[PATCH] nvme: Return BLK_STS_TARGET if the DNR bit is set

[kbusch@kernel.org (Keith Busch)]

On Tue, Aug 06, 2019@03:53:29PM +0200, Hannes Reinecke wrote:
> On 8/6/19 3:50 PM, Nadolski, Edmund wrote:
> > On 8/6/2019 5:10 AM, Hannes Reinecke wrote:
> > > If the DNR bit is set we should not retry the command, even if
> > > the standard status evaluation indicates so.
> > > 
> > > Signed-off-by: Hannes Reinecke <hare at suse.com>
> > > ---
> > > ? drivers/nvme/host/core.c | 7 +++++++
> > > ? 1 file changed, 7 insertions(+)
> > > 
> > > diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> > > index cc09b81fc7f4..c3e9254f4757 100644
> > > --- a/drivers/nvme/host/core.c
> > > +++ b/drivers/nvme/host/core.c
> > > @@ -279,6 +279,13 @@ void nvme_complete_rq(struct request *req)
> > > ????????????? return;
> > > ????????? }
> > > ????? }
> > > +??? /*
> > > +???? * Any pathing error might be retried, but the DNR bit takes
> > > +???? * precedence. So return BLK_STS_TARGET if the DNR bit is set
> > > +???? * to avoid retrying.
> > > +???? */
> > > +??? if (blk_path_error(status) && nvme_req(req)->status & NVME_SC_DNR)
> > > +??????? status = BLK_STS_TARGET;
> > > ????? blk_mq_end_request(req, status);
> > > ? }
> > > ? EXPORT_SYMBOL_GPL(nvme_complete_rq);
> > > 
> > 
> > - If DNR always takes precedence, is the blk_path_error() check still
> > needed?
> > 
> It takes precedence in the sense that it should cause the command not to be
> retried. It should not overwrite any error code indicating a non-retryable
> error.

But we're already past the nvme retry logic, which also handles the DNR
bit. Who is this telling not to retry?