tracing/probes: Avoid setting TPARG_FL_FENTRY and TPARG_FL_RETURN

When parsing a kprobe event, the return probe always sets both
TPARG_FL_RETURN and TPARG_FL_FENTRY, but this is not useful because
some fetchargs are only for return probe and some others only for
function entry. Make it obviously mutual exclusive.

Link: https://lore.kernel.org/all/168507468731.913472.11354553441385410734.stgit@mhiramat.roam.corp.google.com/

Signed-off-by: Masami Hiramatsu (Google) <[email protected]>

Author: mhiramat

kernel/trace/trace_kprobe.c

@@ -825,7 +825,7 @@ static int __trace_kprobe_create(int argc, const char *argv[])
 		if (is_return)
 			flags |= TPARG_FL_RETURN;
 		ret = kprobe_on_func_entry(NULL, symbol, offset);
-		if (ret == 0)
+		if (ret == 0 && !is_return)
 			flags |= TPARG_FL_FENTRY;
 		/* Defer the ENOENT case until register kprobe */
 		if (ret == -EINVAL && is_return) {

kernel/trace/trace_probe.h

@@ -357,6 +357,11 @@ int trace_probe_print_args(struct trace_seq *s, struct probe_arg *args, int nr_a
 #define trace_probe_for_each_link_rcu(pos, tp)	\
 	list_for_each_entry_rcu(pos, &(tp)->event->files, list)
 
+/*
+ * The flags used for parsing trace_probe arguments.
+ * TPARG_FL_RETURN, TPARG_FL_FENTRY and TPARG_FL_TPOINT are mutually exclusive.
+ * TPARG_FL_KERNEL and TPARG_FL_USER are also mutually exclusive.
+ */
 #define TPARG_FL_RETURN BIT(0)
 #define TPARG_FL_KERNEL BIT(1)
 #define TPARG_FL_FENTRY BIT(2)