[RFC 0/2] add unsafe API to get inflight packets

[RFC 0/2] add unsafe API to get inflight packets

[xuan.ding]

From: Xuan Ding <xuan.ding@intel.com>

This patchset introduces an unsafe API to get the number of inflight
packets in DMA engine in some situations. Like vring state changes or
device is destroyed. Compared with rte_vhost_async_get_inflight(),
this is a lock free version.

Xuan Ding (2):
  vhost: add unsafe API to check inflight packets
  examples/vhost: use API to check inflight packets

 doc/guides/prog_guide/vhost_lib.rst    |  5 +++++
 doc/guides/rel_notes/release_22_03.rst |  4 ++++
 examples/vhost/main.c                  | 28 ++++++++++++++------------
 examples/vhost/main.h                  |  1 -
 lib/vhost/rte_vhost_async.h            | 14 +++++++++++++
 lib/vhost/version.map                  |  1 +
 lib/vhost/vhost.c                      | 26 ++++++++++++++++++++++++
 7 files changed, 65 insertions(+), 14 deletions(-)

-- 
2.17.1

[RFC 1/2] vhost: add unsafe API to check inflight packets

[xuan.ding]

From: Xuan Ding <xuan.ding@intel.com>

In async data path, when vring state changes or device is destroyed,
it is necessary to know the number of inflight packets in DMA engine.
This patch provides a thread unsafe API to return the number of
inflight packets for a vhost queue without using any lock.

Signed-off-by: Xuan Ding <xuan.ding@intel.com>
---
 doc/guides/prog_guide/vhost_lib.rst    |  5 +++++
 doc/guides/rel_notes/release_22_03.rst |  4 ++++
 lib/vhost/rte_vhost_async.h            | 14 ++++++++++++++
 lib/vhost/version.map                  |  1 +
 lib/vhost/vhost.c                      | 26 ++++++++++++++++++++++++++
 5 files changed, 50 insertions(+)

diff --git a/doc/guides/prog_guide/vhost_lib.rst b/doc/guides/prog_guide/vhost_lib.rst
index 886f8f5e72..f95288d128 100644
--- a/doc/guides/prog_guide/vhost_lib.rst
+++ b/doc/guides/prog_guide/vhost_lib.rst
@@ -271,6 +271,11 @@ The following is an overview of some key Vhost API functions:
   This function returns the amount of in-flight packets for the vhost
   queue using async acceleration.
 
+ * ``rte_vhost_async_get_inflight_thread_unsafe(vid, queue_id)``
+
+  Get the number of inflight packets for a vhost queue without
+  performing any locking.
+
 * ``rte_vhost_clear_queue_thread_unsafe(vid, queue_id, **pkts, count, dma_id, vchan_id)``
 
   Clear inflight packets which are submitted to DMA engine in vhost async data
diff --git a/doc/guides/rel_notes/release_22_03.rst b/doc/guides/rel_notes/release_22_03.rst
index ff3095d742..37ef37bb20 100644
--- a/doc/guides/rel_notes/release_22_03.rst
+++ b/doc/guides/rel_notes/release_22_03.rst
@@ -149,6 +149,10 @@ New Features
   * Called ``rte_ipv4/6_udptcp_cksum_mbuf()`` functions in testpmd csum mode
     to support software UDP/TCP checksum over multiple segments.
 
+* **Added vhost API to get the number of inflight packets.**
+
+  Added an API which can get the number of inflight packets in
+  vhost async data path without using lock.
 
