From mboxrd@z Thu Jan  1 00:00:00 1970
From: snitzer@redhat.com (Mike Snitzer)
Date: Mon, 4 Jun 2018 08:23:50 -0400
Subject: [PATCH 4/9] nvme: always failover on path or transport errors
In-Reply-To: <20180604062923.GA20662@lst.de>
References: <20180601071128.7630-1-hch@lst.de>
 <20180601071128.7630-5-hch@lst.de>
 <20180601152711.GA98855@redhat.com> <20180604062923.GA20662@lst.de>
Message-ID: <20180604122350.GA6478@redhat.com>

On Mon, Jun 04 2018 at  2:29P -0400,
Christoph Hellwig <hch@lst.de> wrote:

> On Fri, Jun 01, 2018@11:27:11AM -0400, Mike Snitzer wrote:
> > > @@ -59,6 +59,9 @@ bool nvme_req_needs_failover(struct request *req, blk_status_t error)
> > >  {
> > >  	if (!(req->cmd_flags & REQ_NVME_MPATH))
> > >  		return false;
> > > +	/* always failover for path related errors */
> > > +	if ((nvme_req(req)->status & 0x700) == 0x300)
> > > +		return true;
> > >  	return blk_path_error(error);
> > >  }
> > 
> > Shouldn't this go into nvme_error_status(), with a new NVME_SC if
> > needed, so that blk_path_error() just works?
> 
> NVME_SC values are defined in the NVMe spec.  The check is intended
> as a catchall for multiple of theses values, as we've defined a whole
> extensible class for path errors.

Yes, so adding NVME_SC_ANA_INACCESSIBLE, NVME_SC_ANA_PERSISTENT_LOSS,
and NVME_SC_ANA_TRANSITION to nvme_error_status().
 
> But we could probably add a BLK_STS_PATH and I'll map it to that.

And mapping them to BLK_STS_PATH and updating blk_path_error().
Sounds good.

Thanks,
Mike