[v5,1/2] vhost: support inflight share memory protocol feature
diff mbox series

Message ID 20190806182500.22320-2-jin.yu@intel.com
State Superseded, archived
Delegated to: Maxime Coquelin
Headers show
Series
  • vhost: support inflight share memory protocol feature
Related show

Checks

Context Check Description
ci/checkpatch warning coding style issues
ci/Intel-compilation success Compilation OK

Commit Message

Jin Yu Aug. 6, 2019, 6:24 p.m. UTC
This patch introduces two new messages VHOST_USER_GET_INFLIGHT_FD
and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
buffer between qemu and backend.

Firstly, qemu uses VHOST_USER_GET_INFLIGHT_FD to get the
shared buffer from backend. Then qemu should send it back
through VHOST_USER_SET_INFLIGHT_FD each time we start vhost-user.

This shared buffer is used to process inflight I/O when backend
reconnect.

Signed-off-by: Lin Li <lilin24@baidu.com>
Signed-off-by: Xun Ni <nixun@baidu.com>
Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
Signed-off-by: JinYu <jin.yu@intel.com>
---
v1 - specify the APIs are split-ring only
v2 - fix APIs and judge split or packed
v3 - Add rte_vhost_ prefix and fix one issue.
v4 - add the packed ring support
v5 - revise get_vring_base func depend on Tiwei's suggestion
---
 lib/librte_vhost/rte_vhost.h           | 255 ++++++++++++++-
 lib/librte_vhost/rte_vhost_version.map |  12 +
 lib/librte_vhost/vhost.c               | 396 +++++++++++++++++++++-
 lib/librte_vhost/vhost.h               |  61 ++--
 lib/librte_vhost/vhost_user.c          | 437 ++++++++++++++++++++++++-
 lib/librte_vhost/vhost_user.h          |  13 +-
 6 files changed, 1128 insertions(+), 46 deletions(-)

Comments

Tiwei Bie Aug. 26, 2019, 8:28 a.m. UTC | #1
On Wed, Aug 07, 2019 at 02:24:59AM +0800, JinYu wrote:
> This patch introduces two new messages VHOST_USER_GET_INFLIGHT_FD
> and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
> buffer between qemu and backend.
> 
> Firstly, qemu uses VHOST_USER_GET_INFLIGHT_FD to get the
> shared buffer from backend. Then qemu should send it back
> through VHOST_USER_SET_INFLIGHT_FD each time we start vhost-user.
> 
> This shared buffer is used to process inflight I/O when backend
> reconnect.
> 
> Signed-off-by: Lin Li <lilin24@baidu.com>
> Signed-off-by: Xun Ni <nixun@baidu.com>
> Signed-off-by: Yu Zhang <zhangyu31@baidu.com>
> Signed-off-by: JinYu <jin.yu@intel.com>

s/JinYu/Jin Yu/

> ---
> v1 - specify the APIs are split-ring only
> v2 - fix APIs and judge split or packed
> v3 - Add rte_vhost_ prefix and fix one issue.
> v4 - add the packed ring support
> v5 - revise get_vring_base func depend on Tiwei's suggestion
> ---
>  lib/librte_vhost/rte_vhost.h           | 255 ++++++++++++++-
>  lib/librte_vhost/rte_vhost_version.map |  12 +
>  lib/librte_vhost/vhost.c               | 396 +++++++++++++++++++++-
>  lib/librte_vhost/vhost.h               |  61 ++--
>  lib/librte_vhost/vhost_user.c          | 437 ++++++++++++++++++++++++-
>  lib/librte_vhost/vhost_user.h          |  13 +-
>  6 files changed, 1128 insertions(+), 46 deletions(-)

This patch is too big, please divide it into small patches.
E.g. rte_vhost_vq_is_packed() can be introduced in a separate patch.

> 
> diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
> index 0226b3eff..3f01429b1 100644
> --- a/lib/librte_vhost/rte_vhost.h
> +++ b/lib/librte_vhost/rte_vhost.h
> @@ -11,6 +11,7 @@
>   */
>  
>  #include <stdint.h>
> +#include <stdbool.h>
>  #include <sys/eventfd.h>
>  
>  #include <rte_memory.h>
> @@ -71,6 +72,10 @@ extern "C" {
>  #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER 11
>  #endif
>  
> +#ifndef VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
> +#define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD 12
> +#endif
> +
>  /** Indicate whether protocol features negotiation is supported. */
>  #ifndef VHOST_USER_F_PROTOCOL_FEATURES
>  #define VHOST_USER_F_PROTOCOL_FEATURES	30
> @@ -98,10 +103,92 @@ struct rte_vhost_memory {
>  	struct rte_vhost_mem_region regions[];
>  };
>  
> +struct inflight_desc_packed {
> +	uint8_t inflight;
> +	uint8_t padding;
> +	uint16_t next;
> +	uint16_t last;
> +	uint16_t num;
> +	uint64_t counter;
> +	uint16_t id;
> +	uint16_t flags;
> +	uint32_t len;
> +	uint64_t addr;
> +};

Why struct inflight_desc_split doesn't have to be part of
vhost API but struct inflight_desc_packed has to be?

> +
> +struct inflight_info_packed {
> +	uint64_t features;
> +	uint16_t version;
> +	uint16_t desc_num;
> +	uint16_t free_head;
> +	uint16_t old_free_head;
> +	uint16_t used_idx;
> +	uint16_t old_used_idx;
> +	uint8_t used_wrap_counter;
> +	uint8_t old_used_wrap_counter;
> +	uint8_t padding[7];
> +	struct inflight_desc_packed desc[0];
> +};
> +
> +struct rte_vhost_resubmit_desc {
> +	uint16_t index;
> +	uint64_t counter;
> +};
> +
> +struct rte_vhost_resubmit_info {
> +	struct rte_vhost_resubmit_desc	*resubmit_list;
> +	uint16_t resubmit_num;
> +};
> +
> +struct rte_vhost_ring_inflight {
> +	union {
> +		struct inflight_info_split *inflight_split;

struct inflight_info_split is used but not declared.

> +		struct inflight_info_packed *inflight_packed;
> +	};
> +
> +	struct rte_vhost_resubmit_info *resubmit_inflight;
> +};
> +
> +/*
> + * Declare below packed ring defines unconditionally
> + * as Kernel header might use different names.
> + */
> +#ifndef VIRTIO_F_RING_PACKED
> +#define VIRTIO_F_RING_PACKED 34
> +
> +#define VRING_DESC_F_AVAIL	(1ULL << 7)
> +#define VRING_DESC_F_USED	(1ULL << 15)

You shouldn't put above macros under VIRTIO_F_RING_PACKED as
kernel doesn't define them. And the build will be broken when
kernel defines VIRTIO_F_RING_PACKED.

Besides, it seems not a good idea to make them parts of vhost
API.

> +
> +struct vring_packed_desc {
> +	uint64_t addr;
> +	uint32_t len;
> +	uint16_t id;
> +	uint16_t flags;
> +};
> +
> +#define VRING_EVENT_F_ENABLE 0x0
> +#define VRING_EVENT_F_DISABLE 0x1
> +#define VRING_EVENT_F_DESC 0x2
> +struct vring_packed_desc_event {
> +	uint16_t off_wrap;
> +	uint16_t flags;
> +};

You just need to declare instead of defining
vring_packed_desc/vring_packed_desc_event in vhost header.