 Removed Items
 -------------
diff --git a/lib/vhost/rte_vhost_async.h b/lib/vhost/rte_vhost_async.h
index 838c4778cc..06b0b0a579 100644
--- a/lib/vhost/rte_vhost_async.h
+++ b/lib/vhost/rte_vhost_async.h
@@ -135,6 +135,20 @@ uint16_t rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id,
 __rte_experimental
 int rte_vhost_async_get_inflight(int vid, uint16_t queue_id);
 
+/**
+ * This function is lock-free version to return the amount of in-flight
+ * packets for the vhost queue which uses async channel acceleration.
+ *
+ * @param vid
+ * id of vhost device to enqueue data
+ * @param queue_id
+ * queue id to enqueue data
+ * @return
+ * the amount of in-flight packets on success; -1 on failure
+ */
+__rte_experimental
+int rte_vhost_async_get_inflight_thread_unsafe(int vid, uint16_t queue_id);
+
 /**
  * This function checks async completion status and clear packets for
  * a specific vhost device queue. Packets which are inflight will be
diff --git a/lib/vhost/version.map b/lib/vhost/version.map
index 1202ba9c1a..03b46cb10e 100644
--- a/lib/vhost/version.map
+++ b/lib/vhost/version.map
@@ -87,6 +87,7 @@ EXPERIMENTAL {
 
 	# added in 22.03
 	rte_vhost_async_dma_configure;
+	rte_vhost_async_get_inflight_thread_unsafe;
 };
 
 INTERNAL {
diff --git a/lib/vhost/vhost.c b/lib/vhost/vhost.c
index 6bcb716de0..33dca62c6c 100644
--- a/lib/vhost/vhost.c
+++ b/lib/vhost/vhost.c
@@ -1907,6 +1907,32 @@ rte_vhost_async_get_inflight(int vid, uint16_t queue_id)
 	return ret;
 }
 
+int
+rte_vhost_async_get_inflight_thread_unsafe(int vid, uint16_t queue_id)
+{
+	struct vhost_virtqueue *vq;
+	struct virtio_net *dev = get_device(vid);
+	int ret = -1;
+
+	if (dev == NULL)
+		return ret;
+
+	if (queue_id >= VHOST_MAX_VRING)
+		return ret;
+
+	vq = dev->virtqueue[queue_id];
+
+	if (vq == NULL)
+		return ret;
+
+	if (!vq->async)
+		return ret;
+
+	ret = vq->async->pkts_inflight_n;
+
+	return ret;
+}
+
 int
 rte_vhost_get_monitor_addr(int vid, uint16_t queue_id,
 		struct rte_vhost_power_monitor_cond *pmc)
-- 
2.17.1

Re: [RFC 1/2] vhost: add unsafe API to check inflight packets

[Maxime Coquelin]

Hi Xuan,

On 2/16/22 08:04, xuan.ding@intel.com wrote:
> From: Xuan Ding <xuan.ding@intel.com>
> 
> In async data path, when vring state changes or device is destroyed,
> it is necessary to know the number of inflight packets in DMA engine.
> This patch provides a thread unsafe API to return the number of
> inflight packets for a vhost queue without using any lock.
> 
> Signed-off-by: Xuan Ding <xuan.ding@intel.com>
> ---
>   doc/guides/prog_guide/vhost_lib.rst    |  5 +++++
>   doc/guides/rel_notes/release_22_03.rst |  4 ++++
>   lib/vhost/rte_vhost_async.h            | 14 ++++++++++++++
>   lib/vhost/version.map                  |  1 +
>   lib/vhost/vhost.c                      | 26 ++++++++++++++++++++++++++
>   5 files changed, 50 insertions(+)
> 
> diff --git a/doc/guides/prog_guide/vhost_lib.rst b/doc/guides/prog_guide/vhost_lib.rst
> index 886f8f5e72..f95288d128 100644
> --- a/doc/guides/prog_guide/vhost_lib.rst
> +++ b/doc/guides/prog_guide/vhost_lib.rst
> @@ -271,6 +271,11 @@ The following is an overview of some key Vhost API functions:
>     This function returns the amount of in-flight packets for the vhost
>     queue using async acceleration.
>   
> + * ``rte_vhost_async_get_inflight_thread_unsafe(vid, queue_id)``
> +
> +  Get the number of inflight packets for a vhost queue without
> +  performing any locking.

Here and below in the API doc, I would add that it should only be used
within the vhost ops, which already hold the lock.

