[PATCH v6 00/15] Vhost: add support to packed ring layout

[PATCH v6 00/15] Vhost: add support to packed ring layout

[Maxime Coquelin]

This series is a handover from Jen's "[PATCH v4 00/20]
implement packed virtqueues", which only implements the
vhost side. Virtio PMD implementation will follow in a 
next series.

The series applies on top of previous reworks I posted
during this cycle that merges mergeable and non-mergeable
receive paths, and refactors transmit path to re-use
vector buffers.

I haven't run performance tests for now as the Virtio PMD
side isn't ready.

The series has been tested with Tiwei's series implementing
packed ring support to Kernel's virtio-net driver, and
with Wei series implementing the Qemu side.

To test it, I have used testpmd on host side with a vhost
vdev and a tap vdev forwarding to each other. Transferts
of big random files have been done in both ways with
integrity verified.

Tests have been run with Rx mrg ON/OFF and events suppression
ON/OFF.

Tests have also been run with legacy spli ring layout to
ensure no regression have been introduced.

Changes since v5:
=================
- Remove duplicated VHOST_USER_F_PROTOCOL_FEATURES definition (Tiwei)
- Fix vq_is_ready (Maxime)
- Fix local used index overflow in flush_shadow_used_ring_packed (Tiwei)
- Make desc_is_avail() a bool (Tiwei)
- Improve desc_is_avail() logic (Tiwei)
- Remove split rings addr NULL assignment in the right patch (Tiwei)
- Make buffer id a uint16_t (Tiwei)

Jens Freimann (2):
  vhost: add virtio packed virtqueue defines
  vhost: add helpers for packed virtqueues

Maxime Coquelin (12):
  vhost: clear shadow used table index at flush time
  vhost: make indirect desc table copy desc type agnostic
  vhost: clear batch copy index at copy time
  vhost: extract split ring handling from Rx and Tx functions
  vhost: append shadow used ring function names with split
  vhost: add shadow used ring support for packed rings
  vhost: create descriptor mapping function
  vhost: add vector filling support for packed ring
  vhost: add Rx support for packed ring
  vhost: add Tx support for packed ring
  vhost: add notification for packed ring
  vhost: advertize packed ring layout support

Yuanhan Liu (1):
  vhost: vring address setup for packed queues

 lib/librte_vhost/vhost.c         | 126 ++++++-
 lib/librte_vhost/vhost.h         | 104 +++++-
 lib/librte_vhost/vhost_user.c    | 128 ++++++-
 lib/librte_vhost/virtio-packed.h |  41 +++
 lib/librte_vhost/virtio_net.c    | 772 ++++++++++++++++++++++++++++++---------
 5 files changed, 963 insertions(+), 208 deletions(-)
 create mode 100644 lib/librte_vhost/virtio-packed.h

-- 
2.14.4

[PATCH v6 01/15] vhost: add virtio packed virtqueue defines

[Maxime Coquelin]

From: Jens Freimann <jfreimann@redhat.com>

Signed-off-by: Jens Freimann <jfreimann@redhat.com>
---
 lib/librte_vhost/vhost.h         |  3 +++
 lib/librte_vhost/virtio-packed.h | 22 ++++++++++++++++++++++
 2 files changed, 25 insertions(+)
 create mode 100644 lib/librte_vhost/virtio-packed.h

diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index 3f8fa3a78..cecd5c1ee 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -191,6 +191,9 @@ struct vhost_msg {
 #ifndef VIRTIO_F_VERSION_1
  #define VIRTIO_F_VERSION_1 32
 #endif
+#ifndef VIRTIO_F_RING_PACKED
+ #define VIRTIO_F_RING_PACKED 34
+#endif
 
 /* Features supported by this builtin vhost-user net driver. */
 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
new file mode 100644
index 000000000..744b3991b
--- /dev/null
+++ b/lib/librte_vhost/virtio-packed.h
@@ -0,0 +1,22 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) Red Hat Inc.
+ */
+
+#ifndef __VIRTIO_PACKED_H
+#define __VIRTIO_PACKED_H
+
+#define VRING_DESC_F_NEXT       1
+#define VRING_DESC_F_WRITE      2
+#define VRING_DESC_F_INDIRECT   4
+
+#define VRING_DESC_F_AVAIL      (1ULL << 7)
+#define VRING_DESC_F_USED	(1ULL << 15)
+
+struct vring_desc_packed {
+	uint64_t addr;
+	uint32_t len;
+	uint16_t index;
+	uint16_t flags;
+};
+
+#endif /* __VIRTIO_PACKED_H */
-- 
2.14.4

Re: [PATCH v6 01/15] vhost: add virtio packed virtqueue defines

[Tiwei Bie]

On Mon, Jul 02, 2018 at 10:16:15AM +0200, Maxime Coquelin wrote:
> From: Jens Freimann <jfreimann@redhat.com>
> 
> Signed-off-by: Jens Freimann <jfreimann@redhat.com>
> ---
>  lib/librte_vhost/vhost.h         |  3 +++
>  lib/librte_vhost/virtio-packed.h | 22 ++++++++++++++++++++++
>  2 files changed, 25 insertions(+)
>  create mode 100644 lib/librte_vhost/virtio-packed.h
> 
> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
> index 3f8fa3a78..cecd5c1ee 100644
> --- a/lib/librte_vhost/vhost.h
> +++ b/lib/librte_vhost/vhost.h
> @@ -191,6 +191,9 @@ struct vhost_msg {
>  #ifndef VIRTIO_F_VERSION_1
>   #define VIRTIO_F_VERSION_1 32
>  #endif
> +#ifndef VIRTIO_F_RING_PACKED
> + #define VIRTIO_F_RING_PACKED 34
> +#endif
>  
>  /* Features supported by this builtin vhost-user net driver. */
>  #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
> diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
> new file mode 100644
> index 000000000..744b3991b
> --- /dev/null
> +++ b/lib/librte_vhost/virtio-packed.h
> @@ -0,0 +1,22 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) Red Hat Inc.
> + */
> +
> +#ifndef __VIRTIO_PACKED_H
> +#define __VIRTIO_PACKED_H
> +
> +#define VRING_DESC_F_NEXT       1
> +#define VRING_DESC_F_WRITE      2
> +#define VRING_DESC_F_INDIRECT   4
> +
> +#define VRING_DESC_F_AVAIL      (1ULL << 7)
> +#define VRING_DESC_F_USED	(1ULL << 15)

Please use space or tab consistently.

> +
> +struct vring_desc_packed {
> +	uint64_t addr;
> +	uint32_t len;
> +	uint16_t index;
> +	uint16_t flags;
> +};
> +
> +#endif /* __VIRTIO_PACKED_H */
> -- 
> 2.14.4
> 

Re: [PATCH v6 01/15] vhost: add virtio packed virtqueue defines

[Maxime Coquelin]

On 07/04/2018 07:37 AM, Tiwei Bie wrote:
> On Mon, Jul 02, 2018 at 10:16:15AM +0200, Maxime Coquelin wrote:
>> From: Jens Freimann <jfreimann@redhat.com>
>>
>> Signed-off-by: Jens Freimann <jfreimann@redhat.com>
>> ---
>>   lib/librte_vhost/vhost.h         |  3 +++
>>   lib/librte_vhost/virtio-packed.h | 22 ++++++++++++++++++++++
>>   2 files changed, 25 insertions(+)
>>   create mode 100644 lib/librte_vhost/virtio-packed.h
>>
>> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
>> index 3f8fa3a78..cecd5c1ee 100644
>> --- a/lib/librte_vhost/vhost.h
>> +++ b/lib/librte_vhost/vhost.h
>> @@ -191,6 +191,9 @@ struct vhost_msg {
>>   #ifndef VIRTIO_F_VERSION_1
>>    #define VIRTIO_F_VERSION_1 32
>>   #endif
>> +#ifndef VIRTIO_F_RING_PACKED
>> + #define VIRTIO_F_RING_PACKED 34
>> +#endif
>>   
>>   /* Features supported by this builtin vhost-user net driver. */
>>   #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
>> diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
>> new file mode 100644
>> index 000000000..744b3991b
>> --- /dev/null
>> +++ b/lib/librte_vhost/virtio-packed.h
>> @@ -0,0 +1,22 @@
>> +/* SPDX-License-Identifier: BSD-3-Clause
>> + * Copyright(c) Red Hat Inc.
>> + */
>> +
>> +#ifndef __VIRTIO_PACKED_H
>> +#define __VIRTIO_PACKED_H
>> +
>> +#define VRING_DESC_F_NEXT       1
>> +#define VRING_DESC_F_WRITE      2
>> +#define VRING_DESC_F_INDIRECT   4
>> +
>> +#define VRING_DESC_F_AVAIL      (1ULL << 7)
>> +#define VRING_DESC_F_USED	(1ULL << 15)
> 
> Please use space or tab consistently.

Right.

Thanks,
Maxime
>> +
>> +struct vring_desc_packed {
>> +	uint64_t addr;
>> +	uint32_t len;
>> +	uint16_t index;
>> +	uint16_t flags;
>> +};
>> +
>> +#endif /* __VIRTIO_PACKED_H */
>> -- 
>> 2.14.4
>>

[PATCH v6 02/15] vhost: add helpers for packed virtqueues

[Maxime Coquelin]

From: Jens Freimann <jfreimann@redhat.com>

Add some helper functions to check descriptor flags
and check if a vring is of type packed.

Signed-off-by: Jens Freimann <jfreimann@redhat.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/vhost.h         | 6 ++++++
 lib/librte_vhost/virtio-packed.h | 8 ++++++++
 2 files changed, 14 insertions(+)

diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index cecd5c1ee..cabbc3ab7 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -318,6 +318,12 @@ struct virtio_net {
 	struct vhost_user_extern_ops extern_ops;
 } __rte_cache_aligned;
 
+static __rte_always_inline bool
+vq_is_packed(struct virtio_net *dev)
+{
+	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
+}
+
 #define VHOST_LOG_PAGE	4096
 
 /*
diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
index 744b3991b..d386cb6df 100644
--- a/lib/librte_vhost/virtio-packed.h
+++ b/lib/librte_vhost/virtio-packed.h
@@ -19,4 +19,12 @@ struct vring_desc_packed {
 	uint16_t flags;
 };
 
+
+static inline bool
+desc_is_avail(struct vring_desc_packed *desc, bool wrap_counter)
+{
+	return wrap_counter == !!(desc->flags & VRING_DESC_F_AVAIL) &&
+		wrap_counter != !!(desc->flags & VRING_DESC_F_USED);
+}
+
 #endif /* __VIRTIO_PACKED_H */
-- 
2.14.4

Re: [PATCH v6 02/15] vhost: add helpers for packed virtqueues

[Tiwei Bie]

On Mon, Jul 02, 2018 at 10:16:16AM +0200, Maxime Coquelin wrote:
> From: Jens Freimann <jfreimann@redhat.com>
> 
> Add some helper functions to check descriptor flags
> and check if a vring is of type packed.
> 
> Signed-off-by: Jens Freimann <jfreimann@redhat.com>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  lib/librte_vhost/vhost.h         | 6 ++++++
>  lib/librte_vhost/virtio-packed.h | 8 ++++++++
>  2 files changed, 14 insertions(+)
> 
> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
> index cecd5c1ee..cabbc3ab7 100644
> --- a/lib/librte_vhost/vhost.h
> +++ b/lib/librte_vhost/vhost.h
> @@ -318,6 +318,12 @@ struct virtio_net {
>  	struct vhost_user_extern_ops extern_ops;
>  } __rte_cache_aligned;
>  
> +static __rte_always_inline bool
> +vq_is_packed(struct virtio_net *dev)
> +{
> +	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
> +}
> +
>  #define VHOST_LOG_PAGE	4096
>  
>  /*
> diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
> index 744b3991b..d386cb6df 100644
> --- a/lib/librte_vhost/virtio-packed.h
> +++ b/lib/librte_vhost/virtio-packed.h
> @@ -19,4 +19,12 @@ struct vring_desc_packed {
>  	uint16_t flags;
>  };
>  
> +

There is no need to add above blank line.

> +static inline bool
> +desc_is_avail(struct vring_desc_packed *desc, bool wrap_counter)
> +{
> +	return wrap_counter == !!(desc->flags & VRING_DESC_F_AVAIL) &&
> +		wrap_counter != !!(desc->flags & VRING_DESC_F_USED);
> +}
> +
>  #endif /* __VIRTIO_PACKED_H */
> -- 
> 2.14.4
> 

Re: [PATCH v6 02/15] vhost: add helpers for packed virtqueues

[Maxime Coquelin]

On 07/04/2018 07:39 AM, Tiwei Bie wrote:
> On Mon, Jul 02, 2018 at 10:16:16AM +0200, Maxime Coquelin wrote:
>> From: Jens Freimann <jfreimann@redhat.com>
>>
>> Add some helper functions to check descriptor flags
>> and check if a vring is of type packed.
>>
>> Signed-off-by: Jens Freimann <jfreimann@redhat.com>
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>>   lib/librte_vhost/vhost.h         | 6 ++++++
>>   lib/librte_vhost/virtio-packed.h | 8 ++++++++
>>   2 files changed, 14 insertions(+)
>>
>> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
>> index cecd5c1ee..cabbc3ab7 100644
>> --- a/lib/librte_vhost/vhost.h
>> +++ b/lib/librte_vhost/vhost.h
>> @@ -318,6 +318,12 @@ struct virtio_net {
>>   	struct vhost_user_extern_ops extern_ops;
>>   } __rte_cache_aligned;
>>   
>> +static __rte_always_inline bool
>> +vq_is_packed(struct virtio_net *dev)
>> +{
>> +	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
>> +}
>> +
>>   #define VHOST_LOG_PAGE	4096
>>   
>>   /*
>> diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
>> index 744b3991b..d386cb6df 100644
>> --- a/lib/librte_vhost/virtio-packed.h
>> +++ b/lib/librte_vhost/virtio-packed.h
>> @@ -19,4 +19,12 @@ struct vring_desc_packed {
>>   	uint16_t flags;
>>   };
>>   
>> +
> 
> There is no need to add above blank line.

Removed.

>> +static inline bool
>> +desc_is_avail(struct vring_desc_packed *desc, bool wrap_counter)
>> +{
>> +	return wrap_counter == !!(desc->flags & VRING_DESC_F_AVAIL) &&
>> +		wrap_counter != !!(desc->flags & VRING_DESC_F_USED);
>> +}
>> +
>>   #endif /* __VIRTIO_PACKED_H */
>> -- 
>> 2.14.4
>>

[PATCH v6 03/15] vhost: vring address setup for packed queues

[Maxime Coquelin]

From: Yuanhan Liu <yuanhan.liu@linux.intel.com>

Add code to set up packed queues when enabled.

Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: Jens Freimann <jfreiman@redhat.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/vhost.c      | 44 ++++++++++++++++++++++++++++++++++++++-----
 lib/librte_vhost/vhost.h      |  7 ++++++-
 lib/librte_vhost/vhost_user.c | 40 +++++++++++++++++++++++++++++++++++----
 3 files changed, 81 insertions(+), 10 deletions(-)

diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index afded4952..05d2bb182 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -23,6 +23,7 @@
 #include "iotlb.h"
 #include "vhost.h"
 #include "vhost_user.h"
+#include "virtio-packed.h"
 
 struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 
@@ -115,14 +116,11 @@ free_device(struct virtio_net *dev)
 	rte_free(dev);
 }
 
-int
-vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
+static int
+vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
 	uint64_t req_size, size;
 
-	if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
-		goto out;
-
 	req_size = sizeof(struct vring_desc) * vq->size;
 	size = req_size;
 	vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
@@ -153,6 +151,40 @@ vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
 	if (!vq->used || size != req_size)
 		return -1;
 
+	return 0;
+}
+
+static int
+vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+	uint64_t req_size, size;
+
+	req_size = sizeof(struct vring_desc_packed) * vq->size;
+	size = req_size;
+	vq->desc_packed =
+		(struct vring_desc_packed *)(uintptr_t)vhost_iova_to_vva(dev,
+					vq, vq->ring_addrs.desc_user_addr,
+					&size, VHOST_ACCESS_RW);
+	if (!vq->desc_packed || size != req_size)
+		return -1;
+
+	return 0;
+}
+
+int
+vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+
+	if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
+		goto out;
+
+	if (vq_is_packed(dev)) {
+		if (vring_translate_packed(dev, vq) < 0)
+			return -1;
+	} else {
+		if (vring_translate_split(dev, vq) < 0)
+			return -1;
+	}
 out:
 	vq->access_ok = 1;
 
@@ -234,6 +266,8 @@ alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 	dev->virtqueue[vring_idx] = vq;
 	init_vring_queue(dev, vring_idx);
 	rte_spinlock_init(&vq->access_lock);
+	vq->avail_wrap_counter = 1;
+	vq->used_wrap_counter = 1;
 
 	dev->nr_vring += 1;
 
diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index cabbc3ab7..35beff699 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -84,7 +84,10 @@ struct log_cache_entry {
  * Structure contains variables relevant to RX/TX virtqueues.
  */
 struct vhost_virtqueue {
-	struct vring_desc	*desc;
+	union {
+		struct vring_desc	*desc;
+		struct vring_desc_packed   *desc_packed;
+	};
 	struct vring_avail	*avail;
 	struct vring_used	*used;
 	uint32_t		size;
@@ -122,6 +125,8 @@ struct vhost_virtqueue {
 
 	struct batch_copy_elem	*batch_copy_elems;
 	uint16_t		batch_copy_nb_elems;
+	bool			used_wrap_counter;
+	bool			avail_wrap_counter;
 
 	struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
 	uint16_t log_cache_nb_elem;
diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
index 947290fc3..ae03ec29b 100644
--- a/lib/librte_vhost/vhost_user.c
+++ b/lib/librte_vhost/vhost_user.c
@@ -39,6 +39,7 @@
 #include "iotlb.h"
 #include "vhost.h"
 #include "vhost_user.h"
+#include "virtio-packed.h"
 
 #define VIRTIO_MIN_MTU 68
 #define VIRTIO_MAX_MTU 65535
@@ -477,6 +478,27 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index)
 	struct vhost_vring_addr *addr = &vq->ring_addrs;
 	uint64_t len;
 
