[PATCH v3 0/2] virtio-blk: support polling I/O and mq_ops->queue_rqs()

[PATCH v3 0/2] virtio-blk: support polling I/O and mq_ops->queue_rqs()

[Suwan Kim]

This patch serise adds support for polling I/O and mq_ops->queue_rqs()
to virtio-blk driver.

Changes

v2 -> v3
        - Fix warning by kernel test robot
          
            static int virtblk_poll()
                ...
                if (!blk_mq_add_to_batch(req, iob, virtblk_result(vbr),
                                                   -> vbr->status,

v1 -> v2
        - To receive the number of poll queues from user,
          use module parameter instead of QEMU uapi change.

        - Add the comment about virtblk_map_queues().

        - Add support for mq_ops->queue_rqs() to implement submit side
          batch.

Suwan Kim (2):
  virtio-blk: support polling I/O
  virtio-blk: support mq_ops->queue_rqs()

 drivers/block/virtio_blk.c | 194 ++++++++++++++++++++++++++++++++++---
 1 file changed, 181 insertions(+), 13 deletions(-)

-- 
2.26.3

[PATCH v3 1/2] virtio-blk: support polling I/O

[Suwan Kim]

This patch supports polling I/O via virtio-blk driver. Polling
feature is enabled by module parameter "num_poll_queues" and it
sets dedicated polling queues for virtio-blk. This patch improves
the polling I/O throughput and latency.

The virtio-blk driver doesn't not have a poll function and a poll
queue and it has been operating in interrupt driven method even if
the polling function is called in the upper layer.

virtio-blk polling is implemented upon 'batched completion' of block
layer. virtblk_poll() queues completed request to io_comp_batch->req_list
and later, virtblk_complete_batch() calls unmap function and ends
the requests in batch.

virtio-blk reads the number of poll queues from module parameter
"num_poll_queues". If VM sets queue parameter as below,
("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
queues, the poll queues have no callback function.

Regarding HW-SW queue mapping, the default queue mapping uses the
existing method that condsiders MSI irq vector. But the poll queue
doesn't have an irq, so it uses the regular blk-mq cpu mapping.

For verifying the improvement, I did Fio polling I/O performance test
with io_uring engine with the options below.
(io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
queues for VM.

As a result, IOPS and average latency improved about 10%.

Test result:

- Fio io_uring poll without virtio-blk poll support
	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us

- Fio io_uring poll with virtio-blk poll support
	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
---
 drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
 1 file changed, 97 insertions(+), 4 deletions(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 8c415be86732..3d16f8b753e7 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
 		 "0 for no limit. "
 		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
 
+static unsigned int num_poll_queues;
+module_param(num_poll_queues, uint, 0644);
+MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
+
 static int major;
 static DEFINE_IDA(vd_index_ida);
 
@@ -81,6 +85,7 @@ struct virtio_blk {
 
 	/* num of vqs */
 	int num_vqs;
+	int io_queues[HCTX_MAX_TYPES];
 	struct virtio_blk_vq *vqs;
 };
 
@@ -548,6 +553,7 @@ static int init_vq(struct virtio_blk *vblk)
 	const char **names;
 	struct virtqueue **vqs;
 	unsigned short num_vqs;
+	unsigned int num_poll_vqs;
 	struct virtio_device *vdev = vblk->vdev;
 	struct irq_affinity desc = { 0, };
 
@@ -556,6 +562,7 @@ static int init_vq(struct virtio_blk *vblk)
 				   &num_vqs);
 	if (err)
 		num_vqs = 1;
+
 	if (!err && !num_vqs) {
 		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
 		return -EINVAL;
@@ -565,6 +572,13 @@ static int init_vq(struct virtio_blk *vblk)
 			min_not_zero(num_request_queues, nr_cpu_ids),
 			num_vqs);
 
+	num_poll_vqs = min_t(unsigned int, num_poll_queues, num_vqs - 1);
+
+	memset(vblk->io_queues, 0, sizeof(int) * HCTX_MAX_TYPES);
+	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
+	vblk->io_queues[HCTX_TYPE_READ] = 0;
+	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
+
 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 	if (!vblk->vqs)
 		return -ENOMEM;
@@ -578,8 +592,13 @@ static int init_vq(struct virtio_blk *vblk)
 	}
 
 	for (i = 0; i < num_vqs; i++) {
-		callbacks[i] = virtblk_done;
-		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
+		if (i < num_vqs - num_poll_vqs) {
+			callbacks[i] = virtblk_done;
+			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
+		} else {
+			callbacks[i] = NULL;
+			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
+		}
 		names[i] = vblk->vqs[i].name;
 	}
 
@@ -728,16 +747,87 @@ static const struct attribute_group *virtblk_attr_groups[] = {
 static int virtblk_map_queues(struct blk_mq_tag_set *set)
 {
 	struct virtio_blk *vblk = set->driver_data;
+	int i, qoff;
+
+	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
+		struct blk_mq_queue_map *map = &set->map[i];
+
+		map->nr_queues = vblk->io_queues[i];
+		map->queue_offset = qoff;
+		qoff += map->nr_queues;
+
+		if (map->nr_queues == 0)
+			continue;
+
+		/*
+		 * Regular queues have interrupts and hence CPU affinity is
+		 * defined by the core virtio code, but polling queues have
+		 * no interrupts so we let the block layer assign CPU affinity.
+		 */
+		if (i == HCTX_TYPE_DEFAULT)
+			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
+		else
+			blk_mq_map_queues(&set->map[i]);
+	}
+
+	return 0;
+}
+
+static void virtblk_complete_batch(struct io_comp_batch *iob)
+{
+	struct request *req;
+	struct virtblk_req *vbr;
 
-	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
-					vblk->vdev, 0);
+	rq_list_for_each(&iob->req_list, req) {
+		vbr = blk_mq_rq_to_pdu(req);
+		virtblk_unmap_data(req, vbr);
+		virtblk_cleanup_cmd(req);
+	}
+	blk_mq_end_request_batch(iob);
+}
+
+static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
+{
+	struct virtio_blk_vq *vq = hctx->driver_data;
+	struct virtblk_req *vbr;
+	unsigned long flags;
+	unsigned int len;
+	int found = 0;
+
+	spin_lock_irqsave(&vq->lock, flags);
+
+	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
+		struct request *req = blk_mq_rq_from_pdu(vbr);
+
+		found++;
+		if (!blk_mq_add_to_batch(req, iob, vbr->status,
+						virtblk_complete_batch))
+			blk_mq_complete_request(req);
+	}
+
+	spin_unlock_irqrestore(&vq->lock, flags);
+
+	return found;
+}
+
+static int virtblk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
+			  unsigned int hctx_idx)
+{
+	struct virtio_blk *vblk = data;
+	struct virtio_blk_vq *vq = &vblk->vqs[hctx_idx];
+
+	WARN_ON(vblk->tag_set.tags[hctx_idx] != hctx->tags);
+	hctx->driver_data = vq;
+	return 0;
 }
 
 static const struct blk_mq_ops virtio_mq_ops = {
 	.queue_rq	= virtio_queue_rq,
 	.commit_rqs	= virtio_commit_rqs,
+	.init_hctx	= virtblk_init_hctx,
 	.complete	= virtblk_request_done,
 	.map_queues	= virtblk_map_queues,
+	.poll		= virtblk_poll,
 };
 
 static unsigned int virtblk_queue_depth;
@@ -816,6 +906,9 @@ static int virtblk_probe(struct virtio_device *vdev)
 		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
 	vblk->tag_set.driver_data = vblk;
 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
+	vblk->tag_set.nr_maps = 1;
+	if (vblk->io_queues[HCTX_TYPE_POLL])
+		vblk->tag_set.nr_maps = 3;
 
 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
 	if (err)
-- 
2.26.3

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Michael S. Tsirkin]

On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> This patch supports polling I/O via virtio-blk driver. Polling
> feature is enabled by module parameter "num_poll_queues" and it
> sets dedicated polling queues for virtio-blk. This patch improves
> the polling I/O throughput and latency.
> 
> The virtio-blk driver doesn't not have a poll function and a poll
> queue and it has been operating in interrupt driven method even if
> the polling function is called in the upper layer.
> 
> virtio-blk polling is implemented upon 'batched completion' of block
> layer. virtblk_poll() queues completed request to io_comp_batch->req_list
> and later, virtblk_complete_batch() calls unmap function and ends
> the requests in batch.
> 
> virtio-blk reads the number of poll queues from module parameter
> "num_poll_queues". If VM sets queue parameter as below,
> ("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
> It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
> as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
> queues, the poll queues have no callback function.
> 
> Regarding HW-SW queue mapping, the default queue mapping uses the
> existing method that condsiders MSI irq vector. But the poll queue
> doesn't have an irq, so it uses the regular blk-mq cpu mapping.
> 
> For verifying the improvement, I did Fio polling I/O performance test
> with io_uring engine with the options below.
> (io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
> I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
> queues for VM.
> 
> As a result, IOPS and average latency improved about 10%.
> 
> Test result:
> 
> - Fio io_uring poll without virtio-blk poll support
> 	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
> 	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
> 	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us
> 
> - Fio io_uring poll with virtio-blk poll support
> 	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
> 	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
> 	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us
> 
> Reported-by: kernel test robot <lkp@intel.com>
> Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> ---
>  drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
>  1 file changed, 97 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 8c415be86732..3d16f8b753e7 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
>  		 "0 for no limit. "
>  		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
>  
> +static unsigned int num_poll_queues;
> +module_param(num_poll_queues, uint, 0644);
> +MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
> +
>  static int major;
>  static DEFINE_IDA(vd_index_ida);
>

Is there some way to make it work reasonably without need to set
module parameters? I don't see any other devices with a num_poll_queues
parameter - how do they handle this?
  
