Re: [PATCH v6 02/13] user_events: Add minimal support for trace_event into ftrace

[Steven Rostedt]

On Thu, 9 Dec 2021 09:40:50 -0800
Beau Belgrave <beaub@xxxxxxxxxxxxxxxxxxx> wrote:

> No, this is not a fast path, and I don't have a problem moving to a
> mutex if you feel that is better. I've likely become too close to this
> code to know when things are confusing for others.

Yeah. I really dislike the "protection by algorithms" then protection by
locking unless it is a fast path.

If this was a fast path then I'd be more concerned. I dislike global locks
as well, but unless contention becomes a concern, I don't think we should
worry about it.

Also, for this comment:

+static int user_events_release(struct inode *node, struct file *file)
+{
+	struct user_event_refs *refs;
+	struct user_event *user;
+	int i;
+
+	/*
+	 * refs is protected by RCU and could in theory change immediately
+	 * before this call on another core. To ensure we read the latest
+	 * version of refs we acquire the RCU read lock again.
+	 */
+	rcu_read_lock_sched();
+	refs = rcu_dereference_sched(file->private_data);
+	rcu_read_unlock_sched();

How do you see refs changing on another core if this can only be called
when nothing has a reference to it?

I think this comment and grabbing the rcu locks is what is causing me
concern.

-- Steve