From 759167cfc8b8031bf50f4c49bbd4881ebd91cfa9 Mon Sep 17 00:00:00 2001 From: Hou Tao Date: Fri, 14 Oct 2022 19:39:46 +0800 Subject: [PATCH] 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 --- kernel/bpf/core.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c index 25a54e04560e..9c16338bcbe8 100644 --- 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)