Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

[Ming Lin <mlin@kernel.org>]

On Tue, 2016-06-07 at 22:49 -0600, Jens Axboe wrote:
> On 06/06/2016 03:21 PM, Christoph Hellwig wrote:
> > From: Ming Lin <ming.l@ssi.samsung.com>
> > 
> > For some protocols like NVMe over Fabrics we need to be able to
> > send
> > initialization commands to a specific queue.
> > 
> > Based on an earlier patch from Christoph Hellwig <hch@lst.de>.
> > 
> > Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > ---
> >   block/blk-mq.c         | 33 +++++++++++++++++++++++++++++++++
> >   include/linux/blk-mq.h |  2 ++
> >   2 files changed, 35 insertions(+)
> > 
> > diff --git a/block/blk-mq.c b/block/blk-mq.c
> > index 29cbc1b..7bb45ed 100644
> > --- a/block/blk-mq.c
> > +++ b/block/blk-mq.c
> > @@ -266,6 +266,39 @@ struct request *blk_mq_alloc_request(struct
> > request_queue *q, int rw,
> >   }
> >   EXPORT_SYMBOL(blk_mq_alloc_request);
> > 
> > +struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
> > int rw,
> > +		unsigned int flags, unsigned int hctx_idx)
> > +{
> > +	struct blk_mq_hw_ctx *hctx;
> > +	struct blk_mq_ctx *ctx;
> > +	struct request *rq;
> > +	struct blk_mq_alloc_data alloc_data;
> > +	int ret;
> > +
> > +	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
> > +	if (ret)
> > +		return ERR_PTR(ret);
> > +
> > +	hctx = q->queue_hw_ctx[hctx_idx];
> > +	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
> > +
> > +	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
> > +
> > +	rq = __blk_mq_alloc_request(&alloc_data, rw);
> > +	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
> > +		__blk_mq_run_hw_queue(hctx);
> > +
> > +		rq =  __blk_mq_alloc_request(&alloc_data, rw);
> > +	}
> 
> Why are we duplicating this code here? If NOWAIT isn't set, then
> we'll
> always return a request. bt_get() will run the queue for us, if it
> needs
> to. blk_mq_alloc_request() does this too, and I'm guessing that code
> was
> just copied. I'll fix that up. Looks like this should just be:
> 
> 	rq = __blk_mq_alloc_request(&alloc_data, rw);
> 	if (rq)
> 		return rq;
> 
> 	blk_queue_exit(q);
> 	return ERR_PTR(-EWOULDBLOCK);
> 
> for this case.

Yes,

But the bt_get() reminds me that this patch actually has a problem.

blk_mq_alloc_request_hctx() ->
  __blk_mq_alloc_request() ->
    blk_mq_get_tag() -> 
      __blk_mq_get_tag() ->
        bt_get() ->
          blk_mq_put_ctx(data->ctx);

Here are blk_mq_get_ctx() and blk_mq_put_ctx().

static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
{       
        return __blk_mq_get_ctx(q, get_cpu());
} 

static inline void blk_mq_put_ctx(struct blk_mq_ctx *ctx)
{
        put_cpu();
}

blk_mq_alloc_request_hctx() calls __blk_mq_get_ctx() instead
of blk_mq_get_ctx(). Then reason is the "hctx" could belong to other
cpu. So blk_mq_get_ctx() doesn't work.

But then above put_cpu() in blk_mq_put_ctx() will trigger a WARNING
because we didn't do get_cpu() in blk_mq_alloc_request_hctx()