On Tue, Jul 04, 2017 at 10:55:10AM +0300, Sagi Grimberg wrote:
> Unlike blk_mq_stop_hw_queues, blk_mq_quiesce_queue respects the
> submission path rcu grace. quiesce the queue before iterating
> on live tags, or performing device io quiescing.
>
> Signed-off-by: Sagi Grimberg <sagi@xxxxxxxxxxx>
> ---
> drivers/block/mtip32xx/mtip32xx.c | 8 ++++----
> 1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
> index 61b046f256ca..69a62aeace2a 100644
> --- a/drivers/block/mtip32xx/mtip32xx.c
> +++ b/drivers/block/mtip32xx/mtip32xx.c
> @@ -950,7 +950,7 @@ static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
> unsigned long to;
> bool active = true;
>
> - blk_mq_stop_hw_queues(port->dd->queue);
> + blk_mq_quiesce_queue(port->dd->queue);
>
> to = jiffies + msecs_to_jiffies(timeout);
> do {
> @@ -970,10 +970,10 @@ static int mtip_quiesce_io(struct mtip_port *port, unsigned long timeout)
> break;
> } while (time_before(jiffies, to));
>
> - blk_mq_start_stopped_hw_queues(port->dd->queue, true);
> + blk_mq_unquiesce_queue(port->dd->queue);
> return active ? -EBUSY : 0;
> err_fault:
> - blk_mq_start_stopped_hw_queues(port->dd->queue, true);
> + blk_mq_unquiesce_queue(port->dd->queue);
> return -EFAULT;
> }
The above change looks correct.
>
> @@ -3995,7 +3995,7 @@ static int mtip_block_remove(struct driver_data *dd)
> dd->disk->disk_name);
>
> blk_freeze_queue_start(dd->queue);
> - blk_mq_stop_hw_queues(dd->queue);
> + blk_mq_quiesce_queue(dd->queue);
> blk_mq_tagset_busy_iter(&dd->tags, mtip_no_dev_cleanup, dd);
We still need to unquiesce queue for avoiding hanging blk_freeze_queue()
in blk_cleanup_queue() since any new request queued during quiescing
can't be dispatched to driver/device.
There are other blk_mq_tagset_busy_iter() in mtip_service_thread(),
which looks need quiesce too. This case is even worse, because both
mtip_abort_cmd() and mtip_queue_cmd() do not check if the req is
started.
--
Ming
