Re: [PATCH 1/6] blk-mq: make blk_mq_stop_hw_queue() reliably block queue processing

[Tejun Heo <tj@kernel.org>]

Hello, Ming.

On Thu, Jun 19, 2014 at 06:10:28PM +0800, Ming Lei wrote:
> >  void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx)
> >  {
> > -       cancel_delayed_work(&hctx->run_work);
> > -       cancel_delayed_work(&hctx->delay_work);
> >         set_bit(BLK_MQ_S_STOPPED, &hctx->state);
> > +       cancel_delayed_work_sync(&hctx->run_work);
> > +       cancel_delayed_work_sync(&hctx->delay_work);
> 
> The function can be called from atomic context, like virtio-blk.
> 
> Currently blk_mq_stop_hw_queue() is called in case of queue
> being busy, so looks it is ok with cancel_delayed_work().

Hmm... you're right.  So the function is used to stop future
invocations of the queue processing function and thus the order
doesn't matter.  We probably wanna document that.  I don't understand
why virblk_freeze() is invoking it tho.  If freezing is complete,
there's no request in-flight on the driver side, why does it need to
invoke stop too?

Jens, please drop this one.

Thanks.

-- 
tejun