Re: [PATCH v4 6/8] nvme-fc: fix controller reset hang during traffic

[James Smart]

On 8/4/2021 12:23 AM, Hannes Reinecke wrote:
> On 8/2/21 1:26 PM, Daniel Wagner wrote:
> > From: James Smart <jsmart2021@xxxxxxxxx>
> >
> > commit fe35ec58f0d3 ("block: update hctx map when use multiple maps")
> > exposed an issue where we may hang trying to wait for queue freeze
> > during I/O. We call blk_mq_update_nr_hw_queues which may attempt to freeze
> > the queue. However we never started queue freeze when starting the
> > reset, which means that we have inflight pending requests that entered the
> > queue that we will not complete once the queue is quiesced.
> >
> > So start a freeze before we quiesce the queue, and unfreeze the queue
> > after we successfully connected the I/O queues (the unfreeze is already
> > present in the code). blk_mq_update_nr_hw_queues will be called only
> > after we are sure that the queue was already frozen.
> >
> > This follows to how the pci driver handles resets.
> >
> > This patch added logic introduced in commit 9f98772ba307 "nvme-rdma: fix
> > controller reset hang during traffic".
> >
> > Signed-off-by: James Smart <jsmart2021@xxxxxxxxx>
> > CC: Sagi Grimberg <sagi@xxxxxxxxxxx>
> > [dwagner: call nvme_unfreeze() unconditionally in
> >            nvme_fc_recreate_io_queues() to match the nvme_start_freeze()]
> > Tested-by: Daniel Wagner <dwagner@xxxxxxx>
> > Reviewed-by: Daniel Wagner <dwagner@xxxxxxx>
> > ---
> >   drivers/nvme/host/fc.c | 3 ++-
> >   1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/nvme/host/fc.c b/drivers/nvme/host/fc.c
> > index 133b87db4f1d..b292af0fd655 100644
> > --- a/drivers/nvme/host/fc.c
> > +++ b/drivers/nvme/host/fc.c
> > @@ -2486,6 +2486,7 @@ __nvme_fc_abort_outstanding_ios(struct nvme_fc_ctrl *ctrl, bool start_queues)
> >   	 * (but with error status).
> >   	 */
> >   	if (ctrl->ctrl.queue_count > 1) {
> > +		nvme_start_freeze(&ctrl->ctrl);
> >   		nvme_stop_queues(&ctrl->ctrl);
> >   		nvme_sync_io_queues(&ctrl->ctrl);
> >   		blk_mq_tagset_busy_iter(&ctrl->tag_set,
> > @@ -2966,8 +2967,8 @@ nvme_fc_recreate_io_queues(struct nvme_fc_ctrl *ctrl)
> >   			return -ENODEV;
> >   		}
> >   		blk_mq_update_nr_hw_queues(&ctrl->tag_set, nr_io_queues);
> > -		nvme_unfreeze(&ctrl->ctrl);
> >   	}
> > +	nvme_unfreeze(&ctrl->ctrl);
> >   
> >   	ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.sqsize + 1);
> >   	if (ret)
> There still is now an imbalance, as we're always calling
> 'nvme_unfreeze()' (irrespective on the number of queues), but will only
> call 'nvme_start_freeze()' if we have more than one queue.
>
> This might lead to an imbalance on the mq_freeze_depth counter.
> Wouldn't it be better to move the call to 'nvme_start_freeze()' out of
> the if() condition to avoid the imbalance?
>
> Cheers,

Daniel,

try this. Moves the location of the freeze and will always pair with the 
unfreeze.

--- fc.c	2021-08-12 12:33:33.273278611 -0700
+++ fc.c.new	2021-08-12 13:01:16.185817238 -0700
@@ -2965,9 +2965,10 @@ nvme_fc_recreate_io_queues(struct nvme_f
 			prior_ioq_cnt, nr_io_queues);
 		nvme_wait_freeze(&ctrl->ctrl);
 		blk_mq_update_nr_hw_queues(&ctrl->tag_set, nr_io_queues);
-		nvme_unfreeze(&ctrl->ctrl);
 	}

+	nvme_unfreeze(&ctrl->ctrl);
+
 	return 0;

 out_delete_hw_queues:
@@ -3206,6 +3207,9 @@ nvme_fc_delete_association(struct nvme_f
 	ctrl->iocnt = 0;
 	spin_unlock_irqrestore(&ctrl->lock, flags);

+	if (ctrl->ctrl.queue_count > 1)
+		nvme_start_freeze(&ctrl->ctrl);
+
 	__nvme_fc_abort_outstanding_ios(ctrl, false);

 	/* kill the aens as they are a separate path */