> +
>   * ``rte_vhost_clear_queue_thread_unsafe(vid, queue_id, **pkts, count, dma_id, vchan_id)``
>   
>     Clear inflight packets which are submitted to DMA engine in vhost async data
> diff --git a/doc/guides/rel_notes/release_22_03.rst b/doc/guides/rel_notes/release_22_03.rst
> index ff3095d742..37ef37bb20 100644
> --- a/doc/guides/rel_notes/release_22_03.rst
> +++ b/doc/guides/rel_notes/release_22_03.rst
> @@ -149,6 +149,10 @@ New Features
>     * Called ``rte_ipv4/6_udptcp_cksum_mbuf()`` functions in testpmd csum mode
>       to support software UDP/TCP checksum over multiple segments.
>   
> +* **Added vhost API to get the number of inflight packets.**
> +
> +  Added an API which can get the number of inflight packets in
> +  vhost async data path without using lock.
>   
>   Removed Items
>   -------------
> diff --git a/lib/vhost/rte_vhost_async.h b/lib/vhost/rte_vhost_async.h
> index 838c4778cc..06b0b0a579 100644
> --- a/lib/vhost/rte_vhost_async.h
> +++ b/lib/vhost/rte_vhost_async.h
> @@ -135,6 +135,20 @@ uint16_t rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id,
>   __rte_experimental
>   int rte_vhost_async_get_inflight(int vid, uint16_t queue_id);
>   
> +/**
> + * This function is lock-free version to return the amount of in-flight
> + * packets for the vhost queue which uses async channel acceleration.
> + *
> + * @param vid
> + * id of vhost device to enqueue data
> + * @param queue_id
> + * queue id to enqueue data
> + * @return
> + * the amount of in-flight packets on success; -1 on failure
> + */
> +__rte_experimental
> +int rte_vhost_async_get_inflight_thread_unsafe(int vid, uint16_t queue_id);
> +
>   /**
>    * This function checks async completion status and clear packets for
>    * a specific vhost device queue. Packets which are inflight will be
> diff --git a/lib/vhost/version.map b/lib/vhost/version.map
> index 1202ba9c1a..03b46cb10e 100644
> --- a/lib/vhost/version.map
> +++ b/lib/vhost/version.map
> @@ -87,6 +87,7 @@ EXPERIMENTAL {
>   
>   	# added in 22.03
>   	rte_vhost_async_dma_configure;
> +	rte_vhost_async_get_inflight_thread_unsafe;

Don't forget to move it to "added in 22.07".

>   };
>   
>   INTERNAL {
> diff --git a/lib/vhost/vhost.c b/lib/vhost/vhost.c
> index 6bcb716de0..33dca62c6c 100644
> --- a/lib/vhost/vhost.c
> +++ b/lib/vhost/vhost.c
> @@ -1907,6 +1907,32 @@ rte_vhost_async_get_inflight(int vid, uint16_t queue_id)
>   	return ret;
>   }
>   
> +int
> +rte_vhost_async_get_inflight_thread_unsafe(int vid, uint16_t queue_id)
> +{
> +	struct vhost_virtqueue *vq;
> +	struct virtio_net *dev = get_device(vid);
> +	int ret = -1;
> +
> +	if (dev == NULL)
> +		return ret;
> +
> +	if (queue_id >= VHOST_MAX_VRING)
> +		return ret;
> +
> +	vq = dev->virtqueue[queue_id];
> +
> +	if (vq == NULL)
> +		return ret;
> +
> +	if (!vq->async)
> +		return ret;
> +
> +	ret = vq->async->pkts_inflight_n;
> +
> +	return ret;
> +}
> +
>   int
>   rte_vhost_get_monitor_addr(int vid, uint16_t queue_id,
>   		struct rte_vhost_power_monitor_cond *pmc)

RE: [RFC 1/2] vhost: add unsafe API to check inflight packets

[Ding, Xuan]