+	if (vq_is_packed(dev)) {
+		len = sizeof(struct vring_desc_packed) * vq->size;
+		vq->desc_packed = (struct vring_desc_packed *) ring_addr_to_vva
+			(dev, vq, addr->desc_user_addr, &len);
+		vq->log_guest_addr = 0;
+		if (vq->desc_packed == NULL ||
+				len != sizeof(struct vring_desc_packed) *
+				vq->size) {
+			RTE_LOG(DEBUG, VHOST_CONFIG,
+				"(%d) failed to map desc_packed ring.\n",
+				dev->vid);
+			return dev;
+		}
+
+		dev = numa_realloc(dev, vq_index);
+		vq = dev->virtqueue[vq_index];
+		addr = &vq->ring_addrs;
+
+		return dev;
+	}
+
 	/* The addresses are converted from QEMU virtual to Vhost virtual. */
 	if (vq->desc && vq->avail && vq->used)
 		return dev;
@@ -885,10 +907,20 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 	return -1;
 }
 
-static int
-vq_is_ready(struct vhost_virtqueue *vq)
+static bool
+vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
-	return vq && vq->desc && vq->avail && vq->used &&
+	bool rings_ok;
+
+	if (!vq)
+		return false;
+
+	if (vq_is_packed(dev))
+		rings_ok = !!vq->desc_packed;
+	else
+		rings_ok = vq->desc && vq->avail && vq->used;
+
+	return rings_ok &&
 	       vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 	       vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 }
@@ -905,7 +937,7 @@ virtio_is_ready(struct virtio_net *dev)
 	for (i = 0; i < dev->nr_vring; i++) {
 		vq = dev->virtqueue[i];
 
-		if (!vq_is_ready(vq))
+		if (!vq_is_ready(dev, vq))
 			return 0;
 	}
 
-- 
2.14.4

[PATCH v6 04/15] vhost: clear shadow used table index at flush time

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 2c13a59a2..28ce2d073 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -115,6 +115,7 @@ flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq)
 	vhost_log_cache_sync(dev, vq);
 
 	*(volatile uint16_t *)&vq->used->idx += vq->shadow_used_idx;
+	vq->shadow_used_idx = 0;
 	vhost_log_used_vring(dev, vq, offsetof(struct vring_used, idx),
 		sizeof(vq->used->idx));
 }
@@ -561,7 +562,6 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 
 	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 
-	vq->shadow_used_idx = 0;
 	avail_head = *((volatile uint16_t *)&vq->avail->idx);
 	for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
 		uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
@@ -1048,7 +1048,6 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 		goto out_access_unlock;
 
 	vq->batch_copy_nb_elems = 0;
-	vq->shadow_used_idx = 0;
 
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 		vhost_user_iotlb_rd_lock(vq);
@@ -1079,7 +1078,6 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 
 		flush_shadow_used_ring(dev, vq);
 		vhost_vring_call(dev, vq);
-		vq->shadow_used_idx = 0;
 	}
 
 	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
-- 
2.14.4

[PATCH v6 05/15] vhost: make indirect desc table copy desc type agnostic

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 28ce2d073..5cfc100a9 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -37,16 +37,15 @@ is_valid_virt_queue_idx(uint32_t idx, int is_tx, uint32_t nr_vring)
 	return (is_tx ^ (idx & 1)) == 0 && idx < nr_vring;
 }
 
-static __rte_always_inline struct vring_desc *
+static __rte_always_inline void *
 alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
-					 struct vring_desc *desc)
+					 uint64_t desc_addr, uint64_t desc_len)
 {
-	struct vring_desc *idesc;
+	void *idesc;
 	uint64_t src, dst;
-	uint64_t len, remain = desc->len;
-	uint64_t desc_addr = desc->addr;
+	uint64_t len, remain = desc_len;
 
-	idesc = rte_malloc(__func__, desc->len, 0);
+	idesc = rte_malloc(__func__, desc_len, 0);
 	if (unlikely(!idesc))
 		return 0;
 
@@ -72,7 +71,7 @@ alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
 }
 
 static __rte_always_inline void
-free_ind_table(struct vring_desc *idesc)
+free_ind_table(void *idesc)
 {
 	rte_free(idesc);
 }
@@ -251,7 +250,8 @@ fill_vec_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			 * The indirect desc table is not contiguous
 			 * in process VA space, we have to copy it.
 			 */
-			idesc = alloc_copy_ind_table(dev, vq, &vq->desc[idx]);
+			idesc = alloc_copy_ind_table(dev, vq,
+					vq->desc[idx].addr, vq->desc[idx].len);
 			if (unlikely(!idesc))
 				return -1;
 
-- 
2.14.4

[PATCH v6 06/15] vhost: clear batch copy index at copy time

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 5cfc100a9..816d5fc1d 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -141,6 +141,8 @@ do_data_copy_enqueue(struct virtio_net *dev, struct vhost_virtqueue *vq)
 		vhost_log_cache_write(dev, vq, elem[i].log_addr, elem[i].len);
 		PRINT_PACKET(dev, (uintptr_t)elem[i].dst, elem[i].len, 0);
 	}
+
+	vq->batch_copy_nb_elems = 0;
 }
 
 static inline void
@@ -152,6 +154,8 @@ do_data_copy_dequeue(struct vhost_virtqueue *vq)
 
 	for (i = 0; i < count; i++)
 		rte_memcpy(elem[i].dst, elem[i].src, elem[i].len);
+
+	vq->batch_copy_nb_elems = 0;
 }
 
 /* avoid write operation when necessary, to lessen cache issues */
@@ -558,8 +562,6 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 	if (count == 0)
 		goto out;
 
-	vq->batch_copy_nb_elems = 0;
-
 	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 
 	avail_head = *((volatile uint16_t *)&vq->avail->idx);
@@ -1047,8 +1049,6 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 	if (unlikely(vq->enabled == 0))
 		goto out_access_unlock;
 
-	vq->batch_copy_nb_elems = 0;
-
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 		vhost_user_iotlb_rd_lock(vq);
 
-- 
2.14.4

[PATCH v6 07/15] vhost: extract split ring handling from Rx and Tx functions

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 240 +++++++++++++++++++++++-------------------
 1 file changed, 131 insertions(+), 109 deletions(-)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 816d5fc1d..385876527 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -226,13 +226,13 @@ virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
 }
 
 static __rte_always_inline int
-fill_vec_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
-			 uint32_t avail_idx, uint32_t *vec_idx,
+fill_vec_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
+			 uint32_t avail_idx, uint16_t *vec_idx,
 			 struct buf_vector *buf_vec, uint16_t *desc_chain_head,
 			 uint16_t *desc_chain_len, uint8_t perm)
 {
 	uint16_t idx = vq->avail->ring[avail_idx & (vq->size - 1)];
-	uint32_t vec_id = *vec_idx;
+	uint16_t vec_id = *vec_idx;
 	uint32_t len    = 0;
 	uint64_t dlen, desc_avail, desc_iova;
 	struct vring_desc *descs = vq->desc;
@@ -317,13 +317,13 @@ fill_vec_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
  * Returns -1 on fail, 0 on success
  */
 static inline int
-reserve_avail_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
+reserve_avail_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 				uint32_t size, struct buf_vector *buf_vec,
 				uint16_t *num_buffers, uint16_t avail_head,
 				uint16_t *nr_vec)
 {
 	uint16_t cur_idx;
-	uint32_t vec_idx = 0;
+	uint16_t vec_idx = 0;
 	uint16_t max_tries, tries = 0;
 
 	uint16_t head_idx = 0;
@@ -341,7 +341,8 @@ reserve_avail_buf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		if (unlikely(cur_idx == avail_head))
 			return -1;
 
-		if (unlikely(fill_vec_buf(dev, vq, cur_idx, &vec_idx, buf_vec,
+		if (unlikely(fill_vec_buf_split(dev, vq, cur_idx,
+						&vec_idx, buf_vec,
 						&head_idx, &len,
 						VHOST_ACCESS_RO) < 0))
 			return -1;
@@ -528,48 +529,22 @@ copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 }
 
 static __rte_always_inline uint32_t
-virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
+virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	struct rte_mbuf **pkts, uint32_t count)
 {
-	struct vhost_virtqueue *vq;
 	uint32_t pkt_idx = 0;
 	uint16_t num_buffers;
 	struct buf_vector buf_vec[BUF_VECTOR_MAX];
 	uint16_t avail_head;
 
-	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
-	if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
-		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
-			dev->vid, __func__, queue_id);
-		return 0;
-	}
-
-	vq = dev->virtqueue[queue_id];
-
-	rte_spinlock_lock(&vq->access_lock);
-
-	if (unlikely(vq->enabled == 0))
-		goto out_access_unlock;
-
-	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
-		vhost_user_iotlb_rd_lock(vq);
-
-	if (unlikely(vq->access_ok == 0))
-		if (unlikely(vring_translate(dev, vq) < 0))
-			goto out;
-
-	count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
-	if (count == 0)
-		goto out;
-
 	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
-
 	avail_head = *((volatile uint16_t *)&vq->avail->idx);
+
 	for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
 		uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
 		uint16_t nr_vec = 0;
 
-		if (unlikely(reserve_avail_buf(dev, vq,
+		if (unlikely(reserve_avail_buf_split(dev, vq,
 						pkt_len, buf_vec, &num_buffers,
 						avail_head, &nr_vec) < 0)) {
 			VHOST_LOG_DEBUG(VHOST_DATA,
@@ -602,6 +577,42 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 		vhost_vring_call(dev, vq);
 	}
 
+	return pkt_idx;
+}
+
+static __rte_always_inline uint32_t
+virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
+	struct rte_mbuf **pkts, uint32_t count)
+{
+	struct vhost_virtqueue *vq;
+
+	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
+	if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
+		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
+			dev->vid, __func__, queue_id);
+		return 0;
+	}
+
+	vq = dev->virtqueue[queue_id];
+
+	rte_spinlock_lock(&vq->access_lock);
+
+	if (unlikely(vq->enabled == 0))
+		goto out_access_unlock;
+
+	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
+		vhost_user_iotlb_rd_lock(vq);
+
+	if (unlikely(vq->access_ok == 0))
+		if (unlikely(vring_translate(dev, vq) < 0))
+			goto out;
+
+	count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
+	if (count == 0)
+		goto out;
+
+	count = virtio_dev_rx_split(dev, vq, pkts, count);
+
 out:
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 		vhost_user_iotlb_rd_unlock(vq);
@@ -609,7 +620,7 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 out_access_unlock:
 	rte_spinlock_unlock(&vq->access_lock);
 
-	return pkt_idx;
+	return count;
 }
 
 uint16_t
@@ -1014,48 +1025,13 @@ restore_mbuf(struct rte_mbuf *m)
 	}
 }
 
-uint16_t
-rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
+static __rte_always_inline uint16_t
+virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
 {
-	struct virtio_net *dev;
-	struct rte_mbuf *rarp_mbuf = NULL;
-	struct vhost_virtqueue *vq;
-	uint32_t i = 0;
+	uint16_t i;
 	uint16_t free_entries;
 
-	dev = get_device(vid);
-	if (!dev)
-		return 0;
-
-	if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
-		RTE_LOG(ERR, VHOST_DATA,
-			"(%d) %s: built-in vhost net backend is disabled.\n",
-			dev->vid, __func__);
-		return 0;
-	}
-
-	if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->nr_vring))) {
-		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
-			dev->vid, __func__, queue_id);
-		return 0;
-	}
-
-	vq = dev->virtqueue[queue_id];
-
-	if (unlikely(rte_spinlock_trylock(&vq->access_lock) == 0))
-		return 0;
-
-	if (unlikely(vq->enabled == 0))
-		goto out_access_unlock;
-
-	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
-		vhost_user_iotlb_rd_lock(vq);
-
-	if (unlikely(vq->access_ok == 0))
-		if (unlikely(vring_translate(dev, vq) < 0))
-			goto out;
-
 	if (unlikely(dev->dequeue_zero_copy)) {
 		struct zcopy_mbuf *zmbuf, *next;
 		int nr_updated = 0;
@@ -1082,39 +1058,10 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 
 	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 
-	/*
-	 * Construct a RARP broadcast packet, and inject it to the "pkts"
-	 * array, to looks like that guest actually send such packet.
-	 *
-	 * Check user_send_rarp() for more information.
-	 *
-	 * broadcast_rarp shares a cacheline in the virtio_net structure
-	 * with some fields that are accessed during enqueue and
-	 * rte_atomic16_cmpset() causes a write if using cmpxchg. This could
-	 * result in false sharing between enqueue and dequeue.
-	 *
-	 * Prevent unnecessary false sharing by reading broadcast_rarp first
-	 * and only performing cmpset if the read indicates it is likely to
-	 * be set.
-	 */
-
-	if (unlikely(rte_atomic16_read(&dev->broadcast_rarp) &&
-			rte_atomic16_cmpset((volatile uint16_t *)
-				&dev->broadcast_rarp.cnt, 1, 0))) {
-
-		rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
-		if (rarp_mbuf == NULL) {
-			RTE_LOG(ERR, VHOST_DATA,
-				"Failed to make RARP packet.\n");
-			return 0;
-		}
-		count -= 1;
-	}
-
 	free_entries = *((volatile uint16_t *)&vq->avail->idx) -
 		vq->last_avail_idx;
 	if (free_entries == 0)
-		goto out;
+		return 0;
 
 	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
 
@@ -1126,10 +1073,10 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 	for (i = 0; i < count; i++) {
 		struct buf_vector buf_vec[BUF_VECTOR_MAX];
 		uint16_t head_idx, dummy_len;
-		uint32_t nr_vec = 0;
+		uint16_t nr_vec = 0;
 		int err;
 
-		if (unlikely(fill_vec_buf(dev, vq,
+		if (unlikely(fill_vec_buf_split(dev, vq,
 						vq->last_avail_idx + i,
 						&nr_vec, buf_vec,
 						&head_idx, &dummy_len,
@@ -1188,6 +1135,81 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 		vhost_vring_call(dev, vq);
 	}
 
+	return i;
+}
+
+uint16_t
+rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
+	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
+{
+	struct virtio_net *dev;
+	struct rte_mbuf *rarp_mbuf = NULL;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (!dev)
+		return 0;
+
+	if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
+		RTE_LOG(ERR, VHOST_DATA,
+			"(%d) %s: built-in vhost net backend is disabled.\n",
+			dev->vid, __func__);
+		return 0;
+	}
+
+	if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->nr_vring))) {
+		RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
+			dev->vid, __func__, queue_id);
+		return 0;
+	}
+
+	vq = dev->virtqueue[queue_id];
+
+	if (unlikely(rte_spinlock_trylock(&vq->access_lock) == 0))
+		return 0;
+
+	if (unlikely(vq->enabled == 0))
+		goto out_access_unlock;
+
+	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
+		vhost_user_iotlb_rd_lock(vq);
+
+	if (unlikely(vq->access_ok == 0))
+		if (unlikely(vring_translate(dev, vq) < 0))
+			goto out;
+
+
+	/*
+	 * Construct a RARP broadcast packet, and inject it to the "pkts"
+	 * array, to looks like that guest actually send such packet.
+	 *
+	 * Check user_send_rarp() for more information.
+	 *
+	 * broadcast_rarp shares a cacheline in the virtio_net structure
+	 * with some fields that are accessed during enqueue and
+	 * rte_atomic16_cmpset() causes a write if using cmpxchg. This could
+	 * result in false sharing between enqueue and dequeue.
+	 *
+	 * Prevent unnecessary false sharing by reading broadcast_rarp first
+	 * and only performing cmpset if the read indicates it is likely to
+	 * be set.
+	 */
+
+	if (unlikely(rte_atomic16_read(&dev->broadcast_rarp) &&
+			rte_atomic16_cmpset((volatile uint16_t *)
+				&dev->broadcast_rarp.cnt, 1, 0))) {
+
+		rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
+		if (rarp_mbuf == NULL) {
+			RTE_LOG(ERR, VHOST_DATA,
+				"Failed to make RARP packet.\n");
+			return 0;
+		}
+		count -= 1;
+	}
+
+	count = virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count);
+
 out:
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 		vhost_user_iotlb_rd_unlock(vq);
@@ -1200,10 +1222,10 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 		 * Inject it to the head of "pkts" array, so that switch's mac
 		 * learning table will get updated first.
 		 */
-		memmove(&pkts[1], pkts, i * sizeof(struct rte_mbuf *));
+		memmove(&pkts[1], pkts, count * sizeof(struct rte_mbuf *));
 		pkts[0] = rarp_mbuf;
-		i += 1;
+		count += 1;
 	}
 
-	return i;
+	return count;
 }
-- 
2.14.4

Re: [PATCH v6 07/15] vhost: extract split ring handling from Rx and Tx functions

[Tiwei Bie]

On Mon, Jul 02, 2018 at 10:16:21AM +0200, Maxime Coquelin wrote:
[...]
> +
> +	if (unlikely(vq->access_ok == 0))
> +		if (unlikely(vring_translate(dev, vq) < 0))
> +			goto out;
> +
> +

Please just keep one blank line.

> +	/*
> +	 * Construct a RARP broadcast packet, and inject it to the "pkts"
> +	 * array, to looks like that guest actually send such packet.
> +	 *
> +	 * Check user_send_rarp() for more information.
> +	 *
> +	 * broadcast_rarp shares a cacheline in the virtio_net structure
> +	 * with some fields that are accessed during enqueue and
> +	 * rte_atomic16_cmpset() causes a write if using cmpxchg. This could
> +	 * result in false sharing between enqueue and dequeue.
> +	 *
> +	 * Prevent unnecessary false sharing by reading broadcast_rarp first
> +	 * and only performing cmpset if the read indicates it is likely to
> +	 * be set.
> +	 */
> +

Please remove above blank line.