> +#endif
> +
>  struct rte_vhost_vring {
> -	struct vring_desc	*desc;
> -	struct vring_avail	*avail;
> -	struct vring_used	*used;
> +	union {
> +		struct vring_desc	*desc;
> +		struct vring_packed_desc *desc_packed;
> +	};
> +	union {
> +		struct vring_avail	*avail;
> +		struct vring_packed_desc_event *driver_event;
> +	};
> +	union {
> +		struct vring_used	*used;
> +		struct vring_packed_desc_event *device_event;
> +	};
> +
>  	uint64_t		log_guest_addr;
>  
>  	/** Deprecated, use rte_vhost_vring_call() instead. */
> @@ -603,6 +690,33 @@ uint16_t rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
>   */
>  int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
>  
> +/**
> + * Get vq is packed
> + *
> + * @param vid
> + *  vhost device ID
> + * @return
> + *  0 on success, -1 on failure

Return value should be true or false.

> + */
> +int __rte_experimental
> +rte_vhost_vq_is_packed(int vid);
> +
> +/**
> + * Get guest inflight vring info, including inflight ring and resubmit list.
> + *
> + * @param vid
> + *  vhost device ID
> + * @param vring_idx
> + *  vring index
> + * @param vring
> + *  the structure to hold the requested inflight vring info
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental

You should put the `__rte_experimental` tag at the beginning.
See this for more details:
http://git.dpdk.org/dpdk/commit/lib/librte_vhost?id=18218713bf4248c4c6b97a12231e7d59b8a86865

> +rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
> +	struct rte_vhost_ring_inflight *vring);
> +
>  /**
>   * Get guest vring info, including the vring address, vring size, etc.
>   *
> @@ -616,7 +730,7 @@ int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
>   *  0 on success, -1 on failure
>   */
>  int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
> -			      struct rte_vhost_vring *vring);
> +	struct rte_vhost_vring *vring);
>  
>  /**
>   * Notify the guest that used descriptors have been added to the vring.  This
> @@ -631,6 +745,112 @@ int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
>   */
>  int rte_vhost_vring_call(int vid, uint16_t vring_idx);
>  
> +/**
> + * set split inflight descriptor.

Please capitalize the first letter consistently.

> + *
> + * This function save descriptors that has been comsumed in available
> + * ring
> + *
> + * @param vid
> + *  vhost device ID
> + * @param vring_idx
> + *  vring index
> + * @param idx
> + *  inflight entry index
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental
> +rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
> +	uint16_t idx);
> +
> +/**
> + * set packed inflight descriptor and get corresponding inflight entry
> + *
> + * This function save descriptors that has been comsumed
> + *
> + * @param vid
> + *  vhost device ID
> + * @param vring_idx
> + *  vring index
> + * @param idx
> + *  inflight entry index
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental
> +rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
> +	uint16_t head, uint16_t last, uint16_t *inflight_entry);
> +
> +/**
> + * save the head of list that the last batch of used descriptors.
> + *
> + * @param vid
> + *  vhost device ID
> + * @param vring_idx
> + *  vring index
> + * @param idx
> + *  descriptor entry index
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental
> +rte_vhost_set_last_inflight_io_split(int vid,
> +	uint16_t vring_idx, uint16_t idx);
> +
> +/**
> + * update the inflight free_head, used_idx and used_wrap_counter.
> + *
> + * This function will update status first before updating descriptors
> + * to used
> + *
> + * @param vid
> + *  vhost device ID
> + * @param vring_idx
> + *  vring index
> + * @param idx
> + *  inflight entry index
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental
> +rte_vhost_set_last_inflight_io_packed(int vid,
> +	uint16_t vring_idx, uint16_t head);
> +
> +/**
> + * clear the split inflight status.
> + *
> + * @param vid
> + *  vhost device ID
> + * @param vring_idx
> + *  vring index
> + * @param last_used_idx
> + *  last used idx of used ring
> + * @param idx
> + *  inflight entry index
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental
> +rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
> +	uint16_t last_used_idx, uint16_t idx);
> +
> +/**
> + * clear the packed inflight status.
> + *
> + * @param vid
> + *  vhost device ID
> + * @param vring_idx
> + *  vring index
> + * @param head
> + *  inflight entry index
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental
> +rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
> +	uint16_t head);
> +
>  /**
>   * Get vhost RX queue avail count.
>   *
> @@ -656,7 +876,8 @@ uint32_t rte_vhost_rx_queue_count(int vid, uint16_t qid);
>   *  0 on success, -1 on failure
>   */
>  int __rte_experimental
> -rte_vhost_get_log_base(int vid, uint64_t *log_base, uint64_t *log_size);
> +rte_vhost_get_log_base(int vid, uint64_t *log_base,
> +	uint64_t *log_size);

This change isn't necessary.

>  
>  /**
>   * Get last_avail/used_idx of the vhost virtqueue
> @@ -676,6 +897,28 @@ int __rte_experimental
>  rte_vhost_get_vring_base(int vid, uint16_t queue_id,
>  		uint16_t *last_avail_idx, uint16_t *last_used_idx);
>  
> +/**
> + * Get last_avail/last_used of the vhost virtqueue
> + *
> + * This function is designed for the reconnection and it's specific for
> + * the packed ring as we can get the two parameters from the inflight
> + * queueregion
> + *
> + * @param vid
> + *  vhost device ID
> + * @param queue_id
> + *  vhost queue index
> + * @param last_avail_idx
> + *  vhost last_avail_idx to get
> + * @param last_used_idx
> + *  vhost last_used_idx to get
> + * @return
> + *  0 on success, -1 on failure
> + */
> +int __rte_experimental
> +rte_vhost_get_vring_base_from_inflight(int vid,
> +	uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx);
> +
>  /**
>   * Set last_avail/used_idx of the vhost virtqueue
>   *
> @@ -692,7 +935,7 @@ rte_vhost_get_vring_base(int vid, uint16_t queue_id,
>   */
>  int __rte_experimental
>  rte_vhost_set_vring_base(int vid, uint16_t queue_id,
> -		uint16_t last_avail_idx, uint16_t last_used_idx);
> +	uint16_t last_avail_idx, uint16_t last_used_idx);
>  
>  /**
>   * Register external message handling callbacks
> diff --git a/lib/librte_vhost/rte_vhost_version.map b/lib/librte_vhost/rte_vhost_version.map
> index 5f1d4a75c..99f1134ea 100644
> --- a/lib/librte_vhost/rte_vhost_version.map
> +++ b/lib/librte_vhost/rte_vhost_version.map
> @@ -87,4 +87,16 @@ EXPERIMENTAL {
>  	rte_vdpa_relay_vring_used;
>  	rte_vhost_extern_callback_register;
>  	rte_vhost_driver_set_protocol_features;
> +	rte_vhost_set_inflight_desc_split;
> +	rte_vhost_clr_inflight_desc_split;
> +	rte_vhost_set_last_inflight_io_split;
> +	rte_vhost_get_vhost_ring_inflight;
> +	rte_vhost_vq_is_packed;
> +	rte_vhost_set_inflight_desc_packed;
> +	rte_vhost_clr_inflight_desc_packed;
> +	rte_vhost_set_last_inflight_io_packed;
> +	rte_vhost_get_vring_base_counter;
> +	rte_vhost_get_vring_base_from_inflight;
> +	rte_vhost_get_vring_base_counter_from_inflight;
> +	rte_vhost_set_vring_base_counter;
>  };



> @@ -939,13 +1270,48 @@ int rte_vhost_get_log_base(int vid, uint64_t *log_base,
>  int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
>  		uint16_t *last_avail_idx, uint16_t *last_used_idx)
>  {
> +	struct vhost_virtqueue *vq;
>  	struct virtio_net *dev = get_device(vid);
>  
>  	if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
>  		return -1;
>  
> -	*last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
> -	*last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
> +	vq = dev->virtqueue[queue_id];
> +	if (!vq)
> +		return -1;
> +
> +	if (vq_is_packed(dev)) {
> +		*last_avail_idx = (vq->avail_wrap_counter << 15) |
> +					vq->last_avail_idx;
> +		*last_used_idx = (vq->used_wrap_counter << 15) |
> +					vq->last_used_idx;
> +	} else {
> +		*last_avail_idx = vq->last_avail_idx;
> +		*last_used_idx = vq->last_used_idx;
> +	}

This should be a fix. Without this change, the user of this API
can't use packed ring properly.

Please do this in a separate patch, add a fixes line and Cc stable.

> +
> +	return 0;
> +}
> +
> +int rte_vhost_get_vring_base_from_inflight(int vid,
> +	uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx)
> +{
> +	struct inflight_info_packed *inflight_info;
> +	struct virtio_net *dev = get_device(vid);
> +
> +	if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
> +		return -1;
> +
> +	if (!vq_is_packed(dev))
> +		return -1;
> +
> +	inflight_info = dev->virtqueue[queue_id]->inflight_packed;
> +	if (!inflight_info)
> +		return -1;
> +
> +	*last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
> +				inflight_info->old_used_idx;
> +	*last_used_idx = *last_avail_idx;
>  
>  	return 0;
>  }
> @@ -953,13 +1319,25 @@ int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
>  int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
>  		uint16_t last_avail_idx, uint16_t last_used_idx)
>  {
> +	struct vhost_virtqueue *vq;
>  	struct virtio_net *dev = get_device(vid);
>  
>  	if (!dev)
>  		return -1;
>  
> -	dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
> -	dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
> +	vq = dev->virtqueue[queue_id];
> +	if (!vq)
> +		return -1;
> +
> +	if (vq_is_packed(dev)) {
> +		vq->last_avail_idx = last_avail_idx & 0x7fff;
> +		vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
> +		vq->last_used_idx = last_used_idx & 0x7fff;
> +		vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
> +	} else {
> +		vq->last_avail_idx = last_avail_idx;
> +		vq->last_used_idx = last_used_idx;
> +	}

Ditto.

>  
>  	return 0;
>  }
> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
> index 884befa85..e9d0b983d 100644
> --- a/lib/librte_vhost/vhost.h
> +++ b/lib/librte_vhost/vhost.h
> @@ -88,6 +88,22 @@ struct vring_used_elem_packed {
>  	uint32_t count;
>  };
>  
> +struct inflight_desc_split {
> +	uint8_t		inflight;
> +	uint8_t		padding[5];
> +	uint16_t	next;
> +	uint64_t	counter;
> +};
> +
> +struct inflight_info_split {
> +	uint64_t		features;
> +	uint16_t		version;
> +	uint16_t		desc_num;
> +	uint16_t		last_inflight_io;
> +	uint16_t		used_idx;
> +	struct inflight_desc_split desc[0];
> +};
> +

Patch
diff mbox series

diff --git a/lib/librte_vhost/rte_vhost.h b/lib/librte_vhost/rte_vhost.h
index 0226b3eff..3f01429b1 100644
--- a/lib/librte_vhost/rte_vhost.h
+++ b/lib/librte_vhost/rte_vhost.h
@@ -11,6 +11,7 @@ 
  */
 
 #include <stdint.h>
+#include <stdbool.h>
 #include <sys/eventfd.h>
 
 #include <rte_memory.h>
@@ -71,6 +72,10 @@  extern "C" {
 #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER 11
 #endif
 
+#ifndef VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
+#define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD 12
+#endif
+
 /** Indicate whether protocol features negotiation is supported. */
 #ifndef VHOST_USER_F_PROTOCOL_FEATURES
 #define VHOST_USER_F_PROTOCOL_FEATURES	30
@@ -98,10 +103,92 @@  struct rte_vhost_memory {
 	struct rte_vhost_mem_region regions[];
 };
 
+struct inflight_desc_packed {
+	uint8_t inflight;
+	uint8_t padding;
+	uint16_t next;
+	uint16_t last;
+	uint16_t num;
+	uint64_t counter;
+	uint16_t id;
+	uint16_t flags;
+	uint32_t len;
+	uint64_t addr;
+};
+
+struct inflight_info_packed {
+	uint64_t features;
+	uint16_t version;
+	uint16_t desc_num;
+	uint16_t free_head;
+	uint16_t old_free_head;
+	uint16_t used_idx;
+	uint16_t old_used_idx;
+	uint8_t used_wrap_counter;
+	uint8_t old_used_wrap_counter;
+	uint8_t padding[7];
+	struct inflight_desc_packed desc[0];
+};
+
+struct rte_vhost_resubmit_desc {
+	uint16_t index;
+	uint64_t counter;
+};
+
+struct rte_vhost_resubmit_info {
+	struct rte_vhost_resubmit_desc	*resubmit_list;
+	uint16_t resubmit_num;
+};
+
+struct rte_vhost_ring_inflight {
+	union {
+		struct inflight_info_split *inflight_split;
+		struct inflight_info_packed *inflight_packed;
+	};
+
+	struct rte_vhost_resubmit_info *resubmit_inflight;
+};
+
+/*
+ * Declare below packed ring defines unconditionally
+ * as Kernel header might use different names.
+ */
+#ifndef VIRTIO_F_RING_PACKED
+#define VIRTIO_F_RING_PACKED 34
+
+#define VRING_DESC_F_AVAIL	(1ULL << 7)
+#define VRING_DESC_F_USED	(1ULL << 15)
+
+struct vring_packed_desc {
+	uint64_t addr;
+	uint32_t len;
+	uint16_t id;
+	uint16_t flags;
+};
+
+#define VRING_EVENT_F_ENABLE 0x0
+#define VRING_EVENT_F_DISABLE 0x1
+#define VRING_EVENT_F_DESC 0x2
+struct vring_packed_desc_event {
+	uint16_t off_wrap;
+	uint16_t flags;
+};
+#endif
+
 struct rte_vhost_vring {
-	struct vring_desc	*desc;
-	struct vring_avail	*avail;
-	struct vring_used	*used;
+	union {
+		struct vring_desc	*desc;
+		struct vring_packed_desc *desc_packed;
+	};
+	union {
+		struct vring_avail	*avail;
+		struct vring_packed_desc_event *driver_event;
+	};
+	union {
+		struct vring_used	*used;
+		struct vring_packed_desc_event *device_event;
+	};
+
 	uint64_t		log_guest_addr;
 
 	/** Deprecated, use rte_vhost_vring_call() instead. */
@@ -603,6 +690,33 @@  uint16_t rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
  */
 int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
 
+/**
+ * Get vq is packed
+ *
+ * @param vid
+ *  vhost device ID
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_vq_is_packed(int vid);
+
+/**
+ * Get guest inflight vring info, including inflight ring and resubmit list.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param vring
+ *  the structure to hold the requested inflight vring info
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
+	struct rte_vhost_ring_inflight *vring);
+
 /**
  * Get guest vring info, including the vring address, vring size, etc.
  *
@@ -616,7 +730,7 @@  int rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem);
  *  0 on success, -1 on failure
  */
 int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
-			      struct rte_vhost_vring *vring);
+	struct rte_vhost_vring *vring);
 
 /**
  * Notify the guest that used descriptors have been added to the vring.  This
@@ -631,6 +745,112 @@  int rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
  */
 int rte_vhost_vring_call(int vid, uint16_t vring_idx);
 
+/**
+ * set split inflight descriptor.
+ *
+ * This function save descriptors that has been comsumed in available
+ * ring
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param idx
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
+	uint16_t idx);
+
+/**
+ * set packed inflight descriptor and get corresponding inflight entry
+ *
+ * This function save descriptors that has been comsumed
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param idx
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head, uint16_t last, uint16_t *inflight_entry);
+
+/**
+ * save the head of list that the last batch of used descriptors.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param idx
+ *  descriptor entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_set_last_inflight_io_split(int vid,
+	uint16_t vring_idx, uint16_t idx);
+
+/**
+ * update the inflight free_head, used_idx and used_wrap_counter.
+ *
+ * This function will update status first before updating descriptors
+ * to used
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param idx
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_set_last_inflight_io_packed(int vid,
+	uint16_t vring_idx, uint16_t head);
+
+/**
+ * clear the split inflight status.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param last_used_idx
+ *  last used idx of used ring
+ * @param idx
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
+	uint16_t last_used_idx, uint16_t idx);
+
+/**
+ * clear the packed inflight status.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @param head
+ *  inflight entry index
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head);
+
 /**
  * Get vhost RX queue avail count.
  *
@@ -656,7 +876,8 @@  uint32_t rte_vhost_rx_queue_count(int vid, uint16_t qid);
  *  0 on success, -1 on failure
  */
 int __rte_experimental
-rte_vhost_get_log_base(int vid, uint64_t *log_base, uint64_t *log_size);
+rte_vhost_get_log_base(int vid, uint64_t *log_base,
+	uint64_t *log_size);
 
 /**
  * Get last_avail/used_idx of the vhost virtqueue
@@ -676,6 +897,28 @@  int __rte_experimental
 rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 		uint16_t *last_avail_idx, uint16_t *last_used_idx);
 
+/**
+ * Get last_avail/last_used of the vhost virtqueue
+ *
+ * This function is designed for the reconnection and it's specific for
+ * the packed ring as we can get the two parameters from the inflight
+ * queueregion
+ *
+ * @param vid
+ *  vhost device ID
+ * @param queue_id
+ *  vhost queue index
+ * @param last_avail_idx
+ *  vhost last_avail_idx to get
+ * @param last_used_idx
+ *  vhost last_used_idx to get
+ * @return
+ *  0 on success, -1 on failure
+ */
+int __rte_experimental
+rte_vhost_get_vring_base_from_inflight(int vid,
+	uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx);
+
 /**
  * Set last_avail/used_idx of the vhost virtqueue
  *
@@ -692,7 +935,7 @@  rte_vhost_get_vring_base(int vid, uint16_t queue_id,
  */
 int __rte_experimental
 rte_vhost_set_vring_base(int vid, uint16_t queue_id,
-		uint16_t last_avail_idx, uint16_t last_used_idx);
+	uint16_t last_avail_idx, uint16_t last_used_idx);
 
 /**
  * Register external message handling callbacks
diff --git a/lib/librte_vhost/rte_vhost_version.map b/lib/librte_vhost/rte_vhost_version.map
index 5f1d4a75c..99f1134ea 100644
--- a/lib/librte_vhost/rte_vhost_version.map
+++ b/lib/librte_vhost/rte_vhost_version.map
@@ -87,4 +87,16 @@  EXPERIMENTAL {
 	rte_vdpa_relay_vring_used;
 	rte_vhost_extern_callback_register;
 	rte_vhost_driver_set_protocol_features;
+	rte_vhost_set_inflight_desc_split;
+	rte_vhost_clr_inflight_desc_split;
+	rte_vhost_set_last_inflight_io_split;
+	rte_vhost_get_vhost_ring_inflight;
+	rte_vhost_vq_is_packed;
+	rte_vhost_set_inflight_desc_packed;
+	rte_vhost_clr_inflight_desc_packed;
+	rte_vhost_set_last_inflight_io_packed;
+	rte_vhost_get_vring_base_counter;
+	rte_vhost_get_vring_base_from_inflight;
+	rte_vhost_get_vring_base_counter_from_inflight;
+	rte_vhost_set_vring_base_counter;
 };
diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 981837b5d..751817d30 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -242,6 +242,31 @@  cleanup_vq(struct vhost_virtqueue *vq, int destroy)
 		close(vq->kickfd);
 }
 
+void
+cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+	if (!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
+		return;
+
+	if (vq_is_packed(dev)) {
+		if (vq->inflight_packed)
+			vq->inflight_packed = NULL;
+	} else {
+		if (vq->inflight_split)
+			vq->inflight_split = NULL;
+	}
+
+	if (vq->resubmit_inflight) {
+		if (vq->resubmit_inflight->resubmit_list) {
+			free(vq->resubmit_inflight->resubmit_list);
+			vq->resubmit_inflight->resubmit_list = NULL;
+		}
+		free(vq->resubmit_inflight);
+		vq->resubmit_inflight = NULL;
+	}
+}
+
 /*
  * Unmap any memory, close any file descriptors and
  * free any memory owned by a device.
@@ -253,8 +278,10 @@  cleanup_device(struct virtio_net *dev, int destroy)
 
 	vhost_backend_cleanup(dev);
 
-	for (i = 0; i < dev->nr_vring; i++)
+	for (i = 0; i < dev->nr_vring; i++) {
 		cleanup_vq(dev->virtqueue[i], destroy);
+		cleanup_vq_inflight(dev, dev->virtqueue[i]);
+	}
 }
 
 void
@@ -726,6 +753,41 @@  rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
 	return 0;
 }
 
+int
+rte_vhost_get_vhost_ring_inflight(int vid, uint16_t vring_idx,
+			  struct rte_vhost_ring_inflight *vring)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (dev == NULL || vring == NULL)
+		return -1;
+
+	if (vring_idx >= VHOST_MAX_VRING)
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (vq_is_packed(dev)) {
+		if (unlikely(!vq->inflight_packed))
+			return -1;
+
+		vring->inflight_packed = vq->inflight_packed;
+	} else {
+		if (unlikely(!vq->inflight_split))
+			return -1;
+
+		vring->inflight_split = vq->inflight_split;
+	}
+
+	vring->resubmit_inflight = vq->resubmit_inflight;
+
+	return 0;
+}
+
 int
 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 			  struct rte_vhost_vring *vring)
@@ -744,11 +806,17 @@  rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
 	if (!vq)
 		return -1;
 
-	vring->desc  = vq->desc;
-	vring->avail = vq->avail;
-	vring->used  = vq->used;
-	vring->log_guest_addr  = vq->log_guest_addr;
+	if (vq_is_packed(dev)) {
+		vring->desc_packed  = vq->desc_packed;
+		vring->driver_event = vq->driver_event;
+		vring->device_event  = vq->device_event;
+	} else {
+		vring->desc  = vq->desc;
+		vring->avail = vq->avail;
+		vring->used  = vq->used;
+	}
 
+	vring->log_guest_addr  = vq->log_guest_addr;
 	vring->callfd  = vq->callfd;
 	vring->kickfd  = vq->kickfd;
 	vring->size    = vq->size;
@@ -781,6 +849,269 @@  rte_vhost_vring_call(int vid, uint16_t vring_idx)
 	return 0;
 }
 
+int
+rte_vhost_vq_is_packed(int vid)
+{
+	struct virtio_net *dev;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	return vq_is_packed(dev);
+}
+
+int
+rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx, uint16_t idx)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (unlikely(!vq->inflight_split))
+		return -1;
+
+	vq->inflight_split->desc[idx].counter = vq->global_counter++;
+	vq->inflight_split->desc[idx].inflight = 1;
+	return 0;
+}
+
+int
+rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head, uint16_t last, uint16_t *inflight_entry)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+	struct inflight_info_packed *inflight_info;
+	struct vring_packed_desc *desc;
+	uint16_t old_free_head, free_head;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(!vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	inflight_info = vq->inflight_packed;
+	if (unlikely(!inflight_info))
+		return -1;
+
+	desc = vq->desc_packed;
+	old_free_head = inflight_info->old_free_head;
+	free_head = old_free_head;
+
+	/* init header descriptor */
+	inflight_info->desc[old_free_head].num = 0;
+	inflight_info->desc[old_free_head].counter = vq->global_counter++;
+	inflight_info->desc[old_free_head].inflight = 1;
+
+	/* save desc entry in flight entry */
+	while (head != ((last + 1) % vq->size)) {
+		inflight_info->desc[old_free_head].num++;
+		inflight_info->desc[free_head].addr = desc[head].addr;
+		inflight_info->desc[free_head].len = desc[head].len;
+		inflight_info->desc[free_head].flags = desc[head].flags;
+		inflight_info->desc[free_head].id = desc[head].id;
+
+		inflight_info->desc[old_free_head].last = free_head;
+		free_head = inflight_info->desc[free_head].next;
+		inflight_info->free_head = free_head;
+		head = (head + 1) % vq->size;
+	}
+
+	inflight_info->old_free_head = free_head;
+	*inflight_entry = old_free_head;
+
+	return 0;
+}
+
+int
+rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
+	uint16_t last_used_idx, uint16_t idx)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (unlikely(!vq->inflight_split))
+		return -1;
+
+	rte_compiler_barrier();
+
+	vq->inflight_split->desc[idx].inflight = 0;
+
+	rte_compiler_barrier();
+
+	vq->inflight_split->used_idx = last_used_idx;
+	return 0;
+}
+
+int
+rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
+	uint16_t head)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+	struct inflight_info_packed *inflight_info;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(!vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	inflight_info = vq->inflight_packed;
+	if (unlikely(!inflight_info))
+		return -1;
+
+	rte_compiler_barrier();
+
+	inflight_info->desc[head].inflight = 0;
+
+	rte_compiler_barrier();
+
+	inflight_info->old_free_head = inflight_info->free_head;
+	inflight_info->old_used_idx = inflight_info->used_idx;
+	inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
+
+	return 0;
+}
+
+int
+rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
+	uint16_t idx)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	if (unlikely(!vq->inflight_split))
+		return -1;
+
+	vq->inflight_split->last_inflight_io = idx;
+	return 0;
+}
+
+int
+rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
+	uint16_t head)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+	struct inflight_info_packed *inflight_info;
+	uint16_t last;
+
+	dev = get_device(vid);
+	if (unlikely(!dev))
+		return -1;
+
+	if (unlikely(!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
+		return 0;
+
+	if (unlikely(!vq_is_packed(dev)))
+		return -1;
+
+	if (unlikely(vring_idx >= VHOST_MAX_VRING))
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (unlikely(!vq))
+		return -1;
+
+	inflight_info = vq->inflight_packed;
+	if (unlikely(!inflight_info))
+		return -1;
+
+	last = inflight_info->desc[head].last;
+
+	inflight_info->desc[last].next = inflight_info->free_head;
+	inflight_info->free_head = head;
+	inflight_info->used_idx += inflight_info->desc[head].num;
+	if (inflight_info->used_idx >= inflight_info->desc_num) {
+		inflight_info->used_idx -= inflight_info->desc_num;
+		inflight_info->used_wrap_counter =
+			!inflight_info->used_wrap_counter;
+	}
+
+	return 0;
+}
+
 uint16_t
 rte_vhost_avail_entries(int vid, uint16_t queue_id)
 {
@@ -939,13 +1270,48 @@  int rte_vhost_get_log_base(int vid, uint64_t *log_base,
 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 		uint16_t *last_avail_idx, uint16_t *last_used_idx)
 {
+	struct vhost_virtqueue *vq;
 	struct virtio_net *dev = get_device(vid);
 
 	if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
 		return -1;
 
-	*last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
-	*last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
+	vq = dev->virtqueue[queue_id];
+	if (!vq)
+		return -1;
+
+	if (vq_is_packed(dev)) {
+		*last_avail_idx = (vq->avail_wrap_counter << 15) |
+					vq->last_avail_idx;
+		*last_used_idx = (vq->used_wrap_counter << 15) |
+					vq->last_used_idx;
+	} else {
+		*last_avail_idx = vq->last_avail_idx;
+		*last_used_idx = vq->last_used_idx;
+	}
+
+	return 0;
+}
+
+int rte_vhost_get_vring_base_from_inflight(int vid,
+	uint16_t queue_id, uint16_t *last_avail_idx, uint16_t *last_used_idx)
+{
+	struct inflight_info_packed *inflight_info;
+	struct virtio_net *dev = get_device(vid);
+
+	if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
+		return -1;
+
+	if (!vq_is_packed(dev))
+		return -1;
+
+	inflight_info = dev->virtqueue[queue_id]->inflight_packed;
+	if (!inflight_info)
+		return -1;
+
+	*last_avail_idx = (inflight_info->old_used_wrap_counter << 15) |
+				inflight_info->old_used_idx;
+	*last_used_idx = *last_avail_idx;
 
 	return 0;
 }
@@ -953,13 +1319,25 @@  int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
 		uint16_t last_avail_idx, uint16_t last_used_idx)
 {
+	struct vhost_virtqueue *vq;
 	struct virtio_net *dev = get_device(vid);
 
 	if (!dev)
 		return -1;
 
-	dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
-	dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
+	vq = dev->virtqueue[queue_id];
+	if (!vq)
+		return -1;
+
+	if (vq_is_packed(dev)) {
+		vq->last_avail_idx = last_avail_idx & 0x7fff;
+		vq->avail_wrap_counter = !!(last_avail_idx & (1 << 15));
+		vq->last_used_idx = last_used_idx & 0x7fff;
+		vq->used_wrap_counter = !!(last_used_idx & (1 << 15));
+	} else {
+		vq->last_avail_idx = last_avail_idx;
+		vq->last_used_idx = last_used_idx;
+	}
 
 	return 0;
 }
diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index 884befa85..e9d0b983d 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -88,6 +88,22 @@  struct vring_used_elem_packed {
 	uint32_t count;
 };
 
+struct inflight_desc_split {
+	uint8_t		inflight;
+	uint8_t		padding[5];
+	uint16_t	next;
+	uint64_t	counter;
+};
+
+struct inflight_info_split {
+	uint64_t		features;
+	uint16_t		version;
+	uint16_t		desc_num;
+	uint16_t		last_inflight_io;
+	uint16_t		used_idx;
+	struct inflight_desc_split desc[0];
+};
+
 /**
  * Structure contains variables relevant to RX/TX virtqueues.
  */
@@ -128,6 +144,14 @@  struct vhost_virtqueue {
 	/* Physical address of used ring, for logging */
 	uint64_t		log_guest_addr;
 
+	/* inflight share memory info */
+	union {
+		struct inflight_info_split *inflight_split;
+		struct inflight_info_packed *inflight_packed;
+	};
+	struct rte_vhost_resubmit_info *resubmit_inflight;
+	uint64_t		global_counter;
+
 	uint16_t		nr_zmbuf;
 	uint16_t		zmbuf_size;
 	uint16_t		last_zmbuf_idx;
@@ -215,35 +239,6 @@  struct vhost_msg {
  #define VIRTIO_F_VERSION_1 32
 #endif
 
-/* Declare packed ring related bits for older kernels */
-#ifndef VIRTIO_F_RING_PACKED
-
-#define VIRTIO_F_RING_PACKED 34
-
-struct vring_packed_desc {
-	uint64_t addr;
-	uint32_t len;
-	uint16_t id;
-	uint16_t flags;
-};
-
-struct vring_packed_desc_event {
-	uint16_t off_wrap;
-	uint16_t flags;
-};
-#endif
-
-/*
- * Declare below packed ring defines unconditionally
- * as Kernel header might use different names.
- */
-#define VRING_DESC_F_AVAIL	(1ULL << 7)
-#define VRING_DESC_F_USED	(1ULL << 15)
-
-#define VRING_EVENT_F_ENABLE 0x0
-#define VRING_EVENT_F_DISABLE 0x1
-#define VRING_EVENT_F_DESC 0x2
-
 /*
  * Available and used descs are in same order
  */
@@ -286,6 +281,12 @@  struct guest_page {
 	uint64_t size;
 };
 
+struct inflight_mem_info {
+	int		fd;
+	void		*addr;
+	uint64_t	size;
+};
+
 /**
  * Device structure contains all configuration information relating
  * to the device.
@@ -303,6 +304,7 @@  struct virtio_net {
 	uint32_t		nr_vring;
 	int			dequeue_zero_copy;
 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
+	struct inflight_mem_info inflight_info;
 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
 	char			ifname[IF_NAME_SZ];
 	uint64_t		log_size;
@@ -467,6 +469,7 @@  void vhost_destroy_device(int);
 void vhost_destroy_device_notify(struct virtio_net *dev);
 
 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
+void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
 
 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
index c9e29ece8..4984dd874 100644
--- a/lib/librte_vhost/vhost_user.c
+++ b/lib/librte_vhost/vhost_user.c
@@ -31,6 +31,8 @@ 
 #include <sys/stat.h>
 #include <sys/syscall.h>
 #include <assert.h>
+#include <sys/syscall.h>
+#include <asm/unistd.h>
 #ifdef RTE_LIBRTE_VHOST_NUMA
 #include <numaif.h>
 #endif
@@ -49,6 +51,15 @@ 
 #define VIRTIO_MIN_MTU 68
 #define VIRTIO_MAX_MTU 65535
 
+#define INFLIGHT_ALIGNMENT	64
+#define INFLIGHT_VERSION	0xabcd
+#define VIRTQUEUE_MAX_SIZE	1024
+
+#define CLOEXEC	0x0001U
+
+#define ALIGN_DOWN(n, m) ((n) / (m) * (m))
+#define ALIGN_UP(n, m) ALIGN_DOWN((n) + (m) - 1, (m))
+
 static const char *vhost_message_str[VHOST_USER_MAX] = {
 	[VHOST_USER_NONE] = "VHOST_USER_NONE",
 	[VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
@@ -78,6 +89,8 @@  static const char *vhost_message_str[VHOST_USER_MAX] = {
 	[VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
 	[VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
 	[VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
+	[VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
+	[VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
 };
 
 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
@@ -160,6 +173,16 @@  vhost_backend_cleanup(struct virtio_net *dev)
 		dev->log_addr = 0;
 	}
 
+	if (dev->inflight_info.addr) {
+		munmap(dev->inflight_info.addr, dev->inflight_info.size);
+		dev->inflight_info.addr = NULL;
+	}
+
+	if (dev->inflight_info.fd > 0) {
+		close(dev->inflight_info.fd);
+		dev->inflight_info.fd = -1;
+	}
+
 	if (dev->slave_req_fd >= 0) {
 		close(dev->slave_req_fd);
 		dev->slave_req_fd = -1;
@@ -306,6 +329,7 @@  vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 
 			dev->virtqueue[dev->nr_vring] = NULL;
 			cleanup_vq(vq, 1);
+			cleanup_vq_inflight(dev, vq);
 			free_vq(dev, vq);
 		}
 	}
@@ -616,7 +640,6 @@  translate_ring_addresses(struct virtio_net *dev, int vq_index)
 				dev->vid);
 			return dev;
 		}
-
 		return dev;
 	}
 
