Re: [PATCH] block: Fix blk_mq_try_issue_directly()

[Bart Van Assche <bvanassche@acm.org>]

On Thu, 2019-04-04 at 15:08 +0800, Ming Lei wrote:
> On Wed, Apr 03, 2019 at 01:11:26PM -0700, Bart Van Assche wrote:
> > If blk_mq_try_issue_directly() returns BLK_STS*_RESOURCE that means that
> > the request has not been queued and that the caller should retry to submit
> > the request. Both blk_mq_request_bypass_insert() and
> > blk_mq_sched_insert_request() guarantee that a request will be processed.
> > Hence return BLK_STS_OK if one of these functions is called. This patch
> > avoids that blk_mq_dispatch_rq_list() crashes when using dm-mpath.
> > 
> > Cc: Christoph Hellwig <hch@infradead.org>
> > Cc: Hannes Reinecke <hare@suse.com>
> > Cc: James Smart <james.smart@broadcom.com>
> > Cc: Ming Lei <ming.lei@redhat.com>
> > Cc: Jianchao Wang <jianchao.w.wang@oracle.com>
> > Cc: Keith Busch <keith.busch@intel.com>
> > Cc: Dongli Zhang <dongli.zhang@oracle.com>
> > Cc: Laurence Oberman <loberman@redhat.com>
> > Tested-by: Laurence Oberman <loberman@redhat.com>
> > Reviewed-by: Laurence Oberman <loberman@redhat.com>
> > Reported-by: Laurence Oberman <loberman@redhat.com>
> > Fixes: 7f556a44e61d ("blk-mq: refactor the code of issue request directly") # v5.0.
> > Cc: <stable@vger.kernel.org>
> > Signed-off-by: Bart Van Assche <bvanassche@acm.org>
> > ---
> >  block/blk-mq.c | 9 ++-------
> >  1 file changed, 2 insertions(+), 7 deletions(-)
> > 
> > diff --git a/block/blk-mq.c b/block/blk-mq.c
> > index 652d0c6d5945..b2c20dce8a30 100644
> > --- a/block/blk-mq.c
> > +++ b/block/blk-mq.c
> > @@ -1859,16 +1859,11 @@ blk_status_t blk_mq_try_issue_directly(struct blk_mq_hw_ctx *hctx,
> >  	case BLK_STS_RESOURCE:
> >  		if (force) {
> >  			blk_mq_request_bypass_insert(rq, run_queue);
> > -			/*
> > -			 * We have to return BLK_STS_OK for the DM
> > -			 * to avoid livelock. Otherwise, we return
> > -			 * the real result to indicate whether the
> > -			 * request is direct-issued successfully.
> > -			 */
> > -			ret = bypass ? BLK_STS_OK : ret;
> > +			ret = BLK_STS_OK;
> >  		} else if (!bypass) {
> >  			blk_mq_sched_insert_request(rq, false,
> >  						    run_queue, false);
> > +			ret = BLK_STS_OK;
> >  		}
> 
> This change itself is correct.
> 
> However, there is other issue introduced by 7f556a44e61d.
> 
> We need blk_insert_cloned_request() to pass back BLK_STS_RESOURCE/BLK_STS_RESOURCE
> to caller, so that dm-rq driver may see the underlying queue is busy, then tell
> blk-mq to deal with the busy condition from dm-rq queue, so that IO
> merge can get improved.
> 
> That is exactly what 396eaf21ee17c476e8f6 ("blk-mq: improve DM's blk-mq IO merging
> via blk_insert_cloned_request feedback") did.
> 
> There must be performance regression with 7f556a44e61d by cut the feedback.
> 
> So could you fix them all in one patch?

Since commit 7f556a44e61d introduced multiple problems and since fixing
these is nontrivial, how about reverting that commit?

Thanks,

Bart.