Hi Maxime,

> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin@redhat.com>
> Sent: Monday, March 28, 2022 11:05 PM
> To: Ding, Xuan <xuan.ding@intel.com>; Xia, Chenbo <chenbo.xia@intel.com>
> Cc: dev@dpdk.org; Hu, Jiayu <jiayu.hu@intel.com>; Pai G, Sunil
> <sunil.pai.g@intel.com>
> Subject: Re: [RFC 1/2] vhost: add unsafe API to check inflight packets
> 
> Hi Xuan,
> 
> On 2/16/22 08:04, xuan.ding@intel.com wrote:
> > From: Xuan Ding <xuan.ding@intel.com>
> >
> > In async data path, when vring state changes or device is destroyed,
> > it is necessary to know the number of inflight packets in DMA engine.
> > This patch provides a thread unsafe API to return the number of
> > inflight packets for a vhost queue without using any lock.
> >
> > Signed-off-by: Xuan Ding <xuan.ding@intel.com>
> > ---
> >   doc/guides/prog_guide/vhost_lib.rst    |  5 +++++
> >   doc/guides/rel_notes/release_22_03.rst |  4 ++++
> >   lib/vhost/rte_vhost_async.h            | 14 ++++++++++++++
> >   lib/vhost/version.map                  |  1 +
> >   lib/vhost/vhost.c                      | 26 ++++++++++++++++++++++++++
> >   5 files changed, 50 insertions(+)
> >
> > diff --git a/doc/guides/prog_guide/vhost_lib.rst
> > b/doc/guides/prog_guide/vhost_lib.rst
> > index 886f8f5e72..f95288d128 100644
> > --- a/doc/guides/prog_guide/vhost_lib.rst
> > +++ b/doc/guides/prog_guide/vhost_lib.rst
> > @@ -271,6 +271,11 @@ The following is an overview of some key Vhost
> API functions:
> >     This function returns the amount of in-flight packets for the vhost
> >     queue using async acceleration.
> >
> > + * ``rte_vhost_async_get_inflight_thread_unsafe(vid, queue_id)``
> > +
> > +  Get the number of inflight packets for a vhost queue without
> > + performing any locking.
> 
> Here and below in the API doc, I would add that it should only be used within
> the vhost ops, which already hold the lock.

Thanks for the revision on the doc.

> 
> > +
> >   * ``rte_vhost_clear_queue_thread_unsafe(vid, queue_id, **pkts,
> > count, dma_id, vchan_id)``
> >
> >     Clear inflight packets which are submitted to DMA engine in vhost
> > async data diff --git a/doc/guides/rel_notes/release_22_03.rst
> > b/doc/guides/rel_notes/release_22_03.rst
> > index ff3095d742..37ef37bb20 100644
> > --- a/doc/guides/rel_notes/release_22_03.rst
> > +++ b/doc/guides/rel_notes/release_22_03.rst
> > @@ -149,6 +149,10 @@ New Features
> >     * Called ``rte_ipv4/6_udptcp_cksum_mbuf()`` functions in testpmd csum
> mode
> >       to support software UDP/TCP checksum over multiple segments.
> >
> > +* **Added vhost API to get the number of inflight packets.**
> > +
> > +  Added an API which can get the number of inflight packets in  vhost
> > + async data path without using lock.
> >
> >   Removed Items
> >   -------------
> > diff --git a/lib/vhost/rte_vhost_async.h b/lib/vhost/rte_vhost_async.h
> > index 838c4778cc..06b0b0a579 100644
> > --- a/lib/vhost/rte_vhost_async.h
> > +++ b/lib/vhost/rte_vhost_async.h
> > @@ -135,6 +135,20 @@ uint16_t rte_vhost_poll_enqueue_completed(int
> vid, uint16_t queue_id,
> >   __rte_experimental
> >   int rte_vhost_async_get_inflight(int vid, uint16_t queue_id);
> >
> > +/**
> > + * This function is lock-free version to return the amount of
> > +in-flight
> > + * packets for the vhost queue which uses async channel acceleration.
> > + *
> > + * @param vid
> > + * id of vhost device to enqueue data
> > + * @param queue_id
> > + * queue id to enqueue data
> > + * @return
> > + * the amount of in-flight packets on success; -1 on failure  */
> > +__rte_experimental int rte_vhost_async_get_inflight_thread_unsafe(int
> > +vid, uint16_t queue_id);
> > +
> >   /**
> >    * This function checks async completion status and clear packets for
> >    * a specific vhost device queue. Packets which are inflight will be
> > diff --git a/lib/vhost/version.map b/lib/vhost/version.map index
> > 1202ba9c1a..03b46cb10e 100644
> > --- a/lib/vhost/version.map
> > +++ b/lib/vhost/version.map
> > @@ -87,6 +87,7 @@ EXPERIMENTAL {
> >
> >   	# added in 22.03
> >   	rte_vhost_async_dma_configure;
> > +	rte_vhost_async_get_inflight_thread_unsafe;
> 
> Don't forget to move it to "added in 22.07".