@@ -1165,6 +1188,219 @@  virtio_is_ready(struct virtio_net *dev)
 	return 1;
 }
 
+static int mem_create(const char *name, unsigned int flags)
+{
+#ifdef __NR_memfd_create
+	return syscall(__NR_memfd_create, name, flags);
+#else
+	return -1;
+#endif
+}
+
+void *inflight_mem_alloc(const char *name, size_t size, int *fd)
+{
+	void *ptr;
+	int mfd = -1;
+	char fname[20] = "/tmp/memfd-XXXXXX";
+
+	*fd = -1;
+	mfd = mem_create(name, CLOEXEC);
+	if (mfd != -1) {
+		if (ftruncate(mfd, size) == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"ftruncate fail for alloc inflight buffer\n");
+			close(mfd);
+			return NULL;
+		}
+	} else {
+		mfd = mkstemp(fname);
+		unlink(fname);
+
+		if (mfd == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"mkstemp fail for alloc inflight buffer\n");
+			return NULL;
+		}
+
+		if (ftruncate(mfd, size) == -1) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"ftruncate fail for alloc inflight buffer\n");
+			close(mfd);
+			return NULL;
+		}
+	}
+
+	ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
+	if (ptr == MAP_FAILED) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+				"mmap fail for alloc inflight buffer\n");
+		close(mfd);
+		return NULL;
+	}
+
+	*fd = mfd;
+	return ptr;
+}
+
+static uint32_t get_pervq_shm_size_split(uint16_t queue_size)
+{
+	return ALIGN_UP(sizeof(struct inflight_desc_split) * queue_size +
+		sizeof(uint64_t) + sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
+}
+
+static uint32_t get_pervq_shm_size_packed(uint16_t queue_size)
+{
+	return ALIGN_UP(sizeof(struct inflight_desc_packed) * queue_size +
+		sizeof(uint64_t) + sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
+		INFLIGHT_ALIGNMENT);
+}
+
+static int
+vhost_user_get_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
+		int main_fd __rte_unused)
+{
+	int fd, i, j;
+	uint64_t pervq_inflight_size, mmap_size;
+	void *addr;
+	uint16_t num_queues, queue_size;
+	struct virtio_net *dev = *pdev;
+	struct inflight_info_packed *inflight_packed = NULL;
+
+	if (msg->size != sizeof(msg->payload.inflight)) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Invalid get_inflight_fd message size is %d",
+			msg->size);
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	num_queues = msg->payload.inflight.num_queues;
+	queue_size = msg->payload.inflight.queue_size;
+
+	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
+			msg->payload.inflight.num_queues);
+	RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
+			msg->payload.inflight.queue_size);
+
+	if (vq_is_packed(dev))
+		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
+	else
+		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
+
+	mmap_size = num_queues * pervq_inflight_size;
+	addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
+	if (!addr) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Failed to alloc vhost inflight area");
+			msg->payload.inflight.mmap_size = 0;
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+	memset(addr, 0, mmap_size);
+
+	dev->inflight_info.addr = addr;
+	dev->inflight_info.size = msg->payload.inflight.mmap_size = mmap_size;
+	dev->inflight_info.fd = msg->fds[0] = fd;
+	msg->payload.inflight.mmap_offset = 0;
+	msg->fd_num = 1;
+
+	if (vq_is_packed(dev)) {
+		for (i = 0; i < num_queues; i++) {
+			inflight_packed = (struct inflight_info_packed *)addr;
+			inflight_packed->used_wrap_counter = 1;
+			inflight_packed->old_used_wrap_counter = 1;
+			for (j = 0; j < queue_size; j++)
+				inflight_packed->desc[j].next = j + 1;
+			addr = (void *)((char *)addr + pervq_inflight_size);
+		}
+	}
+
+	RTE_LOG(INFO, VHOST_CONFIG,
+			"send inflight mmap_size: %lu\n",
+			msg->payload.inflight.mmap_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+			"send inflight mmap_offset: %lu\n",
+			msg->payload.inflight.mmap_offset);
+	RTE_LOG(INFO, VHOST_CONFIG,
+			"send inflight fd: %d\n", msg->fds[0]);
+
+	return RTE_VHOST_MSG_RESULT_REPLY;
+}
+
+static int
+vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
+		int main_fd __rte_unused)
+{
+	int fd, i;
+	uint64_t mmap_size, mmap_offset;
+	uint16_t num_queues, queue_size;
+	uint32_t pervq_inflight_size;
+	void *addr;
+	struct vhost_virtqueue *vq;
+	struct virtio_net *dev = *pdev;
+
+	fd = msg->fds[0];
+	if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"Invalid set_inflight_fd message size is %d,fd is %d\n",
+			msg->size, fd);
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	mmap_size = msg->payload.inflight.mmap_size;
+	mmap_offset = msg->payload.inflight.mmap_offset;
+	num_queues = msg->payload.inflight.num_queues;
+	queue_size = msg->payload.inflight.queue_size;
+
+	if (vq_is_packed(dev))
+		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
+	else
+		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
+
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd mmap_size: %lu\n", mmap_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd mmap_offset: %lu\n", mmap_offset);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd num_queues: %u\n", num_queues);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd queue_size: %u\n", queue_size);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd fd: %d\n", fd);
+	RTE_LOG(INFO, VHOST_CONFIG,
+		"set_inflight_fd pervq_inflight_size: %d\n",
+		pervq_inflight_size);
+
+	if (dev->inflight_info.addr)
+		munmap(dev->inflight_info.addr, dev->inflight_info.size);
+
+	addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
+			fd, mmap_offset);
+	if (addr == MAP_FAILED) {
+		RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
+		return RTE_VHOST_MSG_RESULT_ERR;
+	}
+
+	if (dev->inflight_info.fd)
+		close(dev->inflight_info.fd);
+
+	dev->inflight_info.fd = fd;
+	dev->inflight_info.addr = addr;
+	dev->inflight_info.size = mmap_size;
+
+	for (i = 0; i < num_queues; i++) {
+		vq = dev->virtqueue[i];
+		if (vq_is_packed(dev)) {
+			vq->inflight_packed = addr;
+			vq->inflight_packed->desc_num = queue_size;
+		} else {
+			vq->inflight_split = addr;
+			vq->inflight_split->desc_num = queue_size;
+		}
+		addr = (void *)((char *)addr + pervq_inflight_size);
+	}
+
+	return RTE_VHOST_MSG_RESULT_OK;
+}
+
 static int
 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
 			int main_fd __rte_unused)
@@ -1201,6 +1437,189 @@  static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
 	return RTE_VHOST_MSG_RESULT_OK;
 }
 