> @@ -81,6 +85,7 @@ struct virtio_blk {
>  
>  	/* num of vqs */
>  	int num_vqs;
> +	int io_queues[HCTX_MAX_TYPES];
>  	struct virtio_blk_vq *vqs;
>  };
>  
> @@ -548,6 +553,7 @@ static int init_vq(struct virtio_blk *vblk)
>  	const char **names;
>  	struct virtqueue **vqs;
>  	unsigned short num_vqs;
> +	unsigned int num_poll_vqs;
>  	struct virtio_device *vdev = vblk->vdev;
>  	struct irq_affinity desc = { 0, };
>  
> @@ -556,6 +562,7 @@ static int init_vq(struct virtio_blk *vblk)
>  				   &num_vqs);
>  	if (err)
>  		num_vqs = 1;
> +
>  	if (!err && !num_vqs) {
>  		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
>  		return -EINVAL;
> @@ -565,6 +572,13 @@ static int init_vq(struct virtio_blk *vblk)
>  			min_not_zero(num_request_queues, nr_cpu_ids),
>  			num_vqs);
>  
> +	num_poll_vqs = min_t(unsigned int, num_poll_queues, num_vqs - 1);
> +
> +	memset(vblk->io_queues, 0, sizeof(int) * HCTX_MAX_TYPES);
> +	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
> +	vblk->io_queues[HCTX_TYPE_READ] = 0;
> +	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
> +
>  	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
>  	if (!vblk->vqs)
>  		return -ENOMEM;
> @@ -578,8 +592,13 @@ static int init_vq(struct virtio_blk *vblk)
>  	}
>  
>  	for (i = 0; i < num_vqs; i++) {
> -		callbacks[i] = virtblk_done;
> -		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> +		if (i < num_vqs - num_poll_vqs) {
> +			callbacks[i] = virtblk_done;
> +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> +		} else {
> +			callbacks[i] = NULL;
> +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
> +		}
>  		names[i] = vblk->vqs[i].name;
>  	}
>  
> @@ -728,16 +747,87 @@ static const struct attribute_group *virtblk_attr_groups[] = {
>  static int virtblk_map_queues(struct blk_mq_tag_set *set)
>  {
>  	struct virtio_blk *vblk = set->driver_data;
> +	int i, qoff;
> +
> +	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
> +		struct blk_mq_queue_map *map = &set->map[i];
> +
> +		map->nr_queues = vblk->io_queues[i];
> +		map->queue_offset = qoff;
> +		qoff += map->nr_queues;
> +
> +		if (map->nr_queues == 0)
> +			continue;
> +
> +		/*
> +		 * Regular queues have interrupts and hence CPU affinity is
> +		 * defined by the core virtio code, but polling queues have
> +		 * no interrupts so we let the block layer assign CPU affinity.
> +		 */
> +		if (i == HCTX_TYPE_DEFAULT)
> +			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
> +		else
> +			blk_mq_map_queues(&set->map[i]);
> +	}
> +
> +	return 0;
> +}
> +
> +static void virtblk_complete_batch(struct io_comp_batch *iob)
> +{
> +	struct request *req;
> +	struct virtblk_req *vbr;
>  
> -	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
> -					vblk->vdev, 0);
> +	rq_list_for_each(&iob->req_list, req) {
> +		vbr = blk_mq_rq_to_pdu(req);
> +		virtblk_unmap_data(req, vbr);
> +		virtblk_cleanup_cmd(req);
> +	}
> +	blk_mq_end_request_batch(iob);
> +}
> +
> +static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
> +{
> +	struct virtio_blk_vq *vq = hctx->driver_data;
> +	struct virtblk_req *vbr;
> +	unsigned long flags;
> +	unsigned int len;
> +	int found = 0;
> +
> +	spin_lock_irqsave(&vq->lock, flags);
> +
> +	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> +		struct request *req = blk_mq_rq_from_pdu(vbr);
> +
> +		found++;
> +		if (!blk_mq_add_to_batch(req, iob, vbr->status,
> +						virtblk_complete_batch))
> +			blk_mq_complete_request(req);
> +	}
> +
> +	spin_unlock_irqrestore(&vq->lock, flags);
> +
> +	return found;
> +}
> +
> +static int virtblk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
> +			  unsigned int hctx_idx)
> +{
> +	struct virtio_blk *vblk = data;
> +	struct virtio_blk_vq *vq = &vblk->vqs[hctx_idx];
> +
> +	WARN_ON(vblk->tag_set.tags[hctx_idx] != hctx->tags);
> +	hctx->driver_data = vq;
> +	return 0;
>  }
>  
>  static const struct blk_mq_ops virtio_mq_ops = {
>  	.queue_rq	= virtio_queue_rq,
>  	.commit_rqs	= virtio_commit_rqs,
> +	.init_hctx	= virtblk_init_hctx,
>  	.complete	= virtblk_request_done,
>  	.map_queues	= virtblk_map_queues,
> +	.poll		= virtblk_poll,
>  };
>  
>  static unsigned int virtblk_queue_depth;
> @@ -816,6 +906,9 @@ static int virtblk_probe(struct virtio_device *vdev)
>  		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
>  	vblk->tag_set.driver_data = vblk;
>  	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
> +	vblk->tag_set.nr_maps = 1;
> +	if (vblk->io_queues[HCTX_TYPE_POLL])
> +		vblk->tag_set.nr_maps = 3;
>  
>  	err = blk_mq_alloc_tag_set(&vblk->tag_set);
>  	if (err)
> -- 
> 2.26.3

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Suwan Kim]

On Thu, Mar 24, 2022 at 10:32:02AM -0400, Michael S. Tsirkin wrote:
> On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> > This patch supports polling I/O via virtio-blk driver. Polling
> > feature is enabled by module parameter "num_poll_queues" and it
> > sets dedicated polling queues for virtio-blk. This patch improves
> > the polling I/O throughput and latency.
> > 
> > The virtio-blk driver doesn't not have a poll function and a poll
> > queue and it has been operating in interrupt driven method even if
> > the polling function is called in the upper layer.
> > 
> > virtio-blk polling is implemented upon 'batched completion' of block
> > layer. virtblk_poll() queues completed request to io_comp_batch->req_list
> > and later, virtblk_complete_batch() calls unmap function and ends
> > the requests in batch.
> > 
> > virtio-blk reads the number of poll queues from module parameter
> > "num_poll_queues". If VM sets queue parameter as below,
> > ("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
> > It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
> > as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
> > queues, the poll queues have no callback function.
> > 
> > Regarding HW-SW queue mapping, the default queue mapping uses the
> > existing method that condsiders MSI irq vector. But the poll queue
> > doesn't have an irq, so it uses the regular blk-mq cpu mapping.
> > 
> > For verifying the improvement, I did Fio polling I/O performance test
> > with io_uring engine with the options below.
> > (io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
> > I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
> > queues for VM.
> > 
> > As a result, IOPS and average latency improved about 10%.
> > 
> > Test result:
> > 
> > - Fio io_uring poll without virtio-blk poll support
> > 	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
> > 	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
> > 	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us
> > 
> > - Fio io_uring poll with virtio-blk poll support
> > 	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
> > 	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
> > 	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us
> > 
> > Reported-by: kernel test robot <lkp@intel.com>
> > Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> > ---
> >  drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
> >  1 file changed, 97 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> > index 8c415be86732..3d16f8b753e7 100644
> > --- a/drivers/block/virtio_blk.c
> > +++ b/drivers/block/virtio_blk.c
> > @@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
> >  		 "0 for no limit. "
> >  		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
> >  
> > +static unsigned int num_poll_queues;
> > +module_param(num_poll_queues, uint, 0644);
> > +MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
> > +
> >  static int major;
> >  static DEFINE_IDA(vd_index_ida);
> >
> 
> Is there some way to make it work reasonably without need to set
> module parameters? I don't see any other devices with a num_poll_queues
> parameter - how do they handle this?

Hi Michael,

NVMe driver uses module parameter.

Please refer to this.
-----
drivers/nvme/host/pci.c

static unsigned int poll_queues;
module_param_cb(poll_queues, &io_queue_count_ops, &poll_queues, 0644);
MODULE_PARM_DESC(poll_queues, "Number of queues to use for polled IO.");
-----

Regards,
Suwan Kim

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Michael S. Tsirkin]

On Thu, Mar 24, 2022 at 11:46:02PM +0900, Suwan Kim wrote:
> On Thu, Mar 24, 2022 at 10:32:02AM -0400, Michael S. Tsirkin wrote:
> > On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> > > This patch supports polling I/O via virtio-blk driver. Polling
> > > feature is enabled by module parameter "num_poll_queues" and it
> > > sets dedicated polling queues for virtio-blk. This patch improves
> > > the polling I/O throughput and latency.
> > > 
> > > The virtio-blk driver doesn't not have a poll function and a poll
> > > queue and it has been operating in interrupt driven method even if
> > > the polling function is called in the upper layer.
> > > 
> > > virtio-blk polling is implemented upon 'batched completion' of block
> > > layer. virtblk_poll() queues completed request to io_comp_batch->req_list
> > > and later, virtblk_complete_batch() calls unmap function and ends
> > > the requests in batch.
> > > 
> > > virtio-blk reads the number of poll queues from module parameter
> > > "num_poll_queues". If VM sets queue parameter as below,
> > > ("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
> > > It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
> > > as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
> > > queues, the poll queues have no callback function.
> > > 
> > > Regarding HW-SW queue mapping, the default queue mapping uses the
> > > existing method that condsiders MSI irq vector. But the poll queue
> > > doesn't have an irq, so it uses the regular blk-mq cpu mapping.
> > > 
> > > For verifying the improvement, I did Fio polling I/O performance test
> > > with io_uring engine with the options below.
> > > (io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
> > > I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
> > > queues for VM.
> > > 
> > > As a result, IOPS and average latency improved about 10%.
> > > 
> > > Test result:
> > > 
> > > - Fio io_uring poll without virtio-blk poll support
> > > 	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
> > > 	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
> > > 	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us
> > > 
> > > - Fio io_uring poll with virtio-blk poll support
> > > 	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
> > > 	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
> > > 	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us
> > > 
> > > Reported-by: kernel test robot <lkp@intel.com>
> > > Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> > > ---
> > >  drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
> > >  1 file changed, 97 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> > > index 8c415be86732..3d16f8b753e7 100644
> > > --- a/drivers/block/virtio_blk.c
> > > +++ b/drivers/block/virtio_blk.c
> > > @@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
> > >  		 "0 for no limit. "
> > >  		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
> > >  
> > > +static unsigned int num_poll_queues;
> > > +module_param(num_poll_queues, uint, 0644);
> > > +MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
> > > +
> > >  static int major;
> > >  static DEFINE_IDA(vd_index_ida);
> > >
> > 
> > Is there some way to make it work reasonably without need to set
> > module parameters? I don't see any other devices with a num_poll_queues
> > parameter - how do they handle this?
> 
> Hi Michael,
> 
> NVMe driver uses module parameter.
> 
> Please refer to this.
> -----
> drivers/nvme/host/pci.c
> 
> static unsigned int poll_queues;
> module_param_cb(poll_queues, &io_queue_count_ops, &poll_queues, 0644);
> MODULE_PARM_DESC(poll_queues, "Number of queues to use for polled IO.");
> -----
> 
> Regards,
> Suwan Kim

OK then. Let's maybe be consistent wrt parameter naming?

-- 
MST

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Suwan Kim]

On Thu, Mar 24, 2022 at 01:56:18PM -0400, Michael S. Tsirkin wrote:
> On Thu, Mar 24, 2022 at 11:46:02PM +0900, Suwan Kim wrote:
> > On Thu, Mar 24, 2022 at 10:32:02AM -0400, Michael S. Tsirkin wrote:
> > > On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> > > > This patch supports polling I/O via virtio-blk driver. Polling
> > > > feature is enabled by module parameter "num_poll_queues" and it
> > > > sets dedicated polling queues for virtio-blk. This patch improves
> > > > the polling I/O throughput and latency.
> > > > 
> > > > The virtio-blk driver doesn't not have a poll function and a poll
> > > > queue and it has been operating in interrupt driven method even if
> > > > the polling function is called in the upper layer.
> > > > 
> > > > virtio-blk polling is implemented upon 'batched completion' of block
> > > > layer. virtblk_poll() queues completed request to io_comp_batch->req_list
> > > > and later, virtblk_complete_batch() calls unmap function and ends
> > > > the requests in batch.
> > > > 
> > > > virtio-blk reads the number of poll queues from module parameter
> > > > "num_poll_queues". If VM sets queue parameter as below,
> > > > ("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
> > > > It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
> > > > as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
> > > > queues, the poll queues have no callback function.
> > > > 
> > > > Regarding HW-SW queue mapping, the default queue mapping uses the
> > > > existing method that condsiders MSI irq vector. But the poll queue
> > > > doesn't have an irq, so it uses the regular blk-mq cpu mapping.
> > > > 
> > > > For verifying the improvement, I did Fio polling I/O performance test
> > > > with io_uring engine with the options below.
> > > > (io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
> > > > I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
> > > > queues for VM.
> > > > 
> > > > As a result, IOPS and average latency improved about 10%.
> > > > 
> > > > Test result:
> > > > 
> > > > - Fio io_uring poll without virtio-blk poll support
> > > > 	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
> > > > 	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
> > > > 	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us
> > > > 
> > > > - Fio io_uring poll with virtio-blk poll support
> > > > 	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
> > > > 	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
> > > > 	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us
> > > > 
> > > > Reported-by: kernel test robot <lkp@intel.com>
> > > > Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> > > > ---
> > > >  drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
> > > >  1 file changed, 97 insertions(+), 4 deletions(-)
> > > > 
> > > > diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> > > > index 8c415be86732..3d16f8b753e7 100644
> > > > --- a/drivers/block/virtio_blk.c
> > > > +++ b/drivers/block/virtio_blk.c
> > > > @@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
> > > >  		 "0 for no limit. "
> > > >  		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
> > > >  
> > > > +static unsigned int num_poll_queues;
> > > > +module_param(num_poll_queues, uint, 0644);
> > > > +MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
> > > > +
> > > >  static int major;
> > > >  static DEFINE_IDA(vd_index_ida);
> > > >
> > > 
> > > Is there some way to make it work reasonably without need to set
> > > module parameters? I don't see any other devices with a num_poll_queues
> > > parameter - how do they handle this?
> > 
> > Hi Michael,
> > 
> > NVMe driver uses module parameter.
> > 
> > Please refer to this.
> > -----
> > drivers/nvme/host/pci.c
> > 
> > static unsigned int poll_queues;
> > module_param_cb(poll_queues, &io_queue_count_ops, &poll_queues, 0644);
> > MODULE_PARM_DESC(poll_queues, "Number of queues to use for polled IO.");
> > -----
> > 
> > Regards,
> > Suwan Kim
> 
> OK then. Let's maybe be consistent wrt parameter naming?
 
Ok. Consistent naming scheme seems to be better for code readability.
I will rename it to 'poll_queues' in next version.

Regards,
Suwan Kim

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Dongli Zhang]

Hi Suwan,

The NVMe prints something like below by nvme_setup_io_queues() to confirm
if the configuration takes effect.

"[ 0.620458] nvme nvme0: 4/0/0 default/read/poll queues".

How about to print in virtio-blk as well?

Thank you very much!

Dongli Zhang


On 3/24/22 7:04 AM, Suwan Kim wrote:
> This patch supports polling I/O via virtio-blk driver. Polling
> feature is enabled by module parameter "num_poll_queues" and it
> sets dedicated polling queues for virtio-blk. This patch improves
> the polling I/O throughput and latency.
> 
> The virtio-blk driver doesn't not have a poll function and a poll
> queue and it has been operating in interrupt driven method even if
> the polling function is called in the upper layer.
> 
> virtio-blk polling is implemented upon 'batched completion' of block
> layer. virtblk_poll() queues completed request to io_comp_batch->req_list
> and later, virtblk_complete_batch() calls unmap function and ends
> the requests in batch.
> 
> virtio-blk reads the number of poll queues from module parameter
> "num_poll_queues". If VM sets queue parameter as below,
> ("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
> It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
> as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
> queues, the poll queues have no callback function.
> 
> Regarding HW-SW queue mapping, the default queue mapping uses the
> existing method that condsiders MSI irq vector. But the poll queue
> doesn't have an irq, so it uses the regular blk-mq cpu mapping.
> 
> For verifying the improvement, I did Fio polling I/O performance test
> with io_uring engine with the options below.
> (io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
> I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
> queues for VM.
> 
> As a result, IOPS and average latency improved about 10%.
> 
> Test result:
> 
> - Fio io_uring poll without virtio-blk poll support
> 	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
> 	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
> 	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us
> 
> - Fio io_uring poll with virtio-blk poll support
> 	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
> 	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
> 	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us
> 
> Reported-by: kernel test robot <lkp@intel.com>
> Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> ---
>  drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
>  1 file changed, 97 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 8c415be86732..3d16f8b753e7 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
>  		 "0 for no limit. "
>  		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
>  
> +static unsigned int num_poll_queues;
> +module_param(num_poll_queues, uint, 0644);
> +MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
> +
>  static int major;
>  static DEFINE_IDA(vd_index_ida);
>  
> @@ -81,6 +85,7 @@ struct virtio_blk {
>  
>  	/* num of vqs */
>  	int num_vqs;
> +	int io_queues[HCTX_MAX_TYPES];
>  	struct virtio_blk_vq *vqs;
>  };
>  
> @@ -548,6 +553,7 @@ static int init_vq(struct virtio_blk *vblk)
>  	const char **names;
>  	struct virtqueue **vqs;
>  	unsigned short num_vqs;
> +	unsigned int num_poll_vqs;
>  	struct virtio_device *vdev = vblk->vdev;
>  	struct irq_affinity desc = { 0, };
>  
> @@ -556,6 +562,7 @@ static int init_vq(struct virtio_blk *vblk)
>  				   &num_vqs);
>  	if (err)
>  		num_vqs = 1;
> +
>  	if (!err && !num_vqs) {
>  		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
>  		return -EINVAL;
> @@ -565,6 +572,13 @@ static int init_vq(struct virtio_blk *vblk)
>  			min_not_zero(num_request_queues, nr_cpu_ids),
>  			num_vqs);
>  
> +	num_poll_vqs = min_t(unsigned int, num_poll_queues, num_vqs - 1);
> +
> +	memset(vblk->io_queues, 0, sizeof(int) * HCTX_MAX_TYPES);
> +	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
> +	vblk->io_queues[HCTX_TYPE_READ] = 0;
> +	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
> +
>  	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
>  	if (!vblk->vqs)
>  		return -ENOMEM;
> @@ -578,8 +592,13 @@ static int init_vq(struct virtio_blk *vblk)
>  	}
>  
>  	for (i = 0; i < num_vqs; i++) {
> -		callbacks[i] = virtblk_done;
> -		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> +		if (i < num_vqs - num_poll_vqs) {
> +			callbacks[i] = virtblk_done;
> +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> +		} else {
> +			callbacks[i] = NULL;
> +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
> +		}
>  		names[i] = vblk->vqs[i].name;
>  	}
>  
> @@ -728,16 +747,87 @@ static const struct attribute_group *virtblk_attr_groups[] = {
>  static int virtblk_map_queues(struct blk_mq_tag_set *set)
>  {
>  	struct virtio_blk *vblk = set->driver_data;
> +	int i, qoff;
> +
> +	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
> +		struct blk_mq_queue_map *map = &set->map[i];
> +
> +		map->nr_queues = vblk->io_queues[i];
> +		map->queue_offset = qoff;
> +		qoff += map->nr_queues;
> +
> +		if (map->nr_queues == 0)
> +			continue;
> +
> +		/*
> +		 * Regular queues have interrupts and hence CPU affinity is
> +		 * defined by the core virtio code, but polling queues have
> +		 * no interrupts so we let the block layer assign CPU affinity.
> +		 */
> +		if (i == HCTX_TYPE_DEFAULT)
> +			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
> +		else
> +			blk_mq_map_queues(&set->map[i]);
> +	}
> +
> +	return 0;
> +}
> +
> +static void virtblk_complete_batch(struct io_comp_batch *iob)
> +{
> +	struct request *req;
> +	struct virtblk_req *vbr;
>  
> -	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
> -					vblk->vdev, 0);
> +	rq_list_for_each(&iob->req_list, req) {
> +		vbr = blk_mq_rq_to_pdu(req);
> +		virtblk_unmap_data(req, vbr);
> +		virtblk_cleanup_cmd(req);
> +	}
> +	blk_mq_end_request_batch(iob);
> +}
> +
> +static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
> +{
> +	struct virtio_blk_vq *vq = hctx->driver_data;
> +	struct virtblk_req *vbr;
> +	unsigned long flags;
> +	unsigned int len;
> +	int found = 0;
> +
> +	spin_lock_irqsave(&vq->lock, flags);
> +
> +	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> +		struct request *req = blk_mq_rq_from_pdu(vbr);
> +
> +		found++;
> +		if (!blk_mq_add_to_batch(req, iob, vbr->status,
> +						virtblk_complete_batch))
> +			blk_mq_complete_request(req);
> +	}
> +
> +	spin_unlock_irqrestore(&vq->lock, flags);
> +
> +	return found;
> +}
> +
> +static int virtblk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
> +			  unsigned int hctx_idx)
> +{
> +	struct virtio_blk *vblk = data;
> +	struct virtio_blk_vq *vq = &vblk->vqs[hctx_idx];
> +
> +	WARN_ON(vblk->tag_set.tags[hctx_idx] != hctx->tags);
> +	hctx->driver_data = vq;
> +	return 0;
>  }
>  
>  static const struct blk_mq_ops virtio_mq_ops = {
>  	.queue_rq	= virtio_queue_rq,
>  	.commit_rqs	= virtio_commit_rqs,
> +	.init_hctx	= virtblk_init_hctx,
>  	.complete	= virtblk_request_done,
>  	.map_queues	= virtblk_map_queues,
> +	.poll		= virtblk_poll,
>  };
>  
>  static unsigned int virtblk_queue_depth;
> @@ -816,6 +906,9 @@ static int virtblk_probe(struct virtio_device *vdev)
>  		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
>  	vblk->tag_set.driver_data = vblk;
>  	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
> +	vblk->tag_set.nr_maps = 1;
> +	if (vblk->io_queues[HCTX_TYPE_POLL])
> +		vblk->tag_set.nr_maps = 3;
>  
>  	err = blk_mq_alloc_tag_set(&vblk->tag_set);
>  	if (err)

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Suwan Kim]

On Thu, Mar 24, 2022 at 10:34:04AM -0700, Dongli Zhang wrote:
> Hi Suwan,
> 
> The NVMe prints something like below by nvme_setup_io_queues() to confirm
> if the configuration takes effect.
> 
> "[ 0.620458] nvme nvme0: 4/0/0 default/read/poll queues".
> 
> How about to print in virtio-blk as well?

Hi Dongli,

Thansk for your feedback. It is good idea.
I will add it in next version.

Regards,
Suwan Kim

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Michael S. Tsirkin]

On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> This patch supports polling I/O via virtio-blk driver. Polling
> feature is enabled by module parameter "num_poll_queues" and it
> sets dedicated polling queues for virtio-blk. This patch improves
> the polling I/O throughput and latency.
> 
> The virtio-blk driver doesn't not have a poll function and a poll
> queue and it has been operating in interrupt driven method even if
> the polling function is called in the upper layer.
> 
> virtio-blk polling is implemented upon 'batched completion' of block
> layer. virtblk_poll() queues completed request to io_comp_batch->req_list
> and later, virtblk_complete_batch() calls unmap function and ends
> the requests in batch.
> 
> virtio-blk reads the number of poll queues from module parameter
> "num_poll_queues". If VM sets queue parameter as below,
> ("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
> It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
> as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
> queues, the poll queues have no callback function.
> 
> Regarding HW-SW queue mapping, the default queue mapping uses the
> existing method that condsiders MSI irq vector. But the poll queue
> doesn't have an irq, so it uses the regular blk-mq cpu mapping.
> 
> For verifying the improvement, I did Fio polling I/O performance test
> with io_uring engine with the options below.
> (io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
> I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
> queues for VM.
> 
> As a result, IOPS and average latency improved about 10%.
> 
> Test result:
> 
> - Fio io_uring poll without virtio-blk poll support
> 	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
> 	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
> 	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us
> 
> - Fio io_uring poll with virtio-blk poll support
> 	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
> 	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
> 	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us
> 
> Reported-by: kernel test robot <lkp@intel.com>
> Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> ---
>  drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
>  1 file changed, 97 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 8c415be86732..3d16f8b753e7 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
>  		 "0 for no limit. "
>  		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
>  
> +static unsigned int num_poll_queues;
> +module_param(num_poll_queues, uint, 0644);
> +MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
> +
>  static int major;
>  static DEFINE_IDA(vd_index_ida);
>  
> @@ -81,6 +85,7 @@ struct virtio_blk {
>  
>  	/* num of vqs */
>  	int num_vqs;
> +	int io_queues[HCTX_MAX_TYPES];
>  	struct virtio_blk_vq *vqs;
>  };
>  
> @@ -548,6 +553,7 @@ static int init_vq(struct virtio_blk *vblk)
>  	const char **names;
>  	struct virtqueue **vqs;
>  	unsigned short num_vqs;
> +	unsigned int num_poll_vqs;
>  	struct virtio_device *vdev = vblk->vdev;
>  	struct irq_affinity desc = { 0, };
>  
> @@ -556,6 +562,7 @@ static int init_vq(struct virtio_blk *vblk)
>  				   &num_vqs);
>  	if (err)
>  		num_vqs = 1;
> +
>  	if (!err && !num_vqs) {
>  		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
>  		return -EINVAL;
> @@ -565,6 +572,13 @@ static int init_vq(struct virtio_blk *vblk)
>  			min_not_zero(num_request_queues, nr_cpu_ids),
>  			num_vqs);
>  
> +	num_poll_vqs = min_t(unsigned int, num_poll_queues, num_vqs - 1);
> +
> +	memset(vblk->io_queues, 0, sizeof(int) * HCTX_MAX_TYPES);
> +	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
> +	vblk->io_queues[HCTX_TYPE_READ] = 0;
> +	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
> +
>  	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
>  	if (!vblk->vqs)
>  		return -ENOMEM;
> @@ -578,8 +592,13 @@ static int init_vq(struct virtio_blk *vblk)
>  	}
>  
>  	for (i = 0; i < num_vqs; i++) {
> -		callbacks[i] = virtblk_done;
> -		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> +		if (i < num_vqs - num_poll_vqs) {
> +			callbacks[i] = virtblk_done;
> +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> +		} else {
> +			callbacks[i] = NULL;
> +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
> +		}
>  		names[i] = vblk->vqs[i].name;
>  	}
>  
> @@ -728,16 +747,87 @@ static const struct attribute_group *virtblk_attr_groups[] = {
>  static int virtblk_map_queues(struct blk_mq_tag_set *set)
>  {
>  	struct virtio_blk *vblk = set->driver_data;
> +	int i, qoff;
> +
> +	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
> +		struct blk_mq_queue_map *map = &set->map[i];
> +
> +		map->nr_queues = vblk->io_queues[i];
> +		map->queue_offset = qoff;
> +		qoff += map->nr_queues;
> +
> +		if (map->nr_queues == 0)
> +			continue;
> +
> +		/*
> +		 * Regular queues have interrupts and hence CPU affinity is
> +		 * defined by the core virtio code, but polling queues have
> +		 * no interrupts so we let the block layer assign CPU affinity.
> +		 */
> +		if (i == HCTX_TYPE_DEFAULT)
> +			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
> +		else
> +			blk_mq_map_queues(&set->map[i]);
> +	}
> +
> +	return 0;
> +}
> +
> +static void virtblk_complete_batch(struct io_comp_batch *iob)
> +{
> +	struct request *req;
> +	struct virtblk_req *vbr;
>  
> -	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
> -					vblk->vdev, 0);
> +	rq_list_for_each(&iob->req_list, req) {
> +		vbr = blk_mq_rq_to_pdu(req);
> +		virtblk_unmap_data(req, vbr);
> +		virtblk_cleanup_cmd(req);
> +	}
> +	blk_mq_end_request_batch(iob);
> +}
> +
> +static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
> +{
> +	struct virtio_blk_vq *vq = hctx->driver_data;
> +	struct virtblk_req *vbr;
> +	unsigned long flags;
> +	unsigned int len;
> +	int found = 0;
> +
> +	spin_lock_irqsave(&vq->lock, flags);
> +
> +	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> +		struct request *req = blk_mq_rq_from_pdu(vbr);
> +
> +		found++;
> +		if (!blk_mq_add_to_batch(req, iob, vbr->status,
> +						virtblk_complete_batch))
> +			blk_mq_complete_request(req);
> +	}
> +
> +	spin_unlock_irqrestore(&vq->lock, flags);
> +
> +	return found;
> +}
> +
> +static int virtblk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
> +			  unsigned int hctx_idx)
> +{
> +	struct virtio_blk *vblk = data;
> +	struct virtio_blk_vq *vq = &vblk->vqs[hctx_idx];
> +
> +	WARN_ON(vblk->tag_set.tags[hctx_idx] != hctx->tags);
> +	hctx->driver_data = vq;
> +	return 0;
>  }
>  
>  static const struct blk_mq_ops virtio_mq_ops = {
>  	.queue_rq	= virtio_queue_rq,
>  	.commit_rqs	= virtio_commit_rqs,
> +	.init_hctx	= virtblk_init_hctx,
>  	.complete	= virtblk_request_done,
>  	.map_queues	= virtblk_map_queues,
> +	.poll		= virtblk_poll,
>  };
>  
>  static unsigned int virtblk_queue_depth;
> @@ -816,6 +906,9 @@ static int virtblk_probe(struct virtio_device *vdev)
>  		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
>  	vblk->tag_set.driver_data = vblk;
>  	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
> +	vblk->tag_set.nr_maps = 1;
> +	if (vblk->io_queues[HCTX_TYPE_POLL])
> +		vblk->tag_set.nr_maps = 3;
>  
>  	err = blk_mq_alloc_tag_set(&vblk->tag_set);
>  	if (err)



So wrt cleanup, does something poll for all buffers to be
used when device is removed?

> -- 
> 2.26.3

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Suwan Kim]

On Thu, Mar 24, 2022 at 01:58:28PM -0400, Michael S. Tsirkin wrote:
> On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> > This patch supports polling I/O via virtio-blk driver. Polling
> > feature is enabled by module parameter "num_poll_queues" and it
> > sets dedicated polling queues for virtio-blk. This patch improves
> > the polling I/O throughput and latency.
> > 
> > The virtio-blk driver doesn't not have a poll function and a poll
> > queue and it has been operating in interrupt driven method even if
> > the polling function is called in the upper layer.
> > 
> > virtio-blk polling is implemented upon 'batched completion' of block
> > layer. virtblk_poll() queues completed request to io_comp_batch->req_list
> > and later, virtblk_complete_batch() calls unmap function and ends
> > the requests in batch.
> > 
> > virtio-blk reads the number of poll queues from module parameter
> > "num_poll_queues". If VM sets queue parameter as below,
> > ("num-queues=N" [QEMU property], "num_poll_queues=M" [module parameter])
> > It allocates N virtqueues to virtio_blk->vqs[N] and it uses [0..(N-M-1)]
> > as default queues and [(N-M)..(N-1)] as poll queues. Unlike the default
> > queues, the poll queues have no callback function.
> > 
> > Regarding HW-SW queue mapping, the default queue mapping uses the
> > existing method that condsiders MSI irq vector. But the poll queue
> > doesn't have an irq, so it uses the regular blk-mq cpu mapping.
> > 
> > For verifying the improvement, I did Fio polling I/O performance test
> > with io_uring engine with the options below.
> > (io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=N)
> > I set 4 vcpu and 4 virtio-blk queues - 2 default queues and 2 poll
> > queues for VM.
> > 
> > As a result, IOPS and average latency improved about 10%.
> > 
> > Test result:
> > 
> > - Fio io_uring poll without virtio-blk poll support
> > 	-- numjobs=1 : IOPS = 339K, avg latency = 188.33us
> > 	-- numjobs=2 : IOPS = 367K, avg latency = 347.33us
> > 	-- numjobs=4 : IOPS = 383K, avg latency = 682.06us
> > 
> > - Fio io_uring poll with virtio-blk poll support
> > 	-- numjobs=1 : IOPS = 380K, avg latency = 167.87us
> > 	-- numjobs=2 : IOPS = 409K, avg latency = 312.6us
> > 	-- numjobs=4 : IOPS = 413K, avg latency = 619.72us
> > 
> > Reported-by: kernel test robot <lkp@intel.com>
> > Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> > ---
> >  drivers/block/virtio_blk.c | 101 +++++++++++++++++++++++++++++++++++--
> >  1 file changed, 97 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> > index 8c415be86732..3d16f8b753e7 100644
> > --- a/drivers/block/virtio_blk.c
> > +++ b/drivers/block/virtio_blk.c
> > @@ -37,6 +37,10 @@ MODULE_PARM_DESC(num_request_queues,
> >  		 "0 for no limit. "
> >  		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
> >  
> > +static unsigned int num_poll_queues;
> > +module_param(num_poll_queues, uint, 0644);
> > +MODULE_PARM_DESC(num_poll_queues, "The number of dedicated virtqueues for polling I/O");
> > +
> >  static int major;
> >  static DEFINE_IDA(vd_index_ida);
> >  
> > @@ -81,6 +85,7 @@ struct virtio_blk {
> >  
> >  	/* num of vqs */
> >  	int num_vqs;
> > +	int io_queues[HCTX_MAX_TYPES];
> >  	struct virtio_blk_vq *vqs;
> >  };
> >  
> > @@ -548,6 +553,7 @@ static int init_vq(struct virtio_blk *vblk)
> >  	const char **names;
> >  	struct virtqueue **vqs;
> >  	unsigned short num_vqs;
> > +	unsigned int num_poll_vqs;
> >  	struct virtio_device *vdev = vblk->vdev;
> >  	struct irq_affinity desc = { 0, };
> >  
> > @@ -556,6 +562,7 @@ static int init_vq(struct virtio_blk *vblk)
> >  				   &num_vqs);
> >  	if (err)
> >  		num_vqs = 1;
> > +
> >  	if (!err && !num_vqs) {
> >  		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
> >  		return -EINVAL;
> > @@ -565,6 +572,13 @@ static int init_vq(struct virtio_blk *vblk)
> >  			min_not_zero(num_request_queues, nr_cpu_ids),
> >  			num_vqs);
> >  
> > +	num_poll_vqs = min_t(unsigned int, num_poll_queues, num_vqs - 1);
> > +
> > +	memset(vblk->io_queues, 0, sizeof(int) * HCTX_MAX_TYPES);
> > +	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
> > +	vblk->io_queues[HCTX_TYPE_READ] = 0;
> > +	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
> > +
> >  	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
> >  	if (!vblk->vqs)
> >  		return -ENOMEM;
> > @@ -578,8 +592,13 @@ static int init_vq(struct virtio_blk *vblk)
> >  	}
> >  
> >  	for (i = 0; i < num_vqs; i++) {
> > -		callbacks[i] = virtblk_done;
> > -		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> > +		if (i < num_vqs - num_poll_vqs) {
> > +			callbacks[i] = virtblk_done;
> > +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
> > +		} else {
> > +			callbacks[i] = NULL;
> > +			snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
> > +		}
> >  		names[i] = vblk->vqs[i].name;
> >  	}
> >  
> > @@ -728,16 +747,87 @@ static const struct attribute_group *virtblk_attr_groups[] = {
> >  static int virtblk_map_queues(struct blk_mq_tag_set *set)
> >  {
> >  	struct virtio_blk *vblk = set->driver_data;
> > +	int i, qoff;
> > +
> > +	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
> > +		struct blk_mq_queue_map *map = &set->map[i];
> > +
> > +		map->nr_queues = vblk->io_queues[i];
> > +		map->queue_offset = qoff;
> > +		qoff += map->nr_queues;
> > +
> > +		if (map->nr_queues == 0)
> > +			continue;
> > +
> > +		/*
> > +		 * Regular queues have interrupts and hence CPU affinity is
> > +		 * defined by the core virtio code, but polling queues have
> > +		 * no interrupts so we let the block layer assign CPU affinity.
> > +		 */
> > +		if (i == HCTX_TYPE_DEFAULT)
> > +			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
> > +		else
> > +			blk_mq_map_queues(&set->map[i]);
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static void virtblk_complete_batch(struct io_comp_batch *iob)
> > +{
> > +	struct request *req;
> > +	struct virtblk_req *vbr;
> >  
> > -	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
> > -					vblk->vdev, 0);
> > +	rq_list_for_each(&iob->req_list, req) {
> > +		vbr = blk_mq_rq_to_pdu(req);
> > +		virtblk_unmap_data(req, vbr);
> > +		virtblk_cleanup_cmd(req);
> > +	}
> > +	blk_mq_end_request_batch(iob);
> > +}
> > +
> > +static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
> > +{
> > +	struct virtio_blk_vq *vq = hctx->driver_data;
> > +	struct virtblk_req *vbr;
> > +	unsigned long flags;
> > +	unsigned int len;
> > +	int found = 0;
> > +
> > +	spin_lock_irqsave(&vq->lock, flags);
> > +
> > +	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> > +		struct request *req = blk_mq_rq_from_pdu(vbr);
> > +
> > +		found++;
> > +		if (!blk_mq_add_to_batch(req, iob, vbr->status,
> > +						virtblk_complete_batch))
> > +			blk_mq_complete_request(req);
> > +	}
> > +
> > +	spin_unlock_irqrestore(&vq->lock, flags);
> > +
> > +	return found;
> > +}
> > +
> > +static int virtblk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
> > +			  unsigned int hctx_idx)
> > +{
> > +	struct virtio_blk *vblk = data;
> > +	struct virtio_blk_vq *vq = &vblk->vqs[hctx_idx];
> > +
> > +	WARN_ON(vblk->tag_set.tags[hctx_idx] != hctx->tags);
> > +	hctx->driver_data = vq;
> > +	return 0;
> >  }
> >  
> >  static const struct blk_mq_ops virtio_mq_ops = {
> >  	.queue_rq	= virtio_queue_rq,
> >  	.commit_rqs	= virtio_commit_rqs,
> > +	.init_hctx	= virtblk_init_hctx,
> >  	.complete	= virtblk_request_done,
> >  	.map_queues	= virtblk_map_queues,
> > +	.poll		= virtblk_poll,
> >  };
> >  
> >  static unsigned int virtblk_queue_depth;
> > @@ -816,6 +906,9 @@ static int virtblk_probe(struct virtio_device *vdev)
> >  		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
> >  	vblk->tag_set.driver_data = vblk;
> >  	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
> > +	vblk->tag_set.nr_maps = 1;
> > +	if (vblk->io_queues[HCTX_TYPE_POLL])
> > +		vblk->tag_set.nr_maps = 3;
> >  
> >  	err = blk_mq_alloc_tag_set(&vblk->tag_set);
> >  	if (err)
> 
> 
> 
> So wrt cleanup, does something poll for all buffers to be
> used when device is removed?
 
Sorry for late reply.

Maybe below function calls iterate each HW queue and flush requests
before device is removed?

-----
virtblk_remove() -> blk_cleanup_disk()/blk_cleanup_queue() ->
blk_queue_start_drain()/blk_freeze_queue() 
-----

Regards,
Suwan Kim

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 926 bytes --]

On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> +static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
> +{
> +	struct virtio_blk_vq *vq = hctx->driver_data;
> +	struct virtblk_req *vbr;
> +	unsigned long flags;
> +	unsigned int len;
> +	int found = 0;
> +
> +	spin_lock_irqsave(&vq->lock, flags);
> +
> +	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> +		struct request *req = blk_mq_rq_from_pdu(vbr);
> +
> +		found++;
> +		if (!blk_mq_add_to_batch(req, iob, vbr->status,
> +						virtblk_complete_batch))
> +			blk_mq_complete_request(req);
> +	}
> +
> +	spin_unlock_irqrestore(&vq->lock, flags);

virtblk_done() does:

  /* In case queue is stopped waiting for more buffers. */
  if (req_done)
          blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);

Is the same thing needed here in virtblk_poll() so that stopped queues
are restarted when requests complete?

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v3 1/2] virtio-blk: support polling I/O

[Suwan Kim]

On Mon, Mar 28, 2022 at 01:53:46PM +0100, Stefan Hajnoczi wrote:
> On Thu, Mar 24, 2022 at 11:04:49PM +0900, Suwan Kim wrote:
> > +static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
> > +{
> > +	struct virtio_blk_vq *vq = hctx->driver_data;
> > +	struct virtblk_req *vbr;
> > +	unsigned long flags;
> > +	unsigned int len;
> > +	int found = 0;
> > +
> > +	spin_lock_irqsave(&vq->lock, flags);
> > +
> > +	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> > +		struct request *req = blk_mq_rq_from_pdu(vbr);
> > +
> > +		found++;
> > +		if (!blk_mq_add_to_batch(req, iob, vbr->status,
> > +						virtblk_complete_batch))
> > +			blk_mq_complete_request(req);
> > +	}
> > +
> > +	spin_unlock_irqrestore(&vq->lock, flags);
> 
> virtblk_done() does:
> 
>   /* In case queue is stopped waiting for more buffers. */
>   if (req_done)
>           blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
> 
> Is the same thing needed here in virtblk_poll() so that stopped queues
> are restarted when requests complete?

I think you are right. I missed that.

I just added blk_mq_start_stopped_hw_queues() to virtblk_poll as
you commented and did performance test again.

It showed higher peak performance than virtblk_poll without
blk_mq_start_stopped_hw_queues().

I will add it in next version.
Thanks for the comment!

Regards,
Suwan Kim

[PATCH v3 2/2] virtio-blk: support mq_ops->queue_rqs()

[Suwan Kim]

This patch supports mq_ops->queue_rqs() hook. It has an advantage of
batch submission to virtio-blk driver. It also helps polling I/O because
polling uses batched completion of block layer. Batch submission in
queue_rqs() can boost polling performance.

In queue_rqs(), it iterates plug->mq_list, collects requests that
belong to same HW queue and adds requests into virtqueue until it
encounters a request from other HW queue or sees the end of the list.
Then, virtio-blk kicks virtqueue to submit requests.

If there is an error, it inserts error request to requeue_list and
passes it to ordinary block layer path.

For verification, I did fio test.
(io_uring, randread, direct=1, bs=4K, iodepth=64 numjobs=N)
I set 4 vcpu and 2 virtio-blk queues for VM and run fio test 5 times.
It shows about 1-4% improvement.

                                 |   numjobs=2   |    numjobs=4
      -----------------------------------------------------------
        fio without queue_rqs()  |   282K IOPS   |    245K IOPS
      -----------------------------------------------------------
        fio with queue_rqs()     |   294K IOPS   |    249K IOPS

For polling I/O performance, I also did fio test as below.
(io_uring, hipri, randread, direct=1, bs=512, iodepth=64 numjobs=4)
I set 4 vcpu and 2 poll queues for VM.
It shows upto 7% improvement in polling I/O.

                                      |   IOPS   |  avg latency
      -----------------------------------------------------------
        fio poll without queue_rqs()  |   413K   |   619.72 usec
      -----------------------------------------------------------
        fio poll with queue_rqs()     |   445K   |   581.2  usec

Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
---
 drivers/block/virtio_blk.c | 93 ++++++++++++++++++++++++++++++++++----
 1 file changed, 84 insertions(+), 9 deletions(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 3d16f8b753e7..4a0a3b2d9caf 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -311,6 +311,28 @@ static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 		virtqueue_notify(vq->vq);
 }
 
+static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
+					struct virtio_blk *vblk,
+					struct request *req,
+					struct virtblk_req *vbr, int *num)
+{
+	blk_status_t status;
+
+	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
+	if (unlikely(status))
+		return status;
+
+	blk_mq_start_request(req);
+
+	*num = virtblk_map_data(hctx, req, vbr);
+	if (unlikely(*num < 0)) {
+		virtblk_cleanup_cmd(req);
+		return BLK_STS_RESOURCE;
+	}
+
+	return BLK_STS_OK;
+}
+
 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 			   const struct blk_mq_queue_data *bd)
 {
@@ -324,18 +346,10 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 	blk_status_t status;
 	int err;
 
-	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
+	status = virtblk_prep_rq(hctx, vblk, req, vbr, &num);
 	if (unlikely(status))
 		return status;
 
-	blk_mq_start_request(req);
-
-	num = virtblk_map_data(hctx, req, vbr);
-	if (unlikely(num < 0)) {
-		virtblk_cleanup_cmd(req);
-		return BLK_STS_RESOURCE;
-	}
-
 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg_table.sgl, num);
 	if (err) {
@@ -367,6 +381,66 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 	return BLK_STS_OK;
 }
 
+static bool virtblk_prep_rq_batch(struct virtio_blk_vq *vq, struct request *req)
+{
+	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
+	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
+	unsigned long flags;
+	int num, err;
+
+	req->mq_hctx->tags->rqs[req->tag] = req;
+
+	if (virtblk_prep_rq(req->mq_hctx, vblk, req, vbr, &num) != BLK_STS_OK)
+		return false;
+
+	spin_lock_irqsave(&vq->lock, flags);
+	err = virtblk_add_req(vq->vq, vbr, vbr->sg_table.sgl, num);
+	if (err) {
+		spin_unlock_irqrestore(&vq->lock, flags);
+		virtblk_unmap_data(req, vbr);
+		virtblk_cleanup_cmd(req);
+		return false;
+	}
+	spin_unlock_irqrestore(&vq->lock, flags);
+
+	return true;
+}
+
+static void virtio_queue_rqs(struct request **rqlist)
+{
+	struct request *req, *next, *prev = NULL;
+	struct request *requeue_list = NULL;
+
+	rq_list_for_each_safe(rqlist, req, next) {
+		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
+		unsigned long flags;
+		bool kick;
+
+		if (!virtblk_prep_rq_batch(vq, req)) {
+			rq_list_move(rqlist, &requeue_list, req, prev);
+			req = prev;
+
+			if (!req)
+				continue;
+		}
+
+		if (!next || req->mq_hctx != next->mq_hctx) {
+			spin_lock_irqsave(&vq->lock, flags);
+			kick = virtqueue_kick_prepare(vq->vq);
+			spin_unlock_irqrestore(&vq->lock, flags);
+			if (kick)
+				virtqueue_notify(vq->vq);
+
+			req->rq_next = NULL;
+			*rqlist = next;
+			prev = NULL;
+		} else
+			prev = req;
+	}
+
+	*rqlist = requeue_list;
+}
+
 /* return id (s/n) string for *disk to *id_str
  */
 static int virtblk_get_id(struct gendisk *disk, char *id_str)
@@ -823,6 +897,7 @@ static int virtblk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
 
 static const struct blk_mq_ops virtio_mq_ops = {
 	.queue_rq	= virtio_queue_rq,
+	.queue_rqs	= virtio_queue_rqs,
 	.commit_rqs	= virtio_commit_rqs,
 	.init_hctx	= virtblk_init_hctx,
 	.complete	= virtblk_request_done,
-- 
2.26.3

Re: [PATCH v3 2/2] virtio-blk: support mq_ops->queue_rqs()

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 2578 bytes --]

On Thu, Mar 24, 2022 at 11:04:50PM +0900, Suwan Kim wrote:
> @@ -367,6 +381,66 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
>  	return BLK_STS_OK;
>  }
>  
> +static bool virtblk_prep_rq_batch(struct virtio_blk_vq *vq, struct request *req)
> +{
> +	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
> +	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
> +	unsigned long flags;
> +	int num, err;
> +
> +	req->mq_hctx->tags->rqs[req->tag] = req;
> +
> +	if (virtblk_prep_rq(req->mq_hctx, vblk, req, vbr, &num) != BLK_STS_OK)
> +		return false;
> +
> +	spin_lock_irqsave(&vq->lock, flags);
> +	err = virtblk_add_req(vq->vq, vbr, vbr->sg_table.sgl, num);
> +	if (err) {
> +		spin_unlock_irqrestore(&vq->lock, flags);
> +		virtblk_unmap_data(req, vbr);
> +		virtblk_cleanup_cmd(req);
> +		return false;
> +	}
> +	spin_unlock_irqrestore(&vq->lock, flags);

Simplification:

  spin_lock_irqsave(&vq->lock, flags);
  err = virtblk_add_req(vq->vq, vbr, vbr->sg_table.sgl, num);
  spin_unlock_irqrestore(&vq->lock, flags);
  if (err) {
      virtblk_unmap_data(req, vbr);
      virtblk_cleanup_cmd(req);
      return false;
  }

> +
> +	return true;
> +}
> +
> +static void virtio_queue_rqs(struct request **rqlist)
> +{
> +	struct request *req, *next, *prev = NULL;
> +	struct request *requeue_list = NULL;
> +
> +	rq_list_for_each_safe(rqlist, req, next) {
> +		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
> +		unsigned long flags;
> +		bool kick;
> +
> +		if (!virtblk_prep_rq_batch(vq, req)) {
> +			rq_list_move(rqlist, &requeue_list, req, prev);
> +			req = prev;
> +
> +			if (!req)
> +				continue;
> +		}
> +
> +		if (!next || req->mq_hctx != next->mq_hctx) {
> +			spin_lock_irqsave(&vq->lock, flags);

Did you try calling virtblk_add_req() here to avoid acquiring and
releasing the lock multiple times? In other words, do virtblk_prep_rq()
but wait until we get here to do virtblk_add_req().

I don't know if it has any measurable effect on performance or maybe the
code would become too complex, but I noticed that we're not fully
exploiting batching.

> +			kick = virtqueue_kick_prepare(vq->vq);
> +			spin_unlock_irqrestore(&vq->lock, flags);
> +			if (kick)
> +				virtqueue_notify(vq->vq);
> +
> +			req->rq_next = NULL;
> +			*rqlist = next;
> +			prev = NULL;
> +		} else
> +			prev = req;

What guarantees that req is still alive after we called
virtblk_add_req()? The device may have seen it and completed it already
by the time we get here.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v3 2/2] virtio-blk: support mq_ops->queue_rqs()

[Suwan Kim]

On Mon, Mar 28, 2022 at 02:16:13PM +0100, Stefan Hajnoczi wrote:
> On Thu, Mar 24, 2022 at 11:04:50PM +0900, Suwan Kim wrote:
> > @@ -367,6 +381,66 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
> >  	return BLK_STS_OK;
> >  }
> >  
> > +static bool virtblk_prep_rq_batch(struct virtio_blk_vq *vq, struct request *req)
> > +{
> > +	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
> > +	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
> > +	unsigned long flags;
> > +	int num, err;
> > +
> > +	req->mq_hctx->tags->rqs[req->tag] = req;
> > +
> > +	if (virtblk_prep_rq(req->mq_hctx, vblk, req, vbr, &num) != BLK_STS_OK)
> > +		return false;
> > +
> > +	spin_lock_irqsave(&vq->lock, flags);
> > +	err = virtblk_add_req(vq->vq, vbr, vbr->sg_table.sgl, num);
> > +	if (err) {
> > +		spin_unlock_irqrestore(&vq->lock, flags);
> > +		virtblk_unmap_data(req, vbr);
> > +		virtblk_cleanup_cmd(req);
> > +		return false;
> > +	}
> > +	spin_unlock_irqrestore(&vq->lock, flags);
> 
> Simplification:
> 
>   spin_lock_irqsave(&vq->lock, flags);
>   err = virtblk_add_req(vq->vq, vbr, vbr->sg_table.sgl, num);
>   spin_unlock_irqrestore(&vq->lock, flags);
>   if (err) {
>       virtblk_unmap_data(req, vbr);
>       virtblk_cleanup_cmd(req);
>       return false;
>   }
> 

Thanks! I will fix it.

> > +
> > +	return true;
> > +}
> > +
> > +static void virtio_queue_rqs(struct request **rqlist)
> > +{
> > +	struct request *req, *next, *prev = NULL;
> > +	struct request *requeue_list = NULL;
> > +
> > +	rq_list_for_each_safe(rqlist, req, next) {
> > +		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
> > +		unsigned long flags;
> > +		bool kick;
> > +
> > +		if (!virtblk_prep_rq_batch(vq, req)) {
> > +			rq_list_move(rqlist, &requeue_list, req, prev);
> > +			req = prev;
> > +
> > +			if (!req)
> > +				continue;
> > +		}
> > +
> > +		if (!next || req->mq_hctx != next->mq_hctx) {
> > +			spin_lock_irqsave(&vq->lock, flags);
> 
> Did you try calling virtblk_add_req() here to avoid acquiring and
> releasing the lock multiple times? In other words, do virtblk_prep_rq()
> but wait until we get here to do virtblk_add_req().
> 
> I don't know if it has any measurable effect on performance or maybe the
> code would become too complex, but I noticed that we're not fully
> exploiting batching.

I tried as you said. I called virtlblk_add_req() and added requests
of rqlist to virtqueue in this if statement with holding the lock
only once.

I attach the code at the end of this mail.
Please refer the code.

But I didn't see improvement. It showed slightly worse performance
than the current patch.

> > +			kick = virtqueue_kick_prepare(vq->vq);
> > +			spin_unlock_irqrestore(&vq->lock, flags);
> > +			if (kick)
> > +				virtqueue_notify(vq->vq);
> > +
> > +			req->rq_next = NULL;

Did you ask this part?

> > +			*rqlist = next;
> > +			prev = NULL;
> > +		} else
> > +			prev = req;
> 
> What guarantees that req is still alive after we called
> virtblk_add_req()? The device may have seen it and completed it already
> by the time we get here.

Isn't request completed after the kick?

If you asked about "req->rq_next = NULL",
I think it should be placed before
"kick = virtqueue_kick_prepare(vq->vq);"

-----------
	req->rq_next = NULL;
	kick = virtqueue_kick_prepare(vq->vq);
	spin_unlock_irqrestore(&vq->lock, flags);
	if (kick)
		virtqueue_notify(vq->vq);
-----------

Regards,
Suwan Kim

---

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 2218cab39c72..d972d3042068 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -92,6 +92,7 @@ struct virtio_blk {
 struct virtblk_req {
 	struct virtio_blk_outhdr out_hdr;
 	u8 status;
+	int sg_num;
 	struct sg_table sg_table;
 	struct scatterlist sg[];
 };
@@ -311,18 +312,13 @@ static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 		virtqueue_notify(vq->vq);
 }

-static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
-			   const struct blk_mq_queue_data *bd)
+static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
+					struct virtio_blk *vblk,
+					struct request *req,
+					struct virtblk_req *vbr)
 {
-	struct virtio_blk *vblk = hctx->queue->queuedata;
-	struct request *req = bd->rq;
-	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
-	unsigned long flags;
-	int num;
-	int qid = hctx->queue_num;
-	bool notify = false;
 	blk_status_t status;
-	int err;
+	int num;

 	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
 	if (unlikely(status))
@@ -335,9 +331,30 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 		virtblk_cleanup_cmd(req);
 		return BLK_STS_RESOURCE;
 	}
+	vbr->sg_num = num;
+
+	return BLK_STS_OK;
+}
+
+static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
+			   const struct blk_mq_queue_data *bd)
+{
+	struct virtio_blk *vblk = hctx->queue->queuedata;
+	struct request *req = bd->rq;
+	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
+	unsigned long flags;
+	int qid = hctx->queue_num;
+	bool notify = false;
+	blk_status_t status;
+	int err;
+
+	status = virtblk_prep_rq(hctx, vblk, req, vbr);
+	if (unlikely(status))
+		return status;

 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
-	err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg_table.sgl, num);
+	err = virtblk_add_req(vblk->vqs[qid].vq, vbr,
+				vbr->sg_table.sgl, vbr->sg_num);
 	if (err) {
 		virtqueue_kick(vblk->vqs[qid].vq);
 		/* Don't stop the queue if -ENOMEM: we may have failed to
@@ -367,6 +384,76 @@ static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 	return BLK_STS_OK;
 }

+static bool virtblk_prep_rq_batch(struct virtio_blk_vq *vq, struct request *req)
+{
+	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
+	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
+
+	req->mq_hctx->tags->rqs[req->tag] = req;
+
+	return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
+}
+
+static bool virtblk_add_req_batch(struct virtio_blk_vq *vq,
+					struct request **rqlist,
+					struct request **requeue_list)
+{
+	unsigned long flags;
+	int err;
+	bool kick;
+
+	spin_lock_irqsave(&vq->lock, flags);
+	while (!rq_list_empty(*rqlist)) {
+		struct request *req = rq_list_pop(rqlist);
+		struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
+
+		err = virtblk_add_req(vq->vq, vbr,
+					vbr->sg_table.sgl, vbr->sg_num);
+		if (err) {
+			virtblk_unmap_data(req, vbr);
+			virtblk_cleanup_cmd(req);
+			rq_list_add(requeue_list, req);
+		}
+	}
+
+	kick = virtqueue_kick_prepare(vq->vq);
+	spin_unlock_irqrestore(&vq->lock, flags);
+
+	return kick;
+}
+
+static void virtio_queue_rqs(struct request **rqlist)
+{
+	struct request *req, *next, *prev = NULL;
+	struct request *requeue_list = NULL;
+
+	rq_list_for_each_safe(rqlist, req, next) {
+		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
+		bool kick;
+
+		if (!virtblk_prep_rq_batch(vq, req)) {
+			rq_list_move(rqlist, &requeue_list, req, prev);
+			req = prev;
+
+			if (!req)
+				continue;
+		}
+
+		if (!next || req->mq_hctx != next->mq_hctx) {
+			kick = virtblk_add_req_batch(vq, rqlist, &requeue_list);
+			if (kick)
+				virtqueue_notify(vq->vq);
+
+			req->rq_next = NULL;
+			*rqlist = next;
+			prev = NULL;
+		} else
+			prev = req;
+	}
+
+	*rqlist = requeue_list;
+}
+
 /* return id (s/n) string for *disk to *id_str
  */
 static int virtblk_get_id(struct gendisk *disk, char *id_str)
@@ -818,6 +905,7 @@ static int virtblk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,

 static const struct blk_mq_ops virtio_mq_ops = {
 	.queue_rq	= virtio_queue_rq,
+	.queue_rqs	= virtio_queue_rqs,
 	.commit_rqs	= virtio_commit_rqs,
 	.init_hctx	= virtblk_init_hctx,
 	.complete	= virtblk_request_done,

Re: [PATCH v3 2/2] virtio-blk: support mq_ops->queue_rqs()

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 2877 bytes --]

On Tue, Mar 29, 2022 at 12:50:33AM +0900, Suwan Kim wrote:
> On Mon, Mar 28, 2022 at 02:16:13PM +0100, Stefan Hajnoczi wrote:
> > On Thu, Mar 24, 2022 at 11:04:50PM +0900, Suwan Kim wrote:
> > > +static void virtio_queue_rqs(struct request **rqlist)
> > > +{
> > > +	struct request *req, *next, *prev = NULL;
> > > +	struct request *requeue_list = NULL;
> > > +
> > > +	rq_list_for_each_safe(rqlist, req, next) {
> > > +		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
> > > +		unsigned long flags;
> > > +		bool kick;
> > > +
> > > +		if (!virtblk_prep_rq_batch(vq, req)) {
> > > +			rq_list_move(rqlist, &requeue_list, req, prev);
> > > +			req = prev;
> > > +
> > > +			if (!req)
> > > +				continue;
> > > +		}
> > > +
> > > +		if (!next || req->mq_hctx != next->mq_hctx) {
> > > +			spin_lock_irqsave(&vq->lock, flags);
> > 
> > Did you try calling virtblk_add_req() here to avoid acquiring and
> > releasing the lock multiple times? In other words, do virtblk_prep_rq()
> > but wait until we get here to do virtblk_add_req().
> > 
> > I don't know if it has any measurable effect on performance or maybe the
> > code would become too complex, but I noticed that we're not fully
> > exploiting batching.
> 
> I tried as you said. I called virtlblk_add_req() and added requests
> of rqlist to virtqueue in this if statement with holding the lock
> only once.
> 
> I attach the code at the end of this mail.
> Please refer the code.
> 
> But I didn't see improvement. It showed slightly worse performance
> than the current patch.

Okay, thanks for trying it!

> > > +			kick = virtqueue_kick_prepare(vq->vq);
> > > +			spin_unlock_irqrestore(&vq->lock, flags);
> > > +			if (kick)
> > > +				virtqueue_notify(vq->vq);
> > > +
> > > +			req->rq_next = NULL;
> 
> Did you ask this part?
> 
> > > +			*rqlist = next;
> > > +			prev = NULL;
> > > +		} else
> > > +			prev = req;
> > 
> > What guarantees that req is still alive after we called
> > virtblk_add_req()? The device may have seen it and completed it already
> > by the time we get here.
> 
> Isn't request completed after the kick?
> 
> If you asked about "req->rq_next = NULL",
> I think it should be placed before
> "kick = virtqueue_kick_prepare(vq->vq);"
> 
> -----------
> 	req->rq_next = NULL;
> 	kick = virtqueue_kick_prepare(vq->vq);
> 	spin_unlock_irqrestore(&vq->lock, flags);
> 	if (kick)
> 		virtqueue_notify(vq->vq);
> -----------

No, virtqueue_add_sgs() exposes vring descriptors to the device. The
device may process immediately. In other words, VIRTIO devices may poll
the vring instead of waiting for virtqueue_notify(). There is no
guarantee that the request is alive until virtqueue_notify() is called.

The code has to handle the case where the request is completed during
virtqueue_add_sgs().

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v3 2/2] virtio-blk: support mq_ops->queue_rqs()

[Suwan Kim]

On Tue, Mar 29, 2022 at 09:45:29AM +0100, Stefan Hajnoczi wrote:
> On Tue, Mar 29, 2022 at 12:50:33AM +0900, Suwan Kim wrote:
> > On Mon, Mar 28, 2022 at 02:16:13PM +0100, Stefan Hajnoczi wrote:
> > > On Thu, Mar 24, 2022 at 11:04:50PM +0900, Suwan Kim wrote:
> > > > +static void virtio_queue_rqs(struct request **rqlist)
> > > > +{
> > > > +	struct request *req, *next, *prev = NULL;
> > > > +	struct request *requeue_list = NULL;
> > > > +
> > > > +	rq_list_for_each_safe(rqlist, req, next) {
> > > > +		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
> > > > +		unsigned long flags;
> > > > +		bool kick;
> > > > +
> > > > +		if (!virtblk_prep_rq_batch(vq, req)) {
> > > > +			rq_list_move(rqlist, &requeue_list, req, prev);
> > > > +			req = prev;
> > > > +
> > > > +			if (!req)
> > > > +				continue;
> > > > +		}
> > > > +
> > > > +		if (!next || req->mq_hctx != next->mq_hctx) {
> > > > +			spin_lock_irqsave(&vq->lock, flags);
> > > 
> > > Did you try calling virtblk_add_req() here to avoid acquiring and
> > > releasing the lock multiple times? In other words, do virtblk_prep_rq()
> > > but wait until we get here to do virtblk_add_req().
> > > 
> > > I don't know if it has any measurable effect on performance or maybe the
> > > code would become too complex, but I noticed that we're not fully
> > > exploiting batching.
> > 
> > I tried as you said. I called virtlblk_add_req() and added requests
> > of rqlist to virtqueue in this if statement with holding the lock
> > only once.
> > 
> > I attach the code at the end of this mail.
> > Please refer the code.
> > 
> > But I didn't see improvement. It showed slightly worse performance
> > than the current patch.
> 
> Okay, thanks for trying it!
> 
> > > > +			kick = virtqueue_kick_prepare(vq->vq);
> > > > +			spin_unlock_irqrestore(&vq->lock, flags);
> > > > +			if (kick)
> > > > +				virtqueue_notify(vq->vq);
> > > > +
> > > > +			req->rq_next = NULL;
> > 
> > Did you ask this part?
> > 
> > > > +			*rqlist = next;
> > > > +			prev = NULL;
> > > > +		} else
> > > > +			prev = req;
> > > 
> > > What guarantees that req is still alive after we called
> > > virtblk_add_req()? The device may have seen it and completed it already
> > > by the time we get here.
> > 
> > Isn't request completed after the kick?
> > 
> > If you asked about "req->rq_next = NULL",
> > I think it should be placed before
> > "kick = virtqueue_kick_prepare(vq->vq);"
> > 
> > -----------
> > 	req->rq_next = NULL;
> > 	kick = virtqueue_kick_prepare(vq->vq);
> > 	spin_unlock_irqrestore(&vq->lock, flags);
> > 	if (kick)
> > 		virtqueue_notify(vq->vq);
> > -----------
> 
> No, virtqueue_add_sgs() exposes vring descriptors to the device. The
> device may process immediately. In other words, VIRTIO devices may poll
> the vring instead of waiting for virtqueue_notify(). There is no
> guarantee that the request is alive until virtqueue_notify() is called.
> 
> The code has to handle the case where the request is completed during
> virtqueue_add_sgs().

Thanks for the explanation.

We should not use req again after virtblk_add_req().
I understand...

Then, as you commented in previous mail, is it ok that we do
virtblk_add_req() in "if (!next || req->mq_hctx != next->mq_hctx)"
statement to avoid use req again after virtblk_add_req() as below code?

In this code, It adds reqs to virtqueue in batch just before
virtqueue_notify(), and it doesn't use req again after calling
virtblk_add_req().

If it is fine, I will try it again.
This code is slightly different from the code I sent in previous mail.