Sure, I will send a v1 version in 22.07.
Thanks for your comments.

Regards,
Xuan

> 
> >   };
> >
> >   INTERNAL {
> > diff --git a/lib/vhost/vhost.c b/lib/vhost/vhost.c index
> > 6bcb716de0..33dca62c6c 100644
> > --- a/lib/vhost/vhost.c
> > +++ b/lib/vhost/vhost.c
> > @@ -1907,6 +1907,32 @@ rte_vhost_async_get_inflight(int vid, uint16_t
> queue_id)
> >   	return ret;
> >   }
> >
> > +int
> > +rte_vhost_async_get_inflight_thread_unsafe(int vid, uint16_t
> > +queue_id) {
> > +	struct vhost_virtqueue *vq;
> > +	struct virtio_net *dev = get_device(vid);
> > +	int ret = -1;
> > +
> > +	if (dev == NULL)
> > +		return ret;
> > +
> > +	if (queue_id >= VHOST_MAX_VRING)
> > +		return ret;
> > +
> > +	vq = dev->virtqueue[queue_id];
> > +
> > +	if (vq == NULL)
> > +		return ret;
> > +
> > +	if (!vq->async)
> > +		return ret;
> > +
> > +	ret = vq->async->pkts_inflight_n;
> > +
> > +	return ret;
> > +}
> > +
> >   int
> >   rte_vhost_get_monitor_addr(int vid, uint16_t queue_id,
> >   		struct rte_vhost_power_monitor_cond *pmc)

[RFC 2/2] examples/vhost: use API to check inflight packets

[xuan.ding]

From: Xuan Ding <xuan.ding@intel.com>

In async data path, call rte_vhost_async_get_inflight_thread_unsafe()
API to directly return the number of inflight packets instead of
maintaining a local variable.

Signed-off-by: Xuan Ding <xuan.ding@intel.com>
---
 examples/vhost/main.c | 28 +++++++++++++++-------------
 examples/vhost/main.h |  1 -
 2 files changed, 15 insertions(+), 14 deletions(-)

diff --git a/examples/vhost/main.c b/examples/vhost/main.c
index 3e784f5c6f..ba7ab23f4e 100644
--- a/examples/vhost/main.c
+++ b/examples/vhost/main.c
@@ -972,10 +972,8 @@ complete_async_pkts(struct vhost_dev *vdev)
 
 	complete_count = rte_vhost_poll_enqueue_completed(vdev->vid,
 					VIRTIO_RXQ, p_cpl, MAX_PKT_BURST, dma_id, 0);
-	if (complete_count) {
+	if (complete_count)
 		free_pkts(p_cpl, complete_count);
-		__atomic_sub_fetch(&vdev->pkts_inflight, complete_count, __ATOMIC_SEQ_CST);
-	}
 
 }
 
@@ -1017,7 +1015,6 @@ drain_vhost(struct vhost_dev *vdev)
 
 		complete_async_pkts(vdev);
 		ret = rte_vhost_submit_enqueue_burst(vdev->vid, VIRTIO_RXQ, m, nr_xmit, dma_id, 0);
-		__atomic_add_fetch(&vdev->pkts_inflight, ret, __ATOMIC_SEQ_CST);
 
 		enqueue_fail = nr_xmit - ret;
 		if (enqueue_fail)
@@ -1346,7 +1343,6 @@ drain_eth_rx(struct vhost_dev *vdev)
 		complete_async_pkts(vdev);
 		enqueue_count = rte_vhost_submit_enqueue_burst(vdev->vid,
 					VIRTIO_RXQ, pkts, rx_count, dma_id, 0);
-		__atomic_add_fetch(&vdev->pkts_inflight, enqueue_count, __ATOMIC_SEQ_CST);
 
 		enqueue_fail = rx_count - enqueue_count;
 		if (enqueue_fail)
@@ -1518,14 +1514,17 @@ destroy_device(int vid)
 
 	if (dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled) {
 		uint16_t n_pkt = 0;
+		int pkts_inflight;
 		int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
-		struct rte_mbuf *m_cpl[vdev->pkts_inflight];
+		pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid, VIRTIO_RXQ);
+		struct rte_mbuf *m_cpl[pkts_inflight];
 
-		while (vdev->pkts_inflight) {
+		while (pkts_inflight) {
 			n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, VIRTIO_RXQ,
-						m_cpl, vdev->pkts_inflight, dma_id, 0);
+						m_cpl, pkts_inflight, dma_id, 0);
 			free_pkts(m_cpl, n_pkt);
-			__atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST);
+			pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid,
+										VIRTIO_RXQ);
 		}
 
 		rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ);
@@ -1629,14 +1628,17 @@ vring_state_changed(int vid, uint16_t queue_id, int enable)
 	if (dma_bind[vid].dmas[queue_id].async_enabled) {
 		if (!enable) {
 			uint16_t n_pkt = 0;
+			int pkts_inflight;
+			pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid, queue_id);
 			int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
-			struct rte_mbuf *m_cpl[vdev->pkts_inflight];
+			struct rte_mbuf *m_cpl[pkts_inflight];
 
-			while (vdev->pkts_inflight) {
+			while (pkts_inflight) {
 				n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, queue_id,
-							m_cpl, vdev->pkts_inflight, dma_id, 0);
+							m_cpl, pkts_inflight, dma_id, 0);
 				free_pkts(m_cpl, n_pkt);
-				__atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST);
+				pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid,
+											queue_id);
 			}
 		}
 	}
diff --git a/examples/vhost/main.h b/examples/vhost/main.h
index b4a453e77e..e7f395c3c9 100644
--- a/examples/vhost/main.h
+++ b/examples/vhost/main.h
@@ -52,7 +52,6 @@ struct vhost_dev {
 	uint64_t features;
 	size_t hdr_len;
 	uint16_t nr_vrings;
-	uint16_t pkts_inflight;
 	struct rte_vhost_memory *mem;
 	struct device_statistics stats;
 	TAILQ_ENTRY(vhost_dev) global_vdev_entry;
-- 
2.17.1

Re: [RFC 2/2] examples/vhost: use API to check inflight packets

[Maxime Coquelin]

Hi Xuan,

On 2/16/22 08:04, xuan.ding@intel.com wrote:
> From: Xuan Ding <xuan.ding@intel.com>
> 
> In async data path, call rte_vhost_async_get_inflight_thread_unsafe()
> API to directly return the number of inflight packets instead of
> maintaining a local variable.
> 
> Signed-off-by: Xuan Ding <xuan.ding@intel.com>
> ---
>   examples/vhost/main.c | 28 +++++++++++++++-------------
>   examples/vhost/main.h |  1 -
>   2 files changed, 15 insertions(+), 14 deletions(-)
> 

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime