x00040000	/* exit just started,
+						   no thread_single() */
 
 /* Flags protected by proctree_lock, kept in p_treeflags. */
 #define	P_TREE_ORPHANED		0x00000001	/* Reparented, on orphan list */
@@ -1155,6 +1159,7 @@
 struct proc *proc_realparent(struct proc *child);
 void	proc_reap(struct thread *td, struct proc *p, int *status, int options);
 void	proc_reparent(struct proc *child, struct proc *newparent, bool set_oppid);
+void	proc_set_p2_wexit(struct proc *p);
 void	proc_set_traced(struct proc *p, bool stop);
 void	proc_wkilled(struct proc *p);
 struct	pstats *pstats_alloc(void);
@@ -1232,7 +1237,7 @@
 void	thread_unsuspend(struct proc *p);
 void	thread_wait(struct proc *p);
 
-void	stop_all_proc_block(void);
+bool	stop_all_proc_block(void);
 void	stop_all_proc_unblock(void);
 void	stop_all_proc(void);
 void	resume_all_proc(void);