---
static void virtio_queue_rqs(struct request **rqlist)
...
	rq_list_for_each_safe(rqlist, req, next) {
...
		if (!next || req->mq_hctx != next->mq_hctx) {
			// Cut the list at current req
			req->rq_next = NULL;
			// Add req list to virtqueue in batch with holding lock once
			kick = virtblk_add_req_batch(vq, rqlist, &requeue_list);
			if (kick)
				virtqueue_notify(vq->vq);

			// setup new req list. Don't use previous req again.
			*rqlist = next;
			prev = NULL;
...
---

Regards,
Suwan Kim

Re: [PATCH v3 2/2] virtio-blk: support mq_ops->queue_rqs()

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 4758 bytes --]

On Tue, Mar 29, 2022 at 10:48:16PM +0900, Suwan Kim wrote:
> On Tue, Mar 29, 2022 at 09:45:29AM +0100, Stefan Hajnoczi wrote:
> > On Tue, Mar 29, 2022 at 12:50:33AM +0900, Suwan Kim wrote:
> > > On Mon, Mar 28, 2022 at 02:16:13PM +0100, Stefan Hajnoczi wrote:
> > > > On Thu, Mar 24, 2022 at 11:04:50PM +0900, Suwan Kim wrote:
> > > > > +static void virtio_queue_rqs(struct request **rqlist)
> > > > > +{
> > > > > +	struct request *req, *next, *prev = NULL;
> > > > > +	struct request *requeue_list = NULL;
> > > > > +
> > > > > +	rq_list_for_each_safe(rqlist, req, next) {
> > > > > +		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
> > > > > +		unsigned long flags;
> > > > > +		bool kick;
> > > > > +
> > > > > +		if (!virtblk_prep_rq_batch(vq, req)) {
> > > > > +			rq_list_move(rqlist, &requeue_list, req, prev);
> > > > > +			req = prev;
> > > > > +
> > > > > +			if (!req)
> > > > > +				continue;
> > > > > +		}
> > > > > +
> > > > > +		if (!next || req->mq_hctx != next->mq_hctx) {
> > > > > +			spin_lock_irqsave(&vq->lock, flags);
> > > > 
> > > > Did you try calling virtblk_add_req() here to avoid acquiring and
> > > > releasing the lock multiple times? In other words, do virtblk_prep_rq()
> > > > but wait until we get here to do virtblk_add_req().
> > > > 
> > > > I don't know if it has any measurable effect on performance or maybe the
> > > > code would become too complex, but I noticed that we're not fully
> > > > exploiting batching.
> > > 
> > > I tried as you said. I called virtlblk_add_req() and added requests
> > > of rqlist to virtqueue in this if statement with holding the lock
> > > only once.
> > > 
> > > I attach the code at the end of this mail.
> > > Please refer the code.
> > > 
> > > But I didn't see improvement. It showed slightly worse performance
> > > than the current patch.
> > 
> > Okay, thanks for trying it!
> > 
> > > > > +			kick = virtqueue_kick_prepare(vq->vq);
> > > > > +			spin_unlock_irqrestore(&vq->lock, flags);
> > > > > +			if (kick)
> > > > > +				virtqueue_notify(vq->vq);
> > > > > +
> > > > > +			req->rq_next = NULL;
> > > 
> > > Did you ask this part?
> > > 
> > > > > +			*rqlist = next;
> > > > > +			prev = NULL;
> > > > > +		} else
> > > > > +			prev = req;
> > > > 
> > > > What guarantees that req is still alive after we called
> > > > virtblk_add_req()? The device may have seen it and completed it already
> > > > by the time we get here.
> > > 
> > > Isn't request completed after the kick?
> > > 
> > > If you asked about "req->rq_next = NULL",
> > > I think it should be placed before
> > > "kick = virtqueue_kick_prepare(vq->vq);"
> > > 
> > > -----------
> > > 	req->rq_next = NULL;
> > > 	kick = virtqueue_kick_prepare(vq->vq);
> > > 	spin_unlock_irqrestore(&vq->lock, flags);
> > > 	if (kick)
> > > 		virtqueue_notify(vq->vq);
> > > -----------
> > 
> > No, virtqueue_add_sgs() exposes vring descriptors to the device. The
> > device may process immediately. In other words, VIRTIO devices may poll
> > the vring instead of waiting for virtqueue_notify(). There is no
> > guarantee that the request is alive until virtqueue_notify() is called.
> > 
> > The code has to handle the case where the request is completed during
> > virtqueue_add_sgs().
> 
> Thanks for the explanation.
> 
> We should not use req again after virtblk_add_req().
> I understand...
> 
> Then, as you commented in previous mail, is it ok that we do
> virtblk_add_req() in "if (!next || req->mq_hctx != next->mq_hctx)"
> statement to avoid use req again after virtblk_add_req() as below code?
> 
> In this code, It adds reqs to virtqueue in batch just before
> virtqueue_notify(), and it doesn't use req again after calling
> virtblk_add_req().
> 
> If it is fine, I will try it again.
> This code is slightly different from the code I sent in previous mail.
> 
> ---
> static void virtio_queue_rqs(struct request **rqlist)
> ...
> 	rq_list_for_each_safe(rqlist, req, next) {
> ...
> 		if (!next || req->mq_hctx != next->mq_hctx) {
> 			// Cut the list at current req
> 			req->rq_next = NULL;
> 			// Add req list to virtqueue in batch with holding lock once
> 			kick = virtblk_add_req_batch(vq, rqlist, &requeue_list);
> 			if (kick)
> 				virtqueue_notify(vq->vq);
> 
> 			// setup new req list. Don't use previous req again.
> 			*rqlist = next;
> 			prev = NULL;
> ...

Yes, that sounds good.

(I noticed struct request has a reference count so that might be a way
to keep requests alive, if necessary, but I haven't investigated. See
req_ref_put_and_test() though it's not used by block drivers and maybe
virtio-blk shouldn't mess with it either.)

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v3 2/2] virtio-blk: support mq_ops->queue_rqs()

[Suwan Kim]

On Tue, Mar 29, 2022 at 04:01:46PM +0100, Stefan Hajnoczi wrote:
> On Tue, Mar 29, 2022 at 10:48:16PM +0900, Suwan Kim wrote:
> > On Tue, Mar 29, 2022 at 09:45:29AM +0100, Stefan Hajnoczi wrote:
> > > On Tue, Mar 29, 2022 at 12:50:33AM +0900, Suwan Kim wrote:
> > > > On Mon, Mar 28, 2022 at 02:16:13PM +0100, Stefan Hajnoczi wrote:
> > > > > On Thu, Mar 24, 2022 at 11:04:50PM +0900, Suwan Kim wrote:
> > > > > > +static void virtio_queue_rqs(struct request **rqlist)
> > > > > > +{
> > > > > > +	struct request *req, *next, *prev = NULL;
> > > > > > +	struct request *requeue_list = NULL;
> > > > > > +
> > > > > > +	rq_list_for_each_safe(rqlist, req, next) {
> > > > > > +		struct virtio_blk_vq *vq = req->mq_hctx->driver_data;
> > > > > > +		unsigned long flags;
> > > > > > +		bool kick;
> > > > > > +
> > > > > > +		if (!virtblk_prep_rq_batch(vq, req)) {
> > > > > > +			rq_list_move(rqlist, &requeue_list, req, prev);
> > > > > > +			req = prev;
> > > > > > +
> > > > > > +			if (!req)
> > > > > > +				continue;
> > > > > > +		}
> > > > > > +
> > > > > > +		if (!next || req->mq_hctx != next->mq_hctx) {
> > > > > > +			spin_lock_irqsave(&vq->lock, flags);
> > > > > 
> > > > > Did you try calling virtblk_add_req() here to avoid acquiring and
> > > > > releasing the lock multiple times? In other words, do virtblk_prep_rq()
> > > > > but wait until we get here to do virtblk_add_req().
> > > > > 
> > > > > I don't know if it has any measurable effect on performance or maybe the
> > > > > code would become too complex, but I noticed that we're not fully
> > > > > exploiting batching.
> > > > 
> > > > I tried as you said. I called virtlblk_add_req() and added requests
> > > > of rqlist to virtqueue in this if statement with holding the lock
> > > > only once.
> > > > 
> > > > I attach the code at the end of this mail.
> > > > Please refer the code.
> > > > 
> > > > But I didn't see improvement. It showed slightly worse performance
> > > > than the current patch.
> > > 
> > > Okay, thanks for trying it!
> > > 
> > > > > > +			kick = virtqueue_kick_prepare(vq->vq);
> > > > > > +			spin_unlock_irqrestore(&vq->lock, flags);
> > > > > > +			if (kick)
> > > > > > +				virtqueue_notify(vq->vq);
> > > > > > +
> > > > > > +			req->rq_next = NULL;
> > > > 
> > > > Did you ask this part?
> > > > 
> > > > > > +			*rqlist = next;
> > > > > > +			prev = NULL;
> > > > > > +		} else
> > > > > > +			prev = req;
> > > > > 
> > > > > What guarantees that req is still alive after we called
> > > > > virtblk_add_req()? The device may have seen it and completed it already
> > > > > by the time we get here.
> > > > 
> > > > Isn't request completed after the kick?
> > > > 
> > > > If you asked about "req->rq_next = NULL",
> > > > I think it should be placed before
> > > > "kick = virtqueue_kick_prepare(vq->vq);"
> > > > 
> > > > -----------
> > > > 	req->rq_next = NULL;
> > > > 	kick = virtqueue_kick_prepare(vq->vq);
> > > > 	spin_unlock_irqrestore(&vq->lock, flags);
> > > > 	if (kick)
> > > > 		virtqueue_notify(vq->vq);
> > > > -----------
> > > 
> > > No, virtqueue_add_sgs() exposes vring descriptors to the device. The
> > > device may process immediately. In other words, VIRTIO devices may poll
> > > the vring instead of waiting for virtqueue_notify(). There is no
> > > guarantee that the request is alive until virtqueue_notify() is called.
> > > 
> > > The code has to handle the case where the request is completed during
> > > virtqueue_add_sgs().
> > 
> > Thanks for the explanation.
> > 
> > We should not use req again after virtblk_add_req().
> > I understand...
> > 
> > Then, as you commented in previous mail, is it ok that we do
> > virtblk_add_req() in "if (!next || req->mq_hctx != next->mq_hctx)"
> > statement to avoid use req again after virtblk_add_req() as below code?
> > 
> > In this code, It adds reqs to virtqueue in batch just before
> > virtqueue_notify(), and it doesn't use req again after calling
> > virtblk_add_req().
> > 
> > If it is fine, I will try it again.
> > This code is slightly different from the code I sent in previous mail.
> > 
> > ---
> > static void virtio_queue_rqs(struct request **rqlist)
> > ...
> > 	rq_list_for_each_safe(rqlist, req, next) {
> > ...
> > 		if (!next || req->mq_hctx != next->mq_hctx) {
> > 			// Cut the list at current req
> > 			req->rq_next = NULL;
> > 			// Add req list to virtqueue in batch with holding lock once
> > 			kick = virtblk_add_req_batch(vq, rqlist, &requeue_list);
> > 			if (kick)
> > 				virtqueue_notify(vq->vq);
> > 
> > 			// setup new req list. Don't use previous req again.
> > 			*rqlist = next;
> > 			prev = NULL;
> > ...
> 
> Yes, that sounds good.
> 
> (I noticed struct request has a reference count so that might be a way
> to keep requests alive, if necessary, but I haven't investigated. See
> req_ref_put_and_test() though it's not used by block drivers and maybe
> virtio-blk shouldn't mess with it either.)

I also think that using ref count is not a good idea.
I will send the next version soon.

Regards,
Suwan Kim