> +	if (unlikely(rte_atomic16_read(&dev->broadcast_rarp) &&
> +			rte_atomic16_cmpset((volatile uint16_t *)
> +				&dev->broadcast_rarp.cnt, 1, 0))) {
> +
> +		rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
> +		if (rarp_mbuf == NULL) {
> +			RTE_LOG(ERR, VHOST_DATA,
> +				"Failed to make RARP packet.\n");
> +			return 0;
> +		}
> +		count -= 1;
> +	}
> +
> +	count = virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count);
> +
>  out:
>  	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
>  		vhost_user_iotlb_rd_unlock(vq);
> @@ -1200,10 +1222,10 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
>  		 * Inject it to the head of "pkts" array, so that switch's mac
>  		 * learning table will get updated first.
>  		 */
> -		memmove(&pkts[1], pkts, i * sizeof(struct rte_mbuf *));
> +		memmove(&pkts[1], pkts, count * sizeof(struct rte_mbuf *));
>  		pkts[0] = rarp_mbuf;
> -		i += 1;
> +		count += 1;
>  	}
>  
> -	return i;
> +	return count;
>  }
> -- 
> 2.14.4
> 

Re: [PATCH v6 07/15] vhost: extract split ring handling from Rx and Tx functions

[Maxime Coquelin]

On 07/04/2018 08:51 AM, Tiwei Bie wrote:
> On Mon, Jul 02, 2018 at 10:16:21AM +0200, Maxime Coquelin wrote:
> [...]
>> +
>> +	if (unlikely(vq->access_ok == 0))
>> +		if (unlikely(vring_translate(dev, vq) < 0))
>> +			goto out;
>> +
>> +
> 
> Please just keep one blank line.
> 
>> +	/*
>> +	 * Construct a RARP broadcast packet, and inject it to the "pkts"
>> +	 * array, to looks like that guest actually send such packet.
>> +	 *
>> +	 * Check user_send_rarp() for more information.
>> +	 *
>> +	 * broadcast_rarp shares a cacheline in the virtio_net structure
>> +	 * with some fields that are accessed during enqueue and
>> +	 * rte_atomic16_cmpset() causes a write if using cmpxchg. This could
>> +	 * result in false sharing between enqueue and dequeue.
>> +	 *
>> +	 * Prevent unnecessary false sharing by reading broadcast_rarp first
>> +	 * and only performing cmpset if the read indicates it is likely to
>> +	 * be set.
>> +	 */
>> +
> 
> Please remove above blank line.
> 
>> +	if (unlikely(rte_atomic16_read(&dev->broadcast_rarp) &&
>> +			rte_atomic16_cmpset((volatile uint16_t *)
>> +				&dev->broadcast_rarp.cnt, 1, 0))) {
>> +
>> +		rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
>> +		if (rarp_mbuf == NULL) {
>> +			RTE_LOG(ERR, VHOST_DATA,
>> +				"Failed to make RARP packet.\n");
>> +			return 0;
>> +		}
>> +		count -= 1;
>> +	}
>> +
>> +	count = virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count);
>> +
>>   out:
>>   	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
>>   		vhost_user_iotlb_rd_unlock(vq);
>> @@ -1200,10 +1222,10 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
>>   		 * Inject it to the head of "pkts" array, so that switch's mac
>>   		 * learning table will get updated first.
>>   		 */
>> -		memmove(&pkts[1], pkts, i * sizeof(struct rte_mbuf *));
>> +		memmove(&pkts[1], pkts, count * sizeof(struct rte_mbuf *));
>>   		pkts[0] = rarp_mbuf;
>> -		i += 1;
>> +		count += 1;
>>   	}
>>   
>> -	return i;
>> +	return count;
>>   }
>> -- 
>> 2.14.4
>>

Done,

Thanks,
Maxime

[PATCH v6 08/15] vhost: append shadow used ring function names with split

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 28 +++++++++++++++-------------
 1 file changed, 15 insertions(+), 13 deletions(-)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 385876527..35f8cf90a 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -77,8 +77,9 @@ free_ind_table(void *idesc)
 }
 
 static __rte_always_inline void
-do_flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq,
-			  uint16_t to, uint16_t from, uint16_t size)
+do_flush_shadow_used_ring_split(struct virtio_net *dev,
+			struct vhost_virtqueue *vq,
+			uint16_t to, uint16_t from, uint16_t size)
 {
 	rte_memcpy(&vq->used->ring[to],
 			&vq->shadow_used_ring[from],
@@ -89,22 +90,22 @@ do_flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq,
 }
 
 static __rte_always_inline void
-flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq)
+flush_shadow_used_ring_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
 	uint16_t used_idx = vq->last_used_idx & (vq->size - 1);
 
 	if (used_idx + vq->shadow_used_idx <= vq->size) {
-		do_flush_shadow_used_ring(dev, vq, used_idx, 0,
+		do_flush_shadow_used_ring_split(dev, vq, used_idx, 0,
 					  vq->shadow_used_idx);
 	} else {
 		uint16_t size;
 
 		/* update used ring interval [used_idx, vq->size] */
 		size = vq->size - used_idx;
-		do_flush_shadow_used_ring(dev, vq, used_idx, 0, size);
+		do_flush_shadow_used_ring_split(dev, vq, used_idx, 0, size);
 
 		/* update the left half used ring interval [0, left_size] */
-		do_flush_shadow_used_ring(dev, vq, 0, size,
+		do_flush_shadow_used_ring_split(dev, vq, 0, size,
 					  vq->shadow_used_idx - size);
 	}
 	vq->last_used_idx += vq->shadow_used_idx;
@@ -120,7 +121,7 @@ flush_shadow_used_ring(struct virtio_net *dev, struct vhost_virtqueue *vq)
 }
 
 static __rte_always_inline void
-update_shadow_used_ring(struct vhost_virtqueue *vq,
+update_shadow_used_ring_split(struct vhost_virtqueue *vq,
 			 uint16_t desc_idx, uint16_t len)
 {
 	uint16_t i = vq->shadow_used_idx++;
@@ -347,7 +348,7 @@ reserve_avail_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 						VHOST_ACCESS_RO) < 0))
 			return -1;
 		len = RTE_MIN(len, size);
-		update_shadow_used_ring(vq, head_idx, len);
+		update_shadow_used_ring_split(vq, head_idx, len);
 		size -= len;
 
 		cur_idx++;
@@ -573,7 +574,7 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	do_data_copy_enqueue(dev, vq);
 
 	if (likely(vq->shadow_used_idx)) {
-		flush_shadow_used_ring(dev, vq);
+		flush_shadow_used_ring_split(dev, vq);
 		vhost_vring_call(dev, vq);
 	}
 
@@ -1041,7 +1042,8 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			next = TAILQ_NEXT(zmbuf, next);
 
 			if (mbuf_is_consumed(zmbuf->mbuf)) {
-				update_shadow_used_ring(vq, zmbuf->desc_idx, 0);
+				update_shadow_used_ring_split(vq,
+						zmbuf->desc_idx, 0);
 				nr_updated += 1;
 
 				TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
@@ -1052,7 +1054,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			}
 		}
 
-		flush_shadow_used_ring(dev, vq);
+		flush_shadow_used_ring_split(dev, vq);
 		vhost_vring_call(dev, vq);
 	}
 
@@ -1084,7 +1086,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			break;
 
 		if (likely(dev->dequeue_zero_copy == 0))
-			update_shadow_used_ring(vq, head_idx, 0);
+			update_shadow_used_ring_split(vq, head_idx, 0);
 
 		rte_prefetch0((void *)(uintptr_t)buf_vec[0].buf_addr);
 
@@ -1131,7 +1133,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		do_data_copy_dequeue(vq);
 		if (unlikely(i < count))
 			vq->shadow_used_idx = i;
-		flush_shadow_used_ring(dev, vq);
+		flush_shadow_used_ring_split(dev, vq);
 		vhost_vring_call(dev, vq);
 	}
 
-- 
2.14.4

[PATCH v6 09/15] vhost: add shadow used ring support for packed rings

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/vhost.c      |  9 ++++--
 lib/librte_vhost/vhost.h      | 13 ++++++--
 lib/librte_vhost/vhost_user.c | 64 ++++++++++++++++++++++++++++----------
 lib/librte_vhost/virtio_net.c | 72 +++++++++++++++++++++++++++++++++++++++++--
 4 files changed, 134 insertions(+), 24 deletions(-)

diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 05d2bb182..8538302c9 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -94,9 +94,12 @@ cleanup_device(struct virtio_net *dev, int destroy)
 }
 
 void
-free_vq(struct vhost_virtqueue *vq)
+free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
-	rte_free(vq->shadow_used_ring);
+	if (vq_is_packed(dev))
+		rte_free(vq->shadow_used_packed);
+	else
+		rte_free(vq->shadow_used_split);
 	rte_free(vq->batch_copy_elems);
 	rte_mempool_free(vq->iotlb_pool);
 	rte_free(vq);
@@ -111,7 +114,7 @@ free_device(struct virtio_net *dev)
 	uint32_t i;
 
 	for (i = 0; i < dev->nr_vring; i++)
-		free_vq(dev->virtqueue[i]);
+		free_vq(dev, dev->virtqueue[i]);
 
 	rte_free(dev);
 }
diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index 35beff699..a7320469a 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -80,6 +80,12 @@ struct log_cache_entry {
 	unsigned long val;
 };
 
+struct vring_used_elem_packed {
+	uint16_t id;
+	uint32_t len;
+	uint32_t count;
+};
+
 /**
  * Structure contains variables relevant to RX/TX virtqueues.
  */
@@ -119,7 +125,10 @@ struct vhost_virtqueue {
 	struct zcopy_mbuf	*zmbufs;
 	struct zcopy_mbuf_list	zmbuf_list;
 
-	struct vring_used_elem  *shadow_used_ring;
+	union {
+		struct vring_used_elem  *shadow_used_split;
+		struct vring_used_elem_packed *shadow_used_packed;
+	};
 	uint16_t                shadow_used_idx;
 	struct vhost_vring_addr ring_addrs;
 
@@ -552,7 +561,7 @@ void reset_device(struct virtio_net *dev);
 void vhost_destroy_device(int);
 
 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
-void free_vq(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 ae03ec29b..b2b57de57 100644
--- a/lib/librte_vhost/vhost_user.c
+++ b/lib/librte_vhost/vhost_user.c
@@ -244,7 +244,7 @@ vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 
 			dev->virtqueue[dev->nr_vring] = NULL;
 			cleanup_vq(vq, 1);
-			free_vq(vq);
+			free_vq(dev, vq);
 		}
 	}
 
@@ -293,13 +293,26 @@ vhost_user_set_vring_num(struct virtio_net *dev,
 		TAILQ_INIT(&vq->zmbuf_list);
 	}
 
-	vq->shadow_used_ring = rte_malloc(NULL,
+	if (vq_is_packed(dev)) {
+		vq->shadow_used_packed = rte_malloc(NULL,
+				vq->size *
+				sizeof(struct vring_used_elem_packed),
+				RTE_CACHE_LINE_SIZE);
+		if (!vq->shadow_used_packed) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"failed to allocate memory for shadow used ring.\n");
+			return -1;
+		}
+
+	} else {
+		vq->shadow_used_split = rte_malloc(NULL,
 				vq->size * sizeof(struct vring_used_elem),
 				RTE_CACHE_LINE_SIZE);
-	if (!vq->shadow_used_ring) {
-		RTE_LOG(ERR, VHOST_CONFIG,
-			"failed to allocate memory for shadow used ring.\n");
-		return -1;
+		if (!vq->shadow_used_split) {
+			RTE_LOG(ERR, VHOST_CONFIG,
+					"failed to allocate memory for shadow used ring.\n");
+			return -1;
+		}
 	}
 
 	vq->batch_copy_elems = rte_malloc(NULL,
@@ -326,7 +339,8 @@ numa_realloc(struct virtio_net *dev, int index)
 	struct virtio_net *old_dev;
 	struct vhost_virtqueue *old_vq, *vq;
 	struct zcopy_mbuf *new_zmbuf;
-	struct vring_used_elem *new_shadow_used_ring;
+	struct vring_used_elem *new_shadow_used_split;
+	struct vring_used_elem_packed *new_shadow_used_packed;
 	struct batch_copy_elem *new_batch_copy_elems;
 	int ret;
 
@@ -361,13 +375,26 @@ numa_realloc(struct virtio_net *dev, int index)
 			vq->zmbufs = new_zmbuf;
 		}
 
-		new_shadow_used_ring = rte_malloc_socket(NULL,
-			vq->size * sizeof(struct vring_used_elem),
-			RTE_CACHE_LINE_SIZE,
-			newnode);
-		if (new_shadow_used_ring) {
-			rte_free(vq->shadow_used_ring);
-			vq->shadow_used_ring = new_shadow_used_ring;
+		if (vq_is_packed(dev)) {
+			new_shadow_used_packed = rte_malloc_socket(NULL,
+					vq->size *
+					sizeof(struct vring_used_elem_packed),
+					RTE_CACHE_LINE_SIZE,
+					newnode);
+			if (new_shadow_used_packed) {
+				rte_free(vq->shadow_used_packed);
+				vq->shadow_used_packed = new_shadow_used_packed;
+			}
+		} else {
+			new_shadow_used_split = rte_malloc_socket(NULL,
+					vq->size *
+					sizeof(struct vring_used_elem),
+					RTE_CACHE_LINE_SIZE,
+					newnode);
+			if (new_shadow_used_split) {
+				rte_free(vq->shadow_used_split);
+				vq->shadow_used_split = new_shadow_used_split;
+			}
 		}
 
 		new_batch_copy_elems = rte_malloc_socket(NULL,
@@ -1067,8 +1094,13 @@ vhost_user_get_vring_base(struct virtio_net *dev,
 
 	if (dev->dequeue_zero_copy)
 		free_zmbufs(vq);
-	rte_free(vq->shadow_used_ring);
-	vq->shadow_used_ring = NULL;
+	if (vq_is_packed(dev)) {
+		rte_free(vq->shadow_used_packed);
+		vq->shadow_used_packed = NULL;
+	} else {
+		rte_free(vq->shadow_used_split);
+		vq->shadow_used_split = NULL;
+	}
 
 	rte_free(vq->batch_copy_elems);
 	vq->batch_copy_elems = NULL;
diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 35f8cf90a..f0e2e6a1f 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -20,6 +20,7 @@
 
 #include "iotlb.h"
 #include "vhost.h"
+#include "virtio-packed.h"
 
 #define MAX_PKT_BURST 32
 
@@ -82,7 +83,7 @@ do_flush_shadow_used_ring_split(struct virtio_net *dev,
 			uint16_t to, uint16_t from, uint16_t size)
 {
 	rte_memcpy(&vq->used->ring[to],
-			&vq->shadow_used_ring[from],
+			&vq->shadow_used_split[from],
 			size * sizeof(struct vring_used_elem));
 	vhost_log_cache_used_vring(dev, vq,
 			offsetof(struct vring_used, ring[to]),
@@ -126,8 +127,73 @@ update_shadow_used_ring_split(struct vhost_virtqueue *vq,
 {
 	uint16_t i = vq->shadow_used_idx++;
 
-	vq->shadow_used_ring[i].id  = desc_idx;
-	vq->shadow_used_ring[i].len = len;
+	vq->shadow_used_split[i].id  = desc_idx;
+	vq->shadow_used_split[i].len = len;
+}
+
+static __rte_always_inline void
+flush_shadow_used_ring_packed(struct virtio_net *dev,
+			struct vhost_virtqueue *vq)
+{
+	int i;
+	uint16_t used_idx = vq->last_used_idx;
+
+	/* Split loop in two to save memory barriers */
+	for (i = 0; i < vq->shadow_used_idx; i++) {
+		vq->desc_packed[used_idx].index = vq->shadow_used_packed[i].id;
+		vq->desc_packed[used_idx].len = vq->shadow_used_packed[i].len;
+
+		used_idx += vq->shadow_used_packed[i].count;
+		if (used_idx >= vq->size)
+			used_idx -= vq->size;
+	}
+
+	rte_smp_wmb();
+
+	for (i = 0; i < vq->shadow_used_idx; i++) {
+		uint16_t flags;
+
+		if (vq->shadow_used_packed[i].len)
+			flags = VRING_DESC_F_WRITE;
+		else
+			flags = 0;
+
+		if (vq->used_wrap_counter) {
+			flags |= VRING_DESC_F_USED;
+			flags |= VRING_DESC_F_AVAIL;
+		} else {
+			flags &= ~VRING_DESC_F_USED;
+			flags &= ~VRING_DESC_F_AVAIL;
+		}
+
+		vq->desc_packed[vq->last_used_idx].flags = flags;
+
+		vhost_log_cache_used_vring(dev, vq,
+					vq->last_used_idx *
+					sizeof(struct vring_desc_packed),
+					sizeof(struct vring_desc_packed));
+
+		vq->last_used_idx += vq->shadow_used_packed[i].count;
+		if (vq->last_used_idx >= vq->size) {
+			vq->used_wrap_counter ^= 1;
+			vq->last_used_idx -= vq->size;
+		}
+	}
+
+	rte_smp_wmb();
+	vq->shadow_used_idx = 0;
+	vhost_log_cache_sync(dev, vq);
+}
+
+static __rte_always_inline void
+update_shadow_used_ring_packed(struct vhost_virtqueue *vq,
+			 uint16_t desc_idx, uint16_t len, uint16_t count)
+{
+	uint16_t i = vq->shadow_used_idx++;
+
+	vq->shadow_used_packed[i].id  = desc_idx;
+	vq->shadow_used_packed[i].len = len;
+	vq->shadow_used_packed[i].count = count;
 }
 
 static inline void
-- 
2.14.4

[PATCH v6 10/15] vhost: create descriptor mapping function

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 61 ++++++++++++++++++++++++++-----------------
 1 file changed, 37 insertions(+), 24 deletions(-)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index f0e2e6a1f..64664b7de 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -292,6 +292,37 @@ virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
 	}
 }
 
+static __rte_always_inline int
+map_one_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
+		struct buf_vector *buf_vec, uint16_t *vec_idx,
+		uint64_t desc_iova, uint64_t desc_len, uint8_t perm)
+{
+	uint16_t vec_id = *vec_idx;
+
+	while (desc_len) {
+		uint64_t desc_addr;
+		uint64_t desc_chunck_len = desc_len;
+
+		desc_addr = vhost_iova_to_vva(dev, vq,
+				desc_iova,
+				&desc_chunck_len,
+				perm);
+		if (unlikely(!desc_addr))
+			return -1;
+
+		buf_vec[vec_id].buf_iova = desc_iova;
+		buf_vec[vec_id].buf_addr = desc_addr;
+		buf_vec[vec_id].buf_len  = desc_chunck_len;
+
+		desc_len -= desc_chunck_len;
+		desc_iova += desc_chunck_len;
+		vec_id++;
+	}
+	*vec_idx = vec_id;
+
+	return 0;
+}
+
 static __rte_always_inline int
 fill_vec_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			 uint32_t avail_idx, uint16_t *vec_idx,
@@ -301,7 +332,7 @@ fill_vec_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	uint16_t idx = vq->avail->ring[avail_idx & (vq->size - 1)];
 	uint16_t vec_id = *vec_idx;
 	uint32_t len    = 0;
-	uint64_t dlen, desc_avail, desc_iova;
+	uint64_t dlen;
 	struct vring_desc *descs = vq->desc;
 	struct vring_desc *idesc = NULL;
 
@@ -339,30 +370,12 @@ fill_vec_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		}
 
 		len += descs[idx].len;
-		desc_avail = descs[idx].len;
-		desc_iova = descs[idx].addr;
-
-		while (desc_avail) {
-			uint64_t desc_addr;
-			uint64_t desc_chunck_len = desc_avail;
-
-			desc_addr = vhost_iova_to_vva(dev, vq,
-					desc_iova,
-					&desc_chunck_len,
-					perm);
-			if (unlikely(!desc_addr)) {
-				free_ind_table(idesc);
-				return -1;
-			}
-
-			buf_vec[vec_id].buf_iova = desc_iova;
-			buf_vec[vec_id].buf_addr = desc_addr;
-			buf_vec[vec_id].buf_len  = desc_chunck_len;
-			buf_vec[vec_id].desc_idx = idx;
 
-			desc_avail -= desc_chunck_len;
-			desc_iova += desc_chunck_len;
-			vec_id++;
+		if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
+						descs[idx].addr, descs[idx].len,
+						perm))) {
+			free_ind_table(idesc);
+			return -1;
 		}
 
 		if ((descs[idx].flags & VRING_DESC_F_NEXT) == 0)
-- 
2.14.4

Re: [PATCH v6 10/15] vhost: create descriptor mapping function

[Tiwei Bie]

On Mon, Jul 02, 2018 at 10:16:24AM +0200, Maxime Coquelin wrote:
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  lib/librte_vhost/virtio_net.c | 61 ++++++++++++++++++++++++++-----------------
>  1 file changed, 37 insertions(+), 24 deletions(-)
> 
> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
> index f0e2e6a1f..64664b7de 100644
> --- a/lib/librte_vhost/virtio_net.c
> +++ b/lib/librte_vhost/virtio_net.c
> @@ -292,6 +292,37 @@ virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
>  	}
>  }
>  
> +static __rte_always_inline int
> +map_one_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
> +		struct buf_vector *buf_vec, uint16_t *vec_idx,
> +		uint64_t desc_iova, uint64_t desc_len, uint8_t perm)
> +{
> +	uint16_t vec_id = *vec_idx;
> +
> +	while (desc_len) {
> +		uint64_t desc_addr;
> +		uint64_t desc_chunck_len = desc_len;
> +
> +		desc_addr = vhost_iova_to_vva(dev, vq,
> +				desc_iova,
> +				&desc_chunck_len,
> +				perm);
> +		if (unlikely(!desc_addr))
> +			return -1;
> +
> +		buf_vec[vec_id].buf_iova = desc_iova;
> +		buf_vec[vec_id].buf_addr = desc_addr;
> +		buf_vec[vec_id].buf_len  = desc_chunck_len;
> +
> +		desc_len -= desc_chunck_len;
> +		desc_iova += desc_chunck_len;
> +		vec_id++;

FYI, a rebase on top of the latest "vhost: generalize
buffer vectors" series is needed to make sure that
access buf_vec[vec_id] won't cause overflow.

> +	}
> +	*vec_idx = vec_id;
> +
> +	return 0;
> +}
> +
[...]

Re: [PATCH v6 10/15] vhost: create descriptor mapping function

[Maxime Coquelin]

On 07/04/2018 07:56 AM, Tiwei Bie wrote:
> On Mon, Jul 02, 2018 at 10:16:24AM +0200, Maxime Coquelin wrote:
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>>   lib/librte_vhost/virtio_net.c | 61 ++++++++++++++++++++++++++-----------------
>>   1 file changed, 37 insertions(+), 24 deletions(-)
>>
>> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
>> index f0e2e6a1f..64664b7de 100644
>> --- a/lib/librte_vhost/virtio_net.c
>> +++ b/lib/librte_vhost/virtio_net.c
>> @@ -292,6 +292,37 @@ virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
>>   	}
>>   }
>>   
>> +static __rte_always_inline int
>> +map_one_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
>> +		struct buf_vector *buf_vec, uint16_t *vec_idx,
>> +		uint64_t desc_iova, uint64_t desc_len, uint8_t perm)
>> +{
>> +	uint16_t vec_id = *vec_idx;
>> +
>> +	while (desc_len) {
>> +		uint64_t desc_addr;
>> +		uint64_t desc_chunck_len = desc_len;
>> +
>> +		desc_addr = vhost_iova_to_vva(dev, vq,
>> +				desc_iova,
>> +				&desc_chunck_len,
>> +				perm);
>> +		if (unlikely(!desc_addr))
>> +			return -1;
>> +
>> +		buf_vec[vec_id].buf_iova = desc_iova;
>> +		buf_vec[vec_id].buf_addr = desc_addr;
>> +		buf_vec[vec_id].buf_len  = desc_chunck_len;
>> +
>> +		desc_len -= desc_chunck_len;
>> +		desc_iova += desc_chunck_len;
>> +		vec_id++;
> 
> FYI, a rebase on top of the latest "vhost: generalize
> buffer vectors" series is needed to make sure that
> access buf_vec[vec_id] won't cause overflow.

Right, this is fixed now that I rebased.

Thanks,
Maxime

>> +	}
>> +	*vec_idx = vec_id;
>> +
>> +	return 0;
>> +}
>> +
> [...]
> 

[PATCH v6 11/15] vhost: add vector filling support for packed ring

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 111 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 111 insertions(+)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 64664b7de..2d867e88e 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -448,6 +448,117 @@ reserve_avail_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	return 0;
 }
 
+static __rte_always_inline int
+fill_vec_buf_packed_indirect(struct virtio_net *dev,
+			struct vhost_virtqueue *vq,
+			struct vring_desc_packed *desc, uint16_t *vec_idx,
+			struct buf_vector *buf_vec, uint16_t *len, uint8_t perm)
+{
+	uint16_t i;
+	uint32_t nr_decs;
+	uint16_t vec_id = *vec_idx;
+	uint64_t dlen;
+	struct vring_desc_packed *descs, *idescs = NULL;
+
+	dlen = desc->len;
+	descs = (struct vring_desc_packed *)(uintptr_t)
+		vhost_iova_to_vva(dev, vq, desc->addr, &dlen, VHOST_ACCESS_RO);
+	if (unlikely(!descs))
+		return -1;
+
+	if (unlikely(dlen < desc->len)) {
+		/*
+		 * The indirect desc table is not contiguous
+		 * in process VA space, we have to copy it.
+		 */
+		idescs = alloc_copy_ind_table(dev, vq, desc->addr, desc->len);
+		if (unlikely(!idescs))
+			return -1;
+
+		descs = idescs;
+	}
+
+	nr_decs =  desc->len / sizeof(struct vring_desc_packed);
+	if (unlikely(nr_decs >= vq->size)) {
+		free_ind_table(idescs);
+		return -1;
+	}
+
+	for (i = 0; i < nr_decs; i++) {
+		if (unlikely(vec_id >= BUF_VECTOR_MAX)) {
+			free_ind_table(idescs);
+			return -1;
+		}
+
+		*len += descs[i].len;
+		if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
+						descs[i].addr, descs[i].len,
+						perm)))
+			return -1;
+	}
+	*vec_idx = vec_id;
+
+	if (unlikely(!!idescs))
+		free_ind_table(idescs);
+
+	return 0;
+}
+
+static inline int
+fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+				uint16_t avail_idx, uint16_t *desc_count,
+				struct buf_vector *buf_vec, uint16_t *vec_idx,
+				uint16_t *buf_id, uint16_t *len, uint8_t perm)
+{
+	bool wrap_counter = vq->avail_wrap_counter;
+	struct vring_desc_packed *descs = vq->desc_packed;
+	uint16_t vec_id = *vec_idx;
+
+	if (avail_idx < vq->last_avail_idx)
+		wrap_counter ^= 1;
+
+	if (unlikely(!desc_is_avail(&descs[avail_idx], wrap_counter)))
+		return -1;
+
+	*desc_count = 0;
+
+	while (1) {
+		if (unlikely(vec_id >= BUF_VECTOR_MAX))
+			return -1;
+
+		*desc_count += 1;
+		*buf_id = descs[avail_idx].index;
+
+		if (descs[avail_idx].flags & VRING_DESC_F_INDIRECT) {
+			if (unlikely(fill_vec_buf_packed_indirect(dev, vq,
+							&descs[avail_idx],
+							&vec_id, buf_vec,
+							len, perm) < 0))
+				return -1;
+		} else {
+			*len += descs[avail_idx].len;
+
+			if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
+							descs[avail_idx].addr,
+							descs[avail_idx].len,
+							perm)))
+				return -1;
+		}
+
+		if ((descs[avail_idx].flags & VRING_DESC_F_NEXT) == 0)
+			break;
+
+		if (++avail_idx >= vq->size) {
+			avail_idx -= vq->size;
+			wrap_counter ^= 1;
+		}
+	}
+
+	*vec_idx = vec_id;
+
+	return 0;
+}
+
 static __rte_always_inline int
 copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			    struct rte_mbuf *m, struct buf_vector *buf_vec,
-- 
2.14.4

Re: [PATCH v6 11/15] vhost: add vector filling support for packed ring

[Tiwei Bie]

On Mon, Jul 02, 2018 at 10:16:25AM +0200, Maxime Coquelin wrote:
[...]
> +static __rte_always_inline int
> +fill_vec_buf_packed_indirect(struct virtio_net *dev,
> +			struct vhost_virtqueue *vq,
> +			struct vring_desc_packed *desc, uint16_t *vec_idx,
> +			struct buf_vector *buf_vec, uint16_t *len, uint8_t perm)
> +{
> +	uint16_t i;
> +	uint32_t nr_decs;
> +	uint16_t vec_id = *vec_idx;
> +	uint64_t dlen;
> +	struct vring_desc_packed *descs, *idescs = NULL;
> +
> +	dlen = desc->len;
> +	descs = (struct vring_desc_packed *)(uintptr_t)
> +		vhost_iova_to_vva(dev, vq, desc->addr, &dlen, VHOST_ACCESS_RO);
> +	if (unlikely(!descs))
> +		return -1;
> +
> +	if (unlikely(dlen < desc->len)) {
> +		/*
> +		 * The indirect desc table is not contiguous
> +		 * in process VA space, we have to copy it.
> +		 */
> +		idescs = alloc_copy_ind_table(dev, vq, desc->addr, desc->len);
> +		if (unlikely(!idescs))
> +			return -1;
> +
> +		descs = idescs;
> +	}
> +
> +	nr_decs =  desc->len / sizeof(struct vring_desc_packed);

s/nr_decs =  /nr_desc = /


> +	if (unlikely(nr_decs >= vq->size)) {
> +		free_ind_table(idescs);
> +		return -1;
> +	}
[...]
> +
> +static inline int
> +fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
> +				uint16_t avail_idx, uint16_t *desc_count,
> +				struct buf_vector *buf_vec, uint16_t *vec_idx,
> +				uint16_t *buf_id, uint16_t *len, uint8_t perm)
> +{
> +	bool wrap_counter = vq->avail_wrap_counter;
> +	struct vring_desc_packed *descs = vq->desc_packed;
> +	uint16_t vec_id = *vec_idx;
> +
> +	if (avail_idx < vq->last_avail_idx)
> +		wrap_counter ^= 1;

In which case avail_idx will be less than vq->last_avail_idx
and we need to wrap the wrap_counter?

> +
> +	if (unlikely(!desc_is_avail(&descs[avail_idx], wrap_counter)))
> +		return -1;
> +
> +	*desc_count = 0;
> +
> +	while (1) {
> +		if (unlikely(vec_id >= BUF_VECTOR_MAX))
> +			return -1;
> +
> +		*desc_count += 1;
> +		*buf_id = descs[avail_idx].index;
> +
> +		if (descs[avail_idx].flags & VRING_DESC_F_INDIRECT) {
> +			if (unlikely(fill_vec_buf_packed_indirect(dev, vq,
> +							&descs[avail_idx],
> +							&vec_id, buf_vec,
> +							len, perm) < 0))
> +				return -1;
> +		} else {
> +			*len += descs[avail_idx].len;
> +
> +			if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
> +							descs[avail_idx].addr,
> +							descs[avail_idx].len,
> +							perm)))
> +				return -1;
> +		}
> +
> +		if ((descs[avail_idx].flags & VRING_DESC_F_NEXT) == 0)
> +			break;
> +
> +		if (++avail_idx >= vq->size) {
> +			avail_idx -= vq->size;
> +			wrap_counter ^= 1;
> +		}
> +	}
> +
> +	*vec_idx = vec_id;
> +
> +	return 0;
> +}
[...]

Re: [PATCH v6 11/15] vhost: add vector filling support for packed ring

[Maxime Coquelin]

On 07/04/2018 07:53 AM, Tiwei Bie wrote:
> On Mon, Jul 02, 2018 at 10:16:25AM +0200, Maxime Coquelin wrote:
> [...]
>> +static __rte_always_inline int
>> +fill_vec_buf_packed_indirect(struct virtio_net *dev,
>> +			struct vhost_virtqueue *vq,
>> +			struct vring_desc_packed *desc, uint16_t *vec_idx,
>> +			struct buf_vector *buf_vec, uint16_t *len, uint8_t perm)
>> +{
>> +	uint16_t i;
>> +	uint32_t nr_decs;
>> +	uint16_t vec_id = *vec_idx;
>> +	uint64_t dlen;
>> +	struct vring_desc_packed *descs, *idescs = NULL;
>> +
>> +	dlen = desc->len;
>> +	descs = (struct vring_desc_packed *)(uintptr_t)
>> +		vhost_iova_to_vva(dev, vq, desc->addr, &dlen, VHOST_ACCESS_RO);
>> +	if (unlikely(!descs))
>> +		return -1;
>> +
>> +	if (unlikely(dlen < desc->len)) {
>> +		/*
>> +		 * The indirect desc table is not contiguous
>> +		 * in process VA space, we have to copy it.
>> +		 */
>> +		idescs = alloc_copy_ind_table(dev, vq, desc->addr, desc->len);
>> +		if (unlikely(!idescs))
>> +			return -1;
>> +
>> +		descs = idescs;
>> +	}
>> +
>> +	nr_decs =  desc->len / sizeof(struct vring_desc_packed);
> 
> s/nr_decs =  /nr_desc = /

Fixed.

> 
> 
>> +	if (unlikely(nr_decs >= vq->size)) {
>> +		free_ind_table(idescs);
>> +		return -1;
>> +	}
> [...]
>> +
>> +static inline int
>> +fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
>> +				uint16_t avail_idx, uint16_t *desc_count,
>> +				struct buf_vector *buf_vec, uint16_t *vec_idx,
>> +				uint16_t *buf_id, uint16_t *len, uint8_t perm)
>> +{
>> +	bool wrap_counter = vq->avail_wrap_counter;
>> +	struct vring_desc_packed *descs = vq->desc_packed;
>> +	uint16_t vec_id = *vec_idx;
>> +
>> +	if (avail_idx < vq->last_avail_idx)
>> +		wrap_counter ^= 1;
> 
> In which case avail_idx will be less than vq->last_avail_idx
> and we need to wrap the wrap_counter?

In the receive mergeable case, it can happen (see patch 12):

static inline int
reserve_avail_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
				uint32_t size, struct buf_vector *buf_vec,
				uint16_t *nr_vec, uint16_t *num_buffers,
				uint16_t *nr_descs)
{
	uint16_t avail_idx;
	uint16_t vec_idx = 0;
	uint16_t max_tries, tries = 0;

	uint16_t buf_id = 0;
	uint16_t len = 0;
	uint16_t desc_count;

	*num_buffers = 0;
	avail_idx = vq->last_avail_idx;

	if (rxvq_is_mergeable(dev))
		max_tries = vq->size;
	else
		max_tries = 1;

	while (size > 0) {
		if (unlikely(fill_vec_buf_packed(dev, vq,
						avail_idx, &desc_count,
						buf_vec, &vec_idx,
						&buf_id, &len,
						VHOST_ACCESS_RO) < 0))
			return -1;

		len = RTE_MIN(len, size);
		update_shadow_used_ring_packed(vq, buf_id, len, desc_count);
		size -= len;

		avail_idx += desc_count;
		if (avail_idx >= vq->size)
			avail_idx -= vq->size;

		*nr_descs += desc_count;
		tries++;
		*num_buffers += 1;

		/*
		 * if we tried all available ring items, and still
		 * can't get enough buf, it means something abnormal
		 * happened.
		 */
		if (unlikely(tries > max_tries))
			return -1;
	}

	*nr_vec = vec_idx;

	return 0;
}

>> +
>> +	if (unlikely(!desc_is_avail(&descs[avail_idx], wrap_counter)))
>> +		return -1;
>> +
>> +	*desc_count = 0;
>> +
>> +	while (1) {
>> +		if (unlikely(vec_id >= BUF_VECTOR_MAX))
>> +			return -1;
>> +
>> +		*desc_count += 1;
>> +		*buf_id = descs[avail_idx].index;
>> +
>> +		if (descs[avail_idx].flags & VRING_DESC_F_INDIRECT) {
>> +			if (unlikely(fill_vec_buf_packed_indirect(dev, vq,
>> +							&descs[avail_idx],
>> +							&vec_id, buf_vec,
>> +							len, perm) < 0))
>> +				return -1;
>> +		} else {
>> +			*len += descs[avail_idx].len;
>> +
>> +			if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
>> +							descs[avail_idx].addr,
>> +							descs[avail_idx].len,
>> +							perm)))
>> +				return -1;
>> +		}
>> +
>> +		if ((descs[avail_idx].flags & VRING_DESC_F_NEXT) == 0)
>> +			break;
>> +
>> +		if (++avail_idx >= vq->size) {
>> +			avail_idx -= vq->size;
>> +			wrap_counter ^= 1;
>> +		}
>> +	}
>> +
>> +	*vec_idx = vec_id;
>> +
>> +	return 0;
>> +}
> [...]
> 

[PATCH v6 12/15] vhost: add Rx support for packed ring

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/virtio_net.c | 117 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 116 insertions(+), 1 deletion(-)

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 2d867e88e..2e286e228 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -559,6 +559,65 @@ fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	return 0;
 }
 
+/*
+ * Returns -1 on fail, 0 on success
+ */
+static inline int
+reserve_avail_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+				uint32_t size, struct buf_vector *buf_vec,
+				uint16_t *nr_vec, uint16_t *num_buffers,
+				uint16_t *nr_descs)
+{
+	uint16_t avail_idx;
+	uint16_t vec_idx = 0;
+	uint16_t max_tries, tries = 0;
+
+	uint16_t buf_id = 0;
+	uint16_t len = 0;
+	uint16_t desc_count;
+
+	*num_buffers = 0;
+	avail_idx = vq->last_avail_idx;
+
+	if (rxvq_is_mergeable(dev))
+		max_tries = vq->size;
+	else
+		max_tries = 1;
+
+	while (size > 0) {
+		if (unlikely(fill_vec_buf_packed(dev, vq,
+						avail_idx, &desc_count,
+						buf_vec, &vec_idx,
+						&buf_id, &len,
+						VHOST_ACCESS_RO) < 0))
+			return -1;
+
+		len = RTE_MIN(len, size);
+		update_shadow_used_ring_packed(vq, buf_id, len, desc_count);
+		size -= len;
+
+		avail_idx += desc_count;
+		if (avail_idx >= vq->size)
+			avail_idx -= vq->size;
+
+		*nr_descs += desc_count;
+		tries++;
+		*num_buffers += 1;
+
+		/*
+		 * if we tried all available ring items, and still
+		 * can't get enough buf, it means something abnormal
+		 * happened.
+		 */
+		if (unlikely(tries > max_tries))
+			return -1;
+	}
+
+	*nr_vec = vec_idx;
+
+	return 0;
+}
+
 static __rte_always_inline int
 copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			    struct rte_mbuf *m, struct buf_vector *buf_vec,
@@ -771,6 +830,59 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	return pkt_idx;
 }
 
+static __rte_always_inline uint32_t
+virtio_dev_rx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+	struct rte_mbuf **pkts, uint32_t count)
+{
+	uint32_t pkt_idx = 0;
+	uint16_t num_buffers;
+	struct buf_vector buf_vec[BUF_VECTOR_MAX];
+
+	for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
+		uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
+		uint16_t nr_vec = 0;
+		uint16_t nr_descs = 0;
+
+		if (unlikely(reserve_avail_buf_packed(dev, vq,
+						pkt_len, buf_vec, &nr_vec,
+						&num_buffers, &nr_descs) < 0)) {
+			VHOST_LOG_DEBUG(VHOST_DATA,
+				"(%d) failed to get enough desc from vring\n",
+				dev->vid);
+			vq->shadow_used_idx -= num_buffers;
+			break;
+		}
+
+		rte_prefetch0((void *)(uintptr_t)buf_vec[0].buf_addr);
+
+		VHOST_LOG_DEBUG(VHOST_DATA, "(%d) current index %d | end index %d\n",
+			dev->vid, vq->last_avail_idx,
+			vq->last_avail_idx + num_buffers);
+
+		if (copy_mbuf_to_desc(dev, vq, pkts[pkt_idx],
+						buf_vec, nr_vec,
+						num_buffers) < 0) {
+			vq->shadow_used_idx -= num_buffers;
+			break;
+		}
+
+		vq->last_avail_idx += nr_descs;
+		if (vq->last_avail_idx >= vq->size) {
+			vq->last_avail_idx -= vq->size;
+			vq->avail_wrap_counter ^= 1;
+		}
+	}
+
+	do_data_copy_enqueue(dev, vq);
+
+	if (likely(vq->shadow_used_idx)) {
+		flush_shadow_used_ring_packed(dev, vq);
+		vhost_vring_call(dev, vq);
+	}
+
+	return pkt_idx;
+}
+
 static __rte_always_inline uint32_t
 virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 	struct rte_mbuf **pkts, uint32_t count)
@@ -802,7 +914,10 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
 	if (count == 0)
 		goto out;
 
-	count = virtio_dev_rx_split(dev, vq, pkts, count);
+	if (vq_is_packed(dev))
+		count = virtio_dev_rx_packed(dev, vq, pkts, count);
+	else
+		count = virtio_dev_rx_split(dev, vq, pkts, count);
 
 out:
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
-- 
2.14.4

[PATCH v6 13/15] vhost: add Tx support for packed ring

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/vhost.h      |   1 +
 lib/librte_vhost/virtio_net.c | 121 +++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 121 insertions(+), 1 deletion(-)

diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index a7320469a..6ea8fb896 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -56,6 +56,7 @@ struct buf_vector {
 struct zcopy_mbuf {
 	struct rte_mbuf *mbuf;
 	uint32_t desc_idx;
+	uint16_t desc_count;
 	uint16_t in_use;
 
 	TAILQ_ENTRY(zcopy_mbuf) next;
diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 2e286e228..03dd38235 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -1445,6 +1445,122 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	return i;
 }
 
+static __rte_always_inline uint16_t
+virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
+{
+	uint16_t i;
+
+	rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
+
+	if (unlikely(dev->dequeue_zero_copy)) {
+		struct zcopy_mbuf *zmbuf, *next;
+		int nr_updated = 0;
+
+		for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
+		     zmbuf != NULL; zmbuf = next) {
+			next = TAILQ_NEXT(zmbuf, next);
+
+			if (mbuf_is_consumed(zmbuf->mbuf)) {
+				update_shadow_used_ring_packed(vq,
+						zmbuf->desc_idx,
+						0,
+						zmbuf->desc_count);
+				nr_updated += 1;
+
+				TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
+				restore_mbuf(zmbuf->mbuf);
+				rte_pktmbuf_free(zmbuf->mbuf);
+				put_zmbuf(zmbuf);
+				vq->nr_zmbuf -= 1;
+			}
+		}
+
+		flush_shadow_used_ring_packed(dev, vq);
+		vhost_vring_call(dev, vq);
+	}
+
+	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
+
+	count = RTE_MIN(count, MAX_PKT_BURST);
+	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) about to dequeue %u buffers\n",
+			dev->vid, count);
+
+	for (i = 0; i < count; i++) {
+		struct buf_vector buf_vec[BUF_VECTOR_MAX];
+		uint16_t buf_id, dummy_len;
+		uint16_t desc_count, nr_vec = 0;
+		int err;
+
+		if (unlikely(fill_vec_buf_packed(dev, vq,
+						vq->last_avail_idx, &desc_count,
+						buf_vec, &nr_vec,
+						&buf_id, &dummy_len,
+						VHOST_ACCESS_RW) < 0))
+			break;
+
+		if (likely(dev->dequeue_zero_copy == 0))
+			update_shadow_used_ring_packed(vq, buf_id, 0,
+					desc_count);
+
+		rte_prefetch0((void *)(uintptr_t)buf_vec[0].buf_addr);
+
+		pkts[i] = rte_pktmbuf_alloc(mbuf_pool);
+		if (unlikely(pkts[i] == NULL)) {
+			RTE_LOG(ERR, VHOST_DATA,
+				"Failed to allocate memory for mbuf.\n");
+			break;
+		}
+
+		err = copy_desc_to_mbuf(dev, vq, buf_vec, nr_vec, pkts[i],
+				mbuf_pool);
+		if (unlikely(err)) {
+			rte_pktmbuf_free(pkts[i]);
+			break;
+		}
+
+		if (unlikely(dev->dequeue_zero_copy)) {
+			struct zcopy_mbuf *zmbuf;
+
+			zmbuf = get_zmbuf(vq);
+			if (!zmbuf) {
+				rte_pktmbuf_free(pkts[i]);
+				break;
+			}
+			zmbuf->mbuf = pkts[i];
+			zmbuf->desc_idx = buf_id;
+			zmbuf->desc_count = desc_count;
+
+			/*
+			 * Pin lock the mbuf; we will check later to see
+			 * whether the mbuf is freed (when we are the last
+			 * user) or not. If that's the case, we then could
+			 * update the used ring safely.
+			 */
+			rte_mbuf_refcnt_update(pkts[i], 1);
+
+			vq->nr_zmbuf += 1;
+			TAILQ_INSERT_TAIL(&vq->zmbuf_list, zmbuf, next);
+		}
+
+		vq->last_avail_idx += desc_count;
+		if (vq->last_avail_idx >= vq->size) {
+			vq->last_avail_idx -= vq->size;
+			vq->avail_wrap_counter ^= 1;
+		}
+	}
+
+	if (likely(dev->dequeue_zero_copy == 0)) {
+		do_data_copy_dequeue(vq);
+		if (unlikely(i < count))
+			vq->shadow_used_idx = i;
+		flush_shadow_used_ring_packed(dev, vq);
+		vhost_vring_call(dev, vq);
+	}
+
+	return i;
+}
+
 uint16_t
 rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
@@ -1515,7 +1631,10 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 		count -= 1;
 	}
 
-	count = virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count);
+	if (vq_is_packed(dev))
+		count = virtio_dev_tx_packed(dev, vq, mbuf_pool, pkts, count);
+	else
+		count = virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count);
 
 out:
 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
-- 
2.14.4

Re: [PATCH v6 13/15] vhost: add Tx support for packed ring

[Tiwei Bie]

On Mon, Jul 02, 2018 at 10:16:27AM +0200, Maxime Coquelin wrote:
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  lib/librte_vhost/vhost.h      |   1 +
>  lib/librte_vhost/virtio_net.c | 121 +++++++++++++++++++++++++++++++++++++++++-
>  2 files changed, 121 insertions(+), 1 deletion(-)
> 
> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
> index a7320469a..6ea8fb896 100644
> --- a/lib/librte_vhost/vhost.h
> +++ b/lib/librte_vhost/vhost.h
> @@ -56,6 +56,7 @@ struct buf_vector {
>  struct zcopy_mbuf {
>  	struct rte_mbuf *mbuf;
>  	uint32_t desc_idx;
> +	uint16_t desc_count;
>  	uint16_t in_use;
>  
>  	TAILQ_ENTRY(zcopy_mbuf) next;
> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
> index 2e286e228..03dd38235 100644
> --- a/lib/librte_vhost/virtio_net.c
> +++ b/lib/librte_vhost/virtio_net.c
> @@ -1445,6 +1445,122 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>  	return i;
>  }
>  
> +static __rte_always_inline uint16_t
> +virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
> +	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
> +{
> +	uint16_t i;
> +
> +	rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
> +
> +	if (unlikely(dev->dequeue_zero_copy)) {
> +		struct zcopy_mbuf *zmbuf, *next;
> +		int nr_updated = 0;
> +
> +		for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
> +		     zmbuf != NULL; zmbuf = next) {
> +			next = TAILQ_NEXT(zmbuf, next);
> +
> +			if (mbuf_is_consumed(zmbuf->mbuf)) {
> +				update_shadow_used_ring_packed(vq,
> +						zmbuf->desc_idx,
> +						0,
> +						zmbuf->desc_count);
> +				nr_updated += 1;

nr_updated isn't really used.

> +
> +				TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
> +				restore_mbuf(zmbuf->mbuf);
> +				rte_pktmbuf_free(zmbuf->mbuf);
> +				put_zmbuf(zmbuf);
> +				vq->nr_zmbuf -= 1;
> +			}
> +		}
> +
> +		flush_shadow_used_ring_packed(dev, vq);
> +		vhost_vring_call(dev, vq);
> +	}
> +
[...]

Re: [PATCH v6 13/15] vhost: add Tx support for packed ring

[Maxime Coquelin]

On 07/04/2018 07:45 AM, Tiwei Bie wrote:
> On Mon, Jul 02, 2018 at 10:16:27AM +0200, Maxime Coquelin wrote:
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>>   lib/librte_vhost/vhost.h      |   1 +
>>   lib/librte_vhost/virtio_net.c | 121 +++++++++++++++++++++++++++++++++++++++++-
>>   2 files changed, 121 insertions(+), 1 deletion(-)
>>
>> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
>> index a7320469a..6ea8fb896 100644
>> --- a/lib/librte_vhost/vhost.h
>> +++ b/lib/librte_vhost/vhost.h
>> @@ -56,6 +56,7 @@ struct buf_vector {
>>   struct zcopy_mbuf {
>>   	struct rte_mbuf *mbuf;
>>   	uint32_t desc_idx;
>> +	uint16_t desc_count;
>>   	uint16_t in_use;
>>   
>>   	TAILQ_ENTRY(zcopy_mbuf) next;
>> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
>> index 2e286e228..03dd38235 100644
>> --- a/lib/librte_vhost/virtio_net.c
>> +++ b/lib/librte_vhost/virtio_net.c
>> @@ -1445,6 +1445,122 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>>   	return i;
>>   }
>>   
>> +static __rte_always_inline uint16_t
>> +virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
>> +	struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
>> +{
>> +	uint16_t i;
>> +
>> +	rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
>> +
>> +	if (unlikely(dev->dequeue_zero_copy)) {
>> +		struct zcopy_mbuf *zmbuf, *next;
>> +		int nr_updated = 0;
>> +
>> +		for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
>> +		     zmbuf != NULL; zmbuf = next) {
>> +			next = TAILQ_NEXT(zmbuf, next);
>> +
>> +			if (mbuf_is_consumed(zmbuf->mbuf)) {
>> +				update_shadow_used_ring_packed(vq,
>> +						zmbuf->desc_idx,
>> +						0,
>> +						zmbuf->desc_count);
>> +				nr_updated += 1;
> 
> nr_updated isn't really used.

Right, it seems it should be removed in the _split version too, but I'll 
do that in a separate patch.

Thanks,
Maxime
>> +
>> +				TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
>> +				restore_mbuf(zmbuf->mbuf);
>> +				rte_pktmbuf_free(zmbuf->mbuf);
>> +				put_zmbuf(zmbuf);
>> +				vq->nr_zmbuf -= 1;
>> +			}
>> +		}
>> +
>> +		flush_shadow_used_ring_packed(dev, vq);
>> +		vhost_vring_call(dev, vq);
>> +	}
>> +
> [...]
> 

[PATCH v6 14/15] vhost: add notification for packed ring

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/vhost.c         | 73 ++++++++++++++++++++++++++++++++++++----
 lib/librte_vhost/vhost.h         | 71 ++++++++++++++++++++++++++++++++++----
 lib/librte_vhost/vhost_user.c    | 24 +++++++++++++
 lib/librte_vhost/virtio-packed.h | 11 ++++++
 lib/librte_vhost/virtio_net.c    | 12 +++----
 5 files changed, 172 insertions(+), 19 deletions(-)

diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
index 8538302c9..78f20c402 100644
--- a/lib/librte_vhost/vhost.c
+++ b/lib/librte_vhost/vhost.c
@@ -171,6 +171,24 @@ vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 	if (!vq->desc_packed || size != req_size)
 		return -1;
 
+	req_size = sizeof(struct vring_packed_desc_event);
+	size = req_size;
+	vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
+		vhost_iova_to_vva(dev,
+					vq,	vq->ring_addrs.avail_user_addr,
+					&size, VHOST_ACCESS_RW);
+	if (!vq->driver_event || size != req_size)
+		return -1;
+
+	req_size = sizeof(struct vring_packed_desc_event);
+	size = req_size;
+	vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
+		vhost_iova_to_vva(dev,
+					vq, vq->ring_addrs.used_user_addr,
+					&size, VHOST_ACCESS_RW);
+	if (!vq->device_event || size != req_size)
+		return -1;
+
 	return 0;
 }
 
@@ -595,7 +613,11 @@ rte_vhost_vring_call(int vid, uint16_t vring_idx)
 	if (!vq)
 		return -1;
 
-	vhost_vring_call(dev, vq);
+	if (vq_is_packed(dev))
+		vhost_vring_call_packed(dev, vq);
+	else
+		vhost_vring_call_split(dev, vq);
+
 	return 0;
 }
 
@@ -616,20 +638,59 @@ rte_vhost_avail_entries(int vid, uint16_t queue_id)
 	return *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
 }
 
+static inline int
+vhost_enable_notify_split(struct vhost_virtqueue *vq, int enable)
+{
+	if (enable)
+		vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
+	else
+		vq->used->flags |= VRING_USED_F_NO_NOTIFY;
+
+	return 0;
+}
+
+static inline int
+vhost_enable_notify_packed(struct virtio_net *dev,
+		struct vhost_virtqueue *vq, int enable)
+{
+	uint16_t flags;
+
+	if (!enable) {
+		vq->device_event->desc_event_flags = RING_EVENT_FLAGS_DISABLE;
+		return 0;
+	}
+
+	flags = RING_EVENT_FLAGS_ENABLE;
+	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
+		flags = RING_EVENT_FLAGS_DESC;
+		vq->device_event->desc_event_off_wrap = vq->last_avail_idx |
+			vq->avail_wrap_counter << 15;
+	}
+
+	rte_smp_wmb();
+
+	vq->device_event->desc_event_flags = flags;
+
+	rte_smp_wmb();
+
+	return 0;
+}
+
 int
 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
 {
 	struct virtio_net *dev = get_device(vid);
+	struct vhost_virtqueue *vq;
 
 	if (!dev)
 		return -1;
 
-	if (enable)
-		dev->virtqueue[queue_id]->used->flags &=
-			~VRING_USED_F_NO_NOTIFY;
+	vq = dev->virtqueue[queue_id];
+
+	if (vq_is_packed(dev))
+		return vhost_enable_notify_packed(dev, vq, enable);
 	else
-		dev->virtqueue[queue_id]->used->flags |= VRING_USED_F_NO_NOTIFY;
-	return 0;
+		return vhost_enable_notify_split(vq, enable);
 }
 
 void
diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index 6ea8fb896..728fd2f6b 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -21,6 +21,7 @@
 
 #include "rte_vhost.h"
 #include "rte_vdpa.h"
+#include "virtio-packed.h"
 
 /* Used to indicate that the device is running on a data core */
 #define VIRTIO_DEV_RUNNING 1
@@ -95,8 +96,14 @@ struct vhost_virtqueue {
 		struct vring_desc	*desc;
 		struct vring_desc_packed   *desc_packed;
 	};
-	struct vring_avail	*avail;
-	struct vring_used	*used;
+	union {
+		struct vring_avail	*avail;
+		struct vring_packed_desc_event *driver_event;
+	};
+	union {
+		struct vring_used	*used;
+		struct vring_packed_desc_event *device_event;
+	};
 	uint32_t		size;
 
 	uint16_t		last_avail_idx;
@@ -613,7 +620,7 @@ vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
 }
 
 static __rte_always_inline void
-vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
+vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
 	/* Flush used->idx update before we read avail->flags. */
 	rte_mb();
@@ -624,11 +631,11 @@ vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
 		uint16_t new = vq->last_used_idx;
 
 		VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
-			__func__,
-			vhost_used_event(vq),
-			old, new);
+				__func__,
+				vhost_used_event(vq),
+				old, new);
 		if (vhost_need_event(vhost_used_event(vq), new, old)
-			&& (vq->callfd >= 0)) {
+				&& (vq->callfd >= 0)) {
 			vq->signalled_used = vq->last_used_idx;
 			eventfd_write(vq->callfd, (eventfd_t) 1);
 		}
@@ -640,4 +647,54 @@ vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
 	}
 }
 
+static __rte_always_inline void
+vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+	uint16_t old, new, off, off_wrap, wrap;
+	bool kick = false;
+
+
+	/*  Flush used desc update. */
+	rte_smp_mb();
+
+	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
+		if (vq->driver_event->desc_event_flags !=
+				RING_EVENT_FLAGS_DISABLE)
+			kick = true;
+		goto kick;
+	}
+
+	old = vq->signalled_used;
+	new = vq->last_used_idx;
+	vq->signalled_used = new;
+
+	if (vq->driver_event->desc_event_flags != RING_EVENT_FLAGS_DESC) {
+		if (vq->driver_event->desc_event_flags !=
+				RING_EVENT_FLAGS_DISABLE)
+			kick = true;
+		goto kick;
+	}
+
+	rte_smp_rmb();
+
+	off_wrap = vq->driver_event->desc_event_off_wrap;
+	off = off_wrap & ~(1 << 15);
+	wrap = vq->used_wrap_counter;
+
+	if (new < old) {
+		new += vq->size;
+		wrap ^= 1;
+	}
+
+	if (wrap != off_wrap >> 15)
+		off += vq->size;
+
+	if (vhost_need_event(off, new, old))
+		kick = true;
+
+kick:
+	if (kick)
+		eventfd_write(vq->callfd, (eventfd_t)1);
+}
+
 #endif /* _VHOST_NET_CDEV_H_ */
diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
index b2b57de57..bda515bdb 100644
--- a/lib/librte_vhost/vhost_user.c
+++ b/lib/librte_vhost/vhost_user.c
@@ -523,6 +523,30 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index)
 		vq = dev->virtqueue[vq_index];
 		addr = &vq->ring_addrs;
 
+		len = sizeof(struct vring_packed_desc_event);
+		vq->driver_event = (struct vring_packed_desc_event *)
+					(uintptr_t)ring_addr_to_vva(dev,
+					vq, addr->avail_user_addr, &len);
+		if (vq->driver_event == 0 ||
+				len != sizeof(struct vring_packed_desc_event)) {
+			RTE_LOG(DEBUG, VHOST_CONFIG,
+				"(%d) failed to find driver area address.\n",
+				dev->vid);
+			return dev;
+		}
+
+		len = sizeof(struct vring_packed_desc_event);
+		vq->device_event = (struct vring_packed_desc_event *)
+					(uintptr_t)ring_addr_to_vva(dev,
+					vq, addr->used_user_addr, &len);
+		if (vq->device_event == 0 ||
+				len != sizeof(struct vring_packed_desc_event)) {
+			RTE_LOG(DEBUG, VHOST_CONFIG,
+				"(%d) failed to find device area address.\n",
+				dev->vid);
+			return dev;
+		}
+
 		return dev;
 	}
 
diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
index d386cb6df..ce3b28313 100644
--- a/lib/librte_vhost/virtio-packed.h
+++ b/lib/librte_vhost/virtio-packed.h
@@ -19,6 +19,17 @@ struct vring_desc_packed {
 	uint16_t flags;
 };
 
+#define RING_EVENT_FLAGS_ENABLE 0x0
+#define RING_EVENT_FLAGS_DISABLE 0x1
+#define RING_EVENT_FLAGS_DESC 0x2
+#define RING_EVENT_FLAGS_MASK 0xFFFC
+#define RING_EVENT_WRAP_MASK 0x8000
+#define RING_EVENT_OFF_MASK 0x7FFF
+
+struct vring_packed_desc_event {
+	uint16_t desc_event_off_wrap;
+	uint16_t desc_event_flags;
+};
 
 static inline bool
 desc_is_avail(struct vring_desc_packed *desc, bool wrap_counter)
diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 03dd38235..11c10aaf8 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -824,7 +824,7 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
 	if (likely(vq->shadow_used_idx)) {
 		flush_shadow_used_ring_split(dev, vq);
-		vhost_vring_call(dev, vq);
+		vhost_vring_call_split(dev, vq);
 	}
 
 	return pkt_idx;
@@ -877,7 +877,7 @@ virtio_dev_rx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
 	if (likely(vq->shadow_used_idx)) {
 		flush_shadow_used_ring_packed(dev, vq);
-		vhost_vring_call(dev, vq);
+		vhost_vring_call_packed(dev, vq);
 	}
 
 	return pkt_idx;
@@ -1360,7 +1360,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		}
 
 		flush_shadow_used_ring_split(dev, vq);
-		vhost_vring_call(dev, vq);
+		vhost_vring_call_split(dev, vq);
 	}
 
 	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
@@ -1439,7 +1439,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		if (unlikely(i < count))
 			vq->shadow_used_idx = i;
 		flush_shadow_used_ring_split(dev, vq);
-		vhost_vring_call(dev, vq);
+		vhost_vring_call_split(dev, vq);
 	}
 
 	return i;
@@ -1477,7 +1477,7 @@ virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		}
 
 		flush_shadow_used_ring_packed(dev, vq);
-		vhost_vring_call(dev, vq);
+		vhost_vring_call_packed(dev, vq);
 	}
 
 	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
@@ -1555,7 +1555,7 @@ virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		if (unlikely(i < count))
 			vq->shadow_used_idx = i;
 		flush_shadow_used_ring_packed(dev, vq);
-		vhost_vring_call(dev, vq);
+		vhost_vring_call_packed(dev, vq);
 	}
 
 	return i;
-- 
2.14.4

Re: [PATCH v6 14/15] vhost: add notification for packed ring

[Jason Wang]

On 2018年07月02日 16:16, Maxime Coquelin wrote:
> +static inline int
> +vhost_enable_notify_split(struct vhost_virtqueue *vq, int enable)
> +{
> +	if (enable)
> +		vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
> +	else
> +		vq->used->flags |= VRING_USED_F_NO_NOTIFY;
> +
> +	return 0;
> +}
> +
> +static inline int
> +vhost_enable_notify_packed(struct virtio_net *dev,
> +		struct vhost_virtqueue *vq, int enable)
> +{
> +	uint16_t flags;
> +
> +	if (!enable) {
> +		vq->device_event->desc_event_flags = RING_EVENT_FLAGS_DISABLE;
> +		return 0;
> +	}
> +
> +	flags = RING_EVENT_FLAGS_ENABLE;
> +	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
> +		flags = RING_EVENT_FLAGS_DESC;
> +		vq->device_event->desc_event_off_wrap = vq->last_avail_idx |
> +			vq->avail_wrap_counter << 15;
> +	}
> +
> +	rte_smp_wmb();
> +
> +	vq->device_event->desc_event_flags = flags;
> +
> +	rte_smp_wmb();
> +

We don't do this for split version. Any specific reason for this?

Btw, looks like we don't care about the return value so using void instead?

Thanks

> +	return 0;
> +}
> +

Re: [PATCH v6 14/15] vhost: add notification for packed ring

[Maxime Coquelin]

On 07/03/2018 07:57 AM, Jason Wang wrote:
> 
> 
> On 2018年07月02日 16:16, Maxime Coquelin wrote:
>> +static inline int
>> +vhost_enable_notify_split(struct vhost_virtqueue *vq, int enable)
>> +{
>> +    if (enable)
>> +        vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
>> +    else
>> +        vq->used->flags |= VRING_USED_F_NO_NOTIFY;
>> +
>> +    return 0;
>> +}
>> +
>> +static inline int
>> +vhost_enable_notify_packed(struct virtio_net *dev,
>> +        struct vhost_virtqueue *vq, int enable)
>> +{
>> +    uint16_t flags;
>> +
>> +    if (!enable) {
>> +        vq->device_event->desc_event_flags = RING_EVENT_FLAGS_DISABLE;
>> +        return 0;
>> +    }
>> +
>> +    flags = RING_EVENT_FLAGS_ENABLE;
>> +    if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
>> +        flags = RING_EVENT_FLAGS_DESC;
>> +        vq->device_event->desc_event_off_wrap = vq->last_avail_idx |
>> +            vq->avail_wrap_counter << 15;
>> +    }
>> +
>> +    rte_smp_wmb();
>> +
>> +    vq->device_event->desc_event_flags = flags;
>> +
>> +    rte_smp_wmb();
>> +
> 
> We don't do this for split version. Any specific reason for this?

Yeah, maybe this is no necessary.

> Btw, looks like we don't care about the return value so using void instead?

Right, it could be a void.

Thanks!
Maxime

> Thanks
> 
>> +    return 0;
>> +}
>> +
> 
> 

Re: [PATCH v6 14/15] vhost: add notification for packed ring

[Jason Wang]

On 2018年07月02日 16:16, Maxime Coquelin wrote:
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>   lib/librte_vhost/vhost.c         | 73 ++++++++++++++++++++++++++++++++++++----
>   lib/librte_vhost/vhost.h         | 71 ++++++++++++++++++++++++++++++++++----
>   lib/librte_vhost/vhost_user.c    | 24 +++++++++++++
>   lib/librte_vhost/virtio-packed.h | 11 ++++++
>   lib/librte_vhost/virtio_net.c    | 12 +++----
>   5 files changed, 172 insertions(+), 19 deletions(-)
>
> diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
> index 8538302c9..78f20c402 100644
> --- a/lib/librte_vhost/vhost.c
> +++ b/lib/librte_vhost/vhost.c
> @@ -171,6 +171,24 @@ vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
>   	if (!vq->desc_packed || size != req_size)
>   		return -1;
>   
> +	req_size = sizeof(struct vring_packed_desc_event);
> +	size = req_size;
> +	vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
> +		vhost_iova_to_vva(dev,
> +					vq,	vq->ring_addrs.avail_user_addr,
> +					&size, VHOST_ACCESS_RW);
> +	if (!vq->driver_event || size != req_size)
> +		return -1;
> +
> +	req_size = sizeof(struct vring_packed_desc_event);
> +	size = req_size;
> +	vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
> +		vhost_iova_to_vva(dev,
> +					vq, vq->ring_addrs.used_user_addr,
> +					&size, VHOST_ACCESS_RW);
> +	if (!vq->device_event || size != req_size)
> +		return -1;
> +
>   	return 0;
>   }
>   
> @@ -595,7 +613,11 @@ rte_vhost_vring_call(int vid, uint16_t vring_idx)
>   	if (!vq)
>   		return -1;
>   
> -	vhost_vring_call(dev, vq);
> +	if (vq_is_packed(dev))
> +		vhost_vring_call_packed(dev, vq);
> +	else
> +		vhost_vring_call_split(dev, vq);
> +
>   	return 0;
>   }
>   
> @@ -616,20 +638,59 @@ rte_vhost_avail_entries(int vid, uint16_t queue_id)
>   	return *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
>   }
>   
> +static inline int
> +vhost_enable_notify_split(struct vhost_virtqueue *vq, int enable)
> +{
> +	if (enable)
> +		vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
> +	else
> +		vq->used->flags |= VRING_USED_F_NO_NOTIFY;
> +
> +	return 0;
> +}
> +
> +static inline int
> +vhost_enable_notify_packed(struct virtio_net *dev,
> +		struct vhost_virtqueue *vq, int enable)
> +{
> +	uint16_t flags;
> +
> +	if (!enable) {
> +		vq->device_event->desc_event_flags = RING_EVENT_FLAGS_DISABLE;
> +		return 0;
> +	}
> +
> +	flags = RING_EVENT_FLAGS_ENABLE;
> +	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
> +		flags = RING_EVENT_FLAGS_DESC;
> +		vq->device_event->desc_event_off_wrap = vq->last_avail_idx |
> +			vq->avail_wrap_counter << 15;
> +	}
> +
> +	rte_smp_wmb();
> +
> +	vq->device_event->desc_event_flags = flags;
> +
> +	rte_smp_wmb();
> +
> +	return 0;
> +}
> +
>   int
>   rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
>   {
>   	struct virtio_net *dev = get_device(vid);
> +	struct vhost_virtqueue *vq;
>   
>   	if (!dev)
>   		return -1;
>   
> -	if (enable)
> -		dev->virtqueue[queue_id]->used->flags &=
> -			~VRING_USED_F_NO_NOTIFY;
> +	vq = dev->virtqueue[queue_id];
> +
> +	if (vq_is_packed(dev))
> +		return vhost_enable_notify_packed(dev, vq, enable);
>   	else
> -		dev->virtqueue[queue_id]->used->flags |= VRING_USED_F_NO_NOTIFY;
> -	return 0;
> +		return vhost_enable_notify_split(vq, enable);
>   }
>   
>   void
> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
> index 6ea8fb896..728fd2f6b 100644
> --- a/lib/librte_vhost/vhost.h
> +++ b/lib/librte_vhost/vhost.h
> @@ -21,6 +21,7 @@
>   
>   #include "rte_vhost.h"
>   #include "rte_vdpa.h"
> +#include "virtio-packed.h"
>   
>   /* Used to indicate that the device is running on a data core */
>   #define VIRTIO_DEV_RUNNING 1
> @@ -95,8 +96,14 @@ struct vhost_virtqueue {
>   		struct vring_desc	*desc;
>   		struct vring_desc_packed   *desc_packed;
>   	};
> -	struct vring_avail	*avail;
> -	struct vring_used	*used;
> +	union {
> +		struct vring_avail	*avail;
> +		struct vring_packed_desc_event *driver_event;
> +	};
> +	union {
> +		struct vring_used	*used;
> +		struct vring_packed_desc_event *device_event;
> +	};
>   	uint32_t		size;
>   
>   	uint16_t		last_avail_idx;
> @@ -613,7 +620,7 @@ vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
>   }
>   
>   static __rte_always_inline void
> -vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
> +vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
>   {
>   	/* Flush used->idx update before we read avail->flags. */
>   	rte_mb();
> @@ -624,11 +631,11 @@ vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
>   		uint16_t new = vq->last_used_idx;
>   
>   		VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
> -			__func__,
> -			vhost_used_event(vq),
> -			old, new);
> +				__func__,
> +				vhost_used_event(vq),
> +				old, new);
>   		if (vhost_need_event(vhost_used_event(vq), new, old)
> -			&& (vq->callfd >= 0)) {
> +				&& (vq->callfd >= 0)) {
>   			vq->signalled_used = vq->last_used_idx;
>   			eventfd_write(vq->callfd, (eventfd_t) 1);
>   		}

Looks like a style fixes, move those to another patch?

> @@ -640,4 +647,54 @@ vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
>   	}
>   }
>   
> +static __rte_always_inline void
> +vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
> +{
> +	uint16_t old, new, off, off_wrap, wrap;
> +	bool kick = false;
> +
> +
> +	/*  Flush used desc update. */
> +	rte_smp_mb();
> +
> +	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
> +		if (vq->driver_event->desc_event_flags !=
> +				RING_EVENT_FLAGS_DISABLE)
> +			kick = true;
> +		goto kick;
> +	}
> +
> +	old = vq->signalled_used;
> +	new = vq->last_used_idx;
> +	vq->signalled_used = new;
> +
> +	if (vq->driver_event->desc_event_flags != RING_EVENT_FLAGS_DESC) {
> +		if (vq->driver_event->desc_event_flags !=
> +				RING_EVENT_FLAGS_DISABLE)
> +			kick = true;
> +		goto kick;
> +	}
> +
> +	rte_smp_rmb();
> +
> +	off_wrap = vq->driver_event->desc_event_off_wrap;
> +	off = off_wrap & ~(1 << 15);
> +	wrap = vq->used_wrap_counter;
> +
> +	if (new < old) {
> +		new += vq->size;
> +		wrap ^= 1;
> +	}
> +
> +	if (wrap != off_wrap >> 15)
> +		off += vq->size;
> +

Jusy FYI. Maybe we can switch to a more compact version like Tiwei used:

...
         wrap_counter = off_wrap >> 15;
         event_idx = off_wrap & ~(1<<15);
         if (wrap_counter != vq->avail_wrap_counter)
                 event_idx -= vq->vring_packed.num;

         if (flags == VRING_EVENT_F_DESC)
                 needs_kick = vring_need_event(event_idx, new, old);
         else
                 needs_kick = (flags != VRING_EVENT_F_DISABLE);

(I've switched to this version in vhost).

Thanks

> +	if (vhost_need_event(off, new, old))
> +		kick = true;
> +
> +kick:
> +	if (kick)
> +		eventfd_write(vq->callfd, (eventfd_t)1);
> +}
> +
>   #endif /* _VHOST_NET_CDEV_H_ */
> diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
> index b2b57de57..bda515bdb 100644
> --- a/lib/librte_vhost/vhost_user.c
> +++ b/lib/librte_vhost/vhost_user.c
> @@ -523,6 +523,30 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index)
>   		vq = dev->virtqueue[vq_index];
>   		addr = &vq->ring_addrs;
>   
> +		len = sizeof(struct vring_packed_desc_event);
> +		vq->driver_event = (struct vring_packed_desc_event *)
> +					(uintptr_t)ring_addr_to_vva(dev,
> +					vq, addr->avail_user_addr, &len);
> +		if (vq->driver_event == 0 ||
> +				len != sizeof(struct vring_packed_desc_event)) {
> +			RTE_LOG(DEBUG, VHOST_CONFIG,
> +				"(%d) failed to find driver area address.\n",
> +				dev->vid);
> +			return dev;
> +		}
> +
> +		len = sizeof(struct vring_packed_desc_event);
> +		vq->device_event = (struct vring_packed_desc_event *)
> +					(uintptr_t)ring_addr_to_vva(dev,
> +					vq, addr->used_user_addr, &len);
> +		if (vq->device_event == 0 ||
> +				len != sizeof(struct vring_packed_desc_event)) {
> +			RTE_LOG(DEBUG, VHOST_CONFIG,
> +				"(%d) failed to find device area address.\n",
> +				dev->vid);
> +			return dev;
> +		}
> +
>   		return dev;
>   	}
>   
> diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
> index d386cb6df..ce3b28313 100644
> --- a/lib/librte_vhost/virtio-packed.h
> +++ b/lib/librte_vhost/virtio-packed.h
> @@ -19,6 +19,17 @@ struct vring_desc_packed {
>   	uint16_t flags;
>   };
>   
> +#define RING_EVENT_FLAGS_ENABLE 0x0
> +#define RING_EVENT_FLAGS_DISABLE 0x1
> +#define RING_EVENT_FLAGS_DESC 0x2
> +#define RING_EVENT_FLAGS_MASK 0xFFFC
> +#define RING_EVENT_WRAP_MASK 0x8000
> +#define RING_EVENT_OFF_MASK 0x7FFF
> +
> +struct vring_packed_desc_event {
> +	uint16_t desc_event_off_wrap;
> +	uint16_t desc_event_flags;
> +};
>   
>   static inline bool
>   desc_is_avail(struct vring_desc_packed *desc, bool wrap_counter)
> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
> index 03dd38235..11c10aaf8 100644
> --- a/lib/librte_vhost/virtio_net.c
> +++ b/lib/librte_vhost/virtio_net.c
> @@ -824,7 +824,7 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>   
>   	if (likely(vq->shadow_used_idx)) {
>   		flush_shadow_used_ring_split(dev, vq);
> -		vhost_vring_call(dev, vq);
> +		vhost_vring_call_split(dev, vq);
>   	}
>   
>   	return pkt_idx;
> @@ -877,7 +877,7 @@ virtio_dev_rx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
>   
>   	if (likely(vq->shadow_used_idx)) {
>   		flush_shadow_used_ring_packed(dev, vq);
> -		vhost_vring_call(dev, vq);
> +		vhost_vring_call_packed(dev, vq);
>   	}
>   
>   	return pkt_idx;
> @@ -1360,7 +1360,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>   		}
>   
>   		flush_shadow_used_ring_split(dev, vq);
> -		vhost_vring_call(dev, vq);
> +		vhost_vring_call_split(dev, vq);
>   	}
>   
>   	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
> @@ -1439,7 +1439,7 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>   		if (unlikely(i < count))
>   			vq->shadow_used_idx = i;
>   		flush_shadow_used_ring_split(dev, vq);
> -		vhost_vring_call(dev, vq);
> +		vhost_vring_call_split(dev, vq);
>   	}
>   
>   	return i;
> @@ -1477,7 +1477,7 @@ virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
>   		}
>   
>   		flush_shadow_used_ring_packed(dev, vq);
> -		vhost_vring_call(dev, vq);
> +		vhost_vring_call_packed(dev, vq);
>   	}
>   
>   	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
> @@ -1555,7 +1555,7 @@ virtio_dev_tx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
>   		if (unlikely(i < count))
>   			vq->shadow_used_idx = i;
>   		flush_shadow_used_ring_packed(dev, vq);
> -		vhost_vring_call(dev, vq);
> +		vhost_vring_call_packed(dev, vq);
>   	}
>   
>   	return i;

Re: [PATCH v6 14/15] vhost: add notification for packed ring

[Maxime Coquelin]

On 07/03/2018 08:06 AM, Jason Wang wrote:
> 
> 
> On 2018年07月02日 16:16, Maxime Coquelin wrote:
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>>   lib/librte_vhost/vhost.c         | 73 
>> ++++++++++++++++++++++++++++++++++++----
>>   lib/librte_vhost/vhost.h         | 71 
>> ++++++++++++++++++++++++++++++++++----
>>   lib/librte_vhost/vhost_user.c    | 24 +++++++++++++
>>   lib/librte_vhost/virtio-packed.h | 11 ++++++
>>   lib/librte_vhost/virtio_net.c    | 12 +++----
>>   5 files changed, 172 insertions(+), 19 deletions(-)
>>
>> diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
>> index 8538302c9..78f20c402 100644
>> --- a/lib/librte_vhost/vhost.c
>> +++ b/lib/librte_vhost/vhost.c
>> @@ -171,6 +171,24 @@ vring_translate_packed(struct virtio_net *dev, 
>> struct vhost_virtqueue *vq)
>>       if (!vq->desc_packed || size != req_size)
>>           return -1;
>> +    req_size = sizeof(struct vring_packed_desc_event);
>> +    size = req_size;
>> +    vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
>> +        vhost_iova_to_vva(dev,
>> +                    vq,    vq->ring_addrs.avail_user_addr,
>> +                    &size, VHOST_ACCESS_RW);
>> +    if (!vq->driver_event || size != req_size)
>> +        return -1;
>> +
>> +    req_size = sizeof(struct vring_packed_desc_event);
>> +    size = req_size;
>> +    vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
>> +        vhost_iova_to_vva(dev,
>> +                    vq, vq->ring_addrs.used_user_addr,
>> +                    &size, VHOST_ACCESS_RW);
>> +    if (!vq->device_event || size != req_size)
>> +        return -1;
>> +
>>       return 0;
>>   }
>> @@ -595,7 +613,11 @@ rte_vhost_vring_call(int vid, uint16_t vring_idx)
>>       if (!vq)
>>           return -1;
>> -    vhost_vring_call(dev, vq);
>> +    if (vq_is_packed(dev))
>> +        vhost_vring_call_packed(dev, vq);
>> +    else
>> +        vhost_vring_call_split(dev, vq);
>> +
>>       return 0;
>>   }
>> @@ -616,20 +638,59 @@ rte_vhost_avail_entries(int vid, uint16_t queue_id)
>>       return *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
>>   }
>> +static inline int
>> +vhost_enable_notify_split(struct vhost_virtqueue *vq, int enable)
>> +{
>> +    if (enable)
>> +        vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
>> +    else
>> +        vq->used->flags |= VRING_USED_F_NO_NOTIFY;
>> +
>> +    return 0;
>> +}
>> +
>> +static inline int
>> +vhost_enable_notify_packed(struct virtio_net *dev,
>> +        struct vhost_virtqueue *vq, int enable)
>> +{
>> +    uint16_t flags;
>> +
>> +    if (!enable) {
>> +        vq->device_event->desc_event_flags = RING_EVENT_FLAGS_DISABLE;
>> +        return 0;
>> +    }
>> +
>> +    flags = RING_EVENT_FLAGS_ENABLE;
>> +    if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
>> +        flags = RING_EVENT_FLAGS_DESC;
>> +        vq->device_event->desc_event_off_wrap = vq->last_avail_idx |
>> +            vq->avail_wrap_counter << 15;
>> +    }
>> +
>> +    rte_smp_wmb();
>> +
>> +    vq->device_event->desc_event_flags = flags;
>> +
>> +    rte_smp_wmb();
>> +
>> +    return 0;
>> +}
>> +
>>   int
>>   rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int 
>> enable)
>>   {
>>       struct virtio_net *dev = get_device(vid);
>> +    struct vhost_virtqueue *vq;
>>       if (!dev)
>>           return -1;
>> -    if (enable)
>> -        dev->virtqueue[queue_id]->used->flags &=
>> -            ~VRING_USED_F_NO_NOTIFY;
>> +    vq = dev->virtqueue[queue_id];
>> +
>> +    if (vq_is_packed(dev))
>> +        return vhost_enable_notify_packed(dev, vq, enable);
>>       else
>> -        dev->virtqueue[queue_id]->used->flags |= VRING_USED_F_NO_NOTIFY;
>> -    return 0;
>> +        return vhost_enable_notify_split(vq, enable);
>>   }
>>   void
>> diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
>> index 6ea8fb896..728fd2f6b 100644
>> --- a/lib/librte_vhost/vhost.h
>> +++ b/lib/librte_vhost/vhost.h
>> @@ -21,6 +21,7 @@
>>   #include "rte_vhost.h"
>>   #include "rte_vdpa.h"
>> +#include "virtio-packed.h"
>>   /* Used to indicate that the device is running on a data core */
>>   #define VIRTIO_DEV_RUNNING 1
>> @@ -95,8 +96,14 @@ struct vhost_virtqueue {
>>           struct vring_desc    *desc;
>>           struct vring_desc_packed   *desc_packed;
>>       };
>> -    struct vring_avail    *avail;
>> -    struct vring_used    *used;
>> +    union {
>> +        struct vring_avail    *avail;
>> +        struct vring_packed_desc_event *driver_event;
>> +    };
>> +    union {
>> +        struct vring_used    *used;
>> +        struct vring_packed_desc_event *device_event;
>> +    };
>>       uint32_t        size;
>>       uint16_t        last_avail_idx;
>> @@ -613,7 +620,7 @@ vhost_need_event(uint16_t event_idx, uint16_t 
>> new_idx, uint16_t old)
>>   }
>>   static __rte_always_inline void
>> -vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
>> +vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue 
>> *vq)
>>   {
>>       /* Flush used->idx update before we read avail->flags. */
>>       rte_mb();
>> @@ -624,11 +631,11 @@ vhost_vring_call(struct virtio_net *dev, struct 
>> vhost_virtqueue *vq)
>>           uint16_t new = vq->last_used_idx;
>>           VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, 
>> new=%d\n",
>> -            __func__,
>> -            vhost_used_event(vq),
>> -            old, new);
>> +                __func__,
>> +                vhost_used_event(vq),
>> +                old, new);
>>           if (vhost_need_event(vhost_used_event(vq), new, old)
>> -            && (vq->callfd >= 0)) {
>> +                && (vq->callfd >= 0)) {
>>               vq->signalled_used = vq->last_used_idx;
>>               eventfd_write(vq->callfd, (eventfd_t) 1);
>>           }
> 
> Looks like a style fixes, move those to another patch?

Indeed, I'll remove this style fixes in next version.

>> @@ -640,4 +647,54 @@ vhost_vring_call(struct virtio_net *dev, struct 
>> vhost_virtqueue *vq)
>>       }
>>   }
>> +static __rte_always_inline void
>> +vhost_vring_call_packed(struct virtio_net *dev, struct 
>> vhost_virtqueue *vq)
>> +{
>> +    uint16_t old, new, off, off_wrap, wrap;
>> +    bool kick = false;
>> +
>> +
>> +    /*  Flush used desc update. */
>> +    rte_smp_mb();
>> +
>> +    if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
>> +        if (vq->driver_event->desc_event_flags !=
>> +                RING_EVENT_FLAGS_DISABLE)
>> +            kick = true;
>> +        goto kick;
>> +    }
>> +
>> +    old = vq->signalled_used;
>> +    new = vq->last_used_idx;
>> +    vq->signalled_used = new;
>> +
>> +    if (vq->driver_event->desc_event_flags != RING_EVENT_FLAGS_DESC) {
>> +        if (vq->driver_event->desc_event_flags !=
>> +                RING_EVENT_FLAGS_DISABLE)
>> +            kick = true;
>> +        goto kick;
>> +    }
>> +
>> +    rte_smp_rmb();
>> +
>> +    off_wrap = vq->driver_event->desc_event_off_wrap;
>> +    off = off_wrap & ~(1 << 15);
>> +    wrap = vq->used_wrap_counter;
>> +
>> +    if (new < old) {
>> +        new += vq->size;
>> +        wrap ^= 1;
>> +    }
>> +
>> +    if (wrap != off_wrap >> 15)
>> +        off += vq->size;
>> +
> 
> Jusy FYI. Maybe we can switch to a more compact version like Tiwei used:
> 
> ...
>          wrap_counter = off_wrap >> 15;
>          event_idx = off_wrap & ~(1<<15);
>          if (wrap_counter != vq->avail_wrap_counter)
>                  event_idx -= vq->vring_packed.num;
> 
>          if (flags == VRING_EVENT_F_DESC)
>                  needs_kick = vring_need_event(event_idx, new, old);
>          else
>                  needs_kick = (flags != VRING_EVENT_F_DISABLE);
> 
> (I've switched to this version in vhost).

Sure, it is indeed simpler.

Thanks,
Maxime

> Thanks
> 
>> +    if (vhost_need_event(off, new, old))
>> +        kick = true;
>> +
>> +kick:
>> +    if (kick)
>> +        eventfd_write(vq->callfd, (eventfd_t)1);
>> +}
>> +
>>   #endif /* _VHOST_NET_CDEV_H_ */
>> diff --git a/lib/librte_vhost/vhost_user.c 
>> b/lib/librte_vhost/vhost_user.c
>> index b2b57de57..bda515bdb 100644
>> --- a/lib/librte_vhost/vhost_user.c
>> +++ b/lib/librte_vhost/vhost_user.c
>> @@ -523,6 +523,30 @@ translate_ring_addresses(struct virtio_net *dev, 
>> int vq_index)
>>           vq = dev->virtqueue[vq_index];
>>           addr = &vq->ring_addrs;
>> +        len = sizeof(struct vring_packed_desc_event);
>> +        vq->driver_event = (struct vring_packed_desc_event *)
>> +                    (uintptr_t)ring_addr_to_vva(dev,
>> +                    vq, addr->avail_user_addr, &len);
>> +        if (vq->driver_event == 0 ||
>> +                len != sizeof(struct vring_packed_desc_event)) {
>> +            RTE_LOG(DEBUG, VHOST_CONFIG,
>> +                "(%d) failed to find driver area address.\n",
>> +                dev->vid);
>> +            return dev;
>> +        }
>> +
>> +        len = sizeof(struct vring_packed_desc_event);
>> +        vq->device_event = (struct vring_packed_desc_event *)
>> +                    (uintptr_t)ring_addr_to_vva(dev,
>> +                    vq, addr->used_user_addr, &len);
>> +        if (vq->device_event == 0 ||
>> +                len != sizeof(struct vring_packed_desc_event)) {
>> +            RTE_LOG(DEBUG, VHOST_CONFIG,
>> +                "(%d) failed to find device area address.\n",
>> +                dev->vid);
>> +            return dev;
>> +        }
>> +
>>           return dev;
>>       }
>> diff --git a/lib/librte_vhost/virtio-packed.h 
>> b/lib/librte_vhost/virtio-packed.h
>> index d386cb6df..ce3b28313 100644
>> --- a/lib/librte_vhost/virtio-packed.h
>> +++ b/lib/librte_vhost/virtio-packed.h
>> @@ -19,6 +19,17 @@ struct vring_desc_packed {
>>       uint16_t flags;
>>   };
>> +#define RING_EVENT_FLAGS_ENABLE 0x0
>> +#define RING_EVENT_FLAGS_DISABLE 0x1
>> +#define RING_EVENT_FLAGS_DESC 0x2
>> +#define RING_EVENT_FLAGS_MASK 0xFFFC
>> +#define RING_EVENT_WRAP_MASK 0x8000
>> +#define RING_EVENT_OFF_MASK 0x7FFF
>> +
>> +struct vring_packed_desc_event {
>> +    uint16_t desc_event_off_wrap;
>> +    uint16_t desc_event_flags;
>> +};
>>   static inline bool
>>   desc_is_avail(struct vring_desc_packed *desc, bool wrap_counter)
>> diff --git a/lib/librte_vhost/virtio_net.c 
>> b/lib/librte_vhost/virtio_net.c
>> index 03dd38235..11c10aaf8 100644
>> --- a/lib/librte_vhost/virtio_net.c
>> +++ b/lib/librte_vhost/virtio_net.c
>> @@ -824,7 +824,7 @@ virtio_dev_rx_split(struct virtio_net *dev, struct 
>> vhost_virtqueue *vq,
>>       if (likely(vq->shadow_used_idx)) {
>>           flush_shadow_used_ring_split(dev, vq);
>> -        vhost_vring_call(dev, vq);
>> +        vhost_vring_call_split(dev, vq);
>>       }
>>       return pkt_idx;
>> @@ -877,7 +877,7 @@ virtio_dev_rx_packed(struct virtio_net *dev, 
>> struct vhost_virtqueue *vq,
>>       if (likely(vq->shadow_used_idx)) {
>>           flush_shadow_used_ring_packed(dev, vq);
>> -        vhost_vring_call(dev, vq);
>> +        vhost_vring_call_packed(dev, vq);
>>       }
>>       return pkt_idx;
>> @@ -1360,7 +1360,7 @@ virtio_dev_tx_split(struct virtio_net *dev, 
>> struct vhost_virtqueue *vq,
>>           }
>>           flush_shadow_used_ring_split(dev, vq);
>> -        vhost_vring_call(dev, vq);
>> +        vhost_vring_call_split(dev, vq);
>>       }
>>       rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 
>> 1)]);
>> @@ -1439,7 +1439,7 @@ virtio_dev_tx_split(struct virtio_net *dev, 
>> struct vhost_virtqueue *vq,
>>           if (unlikely(i < count))
>>               vq->shadow_used_idx = i;
>>           flush_shadow_used_ring_split(dev, vq);
>> -        vhost_vring_call(dev, vq);
>> +        vhost_vring_call_split(dev, vq);
>>       }
>>       return i;
>> @@ -1477,7 +1477,7 @@ virtio_dev_tx_packed(struct virtio_net *dev, 
>> struct vhost_virtqueue *vq,
>>           }
>>           flush_shadow_used_ring_packed(dev, vq);
>> -        vhost_vring_call(dev, vq);
>> +        vhost_vring_call_packed(dev, vq);
>>       }
>>       VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
>> @@ -1555,7 +1555,7 @@ virtio_dev_tx_packed(struct virtio_net *dev, 
>> struct vhost_virtqueue *vq,
>>           if (unlikely(i < count))
>>               vq->shadow_used_idx = i;
>>           flush_shadow_used_ring_packed(dev, vq);
>> -        vhost_vring_call(dev, vq);
>> +        vhost_vring_call_packed(dev, vq);
>>       }
>>       return i;
> 

Re: [PATCH v6 14/15] vhost: add notification for packed ring

[Tiwei Bie]

On Mon, Jul 02, 2018 at 10:16:28AM +0200, Maxime Coquelin wrote:
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  lib/librte_vhost/vhost.c         | 73 ++++++++++++++++++++++++++++++++++++----
>  lib/librte_vhost/vhost.h         | 71 ++++++++++++++++++++++++++++++++++----
>  lib/librte_vhost/vhost_user.c    | 24 +++++++++++++
>  lib/librte_vhost/virtio-packed.h | 11 ++++++
>  lib/librte_vhost/virtio_net.c    | 12 +++----
>  5 files changed, 172 insertions(+), 19 deletions(-)
> 
> diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
> index 8538302c9..78f20c402 100644
> --- a/lib/librte_vhost/vhost.c
> +++ b/lib/librte_vhost/vhost.c
> @@ -171,6 +171,24 @@ vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
>  	if (!vq->desc_packed || size != req_size)
>  		return -1;
>  
> +	req_size = sizeof(struct vring_packed_desc_event);
> +	size = req_size;
> +	vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
> +		vhost_iova_to_vva(dev,
> +					vq,	vq->ring_addrs.avail_user_addr,
> +					&size, VHOST_ACCESS_RW);

It should be a space instead of a tab after "vq,"

Why not put "vq, vq->ring_addrs.avail_user_addr,"
and "vhost_iova_to_vva(dev," on the same line?

> +	if (!vq->driver_event || size != req_size)
> +		return -1;
> +
> +	req_size = sizeof(struct vring_packed_desc_event);
> +	size = req_size;
> +	vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
> +		vhost_iova_to_vva(dev,
> +					vq, vq->ring_addrs.used_user_addr,

It's better to put "vhost_iova_to_vva(dev," and
"vq, vq->ring_addrs.used_user_addr," on the same line.

Currently, it looks like something as this in my editor:

¦       vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
¦       ¦       vhost_iova_to_vva(dev,
¦       ¦       ¦       ¦       ¦       vq, vq->ring_addrs.used_user_addr,
¦       ¦       ¦       ¦       ¦       &size, VHOST_ACCESS_RW);

> +					&size, VHOST_ACCESS_RW);
> +	if (!vq->device_event || size != req_size)
> +		return -1;
> +
>  	return 0;
>  }
>  
[...]
> @@ -640,4 +647,54 @@ vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
>  	}
>  }
>  
> +static __rte_always_inline void
> +vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
> +{
> +	uint16_t old, new, off, off_wrap, wrap;
> +	bool kick = false;
> +
> +

There is no need to have two blank lines.

> +	/*  Flush used desc update. */

Just need one space between "/*" and "Flush".

> +	rte_smp_mb();
> +
> +	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
> +		if (vq->driver_event->desc_event_flags !=
> +				RING_EVENT_FLAGS_DISABLE)
> +			kick = true;
> +		goto kick;
> +	}
> +
> +	old = vq->signalled_used;
> +	new = vq->last_used_idx;
> +	vq->signalled_used = new;
> +
> +	if (vq->driver_event->desc_event_flags != RING_EVENT_FLAGS_DESC) {
> +		if (vq->driver_event->desc_event_flags !=
> +				RING_EVENT_FLAGS_DISABLE)
> +			kick = true;
> +		goto kick;
> +	}
> +
> +	rte_smp_rmb();
> +
> +	off_wrap = vq->driver_event->desc_event_off_wrap;
> +	off = off_wrap & ~(1 << 15);

Maybe it's better to use: RING_EVENT_OFF_MASK

> +	wrap = vq->used_wrap_counter;
> +
> +	if (new < old) {
> +		new += vq->size;
> +		wrap ^= 1;
> +	}
> +
> +	if (wrap != off_wrap >> 15)
> +		off += vq->size;
> +
> +	if (vhost_need_event(off, new, old))
> +		kick = true;
> +
> +kick:
> +	if (kick)
> +		eventfd_write(vq->callfd, (eventfd_t)1);
> +}
> +
>  #endif /* _VHOST_NET_CDEV_H_ */
> diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
> index b2b57de57..bda515bdb 100644
> --- a/lib/librte_vhost/vhost_user.c
> +++ b/lib/librte_vhost/vhost_user.c
> @@ -523,6 +523,30 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index)
>  		vq = dev->virtqueue[vq_index];
>  		addr = &vq->ring_addrs;
>  
> +		len = sizeof(struct vring_packed_desc_event);
> +		vq->driver_event = (struct vring_packed_desc_event *)
> +					(uintptr_t)ring_addr_to_vva(dev,
> +					vq, addr->avail_user_addr, &len);
> +		if (vq->driver_event == 0 ||

It's better to compare with NULL.

> +				len != sizeof(struct vring_packed_desc_event)) {
> +			RTE_LOG(DEBUG, VHOST_CONFIG,
> +				"(%d) failed to find driver area address.\n",
> +				dev->vid);
> +			return dev;
> +		}
> +
> +		len = sizeof(struct vring_packed_desc_event);
> +		vq->device_event = (struct vring_packed_desc_event *)
> +					(uintptr_t)ring_addr_to_vva(dev,
> +					vq, addr->used_user_addr, &len);
> +		if (vq->device_event == 0 ||

It's better to compare with NULL.

> +				len != sizeof(struct vring_packed_desc_event)) {
> +			RTE_LOG(DEBUG, VHOST_CONFIG,
> +				"(%d) failed to find device area address.\n",
> +				dev->vid);
> +			return dev;
> +		}
> +
>  		return dev;
>  	}
>  
> diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
> index d386cb6df..ce3b28313 100644
> --- a/lib/librte_vhost/virtio-packed.h
> +++ b/lib/librte_vhost/virtio-packed.h

Normally, we name c/h files as something like: virtio_packed.h

Besides, it'd be better to move the definitions into
vhost.h and define the types and flags for packed
ring only when e.g. VIRTIO_F_RING_PACKED not defined.
Just like how VIRTIO_F_IOMMU_PLATFORM works:

/* Declare IOMMU related bits for older kernels */
#ifndef VIRTIO_F_IOMMU_PLATFORM

#define VIRTIO_F_IOMMU_PLATFORM 33

struct vhost_iotlb_msg {
	__u64 iova;
	__u64 size;
	__u64 uaddr;
#define VHOST_ACCESS_RO      0x1
#define VHOST_ACCESS_WO      0x2
#define VHOST_ACCESS_RW      0x3
	__u8 perm;
#define VHOST_IOTLB_MISS           1
#define VHOST_IOTLB_UPDATE         2
#define VHOST_IOTLB_INVALIDATE     3
#define VHOST_IOTLB_ACCESS_FAIL    4
	__u8 type;
};

#define VHOST_IOTLB_MSG 0x1

struct vhost_msg {
	int type;
	union {
		struct vhost_iotlb_msg iotlb;
		__u8 padding[64];
	};
};
#endif

> @@ -19,6 +19,17 @@ struct vring_desc_packed {
>  	uint16_t flags;
>  };
>  
> +#define RING_EVENT_FLAGS_ENABLE 0x0
> +#define RING_EVENT_FLAGS_DISABLE 0x1
> +#define RING_EVENT_FLAGS_DESC 0x2
> +#define RING_EVENT_FLAGS_MASK 0xFFFC

RING_EVENT_FLAGS_MASK should be 0x3?

> +#define RING_EVENT_WRAP_MASK 0x8000
> +#define RING_EVENT_OFF_MASK 0x7FFF

It's better to define above macros as VRING_EVENT_*

> +
> +struct vring_packed_desc_event {
> +	uint16_t desc_event_off_wrap;
> +	uint16_t desc_event_flags;
> +};
>  
[...]

Re: [PATCH v6 14/15] vhost: add notification for packed ring

[Maxime Coquelin]

On 07/04/2018 08:25 AM, Tiwei Bie wrote:
> On Mon, Jul 02, 2018 at 10:16:28AM +0200, Maxime Coquelin wrote:
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>>   lib/librte_vhost/vhost.c         | 73 ++++++++++++++++++++++++++++++++++++----
>>   lib/librte_vhost/vhost.h         | 71 ++++++++++++++++++++++++++++++++++----
>>   lib/librte_vhost/vhost_user.c    | 24 +++++++++++++
>>   lib/librte_vhost/virtio-packed.h | 11 ++++++
>>   lib/librte_vhost/virtio_net.c    | 12 +++----
>>   5 files changed, 172 insertions(+), 19 deletions(-)
>>
>> diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c
>> index 8538302c9..78f20c402 100644
>> --- a/lib/librte_vhost/vhost.c
>> +++ b/lib/librte_vhost/vhost.c
>> @@ -171,6 +171,24 @@ vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
>>   	if (!vq->desc_packed || size != req_size)
>>   		return -1;
>>   
>> +	req_size = sizeof(struct vring_packed_desc_event);
>> +	size = req_size;
>> +	vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
>> +		vhost_iova_to_vva(dev,
>> +					vq,	vq->ring_addrs.avail_user_addr,
>> +					&size, VHOST_ACCESS_RW);
> 
> It should be a space instead of a tab after "vq,"
> 
> Why not put "vq, vq->ring_addrs.avail_user_addr,"
> and "vhost_iova_to_vva(dev," on the same line?
> 
>> +	if (!vq->driver_event || size != req_size)
>> +		return -1;
>> +
>> +	req_size = sizeof(struct vring_packed_desc_event);
>> +	size = req_size;
>> +	vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
>> +		vhost_iova_to_vva(dev,
>> +					vq, vq->ring_addrs.used_user_addr,
> 
> It's better to put "vhost_iova_to_vva(dev," and
> "vq, vq->ring_addrs.used_user_addr," on the same line.
> 
> Currently, it looks like something as this in my editor:
> 
> ¦       vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
> ¦       ¦       vhost_iova_to_vva(dev,
> ¦       ¦       ¦       ¦       ¦       vq, vq->ring_addrs.used_user_addr,
> ¦       ¦       ¦       ¦       ¦       &size, VHOST_ACCESS_RW);
> 

It looked a bit cleaner in my editor :) This is fixed now.

>> +					&size, VHOST_ACCESS_RW);
>> +	if (!vq->device_event || size != req_size)
>> +		return -1;
>> +
>>   	return 0;
>>   }
>>   
> [...]
>> @@ -640,4 +647,54 @@ vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
>>   	}
>>   }
>>   
>> +static __rte_always_inline void
>> +vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
>> +{
>> +	uint16_t old, new, off, off_wrap, wrap;
>> +	bool kick = false;
>> +
>> +
> 
> There is no need to have two blank lines.
> 
>> +	/*  Flush used desc update. */
> 
> Just need one space between "/*" and "Flush".
> 
>> +	rte_smp_mb();
>> +
>> +	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
>> +		if (vq->driver_event->desc_event_flags !=
>> +				RING_EVENT_FLAGS_DISABLE)
>> +			kick = true;
>> +		goto kick;
>> +	}
>> +
>> +	old = vq->signalled_used;
>> +	new = vq->last_used_idx;
>> +	vq->signalled_used = new;
>> +
>> +	if (vq->driver_event->desc_event_flags != RING_EVENT_FLAGS_DESC) {
>> +		if (vq->driver_event->desc_event_flags !=
>> +				RING_EVENT_FLAGS_DISABLE)
>> +			kick = true;
>> +		goto kick;
>> +	}
>> +
>> +	rte_smp_rmb();
>> +
>> +	off_wrap = vq->driver_event->desc_event_off_wrap;
>> +	off = off_wrap & ~(1 << 15);
> 
> Maybe it's better to use: RING_EVENT_OFF_MASK

I planned to remove its definition to be consistent with
the virtio_ring.h kernel header.

> 
>> +	wrap = vq->used_wrap_counter;
>> +
>> +	if (new < old) {
>> +		new += vq->size;
>> +		wrap ^= 1;
>> +	}
>> +
>> +	if (wrap != off_wrap >> 15)
>> +		off += vq->size;
>> +
>> +	if (vhost_need_event(off, new, old))
>> +		kick = true;
>> +
>> +kick:
>> +	if (kick)
>> +		eventfd_write(vq->callfd, (eventfd_t)1);
>> +}
>> +
>>   #endif /* _VHOST_NET_CDEV_H_ */
>> diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c
>> index b2b57de57..bda515bdb 100644
>> --- a/lib/librte_vhost/vhost_user.c
>> +++ b/lib/librte_vhost/vhost_user.c
>> @@ -523,6 +523,30 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index)
>>   		vq = dev->virtqueue[vq_index];
>>   		addr = &vq->ring_addrs;
>>   
>> +		len = sizeof(struct vring_packed_desc_event);
>> +		vq->driver_event = (struct vring_packed_desc_event *)
>> +					(uintptr_t)ring_addr_to_vva(dev,
>> +					vq, addr->avail_user_addr, &len);
>> +		if (vq->driver_event == 0 ||
> 
> It's better to compare with NULL.

Yes.

>> +				len != sizeof(struct vring_packed_desc_event)) {
>> +			RTE_LOG(DEBUG, VHOST_CONFIG,
>> +				"(%d) failed to find driver area address.\n",
>> +				dev->vid);
>> +			return dev;
>> +		}
>> +
>> +		len = sizeof(struct vring_packed_desc_event);
>> +		vq->device_event = (struct vring_packed_desc_event *)
>> +					(uintptr_t)ring_addr_to_vva(dev,
>> +					vq, addr->used_user_addr, &len);
>> +		if (vq->device_event == 0 ||
> 
> It's better to compare with NULL.
> 
>> +				len != sizeof(struct vring_packed_desc_event)) {
>> +			RTE_LOG(DEBUG, VHOST_CONFIG,
>> +				"(%d) failed to find device area address.\n",
>> +				dev->vid);
>> +			return dev;
>> +		}
>> +
>>   		return dev;
>>   	}
>>   
>> diff --git a/lib/librte_vhost/virtio-packed.h b/lib/librte_vhost/virtio-packed.h
>> index d386cb6df..ce3b28313 100644
>> --- a/lib/librte_vhost/virtio-packed.h
>> +++ b/lib/librte_vhost/virtio-packed.h
> 
> Normally, we name c/h files as something like: virtio_packed.h
> 
> Besides, it'd be better to move the definitions into
> vhost.h and define the types and flags for packed
> ring only when e.g. VIRTIO_F_RING_PACKED not defined.
> Just like how VIRTIO_F_IOMMU_PLATFORM works:
> 
> /* Declare IOMMU related bits for older kernels */
> #ifndef VIRTIO_F_IOMMU_PLATFORM
> 
> #define VIRTIO_F_IOMMU_PLATFORM 33
> 
> struct vhost_iotlb_msg {
> 	__u64 iova;
> 	__u64 size;
> 	__u64 uaddr;
> #define VHOST_ACCESS_RO      0x1
> #define VHOST_ACCESS_WO      0x2
> #define VHOST_ACCESS_RW      0x3
> 	__u8 perm;
> #define VHOST_IOTLB_MISS           1
> #define VHOST_IOTLB_UPDATE         2
> #define VHOST_IOTLB_INVALIDATE     3
> #define VHOST_IOTLB_ACCESS_FAIL    4
> 	__u8 type;
> };
> 
> #define VHOST_IOTLB_MSG 0x1
> 
> struct vhost_msg {
> 	int type;
> 	union {
> 		struct vhost_iotlb_msg iotlb;
> 		__u8 padding[64];
> 	};
> };
> #endif

I agree. I replied to Jason series so that the defines name are
consistent.

>> @@ -19,6 +19,17 @@ struct vring_desc_packed {
>>   	uint16_t flags;
>>   };
>>   
>> +#define RING_EVENT_FLAGS_ENABLE 0x0
>> +#define RING_EVENT_FLAGS_DISABLE 0x1
>> +#define RING_EVENT_FLAGS_DESC 0x2
>> +#define RING_EVENT_FLAGS_MASK 0xFFFC
> 
> RING_EVENT_FLAGS_MASK should be 0x3?

I actually removed it, but yes I agree.

>> +#define RING_EVENT_WRAP_MASK 0x8000
>> +#define RING_EVENT_OFF_MASK 0x7FFF
> 
> It's better to define above macros as VRING_EVENT_*

Yes, I change it to VRING_EVENT_F_* as you suggested to Jason.

>> +
>> +struct vring_packed_desc_event {
>> +	uint16_t desc_event_off_wrap;
>> +	uint16_t desc_event_flags;
>> +};
>>   
> [...]
> 

[PATCH v6 15/15] vhost: advertize packed ring layout support

[Maxime Coquelin]

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 lib/librte_vhost/vhost.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index 728fd2f6b..ef2558795 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -241,7 +241,8 @@ struct vhost_msg {
 				(1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
 				(1ULL << VIRTIO_RING_F_EVENT_IDX) | \
 				(1ULL << VIRTIO_NET_F_MTU) | \
-				(1ULL << VIRTIO_F_IOMMU_PLATFORM))
+				(1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
+				(1ULL << VIRTIO_F_RING_PACKED))
 
 
 struct guest_page {
-- 
2.14.4