diff --git a/sys/kern/kern_exec.c b/sys/kern/kern_exec.c
index 154dbf8..208187f 100644
--- a/sys/kern/kern_exec.c
+++ b/sys/kern/kern_exec.c
@@ -1323,8 +1323,18 @@ err_exit:
 int
 exec_alloc_args(struct image_args *args)
 {
+	void *buf;
 
-	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
+	critical_enter();
+	if ((buf = PCPU_GET(execargs)) != NULL) {
+		args->buf = buf;
+		PCPU_SET(execargs, NULL);
+		critical_exit();
+	} else {
+		critical_exit();
+		args->buf = (char *)kmap_alloc_wait(exec_map,
+		    PATH_MAX + ARG_MAX);
+	}
 	return (args->buf != NULL ? 0 : ENOMEM);
 }
 
@@ -1333,8 +1343,15 @@ exec_free_args(struct image_args *args)
 {
 
 	if (args->buf != NULL) {
-		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
-		    PATH_MAX + ARG_MAX);
+		critical_enter();
+		if (PCPU_GET(execargs) == NULL) {
+			PCPU_SET(execargs, args->buf);
+			critical_exit();
+		} else {
+			critical_exit();
+			kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
+			    PATH_MAX + ARG_MAX);
+		}
 		args->buf = NULL;
 	}
 	if (args->fname_buf != NULL) {
diff --git a/sys/sys/pcpu.h b/sys/sys/pcpu.h
index d6d1b3d..a9c79b4 100644
--- a/sys/sys/pcpu.h
+++ b/sys/sys/pcpu.h
@@ -166,6 +166,7 @@ struct pcpu {
 	int		pc_domain;		/* Memory domain. */
 	struct rm_queue	pc_rm_queue;		/* rmlock list of trackers */
 	uintptr_t	pc_dynamic;		/* Dynamic per-cpu data area */
+	void		*pc_execargs;		/* cached exec arg buffer */
 
 	/*
 	 * Keep MD fields last, so that CPU-specific variations on a
diff --git a/sys/vm/vm_init.c b/sys/vm/vm_init.c
index e664585..10ffdda 100644
--- a/sys/vm/vm_init.c
+++ b/sys/vm/vm_init.c
@@ -91,8 +91,8 @@ __FBSDID("$FreeBSD$");
 
 long physmem;
 
-static int exec_map_entries = 16;
-SYSCTL_INT(_vm, OID_AUTO, exec_map_entries, CTLFLAG_RDTUN, &exec_map_entries, 0,
+static int exec_map_entries;
+SYSCTL_INT(_vm, OID_AUTO, exec_map_entries, CTLFLAG_RD, &exec_map_entries, 0,
     "Maximum number of simultaneous execs");
 
 /*
@@ -261,8 +261,11 @@ again:
 		panic("Clean map calculation incorrect");
 
 	/*
- 	 * Allocate the pageable submaps.
+	 * Allocate the pageable submaps.  We may cache a buffer of size
+	 * PATH_MAX+ARG_MAX per CPU, so we therefore need to allocate space for
+	 * at least ncpu+1 buffers to avoid deadlock.
 	 */
+	exec_map_entries = mp_ncpus * 2;
 	exec_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr,
 	    exec_map_entries * round_page(PATH_MAX + ARG_MAX), FALSE);
 	pipe_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr, maxpipekva,