+static int
+resubmit_desc_compare(const void *a, const void *b)
+{
+	const struct rte_vhost_resubmit_desc *desc0 =
+		(const struct rte_vhost_resubmit_desc *)a;
+	const struct rte_vhost_resubmit_desc *desc1 =
+		(const struct rte_vhost_resubmit_desc *)b;
+
+	if (desc1->counter > desc0->counter &&
+		(desc1->counter - desc0->counter) < VIRTQUEUE_MAX_SIZE * 2)
+		return 1;
+
+	return -1;
+}
+
+static int
+vhost_check_queue_inflights_split(struct virtio_net *dev,
+	struct vhost_virtqueue *vq)
+{
+	uint16_t i = 0;
+	uint16_t resubmit_num = 0, last_io, num;
+	struct vring_used *used = vq->used;
+	struct rte_vhost_resubmit_info *resubmit = NULL;
+	struct inflight_info_split *inflight_split;
+
+	if (!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
+		return RTE_VHOST_MSG_RESULT_OK;
+
+	if ((!vq->inflight_split))
+		return RTE_VHOST_MSG_RESULT_ERR;
+
+	if (!vq->inflight_split->version) {
+		vq->inflight_split->version = INFLIGHT_VERSION;
+		return RTE_VHOST_MSG_RESULT_OK;
+	}
+
+	inflight_split = vq->inflight_split;
+	vq->resubmit_inflight = NULL;
+	vq->global_counter = 0;
+	last_io = inflight_split->last_inflight_io;
+
+	if (inflight_split->used_idx != used->idx) {
+		inflight_split->desc[last_io].inflight = 0;
+		rte_compiler_barrier();
+		inflight_split->used_idx = used->idx;
+	}
+
+	for (i = 0; i < inflight_split->desc_num; i++) {
+		if (inflight_split->desc[i].inflight == 1)
+			resubmit_num++;
+	}
+
+	vq->last_avail_idx += resubmit_num;
+
+	if (resubmit_num) {
+		resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
+		if (!resubmit) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for resubmit info.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		resubmit->resubmit_list = calloc(resubmit_num,
+			sizeof(struct rte_vhost_resubmit_desc));
+		if (!resubmit->resubmit_list) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for inflight desc.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		num = 0;
+		for (i = 0; i < vq->inflight_split->desc_num; i++) {
+			if (vq->inflight_split->desc[i].inflight == 1) {
+				resubmit->resubmit_list[num].index = i;
+				resubmit->resubmit_list[num].counter =
+					inflight_split->desc[i].counter;
+				num++;
+			}
+		}
+		resubmit->resubmit_num = num;
+
+		if (resubmit->resubmit_num > 1)
+			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
+				sizeof(struct rte_vhost_resubmit_desc),
+				resubmit_desc_compare);
+
+		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
+		vq->resubmit_inflight = resubmit;
+	}
+
+	return RTE_VHOST_MSG_RESULT_OK;
+}
+
+static int
+vhost_check_queue_inflights_packed(struct virtio_net *dev,
+						 struct vhost_virtqueue *vq)
+{
+	uint16_t i = 0;
+	uint16_t resubmit_num = 0, old_used_idx, num;
+	struct rte_vhost_resubmit_info *resubmit = NULL;
+	struct inflight_info_packed *inflight_packed;
+
+	if (!(dev->protocol_features &
+		(1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
+		return RTE_VHOST_MSG_RESULT_OK;
+
+	if (!vq->inflight_packed->version) {
+		vq->inflight_packed->version = INFLIGHT_VERSION;
+		return RTE_VHOST_MSG_RESULT_OK;
+	}
+
+	if ((!vq->inflight_packed))
+		return RTE_VHOST_MSG_RESULT_ERR;
+
+	inflight_packed = vq->inflight_packed;
+	vq->resubmit_inflight = NULL;
+	vq->global_counter = 0;
+	old_used_idx = inflight_packed->old_used_idx;
+
+	if (inflight_packed->used_idx != old_used_idx) {
+		if (inflight_packed->desc[old_used_idx].inflight == 0) {
+			inflight_packed->old_used_idx =
+				inflight_packed->used_idx;
+			inflight_packed->old_used_wrap_counter =
+				inflight_packed->used_wrap_counter;
+			inflight_packed->old_free_head =
+				inflight_packed->free_head;
+		} else {
+			inflight_packed->used_idx =
+				inflight_packed->old_used_idx;
+			inflight_packed->used_wrap_counter =
+				inflight_packed->old_used_wrap_counter;
+			inflight_packed->free_head =
+				inflight_packed->old_free_head;
+		}
+	}
+
+	for (i = 0; i < inflight_packed->desc_num; i++) {
+		if (inflight_packed->desc[i].inflight == 1)
+			resubmit_num++;
+	}
+
+	if (resubmit_num) {
+		resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
+		if (resubmit == NULL) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for resubmit info.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		resubmit->resubmit_list = calloc(resubmit_num,
+			sizeof(struct rte_vhost_resubmit_desc));
+		if (resubmit->resubmit_list == NULL) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+			 "Failed to allocate memory for resubmit desc.\n");
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+
+		num = 0;
+		for (i = 0; i < inflight_packed->desc_num; i++) {
+			if (vq->inflight_packed->desc[i].inflight == 1) {
+				resubmit->resubmit_list[num].index = i;
+				resubmit->resubmit_list[num].counter =
+					inflight_packed->desc[i].counter;
+				num++;
+			}
+		}
+		resubmit->resubmit_num = num;
+
+		if (resubmit->resubmit_num > 1)
+			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
+				sizeof(struct rte_vhost_resubmit_desc),
+				resubmit_desc_compare);
+
+		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
+		vq->resubmit_inflight = resubmit;
+	}
+
+	return RTE_VHOST_MSG_RESULT_OK;
+}
+
+
 static int
 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
 			int main_fd __rte_unused)
@@ -1242,6 +1661,20 @@  vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
 		close(vq->kickfd);
 	vq->kickfd = file.fd;
 
+	if (vq_is_packed(dev)) {
+		if (vhost_check_queue_inflights_packed(dev, vq)) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"Failed to inflights for vq: %d\n", file.index);
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+	} else {
+		if (vhost_check_queue_inflights_split(dev, vq)) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+				"Failed to inflights for vq: %d\n", file.index);
+			return RTE_VHOST_MSG_RESULT_ERR;
+		}
+	}
+
 	return RTE_VHOST_MSG_RESULT_OK;
 }
 
@@ -1762,6 +2195,8 @@  static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
 	[VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
 	[VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
 	[VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
+	[VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
+	[VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
 };
 
 
diff --git a/lib/librte_vhost/vhost_user.h b/lib/librte_vhost/vhost_user.h
index 2a650fe4b..99a773910 100644
--- a/lib/librte_vhost/vhost_user.h
+++ b/lib/librte_vhost/vhost_user.h
@@ -54,7 +54,9 @@  typedef enum VhostUserRequest {
 	VHOST_USER_POSTCOPY_ADVISE = 28,
 	VHOST_USER_POSTCOPY_LISTEN = 29,
 	VHOST_USER_POSTCOPY_END = 30,
-	VHOST_USER_MAX = 31
+	VHOST_USER_GET_INFLIGHT_FD = 31,
+	VHOST_USER_SET_INFLIGHT_FD = 32,
+	VHOST_USER_MAX = 33
 } VhostUserRequest;
 
 typedef enum VhostUserSlaveRequest {
@@ -112,6 +114,13 @@  typedef struct VhostUserVringArea {
 	uint64_t offset;
 } VhostUserVringArea;
 
+typedef struct VhostUserInflight {
+	uint64_t mmap_size;
+	uint64_t mmap_offset;
+	uint16_t num_queues;
+	uint16_t queue_size;
+} VhostUserInflight;
+
 typedef struct VhostUserMsg {
 	union {
 		uint32_t master; /* a VhostUserRequest value */
@@ -131,6 +140,7 @@  typedef struct VhostUserMsg {
 		struct vhost_vring_addr addr;
 		VhostUserMemory memory;
 		VhostUserLog    log;
+		VhostUserInflight inflight;
 		struct vhost_iotlb_msg iotlb;
 		VhostUserCryptoSessionParam crypto_session;
 		VhostUserVringArea area;
@@ -148,6 +158,7 @@  typedef struct VhostUserMsg {
 /* vhost_user.c */
 int vhost_user_msg_handler(int vid, int fd);
 int vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm);
+void *inflight_mem_alloc(const char *name, size_t size, int *fd);
 
 /* socket.c */
 int read_fd_message(int sockfd, char *buf, int buflen, int *fds, int max_fds,