On Tue, May 30, 2017 at 04:54:47PM +0000, Bart Van Assche wrote:

On Tue, 2017-05-30 at 08:22 +0800, Ming Lei wrote:

quoted

On Sun, May 28, 2017 at 04:10:09PM +0000, Bart Van Assche wrote:

quoted

I really would like to see the blk_queue_quiesced() tests as close as possible to
the blk_mq_hctx_stopped() tests. But I agree that we need a way to document and/or

Could you explain why we have to do that? And checking on stopped state
doesn't need to hold RCU/SRCU read lock, and that two states are really
different.

I'm really surprised that you ask me why ... It's because the purpose of the
"stopped" and "quiesced" flags is similar, namely preventing that dispatching

Actually they are not, and you can find it in patch 7.

But I will move the check into the dispatch function, and add comment about
rcu/scru read lock requirement.

requests happens. It doesn't matter that with your patches applied it is no
longer needed to hold an RCU / SRCU lock when testing the stopped flag.

quoted

quoted

verify that these tests occur with an RCU read-side lock held. Have you considered
to use rcu_read_lock_held() to document this?

Then we need to check if it is RCU or SRCU, and make code ugly as
current check on BLOCKING.

How about introducing a macro or inline function in the block layer that tests
whether either the RCU read lock or an SRCU read lock is held depending on the
value of the BLK_MQ_F_BLOCKING flag?

That might make code clean, but need to check the condition two times.

Thanks,
Ming