bpf: Use rcu_trace_implies_rcu_gp() for program array freeing

To support both sleepable and normal uprobe bpf program, the freeing of trace program array chains a RCU-tasks-trace grace period and a normal RCU grace period one after the other. With the introduction of rcu_trace_implies_rcu_gp(), __bpf_prog_array_free_sleepable_cb() can check whether or not a normal RCU grace period has also passed after a RCU-tasks-trace grace period has passed. If it is true, it is safe to invoke kfree() directly. Signed-off-by: Hou Tao <[email protected]>
kernel-patches · Oct 20, 2022 · 5e0d56c · 5e0d56c
1 parent ef72f74
commit 5e0d56c
Showing 1 changed file with 7 additions and 1 deletion.
diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
@@ -2251,8 +2251,14 @@ static void __bpf_prog_array_free_sleepable_cb(struct rcu_head *rcu)
 {
 	struct bpf_prog_array *progs;
 
+	/* If RCU Tasks Trace grace period implies RCU grace period, there is
+	 * no need to call kfree_rcu(), just call kfree() directly.
+	 */
 	progs = container_of(rcu, struct bpf_prog_array, rcu);
-	kfree_rcu(progs, rcu);
+	if (rcu_trace_implies_rcu_gp())
+		kfree(progs);
+	else
+		kfree_rcu(progs, rcu);
 }
 
 void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs)