On Tue, Jan 17, 2017 at 11:33:41AM +0900, Byungchul Park wrote:

On Mon, Jan 16, 2017 at 04:13:19PM +0100, Peter Zijlstra wrote:

quoted

On Fri, Dec 09, 2016 at 02:12:03PM +0900, Byungchul Park wrote:

quoted

+	/*
+	 * We assign class_idx here redundantly even though following
+	 * memcpy will cover it, in order to ensure a rcu reader can
+	 * access the class_idx atomically without lock.
+	 *
+	 * Here we assume setting a word-sized variable is atomic.

which one, where?

I meant xlock_class(xlock) in check_add_plock().

I was not sure about the following two.

1. Is it ordered between following a and b?
   a. memcpy -> list_add_tail_rcu
   b. list_for_each_entry_rcu -> load class_idx (xlock_class)
   I assumed that it's not ordered.
2. Does memcpy guarantee atomic store for each word?
   I assumed that it doesn't.

But I think I was wrong.. The first might be ordered. I will remove
the following redundant statement. It'd be orderd, right?

Yes, a and b are ordered, IOW, they could be paired, meaning when we
got the item in a list_for_each_entry_rcu() loop, all memory operations
before the corresponding list_add_tail_rcu() should be observed by us.

Regards,
Boqun

quoted

quoted

+	 */
+	xlock->hlock.class_idx = hlock->class_idx;
+	gen_id = (unsigned int)atomic_inc_return(&cross_gen_id);
+	WRITE_ONCE(xlock->gen_id, gen_id);
+	memcpy(&xlock->hlock, hlock, sizeof(struct held_lock));
+	INIT_LIST_HEAD(&xlock->xlock_entry);
+	list_add_tail_rcu(&xlock->xlock_entry, &xlocks_head);